4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 * NOTE! It does all its work in the index file, and only cares about
9 * the files in the working directory if you tell it to "merge" the
12 * Even when merging it always takes the source from the index, and
13 * uses the working tree as a "branch" for a 3-way merge.
19 // We default to the merge behaviour, since that's what most people would
22 // --check turns on checking that the working tree matches the
23 // files that are being modified, but doesn't apply the patch
24 // --stat does just a diffstat, and doesn't actually apply
25 // --show-files shows the directory changes
27 static int merge_patch = 1;
28 static int check_index = 0;
29 static int diffstat = 0;
32 static int show_files = 0;
33 static const char apply_usage[] = "git-apply [--stat] [--check] [--show-files] <patch>";
36 * For "diff-stat" like behaviour, we keep track of the biggest change
37 * we've seen, and the longest filename. That allows us to do simple
40 static int max_change, max_len;
43 * Various "current state", notably line numbers and what
44 * file (and how) we're patching right now.. The "is_xxxx"
45 * things are flags, where -1 means "don't know yet".
47 static int linenr = 1;
50 unsigned long oldpos, oldlines;
51 unsigned long newpos, newlines;
54 struct fragment *next;
58 char *new_name, *old_name, *def_name;
59 unsigned int old_mode, new_mode;
60 int is_rename, is_copy, is_new, is_delete;
61 int lines_added, lines_deleted;
62 struct fragment *fragments;
64 unsigned long resultsize;
68 #define CHUNKSIZE (8192)
71 static void *read_patch_file(int fd, unsigned long *sizep)
73 unsigned long size = 0, alloc = CHUNKSIZE;
74 void *buffer = xmalloc(alloc);
77 int nr = alloc - size;
80 buffer = xrealloc(buffer, alloc);
83 nr = read(fd, buffer + size, nr);
89 die("git-apply: read returned %s", strerror(errno));
96 * Make sure that we have some slop in the buffer
97 * so that we can do speculative "memcmp" etc, and
98 * see to it that it is NUL-filled.
100 if (alloc < size + SLOP)
101 buffer = xrealloc(buffer, size + SLOP);
102 memset(buffer + size, 0, SLOP);
106 static unsigned long linelen(const char *buffer, unsigned long size)
108 unsigned long len = 0;
111 if (*buffer++ == '\n')
117 static int is_dev_null(const char *str)
119 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
125 static int name_terminate(const char *name, int namelen, int c, int terminate)
127 if (c == ' ' && !(terminate & TERM_SPACE))
129 if (c == '\t' && !(terminate & TERM_TAB))
135 static char * find_name(const char *line, char *def, int p_value, int terminate)
138 const char *start = line;
147 if (name_terminate(start, line-start, c, terminate))
151 if (c == '/' && !--p_value)
161 * Generally we prefer the shorter name, especially
162 * if the other one is just a variation of that with
163 * something else tacked on to the end (ie "file.orig"
167 int deflen = strlen(def);
168 if (deflen < len && !strncmp(start, def, deflen))
172 name = xmalloc(len + 1);
173 memcpy(name, start, len);
180 * Get the name etc info from the --/+++ lines of a traditional patch header
182 * NOTE! This hardcodes "-p1" behaviour in filename detection.
184 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
185 * files, we can happily check the index for a match, but for creating a
186 * new file we should try to match whatever "patch" does. I have no idea.
188 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
193 first += 4; // skip "--- "
194 second += 4; // skip "+++ "
195 if (is_dev_null(first)) {
197 patch->is_delete = 0;
198 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
199 patch->new_name = name;
200 } else if (is_dev_null(second)) {
202 patch->is_delete = 1;
203 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
204 patch->old_name = name;
206 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
207 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
208 patch->old_name = patch->new_name = name;
211 die("unable to find filename in patch at line %d", linenr);
214 static int gitdiff_hdrend(const char *line, struct patch *patch)
220 * We're anal about diff header consistency, to make
221 * sure that we don't end up having strange ambiguous
222 * patches floating around.
