cvsserver: Determinize output to combat Perl 5.18 hash randomization
[alt-git.git] / builtin / apply.c
blob6c11e8bc73892b15f91fcb65d65e18ee05956ec8
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "diff.h"
18 #include "parse-options.h"
19 #include "xdiff-interface.h"
20 #include "ll-merge.h"
21 #include "rerere.h"
24 * --check turns on checking that the working tree matches the
25 * files that are being modified, but doesn't apply the patch
26 * --stat does just a diffstat, and doesn't actually apply
27 * --numstat does numeric diffstat, and doesn't actually apply
28 * --index-info shows the old and new index info for paths if available.
29 * --index updates the cache as well.
30 * --cached updates only the cache without ever touching the working tree.
32 static const char *prefix;
33 static int prefix_length = -1;
34 static int newfd = -1;
36 static int unidiff_zero;
37 static int p_value = 1;
38 static int p_value_known;
39 static int check_index;
40 static int update_index;
41 static int cached;
42 static int diffstat;
43 static int numstat;
44 static int summary;
45 static int check;
46 static int apply = 1;
47 static int apply_in_reverse;
48 static int apply_with_reject;
49 static int apply_verbosely;
50 static int allow_overlap;
51 static int no_add;
52 static int threeway;
53 static const char *fake_ancestor;
54 static int line_termination = '\n';
55 static unsigned int p_context = UINT_MAX;
56 static const char * const apply_usage[] = {
57 N_("git apply [options] [<patch>...]"),
58 NULL
61 static enum ws_error_action {
62 nowarn_ws_error,
63 warn_on_ws_error,
64 die_on_ws_error,
65 correct_ws_error
66 } ws_error_action = warn_on_ws_error;
67 static int whitespace_error;
68 static int squelch_whitespace_errors = 5;
69 static int applied_after_fixing_ws;
71 static enum ws_ignore {
72 ignore_ws_none,
73 ignore_ws_change
74 } ws_ignore_action = ignore_ws_none;
77 static const char *patch_input_file;
78 static const char *root;
79 static int root_len;
80 static int read_stdin = 1;
81 static int options;
83 static void parse_whitespace_option(const char *option)
85 if (!option) {
86 ws_error_action = warn_on_ws_error;
87 return;
89 if (!strcmp(option, "warn")) {
90 ws_error_action = warn_on_ws_error;
91 return;
93 if (!strcmp(option, "nowarn")) {
94 ws_error_action = nowarn_ws_error;
95 return;
97 if (!strcmp(option, "error")) {
98 ws_error_action = die_on_ws_error;
99 return;
101 if (!strcmp(option, "error-all")) {
102 ws_error_action = die_on_ws_error;
103 squelch_whitespace_errors = 0;
104 return;
106 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
107 ws_error_action = correct_ws_error;
108 return;
110 die(_("unrecognized whitespace option '%s'"), option);
113 static void parse_ignorewhitespace_option(const char *option)
115 if (!option || !strcmp(option, "no") ||
116 !strcmp(option, "false") || !strcmp(option, "never") ||
117 !strcmp(option, "none")) {
118 ws_ignore_action = ignore_ws_none;
119 return;
121 if (!strcmp(option, "change")) {
122 ws_ignore_action = ignore_ws_change;
123 return;
125 die(_("unrecognized whitespace ignore option '%s'"), option);
128 static void set_default_whitespace_mode(const char *whitespace_option)
130 if (!whitespace_option && !apply_default_whitespace)
131 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
135 * For "diff-stat" like behaviour, we keep track of the biggest change
136 * we've seen, and the longest filename. That allows us to do simple
137 * scaling.
139 static int max_change, max_len;
142 * Various "current state", notably line numbers and what
143 * file (and how) we're patching right now.. The "is_xxxx"
144 * things are flags, where -1 means "don't know yet".
146 static int linenr = 1;
149 * This represents one "hunk" from a patch, starting with
150 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
151 * patch text is pointed at by patch, and its byte length
152 * is stored in size. leading and trailing are the number
153 * of context lines.
155 struct fragment {
156 unsigned long leading, trailing;
157 unsigned long oldpos, oldlines;
158 unsigned long newpos, newlines;
160 * 'patch' is usually borrowed from buf in apply_patch(),
161 * but some codepaths store an allocated buffer.
163 const char *patch;
164 unsigned free_patch:1,
165 rejected:1;
166 int size;
167 int linenr;
168 struct fragment *next;
172 * When dealing with a binary patch, we reuse "leading" field
173 * to store the type of the binary hunk, either deflated "delta"
174 * or deflated "literal".
176 #define binary_patch_method leading
177 #define BINARY_DELTA_DEFLATED 1
178 #define BINARY_LITERAL_DEFLATED 2
181 * This represents a "patch" to a file, both metainfo changes
182 * such as creation/deletion, filemode and content changes represented
183 * as a series of fragments.
185 struct patch {
186 char *new_name, *old_name, *def_name;
187 unsigned int old_mode, new_mode;
188 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
189 int rejected;
190 unsigned ws_rule;
191 int lines_added, lines_deleted;
192 int score;
193 unsigned int is_toplevel_relative:1;
194 unsigned int inaccurate_eof:1;
195 unsigned int is_binary:1;
196 unsigned int is_copy:1;
197 unsigned int is_rename:1;
198 unsigned int recount:1;
199 unsigned int conflicted_threeway:1;
200 unsigned int direct_to_threeway:1;
201 struct fragment *fragments;
202 char *result;
203 size_t resultsize;
204 char old_sha1_prefix[41];
205 char new_sha1_prefix[41];
206 struct patch *next;
208 /* three-way fallback result */
209 unsigned char threeway_stage[3][20];
212 static void free_fragment_list(struct fragment *list)
214 while (list) {
215 struct fragment *next = list->next;
216 if (list->free_patch)
217 free((char *)list->patch);
218 free(list);
219 list = next;
223 static void free_patch(struct patch *patch)
225 free_fragment_list(patch->fragments);
226 free(patch->def_name);
227 free(patch->old_name);
228 free(patch->new_name);
229 free(patch->result);
230 free(patch);
233 static void free_patch_list(struct patch *list)
235 while (list) {
236 struct patch *next = list->next;
237 free_patch(list);
238 list = next;
243 * A line in a file, len-bytes long (includes the terminating LF,
244 * except for an incomplete line at the end if the file ends with
245 * one), and its contents hashes to 'hash'.
247 struct line {
248 size_t len;
249 unsigned hash : 24;
250 unsigned flag : 8;
251 #define LINE_COMMON 1
252 #define LINE_PATCHED 2
256 * This represents a "file", which is an array of "lines".
258 struct image {
259 char *buf;
260 size_t len;
261 size_t nr;
262 size_t alloc;
263 struct line *line_allocated;
264 struct line *line;
268 * Records filenames that have been touched, in order to handle
269 * the case where more than one patches touch the same file.
