4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
10 #include "cache-tree.h"
15 #include "string-list.h"
18 #include "parse-options.h"
19 #include "xdiff-interface.h"
24 * --check turns on checking that the working tree matches the
25 * files that are being modified, but doesn't apply the patch
26 * --stat does just a diffstat, and doesn't actually apply
27 * --numstat does numeric diffstat, and doesn't actually apply
28 * --index-info shows the old and new index info for paths if available.
29 * --index updates the cache as well.
30 * --cached updates only the cache without ever touching the working tree.
32 static const char *prefix;
33 static int prefix_length = -1;
34 static int newfd = -1;
36 static int unidiff_zero;
37 static int p_value = 1;
38 static int p_value_known;
39 static int check_index;
40 static int update_index;
47 static int apply_in_reverse;
48 static int apply_with_reject;
49 static int apply_verbosely;
50 static int allow_overlap;
53 static const char *fake_ancestor;
54 static int line_termination = '\n';
55 static unsigned int p_context = UINT_MAX;
56 static const char * const apply_usage[] = {
57 N_("git apply [options] [<patch>...]"),
61 static enum ws_error_action {
66 } ws_error_action = warn_on_ws_error;
67 static int whitespace_error;
68 static int squelch_whitespace_errors = 5;
69 static int applied_after_fixing_ws;
71 static enum ws_ignore {
74 } ws_ignore_action = ignore_ws_none;
77 static const char *patch_input_file;
78 static const char *root;
80 static int read_stdin = 1;
83 static void parse_whitespace_option(const char *option)
86 ws_error_action = warn_on_ws_error;
89 if (!strcmp(option, "warn")) {
90 ws_error_action = warn_on_ws_error;
93 if (!strcmp(option, "nowarn")) {
94 ws_error_action = nowarn_ws_error;
97 if (!strcmp(option, "error")) {
98 ws_error_action = die_on_ws_error;
101 if (!strcmp(option, "error-all")) {
102 ws_error_action = die_on_ws_error;
103 squelch_whitespace_errors = 0;
106 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
107 ws_error_action = correct_ws_error;
110 die(_("unrecognized whitespace option '%s'"), option);
113 static void parse_ignorewhitespace_option(const char *option)
115 if (!option || !strcmp(option, "no") ||
116 !strcmp(option, "false") || !strcmp(option, "never") ||
117 !strcmp(option, "none")) {
118 ws_ignore_action = ignore_ws_none;
121 if (!strcmp(option, "change")) {
122 ws_ignore_action = ignore_ws_change;
125 die(_("unrecognized whitespace ignore option '%s'"), option);
128 static void set_default_whitespace_mode(const char *whitespace_option)
130 if (!whitespace_option && !apply_default_whitespace)
131 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
135 * For "diff-stat" like behaviour, we keep track of the biggest change
136 * we've seen, and the longest filename. That allows us to do simple
139 static int max_change, max_len;
142 * Various "current state", notably line numbers and what
143 * file (and how) we're patching right now.. The "is_xxxx"
144 * things are flags, where -1 means "don't know yet".
146 static int linenr = 1;
149 * This represents one "hunk" from a patch, starting with
150 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
151 * patch text is pointed at by patch, and its byte length
152 * is stored in size. leading and trailing are the number
156 unsigned long leading, trailing;
157 unsigned long oldpos, oldlines;
158 unsigned long newpos, newlines;
160 * 'patch' is usually borrowed from buf in apply_patch(),
161 * but some codepaths store an allocated buffer.
164 unsigned free_patch:1,
168 struct fragment *next;
172 * When dealing with a binary patch, we reuse "leading" field
173 * to store the type of the binary hunk, either deflated "delta"
174 * or deflated "literal".
176 #define binary_patch_method leading
177 #define BINARY_DELTA_DEFLATED 1
178 #define BINARY_LITERAL_DEFLATED 2
181 * This represents a "patch" to a file, both metainfo changes
182 * such as creation/deletion, filemode and content changes represented
183 * as a series of fragments.
186 char *new_name, *old_name, *def_name;
187 unsigned int old_mode, new_mode;
188 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
191 int lines_added, lines_deleted;
193 unsigned int is_toplevel_relative:1;
194 unsigned int inaccurate_eof:1;
195 unsigned int is_binary:1;
196 unsigned int is_copy:1;
197 unsigned int is_rename:1;
198 unsigned int recount:1;
199 unsigned int conflicted_threeway:1;
200 unsigned int direct_to_threeway:1;
201 struct fragment *fragments;
204 char old_sha1_prefix[41];
205 char new_sha1_prefix[41];
208 /* three-way fallback result */
209 unsigned char threeway_stage[3][20];
212 static void free_fragment_list(struct fragment *list)
215 struct fragment *next = list->next;
216 if (list->free_patch)
217 free((char *)list->patch);
223 static void free_patch(struct patch *patch)
225 free_fragment_list(patch->fragments);
226 free(patch->def_name);
227 free(patch->old_name);
228 free(patch->new_name);
233 static void free_patch_list(struct patch *list)
236 struct patch *next = list->next;
243 * A line in a file, len-bytes long (includes the terminating LF,
244 * except for an incomplete line at the end if the file ends with
245 * one), and its contents hashes to 'hash'.
251 #define LINE_COMMON 1
252 #define LINE_PATCHED 2
256 * This represents a "file", which is an array of "lines".
263 struct line *line_allocated;
268 * Records filenames that have been touched, in order to handle
269 * the case where more than one patches touch the same file.
272 static struct string_list fn_table;
274 static uint32_t hash_line(const char *cp, size_t len)
278 for (i = 0, h = 0; i < len; i++) {
279 if (!isspace(cp[i])) {
280 h = h * 3 + (cp[i] & 0xff);
287 * Compare lines s1 of length n1 and s2 of length n2, ignoring
288 * whitespace difference. Returns 1 if they match, 0 otherwise
290 static int fuzzy_matchlines(const char *s1, size_t n1,
291 const char *s2, size_t n2)
293 const char *last1 = s1 + n1 - 1;
294 const char *last2 = s2 + n2 - 1;
297 /* ignore line endings */
298 while ((*last1 == '\r') || (*last1 == '\n'))
300 while ((*last2 == '\r') || (*last2 == '\n'))
303 /* skip leading whitespaces, if both begin with whitespace */
304 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
305 while (isspace(*s1) && (s1 <= last1))
307 while (isspace(*s2) && (s2 <= last2))
310 /* early return if both lines are empty */
311 if ((s1 > last1) && (s2 > last2))
314 result = *s1++ - *s2++;
316 * Skip whitespace inside. We check for whitespace on
317 * both buffers because we don't want "a b" to match
320 if (isspace(*s1) && isspace(*s2)) {
321 while (isspace(*s1) && s1 <= last1)
323 while (isspace(*s2) && s2 <= last2)
327 * If we reached the end on one side only,
331 ((s2 > last2) && (s1 <= last1)) ||
332 ((s1 > last1) && (s2 <= last2)))
334 if ((s1 > last1) && (s2 > last2))
341 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
343 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
344 img->line_allocated[img->nr].len = len;
345 img->line_allocated[img->nr].hash = hash_line(bol, len);
346 img->line_allocated[img->nr].flag = flag;
351 * "buf" has the file contents to be patched (read from various sources).
352 * attach it to "image" and add line-based index to it.
353 * "image" now owns the "buf".
355 static void prepare_image(struct image *image, char *buf, size_t len,
356 int prepare_linetable)
360 memset(image, 0, sizeof(*image));
364 if (!prepare_linetable)
367 ep = image->buf + image->len;
371 for (next = cp; next < ep && *next != '\n'; next++)
375 add_line_info(image, cp, next - cp, 0);
378 image->line = image->line_allocated;
381 static void clear_image(struct image *image)
384 free(image->line_allocated);
385 memset(image, 0, sizeof(*image));
388 /* fmt must contain _one_ %s and no other substitution */
389 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
391 struct strbuf sb = STRBUF_INIT;
393 if (patch->old_name && patch->new_name &&
394 strcmp(patch->old_name, patch->new_name)) {
395 quote_c_style(patch->old_name, &sb, NULL, 0);
396 strbuf_addstr(&sb, " => ");
397 quote_c_style(patch->new_name, &sb, NULL, 0);
399 const char *n = patch->new_name;
402 quote_c_style(n, &sb, NULL, 0);
404 fprintf(output, fmt, sb.buf);
411 static void read_patch_file(struct strbuf *sb, int fd)
413 if (strbuf_read(sb, fd, 0) < 0)
414 die_errno("git apply: failed to read");
417 * Make sure that we have some slop in the buffer
418 * so that we can do speculative "memcmp" etc, and
419 * see to it that it is NUL-filled.
421 strbuf_grow(sb, SLOP);
422 memset(sb->buf + sb->len, 0, SLOP);
425 static unsigned long linelen(const char *buffer, unsigned long size)
427 unsigned long len = 0;
430 if (*buffer++ == '\n')
436 static int is_dev_null(const char *str)
438 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
444 static int name_terminate(const char *name, int namelen, int c, int terminate)
446 if (c == ' ' && !(terminate & TERM_SPACE))
448 if (c == '\t' && !(terminate & TERM_TAB))
454 /* remove double slashes to make --index work with such filenames */
455 static char *squash_slash(char *name)
463 if ((name[j++] = name[i++]) == '/')
464 while (name[i] == '/')
471 static char *find_name_gnu(const char *line, const char *def, int p_value)
473 struct strbuf name = STRBUF_INIT;
477 * Proposed "new-style" GNU patch/diff format; see
478 * http://marc.info/?l=git&m=112927316408690&w=2
480 if (unquote_c_style(&name, line, NULL)) {
481 strbuf_release(&name);
485 for (cp = name.buf; p_value; p_value--) {
486 cp = strchr(cp, '/');
488 strbuf_release(&name);
494 strbuf_remove(&name, 0, cp - name.buf);
496 strbuf_insert(&name, 0, root, root_len);
497 return squash_slash(strbuf_detach(&name, NULL));
500 static size_t sane_tz_len(const char *line, size_t len)
504 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
506 tz = line + len - strlen(" +0500");
508 if (tz[1] != '+' && tz[1] != '-')
511 for (p = tz + 2; p != line + len; p++)
515 return line + len - tz;
518 static size_t tz_with_colon_len(const char *line, size_t len)
522 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
524 tz = line + len - strlen(" +08:00");
526 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
529 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
530 !isdigit(*p++) || !isdigit(*p++))
533 return line + len - tz;
536 static size_t date_len(const char *line, size_t len)
538 const char *date, *p;
540 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
542 p = date = line + len - strlen("72-02-05");
544 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
545 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
546 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
549 if (date - line >= strlen("19") &&
550 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
551 date -= strlen("19");
553 return line + len - date;
556 static size_t short_time_len(const char *line, size_t len)
558 const char *time, *p;
560 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
562 p = time = line + len - strlen(" 07:01:32");
564 /* Permit 1-digit hours? */
566 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
568 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
571 return line + len - time;
574 static size_t fractional_time_len(const char *line, size_t len)
579 /* Expected format: 19:41:17.620000023 */
580 if (!len || !isdigit(line[len - 1]))
584 /* Fractional seconds. */
585 while (p > line && isdigit(*p))
590 /* Hours, minutes, and whole seconds. */
591 n = short_time_len(line, p - line);
595 return line + len - p + n;
598 static size_t trailing_spaces_len(const char *line, size_t len)
602 /* Expected format: ' ' x (1 or more) */
603 if (!len || line[len - 1] != ' ')
610 return line + len - (p + 1);
617 static size_t diff_timestamp_len(const char *line, size_t len)
619 const char *end = line + len;
623 * Posix: 2010-07-05 19:41:17
624 * GNU: 2010-07-05 19:41:17.620000023 -0500
627 if (!isdigit(end[-1]))
630 n = sane_tz_len(line, end - line);
632 n = tz_with_colon_len(line, end - line);
635 n = short_time_len(line, end - line);
637 n = fractional_time_len(line, end - line);
640 n = date_len(line, end - line);
641 if (!n) /* No date. Too bad. */
645 if (end == line) /* No space before date. */
647 if (end[-1] == '\t') { /* Success! */
649 return line + len - end;
651 if (end[-1] != ' ') /* No space before date. */
654 /* Whitespace damage. */
655 end -= trailing_spaces_len(line, end - line);
656 return line + len - end;
659 static char *null_strdup(const char *s)
661 return s ? xstrdup(s) : NULL;
664 static char *find_name_common(const char *line, const char *def,
665 int p_value, const char *end, int terminate)
668 const char *start = NULL;
672 while (line != end) {
675 if (!end && isspace(c)) {
678 if (name_terminate(start, line-start, c, terminate))
682 if (c == '/' && !--p_value)
686 return squash_slash(null_strdup(def));
689 return squash_slash(null_strdup(def));
692 * Generally we prefer the shorter name, especially
693 * if the other one is just a variation of that with
694 * something else tacked on to the end (ie "file.orig"
698 int deflen = strlen(def);
699 if (deflen < len && !strncmp(start, def, deflen))
700 return squash_slash(xstrdup(def));
704 char *ret = xmalloc(root_len + len + 1);
706 memcpy(ret + root_len, start, len);
707 ret[root_len + len] = '\0';
708 return squash_slash(ret);
711 return squash_slash(xmemdupz(start, len));
714 static char *find_name(const char *line, char *def, int p_value, int terminate)
717 char *name = find_name_gnu(line, def, p_value);
722 return find_name_common(line, def, p_value, NULL, terminate);
725 static char *find_name_traditional(const char *line, char *def, int p_value)
731 char *name = find_name_gnu(line, def, p_value);
736 len = strchrnul(line, '\n') - line;
737 date_len = diff_timestamp_len(line, len);
739 return find_name_common(line, def, p_value, NULL, TERM_TAB);
742 return find_name_common(line, def, p_value, line + len, 0);
745 static int count_slashes(const char *cp)
757 * Given the string after "--- " or "+++ ", guess the appropriate
758 * p_value for the given patch.
