Merge branch 'ye/doc-http-proto' into maint
[git.git] / builtin / apply.c
blob622ee1674af592f4f843577f319c6883a1665f02
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "diff.h"
18 #include "parse-options.h"
19 #include "xdiff-interface.h"
20 #include "ll-merge.h"
21 #include "rerere.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;
41 static int cached;
42 static int diffstat;
43 static int numstat;
44 static int summary;
45 static int check;
46 static int apply = 1;
47 static int apply_in_reverse;
48 static int apply_with_reject;
49 static int apply_verbosely;
50 static int allow_overlap;
51 static int no_add;
52 static int threeway;
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>...]"),
58 NULL
61 static enum ws_error_action {
62 nowarn_ws_error,
63 warn_on_ws_error,
64 die_on_ws_error,
65 correct_ws_error
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 {
72 ignore_ws_none,
73 ignore_ws_change
74 } ws_ignore_action = ignore_ws_none;
77 static const char *patch_input_file;
78 static const char *root;
79 static int root_len;
80 static int read_stdin = 1;
81 static int options;
83 static void parse_whitespace_option(const char *option)
85 if (!option) {
86 ws_error_action = warn_on_ws_error;
87 return;
89 if (!strcmp(option, "warn")) {
90 ws_error_action = warn_on_ws_error;
91 return;
93 if (!strcmp(option, "nowarn")) {
94 ws_error_action = nowarn_ws_error;
95 return;
97 if (!strcmp(option, "error")) {
98 ws_error_action = die_on_ws_error;
99 return;
101 if (!strcmp(option, "error-all")) {
102 ws_error_action = die_on_ws_error;
103 squelch_whitespace_errors = 0;
104 return;
106 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
107 ws_error_action = correct_ws_error;
108 return;
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;
119 return;
121 if (!strcmp(option, "change")) {
122 ws_ignore_action = ignore_ws_change;
123 return;
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
137 * scaling.
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
153 * of context lines.
155 struct fragment {
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.
163 const char *patch;
164 unsigned free_patch:1,
165 rejected:1;
166 int size;
167 int linenr;
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.
185 struct patch {
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 */
189 int rejected;
190 unsigned ws_rule;
191 int lines_added, lines_deleted;
192 int score;
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;
202 char *result;
203 size_t resultsize;
204 char old_sha1_prefix[41];
205 char new_sha1_prefix[41];
206 struct patch *next;
208 /* three-way fallback result */
209 unsigned char threeway_stage[3][20];
212 static void free_fragment_list(struct fragment *list)
214 while (list) {
215 struct fragment *next = list->next;
216 if (list->free_patch)
217 free((char *)list->patch);
218 free(list);
219 list = next;
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);
229 free(patch->result);
230 free(patch);
233 static void free_patch_list(struct patch *list)
235 while (list) {
236 struct patch *next = list->next;
237 free_patch(list);
238 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'.
247 struct line {
248 size_t len;
249 unsigned hash : 24;
250 unsigned flag : 8;
251 #define LINE_COMMON 1
252 #define LINE_PATCHED 2
256 * This represents a "file", which is an array of "lines".
258 struct image {
259 char *buf;
260 size_t len;
261 size_t nr;
262 size_t alloc;
263 struct line *line_allocated;
264 struct line *line;
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)
276 size_t i;
277 uint32_t h;
278 for (i = 0, h = 0; i < len; i++) {
279 if (!isspace(cp[i])) {
280 h = h * 3 + (cp[i] & 0xff);
283 return h;
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;
295 int result = 0;
297 /* ignore line endings */
298 while ((*last1 == '\r') || (*last1 == '\n'))
299 last1--;
300 while ((*last2 == '\r') || (*last2 == '\n'))
301 last2--;
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))
306 s1++;
307 while (isspace(*s2) && (s2 <= last2))
308 s2++;
310 /* early return if both lines are empty */
311 if ((s1 > last1) && (s2 > last2))
312 return 1;
313 while (!result) {
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
318 * "ab"
320 if (isspace(*s1) && isspace(*s2)) {
321 while (isspace(*s1) && s1 <= last1)
322 s1++;
323 while (isspace(*s2) && s2 <= last2)
324 s2++;
327 * If we reached the end on one side only,
328 * lines don't match
330 if (
331 ((s2 > last2) && (s1 <= last1)) ||
332 ((s1 > last1) && (s2 <= last2)))
333 return 0;
334 if ((s1 > last1) && (s2 > last2))
335 break;
338 return !result;
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;
347 img->nr++;
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)
358 const char *cp, *ep;
360 memset(image, 0, sizeof(*image));
361 image->buf = buf;
362 image->len = len;
364 if (!prepare_linetable)
365 return;
367 ep = image->buf + image->len;
368 cp = image->buf;
369 while (cp < ep) {
370 const char *next;
371 for (next = cp; next < ep && *next != '\n'; next++)
373 if (next < ep)
374 next++;
375 add_line_info(image, cp, next - cp, 0);
376 cp = next;
378 image->line = image->line_allocated;
381 static void clear_image(struct image *image)
383 free(image->buf);
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);
398 } else {
399 const char *n = patch->new_name;
400 if (!n)
401 n = patch->old_name;
402 quote_c_style(n, &sb, NULL, 0);
404 fprintf(output, fmt, sb.buf);
405 fputc('\n', output);
406 strbuf_release(&sb);
409 #define SLOP (16)
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;
428 while (size--) {
429 len++;
430 if (*buffer++ == '\n')
431 break;
433 return len;
436 static int is_dev_null(const char *str)
438 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
441 #define TERM_SPACE 1
442 #define TERM_TAB 2
444 static int name_terminate(const char *name, int namelen, int c, int terminate)
446 if (c == ' ' && !(terminate & TERM_SPACE))
447 return 0;
448 if (c == '\t' && !(terminate & TERM_TAB))
449 return 0;
451 return 1;
454 /* remove double slashes to make --index work with such filenames */
455 static char *squash_slash(char *name)
457 int i = 0, j = 0;
459 if (!name)
460 return NULL;
462 while (name[i]) {
463 if ((name[j++] = name[i++]) == '/')
464 while (name[i] == '/')
465 i++;
467 name[j] = '\0';
468 return name;
471 static char *find_name_gnu(const char *line, const char *def, int p_value)
473 struct strbuf name = STRBUF_INIT;
474 char *cp;
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);
482 return NULL;
485 for (cp = name.buf; p_value; p_value--) {
486 cp = strchr(cp, '/');
487 if (!cp) {
488 strbuf_release(&name);
489 return NULL;
491 cp++;
494 strbuf_remove(&name, 0, cp - name.buf);
495 if (root)
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)
502 const char *tz, *p;
504 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
505 return 0;
506 tz = line + len - strlen(" +0500");
508 if (tz[1] != '+' && tz[1] != '-')
509 return 0;
511 for (p = tz + 2; p != line + len; p++)
512 if (!isdigit(*p))
513 return 0;
515 return line + len - tz;
518 static size_t tz_with_colon_len(const char *line, size_t len)
520 const char *tz, *p;
522 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
523 return 0;
524 tz = line + len - strlen(" +08:00");
526 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
527 return 0;
528 p = tz + 2;
529 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
530 !isdigit(*p++) || !isdigit(*p++))
531 return 0;
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")] != '-')
541 return 0;
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. */
547 return 0;
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")] != ':')
561 return 0;
562 p = time = line + len - strlen(" 07:01:32");
564 /* Permit 1-digit hours? */
565 if (*p++ != ' ' ||
566 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
568 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
569 return 0;
571 return line + len - time;
574 static size_t fractional_time_len(const char *line, size_t len)
576 const char *p;
577 size_t n;
579 /* Expected format: 19:41:17.620000023 */
580 if (!len || !isdigit(line[len - 1]))
581 return 0;
582 p = line + len - 1;
584 /* Fractional seconds. */
585 while (p > line && isdigit(*p))
586 p--;
587 if (*p != '.')
588 return 0;
590 /* Hours, minutes, and whole seconds. */
591 n = short_time_len(line, p - line);
592 if (!n)
593 return 0;
595 return line + len - p + n;
598 static size_t trailing_spaces_len(const char *line, size_t len)
600 const char *p;
602 /* Expected format: ' ' x (1 or more) */
603 if (!len || line[len - 1] != ' ')
604 return 0;
606 p = line + len;
607 while (p != line) {
608 p--;
609 if (*p != ' ')
610 return line + len - (p + 1);
613 /* All spaces! */
614 return len;
617 static size_t diff_timestamp_len(const char *line, size_t len)
619 const char *end = line + len;
620 size_t n;
623 * Posix: 2010-07-05 19:41:17
624 * GNU: 2010-07-05 19:41:17.620000023 -0500
627 if (!isdigit(end[-1]))
628 return 0;
630 n = sane_tz_len(line, end - line);
631 if (!n)
632 n = tz_with_colon_len(line, end - line);
633 end -= n;
635 n = short_time_len(line, end - line);
636 if (!n)
637 n = fractional_time_len(line, end - line);
638 end -= n;
640 n = date_len(line, end - line);
641 if (!n) /* No date. Too bad. */
642 return 0;
643 end -= n;
645 if (end == line) /* No space before date. */
646 return 0;
647 if (end[-1] == '\t') { /* Success! */
648 end--;
649 return line + len - end;
651 if (end[-1] != ' ') /* No space before date. */
652 return 0;
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)
667 int len;
668 const char *start = NULL;
670 if (p_value == 0)
671 start = line;
672 while (line != end) {
673 char c = *line;
675 if (!end && isspace(c)) {
676 if (c == '\n')
677 break;
678 if (name_terminate(start, line-start, c, terminate))
679 break;
681 line++;
682 if (c == '/' && !--p_value)
683 start = line;
685 if (!start)
686 return squash_slash(null_strdup(def));
687 len = line - start;
688 if (!len)
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"
695 * or "file~").
