submodule: teach rm to remove submodules unless they contain a git directory
[git/jnareb-git.git] / builtin / apply.c
blob3bf71dc4522564463945673aaba0501a51141bb2
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 whitespace */
304 while (isspace(*s1) && (s1 <= last1))
305 s1++;
306 while (isspace(*s2) && (s2 <= last2))
307 s2++;
308 /* early return if both lines are empty */
309 if ((s1 > last1) && (s2 > last2))
310 return 1;
311 while (!result) {
312 result = *s1++ - *s2++;
314 * Skip whitespace inside. We check for whitespace on
315 * both buffers because we don't want "a b" to match
316 * "ab"
318 if (isspace(*s1) && isspace(*s2)) {
319 while (isspace(*s1) && s1 <= last1)
320 s1++;
321 while (isspace(*s2) && s2 <= last2)
322 s2++;
325 * If we reached the end on one side only,
326 * lines don't match
328 if (
329 ((s2 > last2) && (s1 <= last1)) ||
330 ((s1 > last1) && (s2 <= last2)))
331 return 0;
332 if ((s1 > last1) && (s2 > last2))
333 break;
336 return !result;
339 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
341 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
342 img->line_allocated[img->nr].len = len;
343 img->line_allocated[img->nr].hash = hash_line(bol, len);
344 img->line_allocated[img->nr].flag = flag;
345 img->nr++;
349 * "buf" has the file contents to be patched (read from various sources).
350 * attach it to "image" and add line-based index to it.
351 * "image" now owns the "buf".
353 static void prepare_image(struct image *image, char *buf, size_t len,
354 int prepare_linetable)
356 const char *cp, *ep;
358 memset(image, 0, sizeof(*image));
359 image->buf = buf;
360 image->len = len;
362 if (!prepare_linetable)
363 return;
365 ep = image->buf + image->len;
366 cp = image->buf;
367 while (cp < ep) {
368 const char *next;
369 for (next = cp; next < ep && *next != '\n'; next++)
371 if (next < ep)
372 next++;
373 add_line_info(image, cp, next - cp, 0);
374 cp = next;
376 image->line = image->line_allocated;
379 static void clear_image(struct image *image)
381 free(image->buf);
382 free(image->line_allocated);
383 memset(image, 0, sizeof(*image));
386 /* fmt must contain _one_ %s and no other substitution */
387 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
389 struct strbuf sb = STRBUF_INIT;
391 if (patch->old_name && patch->new_name &&
392 strcmp(patch->old_name, patch->new_name)) {
393 quote_c_style(patch->old_name, &sb, NULL, 0);
394 strbuf_addstr(&sb, " => ");
395 quote_c_style(patch->new_name, &sb, NULL, 0);
396 } else {
397 const char *n = patch->new_name;
398 if (!n)
399 n = patch->old_name;
400 quote_c_style(n, &sb, NULL, 0);
402 fprintf(output, fmt, sb.buf);
403 fputc('\n', output);
404 strbuf_release(&sb);
407 #define SLOP (16)
409 static void read_patch_file(struct strbuf *sb, int fd)
411 if (strbuf_read(sb, fd, 0) < 0)
412 die_errno("git apply: failed to read");
415 * Make sure that we have some slop in the buffer
416 * so that we can do speculative "memcmp" etc, and
417 * see to it that it is NUL-filled.
419 strbuf_grow(sb, SLOP);
420 memset(sb->buf + sb->len, 0, SLOP);
423 static unsigned long linelen(const char *buffer, unsigned long size)
425 unsigned long len = 0;
426 while (size--) {
427 len++;
428 if (*buffer++ == '\n')
429 break;
431 return len;
434 static int is_dev_null(const char *str)
436 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
439 #define TERM_SPACE 1
440 #define TERM_TAB 2
442 static int name_terminate(const char *name, int namelen, int c, int terminate)
444 if (c == ' ' && !(terminate & TERM_SPACE))
445 return 0;
446 if (c == '\t' && !(terminate & TERM_TAB))
447 return 0;
449 return 1;
452 /* remove double slashes to make --index work with such filenames */
453 static char *squash_slash(char *name)
455 int i = 0, j = 0;
457 if (!name)
458 return NULL;
460 while (name[i]) {
461 if ((name[j++] = name[i++]) == '/')
462 while (name[i] == '/')
463 i++;
465 name[j] = '\0';
466 return name;
469 static char *find_name_gnu(const char *line, const char *def, int p_value)
471 struct strbuf name = STRBUF_INIT;
472 char *cp;
475 * Proposed "new-style" GNU patch/diff format; see
476 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
478 if (unquote_c_style(&name, line, NULL)) {
479 strbuf_release(&name);
480 return NULL;
483 for (cp = name.buf; p_value; p_value--) {
484 cp = strchr(cp, '/');
485 if (!cp) {
486 strbuf_release(&name);
487 return NULL;
489 cp++;
492 strbuf_remove(&name, 0, cp - name.buf);
493 if (root)
494 strbuf_insert(&name, 0, root, root_len);
495 return squash_slash(strbuf_detach(&name, NULL));
498 static size_t sane_tz_len(const char *line, size_t len)
500 const char *tz, *p;
502 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
503 return 0;
504 tz = line + len - strlen(" +0500");
506 if (tz[1] != '+' && tz[1] != '-')
507 return 0;
509 for (p = tz + 2; p != line + len; p++)
510 if (!isdigit(*p))
511 return 0;
513 return line + len - tz;
516 static size_t tz_with_colon_len(const char *line, size_t len)
518 const char *tz, *p;
520 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
521 return 0;
522 tz = line + len - strlen(" +08:00");
524 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
525 return 0;
526 p = tz + 2;
527 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
528 !isdigit(*p++) || !isdigit(*p++))
529 return 0;
531 return line + len - tz;
534 static size_t date_len(const char *line, size_t len)
536 const char *date, *p;
538 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
539 return 0;
540 p = date = line + len - strlen("72-02-05");
542 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
543 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
544 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
545 return 0;
547 if (date - line >= strlen("19") &&
548 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
549 date -= strlen("19");
551 return line + len - date;
554 static size_t short_time_len(const char *line, size_t len)
556 const char *time, *p;
558 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
559 return 0;
560 p = time = line + len - strlen(" 07:01:32");
562 /* Permit 1-digit hours? */
563 if (*p++ != ' ' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
565 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
566 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
567 return 0;
569 return line + len - time;
572 static size_t fractional_time_len(const char *line, size_t len)
574 const char *p;
575 size_t n;
577 /* Expected format: 19:41:17.620000023 */
578 if (!len || !isdigit(line[len - 1]))
579 return 0;
580 p = line + len - 1;
582 /* Fractional seconds. */
583 while (p > line && isdigit(*p))
584 p--;
585 if (*p != '.')
586 return 0;
588 /* Hours, minutes, and whole seconds. */
589 n = short_time_len(line, p - line);
590 if (!n)
591 return 0;
593 return line + len - p + n;
596 static size_t trailing_spaces_len(const char *line, size_t len)
598 const char *p;
600 /* Expected format: ' ' x (1 or more) */
601 if (!len || line[len - 1] != ' ')
602 return 0;
604 p = line + len;
605 while (p != line) {
606 p--;
607 if (*p != ' ')
608 return line + len - (p + 1);
611 /* All spaces! */
612 return len;
615 static size_t diff_timestamp_len(const char *line, size_t len)
617 const char *end = line + len;
618 size_t n;
621 * Posix: 2010-07-05 19:41:17
622 * GNU: 2010-07-05 19:41:17.620000023 -0500
625 if (!isdigit(end[-1]))
626 return 0;
628 n = sane_tz_len(line, end - line);
629 if (!n)
630 n = tz_with_colon_len(line, end - line);
631 end -= n;
633 n = short_time_len(line, end - line);
634 if (!n)
635 n = fractional_time_len(line, end - line);
636 end -= n;
638 n = date_len(line, end - line);
639 if (!n) /* No date. Too bad. */
640 return 0;
641 end -= n;
643 if (end == line) /* No space before date. */
644 return 0;
645 if (end[-1] == '\t') { /* Success! */
646 end--;
647 return line + len - end;
649 if (end[-1] != ' ') /* No space before date. */
650 return 0;
652 /* Whitespace damage. */
653 end -= trailing_spaces_len(line, end - line);
654 return line + len - end;
657 static char *null_strdup(const char *s)
659 return s ? xstrdup(s) : NULL;
662 static char *find_name_common(const char *line, const char *def,
663 int p_value, const char *end, int terminate)
665 int len;
666 const char *start = NULL;
668 if (p_value == 0)
669 start = line;
670 while (line != end) {
671 char c = *line;
673 if (!end && isspace(c)) {
674 if (c == '\n')
675 break;
676 if (name_terminate(start, line-start, c, terminate))
677 break;
679 line++;
680 if (c == '/' && !--p_value)
681 start = line;
683 if (!start)
684 return squash_slash(null_strdup(def));
685 len = line - start;
686 if (!len)
687 return squash_slash(null_strdup(def));
690 * Generally we prefer the shorter name, especially
691 * if the other one is just a variation of that with
692 * something else tacked on to the end (ie "file.orig"
693 * or "file~").
695 if (def) {
696 int deflen = strlen(def);
697 if (deflen < len && !strncmp(start, def, deflen))
698 return squash_slash(xstrdup(def));
701 if (root) {
702 char *ret = xmalloc(root_len + len + 1);
703 strcpy(ret, root);
704 memcpy(ret + root_len, start, len);
705 ret[root_len + len] = '\0';
706 return squash_slash(ret);
709 return squash_slash(xmemdupz(start, len));
712 static char *find_name(const char *line, char *def, int p_value, int terminate)
714 if (*line == '"') {
715 char *name = find_name_gnu(line, def, p_value);
716 if (name)
717 return name;
720 return find_name_common(line, def, p_value, NULL, terminate);
723 static char *find_name_traditional(const char *line, char *def, int p_value)
725 size_t len = strlen(line);
726 size_t date_len;
728 if (*line == '"') {
729 char *name = find_name_gnu(line, def, p_value);
730 if (name)
731 return name;
734 len = strchrnul(line, '\n') - line;
735 date_len = diff_timestamp_len(line, len);
736 if (!date_len)
737 return find_name_common(line, def, p_value, NULL, TERM_TAB);
738 len -= date_len;
740 return find_name_common(line, def, p_value, line + len, 0);
743 static int count_slashes(const char *cp)
745 int cnt = 0;
746 char ch;
748 while ((ch = *cp++))
749 if (ch == '/')
750 cnt++;
751 return cnt;
755 * Given the string after "--- " or "+++ ", guess the appropriate
756 * p_value for the given patch.
758 static int guess_p_value(const char *nameline)
760 char *name, *cp;
761 int val = -1;
763 if (is_dev_null(nameline))
764 return -1;
765 name = find_name_traditional(nameline, NULL, 0);
766 if (!name)
767 return -1;
768 cp = strchr(name, '/');
769 if (!cp)
770 val = 0;
771 else if (prefix) {
773 * Does it begin with "a/$our-prefix" and such? Then this is
774 * very likely to apply to our directory.
776 if (!strncmp(name, prefix, prefix_length))
777 val = count_slashes(prefix);
778 else {
779 cp++;
780 if (!strncmp(cp, prefix, prefix_length))
781 val = count_slashes(prefix) + 1;
784 free(name);
785 return val;
789 * Does the ---/+++ line has the POSIX timestamp after the last HT?
790 * GNU diff puts epoch there to signal a creation/deletion event. Is
791 * this such a timestamp?
793 static int has_epoch_timestamp(const char *nameline)
796 * We are only interested in epoch timestamp; any non-zero
797 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
798 * For the same reason, the date must be either 1969-12-31 or
799 * 1970-01-01, and the seconds part must be "00".
801 const char stamp_regexp[] =
802 "^(1969-12-31|1970-01-01)"
804 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
806 "([-+][0-2][0-9]:?[0-5][0-9])\n";
807 const char *timestamp = NULL, *cp, *colon;
808 static regex_t *stamp;
809 regmatch_t m[10];
810 int zoneoffset;
811 int hourminute;
812 int status;
814 for (cp = nameline; *cp != '\n'; cp++) {
815 if (*cp == '\t')
816 timestamp = cp + 1;
818 if (!timestamp)
819 return 0;
820 if (!stamp) {
821 stamp = xmalloc(sizeof(*stamp));
822 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
823 warning(_("Cannot prepare timestamp regexp %s"),
824 stamp_regexp);
825 return 0;
829 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
830 if (status) {
831 if (status != REG_NOMATCH)
832 warning(_("regexec returned %d for input: %s"),
833 status, timestamp);
834 return 0;
837 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
838 if (*colon == ':')
839 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
840 else
841 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
842 if (timestamp[m[3].rm_so] == '-')
843 zoneoffset = -zoneoffset;
846 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
847 * (west of GMT) or 1970-01-01 (east of GMT)
849 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
850 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
851 return 0;
853 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
854 strtol(timestamp + 14, NULL, 10) -
855 zoneoffset);
857 return ((zoneoffset < 0 && hourminute == 1440) ||
858 (0 <= zoneoffset && !hourminute));
862 * Get the name etc info from the ---/+++ lines of a traditional patch header
864 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
865 * files, we can happily check the index for a match, but for creating a
866 * new file we should try to match whatever "patch" does. I have no idea.
868 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
870 char *name;
872 first += 4; /* skip "--- " */
873 second += 4; /* skip "+++ " */
874 if (!p_value_known) {
875 int p, q;
876 p = guess_p_value(first);
877 q = guess_p_value(second);
878 if (p < 0) p = q;
879 if (0 <= p && p == q) {
880 p_value = p;
881 p_value_known = 1;
884 if (is_dev_null(first)) {
885 patch->is_new = 1;
886 patch->is_delete = 0;
887 name = find_name_traditional(second, NULL, p_value);
888 patch->new_name = name;
889 } else if (is_dev_null(second)) {
890 patch->is_new = 0;
891 patch->is_delete = 1;
892 name = find_name_traditional(first, NULL, p_value);
893 patch->old_name = name;
894 } else {
895 char *first_name;
896 first_name = find_name_traditional(first, NULL, p_value);
897 name = find_name_traditional(second, first_name, p_value);
898 free(first_name);
899 if (has_epoch_timestamp(first)) {
900 patch->is_new = 1;
901 patch->is_delete = 0;
902 patch->new_name = name;
903 } else if (has_epoch_timestamp(second)) {
904 patch->is_new = 0;
905 patch->is_delete = 1;
906 patch->old_name = name;
907 } else {
908 patch->old_name = name;
909 patch->new_name = xstrdup(name);
912 if (!name)
913 die(_("unable to find filename in patch at line %d"), linenr);
916 static int gitdiff_hdrend(const char *line, struct patch *patch)
918 return -1;
922 * We're anal about diff header consistency, to make
923 * sure that we don't end up having strange ambiguous
924 * patches floating around.
