builtin/apply: tighten (dis)similarity index parsing
[alt-git.git] / builtin / apply.c
blobfd2b40e18af7a7b20e160be5fadbedfd4c188882
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 unsigned long val = strtoul(line, NULL, 10);
1045 if (val <= 100)
1046 patch->score = val;
1047 return 0;
1050 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1052 unsigned long val = strtoul(line, NULL, 10);
1053 if (val <= 100)
1054 patch->score = val;
1055 return 0;
1058 static int gitdiff_index(const char *line, struct patch *patch)
1061 * index line is N hexadecimal, "..", N hexadecimal,
1062 * and optional space with octal mode.
1064 const char *ptr, *eol;
1065 int len;
1067 ptr = strchr(line, '.');
1068 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1069 return 0;
1070 len = ptr - line;
1071 memcpy(patch->old_sha1_prefix, line, len);
1072 patch->old_sha1_prefix[len] = 0;
1074 line = ptr + 2;
1075 ptr = strchr(line, ' ');
1076 eol = strchr(line, '\n');
1078 if (!ptr || eol < ptr)
1079 ptr = eol;
1080 len = ptr - line;
1082 if (40 < len)
1083 return 0;
1084 memcpy(patch->new_sha1_prefix, line, len);
1085 patch->new_sha1_prefix[len] = 0;
1086 if (*ptr == ' ')
1087 patch->old_mode = strtoul(ptr+1, NULL, 8);
1088 return 0;
1092 * This is normal for a diff that doesn't change anything: we'll fall through
1093 * into the next diff. Tell the parser to break out.
1095 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1097 return -1;
1101 * Skip p_value leading components from "line"; as we do not accept
1102 * absolute paths, return NULL in that case.
1104 static const char *skip_tree_prefix(const char *line, int llen)
1106 int nslash;
1107 int i;
1109 if (!p_value)
1110 return (llen && line[0] == '/') ? NULL : line;
1112 nslash = p_value;
1113 for (i = 0; i < llen; i++) {
1114 int ch = line[i];
1115 if (ch == '/' && --nslash <= 0)
1116 return (i == 0) ? NULL : &line[i + 1];
1118 return NULL;
1122 * This is to extract the same name that appears on "diff --git"
1123 * line. We do not find and return anything if it is a rename
1124 * patch, and it is OK because we will find the name elsewhere.
1125 * We need to reliably find name only when it is mode-change only,
1126 * creation or deletion of an empty file. In any of these cases,
1127 * both sides are the same name under a/ and b/ respectively.
1129 static char *git_header_name(const char *line, int llen)
1131 const char *name;
1132 const char *second = NULL;
1133 size_t len, line_len;
1135 line += strlen("diff --git ");
1136 llen -= strlen("diff --git ");
1138 if (*line == '"') {
1139 const char *cp;
1140 struct strbuf first = STRBUF_INIT;
1141 struct strbuf sp = STRBUF_INIT;
1143 if (unquote_c_style(&first, line, &second))
1144 goto free_and_fail1;
1146 /* strip the a/b prefix including trailing slash */
1147 cp = skip_tree_prefix(first.buf, first.len);
1148 if (!cp)
1149 goto free_and_fail1;
1150 strbuf_remove(&first, 0, cp - first.buf);
1153 * second points at one past closing dq of name.
1154 * find the second name.
1156 while ((second < line + llen) && isspace(*second))
1157 second++;
1159 if (line + llen <= second)
1160 goto free_and_fail1;
1161 if (*second == '"') {
1162 if (unquote_c_style(&sp, second, NULL))
1163 goto free_and_fail1;
1164 cp = skip_tree_prefix(sp.buf, sp.len);
1165 if (!cp)
1166 goto free_and_fail1;
1167 /* They must match, otherwise ignore */
1168 if (strcmp(cp, first.buf))
1169 goto free_and_fail1;
1170 strbuf_release(&sp);
1171 return strbuf_detach(&first, NULL);
1174 /* unquoted second */
1175 cp = skip_tree_prefix(second, line + llen - second);
1176 if (!cp)
1177 goto free_and_fail1;
1178 if (line + llen - cp != first.len ||
1179 memcmp(first.buf, cp, first.len))
1180 goto free_and_fail1;
1181 return strbuf_detach(&first, NULL);
1183 free_and_fail1:
1184 strbuf_release(&first);
1185 strbuf_release(&sp);
1186 return NULL;
1189 /* unquoted first name */
1190 name = skip_tree_prefix(line, llen);
1191 if (!name)
1192 return NULL;
1195 * since the first name is unquoted, a dq if exists must be
1196 * the beginning of the second name.
1198 for (second = name; second < line + llen; second++) {
1199 if (*second == '"') {
1200 struct strbuf sp = STRBUF_INIT;
1201 const char *np;
1203 if (unquote_c_style(&sp, second, NULL))
1204 goto free_and_fail2;
1206 np = skip_tree_prefix(sp.buf, sp.len);
1207 if (!np)
1208 goto free_and_fail2;
1210 len = sp.buf + sp.len - np;
1211 if (len < second - name &&
1212 !strncmp(np, name, len) &&
1213 isspace(name[len])) {
1214 /* Good */
1215 strbuf_remove(&sp, 0, np - sp.buf);
1216 return strbuf_detach(&sp, NULL);
1219 free_and_fail2:
1220 strbuf_release(&sp);
1221 return NULL;
1226 * Accept a name only if it shows up twice, exactly the same
1227 * form.
1229 second = strchr(name, '\n');
1230 if (!second)
1231 return NULL;
1232 line_len = second - name;
1233 for (len = 0 ; ; len++) {
1234 switch (name[len]) {
1235 default:
1236 continue;
1237 case '\n':
1238 return NULL;
1239 case '\t': case ' ':
1241 * Is this the separator between the preimage
1242 * and the postimage pathname? Again, we are
1243 * only interested in the case where there is
1244 * no rename, as this is only to set def_name
1245 * and a rename patch has the names elsewhere
1246 * in an unambiguous form.
1248 if (!name[len + 1])
1249 return NULL; /* no postimage name */
1250 second = skip_tree_prefix(name + len + 1,
1251 line_len - (len + 1));
1252 if (!second)
1253 return NULL;
1255 * Does len bytes starting at "name" and "second"
1256 * (that are separated by one HT or SP we just
1257 * found) exactly match?
1259 if (second[len] == '\n' && !strncmp(name, second, len))
1260 return xmemdupz(name, len);
1265 /* Verify that we recognize the lines following a git header */
1266 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1268 unsigned long offset;
1270 /* A git diff has explicit new/delete information, so we don't guess */
1271 patch->is_new = 0;
1272 patch->is_delete = 0;
1275 * Some things may not have the old name in the
1276 * rest of the headers anywhere (pure mode changes,
1277 * or removing or adding empty files), so we get
1278 * the default name from the header.
1280 patch->def_name = git_header_name(line, len);
1281 if (patch->def_name && root) {
1282 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1283 strcpy(s, root);
1284 strcpy(s + root_len, patch->def_name);
1285 free(patch->def_name);
1286 patch->def_name = s;
1289 line += len;
1290 size -= len;
1291 linenr++;
1292 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1293 static const struct opentry {
1294 const char *str;
1295 int (*fn)(const char *, struct patch *);
1296 } optable[] = {
1297 { "@@ -", gitdiff_hdrend },
1298 { "--- ", gitdiff_oldname },
1299 { "+++ ", gitdiff_newname },
1300 { "old mode ", gitdiff_oldmode },
1301 { "new mode ", gitdiff_newmode },
1302 { "deleted file mode ", gitdiff_delete },
1303 { "new file mode ", gitdiff_newfile },
1304 { "copy from ", gitdiff_copysrc },
1305 { "copy to ", gitdiff_copydst },
1306 { "rename old ", gitdiff_renamesrc },
1307 { "rename new ", gitdiff_renamedst },
1308 { "rename from ", gitdiff_renamesrc },
1309 { "rename to ", gitdiff_renamedst },
1310 { "similarity index ", gitdiff_similarity },
1311 { "dissimilarity index ", gitdiff_dissimilarity },
1312 { "index ", gitdiff_index },
1313 { "", gitdiff_unrecognized },
1315 int i;
1317 len = linelen(line, size);
1318 if (!len || line[len-1] != '\n')
1319 break;
1320 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1321 const struct opentry *p = optable + i;
1322 int oplen = strlen(p->str);
1323 if (len < oplen || memcmp(p->str, line, oplen))
1324 continue;
1325 if (p->fn(line + oplen, patch) < 0)
1326 return offset;
1327 break;
1331 return offset;
1334 static int parse_num(const char *line, unsigned long *p)
1336 char *ptr;
1338 if (!isdigit(*line))
1339 return 0;
1340 *p = strtoul(line, &ptr, 10);
1341 return ptr - line;
1344 static int parse_range(const char *line, int len, int offset, const char *expect,
1345 unsigned long *p1, unsigned long *p2)
1347 int digits, ex;
1349 if (offset < 0 || offset >= len)
1350 return -1;
1351 line += offset;
1352 len -= offset;
1354 digits = parse_num(line, p1);
1355 if (!digits)
1356 return -1;
1358 offset += digits;
1359 line += digits;
1360 len -= digits;
1362 *p2 = 1;
1363 if (*line == ',') {
1364 digits = parse_num(line+1, p2);
1365 if (!digits)
1366 return -1;
1368 offset += digits+1;
1369 line += digits+1;
1370 len -= digits+1;
1373 ex = strlen(expect);
1374 if (ex > len)
1375 return -1;
1376 if (memcmp(line, expect, ex))
1377 return -1;
1379 return offset + ex;
1382 static void recount_diff(const char *line, int size, struct fragment *fragment)
1384 int oldlines = 0, newlines = 0, ret = 0;
1386 if (size < 1) {
1387 warning("recount: ignore empty hunk");
1388 return;
1391 for (;;) {
1392 int len = linelen(line, size);
1393 size -= len;
1394 line += len;
1396 if (size < 1)
1397 break;
1399 switch (*line) {
1400 case ' ': case '\n':
1401 newlines++;
1402 /* fall through */
1403 case '-':
1404 oldlines++;
1405 continue;
1406 case '+':
1407 newlines++;
1408 continue;
1409 case '\\':
1410 continue;
1411 case '@':
1412 ret = size < 3 || prefixcmp(line, "@@ ");
1413 break;
1414 case 'd':
1415 ret = size < 5 || prefixcmp(line, "diff ");
1416 break;
1417 default:
1418 ret = -1;
1419 break;
1421 if (ret) {
1422 warning(_("recount: unexpected line: %.*s"),
1423 (int)linelen(line, size), line);
1424 return;
1426 break;
1428 fragment->oldlines = oldlines;
1429 fragment->newlines = newlines;
1433 * Parse a unified diff fragment header of the
1434 * form "@@ -a,b +c,d @@"
1436 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1438 int offset;
1440 if (!len || line[len-1] != '\n')
1441 return -1;
1443 /* Figure out the number of lines in a fragment */
1444 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1445 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1447 return offset;
1450 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1452 unsigned long offset, len;
1454 patch->is_toplevel_relative = 0;
1455 patch->is_rename = patch->is_copy = 0;
1456 patch->is_new = patch->is_delete = -1;
1457 patch->old_mode = patch->new_mode = 0;
1458 patch->old_name = patch->new_name = NULL;
1459 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1460 unsigned long nextlen;
1462 len = linelen(line, size);
1463 if (!len)
1464 break;
1466 /* Testing this early allows us to take a few shortcuts.. */
1467 if (len < 6)
1468 continue;
1471 * Make sure we don't find any unconnected patch fragments.
1472 * That's a sign that we didn't find a header, and that a
1473 * patch has become corrupted/broken up.