697 if (def) {
698 int deflen = strlen(def);
699 if (deflen < len && !strncmp(start, def, deflen))
700 return squash_slash(xstrdup(def));
703 if (root) {
704 char *ret = xmalloc(root_len + len + 1);
705 strcpy(ret, root);
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)
716 if (*line == '"') {
717 char *name = find_name_gnu(line, def, p_value);
718 if (name)
719 return name;
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)
727 size_t len;
728 size_t date_len;
730 if (*line == '"') {
731 char *name = find_name_gnu(line, def, p_value);
732 if (name)
733 return name;
736 len = strchrnul(line, '\n') - line;
737 date_len = diff_timestamp_len(line, len);
738 if (!date_len)
739 return find_name_common(line, def, p_value, NULL, TERM_TAB);
740 len -= date_len;
742 return find_name_common(line, def, p_value, line + len, 0);
745 static int count_slashes(const char *cp)
747 int cnt = 0;
748 char ch;
750 while ((ch = *cp++))
751 if (ch == '/')
752 cnt++;
753 return cnt;
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)
762 char *name, *cp;
763 int val = -1;
765 if (is_dev_null(nameline))
766 return -1;
767 name = find_name_traditional(nameline, NULL, 0);
768 if (!name)
769 return -1;
770 cp = strchr(name, '/');
771 if (!cp)
772 val = 0;
773 else if (prefix) {
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);
780 else {
781 cp++;
782 if (!strncmp(cp, prefix, prefix_length))
783 val = count_slashes(prefix) + 1;
786 free(name);
787 return val;
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;
811 regmatch_t m[10];
812 int zoneoffset;
813 int hourminute;
814 int status;
816 for (cp = nameline; *cp != '\n'; cp++) {
817 if (*cp == '\t')
818 timestamp = cp + 1;
820 if (!timestamp)
821 return 0;
822 if (!stamp) {
823 stamp = xmalloc(sizeof(*stamp));
824 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
825 warning(_("Cannot prepare timestamp regexp %s"),
826 stamp_regexp);
827 return 0;
831 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
832 if (status) {
833 if (status != REG_NOMATCH)
834 warning(_("regexec returned %d for input: %s"),
835 status, timestamp);
836 return 0;
839 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
840 if (*colon == ':')
841 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
842 else
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)))
853 return 0;
855 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
856 strtol(timestamp + 14, NULL, 10) -
857 zoneoffset);
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)
872 char *name;
874 first += 4; /* skip "--- " */
875 second += 4; /* skip "+++ " */
876 if (!p_value_known) {
877 int p, q;
878 p = guess_p_value(first);
879 q = guess_p_value(second);
880 if (p < 0) p = q;
881 if (0 <= p && p == q) {
882 p_value = p;
883 p_value_known = 1;
886 if (is_dev_null(first)) {
887 patch->is_new = 1;
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)) {
892 patch->is_new = 0;
893 patch->is_delete = 1;
894 name = find_name_traditional(first, NULL, p_value);
895 patch->old_name = name;
896 } else {
897 char *first_name;
898 first_name = find_name_traditional(first, NULL, p_value);
899 name = find_name_traditional(second, first_name, p_value);
900 free(first_name);
901 if (has_epoch_timestamp(first)) {
902 patch->is_new = 1;
903 patch->is_delete = 0;
904 patch->new_name = name;
905 } else if (has_epoch_timestamp(second)) {
906 patch->is_new = 0;
907 patch->is_delete = 1;
908 patch->old_name = name;
909 } else {
910 patch->old_name = name;
911 patch->new_name = null_strdup(name);
914 if (!name)
915 die(_("unable to find filename in patch at line %d"), linenr);
918 static int gitdiff_hdrend(const char *line, struct patch *patch)
920 return -1;
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
930 * to make sure..
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);
940 if (orig_name) {
941 int len;
942 const char *name;
943 char *another;
944 name = orig_name;
945 len = strlen(name);
946 if (isnull)
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);
953 free(another);
954 return orig_name;
956 else {
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);
960 return NULL;
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,
968 DIFF_OLD_NAME);
969 if (orig != patch->old_name)
970 free(orig);
971 return 0;
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,
978 DIFF_NEW_NAME);
979 if (orig != patch->new_name)
980 free(orig);
981 return 0;
984 static int gitdiff_oldmode(const char *line, struct patch *patch)
986 patch->old_mode = strtoul(line, NULL, 8);
987 return 0;
990 static int gitdiff_newmode(const char *line, struct patch *patch)
992 patch->new_mode = strtoul(line, NULL, 8);
993 return 0;
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)
1006 patch->is_new = 1;
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)
1014 patch->is_copy = 1;
1015 free(patch->old_name);
1016 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1017 return 0;
1020 static int gitdiff_copydst(const char *line, struct patch *patch)
1022 patch->is_copy = 1;
1023 free(patch->new_name);
1024 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1025 return 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);
1033 return 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);
1041 return 0;
1044 static int gitdiff_similarity(const char *line, struct patch *patch)
1046 unsigned long val = strtoul(line, NULL, 10);
1047 if (val <= 100)
1048 patch->score = val;
1049 return 0;
1052 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1054 unsigned long val = strtoul(line, NULL, 10);
1055 if (val <= 100)
1056 patch->score = val;
1057 return 0;
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;
1067 int len;
1069 ptr = strchr(line, '.');
1070 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1071 return 0;
1072 len = ptr - line;
1073 memcpy(patch->old_sha1_prefix, line, len);
1074 patch->old_sha1_prefix[len] = 0;
1076 line = ptr + 2;
1077 ptr = strchr(line, ' ');
1078 eol = strchr(line, '\n');
1080 if (!ptr || eol < ptr)
1081 ptr = eol;
1082 len = ptr - line;
1084 if (40 < len)
1085 return 0;
1086 memcpy(patch->new_sha1_prefix, line, len);
1087 patch->new_sha1_prefix[len] = 0;
1088 if (*ptr == ' ')
1089 patch->old_mode = strtoul(ptr+1, NULL, 8);
1090 return 0;
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)
1099 return -1;
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)
1108 int nslash;
1109 int i;
1111 if (!p_value)
1112 return (llen && line[0] == '/') ? NULL : line;
1114 nslash = p_value;
1115 for (i = 0; i < llen; i++) {
1116 int ch = line[i];
1117 if (ch == '/' && --nslash <= 0)
1118 return (i == 0) ? NULL : &line[i + 1];
1120 return NULL;
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)
1133 const char *name;
1134 const char *second = NULL;
1135 size_t len, line_len;
1137 line += strlen("diff --git ");
1138 llen -= strlen("diff --git ");
1140 if (*line == '"') {
1141 const char *cp;
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);
1150 if (!cp)
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))
1159 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);
1167 if (!cp)
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);
1178 if (!cp)
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);
1185 free_and_fail1:
1186 strbuf_release(&first);
1187 strbuf_release(&sp);
1188 return NULL;
1191 /* unquoted first name */
1192 name = skip_tree_prefix(line, llen);
1193 if (!name)
1194 return NULL;
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;
1203 const char *np;
1205 if (unquote_c_style(&sp, second, NULL))
1206 goto free_and_fail2;
1208 np = skip_tree_prefix(sp.buf, sp.len);
1209 if (!np)
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])) {
1216 /* Good */
1217 strbuf_remove(&sp, 0, np - sp.buf);
1218 return strbuf_detach(&sp, NULL);
1221 free_and_fail2:
1222 strbuf_release(&sp);
1223 return NULL;
1228 * Accept a name only if it shows up twice, exactly the same
1229 * form.
1231 second = strchr(name, '\n');
1232 if (!second)
1233 return NULL;
1234 line_len = second - name;
1235 for (len = 0 ; ; len++) {
1236 switch (name[len]) {
1237 default:
1238 continue;
1239 case '\n':
1240 return NULL;
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.
1250 if (!name[len + 1])
1251 return NULL; /* no postimage name */
1252 second = skip_tree_prefix(name + len + 1,
1253 line_len - (len + 1));
1254 if (!second)
1255 return NULL;
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 */
1273 patch->is_new = 0;
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);
1285 strcpy(s, root);
1286 strcpy(s + root_len, patch->def_name);
1287 free(patch->def_name);
1288 patch->def_name = s;
1291 line += len;
1292 size -= len;
1293 linenr++;
1294 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1295 static const struct opentry {
1296 const char *str;
1297 int (*fn)(const char *, struct patch *);
1298 } optable[] = {
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 },
1317 int i;
1319 len = linelen(line, size);
1320 if (!len || line[len-1] != '\n')
1321 break;
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))
1326 continue;
1327 if (p->fn(line + oplen, patch) < 0)
1328 return offset;
1329 break;
1333 return offset;
1336 static int parse_num(const char *line, unsigned long *p)
1338 char *ptr;
1340 if (!isdigit(*line))
1341 return 0;
1342 *p = strtoul(line, &ptr, 10);
1343 return ptr - line;
1346 static int parse_range(const char *line, int len, int offset, const char *expect,
1347 unsigned long *p1, unsigned long *p2)
1349 int digits, ex;
1351 if (offset < 0 || offset >= len)
1352 return -1;
1353 line += offset;
1354 len -= offset;
1356 digits = parse_num(line, p1);
1357 if (!digits)
1358 return -1;
1360 offset += digits;
1361 line += digits;
1362 len -= digits;
1364 *p2 = 1;
1365 if (*line == ',') {
1366 digits = parse_num(line+1, p2);
1367 if (!digits)
1368 return -1;
1370 offset += digits+1;
1371 line += digits+1;
1372 len -= digits+1;
1375 ex = strlen(expect);
1376 if (ex > len)
1377 return -1;
1378 if (memcmp(line, expect, ex))
1379 return -1;
1381 return offset + ex;
1384 static void recount_diff(const char *line, int size, struct fragment *fragment)
1386 int oldlines = 0, newlines = 0, ret = 0;
1388 if (size < 1) {
1389 warning("recount: ignore empty hunk");
1390 return;
1393 for (;;) {
1394 int len = linelen(line, size);
1395 size -= len;
1396 line += len;
1398 if (size < 1)
1399 break;
1401 switch (*line) {
1402 case ' ': case '\n':
1403 newlines++;
1404 /* fall through */
1405 case '-':
1406 oldlines++;
1407 continue;
1408 case '+':
1409 newlines++;
1410 continue;
1411 case '\\':
1412 continue;
1413 case '@':
1414 ret = size < 3 || !starts_with(line, "@@ ");
1415 break;
1416 case 'd':
1417 ret = size < 5 || !starts_with(line, "diff ");
1418 break;
1419 default:
1420 ret = -1;
1421 break;
1423 if (ret) {
1424 warning(_("recount: unexpected line: %.*s"),
1425 (int)linelen(line, size), line);
1426 return;
1428 break;
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)
1440 int offset;
1442 if (!len || line[len-1] != '\n')
1443 return -1;
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);
1449 return offset;
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);
1465 if (!len)
1466 break;
1468 /* Testing this early allows us to take a few shortcuts.. */
1469 if (len < 6)
1470 continue;
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)
1480 continue;
1481 die(_("patch fragment without header at line %d: %.*s"),
1482 linenr, (int)len-1, line);
1485 if (size < len + 6)
1486 break;
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)
1495 continue;
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)",
1502 p_value),
1503 p_value, linenr);
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;
1512 return offset;
1515 /* --- followed by +++ ? */
1516 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1517 continue;
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))
1526 continue;
1528 /* Ok, we'll consider it a patch */
1529 parse_traditional_patch(line, line+len, patch);
1530 *hdrsize = len + nextlen;
1531 linenr += 2;
1532 return offset;
1534 return -1;
1537 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1539 char *err;
1541 if (!result)
1542 return;
1544 whitespace_error++;
1545 if (squelch_whitespace_errors &&
1546 squelch_whitespace_errors < whitespace_error)
1547 return;
1549 err = whitespace_error_string(result);
1550 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1551 patch_input_file, linenr, err, len, line);
1552 free(err);
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)
1571 int added, deleted;
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);
1577 if (offset < 0)
1578 return -1;
1579 if (offset > 0 && patch->recount)
1580 recount_diff(line + offset, size - offset, fragment);
1581 oldlines = fragment->oldlines;
1582 newlines = fragment->newlines;
1583 leading = 0;
1584 trailing = 0;
1586 /* Parse the thing.. */
1587 line += len;
1588 size -= len;
1589 linenr++;
1590 added = deleted = 0;
1591 for (offset = len;
1592 0 < size;
1593 offset += len, size -= len, line += len, linenr++) {
1594 if (!oldlines && !newlines)
1595 break;
1596 len = linelen(line, size);
1597 if (!len || line[len-1] != '\n')
1598 return -1;
1599 switch (*line) {
1600 default:
1601 return -1;
1602 case '\n': /* newer GNU diff, an empty context line */
1603 case ' ':
1604 oldlines--;
1605 newlines--;
1606 if (!deleted && !added)
1607 leading++;
1608 trailing++;
1609 break;
1610 case '-':
1611 if (apply_in_reverse &&
1612 ws_error_action != nowarn_ws_error)
1613 check_whitespace(line, len, patch->ws_rule);
1614 deleted++;
1615 oldlines--;
1616 trailing = 0;
1617 break;
1618 case '+':
1619 if (!apply_in_reverse &&
1620 ws_error_action != nowarn_ws_error)
1621 check_whitespace(line, len, patch->ws_rule);
1622 added++;
1623 newlines--;
1624 trailing = 0;
1625 break;
1628 * We allow "\ No newline at end of file". Depending
1629 * on locale settings when the patch was produced we
1630 * don't know what this line looks like. The only
1631 * thing we do know is that it begins with "\ ".