926 * As a result, gitdiff_{old|new}name() will check
927 * their names against any previous information, just
928 * to make sure..
930 #define DIFF_OLD_NAME 0
931 #define DIFF_NEW_NAME 1
933 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
935 if (!orig_name && !isnull)
936 return find_name(line, NULL, p_value, TERM_TAB);
938 if (orig_name) {
939 int len;
940 const char *name;
941 char *another;
942 name = orig_name;
943 len = strlen(name);
944 if (isnull)
945 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
946 another = find_name(line, NULL, p_value, TERM_TAB);
947 if (!another || memcmp(another, name, len + 1))
948 die((side == DIFF_NEW_NAME) ?
949 _("git apply: bad git-diff - inconsistent new filename on line %d") :
950 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
951 free(another);
952 return orig_name;
954 else {
955 /* expect "/dev/null" */
956 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
957 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
958 return NULL;
962 static int gitdiff_oldname(const char *line, struct patch *patch)
964 char *orig = patch->old_name;
965 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
966 DIFF_OLD_NAME);
967 if (orig != patch->old_name)
968 free(orig);
969 return 0;
972 static int gitdiff_newname(const char *line, struct patch *patch)
974 char *orig = patch->new_name;
975 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
976 DIFF_NEW_NAME);
977 if (orig != patch->new_name)
978 free(orig);
979 return 0;
982 static int gitdiff_oldmode(const char *line, struct patch *patch)
984 patch->old_mode = strtoul(line, NULL, 8);
985 return 0;
988 static int gitdiff_newmode(const char *line, struct patch *patch)
990 patch->new_mode = strtoul(line, NULL, 8);
991 return 0;
994 static int gitdiff_delete(const char *line, struct patch *patch)
996 patch->is_delete = 1;
997 free(patch->old_name);
998 patch->old_name = null_strdup(patch->def_name);
999 return gitdiff_oldmode(line, patch);
1002 static int gitdiff_newfile(const char *line, struct patch *patch)
1004 patch->is_new = 1;
1005 free(patch->new_name);
1006 patch->new_name = null_strdup(patch->def_name);
1007 return gitdiff_newmode(line, patch);
1010 static int gitdiff_copysrc(const char *line, struct patch *patch)
1012 patch->is_copy = 1;
1013 free(patch->old_name);
1014 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1015 return 0;
1018 static int gitdiff_copydst(const char *line, struct patch *patch)
1020 patch->is_copy = 1;
1021 free(patch->new_name);
1022 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1023 return 0;
1026 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1028 patch->is_rename = 1;
1029 free(patch->old_name);
1030 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1031 return 0;
1034 static int gitdiff_renamedst(const char *line, struct patch *patch)
1036 patch->is_rename = 1;
1037 free(patch->new_name);
1038 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1039 return 0;
1042 static int gitdiff_similarity(const char *line, struct patch *patch)
1044 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1045 patch->score = 0;
1046 return 0;
1049 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1051 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1052 patch->score = 0;
1053 return 0;
1056 static int gitdiff_index(const char *line, struct patch *patch)
1059 * index line is N hexadecimal, "..", N hexadecimal,
1060 * and optional space with octal mode.
1062 const char *ptr, *eol;
1063 int len;
1065 ptr = strchr(line, '.');
1066 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1067 return 0;
1068 len = ptr - line;
1069 memcpy(patch->old_sha1_prefix, line, len);
1070 patch->old_sha1_prefix[len] = 0;
1072 line = ptr + 2;
1073 ptr = strchr(line, ' ');
1074 eol = strchr(line, '\n');
1076 if (!ptr || eol < ptr)
1077 ptr = eol;
1078 len = ptr - line;
1080 if (40 < len)
1081 return 0;
1082 memcpy(patch->new_sha1_prefix, line, len);
1083 patch->new_sha1_prefix[len] = 0;
1084 if (*ptr == ' ')
1085 patch->old_mode = strtoul(ptr+1, NULL, 8);
1086 return 0;
1090 * This is normal for a diff that doesn't change anything: we'll fall through
1091 * into the next diff. Tell the parser to break out.
1093 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1095 return -1;
1098 static const char *stop_at_slash(const char *line, int llen)
1100 int nslash = p_value;
1101 int i;
1103 for (i = 0; i < llen; i++) {
1104 int ch = line[i];
1105 if (ch == '/' && --nslash <= 0)
1106 return &line[i];
1108 return NULL;
1112 * This is to extract the same name that appears on "diff --git"
1113 * line. We do not find and return anything if it is a rename
1114 * patch, and it is OK because we will find the name elsewhere.
1115 * We need to reliably find name only when it is mode-change only,
1116 * creation or deletion of an empty file. In any of these cases,
1117 * both sides are the same name under a/ and b/ respectively.
1119 static char *git_header_name(const char *line, int llen)
1121 const char *name;
1122 const char *second = NULL;
1123 size_t len, line_len;
1125 line += strlen("diff --git ");
1126 llen -= strlen("diff --git ");
1128 if (*line == '"') {
1129 const char *cp;
1130 struct strbuf first = STRBUF_INIT;
1131 struct strbuf sp = STRBUF_INIT;
1133 if (unquote_c_style(&first, line, &second))
1134 goto free_and_fail1;
1136 /* advance to the first slash */
1137 cp = stop_at_slash(first.buf, first.len);
1138 /* we do not accept absolute paths */
1139 if (!cp || cp == first.buf)
1140 goto free_and_fail1;
1141 strbuf_remove(&first, 0, cp + 1 - first.buf);
1144 * second points at one past closing dq of name.
1145 * find the second name.
1147 while ((second < line + llen) && isspace(*second))
1148 second++;
1150 if (line + llen <= second)
1151 goto free_and_fail1;
1152 if (*second == '"') {
1153 if (unquote_c_style(&sp, second, NULL))
1154 goto free_and_fail1;
1155 cp = stop_at_slash(sp.buf, sp.len);
1156 if (!cp || cp == sp.buf)
1157 goto free_and_fail1;
1158 /* They must match, otherwise ignore */
1159 if (strcmp(cp + 1, first.buf))
1160 goto free_and_fail1;
1161 strbuf_release(&sp);
1162 return strbuf_detach(&first, NULL);
1165 /* unquoted second */
1166 cp = stop_at_slash(second, line + llen - second);
1167 if (!cp || cp == second)
1168 goto free_and_fail1;
1169 cp++;
1170 if (line + llen - cp != first.len + 1 ||
1171 memcmp(first.buf, cp, first.len))
1172 goto free_and_fail1;
1173 return strbuf_detach(&first, NULL);
1175 free_and_fail1:
1176 strbuf_release(&first);
1177 strbuf_release(&sp);
1178 return NULL;
1181 /* unquoted first name */
1182 name = stop_at_slash(line, llen);
1183 if (!name || name == line)
1184 return NULL;
1185 name++;
1188 * since the first name is unquoted, a dq if exists must be
1189 * the beginning of the second name.
1191 for (second = name; second < line + llen; second++) {
1192 if (*second == '"') {
1193 struct strbuf sp = STRBUF_INIT;
1194 const char *np;
1196 if (unquote_c_style(&sp, second, NULL))
1197 goto free_and_fail2;
1199 np = stop_at_slash(sp.buf, sp.len);
1200 if (!np || np == sp.buf)
1201 goto free_and_fail2;
1202 np++;
1204 len = sp.buf + sp.len - np;
1205 if (len < second - name &&
1206 !strncmp(np, name, len) &&
1207 isspace(name[len])) {
1208 /* Good */
1209 strbuf_remove(&sp, 0, np - sp.buf);
1210 return strbuf_detach(&sp, NULL);
1213 free_and_fail2:
1214 strbuf_release(&sp);
1215 return NULL;
1220 * Accept a name only if it shows up twice, exactly the same
1221 * form.
1223 second = strchr(name, '\n');
1224 if (!second)
1225 return NULL;
1226 line_len = second - name;
1227 for (len = 0 ; ; len++) {
1228 switch (name[len]) {
1229 default:
1230 continue;
1231 case '\n':
1232 return NULL;
1233 case '\t': case ' ':
1234 second = stop_at_slash(name + len, line_len - len);
1235 if (!second)
1236 return NULL;
1237 second++;
1238 if (second[len] == '\n' && !strncmp(name, second, len)) {
1239 return xmemdupz(name, len);
1245 /* Verify that we recognize the lines following a git header */
1246 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1248 unsigned long offset;
1250 /* A git diff has explicit new/delete information, so we don't guess */
1251 patch->is_new = 0;
1252 patch->is_delete = 0;
1255 * Some things may not have the old name in the
1256 * rest of the headers anywhere (pure mode changes,
1257 * or removing or adding empty files), so we get
1258 * the default name from the header.
1260 patch->def_name = git_header_name(line, len);
1261 if (patch->def_name && root) {
1262 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1263 strcpy(s, root);
1264 strcpy(s + root_len, patch->def_name);
1265 free(patch->def_name);
1266 patch->def_name = s;
1269 line += len;
1270 size -= len;
1271 linenr++;
1272 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1273 static const struct opentry {
1274 const char *str;
1275 int (*fn)(const char *, struct patch *);
1276 } optable[] = {
1277 { "@@ -", gitdiff_hdrend },
1278 { "--- ", gitdiff_oldname },
1279 { "+++ ", gitdiff_newname },
1280 { "old mode ", gitdiff_oldmode },
1281 { "new mode ", gitdiff_newmode },
1282 { "deleted file mode ", gitdiff_delete },
1283 { "new file mode ", gitdiff_newfile },
1284 { "copy from ", gitdiff_copysrc },
1285 { "copy to ", gitdiff_copydst },
1286 { "rename old ", gitdiff_renamesrc },
1287 { "rename new ", gitdiff_renamedst },
1288 { "rename from ", gitdiff_renamesrc },
1289 { "rename to ", gitdiff_renamedst },
1290 { "similarity index ", gitdiff_similarity },
1291 { "dissimilarity index ", gitdiff_dissimilarity },
1292 { "index ", gitdiff_index },
1293 { "", gitdiff_unrecognized },
1295 int i;
1297 len = linelen(line, size);
1298 if (!len || line[len-1] != '\n')
1299 break;
1300 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1301 const struct opentry *p = optable + i;
1302 int oplen = strlen(p->str);
1303 if (len < oplen || memcmp(p->str, line, oplen))
1304 continue;
1305 if (p->fn(line + oplen, patch) < 0)
1306 return offset;
1307 break;
1311 return offset;
1314 static int parse_num(const char *line, unsigned long *p)
1316 char *ptr;
1318 if (!isdigit(*line))
1319 return 0;
1320 *p = strtoul(line, &ptr, 10);
1321 return ptr - line;
1324 static int parse_range(const char *line, int len, int offset, const char *expect,
1325 unsigned long *p1, unsigned long *p2)
1327 int digits, ex;
1329 if (offset < 0 || offset >= len)
1330 return -1;
1331 line += offset;
1332 len -= offset;
1334 digits = parse_num(line, p1);
1335 if (!digits)
1336 return -1;
1338 offset += digits;
1339 line += digits;
1340 len -= digits;
1342 *p2 = 1;
1343 if (*line == ',') {
1344 digits = parse_num(line+1, p2);
1345 if (!digits)
1346 return -1;
1348 offset += digits+1;
1349 line += digits+1;
1350 len -= digits+1;
1353 ex = strlen(expect);
1354 if (ex > len)
1355 return -1;
1356 if (memcmp(line, expect, ex))
1357 return -1;
1359 return offset + ex;
1362 static void recount_diff(const char *line, int size, struct fragment *fragment)
1364 int oldlines = 0, newlines = 0, ret = 0;
1366 if (size < 1) {
1367 warning("recount: ignore empty hunk");
1368 return;
1371 for (;;) {
1372 int len = linelen(line, size);
1373 size -= len;
1374 line += len;
1376 if (size < 1)
1377 break;
1379 switch (*line) {
1380 case ' ': case '\n':
1381 newlines++;
1382 /* fall through */
1383 case '-':
1384 oldlines++;
1385 continue;
1386 case '+':
1387 newlines++;
1388 continue;
1389 case '\\':
1390 continue;
1391 case '@':
1392 ret = size < 3 || prefixcmp(line, "@@ ");
1393 break;
1394 case 'd':
1395 ret = size < 5 || prefixcmp(line, "diff ");
1396 break;
1397 default:
1398 ret = -1;
1399 break;
1401 if (ret) {
1402 warning(_("recount: unexpected line: %.*s"),
1403 (int)linelen(line, size), line);
1404 return;
1406 break;
1408 fragment->oldlines = oldlines;
1409 fragment->newlines = newlines;
1413 * Parse a unified diff fragment header of the
1414 * form "@@ -a,b +c,d @@"
1416 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1418 int offset;
1420 if (!len || line[len-1] != '\n')
1421 return -1;
1423 /* Figure out the number of lines in a fragment */
1424 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1425 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1427 return offset;
1430 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1432 unsigned long offset, len;
1434 patch->is_toplevel_relative = 0;
1435 patch->is_rename = patch->is_copy = 0;
1436 patch->is_new = patch->is_delete = -1;
1437 patch->old_mode = patch->new_mode = 0;
1438 patch->old_name = patch->new_name = NULL;
1439 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1440 unsigned long nextlen;
1442 len = linelen(line, size);
1443 if (!len)
1444 break;
1446 /* Testing this early allows us to take a few shortcuts.. */
1447 if (len < 6)
1448 continue;
1451 * Make sure we don't find any unconnected patch fragments.
1452 * That's a sign that we didn't find a header, and that a
1453 * patch has become corrupted/broken up.
