git-svn: understand grafts when doing dcommit
[git/builtin-gsoc.git] / builtin-apply.c
blob976ec770417cba4113c4c55f5ca7a5f7b8520f1e
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
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"
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static int show_index_info;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
50 static enum whitespace_eol {
51 nowarn_whitespace,
52 warn_on_whitespace,
53 error_on_whitespace,
54 strip_whitespace,
55 } new_whitespace = warn_on_whitespace;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
63 if (!option) {
64 new_whitespace = warn_on_whitespace;
65 return;
67 if (!strcmp(option, "warn")) {
68 new_whitespace = warn_on_whitespace;
69 return;
71 if (!strcmp(option, "nowarn")) {
72 new_whitespace = nowarn_whitespace;
73 return;
75 if (!strcmp(option, "error")) {
76 new_whitespace = error_on_whitespace;
77 return;
79 if (!strcmp(option, "error-all")) {
80 new_whitespace = error_on_whitespace;
81 squelch_whitespace_errors = 0;
82 return;
84 if (!strcmp(option, "strip")) {
85 new_whitespace = strip_whitespace;
86 return;
88 die("unrecognized whitespace option '%s'", option);
91 static void set_default_whitespace_mode(const char *whitespace_option)
93 if (!whitespace_option && !apply_default_whitespace) {
94 new_whitespace = (apply
95 ? warn_on_whitespace
96 : nowarn_whitespace);
101 * For "diff-stat" like behaviour, we keep track of the biggest change
102 * we've seen, and the longest filename. That allows us to do simple
103 * scaling.
105 static int max_change, max_len;
108 * Various "current state", notably line numbers and what
109 * file (and how) we're patching right now.. The "is_xxxx"
110 * things are flags, where -1 means "don't know yet".
112 static int linenr = 1;
115 * This represents one "hunk" from a patch, starting with
116 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
117 * patch text is pointed at by patch, and its byte length
118 * is stored in size. leading and trailing are the number
119 * of context lines.
121 struct fragment {
122 unsigned long leading, trailing;
123 unsigned long oldpos, oldlines;
124 unsigned long newpos, newlines;
125 const char *patch;
126 int size;
127 int rejected;
128 struct fragment *next;
132 * When dealing with a binary patch, we reuse "leading" field
133 * to store the type of the binary hunk, either deflated "delta"
134 * or deflated "literal".
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED 1
138 #define BINARY_LITERAL_DEFLATED 2
140 struct patch {
141 char *new_name, *old_name, *def_name;
142 unsigned int old_mode, new_mode;
143 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
144 int rejected;
145 unsigned long deflate_origlen;
146 int lines_added, lines_deleted;
147 int score;
148 unsigned int is_toplevel_relative:1;
149 unsigned int inaccurate_eof:1;
150 unsigned int is_binary:1;
151 unsigned int is_copy:1;
152 unsigned int is_rename:1;
153 struct fragment *fragments;
154 char *result;
155 unsigned long resultsize;
156 char old_sha1_prefix[41];
157 char new_sha1_prefix[41];
158 struct patch *next;
161 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
163 fputs(pre, output);
164 if (patch->old_name && patch->new_name &&
165 strcmp(patch->old_name, patch->new_name)) {
166 write_name_quoted(NULL, 0, patch->old_name, 1, output);
167 fputs(" => ", output);
168 write_name_quoted(NULL, 0, patch->new_name, 1, output);
170 else {
171 const char *n = patch->new_name;
172 if (!n)
173 n = patch->old_name;
174 write_name_quoted(NULL, 0, n, 1, output);
176 fputs(post, output);
179 #define CHUNKSIZE (8192)
180 #define SLOP (16)
182 static void *read_patch_file(int fd, unsigned long *sizep)
184 unsigned long size = 0, alloc = CHUNKSIZE;
185 void *buffer = xmalloc(alloc);
187 for (;;) {
188 ssize_t nr = alloc - size;
189 if (nr < 1024) {
190 alloc += CHUNKSIZE;
191 buffer = xrealloc(buffer, alloc);
192 nr = alloc - size;
194 nr = xread(fd, (char *) buffer + size, nr);
195 if (!nr)
196 break;
197 if (nr < 0)
198 die("git-apply: read returned %s", strerror(errno));
199 size += nr;
201 *sizep = size;
204 * Make sure that we have some slop in the buffer
205 * so that we can do speculative "memcmp" etc, and
206 * see to it that it is NUL-filled.
208 if (alloc < size + SLOP)
209 buffer = xrealloc(buffer, size + SLOP);
210 memset((char *) buffer + size, 0, SLOP);
211 return buffer;
214 static unsigned long linelen(const char *buffer, unsigned long size)
216 unsigned long len = 0;
217 while (size--) {
218 len++;
219 if (*buffer++ == '\n')
220 break;
222 return len;
225 static int is_dev_null(const char *str)
227 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
230 #define TERM_SPACE 1
231 #define TERM_TAB 2
233 static int name_terminate(const char *name, int namelen, int c, int terminate)
235 if (c == ' ' && !(terminate & TERM_SPACE))
236 return 0;
237 if (c == '\t' && !(terminate & TERM_TAB))
238 return 0;
240 return 1;
243 static char *find_name(const char *line, char *def, int p_value, int terminate)
245 int len;
246 const char *start = line;
247 char *name;
249 if (*line == '"') {
250 /* Proposed "new-style" GNU patch/diff format; see
251 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
253 name = unquote_c_style(line, NULL);
254 if (name) {
255 char *cp = name;
256 while (p_value) {
257 cp = strchr(name, '/');
258 if (!cp)
259 break;
260 cp++;
261 p_value--;
263 if (cp) {
264 /* name can later be freed, so we need
265 * to memmove, not just return cp
267 memmove(name, cp, strlen(cp) + 1);
268 free(def);
269 return name;
271 else {
272 free(name);
273 name = NULL;
278 for (;;) {
279 char c = *line;
281 if (isspace(c)) {
282 if (c == '\n')
283 break;
284 if (name_terminate(start, line-start, c, terminate))
285 break;
287 line++;
288 if (c == '/' && !--p_value)
289 start = line;
291 if (!start)
292 return def;
293 len = line - start;
294 if (!len)
295 return def;
298 * Generally we prefer the shorter name, especially
299 * if the other one is just a variation of that with
300 * something else tacked on to the end (ie "file.orig"
301 * or "file~").
303 if (def) {
304 int deflen = strlen(def);
305 if (deflen < len && !strncmp(start, def, deflen))
306 return def;
309 name = xmalloc(len + 1);
310 memcpy(name, start, len);
311 name[len] = 0;
312 free(def);
313 return name;
316 static int count_slashes(const char *cp)
318 int cnt = 0;
319 char ch;
321 while ((ch = *cp++))
322 if (ch == '/')
323 cnt++;
324 return cnt;
328 * Given the string after "--- " or "+++ ", guess the appropriate
329 * p_value for the given patch.
331 static int guess_p_value(const char *nameline)
333 char *name, *cp;
334 int val = -1;
336 if (is_dev_null(nameline))
337 return -1;
338 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
339 if (!name)
340 return -1;
341 cp = strchr(name, '/');
342 if (!cp)
343 val = 0;
344 else if (prefix) {
346 * Does it begin with "a/$our-prefix" and such? Then this is
347 * very likely to apply to our directory.
349 if (!strncmp(name, prefix, prefix_length))
350 val = count_slashes(prefix);
351 else {
352 cp++;
353 if (!strncmp(cp, prefix, prefix_length))
354 val = count_slashes(prefix) + 1;
357 free(name);
358 return val;
362 * Get the name etc info from the --/+++ lines of a traditional patch header
364 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
365 * files, we can happily check the index for a match, but for creating a
366 * new file we should try to match whatever "patch" does. I have no idea.
368 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
370 char *name;
372 first += 4; /* skip "--- " */
373 second += 4; /* skip "+++ " */
374 if (!p_value_known) {
375 int p, q;
376 p = guess_p_value(first);
377 q = guess_p_value(second);
378 if (p < 0) p = q;
379 if (0 <= p && p == q) {
380 p_value = p;
381 p_value_known = 1;
384 if (is_dev_null(first)) {
385 patch->is_new = 1;
386 patch->is_delete = 0;
387 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
388 patch->new_name = name;
389 } else if (is_dev_null(second)) {
390 patch->is_new = 0;
391 patch->is_delete = 1;
392 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
393 patch->old_name = name;
394 } else {
395 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
396 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
397 patch->old_name = patch->new_name = name;
399 if (!name)
400 die("unable to find filename in patch at line %d", linenr);
403 static int gitdiff_hdrend(const char *line, struct patch *patch)
405 return -1;
409 * We're anal about diff header consistency, to make
410 * sure that we don't end up having strange ambiguous
411 * patches floating around.
413 * As a result, gitdiff_{old|new}name() will check
414 * their names against any previous information, just
415 * to make sure..