760 static int guess_p_value(const char *nameline)
765 if (is_dev_null(nameline))
767 name = find_name_traditional(nameline, NULL, 0);
770 cp = strchr(name, '/');
775 * Does it begin with "a/$our-prefix" and such? Then this is
776 * very likely to apply to our directory.
778 if (!strncmp(name, prefix, prefix_length))
779 val = count_slashes(prefix);
782 if (!strncmp(cp, prefix, prefix_length))
783 val = count_slashes(prefix) + 1;
791 * Does the ---/+++ line has the POSIX timestamp after the last HT?
792 * GNU diff puts epoch there to signal a creation/deletion event. Is
793 * this such a timestamp?
795 static int has_epoch_timestamp(const char *nameline)
798 * We are only interested in epoch timestamp; any non-zero
799 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
800 * For the same reason, the date must be either 1969-12-31 or
801 * 1970-01-01, and the seconds part must be "00".
803 const char stamp_regexp[] =
804 "^(1969-12-31|1970-01-01)"
806 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
808 "([-+][0-2][0-9]:?[0-5][0-9])\n";
809 const char *timestamp = NULL, *cp, *colon;
810 static regex_t *stamp;
816 for (cp = nameline; *cp != '\n'; cp++) {
823 stamp = xmalloc(sizeof(*stamp));
824 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
825 warning(_("Cannot prepare timestamp regexp %s"),
831 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
833 if (status != REG_NOMATCH)
834 warning(_("regexec returned %d for input: %s"),
839 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
841 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
843 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
844 if (timestamp[m[3].rm_so] == '-')
845 zoneoffset = -zoneoffset;
848 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
849 * (west of GMT) or 1970-01-01 (east of GMT)
851 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
852 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
855 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
856 strtol(timestamp + 14, NULL, 10) -
859 return ((zoneoffset < 0 && hourminute == 1440) ||
860 (0 <= zoneoffset && !hourminute));
864 * Get the name etc info from the ---/+++ lines of a traditional patch header
866 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
867 * files, we can happily check the index for a match, but for creating a
868 * new file we should try to match whatever "patch" does. I have no idea.
870 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
874 first += 4; /* skip "--- " */
875 second += 4; /* skip "+++ " */
876 if (!p_value_known) {
878 p = guess_p_value(first);
879 q = guess_p_value(second);
881 if (0 <= p && p == q) {
886 if (is_dev_null(first)) {
888 patch->is_delete = 0;
889 name = find_name_traditional(second, NULL, p_value);
890 patch->new_name = name;
891 } else if (is_dev_null(second)) {
893 patch->is_delete = 1;
894 name = find_name_traditional(first, NULL, p_value);
895 patch->old_name = name;
898 first_name = find_name_traditional(first, NULL, p_value);
899 name = find_name_traditional(second, first_name, p_value);
901 if (has_epoch_timestamp(first)) {
903 patch->is_delete = 0;
904 patch->new_name = name;
905 } else if (has_epoch_timestamp(second)) {
907 patch->is_delete = 1;
908 patch->old_name = name;
910 patch->old_name = name;
911 patch->new_name = null_strdup(name);
915 die(_("unable to find filename in patch at line %d"), linenr);
918 static int gitdiff_hdrend(const char *line, struct patch *patch)
924 * We're anal about diff header consistency, to make
925 * sure that we don't end up having strange ambiguous
926 * patches floating around.
928 * As a result, gitdiff_{old|new}name() will check
929 * their names against any previous information, just
932 #define DIFF_OLD_NAME 0
933 #define DIFF_NEW_NAME 1
935 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
937 if (!orig_name && !isnull)
938 return find_name(line, NULL, p_value, TERM_TAB);
947 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
948 another = find_name(line, NULL, p_value, TERM_TAB);
949 if (!another || memcmp(another, name, len + 1))
950 die((side == DIFF_NEW_NAME) ?
951 _("git apply: bad git-diff - inconsistent new filename on line %d") :
952 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
957 /* expect "/dev/null" */
958 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
959 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
964 static int gitdiff_oldname(const char *line, struct patch *patch)
966 char *orig = patch->old_name;
967 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
969 if (orig != patch->old_name)
974 static int gitdiff_newname(const char *line, struct patch *patch)
976 char *orig = patch->new_name;
977 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
979 if (orig != patch->new_name)
984 static int gitdiff_oldmode(const char *line, struct patch *patch)
986 patch->old_mode = strtoul(line, NULL, 8);
990 static int gitdiff_newmode(const char *line, struct patch *patch)
992 patch->new_mode = strtoul(line, NULL, 8);
996 static int gitdiff_delete(const char *line, struct patch *patch)
998 patch->is_delete = 1;
999 free(patch->old_name);
1000 patch->old_name = null_strdup(patch->def_name);
1001 return gitdiff_oldmode(line, patch);
1004 static int gitdiff_newfile(const char *line, struct patch *patch)
1007 free(patch->new_name);
1008 patch->new_name = null_strdup(patch->def_name);
1009 return gitdiff_newmode(line, patch);
1012 static int gitdiff_copysrc(const char *line, struct patch *patch)
1015 free(patch->old_name);
1016 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1020 static int gitdiff_copydst(const char *line, struct patch *patch)
1023 free(patch->new_name);
1024 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1028 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1030 patch->is_rename = 1;
1031 free(patch->old_name);
1032 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1036 static int gitdiff_renamedst(const char *line, struct patch *patch)
1038 patch->is_rename = 1;
1039 free(patch->new_name);
1040 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1044 static int gitdiff_similarity(const char *line, struct patch *patch)
1046 unsigned long val = strtoul(line, NULL, 10);
1052 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1054 unsigned long val = strtoul(line, NULL, 10);
1060 static int gitdiff_index(const char *line, struct patch *patch)
1063 * index line is N hexadecimal, "..", N hexadecimal,
1064 * and optional space with octal mode.
1066 const char *ptr, *eol;
1069 ptr = strchr(line, '.');
1070 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1073 memcpy(patch->old_sha1_prefix, line, len);
1074 patch->old_sha1_prefix[len] = 0;
1077 ptr = strchr(line, ' ');
1078 eol = strchr(line, '\n');
1080 if (!ptr || eol < ptr)
1086 memcpy(patch->new_sha1_prefix, line, len);
1087 patch->new_sha1_prefix[len] = 0;
1089 patch->old_mode = strtoul(ptr+1, NULL, 8);
1094 * This is normal for a diff that doesn't change anything: we'll fall through
1095 * into the next diff. Tell the parser to break out.
1097 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1103 * Skip p_value leading components from "line"; as we do not accept
1104 * absolute paths, return NULL in that case.
1106 static const char *skip_tree_prefix(const char *line, int llen)
1112 return (llen && line[0] == '/') ? NULL : line;
1115 for (i = 0; i < llen; i++) {
1117 if (ch == '/' && --nslash <= 0)
1118 return (i == 0) ? NULL : &line[i + 1];
1124 * This is to extract the same name that appears on "diff --git"
1125 * line. We do not find and return anything if it is a rename
1126 * patch, and it is OK because we will find the name elsewhere.
1127 * We need to reliably find name only when it is mode-change only,
1128 * creation or deletion of an empty file. In any of these cases,
1129 * both sides are the same name under a/ and b/ respectively.
1131 static char *git_header_name(const char *line, int llen)
1134 const char *second = NULL;
1135 size_t len, line_len;
1137 line += strlen("diff --git ");
1138 llen -= strlen("diff --git ");
1142 struct strbuf first = STRBUF_INIT;
1143 struct strbuf sp = STRBUF_INIT;
1145 if (unquote_c_style(&first, line, &second))
1146 goto free_and_fail1;
1148 /* strip the a/b prefix including trailing slash */
1149 cp = skip_tree_prefix(first.buf, first.len);
1151 goto free_and_fail1;
1152 strbuf_remove(&first, 0, cp - first.buf);
1155 * second points at one past closing dq of name.
1156 * find the second name.
1158 while ((second < line + llen) && isspace(*second))
1161 if (line + llen <= second)
1162 goto free_and_fail1;
1163 if (*second == '"') {
1164 if (unquote_c_style(&sp, second, NULL))
1165 goto free_and_fail1;
1166 cp = skip_tree_prefix(sp.buf, sp.len);
1168 goto free_and_fail1;
1169 /* They must match, otherwise ignore */
1170 if (strcmp(cp, first.buf))
1171 goto free_and_fail1;
1172 strbuf_release(&sp);
1173 return strbuf_detach(&first, NULL);
1176 /* unquoted second */
1177 cp = skip_tree_prefix(second, line + llen - second);
1179 goto free_and_fail1;
1180 if (line + llen - cp != first.len ||
1181 memcmp(first.buf, cp, first.len))
1182 goto free_and_fail1;
1183 return strbuf_detach(&first, NULL);
1186 strbuf_release(&first);
1187 strbuf_release(&sp);
1191 /* unquoted first name */
1192 name = skip_tree_prefix(line, llen);
1197 * since the first name is unquoted, a dq if exists must be
1198 * the beginning of the second name.
1200 for (second = name; second < line + llen; second++) {
1201 if (*second == '"') {
1202 struct strbuf sp = STRBUF_INIT;
1205 if (unquote_c_style(&sp, second, NULL))
1206 goto free_and_fail2;
1208 np = skip_tree_prefix(sp.buf, sp.len);
1210 goto free_and_fail2;
1212 len = sp.buf + sp.len - np;
1213 if (len < second - name &&
1214 !strncmp(np, name, len) &&
1215 isspace(name[len])) {
1217 strbuf_remove(&sp, 0, np - sp.buf);
1218 return strbuf_detach(&sp, NULL);
1222 strbuf_release(&sp);
1228 * Accept a name only if it shows up twice, exactly the same
1231 second = strchr(name, '\n');
1234 line_len = second - name;
1235 for (len = 0 ; ; len++) {
1236 switch (name[len]) {
1241 case '\t': case ' ':
1243 * Is this the separator between the preimage
1244 * and the postimage pathname? Again, we are
1245 * only interested in the case where there is
1246 * no rename, as this is only to set def_name
1247 * and a rename patch has the names elsewhere
1248 * in an unambiguous form.
1251 return NULL; /* no postimage name */
1252 second = skip_tree_prefix(name + len + 1,
1253 line_len - (len + 1));
1257 * Does len bytes starting at "name" and "second"
1258 * (that are separated by one HT or SP we just
1259 * found) exactly match?
1261 if (second[len] == '\n' && !strncmp(name, second, len))
1262 return xmemdupz(name, len);
1267 /* Verify that we recognize the lines following a git header */
1268 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1270 unsigned long offset;
1272 /* A git diff has explicit new/delete information, so we don't guess */
1274 patch->is_delete = 0;
1277 * Some things may not have the old name in the
1278 * rest of the headers anywhere (pure mode changes,
1279 * or removing or adding empty files), so we get
1280 * the default name from the header.