224 * As a result, gitdiff_{old|new}name() will check
225 * their names against any previous information, just
228 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
233 if (!orig_name && !isnull)
234 return find_name(line, NULL, 1, 0);
242 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
255 if (memcmp(line, name, len) || line[len] != '\n')
259 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
263 static int gitdiff_oldname(const char *line, struct patch *patch)
265 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
269 static int gitdiff_newname(const char *line, struct patch *patch)
271 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
275 static int gitdiff_oldmode(const char *line, struct patch *patch)
277 patch->old_mode = strtoul(line, NULL, 8);
281 static int gitdiff_newmode(const char *line, struct patch *patch)
283 patch->new_mode = strtoul(line, NULL, 8);
287 static int gitdiff_delete(const char *line, struct patch *patch)
289 patch->is_delete = 1;
290 patch->old_name = patch->def_name;
291 return gitdiff_oldmode(line, patch);
294 static int gitdiff_newfile(const char *line, struct patch *patch)
297 patch->new_name = patch->def_name;
298 return gitdiff_newmode(line, patch);
301 static int gitdiff_copysrc(const char *line, struct patch *patch)
304 patch->old_name = find_name(line, NULL, 0, 0);
308 static int gitdiff_copydst(const char *line, struct patch *patch)
311 patch->new_name = find_name(line, NULL, 0, 0);
315 static int gitdiff_renamesrc(const char *line, struct patch *patch)
317 patch->is_rename = 1;
318 patch->old_name = find_name(line, NULL, 0, 0);
322 static int gitdiff_renamedst(const char *line, struct patch *patch)
324 patch->is_rename = 1;
325 patch->new_name = find_name(line, NULL, 0, 0);
329 static int gitdiff_similarity(const char *line, struct patch *patch)
334 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
340 * This is normal for a diff that doesn't change anything: we'll fall through
341 * into the next diff. Tell the parser to break out.
343 static int gitdiff_unrecognized(const char *line, struct patch *patch)
348 static char *git_header_name(char *line)
366 * We don't accept absolute paths (/dev/null) as possibly valid
372 * Accept a name only if it shows up twice, exactly the same
375 for (len = 0 ; ; len++) {
392 if (second[len] == '\n' && !memcmp(name, second, len)) {
393 char *ret = xmalloc(len + 1);
394 memcpy(ret, name, len);
403 /* Verify that we recognize the lines following a git header */
404 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
406 unsigned long offset;
408 /* A git diff has explicit new/delete information, so we don't guess */
410 patch->is_delete = 0;
413 * Some things may not have the old name in the
414 * rest of the headers anywhere (pure mode changes,
415 * or removing or adding empty files), so we get
416 * the default name from the header.
418 patch->def_name = git_header_name(line + strlen("diff --git "));
423 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
424 static const struct opentry {
426 int (*fn)(const char *, struct patch *);
428 { "@@ -", gitdiff_hdrend },
429 { "--- ", gitdiff_oldname },
430 { "+++ ", gitdiff_newname },
431 { "old mode ", gitdiff_oldmode },
432 { "new mode ", gitdiff_newmode },
433 { "deleted file mode ", gitdiff_delete },
434 { "new file mode ", gitdiff_newfile },
435 { "copy from ", gitdiff_copysrc },
436 { "copy to ", gitdiff_copydst },
437 { "rename from ", gitdiff_renamesrc },
438 { "rename to ", gitdiff_renamedst },
439 { "similarity index ", gitdiff_similarity },
440 { "dissimilarity index ", gitdiff_dissimilarity },
441 { "", gitdiff_unrecognized },
445 len = linelen(line, size);
446 if (!len || line[len-1] != '\n')
448 for (i = 0; i < sizeof(optable) / sizeof(optable[0]); i++) {
449 const struct opentry *p = optable + i;
450 int oplen = strlen(p->str);
451 if (len < oplen || memcmp(p->str, line, oplen))
453 if (p->fn(line + oplen, patch) < 0)
462 static int parse_num(const char *line, unsigned long *p)
468 *p = strtoul(line, &ptr, 10);
472 static int parse_range(const char *line, int len, int offset, const char *expect,
473 unsigned long *p1, unsigned long *p2)
477 if (offset < 0 || offset >= len)
482 digits = parse_num(line, p1);
492 digits = parse_num(line+1, p2);
504 if (memcmp(line, expect, ex))
511 * Parse a unified diff fragment header of the
512 * form "@@ -a,b +c,d @@"
514 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
518 if (!len || line[len-1] != '\n')
521 /* Figure out the number of lines in a fragment */
522 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
523 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
528 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
530 unsigned long offset, len;
532 patch->is_rename = patch->is_copy = 0;
533 patch->is_new = patch->is_delete = -1;
534 patch->old_mode = patch->new_mode = 0;
535 patch->old_name = patch->new_name = NULL;
536 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
537 unsigned long nextlen;
539 len = linelen(line, size);
543 /* Testing this early allows us to take a few shortcuts.. */
548 * Make sure we don't find any unconnected patch fragmants.