1475 if (!memcmp("@@ -", line, 4)) {
1476 struct fragment dummy;
1477 if (parse_fragment_header(line, len, &dummy) < 0)
1478 continue;
1479 die(_("patch fragment without header at line %d: %.*s"),
1480 linenr, (int)len-1, line);
1483 if (size < len + 6)
1484 break;
1487 * Git patch? It might not have a real patch, just a rename
1488 * or mode change, so we handle that specially
1490 if (!memcmp("diff --git ", line, 11)) {
1491 int git_hdr_len = parse_git_header(line, len, size, patch);
1492 if (git_hdr_len <= len)
1493 continue;
1494 if (!patch->old_name && !patch->new_name) {
1495 if (!patch->def_name)
1496 die(Q_("git diff header lacks filename information when removing "
1497 "%d leading pathname component (line %d)",
1498 "git diff header lacks filename information when removing "
1499 "%d leading pathname components (line %d)",
1500 p_value),
1501 p_value, linenr);
1502 patch->old_name = xstrdup(patch->def_name);
1503 patch->new_name = xstrdup(patch->def_name);
1505 if (!patch->is_delete && !patch->new_name)
1506 die("git diff header lacks filename information "
1507 "(line %d)", linenr);
1508 patch->is_toplevel_relative = 1;
1509 *hdrsize = git_hdr_len;
1510 return offset;
1513 /* --- followed by +++ ? */
1514 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1515 continue;
1518 * We only accept unified patches, so we want it to
1519 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1520 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1522 nextlen = linelen(line + len, size - len);
1523 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1524 continue;
1526 /* Ok, we'll consider it a patch */
1527 parse_traditional_patch(line, line+len, patch);
1528 *hdrsize = len + nextlen;
1529 linenr += 2;
1530 return offset;
1532 return -1;
1535 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1537 char *err;
1539 if (!result)
1540 return;
1542 whitespace_error++;
1543 if (squelch_whitespace_errors &&
1544 squelch_whitespace_errors < whitespace_error)
1545 return;
1547 err = whitespace_error_string(result);
1548 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1549 patch_input_file, linenr, err, len, line);
1550 free(err);
1553 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1555 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1557 record_ws_error(result, line + 1, len - 2, linenr);
1561 * Parse a unified diff. Note that this really needs to parse each
1562 * fragment separately, since the only way to know the difference
1563 * between a "---" that is part of a patch, and a "---" that starts
1564 * the next patch is to look at the line counts..
1566 static int parse_fragment(const char *line, unsigned long size,
1567 struct patch *patch, struct fragment *fragment)
1569 int added, deleted;
1570 int len = linelen(line, size), offset;
1571 unsigned long oldlines, newlines;
1572 unsigned long leading, trailing;
1574 offset = parse_fragment_header(line, len, fragment);
1575 if (offset < 0)
1576 return -1;
1577 if (offset > 0 && patch->recount)
1578 recount_diff(line + offset, size - offset, fragment);
1579 oldlines = fragment->oldlines;
1580 newlines = fragment->newlines;
1581 leading = 0;
1582 trailing = 0;
1584 /* Parse the thing.. */
1585 line += len;
1586 size -= len;
1587 linenr++;
1588 added = deleted = 0;
1589 for (offset = len;
1590 0 < size;
1591 offset += len, size -= len, line += len, linenr++) {
1592 if (!oldlines && !newlines)
1593 break;
1594 len = linelen(line, size);
1595 if (!len || line[len-1] != '\n')
1596 return -1;
1597 switch (*line) {
1598 default:
1599 return -1;
1600 case '\n': /* newer GNU diff, an empty context line */
1601 case ' ':
1602 oldlines--;
1603 newlines--;
1604 if (!deleted && !added)
1605 leading++;
1606 trailing++;
1607 break;
1608 case '-':
1609 if (apply_in_reverse &&
1610 ws_error_action != nowarn_ws_error)
1611 check_whitespace(line, len, patch->ws_rule);
1612 deleted++;
1613 oldlines--;
1614 trailing = 0;
1615 break;
1616 case '+':
1617 if (!apply_in_reverse &&
1618 ws_error_action != nowarn_ws_error)
1619 check_whitespace(line, len, patch->ws_rule);
1620 added++;
1621 newlines--;
1622 trailing = 0;
1623 break;
1626 * We allow "\ No newline at end of file". Depending
1627 * on locale settings when the patch was produced we
1628 * don't know what this line looks like. The only
1629 * thing we do know is that it begins with "\ ".
1630 * Checking for 12 is just for sanity check -- any
1631 * l10n of "\ No newline..." is at least that long.
1633 case '\\':
1634 if (len < 12 || memcmp(line, "\\ ", 2))
1635 return -1;
1636 break;
1639 if (oldlines || newlines)
1640 return -1;
1641 fragment->leading = leading;
1642 fragment->trailing = trailing;
1645 * If a fragment ends with an incomplete line, we failed to include
1646 * it in the above loop because we hit oldlines == newlines == 0
1647 * before seeing it.
1649 if (12 < size && !memcmp(line, "\\ ", 2))
1650 offset += linelen(line, size);
1652 patch->lines_added += added;
1653 patch->lines_deleted += deleted;
1655 if (0 < patch->is_new && oldlines)
1656 return error(_("new file depends on old contents"));
1657 if (0 < patch->is_delete && newlines)
1658 return error(_("deleted file still has contents"));
1659 return offset;
1663 * We have seen "diff --git a/... b/..." header (or a traditional patch
1664 * header). Read hunks that belong to this patch into fragments and hang
1665 * them to the given patch structure.
1667 * The (fragment->patch, fragment->size) pair points into the memory given
1668 * by the caller, not a copy, when we return.
1670 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1672 unsigned long offset = 0;
1673 unsigned long oldlines = 0, newlines = 0, context = 0;
1674 struct fragment **fragp = &patch->fragments;
1676 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1677 struct fragment *fragment;
1678 int len;
1680 fragment = xcalloc(1, sizeof(*fragment));
1681 fragment->linenr = linenr;
1682 len = parse_fragment(line, size, patch, fragment);
1683 if (len <= 0)
1684 die(_("corrupt patch at line %d"), linenr);
1685 fragment->patch = line;
1686 fragment->size = len;
1687 oldlines += fragment->oldlines;
1688 newlines += fragment->newlines;
1689 context += fragment->leading + fragment->trailing;
1691 *fragp = fragment;
1692 fragp = &fragment->next;
1694 offset += len;
1695 line += len;
1696 size -= len;
1700 * If something was removed (i.e. we have old-lines) it cannot
1701 * be creation, and if something was added it cannot be
1702 * deletion. However, the reverse is not true; --unified=0
1703 * patches that only add are not necessarily creation even
1704 * though they do not have any old lines, and ones that only
1705 * delete are not necessarily deletion.
1707 * Unfortunately, a real creation/deletion patch do _not_ have
1708 * any context line by definition, so we cannot safely tell it
1709 * apart with --unified=0 insanity. At least if the patch has
1710 * more than one hunk it is not creation or deletion.
1712 if (patch->is_new < 0 &&
1713 (oldlines || (patch->fragments && patch->fragments->next)))
1714 patch->is_new = 0;
1715 if (patch->is_delete < 0 &&
1716 (newlines || (patch->fragments && patch->fragments->next)))
1717 patch->is_delete = 0;
1719 if (0 < patch->is_new && oldlines)
1720 die(_("new file %s depends on old contents"), patch->new_name);
1721 if (0 < patch->is_delete && newlines)
1722 die(_("deleted file %s still has contents"), patch->old_name);
1723 if (!patch->is_delete && !newlines && context)
1724 fprintf_ln(stderr,
1725 _("** warning: "
1726 "file %s becomes empty but is not deleted"),
1727 patch->new_name);
1729 return offset;
1732 static inline int metadata_changes(struct patch *patch)
1734 return patch->is_rename > 0 ||
1735 patch->is_copy > 0 ||
1736 patch->is_new > 0 ||
1737 patch->is_delete ||
1738 (patch->old_mode && patch->new_mode &&
1739 patch->old_mode != patch->new_mode);
1742 static char *inflate_it(const void *data, unsigned long size,
1743 unsigned long inflated_size)
1745 git_zstream stream;
1746 void *out;
1747 int st;
1749 memset(&stream, 0, sizeof(stream));
1751 stream.next_in = (unsigned char *)data;
1752 stream.avail_in = size;
1753 stream.next_out = out = xmalloc(inflated_size);
1754 stream.avail_out = inflated_size;
1755 git_inflate_init(&stream);
1756 st = git_inflate(&stream, Z_FINISH);
1757 git_inflate_end(&stream);
1758 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1759 free(out);
1760 return NULL;
1762 return out;
1766 * Read a binary hunk and return a new fragment; fragment->patch
1767 * points at an allocated memory that the caller must free, so
1768 * it is marked as "->free_patch = 1".
1770 static struct fragment *parse_binary_hunk(char **buf_p,
1771 unsigned long *sz_p,
1772 int *status_p,
1773 int *used_p)
1776 * Expect a line that begins with binary patch method ("literal"
1777 * or "delta"), followed by the length of data before deflating.
1778 * a sequence of 'length-byte' followed by base-85 encoded data
1779 * should follow, terminated by a newline.
1781 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1782 * and we would limit the patch line to 66 characters,
1783 * so one line can fit up to 13 groups that would decode
1784 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1785 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1787 int llen, used;
1788 unsigned long size = *sz_p;
1789 char *buffer = *buf_p;
1790 int patch_method;
1791 unsigned long origlen;
1792 char *data = NULL;
1793 int hunk_size = 0;
1794 struct fragment *frag;
1796 llen = linelen(buffer, size);
1797 used = llen;
1799 *status_p = 0;
1801 if (!prefixcmp(buffer, "delta ")) {
1802 patch_method = BINARY_DELTA_DEFLATED;
1803 origlen = strtoul(buffer + 6, NULL, 10);
1805 else if (!prefixcmp(buffer, "literal ")) {
1806 patch_method = BINARY_LITERAL_DEFLATED;
1807 origlen = strtoul(buffer + 8, NULL, 10);
1809 else
1810 return NULL;
1812 linenr++;
1813 buffer += llen;
1814 while (1) {
1815 int byte_length, max_byte_length, newsize;
1816 llen = linelen(buffer, size);
1817 used += llen;
1818 linenr++;
1819 if (llen == 1) {
1820 /* consume the blank line */
1821 buffer++;
1822 size--;
1823 break;
1826 * Minimum line is "A00000\n" which is 7-byte long,
1827 * and the line length must be multiple of 5 plus 2.
1829 if ((llen < 7) || (llen-2) % 5)
1830 goto corrupt;
1831 max_byte_length = (llen - 2) / 5 * 4;
1832 byte_length = *buffer;
1833 if ('A' <= byte_length && byte_length <= 'Z')
1834 byte_length = byte_length - 'A' + 1;
1835 else if ('a' <= byte_length && byte_length <= 'z')
1836 byte_length = byte_length - 'a' + 27;
1837 else
1838 goto corrupt;
1839 /* if the input length was not multiple of 4, we would
1840 * have filler at the end but the filler should never
1841 * exceed 3 bytes
1843 if (max_byte_length < byte_length ||
1844 byte_length <= max_byte_length - 4)
1845 goto corrupt;
1846 newsize = hunk_size + byte_length;
1847 data = xrealloc(data, newsize);
1848 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1849 goto corrupt;
1850 hunk_size = newsize;
1851 buffer += llen;
1852 size -= llen;
1855 frag = xcalloc(1, sizeof(*frag));
1856 frag->patch = inflate_it(data, hunk_size, origlen);
1857 frag->free_patch = 1;
1858 if (!frag->patch)
1859 goto corrupt;
1860 free(data);
1861 frag->size = origlen;
1862 *buf_p = buffer;
1863 *sz_p = size;
1864 *used_p = used;
1865 frag->binary_patch_method = patch_method;
1866 return frag;
1868 corrupt:
1869 free(data);
1870 *status_p = -1;
1871 error(_("corrupt binary patch at line %d: %.*s"),
1872 linenr-1, llen-1, buffer);
1873 return NULL;
1876 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1879 * We have read "GIT binary patch\n"; what follows is a line
1880 * that says the patch method (currently, either "literal" or
1881 * "delta") and the length of data before deflating; a
1882 * sequence of 'length-byte' followed by base-85 encoded data
1883 * follows.
1885 * When a binary patch is reversible, there is another binary
1886 * hunk in the same format, starting with patch method (either
1887 * "literal" or "delta") with the length of data, and a sequence
1888 * of length-byte + base-85 encoded data, terminated with another
1889 * empty line. This data, when applied to the postimage, produces
1890 * the preimage.
1892 struct fragment *forward;
1893 struct fragment *reverse;
1894 int status;
1895 int used, used_1;
1897 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1898 if (!forward && !status)
1899 /* there has to be one hunk (forward hunk) */
1900 return error(_("unrecognized binary patch at line %d"), linenr-1);
1901 if (status)
1902 /* otherwise we already gave an error message */
1903 return status;
1905 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1906 if (reverse)
1907 used += used_1;
1908 else if (status) {
1910 * Not having reverse hunk is not an error, but having
1911 * a corrupt reverse hunk is.
