use REALLOC_ARRAY for changing the allocation size of arrays
[git/mingw.git] / builtin / apply.c
blob8714a887203acd84ad08f3e0917fc72b2c7a3138
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 = strchrnul(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 = xstrfmt("%s%s", root, patch->def_name);
1285 free(patch->def_name);
1286 patch->def_name = s;
1289 line += len;
1290 size -= len;
1291 linenr++;
1292 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1293 static const struct opentry {
1294 const char *str;
1295 int (*fn)(const char *, struct patch *);
1296 } optable[] = {
1297 { "@@ -", gitdiff_hdrend },
1298 { "--- ", gitdiff_oldname },
1299 { "+++ ", gitdiff_newname },
1300 { "old mode ", gitdiff_oldmode },
1301 { "new mode ", gitdiff_newmode },
1302 { "deleted file mode ", gitdiff_delete },
1303 { "new file mode ", gitdiff_newfile },
1304 { "copy from ", gitdiff_copysrc },
1305 { "copy to ", gitdiff_copydst },
1306 { "rename old ", gitdiff_renamesrc },
1307 { "rename new ", gitdiff_renamedst },
1308 { "rename from ", gitdiff_renamesrc },
1309 { "rename to ", gitdiff_renamedst },
1310 { "similarity index ", gitdiff_similarity },
1311 { "dissimilarity index ", gitdiff_dissimilarity },
1312 { "index ", gitdiff_index },
1313 { "", gitdiff_unrecognized },
1315 int i;
1317 len = linelen(line, size);
1318 if (!len || line[len-1] != '\n')
1319 break;
1320 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1321 const struct opentry *p = optable + i;
1322 int oplen = strlen(p->str);
1323 if (len < oplen || memcmp(p->str, line, oplen))
1324 continue;
1325 if (p->fn(line + oplen, patch) < 0)
1326 return offset;
1327 break;
1331 return offset;
1334 static int parse_num(const char *line, unsigned long *p)
1336 char *ptr;
1338 if (!isdigit(*line))
1339 return 0;
1340 *p = strtoul(line, &ptr, 10);
1341 return ptr - line;
1344 static int parse_range(const char *line, int len, int offset, const char *expect,
1345 unsigned long *p1, unsigned long *p2)
1347 int digits, ex;
1349 if (offset < 0 || offset >= len)
1350 return -1;
1351 line += offset;
1352 len -= offset;
1354 digits = parse_num(line, p1);
1355 if (!digits)
1356 return -1;
1358 offset += digits;
1359 line += digits;
1360 len -= digits;
1362 *p2 = 1;
1363 if (*line == ',') {
1364 digits = parse_num(line+1, p2);
1365 if (!digits)
1366 return -1;
1368 offset += digits+1;
1369 line += digits+1;
1370 len -= digits+1;
1373 ex = strlen(expect);
1374 if (ex > len)
1375 return -1;
1376 if (memcmp(line, expect, ex))
1377 return -1;
1379 return offset + ex;
1382 static void recount_diff(const char *line, int size, struct fragment *fragment)
1384 int oldlines = 0, newlines = 0, ret = 0;
1386 if (size < 1) {
1387 warning("recount: ignore empty hunk");
1388 return;
1391 for (;;) {
1392 int len = linelen(line, size);
1393 size -= len;
1394 line += len;
1396 if (size < 1)
1397 break;
1399 switch (*line) {
1400 case ' ': case '\n':
1401 newlines++;
1402 /* fall through */
1403 case '-':
1404 oldlines++;
1405 continue;
1406 case '+':
1407 newlines++;
1408 continue;
1409 case '\\':
1410 continue;
1411 case '@':
1412 ret = size < 3 || !starts_with(line, "@@ ");
1413 break;
1414 case 'd':
1415 ret = size < 5 || !starts_with(line, "diff ");
1416 break;
1417 default:
1418 ret = -1;
1419 break;
1421 if (ret) {
1422 warning(_("recount: unexpected line: %.*s"),
1423 (int)linelen(line, size), line);
1424 return;
1426 break;
1428 fragment->oldlines = oldlines;
1429 fragment->newlines = newlines;
1433 * Parse a unified diff fragment header of the
1434 * form "@@ -a,b +c,d @@"
1436 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1438 int offset;
1440 if (!len || line[len-1] != '\n')
1441 return -1;
1443 /* Figure out the number of lines in a fragment */
1444 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1445 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1447 return offset;
1450 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1452 unsigned long offset, len;
1454 patch->is_toplevel_relative = 0;
1455 patch->is_rename = patch->is_copy = 0;
1456 patch->is_new = patch->is_delete = -1;
1457 patch->old_mode = patch->new_mode = 0;
1458 patch->old_name = patch->new_name = NULL;
1459 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1460 unsigned long nextlen;
1462 len = linelen(line, size);
1463 if (!len)
1464 break;
1466 /* Testing this early allows us to take a few shortcuts.. */
1467 if (len < 6)
1468 continue;
1471 * Make sure we don't find any unconnected patch fragments.
1472 * That's a sign that we didn't find a header, and that a
1473 * patch has become corrupted/broken up.
1475 if (!memcmp("@@ -", line, 4)) {
1476 struct fragment dummy;
1477 if (parse_fragment_header(line, len, &dummy) < 0)
1478 continue;
1479 die(_("patch fragment without header at line %d: %.*s"),
1480 linenr, (int)len-1, line);
1483 if (size < len + 6)
1484 break;
1487 * Git patch? It might not have a real patch, just a rename
1488 * or mode change, so we handle that specially
1490 if (!memcmp("diff --git ", line, 11)) {
1491 int git_hdr_len = parse_git_header(line, len, size, patch);
1492 if (git_hdr_len <= len)
1493 continue;
1494 if (!patch->old_name && !patch->new_name) {
1495 if (!patch->def_name)
1496 die(Q_("git diff header lacks filename information when removing "
1497 "%d leading pathname component (line %d)",
1498 "git diff header lacks filename information when removing "
1499 "%d leading pathname components (line %d)",
1500 p_value),
1501 p_value, linenr);
1502 patch->old_name = xstrdup(patch->def_name);
1503 patch->new_name = xstrdup(patch->def_name);
1505 if (!patch->is_delete && !patch->new_name)
1506 die("git diff header lacks filename information "
1507 "(line %d)", linenr);
1508 patch->is_toplevel_relative = 1;
1509 *hdrsize = git_hdr_len;
1510 return offset;
1513 /* --- followed by +++ ? */
1514 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1515 continue;
1518 * We only accept unified patches, so we want it to
1519 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1520 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1522 nextlen = linelen(line + len, size - len);
1523 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1524 continue;
1526 /* Ok, we'll consider it a patch */
1527 parse_traditional_patch(line, line+len, patch);
1528 *hdrsize = len + nextlen;
1529 linenr += 2;
1530 return offset;
1532 return -1;
1535 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1537 char *err;
1539 if (!result)
1540 return;
1542 whitespace_error++;
1543 if (squelch_whitespace_errors &&
1544 squelch_whitespace_errors < whitespace_error)
1545 return;
1547 err = whitespace_error_string(result);
1548 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1549 patch_input_file, linenr, err, len, line);
1550 free(err);
1553 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1555 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1557 record_ws_error(result, line + 1, len - 2, linenr);
1561 * Parse a unified diff. Note that this really needs to parse each
1562 * fragment separately, since the only way to know the difference
1563 * between a "---" that is part of a patch, and a "---" that starts
1564 * the next patch is to look at the line counts..
1566 static int parse_fragment(const char *line, unsigned long size,
1567 struct patch *patch, struct fragment *fragment)
1569 int added, deleted;
1570 int len = linelen(line, size), offset;
1571 unsigned long oldlines, newlines;
1572 unsigned long leading, trailing;
1574 offset = parse_fragment_header(line, len, fragment);
1575 if (offset < 0)
1576 return -1;
1577 if (offset > 0 && patch->recount)
1578 recount_diff(line + offset, size - offset, fragment);
1579 oldlines = fragment->oldlines;
1580 newlines = fragment->newlines;
1581 leading = 0;
1582 trailing = 0;
1584 /* Parse the thing.. */
1585 line += len;
1586 size -= len;
1587 linenr++;
1588 added = deleted = 0;
1589 for (offset = len;
1590 0 < size;
1591 offset += len, size -= len, line += len, linenr++) {
1592 if (!oldlines && !newlines)
1593 break;
1594 len = linelen(line, size);
1595 if (!len || line[len-1] != '\n')
1596 return -1;
1597 switch (*line) {
1598 default:
1599 return -1;
1600 case '\n': /* newer GNU diff, an empty context line */
1601 case ' ':
1602 oldlines--;
1603 newlines--;
1604 if (!deleted && !added)
1605 leading++;
1606 trailing++;
1607 break;
1608 case '-':
1609 if (apply_in_reverse &&
1610 ws_error_action != nowarn_ws_error)
1611 check_whitespace(line, len, patch->ws_rule);
1612 deleted++;
1613 oldlines--;
1614 trailing = 0;
1615 break;
1616 case '+':
1617 if (!apply_in_reverse &&
1618 ws_error_action != nowarn_ws_error)
1619 check_whitespace(line, len, patch->ws_rule);
1620 added++;
1621 newlines--;
1622 trailing = 0;
1623 break;
1626 * We allow "\ No newline at end of file". Depending
1627 * on locale settings when the patch was produced we
1628 * don't know what this line looks like. The only
1629 * thing we do know is that it begins with "\ ".
1630 * Checking for 12 is just for sanity check -- any
1631 * l10n of "\ No newline..." is at least that long.
1633 case '\\':
1634 if (len < 12 || memcmp(line, "\\ ", 2))
1635 return -1;
1636 break;
1639 if (oldlines || newlines)
1640 return -1;
1641 fragment->leading = leading;
1642 fragment->trailing = trailing;
1645 * If a fragment ends with an incomplete line, we failed to include
1646 * it in the above loop because we hit oldlines == newlines == 0
1647 * before seeing it.
1649 if (12 < size && !memcmp(line, "\\ ", 2))
1650 offset += linelen(line, size);
1652 patch->lines_added += added;
1653 patch->lines_deleted += deleted;
1655 if (0 < patch->is_new && oldlines)
1656 return error(_("new file depends on old contents"));
1657 if (0 < patch->is_delete && newlines)
1658 return error(_("deleted file still has contents"));
1659 return offset;
1663 * We have seen "diff --git a/... b/..." header (or a traditional patch
1664 * header). Read hunks that belong to this patch into fragments and hang
1665 * them to the given patch structure.
1667 * The (fragment->patch, fragment->size) pair points into the memory given
1668 * by the caller, not a copy, when we return.
1670 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1672 unsigned long offset = 0;
1673 unsigned long oldlines = 0, newlines = 0, context = 0;
1674 struct fragment **fragp = &patch->fragments;
1676 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1677 struct fragment *fragment;
1678 int len;
1680 fragment = xcalloc(1, sizeof(*fragment));
1681 fragment->linenr = linenr;
1682 len = parse_fragment(line, size, patch, fragment);
1683 if (len <= 0)
1684 die(_("corrupt patch at line %d"), linenr);
1685 fragment->patch = line;
1686 fragment->size = len;
1687 oldlines += fragment->oldlines;
1688 newlines += fragment->newlines;
1689 context += fragment->leading + fragment->trailing;
1691 *fragp = fragment;
1692 fragp = &fragment->next;
1694 offset += len;
1695 line += len;
1696 size -= len;
1700 * If something was removed (i.e. we have old-lines) it cannot
1701 * be creation, and if something was added it cannot be
1702 * deletion. However, the reverse is not true; --unified=0
1703 * patches that only add are not necessarily creation even
1704 * though they do not have any old lines, and ones that only
1705 * delete are not necessarily deletion.