549 * That's a sign that we didn't find a header, and that a
550 * patch has become corrupted/broken up.
552 if (!memcmp("@@ -", line, 4)) {
553 struct fragment dummy;
554 if (parse_fragment_header(line, len, &dummy) < 0)
556 error("patch fragment without header at line %d: %.*s", linenr, len-1, line);
563 * Git patch? It might not have a real patch, just a rename
564 * or mode change, so we handle that specially
566 if (!memcmp("diff --git ", line, 11)) {
567 int git_hdr_len = parse_git_header(line, len, size, patch);
570 if (!patch->old_name && !patch->new_name)
571 die("git diff header lacks filename information");
572 *hdrsize = git_hdr_len;
576 /** --- followed by +++ ? */
577 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
581 * We only accept unified patches, so we want it to
582 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
585 nextlen = linelen(line + len, size - len);
586 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
589 /* Ok, we'll consider it a patch */
590 parse_traditional_patch(line, line+len, patch);
591 *hdrsize = len + nextlen;
599 * Parse a unified diff. Note that this really needs
600 * to parse each fragment separately, since the only
601 * way to know the difference between a "---" that is
602 * part of a patch, and a "---" that starts the next
603 * patch is to look at the line counts..
605 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
608 int len = linelen(line, size), offset;
609 unsigned long oldlines, newlines;
611 offset = parse_fragment_header(line, len, fragment);
614 oldlines = fragment->oldlines;
615 newlines = fragment->newlines;
617 if (patch->is_new < 0) {
618 patch->is_new = !oldlines;
620 patch->old_name = NULL;
622 if (patch->is_delete < 0) {
623 patch->is_delete = !newlines;
625 patch->new_name = NULL;
628 if (patch->is_new != !oldlines)
629 return error("new file depends on old contents");
630 if (patch->is_delete != !newlines)
631 return error("deleted file still has contents");
633 /* Parse the thing.. */
638 for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
639 if (!oldlines && !newlines)
641 len = linelen(line, size);
642 if (!len || line[len-1] != '\n')
659 /* We allow "\ No newline at end of file" */
661 if (len < 12 || memcmp(line, "\\ No newline", 12))
666 patch->lines_added += added;
667 patch->lines_deleted += deleted;
671 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
673 unsigned long offset = 0;
674 struct fragment **fragp = &patch->fragments;
676 while (size > 4 && !memcmp(line, "@@ -", 4)) {
677 struct fragment *fragment;
680 fragment = xmalloc(sizeof(*fragment));
681 memset(fragment, 0, sizeof(*fragment));
682 len = parse_fragment(line, size, patch, fragment);
684 die("corrupt patch at line %d", linenr);
686 fragment->patch = line;
687 fragment->size = len;
690 fragp = &fragment->next;
699 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
701 int hdrsize, patchsize;
702 int offset = find_header(buffer, size, &hdrsize, patch);
707 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
709 return offset + hdrsize + patchsize;
712 const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
713 const char minuses[]= "----------------------------------------------------------------------";
715 static void show_stats(struct patch *patch)
717 char *name = patch->old_name;
718 int len, max, add, del, total;
721 name = patch->new_name;
724 * "scale" the filename
735 * scale the add/delete
741 add = patch->lines_added;