1913 free((void*) forward->patch);
1914 free(forward);
1915 return status;
1917 forward->next = reverse;
1918 patch->fragments = forward;
1919 patch->is_binary = 1;
1920 return used;
1924 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1925 * reading after seeing a single patch (i.e. changes to a single file).
1926 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1927 * Return the number of bytes consumed, so that the caller can call us
1928 * again for the next patch.
1930 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1932 int hdrsize, patchsize;
1933 int offset = find_header(buffer, size, &hdrsize, patch);
1935 if (offset < 0)
1936 return offset;
1938 patch->ws_rule = whitespace_rule(patch->new_name
1939 ? patch->new_name
1940 : patch->old_name);
1942 patchsize = parse_single_patch(buffer + offset + hdrsize,
1943 size - offset - hdrsize, patch);
1945 if (!patchsize) {
1946 static const char *binhdr[] = {
1947 "Binary files ",
1948 "Files ",
1949 NULL,
1951 static const char git_binary[] = "GIT binary patch\n";
1952 int i;
1953 int hd = hdrsize + offset;
1954 unsigned long llen = linelen(buffer + hd, size - hd);
1956 if (llen == sizeof(git_binary) - 1 &&
1957 !memcmp(git_binary, buffer + hd, llen)) {
1958 int used;
1959 linenr++;
1960 used = parse_binary(buffer + hd + llen,
1961 size - hd - llen, patch);
1962 if (used)
1963 patchsize = used + llen;
1964 else
1965 patchsize = 0;
1967 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1968 for (i = 0; binhdr[i]; i++) {
1969 int len = strlen(binhdr[i]);
1970 if (len < size - hd &&
1971 !memcmp(binhdr[i], buffer + hd, len)) {
1972 linenr++;
1973 patch->is_binary = 1;
1974 patchsize = llen;
1975 break;
1980 /* Empty patch cannot be applied if it is a text patch
1981 * without metadata change. A binary patch appears
1982 * empty to us here.
1984 if ((apply || check) &&
1985 (!patch->is_binary && !metadata_changes(patch)))
1986 die(_("patch with only garbage at line %d"), linenr);
1989 return offset + hdrsize + patchsize;
1992 #define swap(a,b) myswap((a),(b),sizeof(a))
1994 #define myswap(a, b, size) do { \
1995 unsigned char mytmp[size]; \
1996 memcpy(mytmp, &a, size); \
1997 memcpy(&a, &b, size); \
1998 memcpy(&b, mytmp, size); \
1999 } while (0)
2001 static void reverse_patches(struct patch *p)
2003 for (; p; p = p->next) {
2004 struct fragment *frag = p->fragments;
2006 swap(p->new_name, p->old_name);
2007 swap(p->new_mode, p->old_mode);
2008 swap(p->is_new, p->is_delete);
2009 swap(p->lines_added, p->lines_deleted);
2010 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2012 for (; frag; frag = frag->next) {
2013 swap(frag->newpos, frag->oldpos);
2014 swap(frag->newlines, frag->oldlines);
2019 static const char pluses[] =
2020 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2021 static const char minuses[]=
2022 "----------------------------------------------------------------------";
2024 static void show_stats(struct patch *patch)
2026 struct strbuf qname = STRBUF_INIT;
2027 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2028 int max, add, del;
2030 quote_c_style(cp, &qname, NULL, 0);
2033 * "scale" the filename
2035 max = max_len;
2036 if (max > 50)
2037 max = 50;
2039 if (qname.len > max) {
2040 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2041 if (!cp)
2042 cp = qname.buf + qname.len + 3 - max;
2043 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2046 if (patch->is_binary) {
2047 printf(" %-*s | Bin\n", max, qname.buf);
2048 strbuf_release(&qname);
2049 return;
2052 printf(" %-*s |", max, qname.buf);
2053 strbuf_release(&qname);
2056 * scale the add/delete
2058 max = max + max_change > 70 ? 70 - max : max_change;
2059 add = patch->lines_added;
2060 del = patch->lines_deleted;
2062 if (max_change > 0) {
2063 int total = ((add + del) * max + max_change / 2) / max_change;
2064 add = (add * max + max_change / 2) / max_change;
2065 del = total - add;
2067 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2068 add, pluses, del, minuses);
2071 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2073 switch (st->st_mode & S_IFMT) {
2074 case S_IFLNK:
2075 if (strbuf_readlink(buf, path, st->st_size) < 0)
2076 return error(_("unable to read symlink %s"), path);
2077 return 0;
2078 case S_IFREG:
2079 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2080 return error(_("unable to open or read %s"), path);
2081 convert_to_git(path, buf->buf, buf->len, buf, 0);
2082 return 0;
2083 default:
2084 return -1;
2089 * Update the preimage, and the common lines in postimage,
2090 * from buffer buf of length len. If postlen is 0 the postimage
2091 * is updated in place, otherwise it's updated on a new buffer
2092 * of length postlen
2095 static void update_pre_post_images(struct image *preimage,
2096 struct image *postimage,
2097 char *buf,
2098 size_t len, size_t postlen)
2100 int i, ctx;
2101 char *new, *old, *fixed;
2102 struct image fixed_preimage;
2105 * Update the preimage with whitespace fixes. Note that we
2106 * are not losing preimage->buf -- apply_one_fragment() will
2107 * free "oldlines".
2109 prepare_image(&fixed_preimage, buf, len, 1);
2110 assert(fixed_preimage.nr == preimage->nr);
2111 for (i = 0; i < preimage->nr; i++)
2112 fixed_preimage.line[i].flag = preimage->line[i].flag;
2113 free(preimage->line_allocated);
2114 *preimage = fixed_preimage;
2117 * Adjust the common context lines in postimage. This can be
2118 * done in-place when we are just doing whitespace fixing,
2119 * which does not make the string grow, but needs a new buffer
2120 * when ignoring whitespace causes the update, since in this case
2121 * we could have e.g. tabs converted to multiple spaces.
2122 * We trust the caller to tell us if the update can be done
2123 * in place (postlen==0) or not.
2125 old = postimage->buf;
2126 if (postlen)
2127 new = postimage->buf = xmalloc(postlen);
2128 else
2129 new = old;
2130 fixed = preimage->buf;
2131 for (i = ctx = 0; i < postimage->nr; i++) {
2132 size_t len = postimage->line[i].len;
2133 if (!(postimage->line[i].flag & LINE_COMMON)) {
2134 /* an added line -- no counterparts in preimage */
2135 memmove(new, old, len);
2136 old += len;
2137 new += len;
2138 continue;
2141 /* a common context -- skip it in the original postimage */
2142 old += len;
2144 /* and find the corresponding one in the fixed preimage */
2145 while (ctx < preimage->nr &&
2146 !(preimage->line[ctx].flag & LINE_COMMON)) {
2147 fixed += preimage->line[ctx].len;
2148 ctx++;
2150 if (preimage->nr <= ctx)
2151 die(_("oops"));
2153 /* and copy it in, while fixing the line length */
2154 len = preimage->line[ctx].len;
2155 memcpy(new, fixed, len);
2156 new += len;
2157 fixed += len;
2158 postimage->line[i].len = len;
2159 ctx++;
2162 /* Fix the length of the whole thing */
2163 postimage->len = new - postimage->buf;
2166 static int match_fragment(struct image *img,
2167 struct image *preimage,
2168 struct image *postimage,
2169 unsigned long try,
2170 int try_lno,
2171 unsigned ws_rule,
2172 int match_beginning, int match_end)
2174 int i;
2175 char *fixed_buf, *buf, *orig, *target;
2176 struct strbuf fixed;
2177 size_t fixed_len;
2178 int preimage_limit;
2180 if (preimage->nr + try_lno <= img->nr) {
2182 * The hunk falls within the boundaries of img.
2184 preimage_limit = preimage->nr;
2185 if (match_end && (preimage->nr + try_lno != img->nr))
2186 return 0;
2187 } else if (ws_error_action == correct_ws_error &&
2188 (ws_rule & WS_BLANK_AT_EOF)) {
2190 * This hunk extends beyond the end of img, and we are
2191 * removing blank lines at the end of the file. This
2192 * many lines from the beginning of the preimage must
2193 * match with img, and the remainder of the preimage
2194 * must be blank.
2196 preimage_limit = img->nr - try_lno;
2197 } else {
2199 * The hunk extends beyond the end of the img and
2200 * we are not removing blanks at the end, so we
2201 * should reject the hunk at this position.
2203 return 0;
2206 if (match_beginning && try_lno)
2207 return 0;
2209 /* Quick hash check */
2210 for (i = 0; i < preimage_limit; i++)
2211 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2212 (preimage->line[i].hash != img->line[try_lno + i].hash))
2213 return 0;
2215 if (preimage_limit == preimage->nr) {
2217 * Do we have an exact match? If we were told to match
2218 * at the end, size must be exactly at try+fragsize,
2219 * otherwise try+fragsize must be still within the preimage,
2220 * and either case, the old piece should match the preimage
2221 * exactly.
2223 if ((match_end
2224 ? (try + preimage->len == img->len)
2225 : (try + preimage->len <= img->len)) &&
2226 !memcmp(img->buf + try, preimage->buf, preimage->len))
2227 return 1;
2228 } else {
2230 * The preimage extends beyond the end of img, so
2231 * there cannot be an exact match.
2233 * There must be one non-blank context line that match
2234 * a line before the end of img.
2236 char *buf_end;
2238 buf = preimage->buf;
2239 buf_end = buf;
2240 for (i = 0; i < preimage_limit; i++)
2241 buf_end += preimage->line[i].len;
2243 for ( ; buf < buf_end; buf++)
2244 if (!isspace(*buf))
2245 break;
2246 if (buf == buf_end)
2247 return 0;
2251 * No exact match. If we are ignoring whitespace, run a line-by-line
2252 * fuzzy matching. We collect all the line length information because
2253 * we need it to adjust whitespace if we match.
2255 if (ws_ignore_action == ignore_ws_change) {
2256 size_t imgoff = 0;
2257 size_t preoff = 0;
2258 size_t postlen = postimage->len;
2259 size_t extra_chars;
2260 char *preimage_eof;
2261 char *preimage_end;
2262 for (i = 0; i < preimage_limit; i++) {
2263 size_t prelen = preimage->line[i].len;
2264 size_t imglen = img->line[try_lno+i].len;
2266 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2267 preimage->buf + preoff, prelen))
2268 return 0;
2269 if (preimage->line[i].flag & LINE_COMMON)
2270 postlen += imglen - prelen;
2271 imgoff += imglen;
2272 preoff += prelen;
2276 * Ok, the preimage matches with whitespace fuzz.
2278 * imgoff now holds the true length of the target that
2279 * matches the preimage before the end of the file.
2281 * Count the number of characters in the preimage that fall
2282 * beyond the end of the file and make sure that all of them
2283 * are whitespace characters. (This can only happen if
2284 * we are removing blank lines at the end of the file.)
2286 buf = preimage_eof = preimage->buf + preoff;
2287 for ( ; i < preimage->nr; i++)
2288 preoff += preimage->line[i].len;
2289 preimage_end = preimage->buf + preoff;
2290 for ( ; buf < preimage_end; buf++)
2291 if (!isspace(*buf))
2292 return 0;
2295 * Update the preimage and the common postimage context
2296 * lines to use the same whitespace as the target.
2297 * If whitespace is missing in the target (i.e.
2298 * if the preimage extends beyond the end of the file),
2299 * use the whitespace from the preimage.
2301 extra_chars = preimage_end - preimage_eof;
2302 strbuf_init(&fixed, imgoff + extra_chars);
2303 strbuf_add(&fixed, img->buf + try, imgoff);
2304 strbuf_add(&fixed, preimage_eof, extra_chars);
2305 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2306 update_pre_post_images(preimage, postimage,
2307 fixed_buf, fixed_len, postlen);
2308 return 1;
2311 if (ws_error_action != correct_ws_error)
2312 return 0;
2315 * The hunk does not apply byte-by-byte, but the hash says
2316 * it might with whitespace fuzz. We haven't been asked to
2317 * ignore whitespace, we were asked to correct whitespace
2318 * errors, so let's try matching after whitespace correction.