1455 if (!memcmp("@@ -", line, 4)) {
1456 struct fragment dummy;
1457 if (parse_fragment_header(line, len, &dummy) < 0)
1458 continue;
1459 die(_("patch fragment without header at line %d: %.*s"),
1460 linenr, (int)len-1, line);
1463 if (size < len + 6)
1464 break;
1467 * Git patch? It might not have a real patch, just a rename
1468 * or mode change, so we handle that specially
1470 if (!memcmp("diff --git ", line, 11)) {
1471 int git_hdr_len = parse_git_header(line, len, size, patch);
1472 if (git_hdr_len <= len)
1473 continue;
1474 if (!patch->old_name && !patch->new_name) {
1475 if (!patch->def_name)
1476 die(Q_("git diff header lacks filename information when removing "
1477 "%d leading pathname component (line %d)",
1478 "git diff header lacks filename information when removing "
1479 "%d leading pathname components (line %d)",
1480 p_value),
1481 p_value, linenr);
1482 patch->old_name = xstrdup(patch->def_name);
1483 patch->new_name = xstrdup(patch->def_name);
1485 if (!patch->is_delete && !patch->new_name)
1486 die("git diff header lacks filename information "
1487 "(line %d)", linenr);
1488 patch->is_toplevel_relative = 1;
1489 *hdrsize = git_hdr_len;
1490 return offset;
1493 /* --- followed by +++ ? */
1494 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1495 continue;
1498 * We only accept unified patches, so we want it to
1499 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1500 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1502 nextlen = linelen(line + len, size - len);
1503 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1504 continue;
1506 /* Ok, we'll consider it a patch */
1507 parse_traditional_patch(line, line+len, patch);
1508 *hdrsize = len + nextlen;
1509 linenr += 2;
1510 return offset;
1512 return -1;
1515 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1517 char *err;
1519 if (!result)
1520 return;
1522 whitespace_error++;
1523 if (squelch_whitespace_errors &&
1524 squelch_whitespace_errors < whitespace_error)
1525 return;
1527 err = whitespace_error_string(result);
1528 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1529 patch_input_file, linenr, err, len, line);
1530 free(err);
1533 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1535 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1537 record_ws_error(result, line + 1, len - 2, linenr);
1541 * Parse a unified diff. Note that this really needs to parse each
1542 * fragment separately, since the only way to know the difference
1543 * between a "---" that is part of a patch, and a "---" that starts
1544 * the next patch is to look at the line counts..
1546 static int parse_fragment(const char *line, unsigned long size,
1547 struct patch *patch, struct fragment *fragment)
1549 int added, deleted;
1550 int len = linelen(line, size), offset;
1551 unsigned long oldlines, newlines;
1552 unsigned long leading, trailing;
1554 offset = parse_fragment_header(line, len, fragment);
1555 if (offset < 0)
1556 return -1;
1557 if (offset > 0 && patch->recount)
1558 recount_diff(line + offset, size - offset, fragment);
1559 oldlines = fragment->oldlines;
1560 newlines = fragment->newlines;
1561 leading = 0;
1562 trailing = 0;
1564 /* Parse the thing.. */
1565 line += len;
1566 size -= len;
1567 linenr++;
1568 added = deleted = 0;
1569 for (offset = len;
1570 0 < size;
1571 offset += len, size -= len, line += len, linenr++) {
1572 if (!oldlines && !newlines)
1573 break;
1574 len = linelen(line, size);
1575 if (!len || line[len-1] != '\n')
1576 return -1;
1577 switch (*line) {
1578 default:
1579 return -1;
1580 case '\n': /* newer GNU diff, an empty context line */
1581 case ' ':
1582 oldlines--;
1583 newlines--;
1584 if (!deleted && !added)
1585 leading++;
1586 trailing++;
1587 break;
1588 case '-':
1589 if (apply_in_reverse &&
1590 ws_error_action != nowarn_ws_error)
1591 check_whitespace(line, len, patch->ws_rule);
1592 deleted++;
1593 oldlines--;
1594 trailing = 0;
1595 break;
1596 case '+':
1597 if (!apply_in_reverse &&
1598 ws_error_action != nowarn_ws_error)
1599 check_whitespace(line, len, patch->ws_rule);
1600 added++;
1601 newlines--;
1602 trailing = 0;
1603 break;
1606 * We allow "\ No newline at end of file". Depending
1607 * on locale settings when the patch was produced we
1608 * don't know what this line looks like. The only
1609 * thing we do know is that it begins with "\ ".
1610 * Checking for 12 is just for sanity check -- any
1611 * l10n of "\ No newline..." is at least that long.
1613 case '\\':
1614 if (len < 12 || memcmp(line, "\\ ", 2))
1615 return -1;
1616 break;
1619 if (oldlines || newlines)
1620 return -1;
1621 fragment->leading = leading;
1622 fragment->trailing = trailing;
1625 * If a fragment ends with an incomplete line, we failed to include
1626 * it in the above loop because we hit oldlines == newlines == 0
1627 * before seeing it.
1629 if (12 < size && !memcmp(line, "\\ ", 2))
1630 offset += linelen(line, size);
1632 patch->lines_added += added;
1633 patch->lines_deleted += deleted;
1635 if (0 < patch->is_new && oldlines)
1636 return error(_("new file depends on old contents"));
1637 if (0 < patch->is_delete && newlines)
1638 return error(_("deleted file still has contents"));
1639 return offset;
1643 * We have seen "diff --git a/... b/..." header (or a traditional patch
1644 * header). Read hunks that belong to this patch into fragments and hang
1645 * them to the given patch structure.
1647 * The (fragment->patch, fragment->size) pair points into the memory given
1648 * by the caller, not a copy, when we return.
1650 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1652 unsigned long offset = 0;
1653 unsigned long oldlines = 0, newlines = 0, context = 0;
1654 struct fragment **fragp = &patch->fragments;
1656 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1657 struct fragment *fragment;
1658 int len;
1660 fragment = xcalloc(1, sizeof(*fragment));
1661 fragment->linenr = linenr;
1662 len = parse_fragment(line, size, patch, fragment);
1663 if (len <= 0)
1664 die(_("corrupt patch at line %d"), linenr);
1665 fragment->patch = line;
1666 fragment->size = len;
1667 oldlines += fragment->oldlines;
1668 newlines += fragment->newlines;
1669 context += fragment->leading + fragment->trailing;
1671 *fragp = fragment;
1672 fragp = &fragment->next;
1674 offset += len;
1675 line += len;
1676 size -= len;
1680 * If something was removed (i.e. we have old-lines) it cannot
1681 * be creation, and if something was added it cannot be
1682 * deletion. However, the reverse is not true; --unified=0
1683 * patches that only add are not necessarily creation even
1684 * though they do not have any old lines, and ones that only
1685 * delete are not necessarily deletion.
1687 * Unfortunately, a real creation/deletion patch do _not_ have
1688 * any context line by definition, so we cannot safely tell it
1689 * apart with --unified=0 insanity. At least if the patch has
1690 * more than one hunk it is not creation or deletion.
1692 if (patch->is_new < 0 &&
1693 (oldlines || (patch->fragments && patch->fragments->next)))
1694 patch->is_new = 0;
1695 if (patch->is_delete < 0 &&
1696 (newlines || (patch->fragments && patch->fragments->next)))
1697 patch->is_delete = 0;
1699 if (0 < patch->is_new && oldlines)
1700 die(_("new file %s depends on old contents"), patch->new_name);
1701 if (0 < patch->is_delete && newlines)
1702 die(_("deleted file %s still has contents"), patch->old_name);
1703 if (!patch->is_delete && !newlines && context)
1704 fprintf_ln(stderr,
1705 _("** warning: "
1706 "file %s becomes empty but is not deleted"),
1707 patch->new_name);
1709 return offset;
1712 static inline int metadata_changes(struct patch *patch)
1714 return patch->is_rename > 0 ||
1715 patch->is_copy > 0 ||
1716 patch->is_new > 0 ||
1717 patch->is_delete ||
1718 (patch->old_mode && patch->new_mode &&
1719 patch->old_mode != patch->new_mode);
1722 static char *inflate_it(const void *data, unsigned long size,
1723 unsigned long inflated_size)
1725 git_zstream stream;
1726 void *out;
1727 int st;
1729 memset(&stream, 0, sizeof(stream));
1731 stream.next_in = (unsigned char *)data;
1732 stream.avail_in = size;
1733 stream.next_out = out = xmalloc(inflated_size);
1734 stream.avail_out = inflated_size;
1735 git_inflate_init(&stream);
1736 st = git_inflate(&stream, Z_FINISH);
1737 git_inflate_end(&stream);
1738 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1739 free(out);
1740 return NULL;
1742 return out;
1746 * Read a binary hunk and return a new fragment; fragment->patch
1747 * points at an allocated memory that the caller must free, so
1748 * it is marked as "->free_patch = 1".
1750 static struct fragment *parse_binary_hunk(char **buf_p,
1751 unsigned long *sz_p,
1752 int *status_p,
1753 int *used_p)
1756 * Expect a line that begins with binary patch method ("literal"
1757 * or "delta"), followed by the length of data before deflating.
1758 * a sequence of 'length-byte' followed by base-85 encoded data
1759 * should follow, terminated by a newline.
1761 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1762 * and we would limit the patch line to 66 characters,
1763 * so one line can fit up to 13 groups that would decode
1764 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1765 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1767 int llen, used;
1768 unsigned long size = *sz_p;
1769 char *buffer = *buf_p;
1770 int patch_method;
1771 unsigned long origlen;
1772 char *data = NULL;
1773 int hunk_size = 0;
1774 struct fragment *frag;
1776 llen = linelen(buffer, size);
1777 used = llen;
1779 *status_p = 0;
1781 if (!prefixcmp(buffer, "delta ")) {
1782 patch_method = BINARY_DELTA_DEFLATED;
1783 origlen = strtoul(buffer + 6, NULL, 10);
1785 else if (!prefixcmp(buffer, "literal ")) {
1786 patch_method = BINARY_LITERAL_DEFLATED;
1787 origlen = strtoul(buffer + 8, NULL, 10);
1789 else
1790 return NULL;
1792 linenr++;
1793 buffer += llen;
1794 while (1) {
1795 int byte_length, max_byte_length, newsize;
1796 llen = linelen(buffer, size);
1797 used += llen;
1798 linenr++;
1799 if (llen == 1) {
1800 /* consume the blank line */
1801 buffer++;
1802 size--;
1803 break;
1806 * Minimum line is "A00000\n" which is 7-byte long,
1807 * and the line length must be multiple of 5 plus 2.
1809 if ((llen < 7) || (llen-2) % 5)
1810 goto corrupt;
1811 max_byte_length = (llen - 2) / 5 * 4;
1812 byte_length = *buffer;
1813 if ('A' <= byte_length && byte_length <= 'Z')
1814 byte_length = byte_length - 'A' + 1;
1815 else if ('a' <= byte_length && byte_length <= 'z')
1816 byte_length = byte_length - 'a' + 27;
1817 else
1818 goto corrupt;
1819 /* if the input length was not multiple of 4, we would
1820 * have filler at the end but the filler should never
1821 * exceed 3 bytes
1823 if (max_byte_length < byte_length ||
1824 byte_length <= max_byte_length - 4)
1825 goto corrupt;
1826 newsize = hunk_size + byte_length;
1827 data = xrealloc(data, newsize);
1828 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1829 goto corrupt;
1830 hunk_size = newsize;
1831 buffer += llen;
1832 size -= llen;
1835 frag = xcalloc(1, sizeof(*frag));
1836 frag->patch = inflate_it(data, hunk_size, origlen);
1837 frag->free_patch = 1;
1838 if (!frag->patch)
1839 goto corrupt;
1840 free(data);
1841 frag->size = origlen;
1842 *buf_p = buffer;
1843 *sz_p = size;
1844 *used_p = used;
1845 frag->binary_patch_method = patch_method;
1846 return frag;
1848 corrupt:
1849 free(data);
1850 *status_p = -1;
1851 error(_("corrupt binary patch at line %d: %.*s"),
1852 linenr-1, llen-1, buffer);
1853 return NULL;
1856 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1859 * We have read "GIT binary patch\n"; what follows is a line
1860 * that says the patch method (currently, either "literal" or
1861 * "delta") and the length of data before deflating; a
1862 * sequence of 'length-byte' followed by base-85 encoded data
1863 * follows.
1865 * When a binary patch is reversible, there is another binary
1866 * hunk in the same format, starting with patch method (either
1867 * "literal" or "delta") with the length of data, and a sequence
1868 * of length-byte + base-85 encoded data, terminated with another
1869 * empty line. This data, when applied to the postimage, produces
1870 * the preimage.
1872 struct fragment *forward;
1873 struct fragment *reverse;
1874 int status;
1875 int used, used_1;
1877 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1878 if (!forward && !status)
1879 /* there has to be one hunk (forward hunk) */
1880 return error(_("unrecognized binary patch at line %d"), linenr-1);
1881 if (status)
1882 /* otherwise we already gave an error message */
1883 return status;
1885 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1886 if (reverse)
1887 used += used_1;
1888 else if (status) {
1890 * Not having reverse hunk is not an error, but having
1891 * a corrupt reverse hunk is.
1893 free((void*) forward->patch);
1894 free(forward);
1895 return status;
1897 forward->next = reverse;
1898 patch->fragments = forward;
1899 patch->is_binary = 1;
1900 return used;
1904 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1905 * reading after seeing a single patch (i.e. changes to a single file).
1906 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1907 * Return the number of bytes consumed, so that the caller can call us
1908 * again for the next patch.
1910 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1912 int hdrsize, patchsize;
1913 int offset = find_header(buffer, size, &hdrsize, patch);
1915 if (offset < 0)
1916 return offset;
1918 patch->ws_rule = whitespace_rule(patch->new_name
1919 ? patch->new_name
1920 : patch->old_name);
1922 patchsize = parse_single_patch(buffer + offset + hdrsize,
1923 size - offset - hdrsize, patch);
1925 if (!patchsize) {
1926 static const char *binhdr[] = {
1927 "Binary files ",
1928 "Files ",
1929 NULL,
1931 static const char git_binary[] = "GIT binary patch\n";
1932 int i;
1933 int hd = hdrsize + offset;
1934 unsigned long llen = linelen(buffer + hd, size - hd);
1936 if (llen == sizeof(git_binary) - 1 &&
1937 !memcmp(git_binary, buffer + hd, llen)) {
1938 int used;
1939 linenr++;
1940 used = parse_binary(buffer + hd + llen,
1941 size - hd - llen, patch);
1942 if (used)
1943 patchsize = used + llen;
1944 else
1945 patchsize = 0;
1947 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1948 for (i = 0; binhdr[i]; i++) {
1949 int len = strlen(binhdr[i]);
1950 if (len < size - hd &&
1951 !memcmp(binhdr[i], buffer + hd, len)) {
1952 linenr++;
1953 patch->is_binary = 1;
1954 patchsize = llen;
1955 break;
1960 /* Empty patch cannot be applied if it is a text patch
1961 * without metadata change. A binary patch appears
1962 * empty to us here.