417 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
419 if (!orig_name && !isnull)
420 return find_name(line, NULL, p_value, TERM_TAB);
422 if (orig_name) {
423 int len;
424 const char *name;
425 char *another;
426 name = orig_name;
427 len = strlen(name);
428 if (isnull)
429 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
430 another = find_name(line, NULL, p_value, TERM_TAB);
431 if (!another || memcmp(another, name, len))
432 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
433 free(another);
434 return orig_name;
436 else {
437 /* expect "/dev/null" */
438 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
439 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
440 return NULL;
444 static int gitdiff_oldname(const char *line, struct patch *patch)
446 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
447 return 0;
450 static int gitdiff_newname(const char *line, struct patch *patch)
452 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
453 return 0;
456 static int gitdiff_oldmode(const char *line, struct patch *patch)
458 patch->old_mode = strtoul(line, NULL, 8);
459 return 0;
462 static int gitdiff_newmode(const char *line, struct patch *patch)
464 patch->new_mode = strtoul(line, NULL, 8);
465 return 0;
468 static int gitdiff_delete(const char *line, struct patch *patch)
470 patch->is_delete = 1;
471 patch->old_name = patch->def_name;
472 return gitdiff_oldmode(line, patch);
475 static int gitdiff_newfile(const char *line, struct patch *patch)
477 patch->is_new = 1;
478 patch->new_name = patch->def_name;
479 return gitdiff_newmode(line, patch);
482 static int gitdiff_copysrc(const char *line, struct patch *patch)
484 patch->is_copy = 1;
485 patch->old_name = find_name(line, NULL, 0, 0);
486 return 0;
489 static int gitdiff_copydst(const char *line, struct patch *patch)
491 patch->is_copy = 1;
492 patch->new_name = find_name(line, NULL, 0, 0);
493 return 0;
496 static int gitdiff_renamesrc(const char *line, struct patch *patch)
498 patch->is_rename = 1;
499 patch->old_name = find_name(line, NULL, 0, 0);
500 return 0;
503 static int gitdiff_renamedst(const char *line, struct patch *patch)
505 patch->is_rename = 1;
506 patch->new_name = find_name(line, NULL, 0, 0);
507 return 0;
510 static int gitdiff_similarity(const char *line, struct patch *patch)
512 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
513 patch->score = 0;
514 return 0;
517 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
519 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
520 patch->score = 0;
521 return 0;
524 static int gitdiff_index(const char *line, struct patch *patch)
526 /* index line is N hexadecimal, "..", N hexadecimal,
527 * and optional space with octal mode.
529 const char *ptr, *eol;
530 int len;
532 ptr = strchr(line, '.');
533 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
534 return 0;
535 len = ptr - line;
536 memcpy(patch->old_sha1_prefix, line, len);
537 patch->old_sha1_prefix[len] = 0;
539 line = ptr + 2;
540 ptr = strchr(line, ' ');
541 eol = strchr(line, '\n');
543 if (!ptr || eol < ptr)
544 ptr = eol;
545 len = ptr - line;
547 if (40 < len)
548 return 0;
549 memcpy(patch->new_sha1_prefix, line, len);
550 patch->new_sha1_prefix[len] = 0;
551 if (*ptr == ' ')
552 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
553 return 0;
557 * This is normal for a diff that doesn't change anything: we'll fall through
558 * into the next diff. Tell the parser to break out.
560 static int gitdiff_unrecognized(const char *line, struct patch *patch)
562 return -1;
565 static const char *stop_at_slash(const char *line, int llen)
567 int i;
569 for (i = 0; i < llen; i++) {
570 int ch = line[i];
571 if (ch == '/')
572 return line + i;
574 return NULL;
577 /* This is to extract the same name that appears on "diff --git"
578 * line. We do not find and return anything if it is a rename
579 * patch, and it is OK because we will find the name elsewhere.
580 * We need to reliably find name only when it is mode-change only,
581 * creation or deletion of an empty file. In any of these cases,
582 * both sides are the same name under a/ and b/ respectively.
584 static char *git_header_name(char *line, int llen)
586 int len;
587 const char *name;
588 const char *second = NULL;
590 line += strlen("diff --git ");
591 llen -= strlen("diff --git ");
593 if (*line == '"') {
594 const char *cp;
595 char *first = unquote_c_style(line, &second);
596 if (!first)
597 return NULL;
599 /* advance to the first slash */
600 cp = stop_at_slash(first, strlen(first));
601 if (!cp || cp == first) {
602 /* we do not accept absolute paths */
603 free_first_and_fail:
604 free(first);
605 return NULL;
607 len = strlen(cp+1);
608 memmove(first, cp+1, len+1); /* including NUL */
610 /* second points at one past closing dq of name.
611 * find the second name.
613 while ((second < line + llen) && isspace(*second))
614 second++;
616 if (line + llen <= second)
617 goto free_first_and_fail;
618 if (*second == '"') {
619 char *sp = unquote_c_style(second, NULL);
620 if (!sp)
621 goto free_first_and_fail;
622 cp = stop_at_slash(sp, strlen(sp));
623 if (!cp || cp == sp) {
624 free_both_and_fail:
625 free(sp);
626 goto free_first_and_fail;
628 /* They must match, otherwise ignore */
629 if (strcmp(cp+1, first))
630 goto free_both_and_fail;
631 free(sp);
632 return first;
635 /* unquoted second */
636 cp = stop_at_slash(second, line + llen - second);
637 if (!cp || cp == second)
638 goto free_first_and_fail;
639 cp++;
640 if (line + llen - cp != len + 1 ||
641 memcmp(first, cp, len))
642 goto free_first_and_fail;
643 return first;
646 /* unquoted first name */
647 name = stop_at_slash(line, llen);
648 if (!name || name == line)
649 return NULL;
651 name++;
653 /* since the first name is unquoted, a dq if exists must be
654 * the beginning of the second name.
656 for (second = name; second < line + llen; second++) {
657 if (*second == '"') {
658 const char *cp = second;
659 const char *np;
660 char *sp = unquote_c_style(second, NULL);
662 if (!sp)
663 return NULL;
664 np = stop_at_slash(sp, strlen(sp));
665 if (!np || np == sp) {
666 free_second_and_fail:
667 free(sp);
668 return NULL;
670 np++;
671 len = strlen(np);
672 if (len < cp - name &&
673 !strncmp(np, name, len) &&
674 isspace(name[len])) {
675 /* Good */
676 memmove(sp, np, len + 1);
677 return sp;
679 goto free_second_and_fail;
684 * Accept a name only if it shows up twice, exactly the same
685 * form.
687 for (len = 0 ; ; len++) {
688 switch (name[len]) {
689 default:
690 continue;
691 case '\n':
692 return NULL;
693 case '\t': case ' ':
694 second = name+len;
695 for (;;) {
696 char c = *second++;
697 if (c == '\n')
698 return NULL;
699 if (c == '/')
700 break;
702 if (second[len] == '\n' && !memcmp(name, second, len)) {
703 char *ret = xmalloc(len + 1);
704 memcpy(ret, name, len);
705 ret[len] = 0;
706 return ret;
710 return NULL;
713 /* Verify that we recognize the lines following a git header */
714 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
716 unsigned long offset;
718 /* A git diff has explicit new/delete information, so we don't guess */
719 patch->is_new = 0;
720 patch->is_delete = 0;
723 * Some things may not have the old name in the
724 * rest of the headers anywhere (pure mode changes,
725 * or removing or adding empty files), so we get
726 * the default name from the header.
728 patch->def_name = git_header_name(line, len);
730 line += len;
731 size -= len;
732 linenr++;
733 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
734 static const struct opentry {
735 const char *str;
736 int (*fn)(const char *, struct patch *);
737 } optable[] = {
738 { "@@ -", gitdiff_hdrend },
739 { "--- ", gitdiff_oldname },
740 { "+++ ", gitdiff_newname },
741 { "old mode ", gitdiff_oldmode },
742 { "new mode ", gitdiff_newmode },
743 { "deleted file mode ", gitdiff_delete },
744 { "new file mode ", gitdiff_newfile },
745 { "copy from ", gitdiff_copysrc },
746 { "copy to ", gitdiff_copydst },
747 { "rename old ", gitdiff_renamesrc },
748 { "rename new ", gitdiff_renamedst },
749 { "rename from ", gitdiff_renamesrc },
750 { "rename to ", gitdiff_renamedst },
751 { "similarity index ", gitdiff_similarity },
752 { "dissimilarity index ", gitdiff_dissimilarity },
753 { "index ", gitdiff_index },
754 { "", gitdiff_unrecognized },
756 int i;
758 len = linelen(line, size);
759 if (!len || line[len-1] != '\n')
760 break;
761 for (i = 0; i < ARRAY_SIZE(optable); i++) {
762 const struct opentry *p = optable + i;
763 int oplen = strlen(p->str);
764 if (len < oplen || memcmp(p->str, line, oplen))
765 continue;
766 if (p->fn(line + oplen, patch) < 0)
767 return offset;
768 break;
772 return offset;
775 static int parse_num(const char *line, unsigned long *p)
777 char *ptr;
779 if (!isdigit(*line))
780 return 0;
781 *p = strtoul(line, &ptr, 10);
782 return ptr - line;
785 static int parse_range(const char *line, int len, int offset, const char *expect,
786 unsigned long *p1, unsigned long *p2)
788 int digits, ex;
790 if (offset < 0 || offset >= len)
791 return -1;
792 line += offset;
793 len -= offset;
795 digits = parse_num(line, p1);
796 if (!digits)
797 return -1;
799 offset += digits;
800 line += digits;
801 len -= digits;
803 *p2 = 1;
804 if (*line == ',') {
805 digits = parse_num(line+1, p2);
806 if (!digits)
807 return -1;
809 offset += digits+1;
810 line += digits+1;
811 len -= digits+1;
814 ex = strlen(expect);
815 if (ex > len)
816 return -1;
817 if (memcmp(line, expect, ex))
818 return -1;
820 return offset + ex;
824 * Parse a unified diff fragment header of the
825 * form "@@ -a,b +c,d @@"
827 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
829 int offset;
831 if (!len || line[len-1] != '\n')
832 return -1;
834 /* Figure out the number of lines in a fragment */
835 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
836 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
838 return offset;
841 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
843 unsigned long offset, len;
845 patch->is_toplevel_relative = 0;
846 patch->is_rename = patch->is_copy = 0;
847 patch->is_new = patch->is_delete = -1;
848 patch->old_mode = patch->new_mode = 0;
849 patch->old_name = patch->new_name = NULL;
850 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
851 unsigned long nextlen;
853 len = linelen(line, size);
854 if (!len)
855 break;
857 /* Testing this early allows us to take a few shortcuts.. */
858 if (len < 6)
859 continue;
862 * Make sure we don't find any unconnected patch fragments.