2320 * The preimage may extend beyond the end of the file,
2321 * but in this loop we will only handle the part of the
2322 * preimage that falls within the file.
2324 strbuf_init(&fixed, preimage->len + 1);
2325 orig = preimage->buf;
2326 target = img->buf + try;
2327 for (i = 0; i < preimage_limit; i++) {
2328 size_t oldlen = preimage->line[i].len;
2329 size_t tgtlen = img->line[try_lno + i].len;
2330 size_t fixstart = fixed.len;
2331 struct strbuf tgtfix;
2332 int match;
2334 /* Try fixing the line in the preimage */
2335 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2337 /* Try fixing the line in the target */
2338 strbuf_init(&tgtfix, tgtlen);
2339 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2342 * If they match, either the preimage was based on
2343 * a version before our tree fixed whitespace breakage,
2344 * or we are lacking a whitespace-fix patch the tree
2345 * the preimage was based on already had (i.e. target
2346 * has whitespace breakage, the preimage doesn't).
2347 * In either case, we are fixing the whitespace breakages
2348 * so we might as well take the fix together with their
2349 * real change.
2351 match = (tgtfix.len == fixed.len - fixstart &&
2352 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2353 fixed.len - fixstart));
2355 strbuf_release(&tgtfix);
2356 if (!match)
2357 goto unmatch_exit;
2359 orig += oldlen;
2360 target += tgtlen;
2365 * Now handle the lines in the preimage that falls beyond the
2366 * end of the file (if any). They will only match if they are
2367 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2368 * false).
2370 for ( ; i < preimage->nr; i++) {
2371 size_t fixstart = fixed.len; /* start of the fixed preimage */
2372 size_t oldlen = preimage->line[i].len;
2373 int j;
2375 /* Try fixing the line in the preimage */
2376 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2378 for (j = fixstart; j < fixed.len; j++)
2379 if (!isspace(fixed.buf[j]))
2380 goto unmatch_exit;
2382 orig += oldlen;
2386 * Yes, the preimage is based on an older version that still
2387 * has whitespace breakages unfixed, and fixing them makes the
2388 * hunk match. Update the context lines in the postimage.
2390 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2391 update_pre_post_images(preimage, postimage,
2392 fixed_buf, fixed_len, 0);
2393 return 1;
2395 unmatch_exit:
2396 strbuf_release(&fixed);
2397 return 0;
2400 static int find_pos(struct image *img,
2401 struct image *preimage,
2402 struct image *postimage,
2403 int line,
2404 unsigned ws_rule,
2405 int match_beginning, int match_end)
2407 int i;
2408 unsigned long backwards, forwards, try;
2409 int backwards_lno, forwards_lno, try_lno;
2412 * If match_beginning or match_end is specified, there is no
2413 * point starting from a wrong line that will never match and
2414 * wander around and wait for a match at the specified end.
2416 if (match_beginning)
2417 line = 0;
2418 else if (match_end)
2419 line = img->nr - preimage->nr;
2422 * Because the comparison is unsigned, the following test
2423 * will also take care of a negative line number that can
2424 * result when match_end and preimage is larger than the target.
2426 if ((size_t) line > img->nr)
2427 line = img->nr;
2429 try = 0;
2430 for (i = 0; i < line; i++)
2431 try += img->line[i].len;
2434 * There's probably some smart way to do this, but I'll leave
2435 * that to the smart and beautiful people. I'm simple and stupid.
2437 backwards = try;
2438 backwards_lno = line;
2439 forwards = try;
2440 forwards_lno = line;
2441 try_lno = line;
2443 for (i = 0; ; i++) {
2444 if (match_fragment(img, preimage, postimage,
2445 try, try_lno, ws_rule,
2446 match_beginning, match_end))
2447 return try_lno;
2449 again:
2450 if (backwards_lno == 0 && forwards_lno == img->nr)
2451 break;
2453 if (i & 1) {
2454 if (backwards_lno == 0) {
2455 i++;
2456 goto again;
2458 backwards_lno--;
2459 backwards -= img->line[backwards_lno].len;
2460 try = backwards;
2461 try_lno = backwards_lno;
2462 } else {
2463 if (forwards_lno == img->nr) {
2464 i++;
2465 goto again;
2467 forwards += img->line[forwards_lno].len;
2468 forwards_lno++;
2469 try = forwards;
2470 try_lno = forwards_lno;
2474 return -1;
2477 static void remove_first_line(struct image *img)
2479 img->buf += img->line[0].len;
2480 img->len -= img->line[0].len;
2481 img->line++;
2482 img->nr--;
2485 static void remove_last_line(struct image *img)
2487 img->len -= img->line[--img->nr].len;
2491 * The change from "preimage" and "postimage" has been found to
2492 * apply at applied_pos (counts in line numbers) in "img".
2493 * Update "img" to remove "preimage" and replace it with "postimage".
2495 static void update_image(struct image *img,
2496 int applied_pos,
2497 struct image *preimage,
2498 struct image *postimage)
2501 * remove the copy of preimage at offset in img
2502 * and replace it with postimage
2504 int i, nr;
2505 size_t remove_count, insert_count, applied_at = 0;
2506 char *result;
2507 int preimage_limit;
2510 * If we are removing blank lines at the end of img,
2511 * the preimage may extend beyond the end.
2512 * If that is the case, we must be careful only to
2513 * remove the part of the preimage that falls within
2514 * the boundaries of img. Initialize preimage_limit
2515 * to the number of lines in the preimage that falls
2516 * within the boundaries.
2518 preimage_limit = preimage->nr;
2519 if (preimage_limit > img->nr - applied_pos)
2520 preimage_limit = img->nr - applied_pos;
2522 for (i = 0; i < applied_pos; i++)
2523 applied_at += img->line[i].len;
2525 remove_count = 0;
2526 for (i = 0; i < preimage_limit; i++)
2527 remove_count += img->line[applied_pos + i].len;
2528 insert_count = postimage->len;
2530 /* Adjust the contents */
2531 result = xmalloc(img->len + insert_count - remove_count + 1);
2532 memcpy(result, img->buf, applied_at);
2533 memcpy(result + applied_at, postimage->buf, postimage->len);
2534 memcpy(result + applied_at + postimage->len,
2535 img->buf + (applied_at + remove_count),
2536 img->len - (applied_at + remove_count));
2537 free(img->buf);
2538 img->buf = result;
2539 img->len += insert_count - remove_count;
2540 result[img->len] = '\0';
2542 /* Adjust the line table */
2543 nr = img->nr + postimage->nr - preimage_limit;
2544 if (preimage_limit < postimage->nr) {
2546 * NOTE: this knows that we never call remove_first_line()
2547 * on anything other than pre/post image.
2549 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2550 img->line_allocated = img->line;
2552 if (preimage_limit != postimage->nr)
2553 memmove(img->line + applied_pos + postimage->nr,
2554 img->line + applied_pos + preimage_limit,
2555 (img->nr - (applied_pos + preimage_limit)) *
2556 sizeof(*img->line));
2557 memcpy(img->line + applied_pos,
2558 postimage->line,
2559 postimage->nr * sizeof(*img->line));
2560 if (!allow_overlap)
2561 for (i = 0; i < postimage->nr; i++)
2562 img->line[applied_pos + i].flag |= LINE_PATCHED;
2563 img->nr = nr;
2567 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2568 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2569 * replace the part of "img" with "postimage" text.
2571 static int apply_one_fragment(struct image *img, struct fragment *frag,
2572 int inaccurate_eof, unsigned ws_rule,
2573 int nth_fragment)
2575 int match_beginning, match_end;
2576 const char *patch = frag->patch;
2577 int size = frag->size;
2578 char *old, *oldlines;
2579 struct strbuf newlines;
2580 int new_blank_lines_at_end = 0;
2581 int found_new_blank_lines_at_end = 0;
2582 int hunk_linenr = frag->linenr;
2583 unsigned long leading, trailing;
2584 int pos, applied_pos;
2585 struct image preimage;
2586 struct image postimage;
2588 memset(&preimage, 0, sizeof(preimage));
2589 memset(&postimage, 0, sizeof(postimage));
2590 oldlines = xmalloc(size);
2591 strbuf_init(&newlines, size);
2593 old = oldlines;
2594 while (size > 0) {
2595 char first;
2596 int len = linelen(patch, size);
2597 int plen;
2598 int added_blank_line = 0;
2599 int is_blank_context = 0;
2600 size_t start;
2602 if (!len)
2603 break;
2606 * "plen" is how much of the line we should use for
2607 * the actual patch data. Normally we just remove the
2608 * first character on the line, but if the line is
2609 * followed by "\ No newline", then we also remove the
2610 * last one (which is the newline, of course).
2612 plen = len - 1;
2613 if (len < size && patch[len] == '\\')
2614 plen--;
2615 first = *patch;
2616 if (apply_in_reverse) {
2617 if (first == '-')
2618 first = '+';
2619 else if (first == '+')
2620 first = '-';
2623 switch (first) {
2624 case '\n':
2625 /* Newer GNU diff, empty context line */
2626 if (plen < 0)
2627 /* ... followed by '\No newline'; nothing */
2628 break;
2629 *old++ = '\n';
2630 strbuf_addch(&newlines, '\n');
2631 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2632 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2633 is_blank_context = 1;
2634 break;
2635 case ' ':
2636 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2637 ws_blank_line(patch + 1, plen, ws_rule))
2638 is_blank_context = 1;
2639 case '-':
2640 memcpy(old, patch + 1, plen);
2641 add_line_info(&preimage, old, plen,
2642 (first == ' ' ? LINE_COMMON : 0));
2643 old += plen;
2644 if (first == '-')
2645 break;
2646 /* Fall-through for ' ' */
2647 case '+':
2648 /* --no-add does not add new lines */
2649 if (first == '+' && no_add)
2650 break;
2652 start = newlines.len;
2653 if (first != '+' ||
2654 !whitespace_error ||
2655 ws_error_action != correct_ws_error) {
2656 strbuf_add(&newlines, patch + 1, plen);
2658 else {
2659 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2661 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2662 (first == '+' ? 0 : LINE_COMMON));
2663 if (first == '+' &&
2664 (ws_rule & WS_BLANK_AT_EOF) &&
2665 ws_blank_line(patch + 1, plen, ws_rule))
2666 added_blank_line = 1;
2667 break;
2668 case '@': case '\\':
2669 /* Ignore it, we already handled it */
2670 break;
2671 default:
2672 if (apply_verbosely)
2673 error(_("invalid start of line: '%c'"), first);
2674 return -1;
2676 if (added_blank_line) {
2677 if (!new_blank_lines_at_end)
2678 found_new_blank_lines_at_end = hunk_linenr;
2679 new_blank_lines_at_end++;
2681 else if (is_blank_context)
2683 else
2684 new_blank_lines_at_end = 0;
2685 patch += len;
2686 size -= len;
2687 hunk_linenr++;
2689 if (inaccurate_eof &&
2690 old > oldlines && old[-1] == '\n' &&
2691 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2692 old--;
2693 strbuf_setlen(&newlines, newlines.len - 1);
2696 leading = frag->leading;
2697 trailing = frag->trailing;
2700 * A hunk to change lines at the beginning would begin with
2701 * @@ -1,L +N,M @@
2702 * but we need to be careful. -U0 that inserts before the second
2703 * line also has this pattern.
2705 * And a hunk to add to an empty file would begin with
2706 * @@ -0,0 +N,M @@
2708 * In other words, a hunk that is (frag->oldpos <= 1) with or
2709 * without leading context must match at the beginning.
2711 match_beginning = (!frag->oldpos ||
2712 (frag->oldpos == 1 && !unidiff_zero));
2715 * A hunk without trailing lines must match at the end.
2716 * However, we simply cannot tell if a hunk must match end
2717 * from the lack of trailing lines if the patch was generated
2718 * with unidiff without any context.
2720 match_end = !unidiff_zero && !trailing;
2722 pos = frag->newpos ? (frag->newpos - 1) : 0;
2723 preimage.buf = oldlines;
2724 preimage.len = old - oldlines;
2725 postimage.buf = newlines.buf;
2726 postimage.len = newlines.len;
2727 preimage.line = preimage.line_allocated;
2728 postimage.line = postimage.line_allocated;
2730 for (;;) {
2732 applied_pos = find_pos(img, &preimage, &postimage, pos,
2733 ws_rule, match_beginning, match_end);
2735 if (applied_pos >= 0)
2736 break;
2738 /* Am I at my context limits? */
2739 if ((leading <= p_context) && (trailing <= p_context))
2740 break;
2741 if (match_beginning || match_end) {
2742 match_beginning = match_end = 0;
2743 continue;
2747 * Reduce the number of context lines; reduce both
2748 * leading and trailing if they are equal otherwise
2749 * just reduce the larger context.