1707 * Unfortunately, a real creation/deletion patch do _not_ have
1708 * any context line by definition, so we cannot safely tell it
1709 * apart with --unified=0 insanity. At least if the patch has
1710 * more than one hunk it is not creation or deletion.
1712 if (patch->is_new < 0 &&
1713 (oldlines || (patch->fragments && patch->fragments->next)))
1714 patch->is_new = 0;
1715 if (patch->is_delete < 0 &&
1716 (newlines || (patch->fragments && patch->fragments->next)))
1717 patch->is_delete = 0;
1719 if (0 < patch->is_new && oldlines)
1720 die(_("new file %s depends on old contents"), patch->new_name);
1721 if (0 < patch->is_delete && newlines)
1722 die(_("deleted file %s still has contents"), patch->old_name);
1723 if (!patch->is_delete && !newlines && context)
1724 fprintf_ln(stderr,
1725 _("** warning: "
1726 "file %s becomes empty but is not deleted"),
1727 patch->new_name);
1729 return offset;
1732 static inline int metadata_changes(struct patch *patch)
1734 return patch->is_rename > 0 ||
1735 patch->is_copy > 0 ||
1736 patch->is_new > 0 ||
1737 patch->is_delete ||
1738 (patch->old_mode && patch->new_mode &&
1739 patch->old_mode != patch->new_mode);
1742 static char *inflate_it(const void *data, unsigned long size,
1743 unsigned long inflated_size)
1745 git_zstream stream;
1746 void *out;
1747 int st;
1749 memset(&stream, 0, sizeof(stream));
1751 stream.next_in = (unsigned char *)data;
1752 stream.avail_in = size;
1753 stream.next_out = out = xmalloc(inflated_size);
1754 stream.avail_out = inflated_size;
1755 git_inflate_init(&stream);
1756 st = git_inflate(&stream, Z_FINISH);
1757 git_inflate_end(&stream);
1758 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1759 free(out);
1760 return NULL;
1762 return out;
1766 * Read a binary hunk and return a new fragment; fragment->patch
1767 * points at an allocated memory that the caller must free, so
1768 * it is marked as "->free_patch = 1".
1770 static struct fragment *parse_binary_hunk(char **buf_p,
1771 unsigned long *sz_p,
1772 int *status_p,
1773 int *used_p)
1776 * Expect a line that begins with binary patch method ("literal"
1777 * or "delta"), followed by the length of data before deflating.
1778 * a sequence of 'length-byte' followed by base-85 encoded data
1779 * should follow, terminated by a newline.
1781 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1782 * and we would limit the patch line to 66 characters,
1783 * so one line can fit up to 13 groups that would decode
1784 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1785 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1787 int llen, used;
1788 unsigned long size = *sz_p;
1789 char *buffer = *buf_p;
1790 int patch_method;
1791 unsigned long origlen;
1792 char *data = NULL;
1793 int hunk_size = 0;
1794 struct fragment *frag;
1796 llen = linelen(buffer, size);
1797 used = llen;
1799 *status_p = 0;
1801 if (starts_with(buffer, "delta ")) {
1802 patch_method = BINARY_DELTA_DEFLATED;
1803 origlen = strtoul(buffer + 6, NULL, 10);
1805 else if (starts_with(buffer, "literal ")) {
1806 patch_method = BINARY_LITERAL_DEFLATED;
1807 origlen = strtoul(buffer + 8, NULL, 10);
1809 else
1810 return NULL;
1812 linenr++;
1813 buffer += llen;
1814 while (1) {
1815 int byte_length, max_byte_length, newsize;
1816 llen = linelen(buffer, size);
1817 used += llen;
1818 linenr++;
1819 if (llen == 1) {
1820 /* consume the blank line */
1821 buffer++;
1822 size--;
1823 break;
1826 * Minimum line is "A00000\n" which is 7-byte long,
1827 * and the line length must be multiple of 5 plus 2.
1829 if ((llen < 7) || (llen-2) % 5)
1830 goto corrupt;
1831 max_byte_length = (llen - 2) / 5 * 4;
1832 byte_length = *buffer;
1833 if ('A' <= byte_length && byte_length <= 'Z')
1834 byte_length = byte_length - 'A' + 1;
1835 else if ('a' <= byte_length && byte_length <= 'z')
1836 byte_length = byte_length - 'a' + 27;
1837 else
1838 goto corrupt;
1839 /* if the input length was not multiple of 4, we would
1840 * have filler at the end but the filler should never
1841 * exceed 3 bytes
1843 if (max_byte_length < byte_length ||
1844 byte_length <= max_byte_length - 4)
1845 goto corrupt;
1846 newsize = hunk_size + byte_length;
1847 data = xrealloc(data, newsize);
1848 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1849 goto corrupt;
1850 hunk_size = newsize;
1851 buffer += llen;
1852 size -= llen;
1855 frag = xcalloc(1, sizeof(*frag));
1856 frag->patch = inflate_it(data, hunk_size, origlen);
1857 frag->free_patch = 1;
1858 if (!frag->patch)
1859 goto corrupt;
1860 free(data);
1861 frag->size = origlen;
1862 *buf_p = buffer;
1863 *sz_p = size;
1864 *used_p = used;
1865 frag->binary_patch_method = patch_method;
1866 return frag;
1868 corrupt:
1869 free(data);
1870 *status_p = -1;
1871 error(_("corrupt binary patch at line %d: %.*s"),
1872 linenr-1, llen-1, buffer);
1873 return NULL;
1876 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1879 * We have read "GIT binary patch\n"; what follows is a line
1880 * that says the patch method (currently, either "literal" or
1881 * "delta") and the length of data before deflating; a
1882 * sequence of 'length-byte' followed by base-85 encoded data
1883 * follows.
1885 * When a binary patch is reversible, there is another binary
1886 * hunk in the same format, starting with patch method (either
1887 * "literal" or "delta") with the length of data, and a sequence
1888 * of length-byte + base-85 encoded data, terminated with another
1889 * empty line. This data, when applied to the postimage, produces
1890 * the preimage.
1892 struct fragment *forward;
1893 struct fragment *reverse;
1894 int status;
1895 int used, used_1;
1897 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1898 if (!forward && !status)
1899 /* there has to be one hunk (forward hunk) */
1900 return error(_("unrecognized binary patch at line %d"), linenr-1);
1901 if (status)
1902 /* otherwise we already gave an error message */
1903 return status;
1905 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1906 if (reverse)
1907 used += used_1;
1908 else if (status) {
1910 * Not having reverse hunk is not an error, but having
1911 * a corrupt reverse hunk is.
1913 free((void*) forward->patch);
1914 free(forward);
1915 return status;
1917 forward->next = reverse;
1918 patch->fragments = forward;
1919 patch->is_binary = 1;
1920 return used;
1923 static void prefix_one(char **name)
1925 char *old_name = *name;
1926 if (!old_name)
1927 return;
1928 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
1929 free(old_name);
1932 static void prefix_patch(struct patch *p)
1934 if (!prefix || p->is_toplevel_relative)
1935 return;
1936 prefix_one(&p->new_name);
1937 prefix_one(&p->old_name);
1941 * include/exclude
1944 static struct string_list limit_by_name;
1945 static int has_include;
1946 static void add_name_limit(const char *name, int exclude)
1948 struct string_list_item *it;
1950 it = string_list_append(&limit_by_name, name);
1951 it->util = exclude ? NULL : (void *) 1;
1954 static int use_patch(struct patch *p)
1956 const char *pathname = p->new_name ? p->new_name : p->old_name;
1957 int i;
1959 /* Paths outside are not touched regardless of "--include" */
1960 if (0 < prefix_length) {
1961 int pathlen = strlen(pathname);
1962 if (pathlen <= prefix_length ||
1963 memcmp(prefix, pathname, prefix_length))
1964 return 0;
1967 /* See if it matches any of exclude/include rule */
1968 for (i = 0; i < limit_by_name.nr; i++) {
1969 struct string_list_item *it = &limit_by_name.items[i];
1970 if (!wildmatch(it->string, pathname, 0, NULL))
1971 return (it->util != NULL);
1975 * If we had any include, a path that does not match any rule is
1976 * not used. Otherwise, we saw bunch of exclude rules (or none)
1977 * and such a path is used.
1979 return !has_include;
1984 * Read the patch text in "buffer" that extends for "size" bytes; stop
1985 * reading after seeing a single patch (i.e. changes to a single file).
1986 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1987 * Return the number of bytes consumed, so that the caller can call us
1988 * again for the next patch.
1990 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1992 int hdrsize, patchsize;
1993 int offset = find_header(buffer, size, &hdrsize, patch);
1995 if (offset < 0)
1996 return offset;
1998 prefix_patch(patch);
2000 if (!use_patch(patch))
2001 patch->ws_rule = 0;
2002 else
2003 patch->ws_rule = whitespace_rule(patch->new_name
2004 ? patch->new_name
2005 : patch->old_name);
2007 patchsize = parse_single_patch(buffer + offset + hdrsize,
2008 size - offset - hdrsize, patch);
2010 if (!patchsize) {
2011 static const char git_binary[] = "GIT binary patch\n";
2012 int hd = hdrsize + offset;
2013 unsigned long llen = linelen(buffer + hd, size - hd);
2015 if (llen == sizeof(git_binary) - 1 &&
2016 !memcmp(git_binary, buffer + hd, llen)) {
2017 int used;
2018 linenr++;
2019 used = parse_binary(buffer + hd + llen,
2020 size - hd - llen, patch);
2021 if (used)
2022 patchsize = used + llen;
2023 else
2024 patchsize = 0;
2026 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2027 static const char *binhdr[] = {
2028 "Binary files ",
2029 "Files ",
2030 NULL,
2032 int i;
2033 for (i = 0; binhdr[i]; i++) {
2034 int len = strlen(binhdr[i]);
2035 if (len < size - hd &&
2036 !memcmp(binhdr[i], buffer + hd, len)) {
2037 linenr++;
2038 patch->is_binary = 1;
2039 patchsize = llen;
2040 break;
2045 /* Empty patch cannot be applied if it is a text patch
2046 * without metadata change. A binary patch appears
2047 * empty to us here.