1964 if ((apply || check) &&
1965 (!patch->is_binary && !metadata_changes(patch)))
1966 die(_("patch with only garbage at line %d"), linenr);
1969 return offset + hdrsize + patchsize;
1972 #define swap(a,b) myswap((a),(b),sizeof(a))
1974 #define myswap(a, b, size) do { \
1975 unsigned char mytmp[size]; \
1976 memcpy(mytmp, &a, size); \
1977 memcpy(&a, &b, size); \
1978 memcpy(&b, mytmp, size); \
1979 } while (0)
1981 static void reverse_patches(struct patch *p)
1983 for (; p; p = p->next) {
1984 struct fragment *frag = p->fragments;
1986 swap(p->new_name, p->old_name);
1987 swap(p->new_mode, p->old_mode);
1988 swap(p->is_new, p->is_delete);
1989 swap(p->lines_added, p->lines_deleted);
1990 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1992 for (; frag; frag = frag->next) {
1993 swap(frag->newpos, frag->oldpos);
1994 swap(frag->newlines, frag->oldlines);
1999 static const char pluses[] =
2000 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2001 static const char minuses[]=
2002 "----------------------------------------------------------------------";
2004 static void show_stats(struct patch *patch)
2006 struct strbuf qname = STRBUF_INIT;
2007 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2008 int max, add, del;
2010 quote_c_style(cp, &qname, NULL, 0);
2013 * "scale" the filename
2015 max = max_len;
2016 if (max > 50)
2017 max = 50;
2019 if (qname.len > max) {
2020 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2021 if (!cp)
2022 cp = qname.buf + qname.len + 3 - max;
2023 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2026 if (patch->is_binary) {
2027 printf(" %-*s | Bin\n", max, qname.buf);
2028 strbuf_release(&qname);
2029 return;
2032 printf(" %-*s |", max, qname.buf);
2033 strbuf_release(&qname);
2036 * scale the add/delete
2038 max = max + max_change > 70 ? 70 - max : max_change;
2039 add = patch->lines_added;
2040 del = patch->lines_deleted;
2042 if (max_change > 0) {
2043 int total = ((add + del) * max + max_change / 2) / max_change;
2044 add = (add * max + max_change / 2) / max_change;
2045 del = total - add;
2047 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2048 add, pluses, del, minuses);
2051 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2053 switch (st->st_mode & S_IFMT) {
2054 case S_IFLNK:
2055 if (strbuf_readlink(buf, path, st->st_size) < 0)
2056 return error(_("unable to read symlink %s"), path);
2057 return 0;
2058 case S_IFREG:
2059 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2060 return error(_("unable to open or read %s"), path);
2061 convert_to_git(path, buf->buf, buf->len, buf, 0);
2062 return 0;
2063 default:
2064 return -1;
2069 * Update the preimage, and the common lines in postimage,
2070 * from buffer buf of length len. If postlen is 0 the postimage
2071 * is updated in place, otherwise it's updated on a new buffer
2072 * of length postlen
2075 static void update_pre_post_images(struct image *preimage,
2076 struct image *postimage,
2077 char *buf,
2078 size_t len, size_t postlen)
2080 int i, ctx;
2081 char *new, *old, *fixed;
2082 struct image fixed_preimage;
2085 * Update the preimage with whitespace fixes. Note that we
2086 * are not losing preimage->buf -- apply_one_fragment() will
2087 * free "oldlines".
2089 prepare_image(&fixed_preimage, buf, len, 1);
2090 assert(fixed_preimage.nr == preimage->nr);
2091 for (i = 0; i < preimage->nr; i++)
2092 fixed_preimage.line[i].flag = preimage->line[i].flag;
2093 free(preimage->line_allocated);
2094 *preimage = fixed_preimage;
2097 * Adjust the common context lines in postimage. This can be
2098 * done in-place when we are just doing whitespace fixing,
2099 * which does not make the string grow, but needs a new buffer
2100 * when ignoring whitespace causes the update, since in this case
2101 * we could have e.g. tabs converted to multiple spaces.
2102 * We trust the caller to tell us if the update can be done
2103 * in place (postlen==0) or not.
2105 old = postimage->buf;
2106 if (postlen)
2107 new = postimage->buf = xmalloc(postlen);
2108 else
2109 new = old;
2110 fixed = preimage->buf;
2111 for (i = ctx = 0; i < postimage->nr; i++) {
2112 size_t len = postimage->line[i].len;
2113 if (!(postimage->line[i].flag & LINE_COMMON)) {
2114 /* an added line -- no counterparts in preimage */
2115 memmove(new, old, len);
2116 old += len;
2117 new += len;
2118 continue;
2121 /* a common context -- skip it in the original postimage */
2122 old += len;
2124 /* and find the corresponding one in the fixed preimage */
2125 while (ctx < preimage->nr &&
2126 !(preimage->line[ctx].flag & LINE_COMMON)) {
2127 fixed += preimage->line[ctx].len;
2128 ctx++;
2130 if (preimage->nr <= ctx)
2131 die(_("oops"));
2133 /* and copy it in, while fixing the line length */
2134 len = preimage->line[ctx].len;
2135 memcpy(new, fixed, len);
2136 new += len;
2137 fixed += len;
2138 postimage->line[i].len = len;
2139 ctx++;
2142 /* Fix the length of the whole thing */
2143 postimage->len = new - postimage->buf;
2146 static int match_fragment(struct image *img,
2147 struct image *preimage,
2148 struct image *postimage,
2149 unsigned long try,
2150 int try_lno,
2151 unsigned ws_rule,
2152 int match_beginning, int match_end)
2154 int i;
2155 char *fixed_buf, *buf, *orig, *target;
2156 struct strbuf fixed;
2157 size_t fixed_len;
2158 int preimage_limit;
2160 if (preimage->nr + try_lno <= img->nr) {
2162 * The hunk falls within the boundaries of img.
2164 preimage_limit = preimage->nr;
2165 if (match_end && (preimage->nr + try_lno != img->nr))
2166 return 0;
2167 } else if (ws_error_action == correct_ws_error &&
2168 (ws_rule & WS_BLANK_AT_EOF)) {
2170 * This hunk extends beyond the end of img, and we are
2171 * removing blank lines at the end of the file. This
2172 * many lines from the beginning of the preimage must
2173 * match with img, and the remainder of the preimage
2174 * must be blank.
2176 preimage_limit = img->nr - try_lno;
2177 } else {
2179 * The hunk extends beyond the end of the img and
2180 * we are not removing blanks at the end, so we
2181 * should reject the hunk at this position.
2183 return 0;
2186 if (match_beginning && try_lno)
2187 return 0;
2189 /* Quick hash check */
2190 for (i = 0; i < preimage_limit; i++)
2191 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2192 (preimage->line[i].hash != img->line[try_lno + i].hash))
2193 return 0;
2195 if (preimage_limit == preimage->nr) {
2197 * Do we have an exact match? If we were told to match
2198 * at the end, size must be exactly at try+fragsize,
2199 * otherwise try+fragsize must be still within the preimage,
2200 * and either case, the old piece should match the preimage
2201 * exactly.
2203 if ((match_end
2204 ? (try + preimage->len == img->len)
2205 : (try + preimage->len <= img->len)) &&
2206 !memcmp(img->buf + try, preimage->buf, preimage->len))
2207 return 1;
2208 } else {
2210 * The preimage extends beyond the end of img, so
2211 * there cannot be an exact match.
2213 * There must be one non-blank context line that match
2214 * a line before the end of img.
2216 char *buf_end;
2218 buf = preimage->buf;
2219 buf_end = buf;
2220 for (i = 0; i < preimage_limit; i++)
2221 buf_end += preimage->line[i].len;
2223 for ( ; buf < buf_end; buf++)
2224 if (!isspace(*buf))
2225 break;
2226 if (buf == buf_end)
2227 return 0;
2231 * No exact match. If we are ignoring whitespace, run a line-by-line
2232 * fuzzy matching. We collect all the line length information because
2233 * we need it to adjust whitespace if we match.
2235 if (ws_ignore_action == ignore_ws_change) {
2236 size_t imgoff = 0;
2237 size_t preoff = 0;
2238 size_t postlen = postimage->len;
2239 size_t extra_chars;
2240 char *preimage_eof;
2241 char *preimage_end;
2242 for (i = 0; i < preimage_limit; i++) {
2243 size_t prelen = preimage->line[i].len;
2244 size_t imglen = img->line[try_lno+i].len;
2246 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2247 preimage->buf + preoff, prelen))
2248 return 0;
2249 if (preimage->line[i].flag & LINE_COMMON)
2250 postlen += imglen - prelen;
2251 imgoff += imglen;
2252 preoff += prelen;
2256 * Ok, the preimage matches with whitespace fuzz.
2258 * imgoff now holds the true length of the target that
2259 * matches the preimage before the end of the file.
2261 * Count the number of characters in the preimage that fall
2262 * beyond the end of the file and make sure that all of them
2263 * are whitespace characters. (This can only happen if
2264 * we are removing blank lines at the end of the file.)
2266 buf = preimage_eof = preimage->buf + preoff;
2267 for ( ; i < preimage->nr; i++)
2268 preoff += preimage->line[i].len;
2269 preimage_end = preimage->buf + preoff;
2270 for ( ; buf < preimage_end; buf++)
2271 if (!isspace(*buf))
2272 return 0;
2275 * Update the preimage and the common postimage context
2276 * lines to use the same whitespace as the target.
2277 * If whitespace is missing in the target (i.e.
2278 * if the preimage extends beyond the end of the file),
2279 * use the whitespace from the preimage.
2281 extra_chars = preimage_end - preimage_eof;
2282 strbuf_init(&fixed, imgoff + extra_chars);
2283 strbuf_add(&fixed, img->buf + try, imgoff);
2284 strbuf_add(&fixed, preimage_eof, extra_chars);
2285 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2286 update_pre_post_images(preimage, postimage,
2287 fixed_buf, fixed_len, postlen);
2288 return 1;
2291 if (ws_error_action != correct_ws_error)
2292 return 0;
2295 * The hunk does not apply byte-by-byte, but the hash says
2296 * it might with whitespace fuzz. We haven't been asked to
2297 * ignore whitespace, we were asked to correct whitespace
2298 * errors, so let's try matching after whitespace correction.
2300 * The preimage may extend beyond the end of the file,
2301 * but in this loop we will only handle the part of the
2302 * preimage that falls within the file.
2304 strbuf_init(&fixed, preimage->len + 1);
2305 orig = preimage->buf;
2306 target = img->buf + try;
2307 for (i = 0; i < preimage_limit; i++) {
2308 size_t oldlen = preimage->line[i].len;
2309 size_t tgtlen = img->line[try_lno + i].len;
2310 size_t fixstart = fixed.len;
2311 struct strbuf tgtfix;
2312 int match;
2314 /* Try fixing the line in the preimage */
2315 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2317 /* Try fixing the line in the target */
2318 strbuf_init(&tgtfix, tgtlen);
2319 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2322 * If they match, either the preimage was based on
2323 * a version before our tree fixed whitespace breakage,
2324 * or we are lacking a whitespace-fix patch the tree
2325 * the preimage was based on already had (i.e. target
2326 * has whitespace breakage, the preimage doesn't).
2327 * In either case, we are fixing the whitespace breakages
2328 * so we might as well take the fix together with their
2329 * real change.
2331 match = (tgtfix.len == fixed.len - fixstart &&
2332 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2333 fixed.len - fixstart));
2335 strbuf_release(&tgtfix);
2336 if (!match)
2337 goto unmatch_exit;
2339 orig += oldlen;
2340 target += tgtlen;
2345 * Now handle the lines in the preimage that falls beyond the
2346 * end of the file (if any). They will only match if they are
2347 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2348 * false).
2350 for ( ; i < preimage->nr; i++) {
2351 size_t fixstart = fixed.len; /* start of the fixed preimage */
2352 size_t oldlen = preimage->line[i].len;
2353 int j;
2355 /* Try fixing the line in the preimage */
2356 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2358 for (j = fixstart; j < fixed.len; j++)
2359 if (!isspace(fixed.buf[j]))
2360 goto unmatch_exit;
2362 orig += oldlen;
2366 * Yes, the preimage is based on an older version that still
2367 * has whitespace breakages unfixed, and fixing them makes the
2368 * hunk match. Update the context lines in the postimage.
2370 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2371 update_pre_post_images(preimage, postimage,
2372 fixed_buf, fixed_len, 0);
2373 return 1;
2375 unmatch_exit:
2376 strbuf_release(&fixed);
2377 return 0;
2380 static int find_pos(struct image *img,
2381 struct image *preimage,
2382 struct image *postimage,
2383 int line,
2384 unsigned ws_rule,
2385 int match_beginning, int match_end)
2387 int i;
2388 unsigned long backwards, forwards, try;
2389 int backwards_lno, forwards_lno, try_lno;
2392 * If match_beginning or match_end is specified, there is no
2393 * point starting from a wrong line that will never match and
2394 * wander around and wait for a match at the specified end.
2396 if (match_beginning)
2397 line = 0;
2398 else if (match_end)
2399 line = img->nr - preimage->nr;
2402 * Because the comparison is unsigned, the following test
2403 * will also take care of a negative line number that can
2404 * result when match_end and preimage is larger than the target.
2406 if ((size_t) line > img->nr)
2407 line = img->nr;
2409 try = 0;
2410 for (i = 0; i < line; i++)
2411 try += img->line[i].len;
2414 * There's probably some smart way to do this, but I'll leave
2415 * that to the smart and beautiful people. I'm simple and stupid.