1632 * Checking for 12 is just for sanity check -- any
1633 * l10n of "\ No newline..." is at least that long.
1635 case '\\':
1636 if (len < 12 || memcmp(line, "\\ ", 2))
1637 return -1;
1638 break;
1641 if (oldlines || newlines)
1642 return -1;
1643 fragment->leading = leading;
1644 fragment->trailing = trailing;
1647 * If a fragment ends with an incomplete line, we failed to include
1648 * it in the above loop because we hit oldlines == newlines == 0
1649 * before seeing it.
1651 if (12 < size && !memcmp(line, "\\ ", 2))
1652 offset += linelen(line, size);
1654 patch->lines_added += added;
1655 patch->lines_deleted += deleted;
1657 if (0 < patch->is_new && oldlines)
1658 return error(_("new file depends on old contents"));
1659 if (0 < patch->is_delete && newlines)
1660 return error(_("deleted file still has contents"));
1661 return offset;
1665 * We have seen "diff --git a/... b/..." header (or a traditional patch
1666 * header). Read hunks that belong to this patch into fragments and hang
1667 * them to the given patch structure.
1669 * The (fragment->patch, fragment->size) pair points into the memory given
1670 * by the caller, not a copy, when we return.
1672 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1674 unsigned long offset = 0;
1675 unsigned long oldlines = 0, newlines = 0, context = 0;
1676 struct fragment **fragp = &patch->fragments;
1678 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1679 struct fragment *fragment;
1680 int len;
1682 fragment = xcalloc(1, sizeof(*fragment));
1683 fragment->linenr = linenr;
1684 len = parse_fragment(line, size, patch, fragment);
1685 if (len <= 0)
1686 die(_("corrupt patch at line %d"), linenr);
1687 fragment->patch = line;
1688 fragment->size = len;
1689 oldlines += fragment->oldlines;
1690 newlines += fragment->newlines;
1691 context += fragment->leading + fragment->trailing;
1693 *fragp = fragment;
1694 fragp = &fragment->next;
1696 offset += len;
1697 line += len;
1698 size -= len;
1702 * If something was removed (i.e. we have old-lines) it cannot
1703 * be creation, and if something was added it cannot be
1704 * deletion. However, the reverse is not true; --unified=0
1705 * patches that only add are not necessarily creation even
1706 * though they do not have any old lines, and ones that only
1707 * delete are not necessarily deletion.
1709 * Unfortunately, a real creation/deletion patch do _not_ have
1710 * any context line by definition, so we cannot safely tell it
1711 * apart with --unified=0 insanity. At least if the patch has
1712 * more than one hunk it is not creation or deletion.
1714 if (patch->is_new < 0 &&
1715 (oldlines || (patch->fragments && patch->fragments->next)))
1716 patch->is_new = 0;
1717 if (patch->is_delete < 0 &&
1718 (newlines || (patch->fragments && patch->fragments->next)))
1719 patch->is_delete = 0;
1721 if (0 < patch->is_new && oldlines)
1722 die(_("new file %s depends on old contents"), patch->new_name);
1723 if (0 < patch->is_delete && newlines)
1724 die(_("deleted file %s still has contents"), patch->old_name);
1725 if (!patch->is_delete && !newlines && context)
1726 fprintf_ln(stderr,
1727 _("** warning: "
1728 "file %s becomes empty but is not deleted"),
1729 patch->new_name);
1731 return offset;
1734 static inline int metadata_changes(struct patch *patch)
1736 return patch->is_rename > 0 ||
1737 patch->is_copy > 0 ||
1738 patch->is_new > 0 ||
1739 patch->is_delete ||
1740 (patch->old_mode && patch->new_mode &&
1741 patch->old_mode != patch->new_mode);
1744 static char *inflate_it(const void *data, unsigned long size,
1745 unsigned long inflated_size)
1747 git_zstream stream;
1748 void *out;
1749 int st;
1751 memset(&stream, 0, sizeof(stream));
1753 stream.next_in = (unsigned char *)data;
1754 stream.avail_in = size;
1755 stream.next_out = out = xmalloc(inflated_size);
1756 stream.avail_out = inflated_size;
1757 git_inflate_init(&stream);
1758 st = git_inflate(&stream, Z_FINISH);
1759 git_inflate_end(&stream);
1760 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1761 free(out);
1762 return NULL;
1764 return out;
1768 * Read a binary hunk and return a new fragment; fragment->patch
1769 * points at an allocated memory that the caller must free, so
1770 * it is marked as "->free_patch = 1".
1772 static struct fragment *parse_binary_hunk(char **buf_p,
1773 unsigned long *sz_p,
1774 int *status_p,
1775 int *used_p)
1778 * Expect a line that begins with binary patch method ("literal"
1779 * or "delta"), followed by the length of data before deflating.
1780 * a sequence of 'length-byte' followed by base-85 encoded data
1781 * should follow, terminated by a newline.
1783 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1784 * and we would limit the patch line to 66 characters,
1785 * so one line can fit up to 13 groups that would decode
1786 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1787 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1789 int llen, used;
1790 unsigned long size = *sz_p;
1791 char *buffer = *buf_p;
1792 int patch_method;
1793 unsigned long origlen;
1794 char *data = NULL;
1795 int hunk_size = 0;
1796 struct fragment *frag;
1798 llen = linelen(buffer, size);
1799 used = llen;
1801 *status_p = 0;
1803 if (starts_with(buffer, "delta ")) {
1804 patch_method = BINARY_DELTA_DEFLATED;
1805 origlen = strtoul(buffer + 6, NULL, 10);
1807 else if (starts_with(buffer, "literal ")) {
1808 patch_method = BINARY_LITERAL_DEFLATED;
1809 origlen = strtoul(buffer + 8, NULL, 10);
1811 else
1812 return NULL;
1814 linenr++;
1815 buffer += llen;
1816 while (1) {
1817 int byte_length, max_byte_length, newsize;
1818 llen = linelen(buffer, size);
1819 used += llen;
1820 linenr++;
1821 if (llen == 1) {
1822 /* consume the blank line */
1823 buffer++;
1824 size--;
1825 break;
1828 * Minimum line is "A00000\n" which is 7-byte long,
1829 * and the line length must be multiple of 5 plus 2.
1831 if ((llen < 7) || (llen-2) % 5)
1832 goto corrupt;
1833 max_byte_length = (llen - 2) / 5 * 4;
1834 byte_length = *buffer;
1835 if ('A' <= byte_length && byte_length <= 'Z')
1836 byte_length = byte_length - 'A' + 1;
1837 else if ('a' <= byte_length && byte_length <= 'z')
1838 byte_length = byte_length - 'a' + 27;
1839 else
1840 goto corrupt;
1841 /* if the input length was not multiple of 4, we would
1842 * have filler at the end but the filler should never
1843 * exceed 3 bytes
1845 if (max_byte_length < byte_length ||
1846 byte_length <= max_byte_length - 4)
1847 goto corrupt;
1848 newsize = hunk_size + byte_length;
1849 data = xrealloc(data, newsize);
1850 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1851 goto corrupt;
1852 hunk_size = newsize;
1853 buffer += llen;
1854 size -= llen;
1857 frag = xcalloc(1, sizeof(*frag));
1858 frag->patch = inflate_it(data, hunk_size, origlen);
1859 frag->free_patch = 1;
1860 if (!frag->patch)
1861 goto corrupt;
1862 free(data);
1863 frag->size = origlen;
1864 *buf_p = buffer;
1865 *sz_p = size;
1866 *used_p = used;
1867 frag->binary_patch_method = patch_method;
1868 return frag;
1870 corrupt:
1871 free(data);
1872 *status_p = -1;
1873 error(_("corrupt binary patch at line %d: %.*s"),
1874 linenr-1, llen-1, buffer);
1875 return NULL;
1878 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1881 * We have read "GIT binary patch\n"; what follows is a line
1882 * that says the patch method (currently, either "literal" or
1883 * "delta") and the length of data before deflating; a
1884 * sequence of 'length-byte' followed by base-85 encoded data
1885 * follows.
1887 * When a binary patch is reversible, there is another binary
1888 * hunk in the same format, starting with patch method (either
1889 * "literal" or "delta") with the length of data, and a sequence
1890 * of length-byte + base-85 encoded data, terminated with another
1891 * empty line. This data, when applied to the postimage, produces
1892 * the preimage.
1894 struct fragment *forward;
1895 struct fragment *reverse;
1896 int status;
1897 int used, used_1;
1899 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1900 if (!forward && !status)
1901 /* there has to be one hunk (forward hunk) */
1902 return error(_("unrecognized binary patch at line %d"), linenr-1);
1903 if (status)
1904 /* otherwise we already gave an error message */
1905 return status;
1907 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1908 if (reverse)
1909 used += used_1;
1910 else if (status) {
1912 * Not having reverse hunk is not an error, but having
1913 * a corrupt reverse hunk is.
1915 free((void*) forward->patch);
1916 free(forward);
1917 return status;
1919 forward->next = reverse;
1920 patch->fragments = forward;
1921 patch->is_binary = 1;
1922 return used;
1926 * Read the patch text in "buffer" that extends for "size" bytes; stop
1927 * reading after seeing a single patch (i.e. changes to a single file).
1928 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1929 * Return the number of bytes consumed, so that the caller can call us
1930 * again for the next patch.