742 del = patch->lines_deleted;
745 total = (total * max + max_change / 2) / max_change;
746 add = (add * max + max_change / 2) / max_change;
748 printf(" %-*s |%5d %.*s%.*s\n",
749 len, name, patch->lines_added + patch->lines_deleted,
750 add, pluses, del, minuses);
753 static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
758 switch (st->st_mode & S_IFMT) {
760 return readlink(path, buf, size);
762 fd = open(path, O_RDONLY);
764 return error("unable to open %s", path);
767 int ret = read(fd, buf + got, size - got);
785 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line)
788 unsigned long start, backwards, forwards;
795 unsigned long offset = 0;
797 while (offset + fragsize <= size) {
798 if (buf[offset++] == '\n') {
806 /* Exact line number? */
807 if (!memcmp(buf + start, fragment, fragsize))
811 * There's probably some smart way to do this, but I'll leave
812 * that to the smart and beautiful people. I'm simple and stupid.
823 if (forwards + fragsize > size)
829 } while (backwards && buf[backwards-1] != '\n');
832 while (forwards + fragsize <= size) {
833 if (buf[forwards++] == '\n')
839 if (try + fragsize > size)
841 if (memcmp(buf + try, fragment, fragsize))
846 fprintf(stderr, "Fragment applied at offset %d\n", n);
851 * We should start searching forward and backward.
862 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag)
864 char *buf = desc->buffer;
865 const char *patch = frag->patch;
866 int offset, size = frag->size;
867 char *old = xmalloc(size);
868 char *new = xmalloc(size);
869 int oldsize = 0, newsize = 0;
872 int len = linelen(patch, size);
879 * "plen" is how much of the line we should use for
880 * the actual patch data. Normally we just remove the
881 * first character on the line, but if the line is
882 * followed by "\ No newline", then we also remove the
883 * last one (which is the newline, of course).
886 if (len > size && patch[len] == '\\')
891 memcpy(old + oldsize, patch + 1, plen);
895 /* Fall-through for ' ' */
897 memcpy(new + newsize, patch + 1, plen);
901 /* Ignore it, we already handled it */
910 offset = find_offset(buf, desc->size, old, oldsize, frag->newpos);
912 int diff = newsize - oldsize;
913 unsigned long size = desc->size + diff;
914 unsigned long alloc = desc->alloc;
919 buf = xrealloc(buf, alloc);
923 memmove(buf + offset + newsize, buf + offset + oldsize, size - offset - newsize);
924 memcpy(buf + offset, new, newsize);
933 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
935 struct fragment *frag = patch->fragments;
938 if (apply_one_fragment(desc, frag) < 0)
939 return error("patch failed: %s:%d", patch->old_name, frag->oldpos);
945 static int apply_data(struct patch *patch, struct stat *st)
948 unsigned long size, alloc;
949 struct buffer_desc desc;
954 if (patch->old_name) {
957 buf = xmalloc(alloc);
958 if (read_old_data(st, patch->old_name, buf, alloc) != size)
959 return error("read of %s failed", patch->old_name);
965 if (apply_fragments(&desc, patch) < 0)
967 patch->result = desc.buffer;
968 patch->resultsize = desc.size;
972 static int check_patch(struct patch *patch)
975 const char *old_name = patch->old_name;
976 const char *new_name = patch->new_name;
981 if (lstat(old_name, &st) < 0)
982 return error("%s: %s\n", strerror(errno));
984 int pos = cache_name_pos(old_name, strlen(old_name));
986 return error("%s: does not exist in index", old_name);
987 changed = ce_match_stat(active_cache[pos], &st);
989 return error("%s: does not match index", old_name);