2751 if (leading >= trailing) {
2752 remove_first_line(&preimage);
2753 remove_first_line(&postimage);
2754 pos--;
2755 leading--;
2757 if (trailing > leading) {
2758 remove_last_line(&preimage);
2759 remove_last_line(&postimage);
2760 trailing--;
2764 if (applied_pos >= 0) {
2765 if (new_blank_lines_at_end &&
2766 preimage.nr + applied_pos >= img->nr &&
2767 (ws_rule & WS_BLANK_AT_EOF) &&
2768 ws_error_action != nowarn_ws_error) {
2769 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2770 found_new_blank_lines_at_end);
2771 if (ws_error_action == correct_ws_error) {
2772 while (new_blank_lines_at_end--)
2773 remove_last_line(&postimage);
2776 * We would want to prevent write_out_results()
2777 * from taking place in apply_patch() that follows
2778 * the callchain led us here, which is:
2779 * apply_patch->check_patch_list->check_patch->
2780 * apply_data->apply_fragments->apply_one_fragment
2782 if (ws_error_action == die_on_ws_error)
2783 apply = 0;
2786 if (apply_verbosely && applied_pos != pos) {
2787 int offset = applied_pos - pos;
2788 if (apply_in_reverse)
2789 offset = 0 - offset;
2790 fprintf_ln(stderr,
2791 Q_("Hunk #%d succeeded at %d (offset %d line).",
2792 "Hunk #%d succeeded at %d (offset %d lines).",
2793 offset),
2794 nth_fragment, applied_pos + 1, offset);
2798 * Warn if it was necessary to reduce the number
2799 * of context lines.
2801 if ((leading != frag->leading) ||
2802 (trailing != frag->trailing))
2803 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2804 " to apply fragment at %d"),
2805 leading, trailing, applied_pos+1);
2806 update_image(img, applied_pos, &preimage, &postimage);
2807 } else {
2808 if (apply_verbosely)
2809 error(_("while searching for:\n%.*s"),
2810 (int)(old - oldlines), oldlines);
2813 free(oldlines);
2814 strbuf_release(&newlines);
2815 free(preimage.line_allocated);
2816 free(postimage.line_allocated);
2818 return (applied_pos < 0);
2821 static int apply_binary_fragment(struct image *img, struct patch *patch)
2823 struct fragment *fragment = patch->fragments;
2824 unsigned long len;
2825 void *dst;
2827 if (!fragment)
2828 return error(_("missing binary patch data for '%s'"),
2829 patch->new_name ?
2830 patch->new_name :
2831 patch->old_name);
2833 /* Binary patch is irreversible without the optional second hunk */
2834 if (apply_in_reverse) {
2835 if (!fragment->next)
2836 return error("cannot reverse-apply a binary patch "
2837 "without the reverse hunk to '%s'",
2838 patch->new_name
2839 ? patch->new_name : patch->old_name);
2840 fragment = fragment->next;
2842 switch (fragment->binary_patch_method) {
2843 case BINARY_DELTA_DEFLATED:
2844 dst = patch_delta(img->buf, img->len, fragment->patch,
2845 fragment->size, &len);
2846 if (!dst)
2847 return -1;
2848 clear_image(img);
2849 img->buf = dst;
2850 img->len = len;
2851 return 0;
2852 case BINARY_LITERAL_DEFLATED:
2853 clear_image(img);
2854 img->len = fragment->size;
2855 img->buf = xmalloc(img->len+1);
2856 memcpy(img->buf, fragment->patch, img->len);
2857 img->buf[img->len] = '\0';
2858 return 0;
2860 return -1;
2864 * Replace "img" with the result of applying the binary patch.
2865 * The binary patch data itself in patch->fragment is still kept
2866 * but the preimage prepared by the caller in "img" is freed here
2867 * or in the helper function apply_binary_fragment() this calls.
2869 static int apply_binary(struct image *img, struct patch *patch)
2871 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2872 unsigned char sha1[20];
2875 * For safety, we require patch index line to contain
2876 * full 40-byte textual SHA1 for old and new, at least for now.
2878 if (strlen(patch->old_sha1_prefix) != 40 ||
2879 strlen(patch->new_sha1_prefix) != 40 ||
2880 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2881 get_sha1_hex(patch->new_sha1_prefix, sha1))
2882 return error("cannot apply binary patch to '%s' "
2883 "without full index line", name);
2885 if (patch->old_name) {
2887 * See if the old one matches what the patch
2888 * applies to.
2890 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2891 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2892 return error("the patch applies to '%s' (%s), "
2893 "which does not match the "
2894 "current contents.",
2895 name, sha1_to_hex(sha1));
2897 else {
2898 /* Otherwise, the old one must be empty. */
2899 if (img->len)
2900 return error("the patch applies to an empty "
2901 "'%s' but it is not empty", name);
2904 get_sha1_hex(patch->new_sha1_prefix, sha1);
2905 if (is_null_sha1(sha1)) {
2906 clear_image(img);
2907 return 0; /* deletion patch */
2910 if (has_sha1_file(sha1)) {
2911 /* We already have the postimage */
2912 enum object_type type;
2913 unsigned long size;
2914 char *result;
2916 result = read_sha1_file(sha1, &type, &size);
2917 if (!result)
2918 return error("the necessary postimage %s for "
2919 "'%s' cannot be read",
2920 patch->new_sha1_prefix, name);
2921 clear_image(img);
2922 img->buf = result;
2923 img->len = size;
2924 } else {
2926 * We have verified buf matches the preimage;
2927 * apply the patch data to it, which is stored
2928 * in the patch->fragments->{patch,size}.
2930 if (apply_binary_fragment(img, patch))
2931 return error(_("binary patch does not apply to '%s'"),
2932 name);
2934 /* verify that the result matches */
2935 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2936 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2937 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2938 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2941 return 0;
2944 static int apply_fragments(struct image *img, struct patch *patch)
2946 struct fragment *frag = patch->fragments;
2947 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2948 unsigned ws_rule = patch->ws_rule;
2949 unsigned inaccurate_eof = patch->inaccurate_eof;
2950 int nth = 0;
2952 if (patch->is_binary)
2953 return apply_binary(img, patch);
2955 while (frag) {
2956 nth++;
2957 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2958 error(_("patch failed: %s:%ld"), name, frag->oldpos);
2959 if (!apply_with_reject)
2960 return -1;
2961 frag->rejected = 1;
2963 frag = frag->next;
2965 return 0;
2968 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2970 if (S_ISGITLINK(mode)) {
2971 strbuf_grow(buf, 100);
2972 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2973 } else {
2974 enum object_type type;
2975 unsigned long sz;
2976 char *result;
2978 result = read_sha1_file(sha1, &type, &sz);
2979 if (!result)
2980 return -1;
2981 /* XXX read_sha1_file NUL-terminates */
2982 strbuf_attach(buf, result, sz, sz + 1);
2984 return 0;
2987 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2989 if (!ce)
2990 return 0;
2991 return read_blob_object(buf, ce->sha1, ce->ce_mode);
2994 static struct patch *in_fn_table(const char *name)
2996 struct string_list_item *item;
2998 if (name == NULL)
2999 return NULL;
3001 item = string_list_lookup(&fn_table, name);
3002 if (item != NULL)
3003 return (struct patch *)item->util;
3005 return NULL;
3009 * item->util in the filename table records the status of the path.
3010 * Usually it points at a patch (whose result records the contents
3011 * of it after applying it), but it could be PATH_WAS_DELETED for a
3012 * path that a previously applied patch has already removed, or
3013 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3015 * The latter is needed to deal with a case where two paths A and B
3016 * are swapped by first renaming A to B and then renaming B to A;
3017 * moving A to B should not be prevented due to presense of B as we
3018 * will remove it in a later patch.
3020 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3021 #define PATH_WAS_DELETED ((struct patch *) -1)
3023 static int to_be_deleted(struct patch *patch)
3025 return patch == PATH_TO_BE_DELETED;
3028 static int was_deleted(struct patch *patch)
3030 return patch == PATH_WAS_DELETED;
3033 static void add_to_fn_table(struct patch *patch)
3035 struct string_list_item *item;
3038 * Always add new_name unless patch is a deletion
3039 * This should cover the cases for normal diffs,
3040 * file creations and copies
3042 if (patch->new_name != NULL) {
3043 item = string_list_insert(&fn_table, patch->new_name);
3044 item->util = patch;
3048 * store a failure on rename/deletion cases because
3049 * later chunks shouldn't patch old names
3051 if ((patch->new_name == NULL) || (patch->is_rename)) {
3052 item = string_list_insert(&fn_table, patch->old_name);
3053 item->util = PATH_WAS_DELETED;
3057 static void prepare_fn_table(struct patch *patch)
3060 * store information about incoming file deletion
3062 while (patch) {
3063 if ((patch->new_name == NULL) || (patch->is_rename)) {
3064 struct string_list_item *item;
3065 item = string_list_insert(&fn_table, patch->old_name);
3066 item->util = PATH_TO_BE_DELETED;
3068 patch = patch->next;
3072 static int checkout_target(struct cache_entry *ce, struct stat *st)
3074 struct checkout costate;
3076 memset(&costate, 0, sizeof(costate));
3077 costate.base_dir = "";
3078 costate.refresh_cache = 1;
3079 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3080 return error(_("cannot checkout %s"), ce->name);
3081 return 0;
3084 static struct patch *previous_patch(struct patch *patch, int *gone)
3086 struct patch *previous;
3088 *gone = 0;
3089 if (patch->is_copy || patch->is_rename)
3090 return NULL; /* "git" patches do not depend on the order */
3092 previous = in_fn_table(patch->old_name);
3093 if (!previous)
3094 return NULL;
3096 if (to_be_deleted(previous))
3097 return NULL; /* the deletion hasn't happened yet */
3099 if (was_deleted(previous))
3100 *gone = 1;
3102 return previous;
3105 static int verify_index_match(struct cache_entry *ce, struct stat *st)
3107 if (S_ISGITLINK(ce->ce_mode)) {
3108 if (!S_ISDIR(st->st_mode))
3109 return -1;
3110 return 0;
3112 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3115 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3117 static int load_patch_target(struct strbuf *buf,
3118 struct cache_entry *ce,
3119 struct stat *st,
3120 const char *name,
3121 unsigned expected_mode)
3123 if (cached) {
3124 if (read_file_or_gitlink(ce, buf))
3125 return error(_("read of %s failed"), name);
3126 } else if (name) {
3127 if (S_ISGITLINK(expected_mode)) {
3128 if (ce)
3129 return read_file_or_gitlink(ce, buf);
3130 else
3131 return SUBMODULE_PATCH_WITHOUT_INDEX;
3132 } else {
3133 if (read_old_data(st, name, buf))
3134 return error(_("read of %s failed"), name);
3137 return 0;
3141 * We are about to apply "patch"; populate the "image" with the
3142 * current version we have, from the working tree or from the index,
3143 * depending on the situation e.g. --cached/--index. If we are
3144 * applying a non-git patch that incrementally updates the tree,
3145 * we read from the result of a previous diff.
3147 static int load_preimage(struct image *image,
3148 struct patch *patch, struct stat *st, struct cache_entry *ce)
3150 struct strbuf buf = STRBUF_INIT;
3151 size_t len;
3152 char *img;
3153 struct patch *previous;
3154 int status;
3156 previous = previous_patch(patch, &status);
3157 if (status)
3158 return error(_("path %s has been renamed/deleted"),
3159 patch->old_name);
3160 if (previous) {
3161 /* We have a patched copy in memory; use that. */
3162 strbuf_add(&buf, previous->result, previous->resultsize);
3163 } else {
3164 status = load_patch_target(&buf, ce, st,
3165 patch->old_name, patch->old_mode);
3166 if (status < 0)
3167 return status;
3168 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3170 * There is no way to apply subproject
3171 * patch without looking at the index.
3172 * NEEDSWORK: shouldn't this be flagged
3173 * as an error???