863 * That's a sign that we didn't find a header, and that a
864 * patch has become corrupted/broken up.
866 if (!memcmp("@@ -", line, 4)) {
867 struct fragment dummy;
868 if (parse_fragment_header(line, len, &dummy) < 0)
869 continue;
870 die("patch fragment without header at line %d: %.*s",
871 linenr, (int)len-1, line);
874 if (size < len + 6)
875 break;
878 * Git patch? It might not have a real patch, just a rename
879 * or mode change, so we handle that specially
881 if (!memcmp("diff --git ", line, 11)) {
882 int git_hdr_len = parse_git_header(line, len, size, patch);
883 if (git_hdr_len <= len)
884 continue;
885 if (!patch->old_name && !patch->new_name) {
886 if (!patch->def_name)
887 die("git diff header lacks filename information (line %d)", linenr);
888 patch->old_name = patch->new_name = patch->def_name;
890 patch->is_toplevel_relative = 1;
891 *hdrsize = git_hdr_len;
892 return offset;
895 /** --- followed by +++ ? */
896 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
897 continue;
900 * We only accept unified patches, so we want it to
901 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
902 * minimum
904 nextlen = linelen(line + len, size - len);
905 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
906 continue;
908 /* Ok, we'll consider it a patch */
909 parse_traditional_patch(line, line+len, patch);
910 *hdrsize = len + nextlen;
911 linenr += 2;
912 return offset;
914 return -1;
917 static void check_whitespace(const char *line, int len)
919 const char *err = "Adds trailing whitespace";
920 int seen_space = 0;
921 int i;
924 * We know len is at least two, since we have a '+' and we
925 * checked that the last character was a '\n' before calling
926 * this function. That is, an addition of an empty line would
927 * check the '+' here. Sneaky...
929 if (isspace(line[len-2]))
930 goto error;
933 * Make sure that there is no space followed by a tab in
934 * indentation.
936 err = "Space in indent is followed by a tab";
937 for (i = 1; i < len; i++) {
938 if (line[i] == '\t') {
939 if (seen_space)
940 goto error;
942 else if (line[i] == ' ')
943 seen_space = 1;
944 else
945 break;
947 return;
949 error:
950 whitespace_error++;
951 if (squelch_whitespace_errors &&
952 squelch_whitespace_errors < whitespace_error)
954 else
955 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
956 err, patch_input_file, linenr, len-2, line+1);
961 * Parse a unified diff. Note that this really needs to parse each
962 * fragment separately, since the only way to know the difference
963 * between a "---" that is part of a patch, and a "---" that starts
964 * the next patch is to look at the line counts..
966 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
968 int added, deleted;
969 int len = linelen(line, size), offset;
970 unsigned long oldlines, newlines;
971 unsigned long leading, trailing;
973 offset = parse_fragment_header(line, len, fragment);
974 if (offset < 0)
975 return -1;
976 oldlines = fragment->oldlines;
977 newlines = fragment->newlines;
978 leading = 0;
979 trailing = 0;
981 /* Parse the thing.. */
982 line += len;
983 size -= len;
984 linenr++;
985 added = deleted = 0;
986 for (offset = len;
987 0 < size;
988 offset += len, size -= len, line += len, linenr++) {
989 if (!oldlines && !newlines)
990 break;
991 len = linelen(line, size);
992 if (!len || line[len-1] != '\n')
993 return -1;
994 switch (*line) {
995 default:
996 return -1;
997 case '\n': /* newer GNU diff, an empty context line */
998 case ' ':
999 oldlines--;
1000 newlines--;
1001 if (!deleted && !added)
1002 leading++;
1003 trailing++;
1004 break;
1005 case '-':
1006 if (apply_in_reverse &&
1007 new_whitespace != nowarn_whitespace)
1008 check_whitespace(line, len);
1009 deleted++;
1010 oldlines--;
1011 trailing = 0;
1012 break;
1013 case '+':
1014 if (!apply_in_reverse &&
1015 new_whitespace != nowarn_whitespace)
1016 check_whitespace(line, len);
1017 added++;
1018 newlines--;
1019 trailing = 0;
1020 break;
1022 /* We allow "\ No newline at end of file". Depending
1023 * on locale settings when the patch was produced we
1024 * don't know what this line looks like. The only
1025 * thing we do know is that it begins with "\ ".
1026 * Checking for 12 is just for sanity check -- any
1027 * l10n of "\ No newline..." is at least that long.
1029 case '\\':
1030 if (len < 12 || memcmp(line, "\\ ", 2))
1031 return -1;
1032 break;
1035 if (oldlines || newlines)
1036 return -1;
1037 fragment->leading = leading;
1038 fragment->trailing = trailing;
1040 /* If a fragment ends with an incomplete line, we failed to include
1041 * it in the above loop because we hit oldlines == newlines == 0
1042 * before seeing it.
1044 if (12 < size && !memcmp(line, "\\ ", 2))
1045 offset += linelen(line, size);
1047 patch->lines_added += added;
1048 patch->lines_deleted += deleted;
1050 if (0 < patch->is_new && oldlines)
1051 return error("new file depends on old contents");
1052 if (0 < patch->is_delete && newlines)
1053 return error("deleted file still has contents");
1054 return offset;
1057 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1059 unsigned long offset = 0;
1060 unsigned long oldlines = 0, newlines = 0, context = 0;
1061 struct fragment **fragp = &patch->fragments;
1063 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1064 struct fragment *fragment;
1065 int len;
1067 fragment = xcalloc(1, sizeof(*fragment));
1068 len = parse_fragment(line, size, patch, fragment);
1069 if (len <= 0)
1070 die("corrupt patch at line %d", linenr);
1071 fragment->patch = line;
1072 fragment->size = len;
1073 oldlines += fragment->oldlines;
1074 newlines += fragment->newlines;
1075 context += fragment->leading + fragment->trailing;
1077 *fragp = fragment;
1078 fragp = &fragment->next;
1080 offset += len;
1081 line += len;
1082 size -= len;
1086 * If something was removed (i.e. we have old-lines) it cannot
1087 * be creation, and if something was added it cannot be
1088 * deletion. However, the reverse is not true; --unified=0
1089 * patches that only add are not necessarily creation even
1090 * though they do not have any old lines, and ones that only
1091 * delete are not necessarily deletion.
1093 * Unfortunately, a real creation/deletion patch do _not_ have
1094 * any context line by definition, so we cannot safely tell it
1095 * apart with --unified=0 insanity. At least if the patch has
1096 * more than one hunk it is not creation or deletion.
1098 if (patch->is_new < 0 &&
1099 (oldlines || (patch->fragments && patch->fragments->next)))
1100 patch->is_new = 0;
1101 if (patch->is_delete < 0 &&
1102 (newlines || (patch->fragments && patch->fragments->next)))
1103 patch->is_delete = 0;
1104 if (!unidiff_zero || context) {
1105 /* If the user says the patch is not generated with
1106 * --unified=0, or if we have seen context lines,
1107 * then not having oldlines means the patch is creation,
1108 * and not having newlines means the patch is deletion.
1110 if (patch->is_new < 0 && !oldlines) {
1111 patch->is_new = 1;
1112 patch->old_name = NULL;
1114 if (patch->is_delete < 0 && !newlines) {
1115 patch->is_delete = 1;
1116 patch->new_name = NULL;
1120 if (0 < patch->is_new && oldlines)
1121 die("new file %s depends on old contents", patch->new_name);
1122 if (0 < patch->is_delete && newlines)
1123 die("deleted file %s still has contents", patch->old_name);
1124 if (!patch->is_delete && !newlines && context)
1125 fprintf(stderr, "** warning: file %s becomes empty but "
1126 "is not deleted\n", patch->new_name);
1128 return offset;
1131 static inline int metadata_changes(struct patch *patch)
1133 return patch->is_rename > 0 ||
1134 patch->is_copy > 0 ||
1135 patch->is_new > 0 ||
1136 patch->is_delete ||
1137 (patch->old_mode && patch->new_mode &&
1138 patch->old_mode != patch->new_mode);
1141 static char *inflate_it(const void *data, unsigned long size,
1142 unsigned long inflated_size)
1144 z_stream stream;
1145 void *out;
1146 int st;
1148 memset(&stream, 0, sizeof(stream));
1150 stream.next_in = (unsigned char *)data;
1151 stream.avail_in = size;
1152 stream.next_out = out = xmalloc(inflated_size);
1153 stream.avail_out = inflated_size;
1154 inflateInit(&stream);
1155 st = inflate(&stream, Z_FINISH);
1156 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1157 free(out);
1158 return NULL;
1160 return out;
1163 static struct fragment *parse_binary_hunk(char **buf_p,
1164 unsigned long *sz_p,
1165 int *status_p,
1166 int *used_p)
1168 /* Expect a line that begins with binary patch method ("literal"
1169 * or "delta"), followed by the length of data before deflating.
1170 * a sequence of 'length-byte' followed by base-85 encoded data
1171 * should follow, terminated by a newline.
1173 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1174 * and we would limit the patch line to 66 characters,
1175 * so one line can fit up to 13 groups that would decode
1176 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1177 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1179 int llen, used;
1180 unsigned long size = *sz_p;
1181 char *buffer = *buf_p;
1182 int patch_method;
1183 unsigned long origlen;
1184 char *data = NULL;
1185 int hunk_size = 0;
1186 struct fragment *frag;
1188 llen = linelen(buffer, size);
1189 used = llen;
1191 *status_p = 0;
1193 if (!prefixcmp(buffer, "delta ")) {
1194 patch_method = BINARY_DELTA_DEFLATED;
1195 origlen = strtoul(buffer + 6, NULL, 10);
1197 else if (!prefixcmp(buffer, "literal ")) {
1198 patch_method = BINARY_LITERAL_DEFLATED;
1199 origlen = strtoul(buffer + 8, NULL, 10);
1201 else
1202 return NULL;
1204 linenr++;
1205 buffer += llen;
1206 while (1) {
1207 int byte_length, max_byte_length, newsize;
1208 llen = linelen(buffer, size);
1209 used += llen;
1210 linenr++;
1211 if (llen == 1) {
1212 /* consume the blank line */
1213 buffer++;
1214 size--;
1215 break;
1217 /* Minimum line is "A00000\n" which is 7-byte long,
1218 * and the line length must be multiple of 5 plus 2.