2049 if ((apply || check) &&
2050 (!patch->is_binary && !metadata_changes(patch)))
2051 die(_("patch with only garbage at line %d"), linenr);
2054 return offset + hdrsize + patchsize;
2057 #define swap(a,b) myswap((a),(b),sizeof(a))
2059 #define myswap(a, b, size) do { \
2060 unsigned char mytmp[size]; \
2061 memcpy(mytmp, &a, size); \
2062 memcpy(&a, &b, size); \
2063 memcpy(&b, mytmp, size); \
2064 } while (0)
2066 static void reverse_patches(struct patch *p)
2068 for (; p; p = p->next) {
2069 struct fragment *frag = p->fragments;
2071 swap(p->new_name, p->old_name);
2072 swap(p->new_mode, p->old_mode);
2073 swap(p->is_new, p->is_delete);
2074 swap(p->lines_added, p->lines_deleted);
2075 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2077 for (; frag; frag = frag->next) {
2078 swap(frag->newpos, frag->oldpos);
2079 swap(frag->newlines, frag->oldlines);
2084 static const char pluses[] =
2085 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2086 static const char minuses[]=
2087 "----------------------------------------------------------------------";
2089 static void show_stats(struct patch *patch)
2091 struct strbuf qname = STRBUF_INIT;
2092 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2093 int max, add, del;
2095 quote_c_style(cp, &qname, NULL, 0);
2098 * "scale" the filename
2100 max = max_len;
2101 if (max > 50)
2102 max = 50;
2104 if (qname.len > max) {
2105 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2106 if (!cp)
2107 cp = qname.buf + qname.len + 3 - max;
2108 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2111 if (patch->is_binary) {
2112 printf(" %-*s | Bin\n", max, qname.buf);
2113 strbuf_release(&qname);
2114 return;
2117 printf(" %-*s |", max, qname.buf);
2118 strbuf_release(&qname);
2121 * scale the add/delete
2123 max = max + max_change > 70 ? 70 - max : max_change;
2124 add = patch->lines_added;
2125 del = patch->lines_deleted;
2127 if (max_change > 0) {
2128 int total = ((add + del) * max + max_change / 2) / max_change;
2129 add = (add * max + max_change / 2) / max_change;
2130 del = total - add;
2132 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2133 add, pluses, del, minuses);
2136 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2138 switch (st->st_mode & S_IFMT) {
2139 case S_IFLNK:
2140 if (strbuf_readlink(buf, path, st->st_size) < 0)
2141 return error(_("unable to read symlink %s"), path);
2142 return 0;
2143 case S_IFREG:
2144 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2145 return error(_("unable to open or read %s"), path);
2146 convert_to_git(path, buf->buf, buf->len, buf, 0);
2147 return 0;
2148 default:
2149 return -1;
2154 * Update the preimage, and the common lines in postimage,
2155 * from buffer buf of length len. If postlen is 0 the postimage
2156 * is updated in place, otherwise it's updated on a new buffer
2157 * of length postlen
2160 static void update_pre_post_images(struct image *preimage,
2161 struct image *postimage,
2162 char *buf,
2163 size_t len, size_t postlen)
2165 int i, ctx, reduced;
2166 char *new, *old, *fixed;
2167 struct image fixed_preimage;
2170 * Update the preimage with whitespace fixes. Note that we
2171 * are not losing preimage->buf -- apply_one_fragment() will
2172 * free "oldlines".
2174 prepare_image(&fixed_preimage, buf, len, 1);
2175 assert(postlen
2176 ? fixed_preimage.nr == preimage->nr
2177 : fixed_preimage.nr <= preimage->nr);
2178 for (i = 0; i < fixed_preimage.nr; i++)
2179 fixed_preimage.line[i].flag = preimage->line[i].flag;
2180 free(preimage->line_allocated);
2181 *preimage = fixed_preimage;
2184 * Adjust the common context lines in postimage. This can be
2185 * done in-place when we are shrinking it with whitespace
2186 * fixing, but needs a new buffer when ignoring whitespace or
2187 * expanding leading tabs to spaces.
2189 * We trust the caller to tell us if the update can be done
2190 * in place (postlen==0) or not.
2192 old = postimage->buf;
2193 if (postlen)
2194 new = postimage->buf = xmalloc(postlen);
2195 else
2196 new = old;
2197 fixed = preimage->buf;
2199 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2200 size_t len = postimage->line[i].len;
2201 if (!(postimage->line[i].flag & LINE_COMMON)) {
2202 /* an added line -- no counterparts in preimage */
2203 memmove(new, old, len);
2204 old += len;
2205 new += len;
2206 continue;
2209 /* a common context -- skip it in the original postimage */
2210 old += len;
2212 /* and find the corresponding one in the fixed preimage */
2213 while (ctx < preimage->nr &&
2214 !(preimage->line[ctx].flag & LINE_COMMON)) {
2215 fixed += preimage->line[ctx].len;
2216 ctx++;
2220 * preimage is expected to run out, if the caller
2221 * fixed addition of trailing blank lines.
2223 if (preimage->nr <= ctx) {
2224 reduced++;
2225 continue;
2228 /* and copy it in, while fixing the line length */
2229 len = preimage->line[ctx].len;
2230 memcpy(new, fixed, len);
2231 new += len;
2232 fixed += len;
2233 postimage->line[i].len = len;
2234 ctx++;
2237 /* Fix the length of the whole thing */
2238 postimage->len = new - postimage->buf;
2239 postimage->nr -= reduced;
2242 static int match_fragment(struct image *img,
2243 struct image *preimage,
2244 struct image *postimage,
2245 unsigned long try,
2246 int try_lno,
2247 unsigned ws_rule,
2248 int match_beginning, int match_end)
2250 int i;
2251 char *fixed_buf, *buf, *orig, *target;
2252 struct strbuf fixed;
2253 size_t fixed_len, postlen;
2254 int preimage_limit;
2256 if (preimage->nr + try_lno <= img->nr) {
2258 * The hunk falls within the boundaries of img.
2260 preimage_limit = preimage->nr;
2261 if (match_end && (preimage->nr + try_lno != img->nr))
2262 return 0;
2263 } else if (ws_error_action == correct_ws_error &&
2264 (ws_rule & WS_BLANK_AT_EOF)) {
2266 * This hunk extends beyond the end of img, and we are
2267 * removing blank lines at the end of the file. This
2268 * many lines from the beginning of the preimage must
2269 * match with img, and the remainder of the preimage
2270 * must be blank.
2272 preimage_limit = img->nr - try_lno;
2273 } else {
2275 * The hunk extends beyond the end of the img and
2276 * we are not removing blanks at the end, so we
2277 * should reject the hunk at this position.
2279 return 0;
2282 if (match_beginning && try_lno)
2283 return 0;
2285 /* Quick hash check */
2286 for (i = 0; i < preimage_limit; i++)
2287 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2288 (preimage->line[i].hash != img->line[try_lno + i].hash))
2289 return 0;
2291 if (preimage_limit == preimage->nr) {
2293 * Do we have an exact match? If we were told to match
2294 * at the end, size must be exactly at try+fragsize,
2295 * otherwise try+fragsize must be still within the preimage,
2296 * and either case, the old piece should match the preimage
2297 * exactly.
2299 if ((match_end
2300 ? (try + preimage->len == img->len)
2301 : (try + preimage->len <= img->len)) &&
2302 !memcmp(img->buf + try, preimage->buf, preimage->len))
2303 return 1;
2304 } else {
2306 * The preimage extends beyond the end of img, so
2307 * there cannot be an exact match.
2309 * There must be one non-blank context line that match
2310 * a line before the end of img.
2312 char *buf_end;
2314 buf = preimage->buf;
2315 buf_end = buf;
2316 for (i = 0; i < preimage_limit; i++)
2317 buf_end += preimage->line[i].len;
2319 for ( ; buf < buf_end; buf++)
2320 if (!isspace(*buf))
2321 break;
2322 if (buf == buf_end)
2323 return 0;
2327 * No exact match. If we are ignoring whitespace, run a line-by-line
2328 * fuzzy matching. We collect all the line length information because
2329 * we need it to adjust whitespace if we match.
2331 if (ws_ignore_action == ignore_ws_change) {
2332 size_t imgoff = 0;
2333 size_t preoff = 0;
2334 size_t postlen = postimage->len;
2335 size_t extra_chars;
2336 char *preimage_eof;
2337 char *preimage_end;
2338 for (i = 0; i < preimage_limit; i++) {
2339 size_t prelen = preimage->line[i].len;
2340 size_t imglen = img->line[try_lno+i].len;
2342 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2343 preimage->buf + preoff, prelen))
2344 return 0;
2345 if (preimage->line[i].flag & LINE_COMMON)
2346 postlen += imglen - prelen;
2347 imgoff += imglen;
2348 preoff += prelen;
2352 * Ok, the preimage matches with whitespace fuzz.
2354 * imgoff now holds the true length of the target that
2355 * matches the preimage before the end of the file.
2357 * Count the number of characters in the preimage that fall
2358 * beyond the end of the file and make sure that all of them
2359 * are whitespace characters. (This can only happen if
2360 * we are removing blank lines at the end of the file.)
2362 buf = preimage_eof = preimage->buf + preoff;
2363 for ( ; i < preimage->nr; i++)
2364 preoff += preimage->line[i].len;
2365 preimage_end = preimage->buf + preoff;
2366 for ( ; buf < preimage_end; buf++)
2367 if (!isspace(*buf))
2368 return 0;
2371 * Update the preimage and the common postimage context
2372 * lines to use the same whitespace as the target.
2373 * If whitespace is missing in the target (i.e.
2374 * if the preimage extends beyond the end of the file),
2375 * use the whitespace from the preimage.
2377 extra_chars = preimage_end - preimage_eof;
2378 strbuf_init(&fixed, imgoff + extra_chars);
2379 strbuf_add(&fixed, img->buf + try, imgoff);
2380 strbuf_add(&fixed, preimage_eof, extra_chars);
2381 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2382 update_pre_post_images(preimage, postimage,
2383 fixed_buf, fixed_len, postlen);
2384 return 1;
2387 if (ws_error_action != correct_ws_error)
2388 return 0;
2391 * The hunk does not apply byte-by-byte, but the hash says
2392 * it might with whitespace fuzz. We haven't been asked to
2393 * ignore whitespace, we were asked to correct whitespace
2394 * errors, so let's try matching after whitespace correction.
2396 * The preimage may extend beyond the end of the file,
2397 * but in this loop we will only handle the part of the
2398 * preimage that falls within the file.
2400 strbuf_init(&fixed, preimage->len + 1);
2401 orig = preimage->buf;
2402 target = img->buf + try;
2403 postlen = 0;
2404 for (i = 0; i < preimage_limit; i++) {
2405 size_t oldlen = preimage->line[i].len;
2406 size_t tgtlen = img->line[try_lno + i].len;
2407 size_t fixstart = fixed.len;
2408 struct strbuf tgtfix;
2409 int match;
2411 /* Try fixing the line in the preimage */
2412 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2414 /* Try fixing the line in the target */
2415 strbuf_init(&tgtfix, tgtlen);
2416 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2419 * If they match, either the preimage was based on
2420 * a version before our tree fixed whitespace breakage,
2421 * or we are lacking a whitespace-fix patch the tree
2422 * the preimage was based on already had (i.e. target
2423 * has whitespace breakage, the preimage doesn't).
2424 * In either case, we are fixing the whitespace breakages
2425 * so we might as well take the fix together with their
2426 * real change.
2428 match = (tgtfix.len == fixed.len - fixstart &&
2429 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2430 fixed.len - fixstart));
2431 postlen += tgtfix.len;
2433 strbuf_release(&tgtfix);
2434 if (!match)
2435 goto unmatch_exit;
2437 orig += oldlen;
2438 target += tgtlen;
2443 * Now handle the lines in the preimage that falls beyond the
2444 * end of the file (if any). They will only match if they are
2445 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2446 * false).
2448 for ( ; i < preimage->nr; i++) {
2449 size_t fixstart = fixed.len; /* start of the fixed preimage */
2450 size_t oldlen = preimage->line[i].len;
2451 int j;
2453 /* Try fixing the line in the preimage */
2454 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2456 for (j = fixstart; j < fixed.len; j++)
2457 if (!isspace(fixed.buf[j]))
2458 goto unmatch_exit;
2460 orig += oldlen;
2464 * Yes, the preimage is based on an older version that still
2465 * has whitespace breakages unfixed, and fixing them makes the
2466 * hunk match. Update the context lines in the postimage.
2468 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2469 if (postlen < postimage->len)
2470 postlen = 0;
2471 update_pre_post_images(preimage, postimage,
2472 fixed_buf, fixed_len, postlen);
2473 return 1;
2475 unmatch_exit:
2476 strbuf_release(&fixed);
2477 return 0;
2480 static int find_pos(struct image *img,
2481 struct image *preimage,
2482 struct image *postimage,
2483 int line,
2484 unsigned ws_rule,
2485 int match_beginning, int match_end)
2487 int i;
2488 unsigned long backwards, forwards, try;
2489 int backwards_lno, forwards_lno, try_lno;
2492 * If match_beginning or match_end is specified, there is no
2493 * point starting from a wrong line that will never match and
2494 * wander around and wait for a match at the specified end.