3175 free_fragment_list(patch->fragments);
3176 patch->fragments = NULL;
3177 } else if (status) {
3178 return error(_("read of %s failed"), patch->old_name);
3182 img = strbuf_detach(&buf, &len);
3183 prepare_image(image, img, len, !patch->is_binary);
3184 return 0;
3187 static int three_way_merge(struct image *image,
3188 char *path,
3189 const unsigned char *base,
3190 const unsigned char *ours,
3191 const unsigned char *theirs)
3193 mmfile_t base_file, our_file, their_file;
3194 mmbuffer_t result = { NULL };
3195 int status;
3197 read_mmblob(&base_file, base);
3198 read_mmblob(&our_file, ours);
3199 read_mmblob(&their_file, theirs);
3200 status = ll_merge(&result, path,
3201 &base_file, "base",
3202 &our_file, "ours",
3203 &their_file, "theirs", NULL);
3204 free(base_file.ptr);
3205 free(our_file.ptr);
3206 free(their_file.ptr);
3207 if (status < 0 || !result.ptr) {
3208 free(result.ptr);
3209 return -1;
3211 clear_image(image);
3212 image->buf = result.ptr;
3213 image->len = result.size;
3215 return status;
3219 * When directly falling back to add/add three-way merge, we read from
3220 * the current contents of the new_name. In no cases other than that
3221 * this function will be called.
3223 static int load_current(struct image *image, struct patch *patch)
3225 struct strbuf buf = STRBUF_INIT;
3226 int status, pos;
3227 size_t len;
3228 char *img;
3229 struct stat st;
3230 struct cache_entry *ce;
3231 char *name = patch->new_name;
3232 unsigned mode = patch->new_mode;
3234 if (!patch->is_new)
3235 die("BUG: patch to %s is not a creation", patch->old_name);
3237 pos = cache_name_pos(name, strlen(name));
3238 if (pos < 0)
3239 return error(_("%s: does not exist in index"), name);
3240 ce = active_cache[pos];
3241 if (lstat(name, &st)) {
3242 if (errno != ENOENT)
3243 return error(_("%s: %s"), name, strerror(errno));
3244 if (checkout_target(ce, &st))
3245 return -1;
3247 if (verify_index_match(ce, &st))
3248 return error(_("%s: does not match index"), name);
3250 status = load_patch_target(&buf, ce, &st, name, mode);
3251 if (status < 0)
3252 return status;
3253 else if (status)
3254 return -1;
3255 img = strbuf_detach(&buf, &len);
3256 prepare_image(image, img, len, !patch->is_binary);
3257 return 0;
3260 static int try_threeway(struct image *image, struct patch *patch,
3261 struct stat *st, struct cache_entry *ce)
3263 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3264 struct strbuf buf = STRBUF_INIT;
3265 size_t len;
3266 int status;
3267 char *img;
3268 struct image tmp_image;
3270 /* No point falling back to 3-way merge in these cases */
3271 if (patch->is_delete ||
3272 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3273 return -1;
3275 /* Preimage the patch was prepared for */
3276 if (patch->is_new)
3277 write_sha1_file("", 0, blob_type, pre_sha1);
3278 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3279 read_blob_object(&buf, pre_sha1, patch->old_mode))
3280 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3282 fprintf(stderr, "Falling back to three-way merge...\n");
3284 img = strbuf_detach(&buf, &len);
3285 prepare_image(&tmp_image, img, len, 1);
3286 /* Apply the patch to get the post image */
3287 if (apply_fragments(&tmp_image, patch) < 0) {
3288 clear_image(&tmp_image);
3289 return -1;
3291 /* post_sha1[] is theirs */
3292 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3293 clear_image(&tmp_image);
3295 /* our_sha1[] is ours */
3296 if (patch->is_new) {
3297 if (load_current(&tmp_image, patch))
3298 return error("cannot read the current contents of '%s'",
3299 patch->new_name);
3300 } else {
3301 if (load_preimage(&tmp_image, patch, st, ce))
3302 return error("cannot read the current contents of '%s'",
3303 patch->old_name);
3305 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3306 clear_image(&tmp_image);
3308 /* in-core three-way merge between post and our using pre as base */
3309 status = three_way_merge(image, patch->new_name,
3310 pre_sha1, our_sha1, post_sha1);
3311 if (status < 0) {
3312 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3313 return status;
3316 if (status) {
3317 patch->conflicted_threeway = 1;
3318 if (patch->is_new)
3319 hashclr(patch->threeway_stage[0]);
3320 else
3321 hashcpy(patch->threeway_stage[0], pre_sha1);
3322 hashcpy(patch->threeway_stage[1], our_sha1);
3323 hashcpy(patch->threeway_stage[2], post_sha1);
3324 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3325 } else {
3326 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3328 return 0;
3331 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3333 struct image image;
3335 if (load_preimage(&image, patch, st, ce) < 0)
3336 return -1;
3338 if (patch->direct_to_threeway ||
3339 apply_fragments(&image, patch) < 0) {
3340 /* Note: with --reject, apply_fragments() returns 0 */
3341 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3342 return -1;
3344 patch->result = image.buf;
3345 patch->resultsize = image.len;
3346 add_to_fn_table(patch);
3347 free(image.line_allocated);
3349 if (0 < patch->is_delete && patch->resultsize)
3350 return error(_("removal patch leaves file contents"));
3352 return 0;
3356 * If "patch" that we are looking at modifies or deletes what we have,
3357 * we would want it not to lose any local modification we have, either
3358 * in the working tree or in the index.
3360 * This also decides if a non-git patch is a creation patch or a
3361 * modification to an existing empty file. We do not check the state
3362 * of the current tree for a creation patch in this function; the caller
3363 * check_patch() separately makes sure (and errors out otherwise) that
3364 * the path the patch creates does not exist in the current tree.
3366 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3368 const char *old_name = patch->old_name;
3369 struct patch *previous = NULL;
3370 int stat_ret = 0, status;
3371 unsigned st_mode = 0;
3373 if (!old_name)
3374 return 0;
3376 assert(patch->is_new <= 0);
3377 previous = previous_patch(patch, &status);
3379 if (status)
3380 return error(_("path %s has been renamed/deleted"), old_name);
3381 if (previous) {
3382 st_mode = previous->new_mode;
3383 } else if (!cached) {
3384 stat_ret = lstat(old_name, st);
3385 if (stat_ret && errno != ENOENT)
3386 return error(_("%s: %s"), old_name, strerror(errno));
3389 if (check_index && !previous) {
3390 int pos = cache_name_pos(old_name, strlen(old_name));
3391 if (pos < 0) {
3392 if (patch->is_new < 0)
3393 goto is_new;
3394 return error(_("%s: does not exist in index"), old_name);
3396 *ce = active_cache[pos];
3397 if (stat_ret < 0) {
3398 if (checkout_target(*ce, st))
3399 return -1;
3401 if (!cached && verify_index_match(*ce, st))
3402 return error(_("%s: does not match index"), old_name);
3403 if (cached)
3404 st_mode = (*ce)->ce_mode;
3405 } else if (stat_ret < 0) {
3406 if (patch->is_new < 0)
3407 goto is_new;
3408 return error(_("%s: %s"), old_name, strerror(errno));
3411 if (!cached && !previous)
3412 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3414 if (patch->is_new < 0)
3415 patch->is_new = 0;
3416 if (!patch->old_mode)
3417 patch->old_mode = st_mode;
3418 if ((st_mode ^ patch->old_mode) & S_IFMT)
3419 return error(_("%s: wrong type"), old_name);
3420 if (st_mode != patch->old_mode)
3421 warning(_("%s has type %o, expected %o"),
3422 old_name, st_mode, patch->old_mode);
3423 if (!patch->new_mode && !patch->is_delete)
3424 patch->new_mode = st_mode;
3425 return 0;
3427 is_new:
3428 patch->is_new = 1;
3429 patch->is_delete = 0;
3430 free(patch->old_name);
3431 patch->old_name = NULL;
3432 return 0;
3436 #define EXISTS_IN_INDEX 1
3437 #define EXISTS_IN_WORKTREE 2
3439 static int check_to_create(const char *new_name, int ok_if_exists)
3441 struct stat nst;
3443 if (check_index &&
3444 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3445 !ok_if_exists)
3446 return EXISTS_IN_INDEX;
3447 if (cached)
3448 return 0;
3450 if (!lstat(new_name, &nst)) {
3451 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3452 return 0;
3454 * A leading component of new_name might be a symlink
3455 * that is going to be removed with this patch, but
3456 * still pointing at somewhere that has the path.
3457 * In such a case, path "new_name" does not exist as
3458 * far as git is concerned.
3460 if (has_symlink_leading_path(new_name, strlen(new_name)))
3461 return 0;
3463 return EXISTS_IN_WORKTREE;
3464 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3465 return error("%s: %s", new_name, strerror(errno));
3467 return 0;
3471 * Check and apply the patch in-core; leave the result in patch->result
3472 * for the caller to write it out to the final destination.
3474 static int check_patch(struct patch *patch)
3476 struct stat st;
3477 const char *old_name = patch->old_name;
3478 const char *new_name = patch->new_name;
3479 const char *name = old_name ? old_name : new_name;
3480 struct cache_entry *ce = NULL;
3481 struct patch *tpatch;
3482 int ok_if_exists;
3483 int status;
3485 patch->rejected = 1; /* we will drop this after we succeed */
3487 status = check_preimage(patch, &ce, &st);
3488 if (status)
3489 return status;
3490 old_name = patch->old_name;
3493 * A type-change diff is always split into a patch to delete
3494 * old, immediately followed by a patch to create new (see
3495 * diff.c::run_diff()); in such a case it is Ok that the entry
3496 * to be deleted by the previous patch is still in the working
3497 * tree and in the index.
3499 * A patch to swap-rename between A and B would first rename A
3500 * to B and then rename B to A. While applying the first one,
3501 * the presense of B should not stop A from getting renamed to
3502 * B; ask to_be_deleted() about the later rename. Removal of
3503 * B and rename from A to B is handled the same way by asking
3504 * was_deleted().
3506 if ((tpatch = in_fn_table(new_name)) &&
3507 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3508 ok_if_exists = 1;
3509 else
3510 ok_if_exists = 0;
3512 if (new_name &&
3513 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3514 int err = check_to_create(new_name, ok_if_exists);
3516 if (err && threeway) {
3517 patch->direct_to_threeway = 1;
3518 } else switch (err) {
3519 case 0:
3520 break; /* happy */
3521 case EXISTS_IN_INDEX:
3522 return error(_("%s: already exists in index"), new_name);
3523 break;
3524 case EXISTS_IN_WORKTREE:
3525 return error(_("%s: already exists in working directory"),
3526 new_name);
3527 default:
3528 return err;
3531 if (!patch->new_mode) {
3532 if (0 < patch->is_new)
3533 patch->new_mode = S_IFREG | 0644;
3534 else
3535 patch->new_mode = patch->old_mode;
3539 if (new_name && old_name) {
3540 int same = !strcmp(old_name, new_name);
3541 if (!patch->new_mode)
3542 patch->new_mode = patch->old_mode;
3543 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3544 if (same)
3545 return error(_("new mode (%o) of %s does not "
3546 "match old mode (%o)"),
3547 patch->new_mode, new_name,
3548 patch->old_mode);
3549 else
3550 return error(_("new mode (%o) of %s does not "
3551 "match old mode (%o) of %s"),
3552 patch->new_mode, new_name,
3553 patch->old_mode, old_name);
3557 if (apply_data(patch, &st, ce) < 0)
3558 return error(_("%s: patch does not apply"), name);
3559 patch->rejected = 0;
3560 return 0;
3563 static int check_patch_list(struct patch *patch)
3565 int err = 0;
3567 prepare_fn_table(patch);
3568 while (patch) {
3569 if (apply_verbosely)
3570 say_patch_name(stderr,
3571 _("Checking patch %s..."), patch);
3572 err |= check_patch(patch);
3573 patch = patch->next;
3575 return err;
3578 /* This function tries to read the sha1 from the current index */
3579 static int get_current_sha1(const char *path, unsigned char *sha1)
3581 int pos;
3583 if (read_cache() < 0)
3584 return -1;
3585 pos = cache_name_pos(path, strlen(path));
3586 if (pos < 0)
3587 return -1;
3588 hashcpy(sha1, active_cache[pos]->sha1);
3589 return 0;
3592 /* Build an index that contains the just the files needed for a 3way merge */
3593 static void build_fake_ancestor(struct patch *list, const char *filename)
3595 struct patch *patch;
3596 struct index_state result = { NULL };
3597 int fd;
3599 /* Once we start supporting the reverse patch, it may be
3600 * worth showing the new sha1 prefix, but until then...