272 static struct string_list fn_table;
274 static uint32_t hash_line(const char *cp, size_t len)
276 size_t i;
277 uint32_t h;
278 for (i = 0, h = 0; i < len; i++) {
279 if (!isspace(cp[i])) {
280 h = h * 3 + (cp[i] & 0xff);
283 return h;
287 * Compare lines s1 of length n1 and s2 of length n2, ignoring
288 * whitespace difference. Returns 1 if they match, 0 otherwise
290 static int fuzzy_matchlines(const char *s1, size_t n1,
291 const char *s2, size_t n2)
293 const char *last1 = s1 + n1 - 1;
294 const char *last2 = s2 + n2 - 1;
295 int result = 0;
297 /* ignore line endings */
298 while ((*last1 == '\r') || (*last1 == '\n'))
299 last1--;
300 while ((*last2 == '\r') || (*last2 == '\n'))
301 last2--;
303 /* skip leading whitespace */
304 while (isspace(*s1) && (s1 <= last1))
305 s1++;
306 while (isspace(*s2) && (s2 <= last2))
307 s2++;
308 /* early return if both lines are empty */
309 if ((s1 > last1) && (s2 > last2))
310 return 1;
311 while (!result) {
312 result = *s1++ - *s2++;
314 * Skip whitespace inside. We check for whitespace on
315 * both buffers because we don't want "a b" to match
316 * "ab"
318 if (isspace(*s1) && isspace(*s2)) {
319 while (isspace(*s1) && s1 <= last1)
320 s1++;
321 while (isspace(*s2) && s2 <= last2)
322 s2++;
325 * If we reached the end on one side only,
326 * lines don't match
328 if (
329 ((s2 > last2) && (s1 <= last1)) ||
330 ((s1 > last1) && (s2 <= last2)))
331 return 0;
332 if ((s1 > last1) && (s2 > last2))
333 break;
336 return !result;
339 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
341 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
342 img->line_allocated[img->nr].len = len;
343 img->line_allocated[img->nr].hash = hash_line(bol, len);
344 img->line_allocated[img->nr].flag = flag;
345 img->nr++;
349 * "buf" has the file contents to be patched (read from various sources).
350 * attach it to "image" and add line-based index to it.
351 * "image" now owns the "buf".
353 static void prepare_image(struct image *image, char *buf, size_t len,
354 int prepare_linetable)
356 const char *cp, *ep;
358 memset(image, 0, sizeof(*image));
359 image->buf = buf;
360 image->len = len;
362 if (!prepare_linetable)
363 return;
365 ep = image->buf + image->len;
366 cp = image->buf;
367 while (cp < ep) {
368 const char *next;
369 for (next = cp; next < ep && *next != '\n'; next++)
371 if (next < ep)
372 next++;
373 add_line_info(image, cp, next - cp, 0);
374 cp = next;
376 image->line = image->line_allocated;
379 static void clear_image(struct image *image)
381 free(image->buf);
382 free(image->line_allocated);
383 memset(image, 0, sizeof(*image));
386 /* fmt must contain _one_ %s and no other substitution */
387 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
389 struct strbuf sb = STRBUF_INIT;
391 if (patch->old_name && patch->new_name &&
392 strcmp(patch->old_name, patch->new_name)) {
393 quote_c_style(patch->old_name, &sb, NULL, 0);
394 strbuf_addstr(&sb, " => ");
395 quote_c_style(patch->new_name, &sb, NULL, 0);
396 } else {
397 const char *n = patch->new_name;
398 if (!n)
399 n = patch->old_name;
400 quote_c_style(n, &sb, NULL, 0);
402 fprintf(output, fmt, sb.buf);
403 fputc('\n', output);
404 strbuf_release(&sb);
407 #define SLOP (16)
409 static void read_patch_file(struct strbuf *sb, int fd)
411 if (strbuf_read(sb, fd, 0) < 0)
412 die_errno("git apply: failed to read");
415 * Make sure that we have some slop in the buffer
416 * so that we can do speculative "memcmp" etc, and
417 * see to it that it is NUL-filled.
419 strbuf_grow(sb, SLOP);
420 memset(sb->buf + sb->len, 0, SLOP);
423 static unsigned long linelen(const char *buffer, unsigned long size)
425 unsigned long len = 0;
426 while (size--) {
427 len++;
428 if (*buffer++ == '\n')
429 break;
431 return len;
434 static int is_dev_null(const char *str)
436 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
439 #define TERM_SPACE 1
440 #define TERM_TAB 2
442 static int name_terminate(const char *name, int namelen, int c, int terminate)
444 if (c == ' ' && !(terminate & TERM_SPACE))
445 return 0;
446 if (c == '\t' && !(terminate & TERM_TAB))
447 return 0;
449 return 1;
452 /* remove double slashes to make --index work with such filenames */
453 static char *squash_slash(char *name)
455 int i = 0, j = 0;
457 if (!name)
458 return NULL;
460 while (name[i]) {
461 if ((name[j++] = name[i++]) == '/')
462 while (name[i] == '/')
463 i++;
465 name[j] = '\0';
466 return name;
469 static char *find_name_gnu(const char *line, const char *def, int p_value)
471 struct strbuf name = STRBUF_INIT;
472 char *cp;
475 * Proposed "new-style" GNU patch/diff format; see
476 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
478 if (unquote_c_style(&name, line, NULL)) {
479 strbuf_release(&name);
480 return NULL;
483 for (cp = name.buf; p_value; p_value--) {
484 cp = strchr(cp, '/');
485 if (!cp) {
486 strbuf_release(&name);
487 return NULL;
489 cp++;
492 strbuf_remove(&name, 0, cp - name.buf);
493 if (root)
494 strbuf_insert(&name, 0, root, root_len);
495 return squash_slash(strbuf_detach(&name, NULL));
498 static size_t sane_tz_len(const char *line, size_t len)
500 const char *tz, *p;
502 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
503 return 0;
504 tz = line + len - strlen(" +0500");
506 if (tz[1] != '+' && tz[1] != '-')
507 return 0;
509 for (p = tz + 2; p != line + len; p++)
510 if (!isdigit(*p))
511 return 0;
513 return line + len - tz;
516 static size_t tz_with_colon_len(const char *line, size_t len)
518 const char *tz, *p;
520 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
521 return 0;
522 tz = line + len - strlen(" +08:00");
524 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
525 return 0;
526 p = tz + 2;
527 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
528 !isdigit(*p++) || !isdigit(*p++))
529 return 0;
531 return line + len - tz;
534 static size_t date_len(const char *line, size_t len)
536 const char *date, *p;
538 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
539 return 0;
540 p = date = line + len - strlen("72-02-05");
542 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
543 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
544 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
545 return 0;
547 if (date - line >= strlen("19") &&
548 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
549 date -= strlen("19");
551 return line + len - date;
554 static size_t short_time_len(const char *line, size_t len)
556 const char *time, *p;
558 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
559 return 0;
560 p = time = line + len - strlen(" 07:01:32");
562 /* Permit 1-digit hours? */
563 if (*p++ != ' ' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
565 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
566 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
567 return 0;
569 return line + len - time;
572 static size_t fractional_time_len(const char *line, size_t len)
574 const char *p;
575 size_t n;
577 /* Expected format: 19:41:17.620000023 */
578 if (!len || !isdigit(line[len - 1]))
579 return 0;
580 p = line + len - 1;
582 /* Fractional seconds. */
583 while (p > line && isdigit(*p))
584 p--;
585 if (*p != '.')