1932 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1934 int hdrsize, patchsize;
1935 int offset = find_header(buffer, size, &hdrsize, patch);
1937 if (offset < 0)
1938 return offset;
1940 patch->ws_rule = whitespace_rule(patch->new_name
1941 ? patch->new_name
1942 : patch->old_name);
1944 patchsize = parse_single_patch(buffer + offset + hdrsize,
1945 size - offset - hdrsize, patch);
1947 if (!patchsize) {
1948 static const char git_binary[] = "GIT binary patch\n";
1949 int hd = hdrsize + offset;
1950 unsigned long llen = linelen(buffer + hd, size - hd);
1952 if (llen == sizeof(git_binary) - 1 &&
1953 !memcmp(git_binary, buffer + hd, llen)) {
1954 int used;
1955 linenr++;
1956 used = parse_binary(buffer + hd + llen,
1957 size - hd - llen, patch);
1958 if (used)
1959 patchsize = used + llen;
1960 else
1961 patchsize = 0;
1963 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1964 static const char *binhdr[] = {
1965 "Binary files ",
1966 "Files ",
1967 NULL,
1969 int i;
1970 for (i = 0; binhdr[i]; i++) {
1971 int len = strlen(binhdr[i]);
1972 if (len < size - hd &&
1973 !memcmp(binhdr[i], buffer + hd, len)) {
1974 linenr++;
1975 patch->is_binary = 1;
1976 patchsize = llen;
1977 break;
1982 /* Empty patch cannot be applied if it is a text patch
1983 * without metadata change. A binary patch appears
1984 * empty to us here.
1986 if ((apply || check) &&
1987 (!patch->is_binary && !metadata_changes(patch)))
1988 die(_("patch with only garbage at line %d"), linenr);
1991 return offset + hdrsize + patchsize;
1994 #define swap(a,b) myswap((a),(b),sizeof(a))
1996 #define myswap(a, b, size) do { \
1997 unsigned char mytmp[size]; \
1998 memcpy(mytmp, &a, size); \
1999 memcpy(&a, &b, size); \
2000 memcpy(&b, mytmp, size); \
2001 } while (0)
2003 static void reverse_patches(struct patch *p)
2005 for (; p; p = p->next) {
2006 struct fragment *frag = p->fragments;
2008 swap(p->new_name, p->old_name);
2009 swap(p->new_mode, p->old_mode);
2010 swap(p->is_new, p->is_delete);
2011 swap(p->lines_added, p->lines_deleted);
2012 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2014 for (; frag; frag = frag->next) {
2015 swap(frag->newpos, frag->oldpos);
2016 swap(frag->newlines, frag->oldlines);
2021 static const char pluses[] =
2022 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2023 static const char minuses[]=
2024 "----------------------------------------------------------------------";
2026 static void show_stats(struct patch *patch)
2028 struct strbuf qname = STRBUF_INIT;
2029 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2030 int max, add, del;
2032 quote_c_style(cp, &qname, NULL, 0);
2035 * "scale" the filename
2037 max = max_len;
2038 if (max > 50)
2039 max = 50;
2041 if (qname.len > max) {
2042 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2043 if (!cp)
2044 cp = qname.buf + qname.len + 3 - max;
2045 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2048 if (patch->is_binary) {
2049 printf(" %-*s | Bin\n", max, qname.buf);
2050 strbuf_release(&qname);
2051 return;
2054 printf(" %-*s |", max, qname.buf);
2055 strbuf_release(&qname);
2058 * scale the add/delete
2060 max = max + max_change > 70 ? 70 - max : max_change;
2061 add = patch->lines_added;
2062 del = patch->lines_deleted;
2064 if (max_change > 0) {
2065 int total = ((add + del) * max + max_change / 2) / max_change;
2066 add = (add * max + max_change / 2) / max_change;
2067 del = total - add;
2069 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2070 add, pluses, del, minuses);
2073 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2075 switch (st->st_mode & S_IFMT) {
2076 case S_IFLNK:
2077 if (strbuf_readlink(buf, path, st->st_size) < 0)
2078 return error(_("unable to read symlink %s"), path);
2079 return 0;
2080 case S_IFREG:
2081 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2082 return error(_("unable to open or read %s"), path);
2083 convert_to_git(path, buf->buf, buf->len, buf, 0);
2084 return 0;
2085 default:
2086 return -1;
2091 * Update the preimage, and the common lines in postimage,
2092 * from buffer buf of length len. If postlen is 0 the postimage
2093 * is updated in place, otherwise it's updated on a new buffer
2094 * of length postlen
2097 static void update_pre_post_images(struct image *preimage,
2098 struct image *postimage,
2099 char *buf,
2100 size_t len, size_t postlen)
2102 int i, ctx, reduced;
2103 char *new, *old, *fixed;
2104 struct image fixed_preimage;
2107 * Update the preimage with whitespace fixes. Note that we
2108 * are not losing preimage->buf -- apply_one_fragment() will
2109 * free "oldlines".
2111 prepare_image(&fixed_preimage, buf, len, 1);
2112 assert(postlen
2113 ? fixed_preimage.nr == preimage->nr
2114 : fixed_preimage.nr <= preimage->nr);
2115 for (i = 0; i < fixed_preimage.nr; i++)
2116 fixed_preimage.line[i].flag = preimage->line[i].flag;
2117 free(preimage->line_allocated);
2118 *preimage = fixed_preimage;
2121 * Adjust the common context lines in postimage. This can be
2122 * done in-place when we are shrinking it with whitespace
2123 * fixing, but needs a new buffer when ignoring whitespace or
2124 * expanding leading tabs to spaces.
2126 * We trust the caller to tell us if the update can be done
2127 * in place (postlen==0) or not.
2129 old = postimage->buf;
2130 if (postlen)
2131 new = postimage->buf = xmalloc(postlen);
2132 else
2133 new = old;
2134 fixed = preimage->buf;
2136 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2137 size_t len = postimage->line[i].len;
2138 if (!(postimage->line[i].flag & LINE_COMMON)) {
2139 /* an added line -- no counterparts in preimage */
2140 memmove(new, old, len);
2141 old += len;
2142 new += len;
2143 continue;
2146 /* a common context -- skip it in the original postimage */
2147 old += len;
2149 /* and find the corresponding one in the fixed preimage */
2150 while (ctx < preimage->nr &&
2151 !(preimage->line[ctx].flag & LINE_COMMON)) {
2152 fixed += preimage->line[ctx].len;
2153 ctx++;
2157 * preimage is expected to run out, if the caller
2158 * fixed addition of trailing blank lines.
2160 if (preimage->nr <= ctx) {
2161 reduced++;
2162 continue;
2165 /* and copy it in, while fixing the line length */
2166 len = preimage->line[ctx].len;
2167 memcpy(new, fixed, len);
2168 new += len;
2169 fixed += len;
2170 postimage->line[i].len = len;
2171 ctx++;
2174 /* Fix the length of the whole thing */
2175 postimage->len = new - postimage->buf;
2176 postimage->nr -= reduced;
2179 static int match_fragment(struct image *img,
2180 struct image *preimage,
2181 struct image *postimage,
2182 unsigned long try,
2183 int try_lno,
2184 unsigned ws_rule,
2185 int match_beginning, int match_end)
2187 int i;
2188 char *fixed_buf, *buf, *orig, *target;
2189 struct strbuf fixed;
2190 size_t fixed_len, postlen;
2191 int preimage_limit;
2193 if (preimage->nr + try_lno <= img->nr) {
2195 * The hunk falls within the boundaries of img.
2197 preimage_limit = preimage->nr;
2198 if (match_end && (preimage->nr + try_lno != img->nr))
2199 return 0;
2200 } else if (ws_error_action == correct_ws_error &&
2201 (ws_rule & WS_BLANK_AT_EOF)) {
2203 * This hunk extends beyond the end of img, and we are
2204 * removing blank lines at the end of the file. This
2205 * many lines from the beginning of the preimage must
2206 * match with img, and the remainder of the preimage
2207 * must be blank.
2209 preimage_limit = img->nr - try_lno;
2210 } else {
2212 * The hunk extends beyond the end of the img and
2213 * we are not removing blanks at the end, so we
2214 * should reject the hunk at this position.
2216 return 0;
2219 if (match_beginning && try_lno)
2220 return 0;
2222 /* Quick hash check */
2223 for (i = 0; i < preimage_limit; i++)
2224 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2225 (preimage->line[i].hash != img->line[try_lno + i].hash))
2226 return 0;
2228 if (preimage_limit == preimage->nr) {
2230 * Do we have an exact match? If we were told to match
2231 * at the end, size must be exactly at try+fragsize,
2232 * otherwise try+fragsize must be still within the preimage,
2233 * and either case, the old piece should match the preimage
2234 * exactly.
2236 if ((match_end
2237 ? (try + preimage->len == img->len)
2238 : (try + preimage->len <= img->len)) &&
2239 !memcmp(img->buf + try, preimage->buf, preimage->len))
2240 return 1;
2241 } else {
2243 * The preimage extends beyond the end of img, so
2244 * there cannot be an exact match.
2246 * There must be one non-blank context line that match
2247 * a line before the end of img.
2249 char *buf_end;
2251 buf = preimage->buf;
2252 buf_end = buf;
2253 for (i = 0; i < preimage_limit; i++)
2254 buf_end += preimage->line[i].len;
2256 for ( ; buf < buf_end; buf++)
2257 if (!isspace(*buf))
2258 break;
2259 if (buf == buf_end)
2260 return 0;
2264 * No exact match. If we are ignoring whitespace, run a line-by-line
2265 * fuzzy matching. We collect all the line length information because
2266 * we need it to adjust whitespace if we match.
2268 if (ws_ignore_action == ignore_ws_change) {
2269 size_t imgoff = 0;
2270 size_t preoff = 0;
2271 size_t postlen = postimage->len;
2272 size_t extra_chars;
2273 char *preimage_eof;
2274 char *preimage_end;
2275 for (i = 0; i < preimage_limit; i++) {
2276 size_t prelen = preimage->line[i].len;
2277 size_t imglen = img->line[try_lno+i].len;
2279 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2280 preimage->buf + preoff, prelen))
2281 return 0;
2282 if (preimage->line[i].flag & LINE_COMMON)
2283 postlen += imglen - prelen;
2284 imgoff += imglen;
2285 preoff += prelen;
2289 * Ok, the preimage matches with whitespace fuzz.
2291 * imgoff now holds the true length of the target that
2292 * matches the preimage before the end of the file.
2294 * Count the number of characters in the preimage that fall
2295 * beyond the end of the file and make sure that all of them
2296 * are whitespace characters. (This can only happen if
2297 * we are removing blank lines at the end of the file.)
2299 buf = preimage_eof = preimage->buf + preoff;
2300 for ( ; i < preimage->nr; i++)
2301 preoff += preimage->line[i].len;
2302 preimage_end = preimage->buf + preoff;
2303 for ( ; buf < preimage_end; buf++)
2304 if (!isspace(*buf))
2305 return 0;
2308 * Update the preimage and the common postimage context
2309 * lines to use the same whitespace as the target.
2310 * If whitespace is missing in the target (i.e.
2311 * if the preimage extends beyond the end of the file),
2312 * use the whitespace from the preimage.
2314 extra_chars = preimage_end - preimage_eof;
2315 strbuf_init(&fixed, imgoff + extra_chars);
2316 strbuf_add(&fixed, img->buf + try, imgoff);
2317 strbuf_add(&fixed, preimage_eof, extra_chars);
2318 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2319 update_pre_post_images(preimage, postimage,
2320 fixed_buf, fixed_len, postlen);
2321 return 1;
2324 if (ws_error_action != correct_ws_error)
2325 return 0;
2328 * The hunk does not apply byte-by-byte, but the hash says
2329 * it might with whitespace fuzz. We haven't been asked to
2330 * ignore whitespace, we were asked to correct whitespace
2331 * errors, so let's try matching after whitespace correction.