2496 if (match_beginning)
2497 line = 0;
2498 else if (match_end)
2499 line = img->nr - preimage->nr;
2502 * Because the comparison is unsigned, the following test
2503 * will also take care of a negative line number that can
2504 * result when match_end and preimage is larger than the target.
2506 if ((size_t) line > img->nr)
2507 line = img->nr;
2509 try = 0;
2510 for (i = 0; i < line; i++)
2511 try += img->line[i].len;
2514 * There's probably some smart way to do this, but I'll leave
2515 * that to the smart and beautiful people. I'm simple and stupid.
2517 backwards = try;
2518 backwards_lno = line;
2519 forwards = try;
2520 forwards_lno = line;
2521 try_lno = line;
2523 for (i = 0; ; i++) {
2524 if (match_fragment(img, preimage, postimage,
2525 try, try_lno, ws_rule,
2526 match_beginning, match_end))
2527 return try_lno;
2529 again:
2530 if (backwards_lno == 0 && forwards_lno == img->nr)
2531 break;
2533 if (i & 1) {
2534 if (backwards_lno == 0) {
2535 i++;
2536 goto again;
2538 backwards_lno--;
2539 backwards -= img->line[backwards_lno].len;
2540 try = backwards;
2541 try_lno = backwards_lno;
2542 } else {
2543 if (forwards_lno == img->nr) {
2544 i++;
2545 goto again;
2547 forwards += img->line[forwards_lno].len;
2548 forwards_lno++;
2549 try = forwards;
2550 try_lno = forwards_lno;
2554 return -1;
2557 static void remove_first_line(struct image *img)
2559 img->buf += img->line[0].len;
2560 img->len -= img->line[0].len;
2561 img->line++;
2562 img->nr--;
2565 static void remove_last_line(struct image *img)
2567 img->len -= img->line[--img->nr].len;
2571 * The change from "preimage" and "postimage" has been found to
2572 * apply at applied_pos (counts in line numbers) in "img".
2573 * Update "img" to remove "preimage" and replace it with "postimage".
2575 static void update_image(struct image *img,
2576 int applied_pos,
2577 struct image *preimage,
2578 struct image *postimage)
2581 * remove the copy of preimage at offset in img
2582 * and replace it with postimage
2584 int i, nr;
2585 size_t remove_count, insert_count, applied_at = 0;
2586 char *result;
2587 int preimage_limit;
2590 * If we are removing blank lines at the end of img,
2591 * the preimage may extend beyond the end.
2592 * If that is the case, we must be careful only to
2593 * remove the part of the preimage that falls within
2594 * the boundaries of img. Initialize preimage_limit
2595 * to the number of lines in the preimage that falls
2596 * within the boundaries.
2598 preimage_limit = preimage->nr;
2599 if (preimage_limit > img->nr - applied_pos)
2600 preimage_limit = img->nr - applied_pos;
2602 for (i = 0; i < applied_pos; i++)
2603 applied_at += img->line[i].len;
2605 remove_count = 0;
2606 for (i = 0; i < preimage_limit; i++)
2607 remove_count += img->line[applied_pos + i].len;
2608 insert_count = postimage->len;
2610 /* Adjust the contents */
2611 result = xmalloc(img->len + insert_count - remove_count + 1);
2612 memcpy(result, img->buf, applied_at);
2613 memcpy(result + applied_at, postimage->buf, postimage->len);
2614 memcpy(result + applied_at + postimage->len,
2615 img->buf + (applied_at + remove_count),
2616 img->len - (applied_at + remove_count));
2617 free(img->buf);
2618 img->buf = result;
2619 img->len += insert_count - remove_count;
2620 result[img->len] = '\0';
2622 /* Adjust the line table */
2623 nr = img->nr + postimage->nr - preimage_limit;
2624 if (preimage_limit < postimage->nr) {
2626 * NOTE: this knows that we never call remove_first_line()
2627 * on anything other than pre/post image.
2629 REALLOC_ARRAY(img->line, nr);
2630 img->line_allocated = img->line;
2632 if (preimage_limit != postimage->nr)
2633 memmove(img->line + applied_pos + postimage->nr,
2634 img->line + applied_pos + preimage_limit,
2635 (img->nr - (applied_pos + preimage_limit)) *
2636 sizeof(*img->line));
2637 memcpy(img->line + applied_pos,
2638 postimage->line,
2639 postimage->nr * sizeof(*img->line));
2640 if (!allow_overlap)
2641 for (i = 0; i < postimage->nr; i++)
2642 img->line[applied_pos + i].flag |= LINE_PATCHED;
2643 img->nr = nr;
2647 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2648 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2649 * replace the part of "img" with "postimage" text.
2651 static int apply_one_fragment(struct image *img, struct fragment *frag,
2652 int inaccurate_eof, unsigned ws_rule,
2653 int nth_fragment)
2655 int match_beginning, match_end;
2656 const char *patch = frag->patch;
2657 int size = frag->size;
2658 char *old, *oldlines;
2659 struct strbuf newlines;
2660 int new_blank_lines_at_end = 0;
2661 int found_new_blank_lines_at_end = 0;
2662 int hunk_linenr = frag->linenr;
2663 unsigned long leading, trailing;
2664 int pos, applied_pos;
2665 struct image preimage;
2666 struct image postimage;
2668 memset(&preimage, 0, sizeof(preimage));
2669 memset(&postimage, 0, sizeof(postimage));
2670 oldlines = xmalloc(size);
2671 strbuf_init(&newlines, size);
2673 old = oldlines;
2674 while (size > 0) {
2675 char first;
2676 int len = linelen(patch, size);
2677 int plen;
2678 int added_blank_line = 0;
2679 int is_blank_context = 0;
2680 size_t start;
2682 if (!len)
2683 break;
2686 * "plen" is how much of the line we should use for
2687 * the actual patch data. Normally we just remove the
2688 * first character on the line, but if the line is
2689 * followed by "\ No newline", then we also remove the
2690 * last one (which is the newline, of course).
2692 plen = len - 1;
2693 if (len < size && patch[len] == '\\')
2694 plen--;
2695 first = *patch;
2696 if (apply_in_reverse) {
2697 if (first == '-')
2698 first = '+';
2699 else if (first == '+')
2700 first = '-';
2703 switch (first) {
2704 case '\n':
2705 /* Newer GNU diff, empty context line */
2706 if (plen < 0)
2707 /* ... followed by '\No newline'; nothing */
2708 break;
2709 *old++ = '\n';
2710 strbuf_addch(&newlines, '\n');
2711 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2712 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2713 is_blank_context = 1;
2714 break;
2715 case ' ':
2716 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2717 ws_blank_line(patch + 1, plen, ws_rule))
2718 is_blank_context = 1;
2719 case '-':
2720 memcpy(old, patch + 1, plen);
2721 add_line_info(&preimage, old, plen,
2722 (first == ' ' ? LINE_COMMON : 0));
2723 old += plen;
2724 if (first == '-')
2725 break;
2726 /* Fall-through for ' ' */
2727 case '+':
2728 /* --no-add does not add new lines */
2729 if (first == '+' && no_add)
2730 break;
2732 start = newlines.len;
2733 if (first != '+' ||
2734 !whitespace_error ||
2735 ws_error_action != correct_ws_error) {
2736 strbuf_add(&newlines, patch + 1, plen);
2738 else {
2739 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2741 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2742 (first == '+' ? 0 : LINE_COMMON));
2743 if (first == '+' &&
2744 (ws_rule & WS_BLANK_AT_EOF) &&
2745 ws_blank_line(patch + 1, plen, ws_rule))
2746 added_blank_line = 1;
2747 break;
2748 case '@': case '\\':
2749 /* Ignore it, we already handled it */
2750 break;
2751 default:
2752 if (apply_verbosely)
2753 error(_("invalid start of line: '%c'"), first);
2754 return -1;
2756 if (added_blank_line) {
2757 if (!new_blank_lines_at_end)
2758 found_new_blank_lines_at_end = hunk_linenr;
2759 new_blank_lines_at_end++;
2761 else if (is_blank_context)
2763 else
2764 new_blank_lines_at_end = 0;
2765 patch += len;
2766 size -= len;
2767 hunk_linenr++;
2769 if (inaccurate_eof &&
2770 old > oldlines && old[-1] == '\n' &&
2771 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2772 old--;
2773 strbuf_setlen(&newlines, newlines.len - 1);
2776 leading = frag->leading;
2777 trailing = frag->trailing;
2780 * A hunk to change lines at the beginning would begin with
2781 * @@ -1,L +N,M @@
2782 * but we need to be careful. -U0 that inserts before the second
2783 * line also has this pattern.
2785 * And a hunk to add to an empty file would begin with
2786 * @@ -0,0 +N,M @@
2788 * In other words, a hunk that is (frag->oldpos <= 1) with or
2789 * without leading context must match at the beginning.
2791 match_beginning = (!frag->oldpos ||
2792 (frag->oldpos == 1 && !unidiff_zero));
2795 * A hunk without trailing lines must match at the end.
2796 * However, we simply cannot tell if a hunk must match end
2797 * from the lack of trailing lines if the patch was generated
2798 * with unidiff without any context.
2800 match_end = !unidiff_zero && !trailing;
2802 pos = frag->newpos ? (frag->newpos - 1) : 0;
2803 preimage.buf = oldlines;
2804 preimage.len = old - oldlines;
2805 postimage.buf = newlines.buf;
2806 postimage.len = newlines.len;
2807 preimage.line = preimage.line_allocated;
2808 postimage.line = postimage.line_allocated;
2810 for (;;) {
2812 applied_pos = find_pos(img, &preimage, &postimage, pos,
2813 ws_rule, match_beginning, match_end);
2815 if (applied_pos >= 0)
2816 break;
2818 /* Am I at my context limits? */
2819 if ((leading <= p_context) && (trailing <= p_context))
2820 break;
2821 if (match_beginning || match_end) {
2822 match_beginning = match_end = 0;
2823 continue;
2827 * Reduce the number of context lines; reduce both
2828 * leading and trailing if they are equal otherwise
2829 * just reduce the larger context.
2831 if (leading >= trailing) {
2832 remove_first_line(&preimage);
2833 remove_first_line(&postimage);
2834 pos--;
2835 leading--;
2837 if (trailing > leading) {
2838 remove_last_line(&preimage);
2839 remove_last_line(&postimage);
2840 trailing--;
2844 if (applied_pos >= 0) {
2845 if (new_blank_lines_at_end &&
2846 preimage.nr + applied_pos >= img->nr &&
2847 (ws_rule & WS_BLANK_AT_EOF) &&
2848 ws_error_action != nowarn_ws_error) {
2849 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2850 found_new_blank_lines_at_end);
2851 if (ws_error_action == correct_ws_error) {
2852 while (new_blank_lines_at_end--)
2853 remove_last_line(&postimage);
2856 * We would want to prevent write_out_results()
2857 * from taking place in apply_patch() that follows
2858 * the callchain led us here, which is:
2859 * apply_patch->check_patch_list->check_patch->
2860 * apply_data->apply_fragments->apply_one_fragment
2862 if (ws_error_action == die_on_ws_error)
2863 apply = 0;
2866 if (apply_verbosely && applied_pos != pos) {
2867 int offset = applied_pos - pos;
2868 if (apply_in_reverse)
2869 offset = 0 - offset;
2870 fprintf_ln(stderr,
2871 Q_("Hunk #%d succeeded at %d (offset %d line).",
2872 "Hunk #%d succeeded at %d (offset %d lines).",
2873 offset),
2874 nth_fragment, applied_pos + 1, offset);
2878 * Warn if it was necessary to reduce the number
2879 * of context lines.