586 return 0;
588 /* Hours, minutes, and whole seconds. */
589 n = short_time_len(line, p - line);
590 if (!n)
591 return 0;
593 return line + len - p + n;
596 static size_t trailing_spaces_len(const char *line, size_t len)
598 const char *p;
600 /* Expected format: ' ' x (1 or more) */
601 if (!len || line[len - 1] != ' ')
602 return 0;
604 p = line + len;
605 while (p != line) {
606 p--;
607 if (*p != ' ')
608 return line + len - (p + 1);
611 /* All spaces! */
612 return len;
615 static size_t diff_timestamp_len(const char *line, size_t len)
617 const char *end = line + len;
618 size_t n;
621 * Posix: 2010-07-05 19:41:17
622 * GNU: 2010-07-05 19:41:17.620000023 -0500
625 if (!isdigit(end[-1]))
626 return 0;
628 n = sane_tz_len(line, end - line);
629 if (!n)
630 n = tz_with_colon_len(line, end - line);
631 end -= n;
633 n = short_time_len(line, end - line);
634 if (!n)
635 n = fractional_time_len(line, end - line);
636 end -= n;
638 n = date_len(line, end - line);
639 if (!n) /* No date. Too bad. */
640 return 0;
641 end -= n;
643 if (end == line) /* No space before date. */
644 return 0;
645 if (end[-1] == '\t') { /* Success! */
646 end--;
647 return line + len - end;
649 if (end[-1] != ' ') /* No space before date. */
650 return 0;
652 /* Whitespace damage. */
653 end -= trailing_spaces_len(line, end - line);
654 return line + len - end;
657 static char *null_strdup(const char *s)
659 return s ? xstrdup(s) : NULL;
662 static char *find_name_common(const char *line, const char *def,
663 int p_value, const char *end, int terminate)
665 int len;
666 const char *start = NULL;
668 if (p_value == 0)
669 start = line;
670 while (line != end) {
671 char c = *line;
673 if (!end && isspace(c)) {
674 if (c == '\n')
675 break;
676 if (name_terminate(start, line-start, c, terminate))
677 break;
679 line++;
680 if (c == '/' && !--p_value)
681 start = line;
683 if (!start)
684 return squash_slash(null_strdup(def));
685 len = line - start;
686 if (!len)
687 return squash_slash(null_strdup(def));
690 * Generally we prefer the shorter name, especially
691 * if the other one is just a variation of that with
692 * something else tacked on to the end (ie "file.orig"
693 * or "file~").
695 if (def) {
696 int deflen = strlen(def);
697 if (deflen < len && !strncmp(start, def, deflen))
698 return squash_slash(xstrdup(def));
701 if (root) {
702 char *ret = xmalloc(root_len + len + 1);
703 strcpy(ret, root);
704 memcpy(ret + root_len, start, len);
705 ret[root_len + len] = '\0';
706 return squash_slash(ret);
709 return squash_slash(xmemdupz(start, len));
712 static char *find_name(const char *line, char *def, int p_value, int terminate)
714 if (*line == '"') {
715 char *name = find_name_gnu(line, def, p_value);
716 if (name)
717 return name;
720 return find_name_common(line, def, p_value, NULL, terminate);
723 static char *find_name_traditional(const char *line, char *def, int p_value)
725 size_t len = strlen(line);
726 size_t date_len;
728 if (*line == '"') {
729 char *name = find_name_gnu(line, def, p_value);
730 if (name)
731 return name;
734 len = strchrnul(line, '\n') - line;
735 date_len = diff_timestamp_len(line, len);
736 if (!date_len)
737 return find_name_common(line, def, p_value, NULL, TERM_TAB);
738 len -= date_len;
740 return find_name_common(line, def, p_value, line + len, 0);
743 static int count_slashes(const char *cp)
745 int cnt = 0;
746 char ch;
748 while ((ch = *cp++))
749 if (ch == '/')
750 cnt++;
751 return cnt;
755 * Given the string after "--- " or "+++ ", guess the appropriate
756 * p_value for the given patch.
758 static int guess_p_value(const char *nameline)
760 char *name, *cp;
761 int val = -1;
763 if (is_dev_null(nameline))
764 return -1;
765 name = find_name_traditional(nameline, NULL, 0);
766 if (!name)
767 return -1;
768 cp = strchr(name, '/');
769 if (!cp)
770 val = 0;
771 else if (prefix) {
773 * Does it begin with "a/$our-prefix" and such? Then this is
774 * very likely to apply to our directory.
776 if (!strncmp(name, prefix, prefix_length))
777 val = count_slashes(prefix);
778 else {
779 cp++;
780 if (!strncmp(cp, prefix, prefix_length))
781 val = count_slashes(prefix) + 1;
784 free(name);
785 return val;
789 * Does the ---/+++ line has the POSIX timestamp after the last HT?
790 * GNU diff puts epoch there to signal a creation/deletion event. Is
791 * this such a timestamp?
793 static int has_epoch_timestamp(const char *nameline)
796 * We are only interested in epoch timestamp; any non-zero
797 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
798 * For the same reason, the date must be either 1969-12-31 or
799 * 1970-01-01, and the seconds part must be "00".
801 const char stamp_regexp[] =
802 "^(1969-12-31|1970-01-01)"
804 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
806 "([-+][0-2][0-9]:?[0-5][0-9])\n";
807 const char *timestamp = NULL, *cp, *colon;
808 static regex_t *stamp;
809 regmatch_t m[10];
810 int zoneoffset;
811 int hourminute;
812 int status;
814 for (cp = nameline; *cp != '\n'; cp++) {
815 if (*cp == '\t')
816 timestamp = cp + 1;
818 if (!timestamp)
819 return 0;
820 if (!stamp) {
821 stamp = xmalloc(sizeof(*stamp));
822 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
823 warning(_("Cannot prepare timestamp regexp %s"),
824 stamp_regexp);
825 return 0;
829 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
830 if (status) {
831 if (status != REG_NOMATCH)
832 warning(_("regexec returned %d for input: %s"),
833 status, timestamp);
834 return 0;
837 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
838 if (*colon == ':')
839 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
840 else
841 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
842 if (timestamp[m[3].rm_so] == '-')
843 zoneoffset = -zoneoffset;
846 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
847 * (west of GMT) or 1970-01-01 (east of GMT)
849 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
850 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
851 return 0;
853 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
854 strtol(timestamp + 14, NULL, 10) -
855 zoneoffset);
857 return ((zoneoffset < 0 && hourminute == 1440) ||
858 (0 <= zoneoffset && !hourminute));
862 * Get the name etc info from the ---/+++ lines of a traditional patch header
864 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
865 * files, we can happily check the index for a match, but for creating a
866 * new file we should try to match whatever "patch" does. I have no idea.
868 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
870 char *name;
872 first += 4; /* skip "--- " */
873 second += 4; /* skip "+++ " */
874 if (!p_value_known) {
875 int p, q;
876 p = guess_p_value(first);
877 q = guess_p_value(second);
878 if (p < 0) p = q;
879 if (0 <= p && p == q) {
880 p_value = p;
881 p_value_known = 1;
884 if (is_dev_null(first)) {
885 patch->is_new = 1;
886 patch->is_delete = 0;
887 name = find_name_traditional(second, NULL, p_value);
888 patch->new_name = name;
889 } else if (is_dev_null(second)) {
890 patch->is_new = 0;
891 patch->is_delete = 1;
892 name = find_name_traditional(first, NULL, p_value);
893 patch->old_name = name;
894 } else {
895 char *first_name;
896 first_name = find_name_traditional(first, NULL, p_value);
897 name = find_name_traditional(second, first_name, p_value);
898 free(first_name);
899 if (has_epoch_timestamp(first)) {
900 patch->is_new = 1;
901 patch->is_delete = 0;
902 patch->new_name = name;
903 } else if (has_epoch_timestamp(second)) {
904 patch->is_new = 0;
905 patch->is_delete = 1;
906 patch->old_name = name;
907 } else {
908 patch->old_name = name;
909 patch->new_name = xstrdup(name);
912 if (!name)
913 die(_("unable to find filename in patch at line %d"), linenr);
916 static int gitdiff_hdrend(const char *line, struct patch *patch)
918 return -1;
922 * We're anal about diff header consistency, to make
923 * sure that we don't end up having strange ambiguous
924 * patches floating around.