1220 if ((llen < 7) || (llen-2) % 5)
1221 goto corrupt;
1222 max_byte_length = (llen - 2) / 5 * 4;
1223 byte_length = *buffer;
1224 if ('A' <= byte_length && byte_length <= 'Z')
1225 byte_length = byte_length - 'A' + 1;
1226 else if ('a' <= byte_length && byte_length <= 'z')
1227 byte_length = byte_length - 'a' + 27;
1228 else
1229 goto corrupt;
1230 /* if the input length was not multiple of 4, we would
1231 * have filler at the end but the filler should never
1232 * exceed 3 bytes
1234 if (max_byte_length < byte_length ||
1235 byte_length <= max_byte_length - 4)
1236 goto corrupt;
1237 newsize = hunk_size + byte_length;
1238 data = xrealloc(data, newsize);
1239 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1240 goto corrupt;
1241 hunk_size = newsize;
1242 buffer += llen;
1243 size -= llen;
1246 frag = xcalloc(1, sizeof(*frag));
1247 frag->patch = inflate_it(data, hunk_size, origlen);
1248 if (!frag->patch)
1249 goto corrupt;
1250 free(data);
1251 frag->size = origlen;
1252 *buf_p = buffer;
1253 *sz_p = size;
1254 *used_p = used;
1255 frag->binary_patch_method = patch_method;
1256 return frag;
1258 corrupt:
1259 free(data);
1260 *status_p = -1;
1261 error("corrupt binary patch at line %d: %.*s",
1262 linenr-1, llen-1, buffer);
1263 return NULL;
1266 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1268 /* We have read "GIT binary patch\n"; what follows is a line
1269 * that says the patch method (currently, either "literal" or
1270 * "delta") and the length of data before deflating; a
1271 * sequence of 'length-byte' followed by base-85 encoded data
1272 * follows.
1274 * When a binary patch is reversible, there is another binary
1275 * hunk in the same format, starting with patch method (either
1276 * "literal" or "delta") with the length of data, and a sequence
1277 * of length-byte + base-85 encoded data, terminated with another
1278 * empty line. This data, when applied to the postimage, produces
1279 * the preimage.
1281 struct fragment *forward;
1282 struct fragment *reverse;
1283 int status;
1284 int used, used_1;
1286 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1287 if (!forward && !status)
1288 /* there has to be one hunk (forward hunk) */
1289 return error("unrecognized binary patch at line %d", linenr-1);
1290 if (status)
1291 /* otherwise we already gave an error message */
1292 return status;
1294 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1295 if (reverse)
1296 used += used_1;
1297 else if (status) {
1298 /* not having reverse hunk is not an error, but having
1299 * a corrupt reverse hunk is.
1301 free((void*) forward->patch);
1302 free(forward);
1303 return status;
1305 forward->next = reverse;
1306 patch->fragments = forward;
1307 patch->is_binary = 1;
1308 return used;
1311 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1313 int hdrsize, patchsize;
1314 int offset = find_header(buffer, size, &hdrsize, patch);
1316 if (offset < 0)
1317 return offset;
1319 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1321 if (!patchsize) {
1322 static const char *binhdr[] = {
1323 "Binary files ",
1324 "Files ",
1325 NULL,
1327 static const char git_binary[] = "GIT binary patch\n";
1328 int i;
1329 int hd = hdrsize + offset;
1330 unsigned long llen = linelen(buffer + hd, size - hd);
1332 if (llen == sizeof(git_binary) - 1 &&
1333 !memcmp(git_binary, buffer + hd, llen)) {
1334 int used;
1335 linenr++;
1336 used = parse_binary(buffer + hd + llen,
1337 size - hd - llen, patch);
1338 if (used)
1339 patchsize = used + llen;
1340 else
1341 patchsize = 0;
1343 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1344 for (i = 0; binhdr[i]; i++) {
1345 int len = strlen(binhdr[i]);
1346 if (len < size - hd &&
1347 !memcmp(binhdr[i], buffer + hd, len)) {
1348 linenr++;
1349 patch->is_binary = 1;
1350 patchsize = llen;
1351 break;
1356 /* Empty patch cannot be applied if it is a text patch
1357 * without metadata change. A binary patch appears
1358 * empty to us here.
1360 if ((apply || check) &&
1361 (!patch->is_binary && !metadata_changes(patch)))
1362 die("patch with only garbage at line %d", linenr);
1365 return offset + hdrsize + patchsize;
1368 #define swap(a,b) myswap((a),(b),sizeof(a))
1370 #define myswap(a, b, size) do { \
1371 unsigned char mytmp[size]; \
1372 memcpy(mytmp, &a, size); \
1373 memcpy(&a, &b, size); \
1374 memcpy(&b, mytmp, size); \
1375 } while (0)
1377 static void reverse_patches(struct patch *p)
1379 for (; p; p = p->next) {
1380 struct fragment *frag = p->fragments;
1382 swap(p->new_name, p->old_name);
1383 swap(p->new_mode, p->old_mode);
1384 swap(p->is_new, p->is_delete);
1385 swap(p->lines_added, p->lines_deleted);
1386 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1388 for (; frag; frag = frag->next) {
1389 swap(frag->newpos, frag->oldpos);
1390 swap(frag->newlines, frag->oldlines);
1395 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1396 static const char minuses[]= "----------------------------------------------------------------------";
1398 static void show_stats(struct patch *patch)
1400 const char *prefix = "";
1401 char *name = patch->new_name;
1402 char *qname = NULL;
1403 int len, max, add, del, total;
1405 if (!name)
1406 name = patch->old_name;
1408 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1409 qname = xmalloc(len + 1);
1410 quote_c_style(name, qname, NULL, 0);
1411 name = qname;
1415 * "scale" the filename
1417 len = strlen(name);
1418 max = max_len;
1419 if (max > 50)
1420 max = 50;
1421 if (len > max) {
1422 char *slash;
1423 prefix = "...";
1424 max -= 3;
1425 name += len - max;
1426 slash = strchr(name, '/');
1427 if (slash)
1428 name = slash;
1430 len = max;
1433 * scale the add/delete
1435 max = max_change;
1436 if (max + len > 70)
1437 max = 70 - len;
1439 add = patch->lines_added;
1440 del = patch->lines_deleted;
1441 total = add + del;
1443 if (max_change > 0) {
1444 total = (total * max + max_change / 2) / max_change;
1445 add = (add * max + max_change / 2) / max_change;
1446 del = total - add;
1448 if (patch->is_binary)
1449 printf(" %s%-*s | Bin\n", prefix, len, name);
1450 else
1451 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1452 len, name, patch->lines_added + patch->lines_deleted,
1453 add, pluses, del, minuses);
1454 free(qname);
1457 static int read_old_data(struct stat *st, const char *path, char **buf_p, unsigned long *alloc_p, unsigned long *size_p)
1459 int fd;
1460 unsigned long got;
1461 unsigned long nsize;
1462 char *nbuf;
1463 unsigned long size = *size_p;
1464 char *buf = *buf_p;
1466 switch (st->st_mode & S_IFMT) {
1467 case S_IFLNK:
1468 return readlink(path, buf, size) != size;
1469 case S_IFREG:
1470 fd = open(path, O_RDONLY);
1471 if (fd < 0)
1472 return error("unable to open %s", path);
1473 got = 0;
1474 for (;;) {
1475 ssize_t ret = xread(fd, buf + got, size - got);
1476 if (ret <= 0)
1477 break;
1478 got += ret;
1480 close(fd);
1481 nsize = got;
1482 nbuf = convert_to_git(path, buf, &nsize);
1483 if (nbuf) {
1484 free(buf);
1485 *buf_p = nbuf;
1486 *alloc_p = nsize;
1487 *size_p = nsize;
1489 return got != size;
1490 default:
1491 return -1;
1495 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1497 int i;
1498 unsigned long start, backwards, forwards;
1500 if (fragsize > size)
1501 return -1;
1503 start = 0;
1504 if (line > 1) {
1505 unsigned long offset = 0;
1506 i = line-1;
1507 while (offset + fragsize <= size) {
1508 if (buf[offset++] == '\n') {
1509 start = offset;
1510 if (!--i)
1511 break;
1516 /* Exact line number? */
1517 if ((start + fragsize <= size) &&
1518 !memcmp(buf + start, fragment, fragsize))
1519 return start;
1522 * There's probably some smart way to do this, but I'll leave
1523 * that to the smart and beautiful people. I'm simple and stupid.
1525 backwards = start;
1526 forwards = start;
1527 for (i = 0; ; i++) {
1528 unsigned long try;
1529 int n;
1531 /* "backward" */
1532 if (i & 1) {
1533 if (!backwards) {
1534 if (forwards + fragsize > size)
1535 break;
1536 continue;
1538 do {
1539 --backwards;
1540 } while (backwards && buf[backwards-1] != '\n');
1541 try = backwards;
1542 } else {
1543 while (forwards + fragsize <= size) {
1544 if (buf[forwards++] == '\n')
1545 break;
1547 try = forwards;
1550 if (try + fragsize > size)
1551 continue;
1552 if (memcmp(buf + try, fragment, fragsize))
1553 continue;
1554 n = (i >> 1)+1;
1555 if (i & 1)
1556 n = -n;
1557 *lines = n;
1558 return try;
1562 * We should start searching forward and backward.