1282 patch->def_name = git_header_name(line, len);
1283 if (patch->def_name && root) {
1284 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1286 strcpy(s + root_len, patch->def_name);
1287 free(patch->def_name);
1288 patch->def_name = s;
1294 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1295 static const struct opentry {
1297 int (*fn)(const char *, struct patch *);
1299 { "@@ -", gitdiff_hdrend },
1300 { "--- ", gitdiff_oldname },
1301 { "+++ ", gitdiff_newname },
1302 { "old mode ", gitdiff_oldmode },
1303 { "new mode ", gitdiff_newmode },
1304 { "deleted file mode ", gitdiff_delete },
1305 { "new file mode ", gitdiff_newfile },
1306 { "copy from ", gitdiff_copysrc },
1307 { "copy to ", gitdiff_copydst },
1308 { "rename old ", gitdiff_renamesrc },
1309 { "rename new ", gitdiff_renamedst },
1310 { "rename from ", gitdiff_renamesrc },
1311 { "rename to ", gitdiff_renamedst },
1312 { "similarity index ", gitdiff_similarity },
1313 { "dissimilarity index ", gitdiff_dissimilarity },
1314 { "index ", gitdiff_index },
1315 { "", gitdiff_unrecognized },
1319 len = linelen(line, size);
1320 if (!len || line[len-1] != '\n')
1322 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1323 const struct opentry *p = optable + i;
1324 int oplen = strlen(p->str);
1325 if (len < oplen || memcmp(p->str, line, oplen))
1327 if (p->fn(line + oplen, patch) < 0)
1336 static int parse_num(const char *line, unsigned long *p)
1340 if (!isdigit(*line))
1342 *p = strtoul(line, &ptr, 10);
1346 static int parse_range(const char *line, int len, int offset, const char *expect,
1347 unsigned long *p1, unsigned long *p2)
1351 if (offset < 0 || offset >= len)
1356 digits = parse_num(line, p1);
1366 digits = parse_num(line+1, p2);
1375 ex = strlen(expect);
1378 if (memcmp(line, expect, ex))
1384 static void recount_diff(const char *line, int size, struct fragment *fragment)
1386 int oldlines = 0, newlines = 0, ret = 0;
1389 warning("recount: ignore empty hunk");
1394 int len = linelen(line, size);
1402 case ' ': case '\n':
1414 ret = size < 3 || !starts_with(line, "@@ ");
1417 ret = size < 5 || !starts_with(line, "diff ");
1424 warning(_("recount: unexpected line: %.*s"),
1425 (int)linelen(line, size), line);
1430 fragment->oldlines = oldlines;
1431 fragment->newlines = newlines;
1435 * Parse a unified diff fragment header of the
1436 * form "@@ -a,b +c,d @@"
1438 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1442 if (!len || line[len-1] != '\n')
1445 /* Figure out the number of lines in a fragment */
1446 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1447 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1452 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1454 unsigned long offset, len;
1456 patch->is_toplevel_relative = 0;
1457 patch->is_rename = patch->is_copy = 0;
1458 patch->is_new = patch->is_delete = -1;
1459 patch->old_mode = patch->new_mode = 0;
1460 patch->old_name = patch->new_name = NULL;
1461 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1462 unsigned long nextlen;
1464 len = linelen(line, size);
1468 /* Testing this early allows us to take a few shortcuts.. */
1473 * Make sure we don't find any unconnected patch fragments.
1474 * That's a sign that we didn't find a header, and that a
1475 * patch has become corrupted/broken up.
1477 if (!memcmp("@@ -", line, 4)) {
1478 struct fragment dummy;
1479 if (parse_fragment_header(line, len, &dummy) < 0)
1481 die(_("patch fragment without header at line %d: %.*s"),
1482 linenr, (int)len-1, line);
1489 * Git patch? It might not have a real patch, just a rename
1490 * or mode change, so we handle that specially
1492 if (!memcmp("diff --git ", line, 11)) {
1493 int git_hdr_len = parse_git_header(line, len, size, patch);
1494 if (git_hdr_len <= len)
1496 if (!patch->old_name && !patch->new_name) {
1497 if (!patch->def_name)
1498 die(Q_("git diff header lacks filename information when removing "
1499 "%d leading pathname component (line %d)",
1500 "git diff header lacks filename information when removing "
1501 "%d leading pathname components (line %d)",
1504 patch->old_name = xstrdup(patch->def_name);
1505 patch->new_name = xstrdup(patch->def_name);
1507 if (!patch->is_delete && !patch->new_name)
1508 die("git diff header lacks filename information "
1509 "(line %d)", linenr);
1510 patch->is_toplevel_relative = 1;
1511 *hdrsize = git_hdr_len;
1515 /* --- followed by +++ ? */
1516 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1520 * We only accept unified patches, so we want it to
1521 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1522 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1524 nextlen = linelen(line + len, size - len);
1525 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1528 /* Ok, we'll consider it a patch */
1529 parse_traditional_patch(line, line+len, patch);
1530 *hdrsize = len + nextlen;
1537 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1545 if (squelch_whitespace_errors &&
1546 squelch_whitespace_errors < whitespace_error)
1549 err = whitespace_error_string(result);
1550 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1551 patch_input_file, linenr, err, len, line);
1555 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1557 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1559 record_ws_error(result, line + 1, len - 2, linenr);
1563 * Parse a unified diff. Note that this really needs to parse each
1564 * fragment separately, since the only way to know the difference
1565 * between a "---" that is part of a patch, and a "---" that starts
1566 * the next patch is to look at the line counts..
1568 static int parse_fragment(const char *line, unsigned long size,
1569 struct patch *patch, struct fragment *fragment)
1572 int len = linelen(line, size), offset;
1573 unsigned long oldlines, newlines;
1574 unsigned long leading, trailing;
1576 offset = parse_fragment_header(line, len, fragment);
1579 if (offset > 0 && patch->recount)
1580 recount_diff(line + offset, size - offset, fragment);
1581 oldlines = fragment->oldlines;
1582 newlines = fragment->newlines;
1586 /* Parse the thing.. */
1590 added = deleted = 0;
1593 offset += len, size -= len, line += len, linenr++) {
1594 if (!oldlines && !newlines)
1596 len = linelen(line, size);
1597 if (!len || line[len-1] != '\n')
1602 case '\n': /* newer GNU diff, an empty context line */
1606 if (!deleted && !added)
1609 if (!apply_in_reverse &&
1610 ws_error_action == correct_ws_error)
1611 check_whitespace(line, len, patch->ws_rule);
1614 if (apply_in_reverse &&
1615 ws_error_action != nowarn_ws_error)
1616 check_whitespace(line, len, patch->ws_rule);
1622 if (!apply_in_reverse &&
1623 ws_error_action != nowarn_ws_error)
1624 check_whitespace(line, len, patch->ws_rule);
1631 * We allow "\ No newline at end of file". Depending
1632 * on locale settings when the patch was produced we
1633 * don't know what this line looks like. The only
1634 * thing we do know is that it begins with "\ ".
1635 * Checking for 12 is just for sanity check -- any
1636 * l10n of "\ No newline..." is at least that long.
1639 if (len < 12 || memcmp(line, "\\ ", 2))
1644 if (oldlines || newlines)
1646 fragment->leading = leading;
1647 fragment->trailing = trailing;
1650 * If a fragment ends with an incomplete line, we failed to include
1651 * it in the above loop because we hit oldlines == newlines == 0
1654 if (12 < size && !memcmp(line, "\\ ", 2))
1655 offset += linelen(line, size);
1657 patch->lines_added += added;
1658 patch->lines_deleted += deleted;
1660 if (0 < patch->is_new && oldlines)
1661 return error(_("new file depends on old contents"));
1662 if (0 < patch->is_delete && newlines)
1663 return error(_("deleted file still has contents"));
1668 * We have seen "diff --git a/... b/..." header (or a traditional patch
1669 * header). Read hunks that belong to this patch into fragments and hang
1670 * them to the given patch structure.
1672 * The (fragment->patch, fragment->size) pair points into the memory given
1673 * by the caller, not a copy, when we return.
1675 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1677 unsigned long offset = 0;
1678 unsigned long oldlines = 0, newlines = 0, context = 0;
1679 struct fragment **fragp = &patch->fragments;
1681 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1682 struct fragment *fragment;
1685 fragment = xcalloc(1, sizeof(*fragment));
1686 fragment->linenr = linenr;
1687 len = parse_fragment(line, size, patch, fragment);
1689 die(_("corrupt patch at line %d"), linenr);
1690 fragment->patch = line;
1691 fragment->size = len;
1692 oldlines += fragment->oldlines;
1693 newlines += fragment->newlines;
1694 context += fragment->leading + fragment->trailing;
1697 fragp = &fragment->next;
1705 * If something was removed (i.e. we have old-lines) it cannot
1706 * be creation, and if something was added it cannot be
1707 * deletion. However, the reverse is not true; --unified=0
1708 * patches that only add are not necessarily creation even
1709 * though they do not have any old lines, and ones that only
1710 * delete are not necessarily deletion.
1712 * Unfortunately, a real creation/deletion patch do _not_ have
1713 * any context line by definition, so we cannot safely tell it
1714 * apart with --unified=0 insanity. At least if the patch has
1715 * more than one hunk it is not creation or deletion.
1717 if (patch->is_new < 0 &&
1718 (oldlines || (patch->fragments && patch->fragments->next)))
1720 if (patch->is_delete < 0 &&
1721 (newlines || (patch->fragments && patch->fragments->next)))
1722 patch->is_delete = 0;
1724 if (0 < patch->is_new && oldlines)
1725 die(_("new file %s depends on old contents"), patch->new_name);
1726 if (0 < patch->is_delete && newlines)
1727 die(_("deleted file %s still has contents"), patch->old_name);
1728 if (!patch->is_delete && !newlines && context)
1731 "file %s becomes empty but is not deleted"),
1737 static inline int metadata_changes(struct patch *patch)
1739 return patch->is_rename > 0 ||
1740 patch->is_copy > 0 ||
1741 patch->is_new > 0 ||
1743 (patch->old_mode && patch->new_mode &&
1744 patch->old_mode != patch->new_mode);
1747 static char *inflate_it(const void *data, unsigned long size,
1748 unsigned long inflated_size)
1754 memset(&stream, 0, sizeof(stream));
1756 stream.next_in = (unsigned char *)data;
1757 stream.avail_in = size;
1758 stream.next_out = out = xmalloc(inflated_size);
1759 stream.avail_out = inflated_size;
1760 git_inflate_init(&stream);
1761 st = git_inflate(&stream, Z_FINISH);
1762 git_inflate_end(&stream);
1763 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1771 * Read a binary hunk and return a new fragment; fragment->patch
1772 * points at an allocated memory that the caller must free, so
1773 * it is marked as "->free_patch = 1".
1775 static struct fragment *parse_binary_hunk(char **buf_p,
1776 unsigned long *sz_p,
1781 * Expect a line that begins with binary patch method ("literal"
1782 * or "delta"), followed by the length of data before deflating.
1783 * a sequence of 'length-byte' followed by base-85 encoded data
1784 * should follow, terminated by a newline.
1786 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1787 * and we would limit the patch line to 66 characters,
1788 * so one line can fit up to 13 groups that would decode
1789 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1790 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1793 unsigned long size = *sz_p;
1794 char *buffer = *buf_p;
1796 unsigned long origlen;
1799 struct fragment *frag;
1801 llen = linelen(buffer, size);
1806 if (starts_with(buffer, "delta ")) {
1807 patch_method = BINARY_DELTA_DEFLATED;
1808 origlen = strtoul(buffer + 6, NULL, 10);
1810 else if (starts_with(buffer, "literal ")) {
1811 patch_method = BINARY_LITERAL_DEFLATED;
1812 origlen = strtoul(buffer + 8, NULL, 10);
1820 int byte_length, max_byte_length, newsize;
1821 llen = linelen(buffer, size);
1825 /* consume the blank line */
1831 * Minimum line is "A00000\n" which is 7-byte long,
1832 * and the line length must be multiple of 5 plus 2.
1834 if ((llen < 7) || (llen-2) % 5)
1836 max_byte_length = (llen - 2) / 5 * 4;
1837 byte_length = *buffer;
1838 if ('A' <= byte_length && byte_length <= 'Z')
1839 byte_length = byte_length - 'A' + 1;
1840 else if ('a' <= byte_length && byte_length <= 'z')
1841 byte_length = byte_length - 'a' + 27;
1844 /* if the input length was not multiple of 4, we would
1845 * have filler at the end but the filler should never
1848 if (max_byte_length < byte_length ||
1849 byte_length <= max_byte_length - 4)
1851 newsize = hunk_size + byte_length;
1852 data = xrealloc(data, newsize);
1853 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1855 hunk_size = newsize;
1860 frag = xcalloc(1, sizeof(*frag));
1861 frag->patch = inflate_it(data, hunk_size, origlen);
1862 frag->free_patch = 1;
1866 frag->size = origlen;
1870 frag->binary_patch_method = patch_method;
1876 error(_("corrupt binary patch at line %d: %.*s"),
1877 linenr-1, llen-1, buffer);
1881 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1884 * We have read "GIT binary patch\n"; what follows is a line
1885 * that says the patch method (currently, either "literal" or
1886 * "delta") and the length of data before deflating; a
1887 * sequence of 'length-byte' followed by base-85 encoded data
1890 * When a binary patch is reversible, there is another binary
1891 * hunk in the same format, starting with patch method (either
1892 * "literal" or "delta") with the length of data, and a sequence
1893 * of length-byte + base-85 encoded data, terminated with another
1894 * empty line. This data, when applied to the postimage, produces
1897 struct fragment *forward;
1898 struct fragment *reverse;
1902 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1903 if (!forward && !status)
1904 /* there has to be one hunk (forward hunk) */
1905 return error(_("unrecognized binary patch at line %d"), linenr-1);
1907 /* otherwise we already gave an error message */
1910 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1915 * Not having reverse hunk is not an error, but having
1916 * a corrupt reverse hunk is.