926 * As a result, gitdiff_{old|new}name() will check
927 * their names against any previous information, just
928 * to make sure..
930 #define DIFF_OLD_NAME 0
931 #define DIFF_NEW_NAME 1
933 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
935 if (!orig_name && !isnull)
936 return find_name(line, NULL, p_value, TERM_TAB);
938 if (orig_name) {
939 int len;
940 const char *name;
941 char *another;
942 name = orig_name;
943 len = strlen(name);
944 if (isnull)
945 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
946 another = find_name(line, NULL, p_value, TERM_TAB);
947 if (!another || memcmp(another, name, len + 1))
948 die((side == DIFF_NEW_NAME) ?
949 _("git apply: bad git-diff - inconsistent new filename on line %d") :
950 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
951 free(another);
952 return orig_name;
954 else {
955 /* expect "/dev/null" */
956 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
957 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
958 return NULL;
962 static int gitdiff_oldname(const char *line, struct patch *patch)
964 char *orig = patch->old_name;
965 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
966 DIFF_OLD_NAME);
967 if (orig != patch->old_name)
968 free(orig);
969 return 0;
972 static int gitdiff_newname(const char *line, struct patch *patch)
974 char *orig = patch->new_name;
975 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
976 DIFF_NEW_NAME);
977 if (orig != patch->new_name)
978 free(orig);
979 return 0;
982 static int gitdiff_oldmode(const char *line, struct patch *patch)
984 patch->old_mode = strtoul(line, NULL, 8);
985 return 0;
988 static int gitdiff_newmode(const char *line, struct patch *patch)
990 patch->new_mode = strtoul(line, NULL, 8);
991 return 0;
994 static int gitdiff_delete(const char *line, struct patch *patch)
996 patch->is_delete = 1;
997 free(patch->old_name);
998 patch->old_name = null_strdup(patch->def_name);
999 return gitdiff_oldmode(line, patch);
1002 static int gitdiff_newfile(const char *line, struct patch *patch)
1004 patch->is_new = 1;
1005 free(patch->new_name);
1006 patch->new_name = null_strdup(patch->def_name);
1007 return gitdiff_newmode(line, patch);
1010 static int gitdiff_copysrc(const char *line, struct patch *patch)
1012 patch->is_copy = 1;
1013 free(patch->old_name);
1014 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1015 return 0;
1018 static int gitdiff_copydst(const char *line, struct patch *patch)
1020 patch->is_copy = 1;
1021 free(patch->new_name);
1022 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1023 return 0;
1026 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1028 patch->is_rename = 1;
1029 free(patch->old_name);
1030 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1031 return 0;
1034 static int gitdiff_renamedst(const char *line, struct patch *patch)
1036 patch->is_rename = 1;
1037 free(patch->new_name);
1038 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1039 return 0;
1042 static int gitdiff_similarity(const char *line, struct patch *patch)
1044 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1045 patch->score = 0;
1046 return 0;
1049 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1051 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1052 patch->score = 0;
1053 return 0;
1056 static int gitdiff_index(const char *line, struct patch *patch)
1059 * index line is N hexadecimal, "..", N hexadecimal,
1060 * and optional space with octal mode.
1062 const char *ptr, *eol;
1063 int len;
1065 ptr = strchr(line, '.');
1066 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1067 return 0;
1068 len = ptr - line;
1069 memcpy(patch->old_sha1_prefix, line, len);
1070 patch->old_sha1_prefix[len] = 0;
1072 line = ptr + 2;
1073 ptr = strchr(line, ' ');
1074 eol = strchr(line, '\n');
1076 if (!ptr || eol < ptr)
1077 ptr = eol;
1078 len = ptr - line;
1080 if (40 < len)
1081 return 0;
1082 memcpy(patch->new_sha1_prefix, line, len);
1083 patch->new_sha1_prefix[len] = 0;
1084 if (*ptr == ' ')
1085 patch->old_mode = strtoul(ptr+1, NULL, 8);
1086 return 0;
1090 * This is normal for a diff that doesn't change anything: we'll fall through
1091 * into the next diff. Tell the parser to break out.
1093 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1095 return -1;
1099 * Skip p_value leading components from "line"; as we do not accept
1100 * absolute paths, return NULL in that case.
1102 static const char *skip_tree_prefix(const char *line, int llen)
1104 int nslash;
1105 int i;
1107 if (!p_value)
1108 return (llen && line[0] == '/') ? NULL : line;
1110 nslash = p_value;
1111 for (i = 0; i < llen; i++) {
1112 int ch = line[i];
1113 if (ch == '/' && --nslash <= 0)
1114 return (i == 0) ? NULL : &line[i + 1];
1116 return NULL;
1120 * This is to extract the same name that appears on "diff --git"
1121 * line. We do not find and return anything if it is a rename
1122 * patch, and it is OK because we will find the name elsewhere.
1123 * We need to reliably find name only when it is mode-change only,
1124 * creation or deletion of an empty file. In any of these cases,
1125 * both sides are the same name under a/ and b/ respectively.
1127 static char *git_header_name(const char *line, int llen)
1129 const char *name;
1130 const char *second = NULL;
1131 size_t len, line_len;
1133 line += strlen("diff --git ");
1134 llen -= strlen("diff --git ");
1136 if (*line == '"') {
1137 const char *cp;
1138 struct strbuf first = STRBUF_INIT;
1139 struct strbuf sp = STRBUF_INIT;
1141 if (unquote_c_style(&first, line, &second))
1142 goto free_and_fail1;
1144 /* strip the a/b prefix including trailing slash */
1145 cp = skip_tree_prefix(first.buf, first.len);
1146 if (!cp)
1147 goto free_and_fail1;
1148 strbuf_remove(&first, 0, cp - first.buf);
1151 * second points at one past closing dq of name.
1152 * find the second name.
1154 while ((second < line + llen) && isspace(*second))
1155 second++;
1157 if (line + llen <= second)
1158 goto free_and_fail1;
1159 if (*second == '"') {
1160 if (unquote_c_style(&sp, second, NULL))
1161 goto free_and_fail1;
1162 cp = skip_tree_prefix(sp.buf, sp.len);
1163 if (!cp)
1164 goto free_and_fail1;
1165 /* They must match, otherwise ignore */
1166 if (strcmp(cp, first.buf))
1167 goto free_and_fail1;
1168 strbuf_release(&sp);
1169 return strbuf_detach(&first, NULL);
1172 /* unquoted second */
1173 cp = skip_tree_prefix(second, line + llen - second);
1174 if (!cp)
1175 goto free_and_fail1;
1176 if (line + llen - cp != first.len ||
1177 memcmp(first.buf, cp, first.len))
1178 goto free_and_fail1;
1179 return strbuf_detach(&first, NULL);
1181 free_and_fail1:
1182 strbuf_release(&first);
1183 strbuf_release(&sp);
1184 return NULL;
1187 /* unquoted first name */
1188 name = skip_tree_prefix(line, llen);
1189 if (!name)
1190 return NULL;
1193 * since the first name is unquoted, a dq if exists must be
1194 * the beginning of the second name.
1196 for (second = name; second < line + llen; second++) {
1197 if (*second == '"') {
1198 struct strbuf sp = STRBUF_INIT;
1199 const char *np;
1201 if (unquote_c_style(&sp, second, NULL))
1202 goto free_and_fail2;
1204 np = skip_tree_prefix(sp.buf, sp.len);
1205 if (!np)
1206 goto free_and_fail2;
1208 len = sp.buf + sp.len - np;
1209 if (len < second - name &&
1210 !strncmp(np, name, len) &&
1211 isspace(name[len])) {
1212 /* Good */
1213 strbuf_remove(&sp, 0, np - sp.buf);
1214 return strbuf_detach(&sp, NULL);
1217 free_and_fail2:
1218 strbuf_release(&sp);
1219 return NULL;
1224 * Accept a name only if it shows up twice, exactly the same
1225 * form.