2333 * The preimage may extend beyond the end of the file,
2334 * but in this loop we will only handle the part of the
2335 * preimage that falls within the file.
2337 strbuf_init(&fixed, preimage->len + 1);
2338 orig = preimage->buf;
2339 target = img->buf + try;
2340 postlen = 0;
2341 for (i = 0; i < preimage_limit; i++) {
2342 size_t oldlen = preimage->line[i].len;
2343 size_t tgtlen = img->line[try_lno + i].len;
2344 size_t fixstart = fixed.len;
2345 struct strbuf tgtfix;
2346 int match;
2348 /* Try fixing the line in the preimage */
2349 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2351 /* Try fixing the line in the target */
2352 strbuf_init(&tgtfix, tgtlen);
2353 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2356 * If they match, either the preimage was based on
2357 * a version before our tree fixed whitespace breakage,
2358 * or we are lacking a whitespace-fix patch the tree
2359 * the preimage was based on already had (i.e. target
2360 * has whitespace breakage, the preimage doesn't).
2361 * In either case, we are fixing the whitespace breakages
2362 * so we might as well take the fix together with their
2363 * real change.
2365 match = (tgtfix.len == fixed.len - fixstart &&
2366 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2367 fixed.len - fixstart));
2368 postlen += tgtfix.len;
2370 strbuf_release(&tgtfix);
2371 if (!match)
2372 goto unmatch_exit;
2374 orig += oldlen;
2375 target += tgtlen;
2380 * Now handle the lines in the preimage that falls beyond the
2381 * end of the file (if any). They will only match if they are
2382 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2383 * false).
2385 for ( ; i < preimage->nr; i++) {
2386 size_t fixstart = fixed.len; /* start of the fixed preimage */
2387 size_t oldlen = preimage->line[i].len;
2388 int j;
2390 /* Try fixing the line in the preimage */
2391 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2393 for (j = fixstart; j < fixed.len; j++)
2394 if (!isspace(fixed.buf[j]))
2395 goto unmatch_exit;
2397 orig += oldlen;
2401 * Yes, the preimage is based on an older version that still
2402 * has whitespace breakages unfixed, and fixing them makes the
2403 * hunk match. Update the context lines in the postimage.
2405 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2406 if (postlen < postimage->len)
2407 postlen = 0;
2408 update_pre_post_images(preimage, postimage,
2409 fixed_buf, fixed_len, postlen);
2410 return 1;
2412 unmatch_exit:
2413 strbuf_release(&fixed);
2414 return 0;
2417 static int find_pos(struct image *img,
2418 struct image *preimage,
2419 struct image *postimage,
2420 int line,
2421 unsigned ws_rule,
2422 int match_beginning, int match_end)
2424 int i;
2425 unsigned long backwards, forwards, try;
2426 int backwards_lno, forwards_lno, try_lno;
2429 * If match_beginning or match_end is specified, there is no
2430 * point starting from a wrong line that will never match and
2431 * wander around and wait for a match at the specified end.
2433 if (match_beginning)
2434 line = 0;
2435 else if (match_end)
2436 line = img->nr - preimage->nr;
2439 * Because the comparison is unsigned, the following test
2440 * will also take care of a negative line number that can
2441 * result when match_end and preimage is larger than the target.
2443 if ((size_t) line > img->nr)
2444 line = img->nr;
2446 try = 0;
2447 for (i = 0; i < line; i++)
2448 try += img->line[i].len;
2451 * There's probably some smart way to do this, but I'll leave
2452 * that to the smart and beautiful people. I'm simple and stupid.
2454 backwards = try;
2455 backwards_lno = line;
2456 forwards = try;
2457 forwards_lno = line;
2458 try_lno = line;
2460 for (i = 0; ; i++) {
2461 if (match_fragment(img, preimage, postimage,
2462 try, try_lno, ws_rule,
2463 match_beginning, match_end))
2464 return try_lno;
2466 again:
2467 if (backwards_lno == 0 && forwards_lno == img->nr)
2468 break;
2470 if (i & 1) {
2471 if (backwards_lno == 0) {
2472 i++;
2473 goto again;
2475 backwards_lno--;
2476 backwards -= img->line[backwards_lno].len;
2477 try = backwards;
2478 try_lno = backwards_lno;
2479 } else {
2480 if (forwards_lno == img->nr) {
2481 i++;
2482 goto again;
2484 forwards += img->line[forwards_lno].len;
2485 forwards_lno++;
2486 try = forwards;
2487 try_lno = forwards_lno;
2491 return -1;
2494 static void remove_first_line(struct image *img)
2496 img->buf += img->line[0].len;
2497 img->len -= img->line[0].len;
2498 img->line++;
2499 img->nr--;
2502 static void remove_last_line(struct image *img)
2504 img->len -= img->line[--img->nr].len;
2508 * The change from "preimage" and "postimage" has been found to
2509 * apply at applied_pos (counts in line numbers) in "img".
2510 * Update "img" to remove "preimage" and replace it with "postimage".
2512 static void update_image(struct image *img,
2513 int applied_pos,
2514 struct image *preimage,
2515 struct image *postimage)
2518 * remove the copy of preimage at offset in img
2519 * and replace it with postimage
2521 int i, nr;
2522 size_t remove_count, insert_count, applied_at = 0;
2523 char *result;
2524 int preimage_limit;
2527 * If we are removing blank lines at the end of img,
2528 * the preimage may extend beyond the end.
2529 * If that is the case, we must be careful only to
2530 * remove the part of the preimage that falls within
2531 * the boundaries of img. Initialize preimage_limit
2532 * to the number of lines in the preimage that falls
2533 * within the boundaries.
2535 preimage_limit = preimage->nr;
2536 if (preimage_limit > img->nr - applied_pos)
2537 preimage_limit = img->nr - applied_pos;
2539 for (i = 0; i < applied_pos; i++)
2540 applied_at += img->line[i].len;
2542 remove_count = 0;
2543 for (i = 0; i < preimage_limit; i++)
2544 remove_count += img->line[applied_pos + i].len;
2545 insert_count = postimage->len;
2547 /* Adjust the contents */
2548 result = xmalloc(img->len + insert_count - remove_count + 1);
2549 memcpy(result, img->buf, applied_at);
2550 memcpy(result + applied_at, postimage->buf, postimage->len);
2551 memcpy(result + applied_at + postimage->len,
2552 img->buf + (applied_at + remove_count),
2553 img->len - (applied_at + remove_count));
2554 free(img->buf);
2555 img->buf = result;
2556 img->len += insert_count - remove_count;
2557 result[img->len] = '\0';
2559 /* Adjust the line table */
2560 nr = img->nr + postimage->nr - preimage_limit;
2561 if (preimage_limit < postimage->nr) {
2563 * NOTE: this knows that we never call remove_first_line()
2564 * on anything other than pre/post image.
2566 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2567 img->line_allocated = img->line;
2569 if (preimage_limit != postimage->nr)
2570 memmove(img->line + applied_pos + postimage->nr,
2571 img->line + applied_pos + preimage_limit,
2572 (img->nr - (applied_pos + preimage_limit)) *
2573 sizeof(*img->line));
2574 memcpy(img->line + applied_pos,
2575 postimage->line,
2576 postimage->nr * sizeof(*img->line));
2577 if (!allow_overlap)
2578 for (i = 0; i < postimage->nr; i++)
2579 img->line[applied_pos + i].flag |= LINE_PATCHED;
2580 img->nr = nr;
2584 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2585 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2586 * replace the part of "img" with "postimage" text.
2588 static int apply_one_fragment(struct image *img, struct fragment *frag,
2589 int inaccurate_eof, unsigned ws_rule,
2590 int nth_fragment)
2592 int match_beginning, match_end;
2593 const char *patch = frag->patch;
2594 int size = frag->size;
2595 char *old, *oldlines;
2596 struct strbuf newlines;
2597 int new_blank_lines_at_end = 0;
2598 int found_new_blank_lines_at_end = 0;
2599 int hunk_linenr = frag->linenr;
2600 unsigned long leading, trailing;
2601 int pos, applied_pos;
2602 struct image preimage;
2603 struct image postimage;
2605 memset(&preimage, 0, sizeof(preimage));
2606 memset(&postimage, 0, sizeof(postimage));
2607 oldlines = xmalloc(size);
2608 strbuf_init(&newlines, size);
2610 old = oldlines;
2611 while (size > 0) {
2612 char first;
2613 int len = linelen(patch, size);
2614 int plen;
2615 int added_blank_line = 0;
2616 int is_blank_context = 0;
2617 size_t start;
2619 if (!len)
2620 break;
2623 * "plen" is how much of the line we should use for
2624 * the actual patch data. Normally we just remove the
2625 * first character on the line, but if the line is
2626 * followed by "\ No newline", then we also remove the
2627 * last one (which is the newline, of course).
2629 plen = len - 1;
2630 if (len < size && patch[len] == '\\')
2631 plen--;
2632 first = *patch;
2633 if (apply_in_reverse) {
2634 if (first == '-')
2635 first = '+';
2636 else if (first == '+')
2637 first = '-';
2640 switch (first) {
2641 case '\n':
2642 /* Newer GNU diff, empty context line */
2643 if (plen < 0)
2644 /* ... followed by '\No newline'; nothing */
2645 break;
2646 *old++ = '\n';
2647 strbuf_addch(&newlines, '\n');
2648 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2649 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2650 is_blank_context = 1;
2651 break;
2652 case ' ':
2653 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2654 ws_blank_line(patch + 1, plen, ws_rule))
2655 is_blank_context = 1;
2656 case '-':
2657 memcpy(old, patch + 1, plen);
2658 add_line_info(&preimage, old, plen,
2659 (first == ' ' ? LINE_COMMON : 0));
2660 old += plen;
2661 if (first == '-')
2662 break;
2663 /* Fall-through for ' ' */
2664 case '+':
2665 /* --no-add does not add new lines */
2666 if (first == '+' && no_add)
2667 break;
2669 start = newlines.len;
2670 if (first != '+' ||
2671 !whitespace_error ||
2672 ws_error_action != correct_ws_error) {
2673 strbuf_add(&newlines, patch + 1, plen);
2675 else {
2676 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2678 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2679 (first == '+' ? 0 : LINE_COMMON));
2680 if (first == '+' &&
2681 (ws_rule & WS_BLANK_AT_EOF) &&
2682 ws_blank_line(patch + 1, plen, ws_rule))
2683 added_blank_line = 1;
2684 break;
2685 case '@': case '\\':
2686 /* Ignore it, we already handled it */
2687 break;
2688 default:
2689 if (apply_verbosely)
2690 error(_("invalid start of line: '%c'"), first);
2691 return -1;
2693 if (added_blank_line) {
2694 if (!new_blank_lines_at_end)
2695 found_new_blank_lines_at_end = hunk_linenr;
2696 new_blank_lines_at_end++;
2698 else if (is_blank_context)
2700 else
2701 new_blank_lines_at_end = 0;
2702 patch += len;
2703 size -= len;
2704 hunk_linenr++;
2706 if (inaccurate_eof &&
2707 old > oldlines && old[-1] == '\n' &&
2708 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2709 old--;
2710 strbuf_setlen(&newlines, newlines.len - 1);
2713 leading = frag->leading;
2714 trailing = frag->trailing;
2717 * A hunk to change lines at the beginning would begin with
2718 * @@ -1,L +N,M @@
2719 * but we need to be careful. -U0 that inserts before the second
2720 * line also has this pattern.