1918 free((void*) forward->patch);
1922 forward->next = reverse;
1923 patch->fragments = forward;
1924 patch->is_binary = 1;
1929 * Read the patch text in "buffer" that extends for "size" bytes; stop
1930 * reading after seeing a single patch (i.e. changes to a single file).
1931 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1932 * Return the number of bytes consumed, so that the caller can call us
1933 * again for the next patch.
1935 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1937 int hdrsize, patchsize;
1938 int offset = find_header(buffer, size, &hdrsize, patch);
1943 patch->ws_rule = whitespace_rule(patch->new_name
1947 patchsize = parse_single_patch(buffer + offset + hdrsize,
1948 size - offset - hdrsize, patch);
1951 static const char git_binary[] = "GIT binary patch\n";
1952 int hd = hdrsize + offset;
1953 unsigned long llen = linelen(buffer + hd, size - hd);
1955 if (llen == sizeof(git_binary) - 1 &&
1956 !memcmp(git_binary, buffer + hd, llen)) {
1959 used = parse_binary(buffer + hd + llen,
1960 size - hd - llen, patch);
1962 patchsize = used + llen;
1966 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1967 static const char *binhdr[] = {
1973 for (i = 0; binhdr[i]; i++) {
1974 int len = strlen(binhdr[i]);
1975 if (len < size - hd &&
1976 !memcmp(binhdr[i], buffer + hd, len)) {
1978 patch->is_binary = 1;
1985 /* Empty patch cannot be applied if it is a text patch
1986 * without metadata change. A binary patch appears
1989 if ((apply || check) &&
1990 (!patch->is_binary && !metadata_changes(patch)))
1991 die(_("patch with only garbage at line %d"), linenr);
1994 return offset + hdrsize + patchsize;
1997 #define swap(a,b) myswap((a),(b),sizeof(a))
1999 #define myswap(a, b, size) do { \
2000 unsigned char mytmp[size]; \
2001 memcpy(mytmp, &a, size); \
2002 memcpy(&a, &b, size); \
2003 memcpy(&b, mytmp, size); \
2006 static void reverse_patches(struct patch *p)
2008 for (; p; p = p->next) {
2009 struct fragment *frag = p->fragments;
2011 swap(p->new_name, p->old_name);
2012 swap(p->new_mode, p->old_mode);
2013 swap(p->is_new, p->is_delete);
2014 swap(p->lines_added, p->lines_deleted);
2015 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2017 for (; frag; frag = frag->next) {
2018 swap(frag->newpos, frag->oldpos);
2019 swap(frag->newlines, frag->oldlines);
2024 static const char pluses[] =
2025 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2026 static const char minuses[]=
2027 "----------------------------------------------------------------------";
2029 static void show_stats(struct patch *patch)
2031 struct strbuf qname = STRBUF_INIT;
2032 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2035 quote_c_style(cp, &qname, NULL, 0);
2038 * "scale" the filename
2044 if (qname.len > max) {
2045 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2047 cp = qname.buf + qname.len + 3 - max;
2048 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2051 if (patch->is_binary) {
2052 printf(" %-*s | Bin\n", max, qname.buf);
2053 strbuf_release(&qname);
2057 printf(" %-*s |", max, qname.buf);
2058 strbuf_release(&qname);
2061 * scale the add/delete
2063 max = max + max_change > 70 ? 70 - max : max_change;
2064 add = patch->lines_added;
2065 del = patch->lines_deleted;
2067 if (max_change > 0) {
2068 int total = ((add + del) * max + max_change / 2) / max_change;
2069 add = (add * max + max_change / 2) / max_change;
2072 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2073 add, pluses, del, minuses);
2076 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2078 switch (st->st_mode & S_IFMT) {
2080 if (strbuf_readlink(buf, path, st->st_size) < 0)
2081 return error(_("unable to read symlink %s"), path);
2084 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2085 return error(_("unable to open or read %s"), path);
2086 convert_to_git(path, buf->buf, buf->len, buf, 0);
2094 * Update the preimage, and the common lines in postimage,
2095 * from buffer buf of length len. If postlen is 0 the postimage
2096 * is updated in place, otherwise it's updated on a new buffer
2100 static void update_pre_post_images(struct image *preimage,
2101 struct image *postimage,
2103 size_t len, size_t postlen)
2105 int i, ctx, reduced;
2106 char *new, *old, *fixed;
2107 struct image fixed_preimage;
2110 * Update the preimage with whitespace fixes. Note that we
2111 * are not losing preimage->buf -- apply_one_fragment() will
2114 prepare_image(&fixed_preimage, buf, len, 1);
2116 ? fixed_preimage.nr == preimage->nr
2117 : fixed_preimage.nr <= preimage->nr);
2118 for (i = 0; i < fixed_preimage.nr; i++)
2119 fixed_preimage.line[i].flag = preimage->line[i].flag;
2120 free(preimage->line_allocated);
2121 *preimage = fixed_preimage;
2124 * Adjust the common context lines in postimage. This can be
2125 * done in-place when we are shrinking it with whitespace
2126 * fixing, but needs a new buffer when ignoring whitespace or
2127 * expanding leading tabs to spaces.
2129 * We trust the caller to tell us if the update can be done
2130 * in place (postlen==0) or not.
2132 old = postimage->buf;
2134 new = postimage->buf = xmalloc(postlen);
2137 fixed = preimage->buf;
2139 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2140 size_t len = postimage->line[i].len;
2141 if (!(postimage->line[i].flag & LINE_COMMON)) {
2142 /* an added line -- no counterparts in preimage */
2143 memmove(new, old, len);
2149 /* a common context -- skip it in the original postimage */
2152 /* and find the corresponding one in the fixed preimage */
2153 while (ctx < preimage->nr &&
2154 !(preimage->line[ctx].flag & LINE_COMMON)) {
2155 fixed += preimage->line[ctx].len;
2160 * preimage is expected to run out, if the caller
2161 * fixed addition of trailing blank lines.
2163 if (preimage->nr <= ctx) {
2168 /* and copy it in, while fixing the line length */
2169 len = preimage->line[ctx].len;
2170 memcpy(new, fixed, len);
2173 postimage->line[i].len = len;
2178 ? postlen < new - postimage->buf
2179 : postimage->len < new - postimage->buf)
2180 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2181 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2183 /* Fix the length of the whole thing */
2184 postimage->len = new - postimage->buf;
2185 postimage->nr -= reduced;
2188 static int match_fragment(struct image *img,
2189 struct image *preimage,
2190 struct image *postimage,
2194 int match_beginning, int match_end)
2197 char *fixed_buf, *buf, *orig, *target;
2198 struct strbuf fixed;
2199 size_t fixed_len, postlen;
2202 if (preimage->nr + try_lno <= img->nr) {
2204 * The hunk falls within the boundaries of img.
2206 preimage_limit = preimage->nr;
2207 if (match_end && (preimage->nr + try_lno != img->nr))
2209 } else if (ws_error_action == correct_ws_error &&
2210 (ws_rule & WS_BLANK_AT_EOF)) {
2212 * This hunk extends beyond the end of img, and we are
2213 * removing blank lines at the end of the file. This
2214 * many lines from the beginning of the preimage must
2215 * match with img, and the remainder of the preimage
2218 preimage_limit = img->nr - try_lno;
2221 * The hunk extends beyond the end of the img and
2222 * we are not removing blanks at the end, so we
2223 * should reject the hunk at this position.
2228 if (match_beginning && try_lno)
2231 /* Quick hash check */
2232 for (i = 0; i < preimage_limit; i++)
2233 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2234 (preimage->line[i].hash != img->line[try_lno + i].hash))
2237 if (preimage_limit == preimage->nr) {
2239 * Do we have an exact match? If we were told to match
2240 * at the end, size must be exactly at try+fragsize,
2241 * otherwise try+fragsize must be still within the preimage,
2242 * and either case, the old piece should match the preimage
2246 ? (try + preimage->len == img->len)
2247 : (try + preimage->len <= img->len)) &&
2248 !memcmp(img->buf + try, preimage->buf, preimage->len))
2252 * The preimage extends beyond the end of img, so
2253 * there cannot be an exact match.
2255 * There must be one non-blank context line that match
2256 * a line before the end of img.
2260 buf = preimage->buf;
2262 for (i = 0; i < preimage_limit; i++)
2263 buf_end += preimage->line[i].len;
2265 for ( ; buf < buf_end; buf++)
2273 * No exact match. If we are ignoring whitespace, run a line-by-line
2274 * fuzzy matching. We collect all the line length information because
2275 * we need it to adjust whitespace if we match.
2277 if (ws_ignore_action == ignore_ws_change) {
2280 size_t postlen = postimage->len;
2284 for (i = 0; i < preimage_limit; i++) {
2285 size_t prelen = preimage->line[i].len;
2286 size_t imglen = img->line[try_lno+i].len;
2288 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2289 preimage->buf + preoff, prelen))
2291 if (preimage->line[i].flag & LINE_COMMON)
2292 postlen += imglen - prelen;
2298 * Ok, the preimage matches with whitespace fuzz.
2300 * imgoff now holds the true length of the target that
2301 * matches the preimage before the end of the file.
2303 * Count the number of characters in the preimage that fall
2304 * beyond the end of the file and make sure that all of them
2305 * are whitespace characters. (This can only happen if
2306 * we are removing blank lines at the end of the file.)
2308 buf = preimage_eof = preimage->buf + preoff;
2309 for ( ; i < preimage->nr; i++)
2310 preoff += preimage->line[i].len;
2311 preimage_end = preimage->buf + preoff;
2312 for ( ; buf < preimage_end; buf++)
2317 * Update the preimage and the common postimage context
2318 * lines to use the same whitespace as the target.
2319 * If whitespace is missing in the target (i.e.
2320 * if the preimage extends beyond the end of the file),
2321 * use the whitespace from the preimage.
2323 extra_chars = preimage_end - preimage_eof;
2324 strbuf_init(&fixed, imgoff + extra_chars);
2325 strbuf_add(&fixed, img->buf + try, imgoff);
2326 strbuf_add(&fixed, preimage_eof, extra_chars);
2327 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2328 update_pre_post_images(preimage, postimage,
2329 fixed_buf, fixed_len, postlen);
2333 if (ws_error_action != correct_ws_error)
2337 * The hunk does not apply byte-by-byte, but the hash says
2338 * it might with whitespace fuzz. We weren't asked to
2339 * ignore whitespace, we were asked to correct whitespace
2340 * errors, so let's try matching after whitespace correction.
2342 * While checking the preimage against the target, whitespace
2343 * errors in both fixed, we count how large the corresponding
2344 * postimage needs to be. The postimage prepared by
2345 * apply_one_fragment() has whitespace errors fixed on added
2346 * lines already, but the common lines were propagated as-is,
2347 * which may become longer when their whitespace errors are
2351 /* First count added lines in postimage */
2353 for (i = 0; i < postimage->nr; i++) {
2354 if (!(postimage->line[i].flag & LINE_COMMON))
2355 postlen += postimage->line[i].len;
2359 * The preimage may extend beyond the end of the file,
2360 * but in this loop we will only handle the part of the
2361 * preimage that falls within the file.
2363 strbuf_init(&fixed, preimage->len + 1);
2364 orig = preimage->buf;
2365 target = img->buf + try;
2366 for (i = 0; i < preimage_limit; i++) {
2367 size_t oldlen = preimage->line[i].len;
2368 size_t tgtlen = img->line[try_lno + i].len;
2369 size_t fixstart = fixed.len;
2370 struct strbuf tgtfix;
2373 /* Try fixing the line in the preimage */
2374 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2376 /* Try fixing the line in the target */
2377 strbuf_init(&tgtfix, tgtlen);
2378 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2381 * If they match, either the preimage was based on
2382 * a version before our tree fixed whitespace breakage,
2383 * or we are lacking a whitespace-fix patch the tree
2384 * the preimage was based on already had (i.e. target
2385 * has whitespace breakage, the preimage doesn't).