1564 return -1;
1567 static void remove_first_line(const char **rbuf, int *rsize)
1569 const char *buf = *rbuf;
1570 int size = *rsize;
1571 unsigned long offset;
1572 offset = 0;
1573 while (offset <= size) {
1574 if (buf[offset++] == '\n')
1575 break;
1577 *rsize = size - offset;
1578 *rbuf = buf + offset;
1581 static void remove_last_line(const char **rbuf, int *rsize)
1583 const char *buf = *rbuf;
1584 int size = *rsize;
1585 unsigned long offset;
1586 offset = size - 1;
1587 while (offset > 0) {
1588 if (buf[--offset] == '\n')
1589 break;
1591 *rsize = offset + 1;
1594 struct buffer_desc {
1595 char *buffer;
1596 unsigned long size;
1597 unsigned long alloc;
1600 static int apply_line(char *output, const char *patch, int plen)
1602 /* plen is number of bytes to be copied from patch,
1603 * starting at patch+1 (patch[0] is '+'). Typically
1604 * patch[plen] is '\n', unless this is the incomplete
1605 * last line.
1607 int i;
1608 int add_nl_to_tail = 0;
1609 int fixed = 0;
1610 int last_tab_in_indent = -1;
1611 int last_space_in_indent = -1;
1612 int need_fix_leading_space = 0;
1613 char *buf;
1615 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1616 *patch != '+') {
1617 memcpy(output, patch + 1, plen);
1618 return plen;
1621 if (1 < plen && isspace(patch[plen-1])) {
1622 if (patch[plen] == '\n')
1623 add_nl_to_tail = 1;
1624 plen--;
1625 while (0 < plen && isspace(patch[plen]))
1626 plen--;
1627 fixed = 1;
1630 for (i = 1; i < plen; i++) {
1631 char ch = patch[i];
1632 if (ch == '\t') {
1633 last_tab_in_indent = i;
1634 if (0 <= last_space_in_indent)
1635 need_fix_leading_space = 1;
1637 else if (ch == ' ')
1638 last_space_in_indent = i;
1639 else
1640 break;
1643 buf = output;
1644 if (need_fix_leading_space) {
1645 /* between patch[1..last_tab_in_indent] strip the
1646 * funny spaces, updating them to tab as needed.
1648 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1649 char ch = patch[i];
1650 if (ch != ' ')
1651 *output++ = ch;
1652 else if ((i % 8) == 0)
1653 *output++ = '\t';
1655 fixed = 1;
1656 i = last_tab_in_indent;
1658 else
1659 i = 1;
1661 memcpy(output, patch + i, plen);
1662 if (add_nl_to_tail)
1663 output[plen++] = '\n';
1664 if (fixed)
1665 applied_after_fixing_ws++;
1666 return output + plen - buf;
1669 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag, int inaccurate_eof)
1671 int match_beginning, match_end;
1672 char *buf = desc->buffer;
1673 const char *patch = frag->patch;
1674 int offset, size = frag->size;
1675 char *old = xmalloc(size);
1676 char *new = xmalloc(size);
1677 const char *oldlines, *newlines;
1678 int oldsize = 0, newsize = 0;
1679 int new_blank_lines_at_end = 0;
1680 unsigned long leading, trailing;
1681 int pos, lines;
1683 while (size > 0) {
1684 char first;
1685 int len = linelen(patch, size);
1686 int plen;
1687 int added_blank_line = 0;
1689 if (!len)
1690 break;
1693 * "plen" is how much of the line we should use for
1694 * the actual patch data. Normally we just remove the
1695 * first character on the line, but if the line is
1696 * followed by "\ No newline", then we also remove the
1697 * last one (which is the newline, of course).
1699 plen = len-1;
1700 if (len < size && patch[len] == '\\')
1701 plen--;
1702 first = *patch;
1703 if (apply_in_reverse) {
1704 if (first == '-')
1705 first = '+';
1706 else if (first == '+')
1707 first = '-';
1710 switch (first) {
1711 case '\n':
1712 /* Newer GNU diff, empty context line */
1713 if (plen < 0)
1714 /* ... followed by '\No newline'; nothing */
1715 break;
1716 old[oldsize++] = '\n';
1717 new[newsize++] = '\n';
1718 break;
1719 case ' ':
1720 case '-':
1721 memcpy(old + oldsize, patch + 1, plen);
1722 oldsize += plen;
1723 if (first == '-')
1724 break;
1725 /* Fall-through for ' ' */
1726 case '+':
1727 if (first != '+' || !no_add) {
1728 int added = apply_line(new + newsize, patch,
1729 plen);
1730 newsize += added;
1731 if (first == '+' &&
1732 added == 1 && new[newsize-1] == '\n')
1733 added_blank_line = 1;
1735 break;
1736 case '@': case '\\':
1737 /* Ignore it, we already handled it */
1738 break;
1739 default:
1740 if (apply_verbosely)
1741 error("invalid start of line: '%c'", first);
1742 return -1;
1744 if (added_blank_line)
1745 new_blank_lines_at_end++;
1746 else
1747 new_blank_lines_at_end = 0;
1748 patch += len;
1749 size -= len;
1752 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1753 newsize > 0 && new[newsize - 1] == '\n') {
1754 oldsize--;
1755 newsize--;
1758 oldlines = old;
1759 newlines = new;
1760 leading = frag->leading;
1761 trailing = frag->trailing;
1764 * If we don't have any leading/trailing data in the patch,
1765 * we want it to match at the beginning/end of the file.
1767 * But that would break if the patch is generated with
1768 * --unified=0; sane people wouldn't do that to cause us
1769 * trouble, but we try to please not so sane ones as well.
1771 if (unidiff_zero) {
1772 match_beginning = (!leading && !frag->oldpos);
1773 match_end = 0;
1775 else {
1776 match_beginning = !leading && (frag->oldpos == 1);
1777 match_end = !trailing;
1780 lines = 0;
1781 pos = frag->newpos;
1782 for (;;) {
1783 offset = find_offset(buf, desc->size,
1784 oldlines, oldsize, pos, &lines);
1785 if (match_end && offset + oldsize != desc->size)
1786 offset = -1;
1787 if (match_beginning && offset)
1788 offset = -1;
1789 if (offset >= 0) {
1790 int diff;
1791 unsigned long size, alloc;
1793 if (new_whitespace == strip_whitespace &&
1794 (desc->size - oldsize - offset == 0)) /* end of file? */
1795 newsize -= new_blank_lines_at_end;
1797 diff = newsize - oldsize;
1798 size = desc->size + diff;
1799 alloc = desc->alloc;
1801 /* Warn if it was necessary to reduce the number
1802 * of context lines.
1804 if ((leading != frag->leading) ||
1805 (trailing != frag->trailing))
1806 fprintf(stderr, "Context reduced to (%ld/%ld)"
1807 " to apply fragment at %d\n",
1808 leading, trailing, pos + lines);
1810 if (size > alloc) {
1811 alloc = size + 8192;
1812 desc->alloc = alloc;
1813 buf = xrealloc(buf, alloc);
1814 desc->buffer = buf;
1816 desc->size = size;
1817 memmove(buf + offset + newsize,
1818 buf + offset + oldsize,
1819 size - offset - newsize);
1820 memcpy(buf + offset, newlines, newsize);
1821 offset = 0;
1823 break;
1826 /* Am I at my context limits? */
1827 if ((leading <= p_context) && (trailing <= p_context))
1828 break;
1829 if (match_beginning || match_end) {
1830 match_beginning = match_end = 0;
1831 continue;
1833 /* Reduce the number of context lines
1834 * Reduce both leading and trailing if they are equal
1835 * otherwise just reduce the larger context.
1837 if (leading >= trailing) {
1838 remove_first_line(&oldlines, &oldsize);
1839 remove_first_line(&newlines, &newsize);
1840 pos--;
1841 leading--;
1843 if (trailing > leading) {
1844 remove_last_line(&oldlines, &oldsize);
1845 remove_last_line(&newlines, &newsize);
1846 trailing--;
1850 if (offset && apply_verbosely)
1851 error("while searching for:\n%.*s", oldsize, oldlines);
1853 free(old);
1854 free(new);
1855 return offset;
1858 static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1860 unsigned long dst_size;
1861 struct fragment *fragment = patch->fragments;
1862 void *data;
1863 void *result;
1865 /* Binary patch is irreversible without the optional second hunk */
1866 if (apply_in_reverse) {
1867 if (!fragment->next)
1868 return error("cannot reverse-apply a binary patch "
1869 "without the reverse hunk to '%s'",
1870 patch->new_name
1871 ? patch->new_name : patch->old_name);
1872 fragment = fragment->next;
1874 data = (void*) fragment->patch;
1875 switch (fragment->binary_patch_method) {
1876 case BINARY_DELTA_DEFLATED:
1877 result = patch_delta(desc->buffer, desc->size,
1878 data,
1879 fragment->size,
1880 &dst_size);
1881 free(desc->buffer);
1882 desc->buffer = result;
1883 break;
1884 case BINARY_LITERAL_DEFLATED:
1885 free(desc->buffer);
1886 desc->buffer = data;
1887 dst_size = fragment->size;
1888 break;
1890 if (!desc->buffer)
1891 return -1;
1892 desc->size = desc->alloc = dst_size;
1893 return 0;
1896 static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1898 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1899 unsigned char sha1[20];
1901 /* For safety, we require patch index line to contain
1902 * full 40-byte textual SHA1 for old and new, at least for now.