2417 backwards = try;
2418 backwards_lno = line;
2419 forwards = try;
2420 forwards_lno = line;
2421 try_lno = line;
2423 for (i = 0; ; i++) {
2424 if (match_fragment(img, preimage, postimage,
2425 try, try_lno, ws_rule,
2426 match_beginning, match_end))
2427 return try_lno;
2429 again:
2430 if (backwards_lno == 0 && forwards_lno == img->nr)
2431 break;
2433 if (i & 1) {
2434 if (backwards_lno == 0) {
2435 i++;
2436 goto again;
2438 backwards_lno--;
2439 backwards -= img->line[backwards_lno].len;
2440 try = backwards;
2441 try_lno = backwards_lno;
2442 } else {
2443 if (forwards_lno == img->nr) {
2444 i++;
2445 goto again;
2447 forwards += img->line[forwards_lno].len;
2448 forwards_lno++;
2449 try = forwards;
2450 try_lno = forwards_lno;
2454 return -1;
2457 static void remove_first_line(struct image *img)
2459 img->buf += img->line[0].len;
2460 img->len -= img->line[0].len;
2461 img->line++;
2462 img->nr--;
2465 static void remove_last_line(struct image *img)
2467 img->len -= img->line[--img->nr].len;
2471 * The change from "preimage" and "postimage" has been found to
2472 * apply at applied_pos (counts in line numbers) in "img".
2473 * Update "img" to remove "preimage" and replace it with "postimage".
2475 static void update_image(struct image *img,
2476 int applied_pos,
2477 struct image *preimage,
2478 struct image *postimage)
2481 * remove the copy of preimage at offset in img
2482 * and replace it with postimage
2484 int i, nr;
2485 size_t remove_count, insert_count, applied_at = 0;
2486 char *result;
2487 int preimage_limit;
2490 * If we are removing blank lines at the end of img,
2491 * the preimage may extend beyond the end.
2492 * If that is the case, we must be careful only to
2493 * remove the part of the preimage that falls within
2494 * the boundaries of img. Initialize preimage_limit
2495 * to the number of lines in the preimage that falls
2496 * within the boundaries.
2498 preimage_limit = preimage->nr;
2499 if (preimage_limit > img->nr - applied_pos)
2500 preimage_limit = img->nr - applied_pos;
2502 for (i = 0; i < applied_pos; i++)
2503 applied_at += img->line[i].len;
2505 remove_count = 0;
2506 for (i = 0; i < preimage_limit; i++)
2507 remove_count += img->line[applied_pos + i].len;
2508 insert_count = postimage->len;
2510 /* Adjust the contents */
2511 result = xmalloc(img->len + insert_count - remove_count + 1);
2512 memcpy(result, img->buf, applied_at);
2513 memcpy(result + applied_at, postimage->buf, postimage->len);
2514 memcpy(result + applied_at + postimage->len,
2515 img->buf + (applied_at + remove_count),
2516 img->len - (applied_at + remove_count));
2517 free(img->buf);
2518 img->buf = result;
2519 img->len += insert_count - remove_count;
2520 result[img->len] = '\0';
2522 /* Adjust the line table */
2523 nr = img->nr + postimage->nr - preimage_limit;
2524 if (preimage_limit < postimage->nr) {
2526 * NOTE: this knows that we never call remove_first_line()
2527 * on anything other than pre/post image.
2529 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2530 img->line_allocated = img->line;
2532 if (preimage_limit != postimage->nr)
2533 memmove(img->line + applied_pos + postimage->nr,
2534 img->line + applied_pos + preimage_limit,
2535 (img->nr - (applied_pos + preimage_limit)) *
2536 sizeof(*img->line));
2537 memcpy(img->line + applied_pos,
2538 postimage->line,
2539 postimage->nr * sizeof(*img->line));
2540 if (!allow_overlap)
2541 for (i = 0; i < postimage->nr; i++)
2542 img->line[applied_pos + i].flag |= LINE_PATCHED;
2543 img->nr = nr;
2547 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2548 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2549 * replace the part of "img" with "postimage" text.
2551 static int apply_one_fragment(struct image *img, struct fragment *frag,
2552 int inaccurate_eof, unsigned ws_rule,
2553 int nth_fragment)
2555 int match_beginning, match_end;
2556 const char *patch = frag->patch;
2557 int size = frag->size;
2558 char *old, *oldlines;
2559 struct strbuf newlines;
2560 int new_blank_lines_at_end = 0;
2561 int found_new_blank_lines_at_end = 0;
2562 int hunk_linenr = frag->linenr;
2563 unsigned long leading, trailing;
2564 int pos, applied_pos;
2565 struct image preimage;
2566 struct image postimage;
2568 memset(&preimage, 0, sizeof(preimage));
2569 memset(&postimage, 0, sizeof(postimage));
2570 oldlines = xmalloc(size);
2571 strbuf_init(&newlines, size);
2573 old = oldlines;
2574 while (size > 0) {
2575 char first;
2576 int len = linelen(patch, size);
2577 int plen;
2578 int added_blank_line = 0;
2579 int is_blank_context = 0;
2580 size_t start;
2582 if (!len)
2583 break;
2586 * "plen" is how much of the line we should use for
2587 * the actual patch data. Normally we just remove the
2588 * first character on the line, but if the line is
2589 * followed by "\ No newline", then we also remove the
2590 * last one (which is the newline, of course).
2592 plen = len - 1;
2593 if (len < size && patch[len] == '\\')
2594 plen--;
2595 first = *patch;
2596 if (apply_in_reverse) {
2597 if (first == '-')
2598 first = '+';
2599 else if (first == '+')
2600 first = '-';
2603 switch (first) {
2604 case '\n':
2605 /* Newer GNU diff, empty context line */
2606 if (plen < 0)
2607 /* ... followed by '\No newline'; nothing */
2608 break;
2609 *old++ = '\n';
2610 strbuf_addch(&newlines, '\n');
2611 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2612 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2613 is_blank_context = 1;
2614 break;
2615 case ' ':
2616 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2617 ws_blank_line(patch + 1, plen, ws_rule))
2618 is_blank_context = 1;
2619 case '-':
2620 memcpy(old, patch + 1, plen);
2621 add_line_info(&preimage, old, plen,
2622 (first == ' ' ? LINE_COMMON : 0));
2623 old += plen;
2624 if (first == '-')
2625 break;
2626 /* Fall-through for ' ' */
2627 case '+':
2628 /* --no-add does not add new lines */
2629 if (first == '+' && no_add)
2630 break;
2632 start = newlines.len;
2633 if (first != '+' ||
2634 !whitespace_error ||
2635 ws_error_action != correct_ws_error) {
2636 strbuf_add(&newlines, patch + 1, plen);
2638 else {
2639 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2641 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2642 (first == '+' ? 0 : LINE_COMMON));
2643 if (first == '+' &&
2644 (ws_rule & WS_BLANK_AT_EOF) &&
2645 ws_blank_line(patch + 1, plen, ws_rule))
2646 added_blank_line = 1;
2647 break;
2648 case '@': case '\\':
2649 /* Ignore it, we already handled it */
2650 break;
2651 default:
2652 if (apply_verbosely)
2653 error(_("invalid start of line: '%c'"), first);
2654 return -1;
2656 if (added_blank_line) {
2657 if (!new_blank_lines_at_end)
2658 found_new_blank_lines_at_end = hunk_linenr;
2659 new_blank_lines_at_end++;
2661 else if (is_blank_context)
2663 else
2664 new_blank_lines_at_end = 0;
2665 patch += len;
2666 size -= len;
2667 hunk_linenr++;
2669 if (inaccurate_eof &&
2670 old > oldlines && old[-1] == '\n' &&
2671 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2672 old--;
2673 strbuf_setlen(&newlines, newlines.len - 1);
2676 leading = frag->leading;
2677 trailing = frag->trailing;
2680 * A hunk to change lines at the beginning would begin with
2681 * @@ -1,L +N,M @@
2682 * but we need to be careful. -U0 that inserts before the second
2683 * line also has this pattern.
2685 * And a hunk to add to an empty file would begin with
2686 * @@ -0,0 +N,M @@
2688 * In other words, a hunk that is (frag->oldpos <= 1) with or
2689 * without leading context must match at the beginning.
2691 match_beginning = (!frag->oldpos ||
2692 (frag->oldpos == 1 && !unidiff_zero));
2695 * A hunk without trailing lines must match at the end.
2696 * However, we simply cannot tell if a hunk must match end
2697 * from the lack of trailing lines if the patch was generated
2698 * with unidiff without any context.
2700 match_end = !unidiff_zero && !trailing;
2702 pos = frag->newpos ? (frag->newpos - 1) : 0;
2703 preimage.buf = oldlines;
2704 preimage.len = old - oldlines;
2705 postimage.buf = newlines.buf;
2706 postimage.len = newlines.len;
2707 preimage.line = preimage.line_allocated;
2708 postimage.line = postimage.line_allocated;
2710 for (;;) {
2712 applied_pos = find_pos(img, &preimage, &postimage, pos,
2713 ws_rule, match_beginning, match_end);
2715 if (applied_pos >= 0)
2716 break;
2718 /* Am I at my context limits? */
2719 if ((leading <= p_context) && (trailing <= p_context))
2720 break;
2721 if (match_beginning || match_end) {
2722 match_beginning = match_end = 0;
2723 continue;
2727 * Reduce the number of context lines; reduce both
2728 * leading and trailing if they are equal otherwise
2729 * just reduce the larger context.
2731 if (leading >= trailing) {
2732 remove_first_line(&preimage);
2733 remove_first_line(&postimage);
2734 pos--;
2735 leading--;
2737 if (trailing > leading) {
2738 remove_last_line(&preimage);
2739 remove_last_line(&postimage);
2740 trailing--;
2744 if (applied_pos >= 0) {
2745 if (new_blank_lines_at_end &&
2746 preimage.nr + applied_pos >= img->nr &&
2747 (ws_rule & WS_BLANK_AT_EOF) &&
2748 ws_error_action != nowarn_ws_error) {
2749 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2750 found_new_blank_lines_at_end);
2751 if (ws_error_action == correct_ws_error) {
2752 while (new_blank_lines_at_end--)
2753 remove_last_line(&postimage);
2756 * We would want to prevent write_out_results()
2757 * from taking place in apply_patch() that follows
2758 * the callchain led us here, which is:
2759 * apply_patch->check_patch_list->check_patch->
2760 * apply_data->apply_fragments->apply_one_fragment
2762 if (ws_error_action == die_on_ws_error)
2763 apply = 0;
2766 if (apply_verbosely && applied_pos != pos) {
2767 int offset = applied_pos - pos;
2768 if (apply_in_reverse)
2769 offset = 0 - offset;
2770 fprintf_ln(stderr,
2771 Q_("Hunk #%d succeeded at %d (offset %d line).",
2772 "Hunk #%d succeeded at %d (offset %d lines).",
2773 offset),
2774 nth_fragment, applied_pos + 1, offset);
2778 * Warn if it was necessary to reduce the number
2779 * of context lines.
2781 if ((leading != frag->leading) ||
2782 (trailing != frag->trailing))
2783 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2784 " to apply fragment at %d"),
2785 leading, trailing, applied_pos+1);
2786 update_image(img, applied_pos, &preimage, &postimage);
2787 } else {
2788 if (apply_verbosely)
2789 error(_("while searching for:\n%.*s"),
2790 (int)(old - oldlines), oldlines);
2793 free(oldlines);
2794 strbuf_release(&newlines);
2795 free(preimage.line_allocated);
2796 free(postimage.line_allocated);
2798 return (applied_pos < 0);
2801 static int apply_binary_fragment(struct image *img, struct patch *patch)
2803 struct fragment *fragment = patch->fragments;
2804 unsigned long len;
2805 void *dst;
2807 if (!fragment)
2808 return error(_("missing binary patch data for '%s'"),
2809 patch->new_name ?
2810 patch->new_name :
2811 patch->old_name);
2813 /* Binary patch is irreversible without the optional second hunk */
2814 if (apply_in_reverse) {
2815 if (!fragment->next)
2816 return error("cannot reverse-apply a binary patch "
2817 "without the reverse hunk to '%s'",
2818 patch->new_name
2819 ? patch->new_name : patch->old_name);
2820 fragment = fragment->next;
2822 switch (fragment->binary_patch_method) {
2823 case BINARY_DELTA_DEFLATED:
2824 dst = patch_delta(img->buf, img->len, fragment->patch,
2825 fragment->size, &len);
2826 if (!dst)
2827 return -1;
2828 clear_image(img);
2829 img->buf = dst;
2830 img->len = len;
2831 return 0;
2832 case BINARY_LITERAL_DEFLATED:
2833 clear_image(img);
2834 img->len = fragment->size;
2835 img->buf = xmalloc(img->len+1);
2836 memcpy(img->buf, fragment->patch, img->len);
2837 img->buf[img->len] = '\0';
2838 return 0;
2840 return -1;
2844 * Replace "img" with the result of applying the binary patch.
2845 * The binary patch data itself in patch->fragment is still kept
2846 * but the preimage prepared by the caller in "img" is freed here
2847 * or in the helper function apply_binary_fragment() this calls.
2849 static int apply_binary(struct image *img, struct patch *patch)
2851 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2852 unsigned char sha1[20];
2855 * For safety, we require patch index line to contain
2856 * full 40-byte textual SHA1 for old and new, at least for now.
2858 if (strlen(patch->old_sha1_prefix) != 40 ||
2859 strlen(patch->new_sha1_prefix) != 40 ||
2860 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2861 get_sha1_hex(patch->new_sha1_prefix, sha1))
2862 return error("cannot apply binary patch to '%s' "
2863 "without full index line", name);
2865 if (patch->old_name) {
2867 * See if the old one matches what the patch
2868 * applies to.
2870 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2871 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2872 return error("the patch applies to '%s' (%s), "
2873 "which does not match the "
2874 "current contents.",
2875 name, sha1_to_hex(sha1));
2877 else {
2878 /* Otherwise, the old one must be empty. */
2879 if (img->len)
2880 return error("the patch applies to an empty "
2881 "'%s' but it is not empty", name);
2884 get_sha1_hex(patch->new_sha1_prefix, sha1);
2885 if (is_null_sha1(sha1)) {
2886 clear_image(img);
2887 return 0; /* deletion patch */
2890 if (has_sha1_file(sha1)) {
2891 /* We already have the postimage */
2892 enum object_type type;
2893 unsigned long size;
2894 char *result;
2896 result = read_sha1_file(sha1, &type, &size);
2897 if (!result)
2898 return error("the necessary postimage %s for "
2899 "'%s' cannot be read",
2900 patch->new_sha1_prefix, name);
2901 clear_image(img);
2902 img->buf = result;
2903 img->len = size;
2904 } else {
2906 * We have verified buf matches the preimage;
2907 * apply the patch data to it, which is stored
2908 * in the patch->fragments->{patch,size}.