3602 for (patch = list; patch; patch = patch->next) {
3603 const unsigned char *sha1_ptr;
3604 unsigned char sha1[20];
3605 struct cache_entry *ce;
3606 const char *name;
3608 name = patch->old_name ? patch->old_name : patch->new_name;
3609 if (0 < patch->is_new)
3610 continue;
3611 else if (get_sha1_blob(patch->old_sha1_prefix, sha1))
3612 /* git diff has no index line for mode/type changes */
3613 if (!patch->lines_added && !patch->lines_deleted) {
3614 if (get_current_sha1(patch->old_name, sha1))
3615 die("mode change for %s, which is not "
3616 "in current HEAD", name);
3617 sha1_ptr = sha1;
3618 } else
3619 die("sha1 information is lacking or useless "
3620 "(%s).", name);
3621 else
3622 sha1_ptr = sha1;
3624 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3625 if (!ce)
3626 die(_("make_cache_entry failed for path '%s'"), name);
3627 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3628 die ("Could not add %s to temporary index", name);
3631 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3632 if (fd < 0 || write_index(&result, fd) || close(fd))
3633 die ("Could not write temporary index to %s", filename);
3635 discard_index(&result);
3638 static void stat_patch_list(struct patch *patch)
3640 int files, adds, dels;
3642 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3643 files++;
3644 adds += patch->lines_added;
3645 dels += patch->lines_deleted;
3646 show_stats(patch);
3649 print_stat_summary(stdout, files, adds, dels);
3652 static void numstat_patch_list(struct patch *patch)
3654 for ( ; patch; patch = patch->next) {
3655 const char *name;
3656 name = patch->new_name ? patch->new_name : patch->old_name;
3657 if (patch->is_binary)
3658 printf("-\t-\t");
3659 else
3660 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3661 write_name_quoted(name, stdout, line_termination);
3665 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3667 if (mode)
3668 printf(" %s mode %06o %s\n", newdelete, mode, name);
3669 else
3670 printf(" %s %s\n", newdelete, name);
3673 static void show_mode_change(struct patch *p, int show_name)
3675 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3676 if (show_name)
3677 printf(" mode change %06o => %06o %s\n",
3678 p->old_mode, p->new_mode, p->new_name);
3679 else
3680 printf(" mode change %06o => %06o\n",
3681 p->old_mode, p->new_mode);
3685 static void show_rename_copy(struct patch *p)
3687 const char *renamecopy = p->is_rename ? "rename" : "copy";
3688 const char *old, *new;
3690 /* Find common prefix */
3691 old = p->old_name;
3692 new = p->new_name;
3693 while (1) {
3694 const char *slash_old, *slash_new;
3695 slash_old = strchr(old, '/');
3696 slash_new = strchr(new, '/');
3697 if (!slash_old ||
3698 !slash_new ||
3699 slash_old - old != slash_new - new ||
3700 memcmp(old, new, slash_new - new))
3701 break;
3702 old = slash_old + 1;
3703 new = slash_new + 1;
3705 /* p->old_name thru old is the common prefix, and old and new
3706 * through the end of names are renames
3708 if (old != p->old_name)
3709 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3710 (int)(old - p->old_name), p->old_name,
3711 old, new, p->score);
3712 else
3713 printf(" %s %s => %s (%d%%)\n", renamecopy,
3714 p->old_name, p->new_name, p->score);
3715 show_mode_change(p, 0);
3718 static void summary_patch_list(struct patch *patch)
3720 struct patch *p;
3722 for (p = patch; p; p = p->next) {
3723 if (p->is_new)
3724 show_file_mode_name("create", p->new_mode, p->new_name);
3725 else if (p->is_delete)
3726 show_file_mode_name("delete", p->old_mode, p->old_name);
3727 else {
3728 if (p->is_rename || p->is_copy)
3729 show_rename_copy(p);
3730 else {
3731 if (p->score) {
3732 printf(" rewrite %s (%d%%)\n",
3733 p->new_name, p->score);
3734 show_mode_change(p, 0);
3736 else
3737 show_mode_change(p, 1);
3743 static void patch_stats(struct patch *patch)
3745 int lines = patch->lines_added + patch->lines_deleted;
3747 if (lines > max_change)
3748 max_change = lines;
3749 if (patch->old_name) {
3750 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3751 if (!len)
3752 len = strlen(patch->old_name);
3753 if (len > max_len)
3754 max_len = len;
3756 if (patch->new_name) {
3757 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3758 if (!len)
3759 len = strlen(patch->new_name);
3760 if (len > max_len)
3761 max_len = len;
3765 static void remove_file(struct patch *patch, int rmdir_empty)
3767 if (update_index) {
3768 if (remove_file_from_cache(patch->old_name) < 0)
3769 die(_("unable to remove %s from index"), patch->old_name);
3771 if (!cached) {
3772 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3773 remove_path(patch->old_name);
3778 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3780 struct stat st;
3781 struct cache_entry *ce;
3782 int namelen = strlen(path);
3783 unsigned ce_size = cache_entry_size(namelen);
3785 if (!update_index)
3786 return;
3788 ce = xcalloc(1, ce_size);
3789 memcpy(ce->name, path, namelen);
3790 ce->ce_mode = create_ce_mode(mode);
3791 ce->ce_flags = create_ce_flags(0);
3792 ce->ce_namelen = namelen;
3793 if (S_ISGITLINK(mode)) {
3794 const char *s = buf;
3796 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3797 die(_("corrupt patch for subproject %s"), path);
3798 } else {
3799 if (!cached) {
3800 if (lstat(path, &st) < 0)
3801 die_errno(_("unable to stat newly created file '%s'"),
3802 path);
3803 fill_stat_cache_info(ce, &st);
3805 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3806 die(_("unable to create backing store for newly created file %s"), path);
3808 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3809 die(_("unable to add cache entry for %s"), path);
3812 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3814 int fd;
3815 struct strbuf nbuf = STRBUF_INIT;
3817 if (S_ISGITLINK(mode)) {
3818 struct stat st;
3819 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3820 return 0;
3821 return mkdir(path, 0777);
3824 if (has_symlinks && S_ISLNK(mode))
3825 /* Although buf:size is counted string, it also is NUL
3826 * terminated.
3828 return symlink(buf, path);
3830 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3831 if (fd < 0)
3832 return -1;
3834 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3835 size = nbuf.len;
3836 buf = nbuf.buf;
3838 write_or_die(fd, buf, size);
3839 strbuf_release(&nbuf);
3841 if (close(fd) < 0)
3842 die_errno(_("closing file '%s'"), path);
3843 return 0;
3847 * We optimistically assume that the directories exist,
3848 * which is true 99% of the time anyway. If they don't,
3849 * we create them and try again.
3851 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3853 if (cached)
3854 return;
3855 if (!try_create_file(path, mode, buf, size))
3856 return;
3858 if (errno == ENOENT) {
3859 if (safe_create_leading_directories(path))
3860 return;
3861 if (!try_create_file(path, mode, buf, size))
3862 return;
3865 if (errno == EEXIST || errno == EACCES) {
3866 /* We may be trying to create a file where a directory
3867 * used to be.
3869 struct stat st;
3870 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3871 errno = EEXIST;
3874 if (errno == EEXIST) {
3875 unsigned int nr = getpid();
3877 for (;;) {
3878 char newpath[PATH_MAX];
3879 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3880 if (!try_create_file(newpath, mode, buf, size)) {
3881 if (!rename(newpath, path))
3882 return;
3883 unlink_or_warn(newpath);
3884 break;
3886 if (errno != EEXIST)
3887 break;
3888 ++nr;
3891 die_errno(_("unable to write file '%s' mode %o"), path, mode);
3894 static void add_conflicted_stages_file(struct patch *patch)
3896 int stage, namelen;
3897 unsigned ce_size, mode;
3898 struct cache_entry *ce;
3900 if (!update_index)
3901 return;
3902 namelen = strlen(patch->new_name);
3903 ce_size = cache_entry_size(namelen);
3904 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
3906 remove_file_from_cache(patch->new_name);
3907 for (stage = 1; stage < 4; stage++) {
3908 if (is_null_sha1(patch->threeway_stage[stage - 1]))
3909 continue;
3910 ce = xcalloc(1, ce_size);
3911 memcpy(ce->name, patch->new_name, namelen);
3912 ce->ce_mode = create_ce_mode(mode);
3913 ce->ce_flags = create_ce_flags(stage);
3914 ce->ce_namelen = namelen;
3915 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
3916 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3917 die(_("unable to add cache entry for %s"), patch->new_name);
3921 static void create_file(struct patch *patch)
3923 char *path = patch->new_name;
3924 unsigned mode = patch->new_mode;
3925 unsigned long size = patch->resultsize;
3926 char *buf = patch->result;
3928 if (!mode)
3929 mode = S_IFREG | 0644;
3930 create_one_file(path, mode, buf, size);
3932 if (patch->conflicted_threeway)
3933 add_conflicted_stages_file(patch);
3934 else
3935 add_index_file(path, mode, buf, size);
3938 /* phase zero is to remove, phase one is to create */
3939 static void write_out_one_result(struct patch *patch, int phase)
3941 if (patch->is_delete > 0) {
3942 if (phase == 0)
3943 remove_file(patch, 1);
3944 return;
3946 if (patch->is_new > 0 || patch->is_copy) {
3947 if (phase == 1)
3948 create_file(patch);
3949 return;
3952 * Rename or modification boils down to the same
3953 * thing: remove the old, write the new
3955 if (phase == 0)
3956 remove_file(patch, patch->is_rename);
3957 if (phase == 1)
3958 create_file(patch);
3961 static int write_out_one_reject(struct patch *patch)
3963 FILE *rej;
3964 char namebuf[PATH_MAX];
3965 struct fragment *frag;
3966 int cnt = 0;
3967 struct strbuf sb = STRBUF_INIT;
3969 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3970 if (!frag->rejected)
3971 continue;
3972 cnt++;
3975 if (!cnt) {
3976 if (apply_verbosely)
3977 say_patch_name(stderr,
3978 _("Applied patch %s cleanly."), patch);
3979 return 0;
3982 /* This should not happen, because a removal patch that leaves
3983 * contents are marked "rejected" at the patch level.
3985 if (!patch->new_name)
3986 die(_("internal error"));
3988 /* Say this even without --verbose */
3989 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3990 "Applying patch %%s with %d rejects...",
3991 cnt),
3992 cnt);
3993 say_patch_name(stderr, sb.buf, patch);
3994 strbuf_release(&sb);
3996 cnt = strlen(patch->new_name);
3997 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3998 cnt = ARRAY_SIZE(namebuf) - 5;
3999 warning(_("truncating .rej filename to %.*s.rej"),
4000 cnt - 1, patch->new_name);
4002 memcpy(namebuf, patch->new_name, cnt);
4003 memcpy(namebuf + cnt, ".rej", 5);
4005 rej = fopen(namebuf, "w");
4006 if (!rej)
4007 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4009 /* Normal git tools never deal with .rej, so do not pretend
4010 * this is a git patch by saying --git nor give extended
4011 * headers. While at it, maybe please "kompare" that wants
4012 * the trailing TAB and some garbage at the end of line ;-).