2722 * And a hunk to add to an empty file would begin with
2723 * @@ -0,0 +N,M @@
2725 * In other words, a hunk that is (frag->oldpos <= 1) with or
2726 * without leading context must match at the beginning.
2728 match_beginning = (!frag->oldpos ||
2729 (frag->oldpos == 1 && !unidiff_zero));
2732 * A hunk without trailing lines must match at the end.
2733 * However, we simply cannot tell if a hunk must match end
2734 * from the lack of trailing lines if the patch was generated
2735 * with unidiff without any context.
2737 match_end = !unidiff_zero && !trailing;
2739 pos = frag->newpos ? (frag->newpos - 1) : 0;
2740 preimage.buf = oldlines;
2741 preimage.len = old - oldlines;
2742 postimage.buf = newlines.buf;
2743 postimage.len = newlines.len;
2744 preimage.line = preimage.line_allocated;
2745 postimage.line = postimage.line_allocated;
2747 for (;;) {
2749 applied_pos = find_pos(img, &preimage, &postimage, pos,
2750 ws_rule, match_beginning, match_end);
2752 if (applied_pos >= 0)
2753 break;
2755 /* Am I at my context limits? */
2756 if ((leading <= p_context) && (trailing <= p_context))
2757 break;
2758 if (match_beginning || match_end) {
2759 match_beginning = match_end = 0;
2760 continue;
2764 * Reduce the number of context lines; reduce both
2765 * leading and trailing if they are equal otherwise
2766 * just reduce the larger context.
2768 if (leading >= trailing) {
2769 remove_first_line(&preimage);
2770 remove_first_line(&postimage);
2771 pos--;
2772 leading--;
2774 if (trailing > leading) {
2775 remove_last_line(&preimage);
2776 remove_last_line(&postimage);
2777 trailing--;
2781 if (applied_pos >= 0) {
2782 if (new_blank_lines_at_end &&
2783 preimage.nr + applied_pos >= img->nr &&
2784 (ws_rule & WS_BLANK_AT_EOF) &&
2785 ws_error_action != nowarn_ws_error) {
2786 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2787 found_new_blank_lines_at_end);
2788 if (ws_error_action == correct_ws_error) {
2789 while (new_blank_lines_at_end--)
2790 remove_last_line(&postimage);
2793 * We would want to prevent write_out_results()
2794 * from taking place in apply_patch() that follows
2795 * the callchain led us here, which is:
2796 * apply_patch->check_patch_list->check_patch->
2797 * apply_data->apply_fragments->apply_one_fragment
2799 if (ws_error_action == die_on_ws_error)
2800 apply = 0;
2803 if (apply_verbosely && applied_pos != pos) {
2804 int offset = applied_pos - pos;
2805 if (apply_in_reverse)
2806 offset = 0 - offset;
2807 fprintf_ln(stderr,
2808 Q_("Hunk #%d succeeded at %d (offset %d line).",
2809 "Hunk #%d succeeded at %d (offset %d lines).",
2810 offset),
2811 nth_fragment, applied_pos + 1, offset);
2815 * Warn if it was necessary to reduce the number
2816 * of context lines.
2818 if ((leading != frag->leading) ||
2819 (trailing != frag->trailing))
2820 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2821 " to apply fragment at %d"),
2822 leading, trailing, applied_pos+1);
2823 update_image(img, applied_pos, &preimage, &postimage);
2824 } else {
2825 if (apply_verbosely)
2826 error(_("while searching for:\n%.*s"),
2827 (int)(old - oldlines), oldlines);
2830 free(oldlines);
2831 strbuf_release(&newlines);
2832 free(preimage.line_allocated);
2833 free(postimage.line_allocated);
2835 return (applied_pos < 0);
2838 static int apply_binary_fragment(struct image *img, struct patch *patch)
2840 struct fragment *fragment = patch->fragments;
2841 unsigned long len;
2842 void *dst;
2844 if (!fragment)
2845 return error(_("missing binary patch data for '%s'"),
2846 patch->new_name ?
2847 patch->new_name :
2848 patch->old_name);
2850 /* Binary patch is irreversible without the optional second hunk */
2851 if (apply_in_reverse) {
2852 if (!fragment->next)
2853 return error("cannot reverse-apply a binary patch "
2854 "without the reverse hunk to '%s'",
2855 patch->new_name
2856 ? patch->new_name : patch->old_name);
2857 fragment = fragment->next;
2859 switch (fragment->binary_patch_method) {
2860 case BINARY_DELTA_DEFLATED:
2861 dst = patch_delta(img->buf, img->len, fragment->patch,
2862 fragment->size, &len);
2863 if (!dst)
2864 return -1;
2865 clear_image(img);
2866 img->buf = dst;
2867 img->len = len;
2868 return 0;
2869 case BINARY_LITERAL_DEFLATED:
2870 clear_image(img);
2871 img->len = fragment->size;
2872 img->buf = xmemdupz(fragment->patch, img->len);
2873 return 0;
2875 return -1;
2879 * Replace "img" with the result of applying the binary patch.
2880 * The binary patch data itself in patch->fragment is still kept
2881 * but the preimage prepared by the caller in "img" is freed here
2882 * or in the helper function apply_binary_fragment() this calls.
2884 static int apply_binary(struct image *img, struct patch *patch)
2886 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2887 unsigned char sha1[20];
2890 * For safety, we require patch index line to contain
2891 * full 40-byte textual SHA1 for old and new, at least for now.
2893 if (strlen(patch->old_sha1_prefix) != 40 ||
2894 strlen(patch->new_sha1_prefix) != 40 ||
2895 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2896 get_sha1_hex(patch->new_sha1_prefix, sha1))
2897 return error("cannot apply binary patch to '%s' "
2898 "without full index line", name);
2900 if (patch->old_name) {
2902 * See if the old one matches what the patch
2903 * applies to.
2905 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2906 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2907 return error("the patch applies to '%s' (%s), "
2908 "which does not match the "
2909 "current contents.",
2910 name, sha1_to_hex(sha1));
2912 else {
2913 /* Otherwise, the old one must be empty. */
2914 if (img->len)
2915 return error("the patch applies to an empty "
2916 "'%s' but it is not empty", name);
2919 get_sha1_hex(patch->new_sha1_prefix, sha1);
2920 if (is_null_sha1(sha1)) {
2921 clear_image(img);
2922 return 0; /* deletion patch */
2925 if (has_sha1_file(sha1)) {
2926 /* We already have the postimage */
2927 enum object_type type;
2928 unsigned long size;
2929 char *result;
2931 result = read_sha1_file(sha1, &type, &size);
2932 if (!result)
2933 return error("the necessary postimage %s for "
2934 "'%s' cannot be read",
2935 patch->new_sha1_prefix, name);
2936 clear_image(img);
2937 img->buf = result;
2938 img->len = size;
2939 } else {
2941 * We have verified buf matches the preimage;
2942 * apply the patch data to it, which is stored
2943 * in the patch->fragments->{patch,size}.
2945 if (apply_binary_fragment(img, patch))
2946 return error(_("binary patch does not apply to '%s'"),
2947 name);
2949 /* verify that the result matches */
2950 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2951 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2952 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2953 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2956 return 0;
2959 static int apply_fragments(struct image *img, struct patch *patch)
2961 struct fragment *frag = patch->fragments;
2962 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2963 unsigned ws_rule = patch->ws_rule;
2964 unsigned inaccurate_eof = patch->inaccurate_eof;
2965 int nth = 0;
2967 if (patch->is_binary)
2968 return apply_binary(img, patch);
2970 while (frag) {
2971 nth++;
2972 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2973 error(_("patch failed: %s:%ld"), name, frag->oldpos);
2974 if (!apply_with_reject)
2975 return -1;
2976 frag->rejected = 1;
2978 frag = frag->next;
2980 return 0;
2983 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2985 if (S_ISGITLINK(mode)) {
2986 strbuf_grow(buf, 100);
2987 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2988 } else {
2989 enum object_type type;
2990 unsigned long sz;
2991 char *result;
2993 result = read_sha1_file(sha1, &type, &sz);
2994 if (!result)
2995 return -1;
2996 /* XXX read_sha1_file NUL-terminates */
2997 strbuf_attach(buf, result, sz, sz + 1);
2999 return 0;
3002 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3004 if (!ce)
3005 return 0;
3006 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3009 static struct patch *in_fn_table(const char *name)
3011 struct string_list_item *item;
3013 if (name == NULL)
3014 return NULL;
3016 item = string_list_lookup(&fn_table, name);
3017 if (item != NULL)
3018 return (struct patch *)item->util;
3020 return NULL;
3024 * item->util in the filename table records the status of the path.
3025 * Usually it points at a patch (whose result records the contents
3026 * of it after applying it), but it could be PATH_WAS_DELETED for a
3027 * path that a previously applied patch has already removed, or
3028 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3030 * The latter is needed to deal with a case where two paths A and B
3031 * are swapped by first renaming A to B and then renaming B to A;
3032 * moving A to B should not be prevented due to presence of B as we
3033 * will remove it in a later patch.
3035 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3036 #define PATH_WAS_DELETED ((struct patch *) -1)
3038 static int to_be_deleted(struct patch *patch)
3040 return patch == PATH_TO_BE_DELETED;
3043 static int was_deleted(struct patch *patch)
3045 return patch == PATH_WAS_DELETED;
3048 static void add_to_fn_table(struct patch *patch)
3050 struct string_list_item *item;
3053 * Always add new_name unless patch is a deletion
3054 * This should cover the cases for normal diffs,
3055 * file creations and copies
3057 if (patch->new_name != NULL) {
3058 item = string_list_insert(&fn_table, patch->new_name);
3059 item->util = patch;
3063 * store a failure on rename/deletion cases because
3064 * later chunks shouldn't patch old names
3066 if ((patch->new_name == NULL) || (patch->is_rename)) {
3067 item = string_list_insert(&fn_table, patch->old_name);
3068 item->util = PATH_WAS_DELETED;
3072 static void prepare_fn_table(struct patch *patch)
3075 * store information about incoming file deletion
3077 while (patch) {
3078 if ((patch->new_name == NULL) || (patch->is_rename)) {
3079 struct string_list_item *item;
3080 item = string_list_insert(&fn_table, patch->old_name);
3081 item->util = PATH_TO_BE_DELETED;
3083 patch = patch->next;
3087 static int checkout_target(struct cache_entry *ce, struct stat *st)
3089 struct checkout costate;
3091 memset(&costate, 0, sizeof(costate));
3092 costate.base_dir = "";
3093 costate.refresh_cache = 1;
3094 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3095 return error(_("cannot checkout %s"), ce->name);
3096 return 0;
3099 static struct patch *previous_patch(struct patch *patch, int *gone)
3101 struct patch *previous;
3103 *gone = 0;
3104 if (patch->is_copy || patch->is_rename)
3105 return NULL; /* "git" patches do not depend on the order */
3107 previous = in_fn_table(patch->old_name);
3108 if (!previous)
3109 return NULL;
3111 if (to_be_deleted(previous))
3112 return NULL; /* the deletion hasn't happened yet */
3114 if (was_deleted(previous))
3115 *gone = 1;
3117 return previous;
3120 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3122 if (S_ISGITLINK(ce->ce_mode)) {
3123 if (!S_ISDIR(st->st_mode))
3124 return -1;
3125 return 0;
3127 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3130 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3132 static int load_patch_target(struct strbuf *buf,
3133 const struct cache_entry *ce,
3134 struct stat *st,
3135 const char *name,
3136 unsigned expected_mode)
3138 if (cached) {
3139 if (read_file_or_gitlink(ce, buf))
3140 return error(_("read of %s failed"), name);
3141 } else if (name) {
3142 if (S_ISGITLINK(expected_mode)) {
3143 if (ce)
3144 return read_file_or_gitlink(ce, buf);
3145 else
3146 return SUBMODULE_PATCH_WITHOUT_INDEX;
3147 } else {
3148 if (read_old_data(st, name, buf))
3149 return error(_("read of %s failed"), name);
3152 return 0;
3156 * We are about to apply "patch"; populate the "image" with the
3157 * current version we have, from the working tree or from the index,
3158 * depending on the situation e.g. --cached/--index. If we are
3159 * applying a non-git patch that incrementally updates the tree,
3160 * we read from the result of a previous diff.