2881 if ((leading != frag->leading) ||
2882 (trailing != frag->trailing))
2883 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2884 " to apply fragment at %d"),
2885 leading, trailing, applied_pos+1);
2886 update_image(img, applied_pos, &preimage, &postimage);
2887 } else {
2888 if (apply_verbosely)
2889 error(_("while searching for:\n%.*s"),
2890 (int)(old - oldlines), oldlines);
2893 free(oldlines);
2894 strbuf_release(&newlines);
2895 free(preimage.line_allocated);
2896 free(postimage.line_allocated);
2898 return (applied_pos < 0);
2901 static int apply_binary_fragment(struct image *img, struct patch *patch)
2903 struct fragment *fragment = patch->fragments;
2904 unsigned long len;
2905 void *dst;
2907 if (!fragment)
2908 return error(_("missing binary patch data for '%s'"),
2909 patch->new_name ?
2910 patch->new_name :
2911 patch->old_name);
2913 /* Binary patch is irreversible without the optional second hunk */
2914 if (apply_in_reverse) {
2915 if (!fragment->next)
2916 return error("cannot reverse-apply a binary patch "
2917 "without the reverse hunk to '%s'",
2918 patch->new_name
2919 ? patch->new_name : patch->old_name);
2920 fragment = fragment->next;
2922 switch (fragment->binary_patch_method) {
2923 case BINARY_DELTA_DEFLATED:
2924 dst = patch_delta(img->buf, img->len, fragment->patch,
2925 fragment->size, &len);
2926 if (!dst)
2927 return -1;
2928 clear_image(img);
2929 img->buf = dst;
2930 img->len = len;
2931 return 0;
2932 case BINARY_LITERAL_DEFLATED:
2933 clear_image(img);
2934 img->len = fragment->size;
2935 img->buf = xmemdupz(fragment->patch, img->len);
2936 return 0;
2938 return -1;
2942 * Replace "img" with the result of applying the binary patch.
2943 * The binary patch data itself in patch->fragment is still kept
2944 * but the preimage prepared by the caller in "img" is freed here
2945 * or in the helper function apply_binary_fragment() this calls.
2947 static int apply_binary(struct image *img, struct patch *patch)
2949 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2950 unsigned char sha1[20];
2953 * For safety, we require patch index line to contain
2954 * full 40-byte textual SHA1 for old and new, at least for now.
2956 if (strlen(patch->old_sha1_prefix) != 40 ||
2957 strlen(patch->new_sha1_prefix) != 40 ||
2958 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2959 get_sha1_hex(patch->new_sha1_prefix, sha1))
2960 return error("cannot apply binary patch to '%s' "
2961 "without full index line", name);
2963 if (patch->old_name) {
2965 * See if the old one matches what the patch
2966 * applies to.
2968 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2969 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2970 return error("the patch applies to '%s' (%s), "
2971 "which does not match the "
2972 "current contents.",
2973 name, sha1_to_hex(sha1));
2975 else {
2976 /* Otherwise, the old one must be empty. */
2977 if (img->len)
2978 return error("the patch applies to an empty "
2979 "'%s' but it is not empty", name);
2982 get_sha1_hex(patch->new_sha1_prefix, sha1);
2983 if (is_null_sha1(sha1)) {
2984 clear_image(img);
2985 return 0; /* deletion patch */
2988 if (has_sha1_file(sha1)) {
2989 /* We already have the postimage */
2990 enum object_type type;
2991 unsigned long size;
2992 char *result;
2994 result = read_sha1_file(sha1, &type, &size);
2995 if (!result)
2996 return error("the necessary postimage %s for "
2997 "'%s' cannot be read",
2998 patch->new_sha1_prefix, name);
2999 clear_image(img);
3000 img->buf = result;
3001 img->len = size;
3002 } else {
3004 * We have verified buf matches the preimage;
3005 * apply the patch data to it, which is stored
3006 * in the patch->fragments->{patch,size}.
3008 if (apply_binary_fragment(img, patch))
3009 return error(_("binary patch does not apply to '%s'"),
3010 name);
3012 /* verify that the result matches */
3013 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3014 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3015 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3016 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3019 return 0;
3022 static int apply_fragments(struct image *img, struct patch *patch)
3024 struct fragment *frag = patch->fragments;
3025 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3026 unsigned ws_rule = patch->ws_rule;
3027 unsigned inaccurate_eof = patch->inaccurate_eof;
3028 int nth = 0;
3030 if (patch->is_binary)
3031 return apply_binary(img, patch);
3033 while (frag) {
3034 nth++;
3035 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3036 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3037 if (!apply_with_reject)
3038 return -1;
3039 frag->rejected = 1;
3041 frag = frag->next;
3043 return 0;
3046 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3048 if (S_ISGITLINK(mode)) {
3049 strbuf_grow(buf, 100);
3050 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3051 } else {
3052 enum object_type type;
3053 unsigned long sz;
3054 char *result;
3056 result = read_sha1_file(sha1, &type, &sz);
3057 if (!result)
3058 return -1;
3059 /* XXX read_sha1_file NUL-terminates */
3060 strbuf_attach(buf, result, sz, sz + 1);
3062 return 0;
3065 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3067 if (!ce)
3068 return 0;
3069 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3072 static struct patch *in_fn_table(const char *name)
3074 struct string_list_item *item;
3076 if (name == NULL)
3077 return NULL;
3079 item = string_list_lookup(&fn_table, name);
3080 if (item != NULL)
3081 return (struct patch *)item->util;
3083 return NULL;
3087 * item->util in the filename table records the status of the path.
3088 * Usually it points at a patch (whose result records the contents
3089 * of it after applying it), but it could be PATH_WAS_DELETED for a
3090 * path that a previously applied patch has already removed, or
3091 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3093 * The latter is needed to deal with a case where two paths A and B
3094 * are swapped by first renaming A to B and then renaming B to A;
3095 * moving A to B should not be prevented due to presence of B as we
3096 * will remove it in a later patch.
3098 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3099 #define PATH_WAS_DELETED ((struct patch *) -1)
3101 static int to_be_deleted(struct patch *patch)
3103 return patch == PATH_TO_BE_DELETED;
3106 static int was_deleted(struct patch *patch)
3108 return patch == PATH_WAS_DELETED;
3111 static void add_to_fn_table(struct patch *patch)
3113 struct string_list_item *item;
3116 * Always add new_name unless patch is a deletion
3117 * This should cover the cases for normal diffs,
3118 * file creations and copies
3120 if (patch->new_name != NULL) {
3121 item = string_list_insert(&fn_table, patch->new_name);
3122 item->util = patch;
3126 * store a failure on rename/deletion cases because
3127 * later chunks shouldn't patch old names
3129 if ((patch->new_name == NULL) || (patch->is_rename)) {
3130 item = string_list_insert(&fn_table, patch->old_name);
3131 item->util = PATH_WAS_DELETED;
3135 static void prepare_fn_table(struct patch *patch)
3138 * store information about incoming file deletion
3140 while (patch) {
3141 if ((patch->new_name == NULL) || (patch->is_rename)) {
3142 struct string_list_item *item;
3143 item = string_list_insert(&fn_table, patch->old_name);
3144 item->util = PATH_TO_BE_DELETED;
3146 patch = patch->next;
3150 static int checkout_target(struct index_state *istate,
3151 struct cache_entry *ce, struct stat *st)
3153 struct checkout costate;
3155 memset(&costate, 0, sizeof(costate));
3156 costate.base_dir = "";
3157 costate.refresh_cache = 1;
3158 costate.istate = istate;
3159 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3160 return error(_("cannot checkout %s"), ce->name);
3161 return 0;
3164 static struct patch *previous_patch(struct patch *patch, int *gone)
3166 struct patch *previous;
3168 *gone = 0;
3169 if (patch->is_copy || patch->is_rename)
3170 return NULL; /* "git" patches do not depend on the order */
3172 previous = in_fn_table(patch->old_name);
3173 if (!previous)
3174 return NULL;
3176 if (to_be_deleted(previous))
3177 return NULL; /* the deletion hasn't happened yet */
3179 if (was_deleted(previous))
3180 *gone = 1;
3182 return previous;
3185 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3187 if (S_ISGITLINK(ce->ce_mode)) {
3188 if (!S_ISDIR(st->st_mode))
3189 return -1;
3190 return 0;
3192 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3195 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3197 static int load_patch_target(struct strbuf *buf,
3198 const struct cache_entry *ce,
3199 struct stat *st,
3200 const char *name,
3201 unsigned expected_mode)
3203 if (cached) {
3204 if (read_file_or_gitlink(ce, buf))
3205 return error(_("read of %s failed"), name);
3206 } else if (name) {
3207 if (S_ISGITLINK(expected_mode)) {
3208 if (ce)
3209 return read_file_or_gitlink(ce, buf);
3210 else
3211 return SUBMODULE_PATCH_WITHOUT_INDEX;
3212 } else {
3213 if (read_old_data(st, name, buf))
3214 return error(_("read of %s failed"), name);
3217 return 0;
3221 * We are about to apply "patch"; populate the "image" with the
3222 * current version we have, from the working tree or from the index,
3223 * depending on the situation e.g. --cached/--index. If we are
3224 * applying a non-git patch that incrementally updates the tree,
3225 * we read from the result of a previous diff.
3227 static int load_preimage(struct image *image,
3228 struct patch *patch, struct stat *st,
3229 const struct cache_entry *ce)
3231 struct strbuf buf = STRBUF_INIT;
3232 size_t len;
3233 char *img;
3234 struct patch *previous;
3235 int status;
3237 previous = previous_patch(patch, &status);
3238 if (status)
3239 return error(_("path %s has been renamed/deleted"),
3240 patch->old_name);
3241 if (previous) {
3242 /* We have a patched copy in memory; use that. */
3243 strbuf_add(&buf, previous->result, previous->resultsize);
3244 } else {
3245 status = load_patch_target(&buf, ce, st,
3246 patch->old_name, patch->old_mode);
3247 if (status < 0)
3248 return status;
3249 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3251 * There is no way to apply subproject
3252 * patch without looking at the index.
3253 * NEEDSWORK: shouldn't this be flagged
3254 * as an error???
3256 free_fragment_list(patch->fragments);
3257 patch->fragments = NULL;
3258 } else if (status) {
3259 return error(_("read of %s failed"), patch->old_name);
3263 img = strbuf_detach(&buf, &len);
3264 prepare_image(image, img, len, !patch->is_binary);
3265 return 0;
3268 static int three_way_merge(struct image *image,
3269 char *path,
3270 const unsigned char *base,
3271 const unsigned char *ours,
3272 const unsigned char *theirs)
3274 mmfile_t base_file, our_file, their_file;
3275 mmbuffer_t result = { NULL };
3276 int status;
3278 read_mmblob(&base_file, base);
3279 read_mmblob(&our_file, ours);
3280 read_mmblob(&their_file, theirs);
3281 status = ll_merge(&result, path,
3282 &base_file, "base",
3283 &our_file, "ours",
3284 &their_file, "theirs", NULL);
3285 free(base_file.ptr);
3286 free(our_file.ptr);
3287 free(their_file.ptr);
3288 if (status < 0 || !result.ptr) {
3289 free(result.ptr);
3290 return -1;
3292 clear_image(image);
3293 image->buf = result.ptr;
3294 image->len = result.size;
3296 return status;
3300 * When directly falling back to add/add three-way merge, we read from
3301 * the current contents of the new_name. In no cases other than that
3302 * this function will be called.