4014 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4015 patch->new_name, patch->new_name);
4016 for (cnt = 1, frag = patch->fragments;
4017 frag;
4018 cnt++, frag = frag->next) {
4019 if (!frag->rejected) {
4020 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4021 continue;
4023 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4024 fprintf(rej, "%.*s", frag->size, frag->patch);
4025 if (frag->patch[frag->size-1] != '\n')
4026 fputc('\n', rej);
4028 fclose(rej);
4029 return -1;
4032 static int write_out_results(struct patch *list)
4034 int phase;
4035 int errs = 0;
4036 struct patch *l;
4037 struct string_list cpath = STRING_LIST_INIT_DUP;
4039 for (phase = 0; phase < 2; phase++) {
4040 l = list;
4041 while (l) {
4042 if (l->rejected)
4043 errs = 1;
4044 else {
4045 write_out_one_result(l, phase);
4046 if (phase == 1) {
4047 if (write_out_one_reject(l))
4048 errs = 1;
4049 if (l->conflicted_threeway) {
4050 string_list_append(&cpath, l->new_name);
4051 errs = 1;
4055 l = l->next;
4059 if (cpath.nr) {
4060 struct string_list_item *item;
4062 sort_string_list(&cpath);
4063 for_each_string_list_item(item, &cpath)
4064 fprintf(stderr, "U %s\n", item->string);
4065 string_list_clear(&cpath, 0);
4067 rerere(0);
4070 return errs;
4073 static struct lock_file lock_file;
4075 static struct string_list limit_by_name;
4076 static int has_include;
4077 static void add_name_limit(const char *name, int exclude)
4079 struct string_list_item *it;
4081 it = string_list_append(&limit_by_name, name);
4082 it->util = exclude ? NULL : (void *) 1;
4085 static int use_patch(struct patch *p)
4087 const char *pathname = p->new_name ? p->new_name : p->old_name;
4088 int i;
4090 /* Paths outside are not touched regardless of "--include" */
4091 if (0 < prefix_length) {
4092 int pathlen = strlen(pathname);
4093 if (pathlen <= prefix_length ||
4094 memcmp(prefix, pathname, prefix_length))
4095 return 0;
4098 /* See if it matches any of exclude/include rule */
4099 for (i = 0; i < limit_by_name.nr; i++) {
4100 struct string_list_item *it = &limit_by_name.items[i];
4101 if (!fnmatch(it->string, pathname, 0))
4102 return (it->util != NULL);
4106 * If we had any include, a path that does not match any rule is
4107 * not used. Otherwise, we saw bunch of exclude rules (or none)
4108 * and such a path is used.
4110 return !has_include;
4114 static void prefix_one(char **name)
4116 char *old_name = *name;
4117 if (!old_name)
4118 return;
4119 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
4120 free(old_name);
4123 static void prefix_patches(struct patch *p)
4125 if (!prefix || p->is_toplevel_relative)
4126 return;
4127 for ( ; p; p = p->next) {
4128 prefix_one(&p->new_name);
4129 prefix_one(&p->old_name);
4133 #define INACCURATE_EOF (1<<0)
4134 #define RECOUNT (1<<1)
4136 static int apply_patch(int fd, const char *filename, int options)
4138 size_t offset;
4139 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4140 struct patch *list = NULL, **listp = &list;
4141 int skipped_patch = 0;
4143 patch_input_file = filename;
4144 read_patch_file(&buf, fd);
4145 offset = 0;
4146 while (offset < buf.len) {
4147 struct patch *patch;
4148 int nr;
4150 patch = xcalloc(1, sizeof(*patch));
4151 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4152 patch->recount = !!(options & RECOUNT);
4153 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4154 if (nr < 0)
4155 break;
4156 if (apply_in_reverse)
4157 reverse_patches(patch);
4158 if (prefix)
4159 prefix_patches(patch);
4160 if (use_patch(patch)) {
4161 patch_stats(patch);
4162 *listp = patch;
4163 listp = &patch->next;
4165 else {
4166 free_patch(patch);
4167 skipped_patch++;
4169 offset += nr;
4172 if (!list && !skipped_patch)
4173 die(_("unrecognized input"));
4175 if (whitespace_error && (ws_error_action == die_on_ws_error))
4176 apply = 0;
4178 update_index = check_index && apply;
4179 if (update_index && newfd < 0)
4180 newfd = hold_locked_index(&lock_file, 1);
4182 if (check_index) {
4183 if (read_cache() < 0)
4184 die(_("unable to read index file"));
4187 if ((check || apply) &&
4188 check_patch_list(list) < 0 &&
4189 !apply_with_reject)
4190 exit(1);
4192 if (apply && write_out_results(list)) {
4193 if (apply_with_reject)
4194 exit(1);
4195 /* with --3way, we still need to write the index out */
4196 return 1;
4199 if (fake_ancestor)
4200 build_fake_ancestor(list, fake_ancestor);
4202 if (diffstat)
4203 stat_patch_list(list);
4205 if (numstat)
4206 numstat_patch_list(list);
4208 if (summary)
4209 summary_patch_list(list);
4211 free_patch_list(list);
4212 strbuf_release(&buf);
4213 string_list_clear(&fn_table, 0);
4214 return 0;
4217 static int git_apply_config(const char *var, const char *value, void *cb)
4219 if (!strcmp(var, "apply.whitespace"))
4220 return git_config_string(&apply_default_whitespace, var, value);
4221 else if (!strcmp(var, "apply.ignorewhitespace"))
4222 return git_config_string(&apply_default_ignorewhitespace, var, value);
4223 return git_default_config(var, value, cb);
4226 static int option_parse_exclude(const struct option *opt,
4227 const char *arg, int unset)
4229 add_name_limit(arg, 1);
4230 return 0;
4233 static int option_parse_include(const struct option *opt,
4234 const char *arg, int unset)
4236 add_name_limit(arg, 0);
4237 has_include = 1;
4238 return 0;
4241 static int option_parse_p(const struct option *opt,
4242 const char *arg, int unset)
4244 p_value = atoi(arg);
4245 p_value_known = 1;
4246 return 0;
4249 static int option_parse_z(const struct option *opt,
4250 const char *arg, int unset)
4252 if (unset)
4253 line_termination = '\n';
4254 else
4255 line_termination = 0;
4256 return 0;
4259 static int option_parse_space_change(const struct option *opt,
4260 const char *arg, int unset)
4262 if (unset)
4263 ws_ignore_action = ignore_ws_none;
4264 else
4265 ws_ignore_action = ignore_ws_change;
4266 return 0;
4269 static int option_parse_whitespace(const struct option *opt,
4270 const char *arg, int unset)
4272 const char **whitespace_option = opt->value;
4274 *whitespace_option = arg;
4275 parse_whitespace_option(arg);
4276 return 0;
4279 static int option_parse_directory(const struct option *opt,
4280 const char *arg, int unset)
4282 root_len = strlen(arg);
4283 if (root_len && arg[root_len - 1] != '/') {
4284 char *new_root;
4285 root = new_root = xmalloc(root_len + 2);
4286 strcpy(new_root, arg);
4287 strcpy(new_root + root_len++, "/");
4288 } else
4289 root = arg;
4290 return 0;
4293 int cmd_apply(int argc, const char **argv, const char *prefix_)
4295 int i;
4296 int errs = 0;
4297 int is_not_gitdir = !startup_info->have_repository;
4298 int force_apply = 0;
4300 const char *whitespace_option = NULL;
4302 struct option builtin_apply_options[] = {
4303 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4304 N_("don't apply changes matching the given path"),
4305 0, option_parse_exclude },
4306 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4307 N_("apply changes matching the given path"),
4308 0, option_parse_include },
4309 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4310 N_("remove <num> leading slashes from traditional diff paths"),
4311 0, option_parse_p },
4312 OPT_BOOLEAN(0, "no-add", &no_add,
4313 N_("ignore additions made by the patch")),
4314 OPT_BOOLEAN(0, "stat", &diffstat,
4315 N_("instead of applying the patch, output diffstat for the input")),
4316 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4317 OPT_NOOP_NOARG(0, "binary"),
4318 OPT_BOOLEAN(0, "numstat", &numstat,
4319 N_("shows number of added and deleted lines in decimal notation")),
4320 OPT_BOOLEAN(0, "summary", &summary,
4321 N_("instead of applying the patch, output a summary for the input")),
4322 OPT_BOOLEAN(0, "check", &check,
4323 N_("instead of applying the patch, see if the patch is applicable")),
4324 OPT_BOOLEAN(0, "index", &check_index,
4325 N_("make sure the patch is applicable to the current index")),
4326 OPT_BOOLEAN(0, "cached", &cached,
4327 N_("apply a patch without touching the working tree")),
4328 OPT_BOOLEAN(0, "apply", &force_apply,
4329 N_("also apply the patch (use with --stat/--summary/--check)")),
4330 OPT_BOOL('3', "3way", &threeway,
4331 N_( "attempt three-way merge if a patch does not apply")),
4332 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4333 N_("build a temporary index based on embedded index information")),
4334 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4335 N_("paths are separated with NUL character"),
4336 PARSE_OPT_NOARG, option_parse_z },
4337 OPT_INTEGER('C', NULL, &p_context,
4338 N_("ensure at least <n> lines of context match")),
4339 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4340 N_("detect new or modified lines that have whitespace errors"),
4341 0, option_parse_whitespace },
4342 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4343 N_("ignore changes in whitespace when finding context"),
4344 PARSE_OPT_NOARG, option_parse_space_change },
4345 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4346 N_("ignore changes in whitespace when finding context"),
4347 PARSE_OPT_NOARG, option_parse_space_change },
4348 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4349 N_("apply the patch in reverse")),
4350 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4351 N_("don't expect at least one line of context")),
4352 OPT_BOOLEAN(0, "reject", &apply_with_reject,
4353 N_("leave the rejected hunks in corresponding *.rej files")),
4354 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4355 N_("allow overlapping hunks")),
4356 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4357 OPT_BIT(0, "inaccurate-eof", &options,
4358 N_("tolerate incorrectly detected missing new-line at the end of file"),
4359 INACCURATE_EOF),
4360 OPT_BIT(0, "recount", &options,
4361 N_("do not trust the line counts in the hunk headers"),
4362 RECOUNT),
4363 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4364 N_("prepend <root> to all filenames"),
4365 0, option_parse_directory },
4366 OPT_END()
4369 prefix = prefix_;
4370 prefix_length = prefix ? strlen(prefix) : 0;
4371 git_config(git_apply_config, NULL);
4372 if (apply_default_whitespace)
4373 parse_whitespace_option(apply_default_whitespace);
4374 if (apply_default_ignorewhitespace)
4375 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4377 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4378 apply_usage, 0);
4380 if (apply_with_reject && threeway)
4381 die("--reject and --3way cannot be used together.");
4382 if (cached && threeway)
4383 die("--cached and --3way cannot be used together.");
4384 if (threeway) {
4385 if (is_not_gitdir)
4386 die(_("--3way outside a repository"));
4387 check_index = 1;
4389 if (apply_with_reject)
4390 apply = apply_verbosely = 1;
4391 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4392 apply = 0;
4393 if (check_index && is_not_gitdir)
4394 die(_("--index outside a repository"));
4395 if (cached) {
4396 if (is_not_gitdir)
4397 die(_("--cached outside a repository"));
4398 check_index = 1;
4400 for (i = 0; i < argc; i++) {
4401 const char *arg = argv[i];
4402 int fd;
4404 if (!strcmp(arg, "-")) {
4405 errs |= apply_patch(0, "<stdin>", options);
4406 read_stdin = 0;
4407 continue;
4408 } else if (0 < prefix_length)
4409 arg = prefix_filename(prefix, prefix_length, arg);
4411 fd = open(arg, O_RDONLY);
4412 if (fd < 0)
4413 die_errno(_("can't open patch '%s'"), arg);
4414 read_stdin = 0;
4415 set_default_whitespace_mode(whitespace_option);
4416 errs |= apply_patch(fd, arg, options);
4417 close(fd);
4419 set_default_whitespace_mode(whitespace_option);
4420 if (read_stdin)
4421 errs |= apply_patch(0, "<stdin>", options);
4422 if (whitespace_error) {
4423 if (squelch_whitespace_errors &&
4424 squelch_whitespace_errors < whitespace_error) {
4425 int squelched =
4426 whitespace_error - squelch_whitespace_errors;
4427 warning(Q_("squelched %d whitespace error",
4428 "squelched %d whitespace errors",
4429 squelched),
4430 squelched);
4432 if (ws_error_action == die_on_ws_error)
4433 die(Q_("%d line adds whitespace errors.",
4434 "%d lines add whitespace errors.",
4435 whitespace_error),
4436 whitespace_error);
4437 if (applied_after_fixing_ws && apply)
4438 warning("%d line%s applied after"
4439 " fixing whitespace errors.",
4440 applied_after_fixing_ws,
4441 applied_after_fixing_ws == 1 ? "" : "s");
4442 else if (whitespace_error)
4443 warning(Q_("%d line adds whitespace errors.",
4444 "%d lines add whitespace errors.",
4445 whitespace_error),
4446 whitespace_error);
4449 if (update_index) {
4450 if (write_cache(newfd, active_cache, active_nr) ||
4451 commit_locked_index(&lock_file))
4452 die(_("Unable to write new index file"));
4455 return !!errs;