2910 if (apply_binary_fragment(img, patch))
2911 return error(_("binary patch does not apply to '%s'"),
2912 name);
2914 /* verify that the result matches */
2915 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2916 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2917 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2918 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2921 return 0;
2924 static int apply_fragments(struct image *img, struct patch *patch)
2926 struct fragment *frag = patch->fragments;
2927 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2928 unsigned ws_rule = patch->ws_rule;
2929 unsigned inaccurate_eof = patch->inaccurate_eof;
2930 int nth = 0;
2932 if (patch->is_binary)
2933 return apply_binary(img, patch);
2935 while (frag) {
2936 nth++;
2937 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2938 error(_("patch failed: %s:%ld"), name, frag->oldpos);
2939 if (!apply_with_reject)
2940 return -1;
2941 frag->rejected = 1;
2943 frag = frag->next;
2945 return 0;
2948 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2950 if (S_ISGITLINK(mode)) {
2951 strbuf_grow(buf, 100);
2952 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2953 } else {
2954 enum object_type type;
2955 unsigned long sz;
2956 char *result;
2958 result = read_sha1_file(sha1, &type, &sz);
2959 if (!result)
2960 return -1;
2961 /* XXX read_sha1_file NUL-terminates */
2962 strbuf_attach(buf, result, sz, sz + 1);
2964 return 0;
2967 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2969 if (!ce)
2970 return 0;
2971 return read_blob_object(buf, ce->sha1, ce->ce_mode);
2974 static struct patch *in_fn_table(const char *name)
2976 struct string_list_item *item;
2978 if (name == NULL)
2979 return NULL;
2981 item = string_list_lookup(&fn_table, name);
2982 if (item != NULL)
2983 return (struct patch *)item->util;
2985 return NULL;
2989 * item->util in the filename table records the status of the path.
2990 * Usually it points at a patch (whose result records the contents
2991 * of it after applying it), but it could be PATH_WAS_DELETED for a
2992 * path that a previously applied patch has already removed, or
2993 * PATH_TO_BE_DELETED for a path that a later patch would remove.
2995 * The latter is needed to deal with a case where two paths A and B
2996 * are swapped by first renaming A to B and then renaming B to A;
2997 * moving A to B should not be prevented due to presense of B as we
2998 * will remove it in a later patch.
3000 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3001 #define PATH_WAS_DELETED ((struct patch *) -1)
3003 static int to_be_deleted(struct patch *patch)
3005 return patch == PATH_TO_BE_DELETED;
3008 static int was_deleted(struct patch *patch)
3010 return patch == PATH_WAS_DELETED;
3013 static void add_to_fn_table(struct patch *patch)
3015 struct string_list_item *item;
3018 * Always add new_name unless patch is a deletion
3019 * This should cover the cases for normal diffs,
3020 * file creations and copies
3022 if (patch->new_name != NULL) {
3023 item = string_list_insert(&fn_table, patch->new_name);
3024 item->util = patch;
3028 * store a failure on rename/deletion cases because
3029 * later chunks shouldn't patch old names
3031 if ((patch->new_name == NULL) || (patch->is_rename)) {
3032 item = string_list_insert(&fn_table, patch->old_name);
3033 item->util = PATH_WAS_DELETED;
3037 static void prepare_fn_table(struct patch *patch)
3040 * store information about incoming file deletion
3042 while (patch) {
3043 if ((patch->new_name == NULL) || (patch->is_rename)) {
3044 struct string_list_item *item;
3045 item = string_list_insert(&fn_table, patch->old_name);
3046 item->util = PATH_TO_BE_DELETED;
3048 patch = patch->next;
3052 static int checkout_target(struct cache_entry *ce, struct stat *st)
3054 struct checkout costate;
3056 memset(&costate, 0, sizeof(costate));
3057 costate.base_dir = "";
3058 costate.refresh_cache = 1;
3059 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3060 return error(_("cannot checkout %s"), ce->name);
3061 return 0;
3064 static struct patch *previous_patch(struct patch *patch, int *gone)
3066 struct patch *previous;
3068 *gone = 0;
3069 if (patch->is_copy || patch->is_rename)
3070 return NULL; /* "git" patches do not depend on the order */
3072 previous = in_fn_table(patch->old_name);
3073 if (!previous)
3074 return NULL;
3076 if (to_be_deleted(previous))
3077 return NULL; /* the deletion hasn't happened yet */
3079 if (was_deleted(previous))
3080 *gone = 1;
3082 return previous;
3085 static int verify_index_match(struct cache_entry *ce, struct stat *st)
3087 if (S_ISGITLINK(ce->ce_mode)) {
3088 if (!S_ISDIR(st->st_mode))
3089 return -1;
3090 return 0;
3092 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3095 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3097 static int load_patch_target(struct strbuf *buf,
3098 struct cache_entry *ce,
3099 struct stat *st,
3100 const char *name,
3101 unsigned expected_mode)
3103 if (cached) {
3104 if (read_file_or_gitlink(ce, buf))
3105 return error(_("read of %s failed"), name);
3106 } else if (name) {
3107 if (S_ISGITLINK(expected_mode)) {
3108 if (ce)
3109 return read_file_or_gitlink(ce, buf);
3110 else
3111 return SUBMODULE_PATCH_WITHOUT_INDEX;
3112 } else {
3113 if (read_old_data(st, name, buf))
3114 return error(_("read of %s failed"), name);
3117 return 0;
3121 * We are about to apply "patch"; populate the "image" with the
3122 * current version we have, from the working tree or from the index,
3123 * depending on the situation e.g. --cached/--index. If we are
3124 * applying a non-git patch that incrementally updates the tree,
3125 * we read from the result of a previous diff.
3127 static int load_preimage(struct image *image,
3128 struct patch *patch, struct stat *st, struct cache_entry *ce)
3130 struct strbuf buf = STRBUF_INIT;
3131 size_t len;
3132 char *img;
3133 struct patch *previous;
3134 int status;
3136 previous = previous_patch(patch, &status);
3137 if (status)
3138 return error(_("path %s has been renamed/deleted"),
3139 patch->old_name);
3140 if (previous) {
3141 /* We have a patched copy in memory; use that. */
3142 strbuf_add(&buf, previous->result, previous->resultsize);
3143 } else {
3144 status = load_patch_target(&buf, ce, st,
3145 patch->old_name, patch->old_mode);
3146 if (status < 0)
3147 return status;
3148 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3150 * There is no way to apply subproject
3151 * patch without looking at the index.
3152 * NEEDSWORK: shouldn't this be flagged
3153 * as an error???
3155 free_fragment_list(patch->fragments);
3156 patch->fragments = NULL;
3157 } else if (status) {
3158 return error(_("read of %s failed"), patch->old_name);
3162 img = strbuf_detach(&buf, &len);
3163 prepare_image(image, img, len, !patch->is_binary);
3164 return 0;
3167 static int three_way_merge(struct image *image,
3168 char *path,
3169 const unsigned char *base,
3170 const unsigned char *ours,
3171 const unsigned char *theirs)
3173 mmfile_t base_file, our_file, their_file;
3174 mmbuffer_t result = { NULL };
3175 int status;
3177 read_mmblob(&base_file, base);
3178 read_mmblob(&our_file, ours);
3179 read_mmblob(&their_file, theirs);
3180 status = ll_merge(&result, path,
3181 &base_file, "base",
3182 &our_file, "ours",
3183 &their_file, "theirs", NULL);
3184 free(base_file.ptr);
3185 free(our_file.ptr);
3186 free(their_file.ptr);
3187 if (status < 0 || !result.ptr) {
3188 free(result.ptr);
3189 return -1;
3191 clear_image(image);
3192 image->buf = result.ptr;
3193 image->len = result.size;
3195 return status;
3199 * When directly falling back to add/add three-way merge, we read from
3200 * the current contents of the new_name. In no cases other than that
3201 * this function will be called.
3203 static int load_current(struct image *image, struct patch *patch)
3205 struct strbuf buf = STRBUF_INIT;
3206 int status, pos;
3207 size_t len;
3208 char *img;
3209 struct stat st;
3210 struct cache_entry *ce;
3211 char *name = patch->new_name;
3212 unsigned mode = patch->new_mode;
3214 if (!patch->is_new)
3215 die("BUG: patch to %s is not a creation", patch->old_name);
3217 pos = cache_name_pos(name, strlen(name));
3218 if (pos < 0)
3219 return error(_("%s: does not exist in index"), name);
3220 ce = active_cache[pos];
3221 if (lstat(name, &st)) {
3222 if (errno != ENOENT)
3223 return error(_("%s: %s"), name, strerror(errno));
3224 if (checkout_target(ce, &st))
3225 return -1;
3227 if (verify_index_match(ce, &st))
3228 return error(_("%s: does not match index"), name);
3230 status = load_patch_target(&buf, ce, &st, name, mode);
3231 if (status < 0)
3232 return status;
3233 else if (status)
3234 return -1;
3235 img = strbuf_detach(&buf, &len);
3236 prepare_image(image, img, len, !patch->is_binary);
3237 return 0;
3240 static int try_threeway(struct image *image, struct patch *patch,
3241 struct stat *st, struct cache_entry *ce)
3243 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3244 struct strbuf buf = STRBUF_INIT;
3245 size_t len;
3246 int status;
3247 char *img;
3248 struct image tmp_image;
3250 /* No point falling back to 3-way merge in these cases */
3251 if (patch->is_delete ||
3252 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3253 return -1;
3255 /* Preimage the patch was prepared for */
3256 if (patch->is_new)
3257 write_sha1_file("", 0, blob_type, pre_sha1);
3258 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3259 read_blob_object(&buf, pre_sha1, patch->old_mode))
3260 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3262 fprintf(stderr, "Falling back to three-way merge...\n");
3264 img = strbuf_detach(&buf, &len);
3265 prepare_image(&tmp_image, img, len, 1);
3266 /* Apply the patch to get the post image */
3267 if (apply_fragments(&tmp_image, patch) < 0) {
3268 clear_image(&tmp_image);
3269 return -1;
3271 /* post_sha1[] is theirs */
3272 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3273 clear_image(&tmp_image);
3275 /* our_sha1[] is ours */
3276 if (patch->is_new) {
3277 if (load_current(&tmp_image, patch))
3278 return error("cannot read the current contents of '%s'",
3279 patch->new_name);
3280 } else {
3281 if (load_preimage(&tmp_image, patch, st, ce))
3282 return error("cannot read the current contents of '%s'",
3283 patch->old_name);
3285 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3286 clear_image(&tmp_image);
3288 /* in-core three-way merge between post and our using pre as base */
3289 status = three_way_merge(image, patch->new_name,
3290 pre_sha1, our_sha1, post_sha1);
3291 if (status < 0) {
3292 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3293 return status;
3296 if (status) {
3297 patch->conflicted_threeway = 1;
3298 if (patch->is_new)
3299 hashclr(patch->threeway_stage[0]);
3300 else
3301 hashcpy(patch->threeway_stage[0], pre_sha1);
3302 hashcpy(patch->threeway_stage[1], our_sha1);
3303 hashcpy(patch->threeway_stage[2], post_sha1);
3304 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3305 } else {
3306 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3308 return 0;
3311 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3313 struct image image;
3315 if (load_preimage(&image, patch, st, ce) < 0)
3316 return -1;
3318 if (patch->direct_to_threeway ||
3319 apply_fragments(&image, patch) < 0) {
3320 /* Note: with --reject, apply_fragments() returns 0 */
3321 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3322 return -1;
3324 patch->result = image.buf;
3325 patch->resultsize = image.len;
3326 add_to_fn_table(patch);
3327 free(image.line_allocated);
3329 if (0 < patch->is_delete && patch->resultsize)
3330 return error(_("removal patch leaves file contents"));
3332 return 0;
3336 * If "patch" that we are looking at modifies or deletes what we have,
3337 * we would want it not to lose any local modification we have, either
3338 * in the working tree or in the index.
3340 * This also decides if a non-git patch is a creation patch or a
3341 * modification to an existing empty file. We do not check the state
3342 * of the current tree for a creation patch in this function; the caller
3343 * check_patch() separately makes sure (and errors out otherwise) that
3344 * the path the patch creates does not exist in the current tree.
3346 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3348 const char *old_name = patch->old_name;
3349 struct patch *previous = NULL;
3350 int stat_ret = 0, status;
3351 unsigned st_mode = 0;
3353 if (!old_name)
3354 return 0;
3356 assert(patch->is_new <= 0);
3357 previous = previous_patch(patch, &status);
3359 if (status)
3360 return error(_("path %s has been renamed/deleted"), old_name);
3361 if (previous) {
3362 st_mode = previous->new_mode;
3363 } else if (!cached) {
3364 stat_ret = lstat(old_name, st);
3365 if (stat_ret && errno != ENOENT)
3366 return error(_("%s: %s"), old_name, strerror(errno));
3369 if (check_index && !previous) {
3370 int pos = cache_name_pos(old_name, strlen(old_name));
3371 if (pos < 0) {
3372 if (patch->is_new < 0)
3373 goto is_new;
3374 return error(_("%s: does not exist in index"), old_name);
3376 *ce = active_cache[pos];
3377 if (stat_ret < 0) {
3378 if (checkout_target(*ce, st))
3379 return -1;
3381 if (!cached && verify_index_match(*ce, st))
3382 return error(_("%s: does not match index"), old_name);
3383 if (cached)
3384 st_mode = (*ce)->ce_mode;
3385 } else if (stat_ret < 0) {
3386 if (patch->is_new < 0)
3387 goto is_new;
3388 return error(_("%s: %s"), old_name, strerror(errno));
3391 if (!cached && !previous)
3392 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3394 if (patch->is_new < 0)
3395 patch->is_new = 0;
3396 if (!patch->old_mode)
3397 patch->old_mode = st_mode;
3398 if ((st_mode ^ patch->old_mode) & S_IFMT)
3399 return error(_("%s: wrong type"), old_name);
3400 if (st_mode != patch->old_mode)
3401 warning(_("%s has type %o, expected %o"),
3402 old_name, st_mode, patch->old_mode);
3403 if (!patch->new_mode && !patch->is_delete)
3404 patch->new_mode = st_mode;
3405 return 0;
3407 is_new:
3408 patch->is_new = 1;
3409 patch->is_delete = 0;
3410 free(patch->old_name);
3411 patch->old_name = NULL;
3412 return 0;
3416 #define EXISTS_IN_INDEX 1
3417 #define EXISTS_IN_WORKTREE 2
3419 static int check_to_create(const char *new_name, int ok_if_exists)
3421 struct stat nst;
3423 if (check_index &&
3424 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3425 !ok_if_exists)
3426 return EXISTS_IN_INDEX;
3427 if (cached)
3428 return 0;
3430 if (!lstat(new_name, &nst)) {
3431 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3432 return 0;
3434 * A leading component of new_name might be a symlink
3435 * that is going to be removed with this patch, but
3436 * still pointing at somewhere that has the path.