3304 static int load_current(struct image *image, struct patch *patch)
3306 struct strbuf buf = STRBUF_INIT;
3307 int status, pos;
3308 size_t len;
3309 char *img;
3310 struct stat st;
3311 struct cache_entry *ce;
3312 char *name = patch->new_name;
3313 unsigned mode = patch->new_mode;
3315 if (!patch->is_new)
3316 die("BUG: patch to %s is not a creation", patch->old_name);
3318 pos = cache_name_pos(name, strlen(name));
3319 if (pos < 0)
3320 return error(_("%s: does not exist in index"), name);
3321 ce = active_cache[pos];
3322 if (lstat(name, &st)) {
3323 if (errno != ENOENT)
3324 return error(_("%s: %s"), name, strerror(errno));
3325 if (checkout_target(&the_index, ce, &st))
3326 return -1;
3328 if (verify_index_match(ce, &st))
3329 return error(_("%s: does not match index"), name);
3331 status = load_patch_target(&buf, ce, &st, name, mode);
3332 if (status < 0)
3333 return status;
3334 else if (status)
3335 return -1;
3336 img = strbuf_detach(&buf, &len);
3337 prepare_image(image, img, len, !patch->is_binary);
3338 return 0;
3341 static int try_threeway(struct image *image, struct patch *patch,
3342 struct stat *st, const struct cache_entry *ce)
3344 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3345 struct strbuf buf = STRBUF_INIT;
3346 size_t len;
3347 int status;
3348 char *img;
3349 struct image tmp_image;
3351 /* No point falling back to 3-way merge in these cases */
3352 if (patch->is_delete ||
3353 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3354 return -1;
3356 /* Preimage the patch was prepared for */
3357 if (patch->is_new)
3358 write_sha1_file("", 0, blob_type, pre_sha1);
3359 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3360 read_blob_object(&buf, pre_sha1, patch->old_mode))
3361 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3363 fprintf(stderr, "Falling back to three-way merge...\n");
3365 img = strbuf_detach(&buf, &len);
3366 prepare_image(&tmp_image, img, len, 1);
3367 /* Apply the patch to get the post image */
3368 if (apply_fragments(&tmp_image, patch) < 0) {
3369 clear_image(&tmp_image);
3370 return -1;
3372 /* post_sha1[] is theirs */
3373 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3374 clear_image(&tmp_image);
3376 /* our_sha1[] is ours */
3377 if (patch->is_new) {
3378 if (load_current(&tmp_image, patch))
3379 return error("cannot read the current contents of '%s'",
3380 patch->new_name);
3381 } else {
3382 if (load_preimage(&tmp_image, patch, st, ce))
3383 return error("cannot read the current contents of '%s'",
3384 patch->old_name);
3386 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3387 clear_image(&tmp_image);
3389 /* in-core three-way merge between post and our using pre as base */
3390 status = three_way_merge(image, patch->new_name,
3391 pre_sha1, our_sha1, post_sha1);
3392 if (status < 0) {
3393 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3394 return status;
3397 if (status) {
3398 patch->conflicted_threeway = 1;
3399 if (patch->is_new)
3400 hashclr(patch->threeway_stage[0]);
3401 else
3402 hashcpy(patch->threeway_stage[0], pre_sha1);
3403 hashcpy(patch->threeway_stage[1], our_sha1);
3404 hashcpy(patch->threeway_stage[2], post_sha1);
3405 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3406 } else {
3407 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3409 return 0;
3412 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3414 struct image image;
3416 if (load_preimage(&image, patch, st, ce) < 0)
3417 return -1;
3419 if (patch->direct_to_threeway ||
3420 apply_fragments(&image, patch) < 0) {
3421 /* Note: with --reject, apply_fragments() returns 0 */
3422 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3423 return -1;
3425 patch->result = image.buf;
3426 patch->resultsize = image.len;
3427 add_to_fn_table(patch);
3428 free(image.line_allocated);
3430 if (0 < patch->is_delete && patch->resultsize)
3431 return error(_("removal patch leaves file contents"));
3433 return 0;
3437 * If "patch" that we are looking at modifies or deletes what we have,
3438 * we would want it not to lose any local modification we have, either
3439 * in the working tree or in the index.
3441 * This also decides if a non-git patch is a creation patch or a
3442 * modification to an existing empty file. We do not check the state
3443 * of the current tree for a creation patch in this function; the caller
3444 * check_patch() separately makes sure (and errors out otherwise) that
3445 * the path the patch creates does not exist in the current tree.
3447 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3449 const char *old_name = patch->old_name;
3450 struct patch *previous = NULL;
3451 int stat_ret = 0, status;
3452 unsigned st_mode = 0;
3454 if (!old_name)
3455 return 0;
3457 assert(patch->is_new <= 0);
3458 previous = previous_patch(patch, &status);
3460 if (status)
3461 return error(_("path %s has been renamed/deleted"), old_name);
3462 if (previous) {
3463 st_mode = previous->new_mode;
3464 } else if (!cached) {
3465 stat_ret = lstat(old_name, st);
3466 if (stat_ret && errno != ENOENT)
3467 return error(_("%s: %s"), old_name, strerror(errno));
3470 if (check_index && !previous) {
3471 int pos = cache_name_pos(old_name, strlen(old_name));
3472 if (pos < 0) {
3473 if (patch->is_new < 0)
3474 goto is_new;
3475 return error(_("%s: does not exist in index"), old_name);
3477 *ce = active_cache[pos];
3478 if (stat_ret < 0) {
3479 if (checkout_target(&the_index, *ce, st))
3480 return -1;
3482 if (!cached && verify_index_match(*ce, st))
3483 return error(_("%s: does not match index"), old_name);
3484 if (cached)
3485 st_mode = (*ce)->ce_mode;
3486 } else if (stat_ret < 0) {
3487 if (patch->is_new < 0)
3488 goto is_new;
3489 return error(_("%s: %s"), old_name, strerror(errno));
3492 if (!cached && !previous)
3493 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3495 if (patch->is_new < 0)
3496 patch->is_new = 0;
3497 if (!patch->old_mode)
3498 patch->old_mode = st_mode;
3499 if ((st_mode ^ patch->old_mode) & S_IFMT)
3500 return error(_("%s: wrong type"), old_name);
3501 if (st_mode != patch->old_mode)
3502 warning(_("%s has type %o, expected %o"),
3503 old_name, st_mode, patch->old_mode);
3504 if (!patch->new_mode && !patch->is_delete)
3505 patch->new_mode = st_mode;
3506 return 0;
3508 is_new:
3509 patch->is_new = 1;
3510 patch->is_delete = 0;
3511 free(patch->old_name);
3512 patch->old_name = NULL;
3513 return 0;
3517 #define EXISTS_IN_INDEX 1
3518 #define EXISTS_IN_WORKTREE 2
3520 static int check_to_create(const char *new_name, int ok_if_exists)
3522 struct stat nst;
3524 if (check_index &&
3525 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3526 !ok_if_exists)
3527 return EXISTS_IN_INDEX;
3528 if (cached)
3529 return 0;
3531 if (!lstat(new_name, &nst)) {
3532 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3533 return 0;
3535 * A leading component of new_name might be a symlink
3536 * that is going to be removed with this patch, but
3537 * still pointing at somewhere that has the path.
3538 * In such a case, path "new_name" does not exist as
3539 * far as git is concerned.
3541 if (has_symlink_leading_path(new_name, strlen(new_name)))
3542 return 0;
3544 return EXISTS_IN_WORKTREE;
3545 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3546 return error("%s: %s", new_name, strerror(errno));
3548 return 0;
3552 * Check and apply the patch in-core; leave the result in patch->result
3553 * for the caller to write it out to the final destination.
3555 static int check_patch(struct patch *patch)
3557 struct stat st;
3558 const char *old_name = patch->old_name;
3559 const char *new_name = patch->new_name;
3560 const char *name = old_name ? old_name : new_name;
3561 struct cache_entry *ce = NULL;
3562 struct patch *tpatch;
3563 int ok_if_exists;
3564 int status;
3566 patch->rejected = 1; /* we will drop this after we succeed */
3568 status = check_preimage(patch, &ce, &st);
3569 if (status)
3570 return status;
3571 old_name = patch->old_name;
3574 * A type-change diff is always split into a patch to delete
3575 * old, immediately followed by a patch to create new (see
3576 * diff.c::run_diff()); in such a case it is Ok that the entry
3577 * to be deleted by the previous patch is still in the working
3578 * tree and in the index.
3580 * A patch to swap-rename between A and B would first rename A
3581 * to B and then rename B to A. While applying the first one,
3582 * the presence of B should not stop A from getting renamed to
3583 * B; ask to_be_deleted() about the later rename. Removal of
3584 * B and rename from A to B is handled the same way by asking
3585 * was_deleted().
3587 if ((tpatch = in_fn_table(new_name)) &&
3588 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3589 ok_if_exists = 1;
3590 else
3591 ok_if_exists = 0;
3593 if (new_name &&
3594 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3595 int err = check_to_create(new_name, ok_if_exists);
3597 if (err && threeway) {
3598 patch->direct_to_threeway = 1;
3599 } else switch (err) {
3600 case 0:
3601 break; /* happy */
3602 case EXISTS_IN_INDEX:
3603 return error(_("%s: already exists in index"), new_name);
3604 break;
3605 case EXISTS_IN_WORKTREE:
3606 return error(_("%s: already exists in working directory"),
3607 new_name);
3608 default:
3609 return err;
3612 if (!patch->new_mode) {
3613 if (0 < patch->is_new)
3614 patch->new_mode = S_IFREG | 0644;
3615 else
3616 patch->new_mode = patch->old_mode;
3620 if (new_name && old_name) {
3621 int same = !strcmp(old_name, new_name);
3622 if (!patch->new_mode)
3623 patch->new_mode = patch->old_mode;
3624 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3625 if (same)
3626 return error(_("new mode (%o) of %s does not "
3627 "match old mode (%o)"),
3628 patch->new_mode, new_name,
3629 patch->old_mode);
3630 else
3631 return error(_("new mode (%o) of %s does not "
3632 "match old mode (%o) of %s"),
3633 patch->new_mode, new_name,
3634 patch->old_mode, old_name);
3638 if (apply_data(patch, &st, ce) < 0)
3639 return error(_("%s: patch does not apply"), name);
3640 patch->rejected = 0;
3641 return 0;
3644 static int check_patch_list(struct patch *patch)
3646 int err = 0;
3648 prepare_fn_table(patch);
3649 while (patch) {
3650 if (apply_verbosely)
3651 say_patch_name(stderr,
3652 _("Checking patch %s..."), patch);
3653 err |= check_patch(patch);
3654 patch = patch->next;
3656 return err;
3659 /* This function tries to read the sha1 from the current index */
3660 static int get_current_sha1(const char *path, unsigned char *sha1)
3662 int pos;
3664 if (read_cache() < 0)
3665 return -1;
3666 pos = cache_name_pos(path, strlen(path));
3667 if (pos < 0)
3668 return -1;
3669 hashcpy(sha1, active_cache[pos]->sha1);
3670 return 0;
3673 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3676 * A usable gitlink patch has only one fragment (hunk) that looks like:
3677 * @@ -1 +1 @@
3678 * -Subproject commit <old sha1>
3679 * +Subproject commit <new sha1>
3680 * or
3681 * @@ -1 +0,0 @@
3682 * -Subproject commit <old sha1>
3683 * for a removal patch.