1227 second = strchr(name, '\n');
1228 if (!second)
1229 return NULL;
1230 line_len = second - name;
1231 for (len = 0 ; ; len++) {
1232 switch (name[len]) {
1233 default:
1234 continue;
1235 case '\n':
1236 return NULL;
1237 case '\t': case ' ':
1239 * Is this the separator between the preimage
1240 * and the postimage pathname? Again, we are
1241 * only interested in the case where there is
1242 * no rename, as this is only to set def_name
1243 * and a rename patch has the names elsewhere
1244 * in an unambiguous form.
1246 if (!name[len + 1])
1247 return NULL; /* no postimage name */
1248 second = skip_tree_prefix(name + len + 1,
1249 line_len - (len + 1));
1250 if (!second)
1251 return NULL;
1253 * Does len bytes starting at "name" and "second"
1254 * (that are separated by one HT or SP we just
1255 * found) exactly match?
1257 if (second[len] == '\n' && !strncmp(name, second, len))
1258 return xmemdupz(name, len);
1263 /* Verify that we recognize the lines following a git header */
1264 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1266 unsigned long offset;
1268 /* A git diff has explicit new/delete information, so we don't guess */
1269 patch->is_new = 0;
1270 patch->is_delete = 0;
1273 * Some things may not have the old name in the
1274 * rest of the headers anywhere (pure mode changes,
1275 * or removing or adding empty files), so we get
1276 * the default name from the header.
1278 patch->def_name = git_header_name(line, len);
1279 if (patch->def_name && root) {
1280 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1281 strcpy(s, root);
1282 strcpy(s + root_len, patch->def_name);
1283 free(patch->def_name);
1284 patch->def_name = s;
1287 line += len;
1288 size -= len;
1289 linenr++;
1290 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1291 static const struct opentry {
1292 const char *str;
1293 int (*fn)(const char *, struct patch *);
1294 } optable[] = {
1295 { "@@ -", gitdiff_hdrend },
1296 { "--- ", gitdiff_oldname },
1297 { "+++ ", gitdiff_newname },
1298 { "old mode ", gitdiff_oldmode },
1299 { "new mode ", gitdiff_newmode },
1300 { "deleted file mode ", gitdiff_delete },
1301 { "new file mode ", gitdiff_newfile },
1302 { "copy from ", gitdiff_copysrc },
1303 { "copy to ", gitdiff_copydst },
1304 { "rename old ", gitdiff_renamesrc },
1305 { "rename new ", gitdiff_renamedst },
1306 { "rename from ", gitdiff_renamesrc },
1307 { "rename to ", gitdiff_renamedst },
1308 { "similarity index ", gitdiff_similarity },
1309 { "dissimilarity index ", gitdiff_dissimilarity },
1310 { "index ", gitdiff_index },
1311 { "", gitdiff_unrecognized },
1313 int i;
1315 len = linelen(line, size);
1316 if (!len || line[len-1] != '\n')
1317 break;
1318 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1319 const struct opentry *p = optable + i;
1320 int oplen = strlen(p->str);
1321 if (len < oplen || memcmp(p->str, line, oplen))
1322 continue;
1323 if (p->fn(line + oplen, patch) < 0)
1324 return offset;
1325 break;
1329 return offset;
1332 static int parse_num(const char *line, unsigned long *p)
1334 char *ptr;
1336 if (!isdigit(*line))
1337 return 0;
1338 *p = strtoul(line, &ptr, 10);
1339 return ptr - line;
1342 static int parse_range(const char *line, int len, int offset, const char *expect,
1343 unsigned long *p1, unsigned long *p2)
1345 int digits, ex;
1347 if (offset < 0 || offset >= len)
1348 return -1;
1349 line += offset;
1350 len -= offset;
1352 digits = parse_num(line, p1);
1353 if (!digits)
1354 return -1;
1356 offset += digits;
1357 line += digits;
1358 len -= digits;
1360 *p2 = 1;
1361 if (*line == ',') {
1362 digits = parse_num(line+1, p2);
1363 if (!digits)
1364 return -1;
1366 offset += digits+1;
1367 line += digits+1;
1368 len -= digits+1;
1371 ex = strlen(expect);
1372 if (ex > len)
1373 return -1;
1374 if (memcmp(line, expect, ex))
1375 return -1;
1377 return offset + ex;
1380 static void recount_diff(const char *line, int size, struct fragment *fragment)
1382 int oldlines = 0, newlines = 0, ret = 0;
1384 if (size < 1) {
1385 warning("recount: ignore empty hunk");
1386 return;
1389 for (;;) {
1390 int len = linelen(line, size);
1391 size -= len;
1392 line += len;
1394 if (size < 1)
1395 break;
1397 switch (*line) {
1398 case ' ': case '\n':
1399 newlines++;
1400 /* fall through */
1401 case '-':
1402 oldlines++;
1403 continue;
1404 case '+':
1405 newlines++;
1406 continue;
1407 case '\\':
1408 continue;
1409 case '@':
1410 ret = size < 3 || prefixcmp(line, "@@ ");
1411 break;
1412 case 'd':
1413 ret = size < 5 || prefixcmp(line, "diff ");
1414 break;
1415 default:
1416 ret = -1;
1417 break;
1419 if (ret) {
1420 warning(_("recount: unexpected line: %.*s"),
1421 (int)linelen(line, size), line);
1422 return;
1424 break;
1426 fragment->oldlines = oldlines;
1427 fragment->newlines = newlines;
1431 * Parse a unified diff fragment header of the
1432 * form "@@ -a,b +c,d @@"
1434 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1436 int offset;
1438 if (!len || line[len-1] != '\n')
1439 return -1;
1441 /* Figure out the number of lines in a fragment */
1442 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1443 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1445 return offset;
1448 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1450 unsigned long offset, len;
1452 patch->is_toplevel_relative = 0;
1453 patch->is_rename = patch->is_copy = 0;
1454 patch->is_new = patch->is_delete = -1;
1455 patch->old_mode = patch->new_mode = 0;
1456 patch->old_name = patch->new_name = NULL;
1457 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1458 unsigned long nextlen;
1460 len = linelen(line, size);
1461 if (!len)
1462 break;
1464 /* Testing this early allows us to take a few shortcuts.. */
1465 if (len < 6)
1466 continue;
1469 * Make sure we don't find any unconnected patch fragments.
1470 * That's a sign that we didn't find a header, and that a
1471 * patch has become corrupted/broken up.