1904 if (strlen(patch->old_sha1_prefix) != 40 ||
1905 strlen(patch->new_sha1_prefix) != 40 ||
1906 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1907 get_sha1_hex(patch->new_sha1_prefix, sha1))
1908 return error("cannot apply binary patch to '%s' "
1909 "without full index line", name);
1911 if (patch->old_name) {
1912 /* See if the old one matches what the patch
1913 * applies to.
1915 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1916 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1917 return error("the patch applies to '%s' (%s), "
1918 "which does not match the "
1919 "current contents.",
1920 name, sha1_to_hex(sha1));
1922 else {
1923 /* Otherwise, the old one must be empty. */
1924 if (desc->size)
1925 return error("the patch applies to an empty "
1926 "'%s' but it is not empty", name);
1929 get_sha1_hex(patch->new_sha1_prefix, sha1);
1930 if (is_null_sha1(sha1)) {
1931 free(desc->buffer);
1932 desc->alloc = desc->size = 0;
1933 desc->buffer = NULL;
1934 return 0; /* deletion patch */
1937 if (has_sha1_file(sha1)) {
1938 /* We already have the postimage */
1939 enum object_type type;
1940 unsigned long size;
1942 free(desc->buffer);
1943 desc->buffer = read_sha1_file(sha1, &type, &size);
1944 if (!desc->buffer)
1945 return error("the necessary postimage %s for "
1946 "'%s' cannot be read",
1947 patch->new_sha1_prefix, name);
1948 desc->alloc = desc->size = size;
1950 else {
1951 /* We have verified desc matches the preimage;
1952 * apply the patch data to it, which is stored
1953 * in the patch->fragments->{patch,size}.
1955 if (apply_binary_fragment(desc, patch))
1956 return error("binary patch does not apply to '%s'",
1957 name);
1959 /* verify that the result matches */
1960 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1961 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1962 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)", name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1965 return 0;
1968 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1970 struct fragment *frag = patch->fragments;
1971 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1973 if (patch->is_binary)
1974 return apply_binary(desc, patch);
1976 while (frag) {
1977 if (apply_one_fragment(desc, frag, patch->inaccurate_eof)) {
1978 error("patch failed: %s:%ld", name, frag->oldpos);
1979 if (!apply_with_reject)
1980 return -1;
1981 frag->rejected = 1;
1983 frag = frag->next;
1985 return 0;
1988 static int read_file_or_gitlink(struct cache_entry *ce, char **buf_p,
1989 unsigned long *size_p)
1991 if (!ce)
1992 return 0;
1994 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1995 *buf_p = xmalloc(100);
1996 *size_p = snprintf(*buf_p, 100,
1997 "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1998 } else {
1999 enum object_type type;
2000 *buf_p = read_sha1_file(ce->sha1, &type, size_p);
2001 if (!*buf_p)
2002 return -1;
2004 return 0;
2007 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2009 char *buf;
2010 unsigned long size, alloc;
2011 struct buffer_desc desc;
2013 size = 0;
2014 alloc = 0;
2015 buf = NULL;
2016 if (cached) {
2017 if (read_file_or_gitlink(ce, &buf, &size))
2018 return error("read of %s failed", patch->old_name);
2019 alloc = size;
2020 } else if (patch->old_name) {
2021 if (S_ISGITLINK(patch->old_mode)) {
2022 if (ce)
2023 read_file_or_gitlink(ce, &buf, &size);
2024 else {
2026 * There is no way to apply subproject
2027 * patch without looking at the index.
2029 patch->fragments = NULL;
2030 size = 0;
2033 else {
2034 size = xsize_t(st->st_size);
2035 alloc = size + 8192;
2036 buf = xmalloc(alloc);
2037 if (read_old_data(st, patch->old_name,
2038 &buf, &alloc, &size))
2039 return error("read of %s failed",
2040 patch->old_name);
2044 desc.size = size;
2045 desc.alloc = alloc;
2046 desc.buffer = buf;
2048 if (apply_fragments(&desc, patch) < 0)
2049 return -1; /* note with --reject this succeeds. */
2051 /* NUL terminate the result */
2052 if (desc.alloc <= desc.size)
2053 desc.buffer = xrealloc(desc.buffer, desc.size + 1);
2054 desc.buffer[desc.size] = 0;
2056 patch->result = desc.buffer;
2057 patch->resultsize = desc.size;
2059 if (0 < patch->is_delete && patch->resultsize)
2060 return error("removal patch leaves file contents");
2062 return 0;
2065 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2067 struct stat nst;
2068 if (!lstat(new_name, &nst)) {
2069 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2070 return 0;
2072 * A leading component of new_name might be a symlink
2073 * that is going to be removed with this patch, but
2074 * still pointing at somewhere that has the path.
2075 * In such a case, path "new_name" does not exist as
2076 * far as git is concerned.
2078 if (has_symlink_leading_path(new_name, NULL))
2079 return 0;
2081 return error("%s: already exists in working directory", new_name);
2083 else if ((errno != ENOENT) && (errno != ENOTDIR))
2084 return error("%s: %s", new_name, strerror(errno));
2085 return 0;
2088 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2090 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2091 if (!S_ISDIR(st->st_mode))
2092 return -1;
2093 return 0;
2095 return ce_match_stat(ce, st, 1);
2098 static int check_patch(struct patch *patch, struct patch *prev_patch)
2100 struct stat st;
2101 const char *old_name = patch->old_name;
2102 const char *new_name = patch->new_name;
2103 const char *name = old_name ? old_name : new_name;
2104 struct cache_entry *ce = NULL;
2105 int ok_if_exists;
2107 patch->rejected = 1; /* we will drop this after we succeed */
2110 * Make sure that we do not have local modifications from the
2111 * index when we are looking at the index. Also make sure
2112 * we have the preimage file to be patched in the work tree,
2113 * unless --cached, which tells git to apply only in the index.
2115 if (old_name) {
2116 int stat_ret = 0;
2117 unsigned st_mode = 0;
2119 if (!cached)
2120 stat_ret = lstat(old_name, &st);
2121 if (check_index) {
2122 int pos = cache_name_pos(old_name, strlen(old_name));
2123 if (pos < 0)
2124 return error("%s: does not exist in index",
2125 old_name);
2126 ce = active_cache[pos];
2127 if (stat_ret < 0) {
2128 struct checkout costate;
2129 if (errno != ENOENT)
2130 return error("%s: %s", old_name,
2131 strerror(errno));
2132 /* checkout */
2133 costate.base_dir = "";
2134 costate.base_dir_len = 0;
2135 costate.force = 0;
2136 costate.quiet = 0;
2137 costate.not_new = 0;
2138 costate.refresh_cache = 1;
2139 if (checkout_entry(ce,
2140 &costate,
2141 NULL) ||
2142 lstat(old_name, &st))
2143 return -1;
2145 if (!cached && verify_index_match(ce, &st))
2146 return error("%s: does not match index",
2147 old_name);
2148 if (cached)
2149 st_mode = ntohl(ce->ce_mode);
2150 } else if (stat_ret < 0)
2151 return error("%s: %s", old_name, strerror(errno));
2153 if (!cached)
2154 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2156 if (patch->is_new < 0)
2157 patch->is_new = 0;
2158 if (!patch->old_mode)
2159 patch->old_mode = st_mode;
2160 if ((st_mode ^ patch->old_mode) & S_IFMT)
2161 return error("%s: wrong type", old_name);
2162 if (st_mode != patch->old_mode)
2163 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2164 old_name, st_mode, patch->old_mode);
2167 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2168 !strcmp(prev_patch->old_name, new_name))
2169 /* A type-change diff is always split into a patch to
2170 * delete old, immediately followed by a patch to
2171 * create new (see diff.c::run_diff()); in such a case
2172 * it is Ok that the entry to be deleted by the
2173 * previous patch is still in the working tree and in
2174 * the index.
2176 ok_if_exists = 1;
2177 else
2178 ok_if_exists = 0;
2180 if (new_name &&
2181 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2182 if (check_index &&
2183 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2184 !ok_if_exists)
2185 return error("%s: already exists in index", new_name);
2186 if (!cached) {
2187 int err = check_to_create_blob(new_name, ok_if_exists);
2188 if (err)
2189 return err;
2191 if (!patch->new_mode) {
2192 if (0 < patch->is_new)
2193 patch->new_mode = S_IFREG | 0644;
2194 else
2195 patch->new_mode = patch->old_mode;
2199 if (new_name && old_name) {
2200 int same = !strcmp(old_name, new_name);
2201 if (!patch->new_mode)
2202 patch->new_mode = patch->old_mode;
2203 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2204 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2205 patch->new_mode, new_name, patch->old_mode,
2206 same ? "" : " of ", same ? "" : old_name);
2209 if (apply_data(patch, &st, ce) < 0)
2210 return error("%s: patch does not apply", name);
2211 patch->rejected = 0;
2212 return 0;
2215 static int check_patch_list(struct patch *patch)
2217 struct patch *prev_patch = NULL;
2218 int err = 0;
2220 for (prev_patch = NULL; patch ; patch = patch->next) {
2221 if (apply_verbosely)
2222 say_patch_name(stderr,
2223 "Checking patch ", patch, "...\n");
2224 err |= check_patch(patch, prev_patch);
2225 prev_patch = patch;
2227 return err;
2230 static void show_index_list(struct patch *list)
2232 struct patch *patch;
2234 /* Once we start supporting the reverse patch, it may be
2235 * worth showing the new sha1 prefix, but until then...