3685 struct fragment *hunk = p->fragments;
3686 static const char heading[] = "-Subproject commit ";
3687 char *preimage;
3689 if (/* does the patch have only one hunk? */
3690 hunk && !hunk->next &&
3691 /* is its preimage one line? */
3692 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3693 /* does preimage begin with the heading? */
3694 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3695 starts_with(++preimage, heading) &&
3696 /* does it record full SHA-1? */
3697 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3698 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3699 /* does the abbreviated name on the index line agree with it? */
3700 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3701 return 0; /* it all looks fine */
3703 /* we may have full object name on the index line */
3704 return get_sha1_hex(p->old_sha1_prefix, sha1);
3707 /* Build an index that contains the just the files needed for a 3way merge */
3708 static void build_fake_ancestor(struct patch *list, const char *filename)
3710 struct patch *patch;
3711 struct index_state result = { NULL };
3712 static struct lock_file lock;
3714 /* Once we start supporting the reverse patch, it may be
3715 * worth showing the new sha1 prefix, but until then...
3717 for (patch = list; patch; patch = patch->next) {
3718 unsigned char sha1[20];
3719 struct cache_entry *ce;
3720 const char *name;
3722 name = patch->old_name ? patch->old_name : patch->new_name;
3723 if (0 < patch->is_new)
3724 continue;
3726 if (S_ISGITLINK(patch->old_mode)) {
3727 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3728 ; /* ok, the textual part looks sane */
3729 else
3730 die("sha1 information is lacking or useless for submoule %s",
3731 name);
3732 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3733 ; /* ok */
3734 } else if (!patch->lines_added && !patch->lines_deleted) {
3735 /* mode-only change: update the current */
3736 if (get_current_sha1(patch->old_name, sha1))
3737 die("mode change for %s, which is not "
3738 "in current HEAD", name);
3739 } else
3740 die("sha1 information is lacking or useless "
3741 "(%s).", name);
3743 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3744 if (!ce)
3745 die(_("make_cache_entry failed for path '%s'"), name);
3746 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3747 die ("Could not add %s to temporary index", name);
3750 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3751 if (write_locked_index(&result, &lock, COMMIT_LOCK))
3752 die ("Could not write temporary index to %s", filename);
3754 discard_index(&result);
3757 static void stat_patch_list(struct patch *patch)
3759 int files, adds, dels;
3761 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3762 files++;
3763 adds += patch->lines_added;
3764 dels += patch->lines_deleted;
3765 show_stats(patch);
3768 print_stat_summary(stdout, files, adds, dels);
3771 static void numstat_patch_list(struct patch *patch)
3773 for ( ; patch; patch = patch->next) {
3774 const char *name;
3775 name = patch->new_name ? patch->new_name : patch->old_name;
3776 if (patch->is_binary)
3777 printf("-\t-\t");
3778 else
3779 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3780 write_name_quoted(name, stdout, line_termination);
3784 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3786 if (mode)
3787 printf(" %s mode %06o %s\n", newdelete, mode, name);
3788 else
3789 printf(" %s %s\n", newdelete, name);
3792 static void show_mode_change(struct patch *p, int show_name)
3794 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3795 if (show_name)
3796 printf(" mode change %06o => %06o %s\n",
3797 p->old_mode, p->new_mode, p->new_name);
3798 else
3799 printf(" mode change %06o => %06o\n",
3800 p->old_mode, p->new_mode);
3804 static void show_rename_copy(struct patch *p)
3806 const char *renamecopy = p->is_rename ? "rename" : "copy";
3807 const char *old, *new;
3809 /* Find common prefix */
3810 old = p->old_name;
3811 new = p->new_name;
3812 while (1) {
3813 const char *slash_old, *slash_new;
3814 slash_old = strchr(old, '/');
3815 slash_new = strchr(new, '/');
3816 if (!slash_old ||
3817 !slash_new ||
3818 slash_old - old != slash_new - new ||
3819 memcmp(old, new, slash_new - new))
3820 break;
3821 old = slash_old + 1;
3822 new = slash_new + 1;
3824 /* p->old_name thru old is the common prefix, and old and new
3825 * through the end of names are renames
3827 if (old != p->old_name)
3828 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3829 (int)(old - p->old_name), p->old_name,
3830 old, new, p->score);
3831 else
3832 printf(" %s %s => %s (%d%%)\n", renamecopy,
3833 p->old_name, p->new_name, p->score);
3834 show_mode_change(p, 0);
3837 static void summary_patch_list(struct patch *patch)
3839 struct patch *p;
3841 for (p = patch; p; p = p->next) {
3842 if (p->is_new)
3843 show_file_mode_name("create", p->new_mode, p->new_name);
3844 else if (p->is_delete)
3845 show_file_mode_name("delete", p->old_mode, p->old_name);
3846 else {
3847 if (p->is_rename || p->is_copy)
3848 show_rename_copy(p);
3849 else {
3850 if (p->score) {
3851 printf(" rewrite %s (%d%%)\n",
3852 p->new_name, p->score);
3853 show_mode_change(p, 0);
3855 else
3856 show_mode_change(p, 1);
3862 static void patch_stats(struct patch *patch)
3864 int lines = patch->lines_added + patch->lines_deleted;
3866 if (lines > max_change)
3867 max_change = lines;
3868 if (patch->old_name) {
3869 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3870 if (!len)
3871 len = strlen(patch->old_name);
3872 if (len > max_len)
3873 max_len = len;
3875 if (patch->new_name) {
3876 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3877 if (!len)
3878 len = strlen(patch->new_name);
3879 if (len > max_len)
3880 max_len = len;
3884 static void remove_file(struct patch *patch, int rmdir_empty)
3886 if (update_index) {
3887 if (remove_file_from_cache(patch->old_name) < 0)
3888 die(_("unable to remove %s from index"), patch->old_name);
3890 if (!cached) {
3891 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3892 remove_path(patch->old_name);
3897 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3899 struct stat st;
3900 struct cache_entry *ce;
3901 int namelen = strlen(path);
3902 unsigned ce_size = cache_entry_size(namelen);
3904 if (!update_index)
3905 return;
3907 ce = xcalloc(1, ce_size);
3908 memcpy(ce->name, path, namelen);
3909 ce->ce_mode = create_ce_mode(mode);
3910 ce->ce_flags = create_ce_flags(0);
3911 ce->ce_namelen = namelen;
3912 if (S_ISGITLINK(mode)) {
3913 const char *s;
3915 if (!skip_prefix(buf, "Subproject commit ", &s) ||
3916 get_sha1_hex(s, ce->sha1))
3917 die(_("corrupt patch for submodule %s"), path);
3918 } else {
3919 if (!cached) {
3920 if (lstat(path, &st) < 0)
3921 die_errno(_("unable to stat newly created file '%s'"),
3922 path);
3923 fill_stat_cache_info(ce, &st);
3925 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3926 die(_("unable to create backing store for newly created file %s"), path);
3928 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3929 die(_("unable to add cache entry for %s"), path);
3932 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3934 int fd;
3935 struct strbuf nbuf = STRBUF_INIT;
3937 if (S_ISGITLINK(mode)) {
3938 struct stat st;
3939 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3940 return 0;
3941 return mkdir(path, 0777);
3944 if (has_symlinks && S_ISLNK(mode))
3945 /* Although buf:size is counted string, it also is NUL
3946 * terminated.
3948 return symlink(buf, path);
3950 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3951 if (fd < 0)
3952 return -1;
3954 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3955 size = nbuf.len;
3956 buf = nbuf.buf;
3958 write_or_die(fd, buf, size);
3959 strbuf_release(&nbuf);
3961 if (close(fd) < 0)
3962 die_errno(_("closing file '%s'"), path);
3963 return 0;
3967 * We optimistically assume that the directories exist,
3968 * which is true 99% of the time anyway. If they don't,
3969 * we create them and try again.
3971 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3973 if (cached)
3974 return;
3975 if (!try_create_file(path, mode, buf, size))
3976 return;
3978 if (errno == ENOENT) {
3979 if (safe_create_leading_directories(path))
3980 return;
3981 if (!try_create_file(path, mode, buf, size))
3982 return;
3985 if (errno == EEXIST || errno == EACCES) {
3986 /* We may be trying to create a file where a directory
3987 * used to be.
3989 struct stat st;
3990 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3991 errno = EEXIST;
3994 if (errno == EEXIST) {
3995 unsigned int nr = getpid();
3997 for (;;) {
3998 char newpath[PATH_MAX];
3999 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4000 if (!try_create_file(newpath, mode, buf, size)) {
4001 if (!rename(newpath, path))
4002 return;
4003 unlink_or_warn(newpath);
4004 break;
4006 if (errno != EEXIST)
4007 break;
4008 ++nr;
4011 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4014 static void add_conflicted_stages_file(struct patch *patch)
4016 int stage, namelen;
4017 unsigned ce_size, mode;
4018 struct cache_entry *ce;
4020 if (!update_index)
4021 return;
4022 namelen = strlen(patch->new_name);
4023 ce_size = cache_entry_size(namelen);
4024 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4026 remove_file_from_cache(patch->new_name);
4027 for (stage = 1; stage < 4; stage++) {
4028 if (is_null_sha1(patch->threeway_stage[stage - 1]))
4029 continue;
4030 ce = xcalloc(1, ce_size);
4031 memcpy(ce->name, patch->new_name, namelen);
4032 ce->ce_mode = create_ce_mode(mode);
4033 ce->ce_flags = create_ce_flags(stage);
4034 ce->ce_namelen = namelen;
4035 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
4036 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4037 die(_("unable to add cache entry for %s"), patch->new_name);
4041 static void create_file(struct patch *patch)
4043 char *path = patch->new_name;
4044 unsigned mode = patch->new_mode;
4045 unsigned long size = patch->resultsize;
4046 char *buf = patch->result;
4048 if (!mode)
4049 mode = S_IFREG | 0644;
4050 create_one_file(path, mode, buf, size);
4052 if (patch->conflicted_threeway)
4053 add_conflicted_stages_file(patch);
4054 else
4055 add_index_file(path, mode, buf, size);
4058 /* phase zero is to remove, phase one is to create */
4059 static void write_out_one_result(struct patch *patch, int phase)
4061 if (patch->is_delete > 0) {
4062 if (phase == 0)
4063 remove_file(patch, 1);
4064 return;
4066 if (patch->is_new > 0 || patch->is_copy) {
4067 if (phase == 1)
4068 create_file(patch);
4069 return;
4072 * Rename or modification boils down to the same
4073 * thing: remove the old, write the new
4075 if (phase == 0)
4076 remove_file(patch, patch->is_rename);
4077 if (phase == 1)
4078 create_file(patch);
4081 static int write_out_one_reject(struct patch *patch)
4083 FILE *rej;
4084 char namebuf[PATH_MAX];
4085 struct fragment *frag;
4086 int cnt = 0;
4087 struct strbuf sb = STRBUF_INIT;
4089 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4090 if (!frag->rejected)
4091 continue;
4092 cnt++;
4095 if (!cnt) {
4096 if (apply_verbosely)
4097 say_patch_name(stderr,
4098 _("Applied patch %s cleanly."), patch);
4099 return 0;
4102 /* This should not happen, because a removal patch that leaves
4103 * contents are marked "rejected" at the patch level.
4105 if (!patch->new_name)
4106 die(_("internal error"));
4108 /* Say this even without --verbose */
4109 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4110 "Applying patch %%s with %d rejects...",
4111 cnt),
4112 cnt);
4113 say_patch_name(stderr, sb.buf, patch);
4114 strbuf_release(&sb);
4116 cnt = strlen(patch->new_name);
4117 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4118 cnt = ARRAY_SIZE(namebuf) - 5;
4119 warning(_("truncating .rej filename to %.*s.rej"),
4120 cnt - 1, patch->new_name);
4122 memcpy(namebuf, patch->new_name, cnt);
4123 memcpy(namebuf + cnt, ".rej", 5);
4125 rej = fopen(namebuf, "w");
4126 if (!rej)
4127 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4129 /* Normal git tools never deal with .rej, so do not pretend
4130 * this is a git patch by saying --git or giving extended
4131 * headers. While at it, maybe please "kompare" that wants
4132 * the trailing TAB and some garbage at the end of line ;-).