1473 if (!memcmp("@@ -", line, 4)) {
1474 struct fragment dummy;
1475 if (parse_fragment_header(line, len, &dummy) < 0)
1476 continue;
1477 die(_("patch fragment without header at line %d: %.*s"),
1478 linenr, (int)len-1, line);
1481 if (size < len + 6)
1482 break;
1485 * Git patch? It might not have a real patch, just a rename
1486 * or mode change, so we handle that specially
1488 if (!memcmp("diff --git ", line, 11)) {
1489 int git_hdr_len = parse_git_header(line, len, size, patch);
1490 if (git_hdr_len <= len)
1491 continue;
1492 if (!patch->old_name && !patch->new_name) {
1493 if (!patch->def_name)
1494 die(Q_("git diff header lacks filename information when removing "
1495 "%d leading pathname component (line %d)",
1496 "git diff header lacks filename information when removing "
1497 "%d leading pathname components (line %d)",
1498 p_value),
1499 p_value, linenr);
1500 patch->old_name = xstrdup(patch->def_name);
1501 patch->new_name = xstrdup(patch->def_name);
1503 if (!patch->is_delete && !patch->new_name)
1504 die("git diff header lacks filename information "
1505 "(line %d)", linenr);
1506 patch->is_toplevel_relative = 1;
1507 *hdrsize = git_hdr_len;
1508 return offset;
1511 /* --- followed by +++ ? */
1512 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1513 continue;
1516 * We only accept unified patches, so we want it to
1517 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1518 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1520 nextlen = linelen(line + len, size - len);
1521 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1522 continue;
1524 /* Ok, we'll consider it a patch */
1525 parse_traditional_patch(line, line+len, patch);
1526 *hdrsize = len + nextlen;
1527 linenr += 2;
1528 return offset;
1530 return -1;
1533 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1535 char *err;
1537 if (!result)
1538 return;
1540 whitespace_error++;
1541 if (squelch_whitespace_errors &&
1542 squelch_whitespace_errors < whitespace_error)
1543 return;
1545 err = whitespace_error_string(result);
1546 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1547 patch_input_file, linenr, err, len, line);
1548 free(err);
1551 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1553 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1555 record_ws_error(result, line + 1, len - 2, linenr);
1559 * Parse a unified diff. Note that this really needs to parse each
1560 * fragment separately, since the only way to know the difference
1561 * between a "---" that is part of a patch, and a "---" that starts
1562 * the next patch is to look at the line counts..
1564 static int parse_fragment(const char *line, unsigned long size,
1565 struct patch *patch, struct fragment *fragment)
1567 int added, deleted;
1568 int len = linelen(line, size), offset;
1569 unsigned long oldlines, newlines;
1570 unsigned long leading, trailing;
1572 offset = parse_fragment_header(line, len, fragment);
1573 if (offset < 0)
1574 return -1;
1575 if (offset > 0 && patch->recount)
1576 recount_diff(line + offset, size - offset, fragment);
1577 oldlines = fragment->oldlines;
1578 newlines = fragment->newlines;
1579 leading = 0;
1580 trailing = 0;
1582 /* Parse the thing.. */
1583 line += len;
1584 size -= len;
1585 linenr++;
1586 added = deleted = 0;
1587 for (offset = len;
1588 0 < size;
1589 offset += len, size -= len, line += len, linenr++) {
1590 if (!oldlines && !newlines)
1591 break;
1592 len = linelen(line, size);
1593 if (!len || line[len-1] != '\n')
1594 return -1;
1595 switch (*line) {
1596 default:
1597 return -1;
1598 case '\n': /* newer GNU diff, an empty context line */
1599 case ' ':
1600 oldlines--;
1601 newlines--;
1602 if (!deleted && !added)
1603 leading++;
1604 trailing++;
1605 break;
1606 case '-':
1607 if (apply_in_reverse &&
1608 ws_error_action != nowarn_ws_error)
1609 check_whitespace(line, len, patch->ws_rule);
1610 deleted++;
1611 oldlines--;
1612 trailing = 0;
1613 break;
1614 case '+':
1615 if (!apply_in_reverse &&
1616 ws_error_action != nowarn_ws_error)
1617 check_whitespace(line, len, patch->ws_rule);
1618 added++;
1619 newlines--;
1620 trailing = 0;
1621 break;
1624 * We allow "\ No newline at end of file". Depending
1625 * on locale settings when the patch was produced we
1626 * don't know what this line looks like. The only
1627 * thing we do know is that it begins with "\ ".
1628 * Checking for 12 is just for sanity check -- any
1629 * l10n of "\ No newline..." is at least that long.
1631 case '\\':
1632 if (len < 12 || memcmp(line, "\\ ", 2))
1633 return -1;
1634 break;
1637 if (oldlines || newlines)
1638 return -1;
1639 fragment->leading = leading;
1640 fragment->trailing = trailing;
1643 * If a fragment ends with an incomplete line, we failed to include
1644 * it in the above loop because we hit oldlines == newlines == 0
1645 * before seeing it.
1647 if (12 < size && !memcmp(line, "\\ ", 2))
1648 offset += linelen(line, size);
1650 patch->lines_added += added;
1651 patch->lines_deleted += deleted;
1653 if (0 < patch->is_new && oldlines)
1654 return error(_("new file depends on old contents"));
1655 if (0 < patch->is_delete && newlines)
1656 return error(_("deleted file still has contents"));
1657 return offset;
1661 * We have seen "diff --git a/... b/..." header (or a traditional patch
1662 * header). Read hunks that belong to this patch into fragments and hang
1663 * them to the given patch structure.
1665 * The (fragment->patch, fragment->size) pair points into the memory given
1666 * by the caller, not a copy, when we return.
1668 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1670 unsigned long offset = 0;
1671 unsigned long oldlines = 0, newlines = 0, context = 0;
1672 struct fragment **fragp = &patch->fragments;
1674 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1675 struct fragment *fragment;
1676 int len;
1678 fragment = xcalloc(1, sizeof(*fragment));
1679 fragment->linenr = linenr;
1680 len = parse_fragment(line, size, patch, fragment);
1681 if (len <= 0)
1682 die(_("corrupt patch at line %d"), linenr);
1683 fragment->patch = line;
1684 fragment->size = len;
1685 oldlines += fragment->oldlines;
1686 newlines += fragment->newlines;
1687 context += fragment->leading + fragment->trailing;
1689 *fragp = fragment;
1690 fragp = &fragment->next;
1692 offset += len;
1693 line += len;
1694 size -= len;
1698 * If something was removed (i.e. we have old-lines) it cannot
1699 * be creation, and if something was added it cannot be
1700 * deletion. However, the reverse is not true; --unified=0
1701 * patches that only add are not necessarily creation even
1702 * though they do not have any old lines, and ones that only
1703 * delete are not necessarily deletion.
1705 * Unfortunately, a real creation/deletion patch do _not_ have
1706 * any context line by definition, so we cannot safely tell it
1707 * apart with --unified=0 insanity. At least if the patch has
1708 * more than one hunk it is not creation or deletion.
1710 if (patch->is_new < 0 &&
1711 (oldlines || (patch->fragments && patch->fragments->next)))
1712 patch->is_new = 0;
1713 if (patch->is_delete < 0 &&
1714 (newlines || (patch->fragments && patch->fragments->next)))
1715 patch->is_delete = 0;
1717 if (0 < patch->is_new && oldlines)
1718 die(_("new file %s depends on old contents"), patch->new_name);
1719 if (0 < patch->is_delete && newlines)
1720 die(_("deleted file %s still has contents"), patch->old_name);
1721 if (!patch->is_delete && !newlines && context)
1722 fprintf_ln(stderr,
1723 _("** warning: "
1724 "file %s becomes empty but is not deleted"),
1725 patch->new_name);
1727 return offset;
1730 static inline int metadata_changes(struct patch *patch)
1732 return patch->is_rename > 0 ||
1733 patch->is_copy > 0 ||
1734 patch->is_new > 0 ||
1735 patch->is_delete ||
1736 (patch->old_mode && patch->new_mode &&
1737 patch->old_mode != patch->new_mode);
1740 static char *inflate_it(const void *data, unsigned long size,
1741 unsigned long inflated_size)
1743 git_zstream stream;
1744 void *out;
1745 int st;
1747 memset(&stream, 0, sizeof(stream));
1749 stream.next_in = (unsigned char *)data;
1750 stream.avail_in = size;
1751 stream.next_out = out = xmalloc(inflated_size);
1752 stream.avail_out = inflated_size;
1753 git_inflate_init(&stream);
1754 st = git_inflate(&stream, Z_FINISH);
1755 git_inflate_end(&stream);
1756 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1757 free(out);
1758 return NULL;
1760 return out;
1764 * Read a binary hunk and return a new fragment; fragment->patch
1765 * points at an allocated memory that the caller must free, so
1766 * it is marked as "->free_patch = 1".