3162 static int load_preimage(struct image *image,
3163 struct patch *patch, struct stat *st,
3164 const struct cache_entry *ce)
3166 struct strbuf buf = STRBUF_INIT;
3167 size_t len;
3168 char *img;
3169 struct patch *previous;
3170 int status;
3172 previous = previous_patch(patch, &status);
3173 if (status)
3174 return error(_("path %s has been renamed/deleted"),
3175 patch->old_name);
3176 if (previous) {
3177 /* We have a patched copy in memory; use that. */
3178 strbuf_add(&buf, previous->result, previous->resultsize);
3179 } else {
3180 status = load_patch_target(&buf, ce, st,
3181 patch->old_name, patch->old_mode);
3182 if (status < 0)
3183 return status;
3184 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3186 * There is no way to apply subproject
3187 * patch without looking at the index.
3188 * NEEDSWORK: shouldn't this be flagged
3189 * as an error???
3191 free_fragment_list(patch->fragments);
3192 patch->fragments = NULL;
3193 } else if (status) {
3194 return error(_("read of %s failed"), patch->old_name);
3198 img = strbuf_detach(&buf, &len);
3199 prepare_image(image, img, len, !patch->is_binary);
3200 return 0;
3203 static int three_way_merge(struct image *image,
3204 char *path,
3205 const unsigned char *base,
3206 const unsigned char *ours,
3207 const unsigned char *theirs)
3209 mmfile_t base_file, our_file, their_file;
3210 mmbuffer_t result = { NULL };
3211 int status;
3213 read_mmblob(&base_file, base);
3214 read_mmblob(&our_file, ours);
3215 read_mmblob(&their_file, theirs);
3216 status = ll_merge(&result, path,
3217 &base_file, "base",
3218 &our_file, "ours",
3219 &their_file, "theirs", NULL);
3220 free(base_file.ptr);
3221 free(our_file.ptr);
3222 free(their_file.ptr);
3223 if (status < 0 || !result.ptr) {
3224 free(result.ptr);
3225 return -1;
3227 clear_image(image);
3228 image->buf = result.ptr;
3229 image->len = result.size;
3231 return status;
3235 * When directly falling back to add/add three-way merge, we read from
3236 * the current contents of the new_name. In no cases other than that
3237 * this function will be called.
3239 static int load_current(struct image *image, struct patch *patch)
3241 struct strbuf buf = STRBUF_INIT;
3242 int status, pos;
3243 size_t len;
3244 char *img;
3245 struct stat st;
3246 struct cache_entry *ce;
3247 char *name = patch->new_name;
3248 unsigned mode = patch->new_mode;
3250 if (!patch->is_new)
3251 die("BUG: patch to %s is not a creation", patch->old_name);
3253 pos = cache_name_pos(name, strlen(name));
3254 if (pos < 0)
3255 return error(_("%s: does not exist in index"), name);
3256 ce = active_cache[pos];
3257 if (lstat(name, &st)) {
3258 if (errno != ENOENT)
3259 return error(_("%s: %s"), name, strerror(errno));
3260 if (checkout_target(ce, &st))
3261 return -1;
3263 if (verify_index_match(ce, &st))
3264 return error(_("%s: does not match index"), name);
3266 status = load_patch_target(&buf, ce, &st, name, mode);
3267 if (status < 0)
3268 return status;
3269 else if (status)
3270 return -1;
3271 img = strbuf_detach(&buf, &len);
3272 prepare_image(image, img, len, !patch->is_binary);
3273 return 0;
3276 static int try_threeway(struct image *image, struct patch *patch,
3277 struct stat *st, const struct cache_entry *ce)
3279 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3280 struct strbuf buf = STRBUF_INIT;
3281 size_t len;
3282 int status;
3283 char *img;
3284 struct image tmp_image;
3286 /* No point falling back to 3-way merge in these cases */
3287 if (patch->is_delete ||
3288 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3289 return -1;
3291 /* Preimage the patch was prepared for */
3292 if (patch->is_new)
3293 write_sha1_file("", 0, blob_type, pre_sha1);
3294 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3295 read_blob_object(&buf, pre_sha1, patch->old_mode))
3296 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3298 fprintf(stderr, "Falling back to three-way merge...\n");
3300 img = strbuf_detach(&buf, &len);
3301 prepare_image(&tmp_image, img, len, 1);
3302 /* Apply the patch to get the post image */
3303 if (apply_fragments(&tmp_image, patch) < 0) {
3304 clear_image(&tmp_image);
3305 return -1;
3307 /* post_sha1[] is theirs */
3308 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3309 clear_image(&tmp_image);
3311 /* our_sha1[] is ours */
3312 if (patch->is_new) {
3313 if (load_current(&tmp_image, patch))
3314 return error("cannot read the current contents of '%s'",
3315 patch->new_name);
3316 } else {
3317 if (load_preimage(&tmp_image, patch, st, ce))
3318 return error("cannot read the current contents of '%s'",
3319 patch->old_name);
3321 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3322 clear_image(&tmp_image);
3324 /* in-core three-way merge between post and our using pre as base */
3325 status = three_way_merge(image, patch->new_name,
3326 pre_sha1, our_sha1, post_sha1);
3327 if (status < 0) {
3328 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3329 return status;
3332 if (status) {
3333 patch->conflicted_threeway = 1;
3334 if (patch->is_new)
3335 hashclr(patch->threeway_stage[0]);
3336 else
3337 hashcpy(patch->threeway_stage[0], pre_sha1);
3338 hashcpy(patch->threeway_stage[1], our_sha1);
3339 hashcpy(patch->threeway_stage[2], post_sha1);
3340 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3341 } else {
3342 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3344 return 0;
3347 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3349 struct image image;
3351 if (load_preimage(&image, patch, st, ce) < 0)
3352 return -1;
3354 if (patch->direct_to_threeway ||
3355 apply_fragments(&image, patch) < 0) {
3356 /* Note: with --reject, apply_fragments() returns 0 */
3357 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3358 return -1;
3360 patch->result = image.buf;
3361 patch->resultsize = image.len;
3362 add_to_fn_table(patch);
3363 free(image.line_allocated);
3365 if (0 < patch->is_delete && patch->resultsize)
3366 return error(_("removal patch leaves file contents"));
3368 return 0;
3372 * If "patch" that we are looking at modifies or deletes what we have,
3373 * we would want it not to lose any local modification we have, either
3374 * in the working tree or in the index.
3376 * This also decides if a non-git patch is a creation patch or a
3377 * modification to an existing empty file. We do not check the state
3378 * of the current tree for a creation patch in this function; the caller
3379 * check_patch() separately makes sure (and errors out otherwise) that
3380 * the path the patch creates does not exist in the current tree.
3382 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3384 const char *old_name = patch->old_name;
3385 struct patch *previous = NULL;
3386 int stat_ret = 0, status;
3387 unsigned st_mode = 0;
3389 if (!old_name)
3390 return 0;
3392 assert(patch->is_new <= 0);
3393 previous = previous_patch(patch, &status);
3395 if (status)
3396 return error(_("path %s has been renamed/deleted"), old_name);
3397 if (previous) {
3398 st_mode = previous->new_mode;
3399 } else if (!cached) {
3400 stat_ret = lstat(old_name, st);
3401 if (stat_ret && errno != ENOENT)
3402 return error(_("%s: %s"), old_name, strerror(errno));
3405 if (check_index && !previous) {
3406 int pos = cache_name_pos(old_name, strlen(old_name));
3407 if (pos < 0) {
3408 if (patch->is_new < 0)
3409 goto is_new;
3410 return error(_("%s: does not exist in index"), old_name);
3412 *ce = active_cache[pos];
3413 if (stat_ret < 0) {
3414 if (checkout_target(*ce, st))
3415 return -1;
3417 if (!cached && verify_index_match(*ce, st))
3418 return error(_("%s: does not match index"), old_name);
3419 if (cached)
3420 st_mode = (*ce)->ce_mode;
3421 } else if (stat_ret < 0) {
3422 if (patch->is_new < 0)
3423 goto is_new;
3424 return error(_("%s: %s"), old_name, strerror(errno));
3427 if (!cached && !previous)
3428 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3430 if (patch->is_new < 0)
3431 patch->is_new = 0;
3432 if (!patch->old_mode)
3433 patch->old_mode = st_mode;
3434 if ((st_mode ^ patch->old_mode) & S_IFMT)
3435 return error(_("%s: wrong type"), old_name);
3436 if (st_mode != patch->old_mode)
3437 warning(_("%s has type %o, expected %o"),
3438 old_name, st_mode, patch->old_mode);
3439 if (!patch->new_mode && !patch->is_delete)
3440 patch->new_mode = st_mode;
3441 return 0;
3443 is_new:
3444 patch->is_new = 1;
3445 patch->is_delete = 0;
3446 free(patch->old_name);
3447 patch->old_name = NULL;
3448 return 0;
3452 #define EXISTS_IN_INDEX 1
3453 #define EXISTS_IN_WORKTREE 2
3455 static int check_to_create(const char *new_name, int ok_if_exists)
3457 struct stat nst;
3459 if (check_index &&
3460 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3461 !ok_if_exists)
3462 return EXISTS_IN_INDEX;
3463 if (cached)
3464 return 0;
3466 if (!lstat(new_name, &nst)) {
3467 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3468 return 0;
3470 * A leading component of new_name might be a symlink
3471 * that is going to be removed with this patch, but
3472 * still pointing at somewhere that has the path.
3473 * In such a case, path "new_name" does not exist as
3474 * far as git is concerned.
3476 if (has_symlink_leading_path(new_name, strlen(new_name)))
3477 return 0;
3479 return EXISTS_IN_WORKTREE;
3480 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3481 return error("%s: %s", new_name, strerror(errno));
3483 return 0;
3487 * Check and apply the patch in-core; leave the result in patch->result
3488 * for the caller to write it out to the final destination.