3437 * In such a case, path "new_name" does not exist as
3438 * far as git is concerned.
3440 if (has_symlink_leading_path(new_name, strlen(new_name)))
3441 return 0;
3443 return EXISTS_IN_WORKTREE;
3444 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3445 return error("%s: %s", new_name, strerror(errno));
3447 return 0;
3451 * Check and apply the patch in-core; leave the result in patch->result
3452 * for the caller to write it out to the final destination.
3454 static int check_patch(struct patch *patch)
3456 struct stat st;
3457 const char *old_name = patch->old_name;
3458 const char *new_name = patch->new_name;
3459 const char *name = old_name ? old_name : new_name;
3460 struct cache_entry *ce = NULL;
3461 struct patch *tpatch;
3462 int ok_if_exists;
3463 int status;
3465 patch->rejected = 1; /* we will drop this after we succeed */
3467 status = check_preimage(patch, &ce, &st);
3468 if (status)
3469 return status;
3470 old_name = patch->old_name;
3473 * A type-change diff is always split into a patch to delete
3474 * old, immediately followed by a patch to create new (see
3475 * diff.c::run_diff()); in such a case it is Ok that the entry
3476 * to be deleted by the previous patch is still in the working
3477 * tree and in the index.
3479 * A patch to swap-rename between A and B would first rename A
3480 * to B and then rename B to A. While applying the first one,
3481 * the presense of B should not stop A from getting renamed to
3482 * B; ask to_be_deleted() about the later rename. Removal of
3483 * B and rename from A to B is handled the same way by asking
3484 * was_deleted().
3486 if ((tpatch = in_fn_table(new_name)) &&
3487 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3488 ok_if_exists = 1;
3489 else
3490 ok_if_exists = 0;
3492 if (new_name &&
3493 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3494 int err = check_to_create(new_name, ok_if_exists);
3496 if (err && threeway) {
3497 patch->direct_to_threeway = 1;
3498 } else switch (err) {
3499 case 0:
3500 break; /* happy */
3501 case EXISTS_IN_INDEX:
3502 return error(_("%s: already exists in index"), new_name);
3503 break;
3504 case EXISTS_IN_WORKTREE:
3505 return error(_("%s: already exists in working directory"),
3506 new_name);
3507 default:
3508 return err;
3511 if (!patch->new_mode) {
3512 if (0 < patch->is_new)
3513 patch->new_mode = S_IFREG | 0644;
3514 else
3515 patch->new_mode = patch->old_mode;
3519 if (new_name && old_name) {
3520 int same = !strcmp(old_name, new_name);
3521 if (!patch->new_mode)
3522 patch->new_mode = patch->old_mode;
3523 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3524 if (same)
3525 return error(_("new mode (%o) of %s does not "
3526 "match old mode (%o)"),
3527 patch->new_mode, new_name,
3528 patch->old_mode);
3529 else
3530 return error(_("new mode (%o) of %s does not "
3531 "match old mode (%o) of %s"),
3532 patch->new_mode, new_name,
3533 patch->old_mode, old_name);
3537 if (apply_data(patch, &st, ce) < 0)
3538 return error(_("%s: patch does not apply"), name);
3539 patch->rejected = 0;
3540 return 0;
3543 static int check_patch_list(struct patch *patch)
3545 int err = 0;
3547 prepare_fn_table(patch);
3548 while (patch) {
3549 if (apply_verbosely)
3550 say_patch_name(stderr,
3551 _("Checking patch %s..."), patch);
3552 err |= check_patch(patch);
3553 patch = patch->next;
3555 return err;
3558 /* This function tries to read the sha1 from the current index */
3559 static int get_current_sha1(const char *path, unsigned char *sha1)
3561 int pos;
3563 if (read_cache() < 0)
3564 return -1;
3565 pos = cache_name_pos(path, strlen(path));
3566 if (pos < 0)
3567 return -1;
3568 hashcpy(sha1, active_cache[pos]->sha1);
3569 return 0;
3572 /* Build an index that contains the just the files needed for a 3way merge */
3573 static void build_fake_ancestor(struct patch *list, const char *filename)
3575 struct patch *patch;
3576 struct index_state result = { NULL };
3577 int fd;
3579 /* Once we start supporting the reverse patch, it may be
3580 * worth showing the new sha1 prefix, but until then...
3582 for (patch = list; patch; patch = patch->next) {
3583 const unsigned char *sha1_ptr;
3584 unsigned char sha1[20];
3585 struct cache_entry *ce;
3586 const char *name;
3588 name = patch->old_name ? patch->old_name : patch->new_name;
3589 if (0 < patch->is_new)
3590 continue;
3591 else if (get_sha1_blob(patch->old_sha1_prefix, sha1))
3592 /* git diff has no index line for mode/type changes */
3593 if (!patch->lines_added && !patch->lines_deleted) {
3594 if (get_current_sha1(patch->old_name, sha1))
3595 die("mode change for %s, which is not "
3596 "in current HEAD", name);
3597 sha1_ptr = sha1;
3598 } else
3599 die("sha1 information is lacking or useless "
3600 "(%s).", name);
3601 else
3602 sha1_ptr = sha1;
3604 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3605 if (!ce)
3606 die(_("make_cache_entry failed for path '%s'"), name);
3607 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3608 die ("Could not add %s to temporary index", name);
3611 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3612 if (fd < 0 || write_index(&result, fd) || close(fd))
3613 die ("Could not write temporary index to %s", filename);
3615 discard_index(&result);
3618 static void stat_patch_list(struct patch *patch)
3620 int files, adds, dels;
3622 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3623 files++;
3624 adds += patch->lines_added;
3625 dels += patch->lines_deleted;
3626 show_stats(patch);
3629 print_stat_summary(stdout, files, adds, dels);
3632 static void numstat_patch_list(struct patch *patch)
3634 for ( ; patch; patch = patch->next) {
3635 const char *name;
3636 name = patch->new_name ? patch->new_name : patch->old_name;
3637 if (patch->is_binary)
3638 printf("-\t-\t");
3639 else
3640 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3641 write_name_quoted(name, stdout, line_termination);
3645 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3647 if (mode)
3648 printf(" %s mode %06o %s\n", newdelete, mode, name);
3649 else
3650 printf(" %s %s\n", newdelete, name);
3653 static void show_mode_change(struct patch *p, int show_name)
3655 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3656 if (show_name)
3657 printf(" mode change %06o => %06o %s\n",
3658 p->old_mode, p->new_mode, p->new_name);
3659 else
3660 printf(" mode change %06o => %06o\n",
3661 p->old_mode, p->new_mode);
3665 static void show_rename_copy(struct patch *p)
3667 const char *renamecopy = p->is_rename ? "rename" : "copy";
3668 const char *old, *new;
3670 /* Find common prefix */
3671 old = p->old_name;
3672 new = p->new_name;
3673 while (1) {
3674 const char *slash_old, *slash_new;
3675 slash_old = strchr(old, '/');
3676 slash_new = strchr(new, '/');
3677 if (!slash_old ||
3678 !slash_new ||
3679 slash_old - old != slash_new - new ||
3680 memcmp(old, new, slash_new - new))
3681 break;
3682 old = slash_old + 1;
3683 new = slash_new + 1;
3685 /* p->old_name thru old is the common prefix, and old and new
3686 * through the end of names are renames
3688 if (old != p->old_name)
3689 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3690 (int)(old - p->old_name), p->old_name,
3691 old, new, p->score);
3692 else
3693 printf(" %s %s => %s (%d%%)\n", renamecopy,
3694 p->old_name, p->new_name, p->score);
3695 show_mode_change(p, 0);
3698 static void summary_patch_list(struct patch *patch)
3700 struct patch *p;
3702 for (p = patch; p; p = p->next) {
3703 if (p->is_new)
3704 show_file_mode_name("create", p->new_mode, p->new_name);
3705 else if (p->is_delete)
3706 show_file_mode_name("delete", p->old_mode, p->old_name);
3707 else {
3708 if (p->is_rename || p->is_copy)
3709 show_rename_copy(p);
3710 else {
3711 if (p->score) {
3712 printf(" rewrite %s (%d%%)\n",
3713 p->new_name, p->score);
3714 show_mode_change(p, 0);
3716 else
3717 show_mode_change(p, 1);
3723 static void patch_stats(struct patch *patch)
3725 int lines = patch->lines_added + patch->lines_deleted;
3727 if (lines > max_change)
3728 max_change = lines;
3729 if (patch->old_name) {
3730 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3731 if (!len)
3732 len = strlen(patch->old_name);
3733 if (len > max_len)
3734 max_len = len;
3736 if (patch->new_name) {
3737 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3738 if (!len)
3739 len = strlen(patch->new_name);
3740 if (len > max_len)
3741 max_len = len;
3745 static void remove_file(struct patch *patch, int rmdir_empty)
3747 if (update_index) {
3748 if (remove_file_from_cache(patch->old_name) < 0)
3749 die(_("unable to remove %s from index"), patch->old_name);
3751 if (!cached) {
3752 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3753 remove_path(patch->old_name);
3758 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3760 struct stat st;
3761 struct cache_entry *ce;
3762 int namelen = strlen(path);
3763 unsigned ce_size = cache_entry_size(namelen);
3765 if (!update_index)
3766 return;
3768 ce = xcalloc(1, ce_size);
3769 memcpy(ce->name, path, namelen);
3770 ce->ce_mode = create_ce_mode(mode);
3771 ce->ce_flags = create_ce_flags(0);
3772 ce->ce_namelen = namelen;
3773 if (S_ISGITLINK(mode)) {
3774 const char *s = buf;
3776 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3777 die(_("corrupt patch for subproject %s"), path);
3778 } else {
3779 if (!cached) {
3780 if (lstat(path, &st) < 0)
3781 die_errno(_("unable to stat newly created file '%s'"),
3782 path);
3783 fill_stat_cache_info(ce, &st);
3785 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3786 die(_("unable to create backing store for newly created file %s"), path);
3788 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3789 die(_("unable to add cache entry for %s"), path);
3792 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3794 int fd;
3795 struct strbuf nbuf = STRBUF_INIT;
3797 if (S_ISGITLINK(mode)) {
3798 struct stat st;
3799 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3800 return 0;
3801 return mkdir(path, 0777);
3804 if (has_symlinks && S_ISLNK(mode))
3805 /* Although buf:size is counted string, it also is NUL
3806 * terminated.
3808 return symlink(buf, path);
3810 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3811 if (fd < 0)
3812 return -1;
3814 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3815 size = nbuf.len;
3816 buf = nbuf.buf;
3818 write_or_die(fd, buf, size);
3819 strbuf_release(&nbuf);
3821 if (close(fd) < 0)
3822 die_errno(_("closing file '%s'"), path);
3823 return 0;
3827 * We optimistically assume that the directories exist,
3828 * which is true 99% of the time anyway. If they don't,
3829 * we create them and try again.
3831 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3833 if (cached)
3834 return;
3835 if (!try_create_file(path, mode, buf, size))
3836 return;
3838 if (errno == ENOENT) {
3839 if (safe_create_leading_directories(path))
3840 return;
3841 if (!try_create_file(path, mode, buf, size))
3842 return;
3845 if (errno == EEXIST || errno == EACCES) {
3846 /* We may be trying to create a file where a directory
3847 * used to be.
3849 struct stat st;
3850 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3851 errno = EEXIST;
3854 if (errno == EEXIST) {
3855 unsigned int nr = getpid();
3857 for (;;) {
3858 char newpath[PATH_MAX];
3859 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3860 if (!try_create_file(newpath, mode, buf, size)) {
3861 if (!rename(newpath, path))
3862 return;
3863 unlink_or_warn(newpath);
3864 break;
3866 if (errno != EEXIST)
3867 break;
3868 ++nr;
3871 die_errno(_("unable to write file '%s' mode %o"), path, mode);
3874 static void add_conflicted_stages_file(struct patch *patch)
3876 int stage, namelen;
3877 unsigned ce_size, mode;
3878 struct cache_entry *ce;
3880 if (!update_index)
3881 return;
3882 namelen = strlen(patch->new_name);
3883 ce_size = cache_entry_size(namelen);
3884 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
3886 remove_file_from_cache(patch->new_name);
3887 for (stage = 1; stage < 4; stage++) {
3888 if (is_null_sha1(patch->threeway_stage[stage - 1]))
3889 continue;
3890 ce = xcalloc(1, ce_size);
3891 memcpy(ce->name, patch->new_name, namelen);
3892 ce->ce_mode = create_ce_mode(mode);
3893 ce->ce_flags = create_ce_flags(stage);
3894 ce->ce_namelen = namelen;
3895 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
3896 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3897 die(_("unable to add cache entry for %s"), patch->new_name);
3901 static void create_file(struct patch *patch)
3903 char *path = patch->new_name;
3904 unsigned mode = patch->new_mode;
3905 unsigned long size = patch->resultsize;
3906 char *buf = patch->result;
3908 if (!mode)
3909 mode = S_IFREG | 0644;
3910 create_one_file(path, mode, buf, size);
3912 if (patch->conflicted_threeway)
3913 add_conflicted_stages_file(patch);
3914 else
3915 add_index_file(path, mode, buf, size);
3918 /* phase zero is to remove, phase one is to create */
3919 static void write_out_one_result(struct patch *patch, int phase)
3921 if (patch->is_delete > 0) {
3922 if (phase == 0)
3923 remove_file(patch, 1);
3924 return;
3926 if (patch->is_new > 0 || patch->is_copy) {
3927 if (phase == 1)
3928 create_file(patch);
3929 return;
3932 * Rename or modification boils down to the same
3933 * thing: remove the old, write the new
3935 if (phase == 0)
3936 remove_file(patch, patch->is_rename);
3937 if (phase == 1)
3938 create_file(patch);
3941 static int write_out_one_reject(struct patch *patch)
3943 FILE *rej;
3944 char namebuf[PATH_MAX];
3945 struct fragment *frag;
3946 int cnt = 0;
3947 struct strbuf sb = STRBUF_INIT;
3949 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3950 if (!frag->rejected)
3951 continue;
3952 cnt++;
3955 if (!cnt) {
3956 if (apply_verbosely)
3957 say_patch_name(stderr,
3958 _("Applied patch %s cleanly."), patch);
3959 return 0;
3962 /* This should not happen, because a removal patch that leaves
3963 * contents are marked "rejected" at the patch level.