1768 static struct fragment *parse_binary_hunk(char **buf_p,
1769 unsigned long *sz_p,
1770 int *status_p,
1771 int *used_p)
1774 * Expect a line that begins with binary patch method ("literal"
1775 * or "delta"), followed by the length of data before deflating.
1776 * a sequence of 'length-byte' followed by base-85 encoded data
1777 * should follow, terminated by a newline.
1779 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1780 * and we would limit the patch line to 66 characters,
1781 * so one line can fit up to 13 groups that would decode
1782 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1783 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1785 int llen, used;
1786 unsigned long size = *sz_p;
1787 char *buffer = *buf_p;
1788 int patch_method;
1789 unsigned long origlen;
1790 char *data = NULL;
1791 int hunk_size = 0;
1792 struct fragment *frag;
1794 llen = linelen(buffer, size);
1795 used = llen;
1797 *status_p = 0;
1799 if (!prefixcmp(buffer, "delta ")) {
1800 patch_method = BINARY_DELTA_DEFLATED;
1801 origlen = strtoul(buffer + 6, NULL, 10);
1803 else if (!prefixcmp(buffer, "literal ")) {
1804 patch_method = BINARY_LITERAL_DEFLATED;
1805 origlen = strtoul(buffer + 8, NULL, 10);
1807 else
1808 return NULL;
1810 linenr++;
1811 buffer += llen;
1812 while (1) {
1813 int byte_length, max_byte_length, newsize;
1814 llen = linelen(buffer, size);
1815 used += llen;
1816 linenr++;
1817 if (llen == 1) {
1818 /* consume the blank line */
1819 buffer++;
1820 size--;
1821 break;
1824 * Minimum line is "A00000\n" which is 7-byte long,
1825 * and the line length must be multiple of 5 plus 2.
1827 if ((llen < 7) || (llen-2) % 5)
1828 goto corrupt;
1829 max_byte_length = (llen - 2) / 5 * 4;
1830 byte_length = *buffer;
1831 if ('A' <= byte_length && byte_length <= 'Z')
1832 byte_length = byte_length - 'A' + 1;
1833 else if ('a' <= byte_length && byte_length <= 'z')
1834 byte_length = byte_length - 'a' + 27;
1835 else
1836 goto corrupt;
1837 /* if the input length was not multiple of 4, we would
1838 * have filler at the end but the filler should never
1839 * exceed 3 bytes
1841 if (max_byte_length < byte_length ||
1842 byte_length <= max_byte_length - 4)
1843 goto corrupt;
1844 newsize = hunk_size + byte_length;
1845 data = xrealloc(data, newsize);
1846 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1847 goto corrupt;
1848 hunk_size = newsize;
1849 buffer += llen;
1850 size -= llen;
1853 frag = xcalloc(1, sizeof(*frag));
1854 frag->patch = inflate_it(data, hunk_size, origlen);
1855 frag->free_patch = 1;
1856 if (!frag->patch)
1857 goto corrupt;
1858 free(data);
1859 frag->size = origlen;
1860 *buf_p = buffer;
1861 *sz_p = size;
1862 *used_p = used;
1863 frag->binary_patch_method = patch_method;
1864 return frag;
1866 corrupt:
1867 free(data);
1868 *status_p = -1;
1869 error(_("corrupt binary patch at line %d: %.*s"),
1870 linenr-1, llen-1, buffer);
1871 return NULL;
1874 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1877 * We have read "GIT binary patch\n"; what follows is a line
1878 * that says the patch method (currently, either "literal" or
1879 * "delta") and the length of data before deflating; a
1880 * sequence of 'length-byte' followed by base-85 encoded data
1881 * follows.
1883 * When a binary patch is reversible, there is another binary
1884 * hunk in the same format, starting with patch method (either
1885 * "literal" or "delta") with the length of data, and a sequence
1886 * of length-byte + base-85 encoded data, terminated with another
1887 * empty line. This data, when applied to the postimage, produces
1888 * the preimage.
1890 struct fragment *forward;
1891 struct fragment *reverse;
1892 int status;
1893 int used, used_1;
1895 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1896 if (!forward && !status)
1897 /* there has to be one hunk (forward hunk) */
1898 return error(_("unrecognized binary patch at line %d"), linenr-1);
1899 if (status)
1900 /* otherwise we already gave an error message */
1901 return status;
1903 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1904 if (reverse)
1905 used += used_1;
1906 else if (status) {
1908 * Not having reverse hunk is not an error, but having
1909 * a corrupt reverse hunk is.
1911 free((void*) forward->patch);
1912 free(forward);
1913 return status;
1915 forward->next = reverse;
1916 patch->fragments = forward;
1917 patch->is_binary = 1;
1918 return used;
1922 * Read the patch text in "buffer" taht extends for "size" bytes; stop
1923 * reading after seeing a single patch (i.e. changes to a single file).
1924 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1925 * Return the number of bytes consumed, so that the caller can call us
1926 * again for the next patch.
1928 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1930 int hdrsize, patchsize;
1931 int offset = find_header(buffer, size, &hdrsize, patch);
1933 if (offset < 0)
1934 return offset;
1936 patch->ws_rule = whitespace_rule(patch->new_name
1937 ? patch->new_name
1938 : patch->old_name);
1940 patchsize = parse_single_patch(buffer + offset + hdrsize,
1941 size - offset - hdrsize, patch);
1943 if (!patchsize) {
1944 static const char *binhdr[] = {
1945 "Binary files ",
1946 "Files ",
1947 NULL,
1949 static const char git_binary[] = "GIT binary patch\n";
1950 int i;
1951 int hd = hdrsize + offset;
1952 unsigned long llen = linelen(buffer + hd, size - hd);
1954 if (llen == sizeof(git_binary) - 1 &&
1955 !memcmp(git_binary, buffer + hd, llen)) {
1956 int used;
1957 linenr++;
1958 used = parse_binary(buffer + hd + llen,
1959 size - hd - llen, patch);
1960 if (used)
1961 patchsize = used + llen;
1962 else
1963 patchsize = 0;
1965 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1966 for (i = 0; binhdr[i]; i++) {
1967 int len = strlen(binhdr[i]);
1968 if (len < size - hd &&
1969 !memcmp(binhdr[i], buffer + hd, len)) {
1970 linenr++;
1971 patch->is_binary = 1;
1972 patchsize = llen;
1973 break;
1978 /* Empty patch cannot be applied if it is a text patch
1979 * without metadata change. A binary patch appears
1980 * empty to us here.