991 if (patch->is_new < 0)
993 if (!patch->old_mode)
994 patch->old_mode = st.st_mode;
995 if ((st.st_mode ^ patch->old_mode) & S_IFMT)
996 return error("%s: wrong type", old_name);
997 if (st.st_mode != patch->old_mode)
998 fprintf(stderr, "warning: %s has type %o, expected %o\n",
999 old_name, st.st_mode, patch->old_mode);
1002 if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
1003 if (check_index && cache_name_pos(new_name, strlen(new_name)) >= 0)
1004 return error("%s: already exists in index", new_name);
1005 if (!lstat(new_name, &st))
1006 return error("%s: already exists in working directory", new_name);
1007 if (errno != ENOENT)
1008 return error("%s: %s", new_name, strerror(errno));
1011 if (new_name && old_name) {
1012 int same = !strcmp(old_name, new_name);
1013 if (!patch->new_mode)
1014 patch->new_mode = patch->old_mode;
1015 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
1016 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
1017 patch->new_mode, new_name, patch->old_mode,
1018 same ? "" : " of ", same ? "" : old_name);
1021 if (apply_data(patch, &st) < 0)
1022 return error("%s: patch does not apply", old_name);
1026 static int check_patch_list(struct patch *patch)
1030 for (;patch ; patch = patch->next)
1031 error |= check_patch(patch);
1035 static void show_file(int c, unsigned int mode, const char *name)
1037 printf("%c %o %s\n", c, mode, name);
1040 static void show_file_list(struct patch *patch)
1042 for (;patch ; patch = patch->next) {
1043 if (patch->is_rename) {
1044 show_file('-', patch->old_mode, patch->old_name);
1045 show_file('+', patch->new_mode, patch->new_name);
1048 if (patch->is_copy || patch->is_new) {
1049 show_file('+', patch->new_mode, patch->new_name);
1052 if (patch->is_delete) {
1053 show_file('-', patch->old_mode, patch->old_name);
1056 if (patch->old_mode && patch->new_mode && patch->old_mode != patch->new_mode) {
1057 printf("M %o:%o %s\n", patch->old_mode, patch->new_mode, patch->old_name);
1060 printf("M %o %s\n", patch->old_mode, patch->old_name);
1064 static void stat_patch_list(struct patch *patch)
1066 int files, adds, dels;
1068 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
1070 adds += patch->lines_added;
1071 dels += patch->lines_deleted;
1075 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
1078 static void patch_stats(struct patch *patch)
1080 int lines = patch->lines_added + patch->lines_deleted;
1082 if (lines > max_change)
1084 if (patch->old_name) {
1085 int len = strlen(patch->old_name);
1089 if (patch->new_name) {
1090 int len = strlen(patch->new_name);
1096 static int apply_patch(int fd)
1098 unsigned long offset, size;
1099 char *buffer = read_patch_file(fd, &size);
1100 struct patch *list = NULL, **listp = &list;
1106 struct patch *patch;
1109 patch = xmalloc(sizeof(*patch));
1110 memset(patch, 0, sizeof(*patch));
1111 nr = parse_chunk(buffer + offset, size, patch);
1116 listp = &patch->next;
1121 if ((check || apply) && check_patch_list(list) < 0)
1125 show_file_list(list);
1128 stat_patch_list(list);
1134 int main(int argc, char **argv)
1139 if (read_cache() < 0)
1140 die("unable to read index file");
1142 for (i = 1; i < argc; i++) {
1143 const char *arg = argv[i];
1146 if (!strcmp(arg, "-")) {
1151 if (!strcmp(arg, "--no-merge")) {
1155 if (!strcmp(arg, "--stat")) {
1160 if (!strcmp(arg, "--check")) {
1165 if (!strcmp(arg, "--index")) {
1169 if (!strcmp(arg, "--show-files")) {
1173 fd = open(arg, O_RDONLY);