2386 * In either case, we are fixing the whitespace breakages
2387 * so we might as well take the fix together with their
2390 match = (tgtfix.len == fixed.len - fixstart &&
2391 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2392 fixed.len - fixstart));
2394 /* Add the length if this is common with the postimage */
2395 if (preimage->line[i].flag & LINE_COMMON)
2396 postlen += tgtfix.len;
2398 strbuf_release(&tgtfix);
2408 * Now handle the lines in the preimage that falls beyond the
2409 * end of the file (if any). They will only match if they are
2410 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2413 for ( ; i < preimage->nr; i++) {
2414 size_t fixstart = fixed.len; /* start of the fixed preimage */
2415 size_t oldlen = preimage->line[i].len;
2418 /* Try fixing the line in the preimage */
2419 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2421 for (j = fixstart; j < fixed.len; j++)
2422 if (!isspace(fixed.buf[j]))
2429 * Yes, the preimage is based on an older version that still
2430 * has whitespace breakages unfixed, and fixing them makes the
2431 * hunk match. Update the context lines in the postimage.
2433 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2434 if (postlen < postimage->len)
2436 update_pre_post_images(preimage, postimage,
2437 fixed_buf, fixed_len, postlen);
2441 strbuf_release(&fixed);
2445 static int find_pos(struct image *img,
2446 struct image *preimage,
2447 struct image *postimage,
2450 int match_beginning, int match_end)
2453 unsigned long backwards, forwards, try;
2454 int backwards_lno, forwards_lno, try_lno;
2457 * If match_beginning or match_end is specified, there is no
2458 * point starting from a wrong line that will never match and
2459 * wander around and wait for a match at the specified end.
2461 if (match_beginning)
2464 line = img->nr - preimage->nr;
2467 * Because the comparison is unsigned, the following test
2468 * will also take care of a negative line number that can
2469 * result when match_end and preimage is larger than the target.
2471 if ((size_t) line > img->nr)
2475 for (i = 0; i < line; i++)
2476 try += img->line[i].len;
2479 * There's probably some smart way to do this, but I'll leave
2480 * that to the smart and beautiful people. I'm simple and stupid.
2483 backwards_lno = line;
2485 forwards_lno = line;
2488 for (i = 0; ; i++) {
2489 if (match_fragment(img, preimage, postimage,
2490 try, try_lno, ws_rule,
2491 match_beginning, match_end))
2495 if (backwards_lno == 0 && forwards_lno == img->nr)
2499 if (backwards_lno == 0) {
2504 backwards -= img->line[backwards_lno].len;
2506 try_lno = backwards_lno;
2508 if (forwards_lno == img->nr) {
2512 forwards += img->line[forwards_lno].len;
2515 try_lno = forwards_lno;
2522 static void remove_first_line(struct image *img)
2524 img->buf += img->line[0].len;
2525 img->len -= img->line[0].len;
2530 static void remove_last_line(struct image *img)
2532 img->len -= img->line[--img->nr].len;
2536 * The change from "preimage" and "postimage" has been found to
2537 * apply at applied_pos (counts in line numbers) in "img".
2538 * Update "img" to remove "preimage" and replace it with "postimage".
2540 static void update_image(struct image *img,
2542 struct image *preimage,
2543 struct image *postimage)
2546 * remove the copy of preimage at offset in img
2547 * and replace it with postimage
2550 size_t remove_count, insert_count, applied_at = 0;
2555 * If we are removing blank lines at the end of img,
2556 * the preimage may extend beyond the end.
2557 * If that is the case, we must be careful only to
2558 * remove the part of the preimage that falls within
2559 * the boundaries of img. Initialize preimage_limit
2560 * to the number of lines in the preimage that falls
2561 * within the boundaries.
2563 preimage_limit = preimage->nr;
2564 if (preimage_limit > img->nr - applied_pos)
2565 preimage_limit = img->nr - applied_pos;
2567 for (i = 0; i < applied_pos; i++)
2568 applied_at += img->line[i].len;
2571 for (i = 0; i < preimage_limit; i++)
2572 remove_count += img->line[applied_pos + i].len;
2573 insert_count = postimage->len;
2575 /* Adjust the contents */
2576 result = xmalloc(img->len + insert_count - remove_count + 1);
2577 memcpy(result, img->buf, applied_at);
2578 memcpy(result + applied_at, postimage->buf, postimage->len);
2579 memcpy(result + applied_at + postimage->len,
2580 img->buf + (applied_at + remove_count),
2581 img->len - (applied_at + remove_count));
2584 img->len += insert_count - remove_count;
2585 result[img->len] = '\0';
2587 /* Adjust the line table */
2588 nr = img->nr + postimage->nr - preimage_limit;
2589 if (preimage_limit < postimage->nr) {
2591 * NOTE: this knows that we never call remove_first_line()
2592 * on anything other than pre/post image.
2594 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2595 img->line_allocated = img->line;
2597 if (preimage_limit != postimage->nr)
2598 memmove(img->line + applied_pos + postimage->nr,
2599 img->line + applied_pos + preimage_limit,
2600 (img->nr - (applied_pos + preimage_limit)) *
2601 sizeof(*img->line));
2602 memcpy(img->line + applied_pos,
2604 postimage->nr * sizeof(*img->line));
2606 for (i = 0; i < postimage->nr; i++)
2607 img->line[applied_pos + i].flag |= LINE_PATCHED;
2612 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2613 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2614 * replace the part of "img" with "postimage" text.
2616 static int apply_one_fragment(struct image *img, struct fragment *frag,
2617 int inaccurate_eof, unsigned ws_rule,
2620 int match_beginning, match_end;
2621 const char *patch = frag->patch;
2622 int size = frag->size;
2623 char *old, *oldlines;
2624 struct strbuf newlines;
2625 int new_blank_lines_at_end = 0;
2626 int found_new_blank_lines_at_end = 0;
2627 int hunk_linenr = frag->linenr;
2628 unsigned long leading, trailing;
2629 int pos, applied_pos;
2630 struct image preimage;
2631 struct image postimage;
2633 memset(&preimage, 0, sizeof(preimage));
2634 memset(&postimage, 0, sizeof(postimage));
2635 oldlines = xmalloc(size);
2636 strbuf_init(&newlines, size);
2641 int len = linelen(patch, size);
2643 int added_blank_line = 0;
2644 int is_blank_context = 0;
2651 * "plen" is how much of the line we should use for
2652 * the actual patch data. Normally we just remove the
2653 * first character on the line, but if the line is
2654 * followed by "\ No newline", then we also remove the
2655 * last one (which is the newline, of course).
2658 if (len < size && patch[len] == '\\')
2661 if (apply_in_reverse) {
2664 else if (first == '+')
2670 /* Newer GNU diff, empty context line */
2672 /* ... followed by '\No newline'; nothing */
2675 strbuf_addch(&newlines, '\n');
2676 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2677 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2678 is_blank_context = 1;
2681 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2682 ws_blank_line(patch + 1, plen, ws_rule))
2683 is_blank_context = 1;
2685 memcpy(old, patch + 1, plen);
2686 add_line_info(&preimage, old, plen,
2687 (first == ' ' ? LINE_COMMON : 0));
2691 /* Fall-through for ' ' */
2693 /* --no-add does not add new lines */
2694 if (first == '+' && no_add)
2697 start = newlines.len;
2699 !whitespace_error ||
2700 ws_error_action != correct_ws_error) {
2701 strbuf_add(&newlines, patch + 1, plen);
2704 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2706 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2707 (first == '+' ? 0 : LINE_COMMON));
2709 (ws_rule & WS_BLANK_AT_EOF) &&
2710 ws_blank_line(patch + 1, plen, ws_rule))
2711 added_blank_line = 1;
2713 case '@': case '\\':
2714 /* Ignore it, we already handled it */
2717 if (apply_verbosely)
2718 error(_("invalid start of line: '%c'"), first);
2721 if (added_blank_line) {
2722 if (!new_blank_lines_at_end)
2723 found_new_blank_lines_at_end = hunk_linenr;
2724 new_blank_lines_at_end++;
2726 else if (is_blank_context)
2729 new_blank_lines_at_end = 0;
2734 if (inaccurate_eof &&
2735 old > oldlines && old[-1] == '\n' &&
2736 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2738 strbuf_setlen(&newlines, newlines.len - 1);
2741 leading = frag->leading;
2742 trailing = frag->trailing;
2745 * A hunk to change lines at the beginning would begin with
2747 * but we need to be careful. -U0 that inserts before the second
2748 * line also has this pattern.
2750 * And a hunk to add to an empty file would begin with
2753 * In other words, a hunk that is (frag->oldpos <= 1) with or
2754 * without leading context must match at the beginning.
2756 match_beginning = (!frag->oldpos ||
2757 (frag->oldpos == 1 && !unidiff_zero));
2760 * A hunk without trailing lines must match at the end.
2761 * However, we simply cannot tell if a hunk must match end
2762 * from the lack of trailing lines if the patch was generated
2763 * with unidiff without any context.
2765 match_end = !unidiff_zero && !trailing;
2767 pos = frag->newpos ? (frag->newpos - 1) : 0;
2768 preimage.buf = oldlines;
2769 preimage.len = old - oldlines;
2770 postimage.buf = newlines.buf;
2771 postimage.len = newlines.len;
2772 preimage.line = preimage.line_allocated;
2773 postimage.line = postimage.line_allocated;
2777 applied_pos = find_pos(img, &preimage, &postimage, pos,
2778 ws_rule, match_beginning, match_end);
2780 if (applied_pos >= 0)
2783 /* Am I at my context limits? */
2784 if ((leading <= p_context) && (trailing <= p_context))
2786 if (match_beginning || match_end) {
2787 match_beginning = match_end = 0;
2792 * Reduce the number of context lines; reduce both
2793 * leading and trailing if they are equal otherwise
2794 * just reduce the larger context.
2796 if (leading >= trailing) {
2797 remove_first_line(&preimage);
2798 remove_first_line(&postimage);
2802 if (trailing > leading) {
2803 remove_last_line(&preimage);
2804 remove_last_line(&postimage);
2809 if (applied_pos >= 0) {
2810 if (new_blank_lines_at_end &&
2811 preimage.nr + applied_pos >= img->nr &&
2812 (ws_rule & WS_BLANK_AT_EOF) &&
2813 ws_error_action != nowarn_ws_error) {
2814 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2815 found_new_blank_lines_at_end);
2816 if (ws_error_action == correct_ws_error) {
2817 while (new_blank_lines_at_end--)
2818 remove_last_line(&postimage);
2821 * We would want to prevent write_out_results()
2822 * from taking place in apply_patch() that follows
2823 * the callchain led us here, which is:
2824 * apply_patch->check_patch_list->check_patch->
2825 * apply_data->apply_fragments->apply_one_fragment
2827 if (ws_error_action == die_on_ws_error)
2831 if (apply_verbosely && applied_pos != pos) {
2832 int offset = applied_pos - pos;
2833 if (apply_in_reverse)
2834 offset = 0 - offset;
2836 Q_("Hunk #%d succeeded at %d (offset %d line).",
2837 "Hunk #%d succeeded at %d (offset %d lines).",
2839 nth_fragment, applied_pos + 1, offset);
2843 * Warn if it was necessary to reduce the number
2846 if ((leading != frag->leading) ||
2847 (trailing != frag->trailing))
2848 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2849 " to apply fragment at %d"),
2850 leading, trailing, applied_pos+1);
2851 update_image(img, applied_pos, &preimage, &postimage);
2853 if (apply_verbosely)
2854 error(_("while searching for:\n%.*s"),
2855 (int)(old - oldlines), oldlines);
2859 strbuf_release(&newlines);
2860 free(preimage.line_allocated);
2861 free(postimage.line_allocated);
2863 return (applied_pos < 0);
2866 static int apply_binary_fragment(struct image *img, struct patch *patch)
2868 struct fragment *fragment = patch->fragments;
2873 return error(_("missing binary patch data for '%s'"),
2878 /* Binary patch is irreversible without the optional second hunk */
2879 if (apply_in_reverse) {
2880 if (!fragment->next)
2881 return error("cannot reverse-apply a binary patch "
2882 "without the reverse hunk to '%s'",
2884 ? patch->new_name : patch->old_name);
2885 fragment = fragment->next;
2887 switch (fragment->binary_patch_method) {
2888 case BINARY_DELTA_DEFLATED:
2889 dst = patch_delta(img->buf, img->len, fragment->patch,
2890 fragment->size, &len);
2897 case BINARY_LITERAL_DEFLATED:
2899 img->len = fragment->size;
2900 img->buf = xmemdupz(fragment->patch, img->len);
2907 * Replace "img" with the result of applying the binary patch.
2908 * The binary patch data itself in patch->fragment is still kept
2909 * but the preimage prepared by the caller in "img" is freed here
2910 * or in the helper function apply_binary_fragment() this calls.