3965 if (!patch->new_name)
3966 die(_("internal error"));
3968 /* Say this even without --verbose */
3969 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3970 "Applying patch %%s with %d rejects...",
3971 cnt),
3972 cnt);
3973 say_patch_name(stderr, sb.buf, patch);
3974 strbuf_release(&sb);
3976 cnt = strlen(patch->new_name);
3977 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3978 cnt = ARRAY_SIZE(namebuf) - 5;
3979 warning(_("truncating .rej filename to %.*s.rej"),
3980 cnt - 1, patch->new_name);
3982 memcpy(namebuf, patch->new_name, cnt);
3983 memcpy(namebuf + cnt, ".rej", 5);
3985 rej = fopen(namebuf, "w");
3986 if (!rej)
3987 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
3989 /* Normal git tools never deal with .rej, so do not pretend
3990 * this is a git patch by saying --git nor give extended
3991 * headers. While at it, maybe please "kompare" that wants
3992 * the trailing TAB and some garbage at the end of line ;-).
3994 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3995 patch->new_name, patch->new_name);
3996 for (cnt = 1, frag = patch->fragments;
3997 frag;
3998 cnt++, frag = frag->next) {
3999 if (!frag->rejected) {
4000 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4001 continue;
4003 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4004 fprintf(rej, "%.*s", frag->size, frag->patch);
4005 if (frag->patch[frag->size-1] != '\n')
4006 fputc('\n', rej);
4008 fclose(rej);
4009 return -1;
4012 static int write_out_results(struct patch *list)
4014 int phase;
4015 int errs = 0;
4016 struct patch *l;
4017 struct string_list cpath = STRING_LIST_INIT_DUP;
4019 for (phase = 0; phase < 2; phase++) {
4020 l = list;
4021 while (l) {
4022 if (l->rejected)
4023 errs = 1;
4024 else {
4025 write_out_one_result(l, phase);
4026 if (phase == 1) {
4027 if (write_out_one_reject(l))
4028 errs = 1;
4029 if (l->conflicted_threeway) {
4030 string_list_append(&cpath, l->new_name);
4031 errs = 1;
4035 l = l->next;
4039 if (cpath.nr) {
4040 struct string_list_item *item;
4042 sort_string_list(&cpath);
4043 for_each_string_list_item(item, &cpath)
4044 fprintf(stderr, "U %s\n", item->string);
4045 string_list_clear(&cpath, 0);
4047 rerere(0);
4050 return errs;
4053 static struct lock_file lock_file;
4055 static struct string_list limit_by_name;
4056 static int has_include;
4057 static void add_name_limit(const char *name, int exclude)
4059 struct string_list_item *it;
4061 it = string_list_append(&limit_by_name, name);
4062 it->util = exclude ? NULL : (void *) 1;
4065 static int use_patch(struct patch *p)
4067 const char *pathname = p->new_name ? p->new_name : p->old_name;
4068 int i;
4070 /* Paths outside are not touched regardless of "--include" */
4071 if (0 < prefix_length) {
4072 int pathlen = strlen(pathname);
4073 if (pathlen <= prefix_length ||
4074 memcmp(prefix, pathname, prefix_length))
4075 return 0;
4078 /* See if it matches any of exclude/include rule */
4079 for (i = 0; i < limit_by_name.nr; i++) {
4080 struct string_list_item *it = &limit_by_name.items[i];
4081 if (!fnmatch(it->string, pathname, 0))
4082 return (it->util != NULL);
4086 * If we had any include, a path that does not match any rule is
4087 * not used. Otherwise, we saw bunch of exclude rules (or none)
4088 * and such a path is used.
4090 return !has_include;
4094 static void prefix_one(char **name)
4096 char *old_name = *name;
4097 if (!old_name)
4098 return;
4099 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
4100 free(old_name);
4103 static void prefix_patches(struct patch *p)
4105 if (!prefix || p->is_toplevel_relative)
4106 return;
4107 for ( ; p; p = p->next) {
4108 prefix_one(&p->new_name);
4109 prefix_one(&p->old_name);
4113 #define INACCURATE_EOF (1<<0)
4114 #define RECOUNT (1<<1)
4116 static int apply_patch(int fd, const char *filename, int options)
4118 size_t offset;
4119 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4120 struct patch *list = NULL, **listp = &list;
4121 int skipped_patch = 0;
4123 patch_input_file = filename;
4124 read_patch_file(&buf, fd);
4125 offset = 0;
4126 while (offset < buf.len) {
4127 struct patch *patch;
4128 int nr;
4130 patch = xcalloc(1, sizeof(*patch));
4131 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4132 patch->recount = !!(options & RECOUNT);
4133 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4134 if (nr < 0)
4135 break;
4136 if (apply_in_reverse)
4137 reverse_patches(patch);
4138 if (prefix)
4139 prefix_patches(patch);
4140 if (use_patch(patch)) {
4141 patch_stats(patch);
4142 *listp = patch;
4143 listp = &patch->next;
4145 else {
4146 free_patch(patch);
4147 skipped_patch++;
4149 offset += nr;
4152 if (!list && !skipped_patch)
4153 die(_("unrecognized input"));
4155 if (whitespace_error && (ws_error_action == die_on_ws_error))
4156 apply = 0;
4158 update_index = check_index && apply;
4159 if (update_index && newfd < 0)
4160 newfd = hold_locked_index(&lock_file, 1);
4162 if (check_index) {
4163 if (read_cache() < 0)
4164 die(_("unable to read index file"));
4167 if ((check || apply) &&
4168 check_patch_list(list) < 0 &&
4169 !apply_with_reject)
4170 exit(1);
4172 if (apply && write_out_results(list)) {
4173 if (apply_with_reject)
4174 exit(1);
4175 /* with --3way, we still need to write the index out */
4176 return 1;
4179 if (fake_ancestor)
4180 build_fake_ancestor(list, fake_ancestor);
4182 if (diffstat)
4183 stat_patch_list(list);
4185 if (numstat)
4186 numstat_patch_list(list);
4188 if (summary)
4189 summary_patch_list(list);
4191 free_patch_list(list);
4192 strbuf_release(&buf);
4193 string_list_clear(&fn_table, 0);
4194 return 0;
4197 static int git_apply_config(const char *var, const char *value, void *cb)
4199 if (!strcmp(var, "apply.whitespace"))
4200 return git_config_string(&apply_default_whitespace, var, value);
4201 else if (!strcmp(var, "apply.ignorewhitespace"))
4202 return git_config_string(&apply_default_ignorewhitespace, var, value);
4203 return git_default_config(var, value, cb);
4206 static int option_parse_exclude(const struct option *opt,
4207 const char *arg, int unset)
4209 add_name_limit(arg, 1);
4210 return 0;
4213 static int option_parse_include(const struct option *opt,
4214 const char *arg, int unset)
4216 add_name_limit(arg, 0);
4217 has_include = 1;
4218 return 0;
4221 static int option_parse_p(const struct option *opt,
4222 const char *arg, int unset)
4224 p_value = atoi(arg);
4225 p_value_known = 1;
4226 return 0;
4229 static int option_parse_z(const struct option *opt,
4230 const char *arg, int unset)
4232 if (unset)
4233 line_termination = '\n';
4234 else
4235 line_termination = 0;
4236 return 0;
4239 static int option_parse_space_change(const struct option *opt,
4240 const char *arg, int unset)
4242 if (unset)
4243 ws_ignore_action = ignore_ws_none;
4244 else
4245 ws_ignore_action = ignore_ws_change;
4246 return 0;
4249 static int option_parse_whitespace(const struct option *opt,
4250 const char *arg, int unset)
4252 const char **whitespace_option = opt->value;
4254 *whitespace_option = arg;
4255 parse_whitespace_option(arg);
4256 return 0;
4259 static int option_parse_directory(const struct option *opt,
4260 const char *arg, int unset)
4262 root_len = strlen(arg);
4263 if (root_len && arg[root_len - 1] != '/') {
4264 char *new_root;
4265 root = new_root = xmalloc(root_len + 2);
4266 strcpy(new_root, arg);
4267 strcpy(new_root + root_len++, "/");
4268 } else
4269 root = arg;
4270 return 0;
4273 int cmd_apply(int argc, const char **argv, const char *prefix_)
4275 int i;
4276 int errs = 0;
4277 int is_not_gitdir = !startup_info->have_repository;
4278 int force_apply = 0;
4280 const char *whitespace_option = NULL;
4282 struct option builtin_apply_options[] = {
4283 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4284 N_("don't apply changes matching the given path"),
4285 0, option_parse_exclude },
4286 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4287 N_("apply changes matching the given path"),
4288 0, option_parse_include },
4289 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4290 N_("remove <num> leading slashes from traditional diff paths"),
4291 0, option_parse_p },
4292 OPT_BOOLEAN(0, "no-add", &no_add,
4293 N_("ignore additions made by the patch")),
4294 OPT_BOOLEAN(0, "stat", &diffstat,
4295 N_("instead of applying the patch, output diffstat for the input")),
4296 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4297 OPT_NOOP_NOARG(0, "binary"),
4298 OPT_BOOLEAN(0, "numstat", &numstat,
4299 N_("shows number of added and deleted lines in decimal notation")),
4300 OPT_BOOLEAN(0, "summary", &summary,
4301 N_("instead of applying the patch, output a summary for the input")),
4302 OPT_BOOLEAN(0, "check", &check,
4303 N_("instead of applying the patch, see if the patch is applicable")),
4304 OPT_BOOLEAN(0, "index", &check_index,
4305 N_("make sure the patch is applicable to the current index")),
4306 OPT_BOOLEAN(0, "cached", &cached,
4307 N_("apply a patch without touching the working tree")),
4308 OPT_BOOLEAN(0, "apply", &force_apply,
4309 N_("also apply the patch (use with --stat/--summary/--check)")),
4310 OPT_BOOL('3', "3way", &threeway,
4311 N_( "attempt three-way merge if a patch does not apply")),
4312 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4313 N_("build a temporary index based on embedded index information")),
4314 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4315 N_("paths are separated with NUL character"),
4316 PARSE_OPT_NOARG, option_parse_z },
4317 OPT_INTEGER('C', NULL, &p_context,
4318 N_("ensure at least <n> lines of context match")),
4319 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4320 N_("detect new or modified lines that have whitespace errors"),
4321 0, option_parse_whitespace },
4322 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4323 N_("ignore changes in whitespace when finding context"),
4324 PARSE_OPT_NOARG, option_parse_space_change },
4325 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4326 N_("ignore changes in whitespace when finding context"),
4327 PARSE_OPT_NOARG, option_parse_space_change },
4328 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4329 N_("apply the patch in reverse")),
4330 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4331 N_("don't expect at least one line of context")),
4332 OPT_BOOLEAN(0, "reject", &apply_with_reject,
4333 N_("leave the rejected hunks in corresponding *.rej files")),
4334 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4335 N_("allow overlapping hunks")),
4336 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4337 OPT_BIT(0, "inaccurate-eof", &options,
4338 N_("tolerate incorrectly detected missing new-line at the end of file"),
4339 INACCURATE_EOF),
4340 OPT_BIT(0, "recount", &options,
4341 N_("do not trust the line counts in the hunk headers"),
4342 RECOUNT),
4343 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4344 N_("prepend <root> to all filenames"),
4345 0, option_parse_directory },
4346 OPT_END()
4349 prefix = prefix_;
4350 prefix_length = prefix ? strlen(prefix) : 0;
4351 git_config(git_apply_config, NULL);
4352 if (apply_default_whitespace)
4353 parse_whitespace_option(apply_default_whitespace);
4354 if (apply_default_ignorewhitespace)
4355 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4357 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4358 apply_usage, 0);
4360 if (apply_with_reject && threeway)
4361 die("--reject and --3way cannot be used together.");
4362 if (cached && threeway)
4363 die("--cached and --3way cannot be used together.");
4364 if (threeway) {
4365 if (is_not_gitdir)
4366 die(_("--3way outside a repository"));
4367 check_index = 1;
4369 if (apply_with_reject)
4370 apply = apply_verbosely = 1;
4371 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4372 apply = 0;
4373 if (check_index && is_not_gitdir)
4374 die(_("--index outside a repository"));
4375 if (cached) {
4376 if (is_not_gitdir)
4377 die(_("--cached outside a repository"));
4378 check_index = 1;
4380 for (i = 0; i < argc; i++) {
4381 const char *arg = argv[i];
4382 int fd;
4384 if (!strcmp(arg, "-")) {
4385 errs |= apply_patch(0, "<stdin>", options);
4386 read_stdin = 0;
4387 continue;
4388 } else if (0 < prefix_length)
4389 arg = prefix_filename(prefix, prefix_length, arg);
4391 fd = open(arg, O_RDONLY);
4392 if (fd < 0)
4393 die_errno(_("can't open patch '%s'"), arg);
4394 read_stdin = 0;
4395 set_default_whitespace_mode(whitespace_option);
4396 errs |= apply_patch(fd, arg, options);
4397 close(fd);
4399 set_default_whitespace_mode(whitespace_option);
4400 if (read_stdin)
4401 errs |= apply_patch(0, "<stdin>", options);
4402 if (whitespace_error) {
4403 if (squelch_whitespace_errors &&
4404 squelch_whitespace_errors < whitespace_error) {
4405 int squelched =
4406 whitespace_error - squelch_whitespace_errors;
4407 warning(Q_("squelched %d whitespace error",
4408 "squelched %d whitespace errors",
4409 squelched),
4410 squelched);
4412 if (ws_error_action == die_on_ws_error)
4413 die(Q_("%d line adds whitespace errors.",
4414 "%d lines add whitespace errors.",
4415 whitespace_error),
4416 whitespace_error);
4417 if (applied_after_fixing_ws && apply)
4418 warning("%d line%s applied after"
4419 " fixing whitespace errors.",
4420 applied_after_fixing_ws,
4421 applied_after_fixing_ws == 1 ? "" : "s");
4422 else if (whitespace_error)
4423 warning(Q_("%d line adds whitespace errors.",
4424 "%d lines add whitespace errors.",
4425 whitespace_error),
4426 whitespace_error);
4429 if (update_index) {
4430 if (write_cache(newfd, active_cache, active_nr) ||
4431 commit_locked_index(&lock_file))
4432 die(_("Unable to write new index file"));
4435 return !!errs;