3490 static int check_patch(struct patch *patch)
3492 struct stat st;
3493 const char *old_name = patch->old_name;
3494 const char *new_name = patch->new_name;
3495 const char *name = old_name ? old_name : new_name;
3496 struct cache_entry *ce = NULL;
3497 struct patch *tpatch;
3498 int ok_if_exists;
3499 int status;
3501 patch->rejected = 1; /* we will drop this after we succeed */
3503 status = check_preimage(patch, &ce, &st);
3504 if (status)
3505 return status;
3506 old_name = patch->old_name;
3509 * A type-change diff is always split into a patch to delete
3510 * old, immediately followed by a patch to create new (see
3511 * diff.c::run_diff()); in such a case it is Ok that the entry
3512 * to be deleted by the previous patch is still in the working
3513 * tree and in the index.
3515 * A patch to swap-rename between A and B would first rename A
3516 * to B and then rename B to A. While applying the first one,
3517 * the presence of B should not stop A from getting renamed to
3518 * B; ask to_be_deleted() about the later rename. Removal of
3519 * B and rename from A to B is handled the same way by asking
3520 * was_deleted().
3522 if ((tpatch = in_fn_table(new_name)) &&
3523 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3524 ok_if_exists = 1;
3525 else
3526 ok_if_exists = 0;
3528 if (new_name &&
3529 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3530 int err = check_to_create(new_name, ok_if_exists);
3532 if (err && threeway) {
3533 patch->direct_to_threeway = 1;
3534 } else switch (err) {
3535 case 0:
3536 break; /* happy */
3537 case EXISTS_IN_INDEX:
3538 return error(_("%s: already exists in index"), new_name);
3539 break;
3540 case EXISTS_IN_WORKTREE:
3541 return error(_("%s: already exists in working directory"),
3542 new_name);
3543 default:
3544 return err;
3547 if (!patch->new_mode) {
3548 if (0 < patch->is_new)
3549 patch->new_mode = S_IFREG | 0644;
3550 else
3551 patch->new_mode = patch->old_mode;
3555 if (new_name && old_name) {
3556 int same = !strcmp(old_name, new_name);
3557 if (!patch->new_mode)
3558 patch->new_mode = patch->old_mode;
3559 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3560 if (same)
3561 return error(_("new mode (%o) of %s does not "
3562 "match old mode (%o)"),
3563 patch->new_mode, new_name,
3564 patch->old_mode);
3565 else
3566 return error(_("new mode (%o) of %s does not "
3567 "match old mode (%o) of %s"),
3568 patch->new_mode, new_name,
3569 patch->old_mode, old_name);
3573 if (apply_data(patch, &st, ce) < 0)
3574 return error(_("%s: patch does not apply"), name);
3575 patch->rejected = 0;
3576 return 0;
3579 static int check_patch_list(struct patch *patch)
3581 int err = 0;
3583 prepare_fn_table(patch);
3584 while (patch) {
3585 if (apply_verbosely)
3586 say_patch_name(stderr,
3587 _("Checking patch %s..."), patch);
3588 err |= check_patch(patch);
3589 patch = patch->next;
3591 return err;
3594 /* This function tries to read the sha1 from the current index */
3595 static int get_current_sha1(const char *path, unsigned char *sha1)
3597 int pos;
3599 if (read_cache() < 0)
3600 return -1;
3601 pos = cache_name_pos(path, strlen(path));
3602 if (pos < 0)
3603 return -1;
3604 hashcpy(sha1, active_cache[pos]->sha1);
3605 return 0;
3608 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3611 * A usable gitlink patch has only one fragment (hunk) that looks like:
3612 * @@ -1 +1 @@
3613 * -Subproject commit <old sha1>
3614 * +Subproject commit <new sha1>
3615 * or
3616 * @@ -1 +0,0 @@
3617 * -Subproject commit <old sha1>
3618 * for a removal patch.
3620 struct fragment *hunk = p->fragments;
3621 static const char heading[] = "-Subproject commit ";
3622 char *preimage;
3624 if (/* does the patch have only one hunk? */
3625 hunk && !hunk->next &&
3626 /* is its preimage one line? */
3627 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3628 /* does preimage begin with the heading? */
3629 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3630 starts_with(++preimage, heading) &&
3631 /* does it record full SHA-1? */
3632 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3633 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3634 /* does the abbreviated name on the index line agree with it? */
3635 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3636 return 0; /* it all looks fine */
3638 /* we may have full object name on the index line */
3639 return get_sha1_hex(p->old_sha1_prefix, sha1);
3642 /* Build an index that contains the just the files needed for a 3way merge */
3643 static void build_fake_ancestor(struct patch *list, const char *filename)
3645 struct patch *patch;
3646 struct index_state result = { NULL };
3647 int fd;
3649 /* Once we start supporting the reverse patch, it may be
3650 * worth showing the new sha1 prefix, but until then...
3652 for (patch = list; patch; patch = patch->next) {
3653 unsigned char sha1[20];
3654 struct cache_entry *ce;
3655 const char *name;
3657 name = patch->old_name ? patch->old_name : patch->new_name;
3658 if (0 < patch->is_new)
3659 continue;
3661 if (S_ISGITLINK(patch->old_mode)) {
3662 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3663 ; /* ok, the textual part looks sane */
3664 else
3665 die("sha1 information is lacking or useless for submoule %s",
3666 name);
3667 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3668 ; /* ok */
3669 } else if (!patch->lines_added && !patch->lines_deleted) {
3670 /* mode-only change: update the current */
3671 if (get_current_sha1(patch->old_name, sha1))
3672 die("mode change for %s, which is not "
3673 "in current HEAD", name);
3674 } else
3675 die("sha1 information is lacking or useless "
3676 "(%s).", name);
3678 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3679 if (!ce)
3680 die(_("make_cache_entry failed for path '%s'"), name);
3681 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3682 die ("Could not add %s to temporary index", name);
3685 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3686 if (fd < 0 || write_index(&result, fd) || close(fd))
3687 die ("Could not write temporary index to %s", filename);
3689 discard_index(&result);
3692 static void stat_patch_list(struct patch *patch)
3694 int files, adds, dels;
3696 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3697 files++;
3698 adds += patch->lines_added;
3699 dels += patch->lines_deleted;
3700 show_stats(patch);
3703 print_stat_summary(stdout, files, adds, dels);
3706 static void numstat_patch_list(struct patch *patch)
3708 for ( ; patch; patch = patch->next) {
3709 const char *name;
3710 name = patch->new_name ? patch->new_name : patch->old_name;
3711 if (patch->is_binary)
3712 printf("-\t-\t");
3713 else
3714 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3715 write_name_quoted(name, stdout, line_termination);
3719 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3721 if (mode)
3722 printf(" %s mode %06o %s\n", newdelete, mode, name);
3723 else
3724 printf(" %s %s\n", newdelete, name);
3727 static void show_mode_change(struct patch *p, int show_name)
3729 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3730 if (show_name)
3731 printf(" mode change %06o => %06o %s\n",
3732 p->old_mode, p->new_mode, p->new_name);
3733 else
3734 printf(" mode change %06o => %06o\n",
3735 p->old_mode, p->new_mode);
3739 static void show_rename_copy(struct patch *p)
3741 const char *renamecopy = p->is_rename ? "rename" : "copy";
3742 const char *old, *new;
3744 /* Find common prefix */
3745 old = p->old_name;
3746 new = p->new_name;
3747 while (1) {
3748 const char *slash_old, *slash_new;
3749 slash_old = strchr(old, '/');
3750 slash_new = strchr(new, '/');
3751 if (!slash_old ||
3752 !slash_new ||
3753 slash_old - old != slash_new - new ||
3754 memcmp(old, new, slash_new - new))
3755 break;
3756 old = slash_old + 1;
3757 new = slash_new + 1;
3759 /* p->old_name thru old is the common prefix, and old and new
3760 * through the end of names are renames
3762 if (old != p->old_name)
3763 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3764 (int)(old - p->old_name), p->old_name,
3765 old, new, p->score);
3766 else
3767 printf(" %s %s => %s (%d%%)\n", renamecopy,
3768 p->old_name, p->new_name, p->score);
3769 show_mode_change(p, 0);
3772 static void summary_patch_list(struct patch *patch)
3774 struct patch *p;
3776 for (p = patch; p; p = p->next) {
3777 if (p->is_new)
3778 show_file_mode_name("create", p->new_mode, p->new_name);
3779 else if (p->is_delete)
3780 show_file_mode_name("delete", p->old_mode, p->old_name);
3781 else {
3782 if (p->is_rename || p->is_copy)
3783 show_rename_copy(p);
3784 else {
3785 if (p->score) {
3786 printf(" rewrite %s (%d%%)\n",
3787 p->new_name, p->score);
3788 show_mode_change(p, 0);
3790 else
3791 show_mode_change(p, 1);
3797 static void patch_stats(struct patch *patch)
3799 int lines = patch->lines_added + patch->lines_deleted;
3801 if (lines > max_change)
3802 max_change = lines;
3803 if (patch->old_name) {
3804 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3805 if (!len)
3806 len = strlen(patch->old_name);
3807 if (len > max_len)
3808 max_len = len;
3810 if (patch->new_name) {
3811 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3812 if (!len)
3813 len = strlen(patch->new_name);
3814 if (len > max_len)
3815 max_len = len;
3819 static void remove_file(struct patch *patch, int rmdir_empty)
3821 if (update_index) {
3822 if (remove_file_from_cache(patch->old_name) < 0)
3823 die(_("unable to remove %s from index"), patch->old_name);
3825 if (!cached) {
3826 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3827 remove_path(patch->old_name);
3832 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3834 struct stat st;
3835 struct cache_entry *ce;
3836 int namelen = strlen(path);
3837 unsigned ce_size = cache_entry_size(namelen);
3839 if (!update_index)
3840 return;
3842 ce = xcalloc(1, ce_size);
3843 memcpy(ce->name, path, namelen);
3844 ce->ce_mode = create_ce_mode(mode);
3845 ce->ce_flags = create_ce_flags(0);
3846 ce->ce_namelen = namelen;
3847 if (S_ISGITLINK(mode)) {
3848 const char *s = buf;
3850 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3851 die(_("corrupt patch for submodule %s"), path);
3852 } else {
3853 if (!cached) {
3854 if (lstat(path, &st) < 0)
3855 die_errno(_("unable to stat newly created file '%s'"),
3856 path);
3857 fill_stat_cache_info(ce, &st);
3859 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3860 die(_("unable to create backing store for newly created file %s"), path);
3862 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3863 die(_("unable to add cache entry for %s"), path);
3866 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3868 int fd;
3869 struct strbuf nbuf = STRBUF_INIT;
3871 if (S_ISGITLINK(mode)) {
3872 struct stat st;
3873 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3874 return 0;
3875 return mkdir(path, 0777);
3878 if (has_symlinks && S_ISLNK(mode))
3879 /* Although buf:size is counted string, it also is NUL
3880 * terminated.
3882 return symlink(buf, path);
3884 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);