2912 static int apply_binary(struct image *img, struct patch *patch)
2914 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2915 unsigned char sha1[20];
2918 * For safety, we require patch index line to contain
2919 * full 40-byte textual SHA1 for old and new, at least for now.
2921 if (strlen(patch->old_sha1_prefix) != 40 ||
2922 strlen(patch->new_sha1_prefix) != 40 ||
2923 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2924 get_sha1_hex(patch->new_sha1_prefix, sha1))
2925 return error("cannot apply binary patch to '%s' "
2926 "without full index line", name);
2928 if (patch->old_name) {
2930 * See if the old one matches what the patch
2933 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2934 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2935 return error("the patch applies to '%s' (%s), "
2936 "which does not match the "
2937 "current contents.",
2938 name, sha1_to_hex(sha1));
2941 /* Otherwise, the old one must be empty. */
2943 return error("the patch applies to an empty "
2944 "'%s' but it is not empty", name);
2947 get_sha1_hex(patch->new_sha1_prefix, sha1);
2948 if (is_null_sha1(sha1)) {
2950 return 0; /* deletion patch */
2953 if (has_sha1_file(sha1)) {
2954 /* We already have the postimage */
2955 enum object_type type;
2959 result = read_sha1_file(sha1, &type, &size);
2961 return error("the necessary postimage %s for "
2962 "'%s' cannot be read",
2963 patch->new_sha1_prefix, name);
2969 * We have verified buf matches the preimage;
2970 * apply the patch data to it, which is stored
2971 * in the patch->fragments->{patch,size}.
2973 if (apply_binary_fragment(img, patch))
2974 return error(_("binary patch does not apply to '%s'"),
2977 /* verify that the result matches */
2978 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2979 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2980 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2981 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2987 static int apply_fragments(struct image *img, struct patch *patch)
2989 struct fragment *frag = patch->fragments;
2990 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2991 unsigned ws_rule = patch->ws_rule;
2992 unsigned inaccurate_eof = patch->inaccurate_eof;
2995 if (patch->is_binary)
2996 return apply_binary(img, patch);
3000 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3001 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3002 if (!apply_with_reject)
3011 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3013 if (S_ISGITLINK(mode)) {
3014 strbuf_grow(buf, 100);
3015 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3017 enum object_type type;
3021 result = read_sha1_file(sha1, &type, &sz);
3024 /* XXX read_sha1_file NUL-terminates */
3025 strbuf_attach(buf, result, sz, sz + 1);
3030 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3034 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3037 static struct patch *in_fn_table(const char *name)
3039 struct string_list_item *item;
3044 item = string_list_lookup(&fn_table, name);
3046 return (struct patch *)item->util;
3052 * item->util in the filename table records the status of the path.
3053 * Usually it points at a patch (whose result records the contents
3054 * of it after applying it), but it could be PATH_WAS_DELETED for a
3055 * path that a previously applied patch has already removed, or
3056 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3058 * The latter is needed to deal with a case where two paths A and B
3059 * are swapped by first renaming A to B and then renaming B to A;
3060 * moving A to B should not be prevented due to presence of B as we
3061 * will remove it in a later patch.
3063 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3064 #define PATH_WAS_DELETED ((struct patch *) -1)
3066 static int to_be_deleted(struct patch *patch)
3068 return patch == PATH_TO_BE_DELETED;
3071 static int was_deleted(struct patch *patch)
3073 return patch == PATH_WAS_DELETED;
3076 static void add_to_fn_table(struct patch *patch)
3078 struct string_list_item *item;
3081 * Always add new_name unless patch is a deletion
3082 * This should cover the cases for normal diffs,
3083 * file creations and copies
3085 if (patch->new_name != NULL) {
3086 item = string_list_insert(&fn_table, patch->new_name);
3091 * store a failure on rename/deletion cases because
3092 * later chunks shouldn't patch old names
3094 if ((patch->new_name == NULL) || (patch->is_rename)) {
3095 item = string_list_insert(&fn_table, patch->old_name);
3096 item->util = PATH_WAS_DELETED;
3100 static void prepare_fn_table(struct patch *patch)
3103 * store information about incoming file deletion
3106 if ((patch->new_name == NULL) || (patch->is_rename)) {
3107 struct string_list_item *item;
3108 item = string_list_insert(&fn_table, patch->old_name);
3109 item->util = PATH_TO_BE_DELETED;
3111 patch = patch->next;
3115 static int checkout_target(struct cache_entry *ce, struct stat *st)
3117 struct checkout costate;
3119 memset(&costate, 0, sizeof(costate));
3120 costate.base_dir = "";
3121 costate.refresh_cache = 1;
3122 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3123 return error(_("cannot checkout %s"), ce->name);
3127 static struct patch *previous_patch(struct patch *patch, int *gone)
3129 struct patch *previous;
3132 if (patch->is_copy || patch->is_rename)
3133 return NULL; /* "git" patches do not depend on the order */
3135 previous = in_fn_table(patch->old_name);
3139 if (to_be_deleted(previous))
3140 return NULL; /* the deletion hasn't happened yet */
3142 if (was_deleted(previous))
3148 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3150 if (S_ISGITLINK(ce->ce_mode)) {
3151 if (!S_ISDIR(st->st_mode))
3155 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3158 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3160 static int load_patch_target(struct strbuf *buf,
3161 const struct cache_entry *ce,
3164 unsigned expected_mode)
3167 if (read_file_or_gitlink(ce, buf))
3168 return error(_("read of %s failed"), name);
3170 if (S_ISGITLINK(expected_mode)) {
3172 return read_file_or_gitlink(ce, buf);
3174 return SUBMODULE_PATCH_WITHOUT_INDEX;
3176 if (read_old_data(st, name, buf))
3177 return error(_("read of %s failed"), name);
3184 * We are about to apply "patch"; populate the "image" with the
3185 * current version we have, from the working tree or from the index,
3186 * depending on the situation e.g. --cached/--index. If we are
3187 * applying a non-git patch that incrementally updates the tree,
3188 * we read from the result of a previous diff.
3190 static int load_preimage(struct image *image,
3191 struct patch *patch, struct stat *st,
3192 const struct cache_entry *ce)
3194 struct strbuf buf = STRBUF_INIT;
3197 struct patch *previous;
3200 previous = previous_patch(patch, &status);
3202 return error(_("path %s has been renamed/deleted"),
3205 /* We have a patched copy in memory; use that. */
3206 strbuf_add(&buf, previous->result, previous->resultsize);
3208 status = load_patch_target(&buf, ce, st,
3209 patch->old_name, patch->old_mode);
3212 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3214 * There is no way to apply subproject
3215 * patch without looking at the index.
3216 * NEEDSWORK: shouldn't this be flagged
3219 free_fragment_list(patch->fragments);
3220 patch->fragments = NULL;
3221 } else if (status) {
3222 return error(_("read of %s failed"), patch->old_name);
3226 img = strbuf_detach(&buf, &len);
3227 prepare_image(image, img, len, !patch->is_binary);
3231 static int three_way_merge(struct image *image,
3233 const unsigned char *base,
3234 const unsigned char *ours,
3235 const unsigned char *theirs)
3237 mmfile_t base_file, our_file, their_file;
3238 mmbuffer_t result = { NULL };
3241 read_mmblob(&base_file, base);
3242 read_mmblob(&our_file, ours);
3243 read_mmblob(&their_file, theirs);
3244 status = ll_merge(&result, path,
3247 &their_file, "theirs", NULL);
3248 free(base_file.ptr);
3250 free(their_file.ptr);
3251 if (status < 0 || !result.ptr) {
3256 image->buf = result.ptr;
3257 image->len = result.size;
3263 * When directly falling back to add/add three-way merge, we read from
3264 * the current contents of the new_name. In no cases other than that
3265 * this function will be called.
3267 static int load_current(struct image *image, struct patch *patch)
3269 struct strbuf buf = STRBUF_INIT;
3274 struct cache_entry *ce;
3275 char *name = patch->new_name;
3276 unsigned mode = patch->new_mode;
3279 die("BUG: patch to %s is not a creation", patch->old_name);
3281 pos = cache_name_pos(name, strlen(name));
3283 return error(_("%s: does not exist in index"), name);
3284 ce = active_cache[pos];
3285 if (lstat(name, &st)) {
3286 if (errno != ENOENT)
3287 return error(_("%s: %s"), name, strerror(errno));
3288 if (checkout_target(ce, &st))
3291 if (verify_index_match(ce, &st))
3292 return error(_("%s: does not match index"), name);
3294 status = load_patch_target(&buf, ce, &st, name, mode);
3299 img = strbuf_detach(&buf, &len);
3300 prepare_image(image, img, len, !patch->is_binary);
3304 static int try_threeway(struct image *image, struct patch *patch,
3305 struct stat *st, const struct cache_entry *ce)
3307 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3308 struct strbuf buf = STRBUF_INIT;
3312 struct image tmp_image;
3314 /* No point falling back to 3-way merge in these cases */
3315 if (patch->is_delete ||
3316 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3319 /* Preimage the patch was prepared for */
3321 write_sha1_file("", 0, blob_type, pre_sha1);
3322 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3323 read_blob_object(&buf, pre_sha1, patch->old_mode))
3324 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3326 fprintf(stderr, "Falling back to three-way merge...\n");
3328 img = strbuf_detach(&buf, &len);
3329 prepare_image(&tmp_image, img, len, 1);
3330 /* Apply the patch to get the post image */
3331 if (apply_fragments(&tmp_image, patch) < 0) {
3332 clear_image(&tmp_image);
3335 /* post_sha1[] is theirs */
3336 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3337 clear_image(&tmp_image);
3339 /* our_sha1[] is ours */
3340 if (patch->is_new) {
3341 if (load_current(&tmp_image, patch))
3342 return error("cannot read the current contents of '%s'",
3345 if (load_preimage(&tmp_image, patch, st, ce))
3346 return error("cannot read the current contents of '%s'",
3349 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3350 clear_image(&tmp_image);
3352 /* in-core three-way merge between post and our using pre as base */
3353 status = three_way_merge(image, patch->new_name,
3354 pre_sha1, our_sha1, post_sha1);
3356 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3361 patch->conflicted_threeway = 1;
3363 hashclr(patch->threeway_stage[0]);
3365 hashcpy(patch->threeway_stage[0], pre_sha1);
3366 hashcpy(patch->threeway_stage[1], our_sha1);
3367 hashcpy(patch->threeway_stage[2], post_sha1);
3368 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3370 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3375 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3379 if (load_preimage(&image, patch, st, ce) < 0)
3382 if (patch->direct_to_threeway ||
3383 apply_fragments(&image, patch) < 0) {
3384 /* Note: with --reject, apply_fragments() returns 0 */
3385 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3388 patch->result = image.buf;
3389 patch->resultsize = image.len;
3390 add_to_fn_table(patch);
3391 free(image.line_allocated);
3393 if (0 < patch->is_delete && patch->resultsize)
3394 return error(_("removal patch leaves file contents"));
3400 * If "patch" that we are looking at modifies or deletes what we have,
3401 * we would want it not to lose any local modification we have, either
3402 * in the working tree or in the index.
3404 * This also decides if a non-git patch is a creation patch or a
3405 * modification to an existing empty file. We do not check the state
3406 * of the current tree for a creation patch in this function; the caller
3407 * check_patch() separately makes sure (and errors out otherwise) that
3408 * the path the patch creates does not exist in the current tree.
3410 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3412 const char *old_name = patch->old_name;
3413 struct patch *previous = NULL;
3414 int stat_ret = 0, status;
3415 unsigned st_mode = 0;
3420 assert(patch->is_new <= 0);
3421 previous = previous_patch(patch, &status);
3424 return error(_("path %s has been renamed/deleted"), old_name);
3426 st_mode = previous->new_mode;
3427 } else if (!cached) {
3428 stat_ret = lstat(old_name, st);
3429 if (stat_ret && errno != ENOENT)
3430 return error(_("%s: %s"), old_name, strerror(errno));
3433 if (check_index && !previous) {
3434 int pos = cache_name_pos(old_name, strlen(old_name));
3436 if (patch->is_new < 0)
3438 return error(_("%s: does not exist in index"), old_name);
3440 *ce = active_cache[pos];
3442 if (checkout_target(*ce, st))
3445 if (!cached && verify_index_match(*ce, st))
3446 return error(_("%s: does not match index"), old_name);
3448 st_mode = (*ce)->ce_mode;
3449 } else if (stat_ret < 0) {
3450 if (patch->is_new < 0)
3452 return error(_("%s: %s"), old_name, strerror(errno));
3455 if (!cached && !previous)
3456 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3458 if (patch->is_new < 0)
3460 if (!patch->old_mode)
3461 patch->old_mode = st_mode;
3462 if ((st_mode ^ patch->old_mode) & S_IFMT)
3463 return error(_("%s: wrong type"), old_name);
3464 if (st_mode != patch->old_mode)
3465 warning(_("%s has type %o, expected %o"),
3466 old_name, st_mode, patch->old_mode);
3467 if (!patch->new_mode && !patch->is_delete)
3468 patch->new_mode = st_mode;
3473 patch->is_delete = 0;
3474 free(patch->old_name);
3475 patch->old_name = NULL;
3480 #define EXISTS_IN_INDEX 1
3481 #define EXISTS_IN_WORKTREE 2
3483 static int check_to_create(const char *new_name, int ok_if_exists)
3488 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3490 return EXISTS_IN_INDEX;
3494 if (!lstat(new_name, &nst)) {
3495 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3498 * A leading component of new_name might be a symlink
3499 * that is going to be removed with this patch, but
3500 * still pointing at somewhere that has the path.