2237 for (patch = list; patch; patch = patch->next) {
2238 const unsigned char *sha1_ptr;
2239 unsigned char sha1[20];
2240 const char *name;
2242 name = patch->old_name ? patch->old_name : patch->new_name;
2243 if (0 < patch->is_new)
2244 sha1_ptr = null_sha1;
2245 else if (get_sha1(patch->old_sha1_prefix, sha1))
2246 die("sha1 information is lacking or useless (%s).",
2247 name);
2248 else
2249 sha1_ptr = sha1;
2251 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2252 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2253 quote_c_style(name, NULL, stdout, 0);
2254 else
2255 fputs(name, stdout);
2256 putchar(line_termination);
2260 static void stat_patch_list(struct patch *patch)
2262 int files, adds, dels;
2264 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2265 files++;
2266 adds += patch->lines_added;
2267 dels += patch->lines_deleted;
2268 show_stats(patch);
2271 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2274 static void numstat_patch_list(struct patch *patch)
2276 for ( ; patch; patch = patch->next) {
2277 const char *name;
2278 name = patch->new_name ? patch->new_name : patch->old_name;
2279 if (patch->is_binary)
2280 printf("-\t-\t");
2281 else
2282 printf("%d\t%d\t",
2283 patch->lines_added, patch->lines_deleted);
2284 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2285 quote_c_style(name, NULL, stdout, 0);
2286 else
2287 fputs(name, stdout);
2288 putchar(line_termination);
2292 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2294 if (mode)
2295 printf(" %s mode %06o %s\n", newdelete, mode, name);
2296 else
2297 printf(" %s %s\n", newdelete, name);
2300 static void show_mode_change(struct patch *p, int show_name)
2302 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2303 if (show_name)
2304 printf(" mode change %06o => %06o %s\n",
2305 p->old_mode, p->new_mode, p->new_name);
2306 else
2307 printf(" mode change %06o => %06o\n",
2308 p->old_mode, p->new_mode);
2312 static void show_rename_copy(struct patch *p)
2314 const char *renamecopy = p->is_rename ? "rename" : "copy";
2315 const char *old, *new;
2317 /* Find common prefix */
2318 old = p->old_name;
2319 new = p->new_name;
2320 while (1) {
2321 const char *slash_old, *slash_new;
2322 slash_old = strchr(old, '/');
2323 slash_new = strchr(new, '/');
2324 if (!slash_old ||
2325 !slash_new ||
2326 slash_old - old != slash_new - new ||
2327 memcmp(old, new, slash_new - new))
2328 break;
2329 old = slash_old + 1;
2330 new = slash_new + 1;
2332 /* p->old_name thru old is the common prefix, and old and new
2333 * through the end of names are renames
2335 if (old != p->old_name)
2336 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2337 (int)(old - p->old_name), p->old_name,
2338 old, new, p->score);
2339 else
2340 printf(" %s %s => %s (%d%%)\n", renamecopy,
2341 p->old_name, p->new_name, p->score);
2342 show_mode_change(p, 0);
2345 static void summary_patch_list(struct patch *patch)
2347 struct patch *p;
2349 for (p = patch; p; p = p->next) {
2350 if (p->is_new)
2351 show_file_mode_name("create", p->new_mode, p->new_name);
2352 else if (p->is_delete)
2353 show_file_mode_name("delete", p->old_mode, p->old_name);
2354 else {
2355 if (p->is_rename || p->is_copy)
2356 show_rename_copy(p);
2357 else {
2358 if (p->score) {
2359 printf(" rewrite %s (%d%%)\n",
2360 p->new_name, p->score);
2361 show_mode_change(p, 0);
2363 else
2364 show_mode_change(p, 1);
2370 static void patch_stats(struct patch *patch)
2372 int lines = patch->lines_added + patch->lines_deleted;
2374 if (lines > max_change)
2375 max_change = lines;
2376 if (patch->old_name) {
2377 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2378 if (!len)
2379 len = strlen(patch->old_name);
2380 if (len > max_len)
2381 max_len = len;
2383 if (patch->new_name) {
2384 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2385 if (!len)
2386 len = strlen(patch->new_name);
2387 if (len > max_len)
2388 max_len = len;
2392 static void remove_file(struct patch *patch, int rmdir_empty)
2394 if (update_index) {
2395 if (remove_file_from_cache(patch->old_name) < 0)
2396 die("unable to remove %s from index", patch->old_name);
2397 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2399 if (!cached) {
2400 if (S_ISGITLINK(patch->old_mode)) {
2401 if (rmdir(patch->old_name))
2402 warning("unable to remove submodule %s",
2403 patch->old_name);
2404 } else if (!unlink(patch->old_name) && rmdir_empty) {
2405 char *name = xstrdup(patch->old_name);
2406 char *end = strrchr(name, '/');
2407 while (end) {
2408 *end = 0;
2409 if (rmdir(name))
2410 break;
2411 end = strrchr(name, '/');
2413 free(name);
2418 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2420 struct stat st;
2421 struct cache_entry *ce;
2422 int namelen = strlen(path);
2423 unsigned ce_size = cache_entry_size(namelen);
2425 if (!update_index)
2426 return;
2428 ce = xcalloc(1, ce_size);
2429 memcpy(ce->name, path, namelen);
2430 ce->ce_mode = create_ce_mode(mode);
2431 ce->ce_flags = htons(namelen);
2432 if (S_ISGITLINK(mode)) {
2433 const char *s = buf;
2435 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2436 die("corrupt patch for subproject %s", path);
2437 } else {
2438 if (!cached) {
2439 if (lstat(path, &st) < 0)
2440 die("unable to stat newly created file %s",
2441 path);
2442 fill_stat_cache_info(ce, &st);
2444 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2445 die("unable to create backing store for newly created file %s", path);
2447 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2448 die("unable to add cache entry for %s", path);
2451 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2453 int fd;
2454 char *nbuf;
2456 if (S_ISGITLINK(mode)) {
2457 struct stat st;
2458 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2459 return 0;
2460 return mkdir(path, 0777);
2463 if (has_symlinks && S_ISLNK(mode))
2464 /* Although buf:size is counted string, it also is NUL
2465 * terminated.
2467 return symlink(buf, path);
2469 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2470 if (fd < 0)
2471 return -1;
2473 nbuf = convert_to_working_tree(path, buf, &size);
2474 if (nbuf)
2475 buf = nbuf;
2477 while (size) {
2478 int written = xwrite(fd, buf, size);
2479 if (written < 0)
2480 die("writing file %s: %s", path, strerror(errno));
2481 if (!written)
2482 die("out of space writing file %s", path);
2483 buf += written;
2484 size -= written;
2486 if (close(fd) < 0)
2487 die("closing file %s: %s", path, strerror(errno));
2488 if (nbuf)
2489 free(nbuf);
2490 return 0;
2494 * We optimistically assume that the directories exist,
2495 * which is true 99% of the time anyway. If they don't,
2496 * we create them and try again.
2498 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2500 if (cached)
2501 return;
2502 if (!try_create_file(path, mode, buf, size))
2503 return;
2505 if (errno == ENOENT) {
2506 if (safe_create_leading_directories(path))
2507 return;
2508 if (!try_create_file(path, mode, buf, size))
2509 return;
2512 if (errno == EEXIST || errno == EACCES) {
2513 /* We may be trying to create a file where a directory
2514 * used to be.
2516 struct stat st;
2517 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2518 errno = EEXIST;
2521 if (errno == EEXIST) {
2522 unsigned int nr = getpid();
2524 for (;;) {
2525 const char *newpath;
2526 newpath = mkpath("%s~%u", path, nr);
2527 if (!try_create_file(newpath, mode, buf, size)) {
2528 if (!rename(newpath, path))
2529 return;
2530 unlink(newpath);
2531 break;
2533 if (errno != EEXIST)
2534 break;
2535 ++nr;
2538 die("unable to write file %s mode %o", path, mode);
2541 static void create_file(struct patch *patch)
2543 char *path = patch->new_name;
2544 unsigned mode = patch->new_mode;
2545 unsigned long size = patch->resultsize;
2546 char *buf = patch->result;
2548 if (!mode)
2549 mode = S_IFREG | 0644;
2550 create_one_file(path, mode, buf, size);
2551 add_index_file(path, mode, buf, size);
2552 cache_tree_invalidate_path(active_cache_tree, path);
2555 /* phase zero is to remove, phase one is to create */
2556 static void write_out_one_result(struct patch *patch, int phase)
2558 if (patch->is_delete > 0) {
2559 if (phase == 0)
2560 remove_file(patch, 1);
2561 return;
2563 if (patch->is_new > 0 || patch->is_copy) {
2564 if (phase == 1)
2565 create_file(patch);
2566 return;
2569 * Rename or modification boils down to the same
2570 * thing: remove the old, write the new
2572 if (phase == 0)
2573 remove_file(patch, patch->is_rename);
2574 if (phase == 1)
2575 create_file(patch);
2578 static int write_out_one_reject(struct patch *patch)
2580 FILE *rej;
2581 char namebuf[PATH_MAX];
2582 struct fragment *frag;
2583 int cnt = 0;
2585 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2586 if (!frag->rejected)
2587 continue;
2588 cnt++;
2591 if (!cnt) {
2592 if (apply_verbosely)
2593 say_patch_name(stderr,
2594 "Applied patch ", patch, " cleanly.\n");
2595 return 0;
2598 /* This should not happen, because a removal patch that leaves
2599 * contents are marked "rejected" at the patch level.
2601 if (!patch->new_name)
2602 die("internal error");
2604 /* Say this even without --verbose */
2605 say_patch_name(stderr, "Applying patch ", patch, " with");
2606 fprintf(stderr, " %d rejects...\n", cnt);
2608 cnt = strlen(patch->new_name);
2609 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2610 cnt = ARRAY_SIZE(namebuf) - 5;
2611 fprintf(stderr,
2612 "warning: truncating .rej filename to %.*s.rej",
2613 cnt - 1, patch->new_name);
2615 memcpy(namebuf, patch->new_name, cnt);
2616 memcpy(namebuf + cnt, ".rej", 5);
2618 rej = fopen(namebuf, "w");
2619 if (!rej)
2620 return error("cannot open %s: %s", namebuf, strerror(errno));
2622 /* Normal git tools never deal with .rej, so do not pretend
2623 * this is a git patch by saying --git nor give extended
2624 * headers. While at it, maybe please "kompare" that wants
2625 * the trailing TAB and some garbage at the end of line ;-).