4134 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4135 patch->new_name, patch->new_name);
4136 for (cnt = 1, frag = patch->fragments;
4137 frag;
4138 cnt++, frag = frag->next) {
4139 if (!frag->rejected) {
4140 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4141 continue;
4143 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4144 fprintf(rej, "%.*s", frag->size, frag->patch);
4145 if (frag->patch[frag->size-1] != '\n')
4146 fputc('\n', rej);
4148 fclose(rej);
4149 return -1;
4152 static int write_out_results(struct patch *list)
4154 int phase;
4155 int errs = 0;
4156 struct patch *l;
4157 struct string_list cpath = STRING_LIST_INIT_DUP;
4159 for (phase = 0; phase < 2; phase++) {
4160 l = list;
4161 while (l) {
4162 if (l->rejected)
4163 errs = 1;
4164 else {
4165 write_out_one_result(l, phase);
4166 if (phase == 1) {
4167 if (write_out_one_reject(l))
4168 errs = 1;
4169 if (l->conflicted_threeway) {
4170 string_list_append(&cpath, l->new_name);
4171 errs = 1;
4175 l = l->next;
4179 if (cpath.nr) {
4180 struct string_list_item *item;
4182 sort_string_list(&cpath);
4183 for_each_string_list_item(item, &cpath)
4184 fprintf(stderr, "U %s\n", item->string);
4185 string_list_clear(&cpath, 0);
4187 rerere(0);
4190 return errs;
4193 static struct lock_file lock_file;
4195 #define INACCURATE_EOF (1<<0)
4196 #define RECOUNT (1<<1)
4198 static int apply_patch(int fd, const char *filename, int options)
4200 size_t offset;
4201 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4202 struct patch *list = NULL, **listp = &list;
4203 int skipped_patch = 0;
4205 patch_input_file = filename;
4206 read_patch_file(&buf, fd);
4207 offset = 0;
4208 while (offset < buf.len) {
4209 struct patch *patch;
4210 int nr;
4212 patch = xcalloc(1, sizeof(*patch));
4213 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4214 patch->recount = !!(options & RECOUNT);
4215 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4216 if (nr < 0)
4217 break;
4218 if (apply_in_reverse)
4219 reverse_patches(patch);
4220 if (use_patch(patch)) {
4221 patch_stats(patch);
4222 *listp = patch;
4223 listp = &patch->next;
4225 else {
4226 free_patch(patch);
4227 skipped_patch++;
4229 offset += nr;
4232 if (!list && !skipped_patch)
4233 die(_("unrecognized input"));
4235 if (whitespace_error && (ws_error_action == die_on_ws_error))
4236 apply = 0;
4238 update_index = check_index && apply;
4239 if (update_index && newfd < 0)
4240 newfd = hold_locked_index(&lock_file, 1);
4242 if (check_index) {
4243 if (read_cache() < 0)
4244 die(_("unable to read index file"));
4247 if ((check || apply) &&
4248 check_patch_list(list) < 0 &&
4249 !apply_with_reject)
4250 exit(1);
4252 if (apply && write_out_results(list)) {
4253 if (apply_with_reject)
4254 exit(1);
4255 /* with --3way, we still need to write the index out */
4256 return 1;
4259 if (fake_ancestor)
4260 build_fake_ancestor(list, fake_ancestor);
4262 if (diffstat)
4263 stat_patch_list(list);
4265 if (numstat)
4266 numstat_patch_list(list);
4268 if (summary)
4269 summary_patch_list(list);
4271 free_patch_list(list);
4272 strbuf_release(&buf);
4273 string_list_clear(&fn_table, 0);
4274 return 0;
4277 static void git_apply_config(void)
4279 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4280 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4281 git_config(git_default_config, NULL);
4284 static int option_parse_exclude(const struct option *opt,
4285 const char *arg, int unset)
4287 add_name_limit(arg, 1);
4288 return 0;
4291 static int option_parse_include(const struct option *opt,
4292 const char *arg, int unset)
4294 add_name_limit(arg, 0);
4295 has_include = 1;
4296 return 0;
4299 static int option_parse_p(const struct option *opt,
4300 const char *arg, int unset)
4302 p_value = atoi(arg);
4303 p_value_known = 1;
4304 return 0;
4307 static int option_parse_z(const struct option *opt,
4308 const char *arg, int unset)
4310 if (unset)
4311 line_termination = '\n';
4312 else
4313 line_termination = 0;
4314 return 0;
4317 static int option_parse_space_change(const struct option *opt,
4318 const char *arg, int unset)
4320 if (unset)
4321 ws_ignore_action = ignore_ws_none;
4322 else
4323 ws_ignore_action = ignore_ws_change;
4324 return 0;
4327 static int option_parse_whitespace(const struct option *opt,
4328 const char *arg, int unset)
4330 const char **whitespace_option = opt->value;
4332 *whitespace_option = arg;
4333 parse_whitespace_option(arg);
4334 return 0;
4337 static int option_parse_directory(const struct option *opt,
4338 const char *arg, int unset)
4340 root_len = strlen(arg);
4341 if (root_len && arg[root_len - 1] != '/') {
4342 char *new_root;
4343 root = new_root = xmalloc(root_len + 2);
4344 strcpy(new_root, arg);
4345 strcpy(new_root + root_len++, "/");
4346 } else
4347 root = arg;
4348 return 0;
4351 int cmd_apply(int argc, const char **argv, const char *prefix_)
4353 int i;
4354 int errs = 0;
4355 int is_not_gitdir = !startup_info->have_repository;
4356 int force_apply = 0;
4358 const char *whitespace_option = NULL;
4360 struct option builtin_apply_options[] = {
4361 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4362 N_("don't apply changes matching the given path"),
4363 0, option_parse_exclude },
4364 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4365 N_("apply changes matching the given path"),
4366 0, option_parse_include },
4367 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4368 N_("remove <num> leading slashes from traditional diff paths"),
4369 0, option_parse_p },
4370 OPT_BOOL(0, "no-add", &no_add,
4371 N_("ignore additions made by the patch")),
4372 OPT_BOOL(0, "stat", &diffstat,
4373 N_("instead of applying the patch, output diffstat for the input")),
4374 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4375 OPT_NOOP_NOARG(0, "binary"),
4376 OPT_BOOL(0, "numstat", &numstat,
4377 N_("show number of added and deleted lines in decimal notation")),
4378 OPT_BOOL(0, "summary", &summary,
4379 N_("instead of applying the patch, output a summary for the input")),
4380 OPT_BOOL(0, "check", &check,
4381 N_("instead of applying the patch, see if the patch is applicable")),
4382 OPT_BOOL(0, "index", &check_index,
4383 N_("make sure the patch is applicable to the current index")),
4384 OPT_BOOL(0, "cached", &cached,
4385 N_("apply a patch without touching the working tree")),
4386 OPT_BOOL(0, "apply", &force_apply,
4387 N_("also apply the patch (use with --stat/--summary/--check)")),
4388 OPT_BOOL('3', "3way", &threeway,
4389 N_( "attempt three-way merge if a patch does not apply")),
4390 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4391 N_("build a temporary index based on embedded index information")),
4392 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4393 N_("paths are separated with NUL character"),
4394 PARSE_OPT_NOARG, option_parse_z },
4395 OPT_INTEGER('C', NULL, &p_context,
4396 N_("ensure at least <n> lines of context match")),
4397 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4398 N_("detect new or modified lines that have whitespace errors"),
4399 0, option_parse_whitespace },
4400 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4401 N_("ignore changes in whitespace when finding context"),
4402 PARSE_OPT_NOARG, option_parse_space_change },
4403 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4404 N_("ignore changes in whitespace when finding context"),
4405 PARSE_OPT_NOARG, option_parse_space_change },
4406 OPT_BOOL('R', "reverse", &apply_in_reverse,
4407 N_("apply the patch in reverse")),
4408 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4409 N_("don't expect at least one line of context")),
4410 OPT_BOOL(0, "reject", &apply_with_reject,
4411 N_("leave the rejected hunks in corresponding *.rej files")),
4412 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4413 N_("allow overlapping hunks")),
4414 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4415 OPT_BIT(0, "inaccurate-eof", &options,
4416 N_("tolerate incorrectly detected missing new-line at the end of file"),
4417 INACCURATE_EOF),
4418 OPT_BIT(0, "recount", &options,
4419 N_("do not trust the line counts in the hunk headers"),
4420 RECOUNT),
4421 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4422 N_("prepend <root> to all filenames"),
4423 0, option_parse_directory },
4424 OPT_END()
4427 prefix = prefix_;
4428 prefix_length = prefix ? strlen(prefix) : 0;
4429 git_apply_config();
4430 if (apply_default_whitespace)
4431 parse_whitespace_option(apply_default_whitespace);
4432 if (apply_default_ignorewhitespace)
4433 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4435 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4436 apply_usage, 0);
4438 if (apply_with_reject && threeway)
4439 die("--reject and --3way cannot be used together.");
4440 if (cached && threeway)
4441 die("--cached and --3way cannot be used together.");
4442 if (threeway) {
4443 if (is_not_gitdir)
4444 die(_("--3way outside a repository"));
4445 check_index = 1;
4447 if (apply_with_reject)
4448 apply = apply_verbosely = 1;
4449 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4450 apply = 0;
4451 if (check_index && is_not_gitdir)
4452 die(_("--index outside a repository"));
4453 if (cached) {
4454 if (is_not_gitdir)
4455 die(_("--cached outside a repository"));
4456 check_index = 1;
4458 for (i = 0; i < argc; i++) {
4459 const char *arg = argv[i];
4460 int fd;
4462 if (!strcmp(arg, "-")) {
4463 errs |= apply_patch(0, "<stdin>", options);
4464 read_stdin = 0;
4465 continue;
4466 } else if (0 < prefix_length)
4467 arg = prefix_filename(prefix, prefix_length, arg);
4469 fd = open(arg, O_RDONLY);
4470 if (fd < 0)
4471 die_errno(_("can't open patch '%s'"), arg);
4472 read_stdin = 0;
4473 set_default_whitespace_mode(whitespace_option);
4474 errs |= apply_patch(fd, arg, options);
4475 close(fd);
4477 set_default_whitespace_mode(whitespace_option);
4478 if (read_stdin)
4479 errs |= apply_patch(0, "<stdin>", options);
4480 if (whitespace_error) {
4481 if (squelch_whitespace_errors &&
4482 squelch_whitespace_errors < whitespace_error) {
4483 int squelched =
4484 whitespace_error - squelch_whitespace_errors;
4485 warning(Q_("squelched %d whitespace error",
4486 "squelched %d whitespace errors",
4487 squelched),
4488 squelched);
4490 if (ws_error_action == die_on_ws_error)
4491 die(Q_("%d line adds whitespace errors.",
4492 "%d lines add whitespace errors.",
4493 whitespace_error),
4494 whitespace_error);
4495 if (applied_after_fixing_ws && apply)
4496 warning("%d line%s applied after"
4497 " fixing whitespace errors.",
4498 applied_after_fixing_ws,
4499 applied_after_fixing_ws == 1 ? "" : "s");
4500 else if (whitespace_error)
4501 warning(Q_("%d line adds whitespace errors.",
4502 "%d lines add whitespace errors.",
4503 whitespace_error),
4504 whitespace_error);
4507 if (update_index) {
4508 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4509 die(_("Unable to write new index file"));
4512 return !!errs;