3501 * In such a case, path "new_name" does not exist as
3502 * far as git is concerned.
3504 if (has_symlink_leading_path(new_name, strlen(new_name)))
3507 return EXISTS_IN_WORKTREE;
3508 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3509 return error("%s: %s", new_name, strerror(errno));
3515 * Check and apply the patch in-core; leave the result in patch->result
3516 * for the caller to write it out to the final destination.
3518 static int check_patch(struct patch *patch)
3521 const char *old_name = patch->old_name;
3522 const char *new_name = patch->new_name;
3523 const char *name = old_name ? old_name : new_name;
3524 struct cache_entry *ce = NULL;
3525 struct patch *tpatch;
3529 patch->rejected = 1; /* we will drop this after we succeed */
3531 status = check_preimage(patch, &ce, &st);
3534 old_name = patch->old_name;
3537 * A type-change diff is always split into a patch to delete
3538 * old, immediately followed by a patch to create new (see
3539 * diff.c::run_diff()); in such a case it is Ok that the entry
3540 * to be deleted by the previous patch is still in the working
3541 * tree and in the index.
3543 * A patch to swap-rename between A and B would first rename A
3544 * to B and then rename B to A. While applying the first one,
3545 * the presence of B should not stop A from getting renamed to
3546 * B; ask to_be_deleted() about the later rename. Removal of
3547 * B and rename from A to B is handled the same way by asking
3550 if ((tpatch = in_fn_table(new_name)) &&
3551 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3557 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3558 int err = check_to_create(new_name, ok_if_exists);
3560 if (err && threeway) {
3561 patch->direct_to_threeway = 1;
3562 } else switch (err) {
3565 case EXISTS_IN_INDEX:
3566 return error(_("%s: already exists in index"), new_name);
3568 case EXISTS_IN_WORKTREE:
3569 return error(_("%s: already exists in working directory"),
3575 if (!patch->new_mode) {
3576 if (0 < patch->is_new)
3577 patch->new_mode = S_IFREG | 0644;
3579 patch->new_mode = patch->old_mode;
3583 if (new_name && old_name) {
3584 int same = !strcmp(old_name, new_name);
3585 if (!patch->new_mode)
3586 patch->new_mode = patch->old_mode;
3587 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3589 return error(_("new mode (%o) of %s does not "
3590 "match old mode (%o)"),
3591 patch->new_mode, new_name,
3594 return error(_("new mode (%o) of %s does not "
3595 "match old mode (%o) of %s"),
3596 patch->new_mode, new_name,
3597 patch->old_mode, old_name);
3601 if (apply_data(patch, &st, ce) < 0)
3602 return error(_("%s: patch does not apply"), name);
3603 patch->rejected = 0;
3607 static int check_patch_list(struct patch *patch)
3611 prepare_fn_table(patch);
3613 if (apply_verbosely)
3614 say_patch_name(stderr,
3615 _("Checking patch %s..."), patch);
3616 err |= check_patch(patch);
3617 patch = patch->next;
3622 /* This function tries to read the sha1 from the current index */
3623 static int get_current_sha1(const char *path, unsigned char *sha1)
3627 if (read_cache() < 0)
3629 pos = cache_name_pos(path, strlen(path));
3632 hashcpy(sha1, active_cache[pos]->sha1);
3636 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3639 * A usable gitlink patch has only one fragment (hunk) that looks like:
3641 * -Subproject commit <old sha1>
3642 * +Subproject commit <new sha1>
3645 * -Subproject commit <old sha1>
3646 * for a removal patch.
3648 struct fragment *hunk = p->fragments;
3649 static const char heading[] = "-Subproject commit ";
3652 if (/* does the patch have only one hunk? */
3653 hunk && !hunk->next &&
3654 /* is its preimage one line? */
3655 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3656 /* does preimage begin with the heading? */
3657 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3658 starts_with(++preimage, heading) &&
3659 /* does it record full SHA-1? */
3660 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3661 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3662 /* does the abbreviated name on the index line agree with it? */
3663 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3664 return 0; /* it all looks fine */
3666 /* we may have full object name on the index line */
3667 return get_sha1_hex(p->old_sha1_prefix, sha1);
3670 /* Build an index that contains the just the files needed for a 3way merge */
3671 static void build_fake_ancestor(struct patch *list, const char *filename)
3673 struct patch *patch;
3674 struct index_state result = { NULL };
3677 /* Once we start supporting the reverse patch, it may be
3678 * worth showing the new sha1 prefix, but until then...
3680 for (patch = list; patch; patch = patch->next) {
3681 unsigned char sha1[20];
3682 struct cache_entry *ce;
3685 name = patch->old_name ? patch->old_name : patch->new_name;
3686 if (0 < patch->is_new)
3689 if (S_ISGITLINK(patch->old_mode)) {
3690 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3691 ; /* ok, the textual part looks sane */
3693 die("sha1 information is lacking or useless for submoule %s",
3695 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3697 } else if (!patch->lines_added && !patch->lines_deleted) {
3698 /* mode-only change: update the current */
3699 if (get_current_sha1(patch->old_name, sha1))
3700 die("mode change for %s, which is not "
3701 "in current HEAD", name);
3703 die("sha1 information is lacking or useless "
3706 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3708 die(_("make_cache_entry failed for path '%s'"), name);
3709 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3710 die ("Could not add %s to temporary index", name);
3713 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3714 if (fd < 0 || write_index(&result, fd) || close(fd))
3715 die ("Could not write temporary index to %s", filename);
3717 discard_index(&result);
3720 static void stat_patch_list(struct patch *patch)
3722 int files, adds, dels;
3724 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3726 adds += patch->lines_added;
3727 dels += patch->lines_deleted;
3731 print_stat_summary(stdout, files, adds, dels);
3734 static void numstat_patch_list(struct patch *patch)
3736 for ( ; patch; patch = patch->next) {
3738 name = patch->new_name ? patch->new_name : patch->old_name;
3739 if (patch->is_binary)
3742 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3743 write_name_quoted(name, stdout, line_termination);
3747 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3750 printf(" %s mode %06o %s\n", newdelete, mode, name);
3752 printf(" %s %s\n", newdelete, name);
3755 static void show_mode_change(struct patch *p, int show_name)
3757 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3759 printf(" mode change %06o => %06o %s\n",
3760 p->old_mode, p->new_mode, p->new_name);
3762 printf(" mode change %06o => %06o\n",
3763 p->old_mode, p->new_mode);
3767 static void show_rename_copy(struct patch *p)
3769 const char *renamecopy = p->is_rename ? "rename" : "copy";
3770 const char *old, *new;
3772 /* Find common prefix */
3776 const char *slash_old, *slash_new;
3777 slash_old = strchr(old, '/');
3778 slash_new = strchr(new, '/');
3781 slash_old - old != slash_new - new ||
3782 memcmp(old, new, slash_new - new))
3784 old = slash_old + 1;
3785 new = slash_new + 1;
3787 /* p->old_name thru old is the common prefix, and old and new
3788 * through the end of names are renames
3790 if (old != p->old_name)
3791 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3792 (int)(old - p->old_name), p->old_name,
3793 old, new, p->score);
3795 printf(" %s %s => %s (%d%%)\n", renamecopy,
3796 p->old_name, p->new_name, p->score);
3797 show_mode_change(p, 0);
3800 static void summary_patch_list(struct patch *patch)
3804 for (p = patch; p; p = p->next) {
3806 show_file_mode_name("create", p->new_mode, p->new_name);
3807 else if (p->is_delete)
3808 show_file_mode_name("delete", p->old_mode, p->old_name);
3810 if (p->is_rename || p->is_copy)
3811 show_rename_copy(p);
3814 printf(" rewrite %s (%d%%)\n",
3815 p->new_name, p->score);
3816 show_mode_change(p, 0);
3819 show_mode_change(p, 1);
3825 static void patch_stats(struct patch *patch)
3827 int lines = patch->lines_added + patch->lines_deleted;
3829 if (lines > max_change)
3831 if (patch->old_name) {
3832 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3834 len = strlen(patch->old_name);
3838 if (patch->new_name) {
3839 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3841 len = strlen(patch->new_name);
3847 static void remove_file(struct patch *patch, int rmdir_empty)
3850 if (remove_file_from_cache(patch->old_name) < 0)
3851 die(_("unable to remove %s from index"), patch->old_name);
3854 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3855 remove_path(patch->old_name);
3860 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3863 struct cache_entry *ce;
3864 int namelen = strlen(path);
3865 unsigned ce_size = cache_entry_size(namelen);
3870 ce = xcalloc(1, ce_size);
3871 memcpy(ce->name, path, namelen);
3872 ce->ce_mode = create_ce_mode(mode);
3873 ce->ce_flags = create_ce_flags(0);
3874 ce->ce_namelen = namelen;
3875 if (S_ISGITLINK(mode)) {
3876 const char *s = buf;
3878 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3879 die(_("corrupt patch for submodule %s"), path);
3882 if (lstat(path, &st) < 0)
3883 die_errno(_("unable to stat newly created file '%s'"),
3885 fill_stat_cache_info(ce, &st);
3887 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3888 die(_("unable to create backing store for newly created file %s"), path);
3890 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3891 die(_("unable to add cache entry for %s"), path);
3894 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3897 struct strbuf nbuf = STRBUF_INIT;
3899 if (S_ISGITLINK(mode)) {
3901 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3903 return mkdir(path, 0777);
3906 if (has_symlinks && S_ISLNK(mode))
3907 /* Although buf:size is counted string, it also is NUL
3910 return symlink(buf, path);
3912 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3916 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3920 write_or_die(fd, buf, size);
3921 strbuf_release(&nbuf);
3924 die_errno(_("closing file '%s'"), path);
3929 * We optimistically assume that the directories exist,
3930 * which is true 99% of the time anyway. If they don't,
3931 * we create them and try again.
3933 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3937 if (!try_create_file(path, mode, buf, size))
3940 if (errno == ENOENT) {
3941 if (safe_create_leading_directories(path))
3943 if (!try_create_file(path, mode, buf, size))
3947 if (errno == EEXIST || errno == EACCES) {
3948 /* We may be trying to create a file where a directory
3952 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3956 if (errno == EEXIST) {
3957 unsigned int nr = getpid();
3960 char newpath[PATH_MAX];
3961 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3962 if (!try_create_file(newpath, mode, buf, size)) {
3963 if (!rename(newpath, path))
3965 unlink_or_warn(newpath);
3968 if (errno != EEXIST)
3973 die_errno(_("unable to write file '%s' mode %o"), path, mode);
3976 static void add_conflicted_stages_file(struct patch *patch)
3979 unsigned ce_size, mode;
3980 struct cache_entry *ce;
3984 namelen = strlen(patch->new_name);
3985 ce_size = cache_entry_size(namelen);
3986 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
3988 remove_file_from_cache(patch->new_name);
3989 for (stage = 1; stage < 4; stage++) {
3990 if (is_null_sha1(patch->threeway_stage[stage - 1]))
3992 ce = xcalloc(1, ce_size);
3993 memcpy(ce->name, patch->new_name, namelen);
3994 ce->ce_mode = create_ce_mode(mode);
3995 ce->ce_flags = create_ce_flags(stage);
3996 ce->ce_namelen = namelen;
3997 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
3998 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3999 die(_("unable to add cache entry for %s"), patch->new_name);
4003 static void create_file(struct patch *patch)
4005 char *path = patch->new_name;
4006 unsigned mode = patch->new_mode;
4007 unsigned long size = patch->resultsize;
4008 char *buf = patch->result;
4011 mode = S_IFREG | 0644;
4012 create_one_file(path, mode, buf, size);
4014 if (patch->conflicted_threeway)
4015 add_conflicted_stages_file(patch);
4017 add_index_file(path, mode, buf, size);
4020 /* phase zero is to remove, phase one is to create */
4021 static void write_out_one_result(struct patch *patch, int phase)
4023 if (patch->is_delete > 0) {
4025 remove_file(patch, 1);
4028 if (patch->is_new > 0 || patch->is_copy) {
4034 * Rename or modification boils down to the same
4035 * thing: remove the old, write the new
4038 remove_file(patch, patch->is_rename);
4043 static int write_out_one_reject(struct patch *patch)
4046 char namebuf[PATH_MAX];
4047 struct fragment *frag;
4049 struct strbuf sb = STRBUF_INIT;
4051 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4052 if (!frag->rejected)
4058 if (apply_verbosely)
4059 say_patch_name(stderr,
4060 _("Applied patch %s cleanly."), patch);
4064 /* This should not happen, because a removal patch that leaves
4065 * contents are marked "rejected" at the patch level.