2627 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2628 patch->new_name, patch->new_name);
2629 for (cnt = 1, frag = patch->fragments;
2630 frag;
2631 cnt++, frag = frag->next) {
2632 if (!frag->rejected) {
2633 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2634 continue;
2636 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2637 fprintf(rej, "%.*s", frag->size, frag->patch);
2638 if (frag->patch[frag->size-1] != '\n')
2639 fputc('\n', rej);
2641 fclose(rej);
2642 return -1;
2645 static int write_out_results(struct patch *list, int skipped_patch)
2647 int phase;
2648 int errs = 0;
2649 struct patch *l;
2651 if (!list && !skipped_patch)
2652 return error("No changes");
2654 for (phase = 0; phase < 2; phase++) {
2655 l = list;
2656 while (l) {
2657 if (l->rejected)
2658 errs = 1;
2659 else {
2660 write_out_one_result(l, phase);
2661 if (phase == 1 && write_out_one_reject(l))
2662 errs = 1;
2664 l = l->next;
2667 return errs;
2670 static struct lock_file lock_file;
2672 static struct excludes {
2673 struct excludes *next;
2674 const char *path;
2675 } *excludes;
2677 static int use_patch(struct patch *p)
2679 const char *pathname = p->new_name ? p->new_name : p->old_name;
2680 struct excludes *x = excludes;
2681 while (x) {
2682 if (fnmatch(x->path, pathname, 0) == 0)
2683 return 0;
2684 x = x->next;
2686 if (0 < prefix_length) {
2687 int pathlen = strlen(pathname);
2688 if (pathlen <= prefix_length ||
2689 memcmp(prefix, pathname, prefix_length))
2690 return 0;
2692 return 1;
2695 static void prefix_one(char **name)
2697 char *old_name = *name;
2698 if (!old_name)
2699 return;
2700 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2701 free(old_name);
2704 static void prefix_patches(struct patch *p)
2706 if (!prefix || p->is_toplevel_relative)
2707 return;
2708 for ( ; p; p = p->next) {
2709 if (p->new_name == p->old_name) {
2710 char *prefixed = p->new_name;
2711 prefix_one(&prefixed);
2712 p->new_name = p->old_name = prefixed;
2714 else {
2715 prefix_one(&p->new_name);
2716 prefix_one(&p->old_name);
2721 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2723 unsigned long offset, size;
2724 char *buffer = read_patch_file(fd, &size);
2725 struct patch *list = NULL, **listp = &list;
2726 int skipped_patch = 0;
2728 patch_input_file = filename;
2729 if (!buffer)
2730 return -1;
2731 offset = 0;
2732 while (size > 0) {
2733 struct patch *patch;
2734 int nr;
2736 patch = xcalloc(1, sizeof(*patch));
2737 patch->inaccurate_eof = inaccurate_eof;
2738 nr = parse_chunk(buffer + offset, size, patch);
2739 if (nr < 0)
2740 break;
2741 if (apply_in_reverse)
2742 reverse_patches(patch);
2743 if (prefix)
2744 prefix_patches(patch);
2745 if (use_patch(patch)) {
2746 patch_stats(patch);
2747 *listp = patch;
2748 listp = &patch->next;
2750 else {
2751 /* perhaps free it a bit better? */
2752 free(patch);
2753 skipped_patch++;
2755 offset += nr;
2756 size -= nr;
2759 if (whitespace_error && (new_whitespace == error_on_whitespace))
2760 apply = 0;
2762 update_index = check_index && apply;
2763 if (update_index && newfd < 0)
2764 newfd = hold_locked_index(&lock_file, 1);
2766 if (check_index) {
2767 if (read_cache() < 0)
2768 die("unable to read index file");
2771 if ((check || apply) &&
2772 check_patch_list(list) < 0 &&
2773 !apply_with_reject)
2774 exit(1);
2776 if (apply && write_out_results(list, skipped_patch))
2777 exit(1);
2779 if (show_index_info)
2780 show_index_list(list);
2782 if (diffstat)
2783 stat_patch_list(list);
2785 if (numstat)
2786 numstat_patch_list(list);
2788 if (summary)
2789 summary_patch_list(list);
2791 free(buffer);
2792 return 0;
2795 static int git_apply_config(const char *var, const char *value)
2797 if (!strcmp(var, "apply.whitespace")) {
2798 apply_default_whitespace = xstrdup(value);
2799 return 0;
2801 return git_default_config(var, value);
2805 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2807 int i;
2808 int read_stdin = 1;
2809 int inaccurate_eof = 0;
2810 int errs = 0;
2811 int is_not_gitdir = 0;
2813 const char *whitespace_option = NULL;
2815 prefix = setup_git_directory_gently(&is_not_gitdir);
2816 prefix_length = prefix ? strlen(prefix) : 0;
2817 git_config(git_apply_config);
2818 if (apply_default_whitespace)
2819 parse_whitespace_option(apply_default_whitespace);
2821 for (i = 1; i < argc; i++) {
2822 const char *arg = argv[i];
2823 char *end;
2824 int fd;
2826 if (!strcmp(arg, "-")) {
2827 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2828 read_stdin = 0;
2829 continue;
2831 if (!prefixcmp(arg, "--exclude=")) {
2832 struct excludes *x = xmalloc(sizeof(*x));
2833 x->path = arg + 10;
2834 x->next = excludes;
2835 excludes = x;
2836 continue;
2838 if (!prefixcmp(arg, "-p")) {
2839 p_value = atoi(arg + 2);
2840 p_value_known = 1;
2841 continue;
2843 if (!strcmp(arg, "--no-add")) {
2844 no_add = 1;
2845 continue;
2847 if (!strcmp(arg, "--stat")) {
2848 apply = 0;
2849 diffstat = 1;
2850 continue;
2852 if (!strcmp(arg, "--allow-binary-replacement") ||
2853 !strcmp(arg, "--binary")) {
2854 continue; /* now no-op */
2856 if (!strcmp(arg, "--numstat")) {
2857 apply = 0;
2858 numstat = 1;
2859 continue;
2861 if (!strcmp(arg, "--summary")) {
2862 apply = 0;
2863 summary = 1;
2864 continue;
2866 if (!strcmp(arg, "--check")) {
2867 apply = 0;
2868 check = 1;
2869 continue;
2871 if (!strcmp(arg, "--index")) {
2872 if (is_not_gitdir)
2873 die("--index outside a repository");
2874 check_index = 1;
2875 continue;
2877 if (!strcmp(arg, "--cached")) {
2878 if (is_not_gitdir)
2879 die("--cached outside a repository");
2880 check_index = 1;
2881 cached = 1;
2882 continue;
2884 if (!strcmp(arg, "--apply")) {
2885 apply = 1;
2886 continue;
2888 if (!strcmp(arg, "--index-info")) {
2889 apply = 0;
2890 show_index_info = 1;
2891 continue;
2893 if (!strcmp(arg, "-z")) {
2894 line_termination = 0;
2895 continue;
2897 if (!prefixcmp(arg, "-C")) {
2898 p_context = strtoul(arg + 2, &end, 0);
2899 if (*end != '\0')
2900 die("unrecognized context count '%s'", arg + 2);
2901 continue;
2903 if (!prefixcmp(arg, "--whitespace=")) {
2904 whitespace_option = arg + 13;
2905 parse_whitespace_option(arg + 13);
2906 continue;
2908 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2909 apply_in_reverse = 1;
2910 continue;
2912 if (!strcmp(arg, "--unidiff-zero")) {
2913 unidiff_zero = 1;
2914 continue;
2916 if (!strcmp(arg, "--reject")) {
2917 apply = apply_with_reject = apply_verbosely = 1;
2918 continue;
2920 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2921 apply_verbosely = 1;
2922 continue;
2924 if (!strcmp(arg, "--inaccurate-eof")) {
2925 inaccurate_eof = 1;
2926 continue;
2928 if (0 < prefix_length)
2929 arg = prefix_filename(prefix, prefix_length, arg);
2931 fd = open(arg, O_RDONLY);
2932 if (fd < 0)
2933 usage(apply_usage);
2934 read_stdin = 0;
2935 set_default_whitespace_mode(whitespace_option);
2936 errs |= apply_patch(fd, arg, inaccurate_eof);
2937 close(fd);
2939 set_default_whitespace_mode(whitespace_option);
2940 if (read_stdin)
2941 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2942 if (whitespace_error) {
2943 if (squelch_whitespace_errors &&
2944 squelch_whitespace_errors < whitespace_error) {
2945 int squelched =
2946 whitespace_error - squelch_whitespace_errors;
2947 fprintf(stderr, "warning: squelched %d "
2948 "whitespace error%s\n",
2949 squelched,
2950 squelched == 1 ? "" : "s");
2952 if (new_whitespace == error_on_whitespace)
2953 die("%d line%s add%s whitespace errors.",
2954 whitespace_error,
2955 whitespace_error == 1 ? "" : "s",
2956 whitespace_error == 1 ? "s" : "");
2957 if (applied_after_fixing_ws)
2958 fprintf(stderr, "warning: %d line%s applied after"
2959 " fixing whitespace errors.\n",
2960 applied_after_fixing_ws,
2961 applied_after_fixing_ws == 1 ? "" : "s");
2962 else if (whitespace_error)
2963 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2964 whitespace_error,
2965 whitespace_error == 1 ? "" : "s",
2966 whitespace_error == 1 ? "s" : "");
2969 if (update_index) {
2970 if (write_cache(newfd, active_cache, active_nr) ||
2971 close(newfd) || commit_locked_index(&lock_file))
2972 die("Unable to write new index file");
2975 return !!errs;