4067 if (!patch->new_name)
4068 die(_("internal error"));
4070 /* Say this even without --verbose */
4071 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4072 "Applying patch %%s with %d rejects...",
4075 say_patch_name(stderr, sb.buf, patch);
4076 strbuf_release(&sb);
4078 cnt = strlen(patch->new_name);
4079 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4080 cnt = ARRAY_SIZE(namebuf) - 5;
4081 warning(_("truncating .rej filename to %.*s.rej"),
4082 cnt - 1, patch->new_name);
4084 memcpy(namebuf, patch->new_name, cnt);
4085 memcpy(namebuf + cnt, ".rej", 5);
4087 rej = fopen(namebuf, "w");
4089 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4091 /* Normal git tools never deal with .rej, so do not pretend
4092 * this is a git patch by saying --git or giving extended
4093 * headers. While at it, maybe please "kompare" that wants
4094 * the trailing TAB and some garbage at the end of line ;-).
4096 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4097 patch->new_name, patch->new_name);
4098 for (cnt = 1, frag = patch->fragments;
4100 cnt++, frag = frag->next) {
4101 if (!frag->rejected) {
4102 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4105 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4106 fprintf(rej, "%.*s", frag->size, frag->patch);
4107 if (frag->patch[frag->size-1] != '\n')
4114 static int write_out_results(struct patch *list)
4119 struct string_list cpath = STRING_LIST_INIT_DUP;
4121 for (phase = 0; phase < 2; phase++) {
4127 write_out_one_result(l, phase);
4129 if (write_out_one_reject(l))
4131 if (l->conflicted_threeway) {
4132 string_list_append(&cpath, l->new_name);
4142 struct string_list_item *item;
4144 sort_string_list(&cpath);
4145 for_each_string_list_item(item, &cpath)
4146 fprintf(stderr, "U %s\n", item->string);
4147 string_list_clear(&cpath, 0);
4155 static struct lock_file lock_file;
4157 static struct string_list limit_by_name;
4158 static int has_include;
4159 static void add_name_limit(const char *name, int exclude)
4161 struct string_list_item *it;
4163 it = string_list_append(&limit_by_name, name);
4164 it->util = exclude ? NULL : (void *) 1;
4167 static int use_patch(struct patch *p)
4169 const char *pathname = p->new_name ? p->new_name : p->old_name;
4172 /* Paths outside are not touched regardless of "--include" */
4173 if (0 < prefix_length) {
4174 int pathlen = strlen(pathname);
4175 if (pathlen <= prefix_length ||
4176 memcmp(prefix, pathname, prefix_length))
4180 /* See if it matches any of exclude/include rule */
4181 for (i = 0; i < limit_by_name.nr; i++) {
4182 struct string_list_item *it = &limit_by_name.items[i];
4183 if (!wildmatch(it->string, pathname, 0, NULL))
4184 return (it->util != NULL);
4188 * If we had any include, a path that does not match any rule is
4189 * not used. Otherwise, we saw bunch of exclude rules (or none)
4190 * and such a path is used.
4192 return !has_include;
4196 static void prefix_one(char **name)
4198 char *old_name = *name;
4201 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
4205 static void prefix_patches(struct patch *p)
4207 if (!prefix || p->is_toplevel_relative)
4209 for ( ; p; p = p->next) {
4210 prefix_one(&p->new_name);
4211 prefix_one(&p->old_name);
4215 #define INACCURATE_EOF (1<<0)
4216 #define RECOUNT (1<<1)
4218 static int apply_patch(int fd, const char *filename, int options)
4221 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4222 struct patch *list = NULL, **listp = &list;
4223 int skipped_patch = 0;
4225 patch_input_file = filename;
4226 read_patch_file(&buf, fd);
4228 while (offset < buf.len) {
4229 struct patch *patch;
4232 patch = xcalloc(1, sizeof(*patch));
4233 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4234 patch->recount = !!(options & RECOUNT);
4235 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4238 if (apply_in_reverse)
4239 reverse_patches(patch);
4241 prefix_patches(patch);
4242 if (use_patch(patch)) {
4245 listp = &patch->next;
4254 if (!list && !skipped_patch)
4255 die(_("unrecognized input"));
4257 if (whitespace_error && (ws_error_action == die_on_ws_error))
4260 update_index = check_index && apply;
4261 if (update_index && newfd < 0)
4262 newfd = hold_locked_index(&lock_file, 1);
4265 if (read_cache() < 0)
4266 die(_("unable to read index file"));
4269 if ((check || apply) &&
4270 check_patch_list(list) < 0 &&
4274 if (apply && write_out_results(list)) {
4275 if (apply_with_reject)
4277 /* with --3way, we still need to write the index out */
4282 build_fake_ancestor(list, fake_ancestor);
4285 stat_patch_list(list);
4288 numstat_patch_list(list);
4291 summary_patch_list(list);
4293 free_patch_list(list);
4294 strbuf_release(&buf);
4295 string_list_clear(&fn_table, 0);
4299 static int git_apply_config(const char *var, const char *value, void *cb)
4301 if (!strcmp(var, "apply.whitespace"))
4302 return git_config_string(&apply_default_whitespace, var, value);
4303 else if (!strcmp(var, "apply.ignorewhitespace"))
4304 return git_config_string(&apply_default_ignorewhitespace, var, value);
4305 return git_default_config(var, value, cb);
4308 static int option_parse_exclude(const struct option *opt,
4309 const char *arg, int unset)
4311 add_name_limit(arg, 1);
4315 static int option_parse_include(const struct option *opt,
4316 const char *arg, int unset)
4318 add_name_limit(arg, 0);
4323 static int option_parse_p(const struct option *opt,
4324 const char *arg, int unset)
4326 p_value = atoi(arg);
4331 static int option_parse_z(const struct option *opt,
4332 const char *arg, int unset)
4335 line_termination = '\n';
4337 line_termination = 0;
4341 static int option_parse_space_change(const struct option *opt,
4342 const char *arg, int unset)
4345 ws_ignore_action = ignore_ws_none;
4347 ws_ignore_action = ignore_ws_change;
4351 static int option_parse_whitespace(const struct option *opt,
4352 const char *arg, int unset)
4354 const char **whitespace_option = opt->value;
4356 *whitespace_option = arg;
4357 parse_whitespace_option(arg);
4361 static int option_parse_directory(const struct option *opt,
4362 const char *arg, int unset)
4364 root_len = strlen(arg);
4365 if (root_len && arg[root_len - 1] != '/') {
4367 root = new_root = xmalloc(root_len + 2);
4368 strcpy(new_root, arg);
4369 strcpy(new_root + root_len++, "/");
4375 int cmd_apply(int argc, const char **argv, const char *prefix_)
4379 int is_not_gitdir = !startup_info->have_repository;
4380 int force_apply = 0;
4382 const char *whitespace_option = NULL;
4384 struct option builtin_apply_options[] = {
4385 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4386 N_("don't apply changes matching the given path"),
4387 0, option_parse_exclude },
4388 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4389 N_("apply changes matching the given path"),
4390 0, option_parse_include },
4391 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4392 N_("remove <num> leading slashes from traditional diff paths"),
4393 0, option_parse_p },
4394 OPT_BOOL(0, "no-add", &no_add,
4395 N_("ignore additions made by the patch")),
4396 OPT_BOOL(0, "stat", &diffstat,
4397 N_("instead of applying the patch, output diffstat for the input")),
4398 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4399 OPT_NOOP_NOARG(0, "binary"),
4400 OPT_BOOL(0, "numstat", &numstat,
4401 N_("show number of added and deleted lines in decimal notation")),
4402 OPT_BOOL(0, "summary", &summary,
4403 N_("instead of applying the patch, output a summary for the input")),
4404 OPT_BOOL(0, "check", &check,
4405 N_("instead of applying the patch, see if the patch is applicable")),
4406 OPT_BOOL(0, "index", &check_index,
4407 N_("make sure the patch is applicable to the current index")),
4408 OPT_BOOL(0, "cached", &cached,
4409 N_("apply a patch without touching the working tree")),
4410 OPT_BOOL(0, "apply", &force_apply,
4411 N_("also apply the patch (use with --stat/--summary/--check)")),
4412 OPT_BOOL('3', "3way", &threeway,
4413 N_( "attempt three-way merge if a patch does not apply")),
4414 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4415 N_("build a temporary index based on embedded index information")),
4416 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4417 N_("paths are separated with NUL character"),
4418 PARSE_OPT_NOARG, option_parse_z },
4419 OPT_INTEGER('C', NULL, &p_context,
4420 N_("ensure at least <n> lines of context match")),
4421 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4422 N_("detect new or modified lines that have whitespace errors"),
4423 0, option_parse_whitespace },
4424 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4425 N_("ignore changes in whitespace when finding context"),
4426 PARSE_OPT_NOARG, option_parse_space_change },
4427 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4428 N_("ignore changes in whitespace when finding context"),
4429 PARSE_OPT_NOARG, option_parse_space_change },
4430 OPT_BOOL('R', "reverse", &apply_in_reverse,
4431 N_("apply the patch in reverse")),
4432 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4433 N_("don't expect at least one line of context")),
4434 OPT_BOOL(0, "reject", &apply_with_reject,
4435 N_("leave the rejected hunks in corresponding *.rej files")),
4436 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4437 N_("allow overlapping hunks")),
4438 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4439 OPT_BIT(0, "inaccurate-eof", &options,
4440 N_("tolerate incorrectly detected missing new-line at the end of file"),
4442 OPT_BIT(0, "recount", &options,
4443 N_("do not trust the line counts in the hunk headers"),
4445 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4446 N_("prepend <root> to all filenames"),
4447 0, option_parse_directory },
4452 prefix_length = prefix ? strlen(prefix) : 0;
4453 git_config(git_apply_config, NULL);
4454 if (apply_default_whitespace)
4455 parse_whitespace_option(apply_default_whitespace);
4456 if (apply_default_ignorewhitespace)
4457 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4459 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4462 if (apply_with_reject && threeway)
4463 die("--reject and --3way cannot be used together.");
4464 if (cached && threeway)
4465 die("--cached and --3way cannot be used together.");
4468 die(_("--3way outside a repository"));
4471 if (apply_with_reject)
4472 apply = apply_verbosely = 1;
4473 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4475 if (check_index && is_not_gitdir)
4476 die(_("--index outside a repository"));
4479 die(_("--cached outside a repository"));
4482 for (i = 0; i < argc; i++) {
4483 const char *arg = argv[i];
4486 if (!strcmp(arg, "-")) {
4487 errs |= apply_patch(0, "<stdin>", options);
4490 } else if (0 < prefix_length)
4491 arg = prefix_filename(prefix, prefix_length, arg);
4493 fd = open(arg, O_RDONLY);
4495 die_errno(_("can't open patch '%s'"), arg);
4497 set_default_whitespace_mode(whitespace_option);
4498 errs |= apply_patch(fd, arg, options);
4501 set_default_whitespace_mode(whitespace_option);
4503 errs |= apply_patch(0, "<stdin>", options);
4504 if (whitespace_error) {
4505 if (squelch_whitespace_errors &&
4506 squelch_whitespace_errors < whitespace_error) {
4508 whitespace_error - squelch_whitespace_errors;
4509 warning(Q_("squelched %d whitespace error",
4510 "squelched %d whitespace errors",
4514 if (ws_error_action == die_on_ws_error)
4515 die(Q_("%d line adds whitespace errors.",
4516 "%d lines add whitespace errors.",
4519 if (applied_after_fixing_ws && apply)
4520 warning("%d line%s applied after"
4521 " fixing whitespace errors.",
4522 applied_after_fixing_ws,
4523 applied_after_fixing_ws == 1 ? "" : "s");
4524 else if (whitespace_error)
4525 warning(Q_("%d line adds whitespace errors.",
4526 "%d lines add whitespace errors.",
4532 if (write_cache(newfd, active_cache, active_nr) ||
4533 commit_locked_index(&lock_file))
4534 die(_("Unable to write new index file"));