1982 if ((apply || check) &&
1983 (!patch->is_binary && !metadata_changes(patch)))
1984 die(_("patch with only garbage at line %d"), linenr);
1987 return offset + hdrsize + patchsize;
1990 #define swap(a,b) myswap((a),(b),sizeof(a))
1992 #define myswap(a, b, size) do { \
1993 unsigned char mytmp[size]; \
1994 memcpy(mytmp, &a, size); \
1995 memcpy(&a, &b, size); \
1996 memcpy(&b, mytmp, size); \
1997 } while (0)
1999 static void reverse_patches(struct patch *p)
2001 for (; p; p = p->next) {
2002 struct fragment *frag = p->fragments;
2004 swap(p->new_name, p->old_name);
2005 swap(p->new_mode, p->old_mode);
2006 swap(p->is_new, p->is_delete);
2007 swap(p->lines_added, p->lines_deleted);
2008 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2010 for (; frag; frag = frag->next) {
2011 swap(frag->newpos, frag->oldpos);
2012 swap(frag->newlines, frag->oldlines);
2017 static const char pluses[] =
2018 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2019 static const char minuses[]=
2020 "----------------------------------------------------------------------";
2022 static void show_stats(struct patch *patch)
2024 struct strbuf qname = STRBUF_INIT;
2025 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2026 int max, add, del;
2028 quote_c_style(cp, &qname, NULL, 0);
2031 * "scale" the filename
2033 max = max_len;
2034 if (max > 50)
2035 max = 50;
2037 if (qname.len > max) {
2038 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2039 if (!cp)
2040 cp = qname.buf + qname.len + 3 - max;
2041 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2044 if (patch->is_binary) {
2045 printf(" %-*s | Bin\n", max, qname.buf);
2046 strbuf_release(&qname);
2047 return;
2050 printf(" %-*s |", max, qname.buf);
2051 strbuf_release(&qname);
2054 * scale the add/delete
2056 max = max + max_change > 70 ? 70 - max : max_change;
2057 add = patch->lines_added;
2058 del = patch->lines_deleted;
2060 if (max_change > 0) {
2061 int total = ((add + del) * max + max_change / 2) / max_change;
2062 add = (add * max + max_change / 2) / max_change;
2063 del = total - add;
2065 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2066 add, pluses, del, minuses);
2069 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2071 switch (st->st_mode & S_IFMT) {
2072 case S_IFLNK:
2073 if (strbuf_readlink(buf, path, st->st_size) < 0)
2074 return error(_("unable to read symlink %s"), path);
2075 return 0;
2076 case S_IFREG:
2077 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2078 return error(_("unable to open or read %s"), path);
2079 convert_to_git(path, buf->buf, buf->len, buf, 0);
2080 return 0;
2081 default:
2082 return -1;
2087 * Update the preimage, and the common lines in postimage,
2088 * from buffer buf of length len. If postlen is 0 the postimage
2089 * is updated in place, otherwise it's updated on a new buffer
2090 * of length postlen
2093 static void update_pre_post_images(struct image *preimage,
2094 struct image *postimage,
2095 char *buf,
2096 size_t len, size_t postlen)
2098 int i, ctx, reduced;
2099 char *new, *old, *fixed;
2100 struct image fixed_preimage;
2103 * Update the preimage with whitespace fixes. Note that we
2104 * are not losing preimage->buf -- apply_one_fragment() will
2105 * free "oldlines".
2107 prepare_image(&fixed_preimage, buf, len, 1);
2108 assert(postlen
2109 ? fixed_preimage.nr == preimage->nr
2110 : fixed_preimage.nr <= preimage->nr);
2111 for (i = 0; i < fixed_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;
2132 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2133 size_t len = postimage->line[i].len;
2134 if (!(postimage->line[i].flag & LINE_COMMON)) {
2135 /* an added line -- no counterparts in preimage */
2136 memmove(new, old, len);
2137 old += len;
2138 new += len;
2139 continue;
2142 /* a common context -- skip it in the original postimage */
2143 old += len;
2145 /* and find the corresponding one in the fixed preimage */
2146 while (ctx < preimage->nr &&
2147 !(preimage->line[ctx].flag & LINE_COMMON)) {
2148 fixed += preimage->line[ctx].len;
2149 ctx++;
2153 * preimage is expected to run out, if the caller
2154 * fixed addition of trailing blank lines.
2156 if (preimage->nr <= ctx) {
2157 reduced++;
2158 continue;
2161 /* and copy it in, while fixing the line length */
2162 len = preimage->line[ctx].len;
2163 memcpy(new, fixed, len);
2164 new += len;
2165 fixed += len;
2166 postimage->line[i].len = len;
2167 ctx++;
2170 /* Fix the length of the whole thing */
2171 postimage->len = new - postimage->buf;
2172 postimage->nr -= reduced;
2175 static int match_fragment(struct image *img,
2176 struct image *preimage,
2177 struct image *postimage,
2178 unsigned long try,
2179 int try_lno,
2180 unsigned ws_rule,
2181 int match_beginning, int match_end)
2183 int i;
2184 char *fixed_buf, *buf, *orig, *target;
2185 struct strbuf fixed;
2186 size_t fixed_len;
2187 int preimage_limit;
2189 if (preimage->nr + try_lno <= img->nr) {
2191 * The hunk falls within the boundaries of img.
2193 preimage_limit = preimage->nr;
2194 if (match_end && (preimage->nr + try_lno != img->nr))
2195 return 0;
2196 } else if (ws_error_action == correct_ws_error &&
2197 (ws_rule & WS_BLANK_AT_EOF)) {
2199 * This hunk extends beyond the end of img, and we are
2200 * removing blank lines at the end of the file. This
2201 * many lines from the beginning of the preimage must
2202 * match with img, and the remainder of the preimage
2203 * must be blank.
2205 preimage_limit = img->nr - try_lno;
2206 } else {
2208 * The hunk extends beyond the end of the img and
2209 * we are not removing blanks at the end, so we
2210 * should reject the hunk at this position.
2212 return 0;
2215 if (match_beginning && try_lno)
2216 return 0;
2218 /* Quick hash check */
2219 for (i = 0; i < preimage_limit; i++)
2220 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2221 (preimage->line[i].hash != img->line[try_lno + i].hash))
2222 return 0;
2224 if (preimage_limit == preimage->nr) {
2226 * Do we have an exact match? If we were told to match
2227 * at the end, size must be exactly at try+fragsize,
2228 * otherwise try+fragsize must be still within the preimage,
2229 * and either case, the old piece should match the preimage
2230 * exactly.
2232 if ((match_end
2233 ? (try + preimage->len == img->len)
2234 : (try + preimage->len <= img->len)) &&
2235 !memcmp(img->buf + try, preimage->buf, preimage->len))
2236 return 1;
2237 } else {
2239 * The preimage extends beyond the end of img, so
2240 * there cannot be an exact match.
2242 * There must be one non-blank context line that match
2243 * a line before the end of img.
2245 char *buf_end;
2247 buf = preimage->buf;
2248 buf_end = buf;
2249 for (i = 0; i < preimage_limit; i++)
2250 buf_end += preimage->line[i].len;
2252 for ( ; buf < buf_end; buf++)
2253 if (!isspace(*buf))
2254 break;
2255 if (buf == buf_end)
2256 return 0;
2260 * No exact match. If we are ignoring whitespace, run a line-by-line
2261 * fuzzy matching. We collect all the line length information because
2262 * we need it to adjust whitespace if we match.
2264 if (ws_ignore_action == ignore_ws_change) {
2265 size_t imgoff = 0;
2266 size_t preoff = 0;
2267 size_t postlen = postimage->len;
2268 size_t extra_chars;
2269 char *preimage_eof;
2270 char *preimage_end;
2271 for (i = 0; i < preimage_limit; i++) {
2272 size_t prelen = preimage->line[i].len;
2273 size_t imglen = img->line[try_lno+i].len;
2275 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2276 preimage->buf + preoff, prelen))
2277 return 0;
2278 if (preimage->line[i].flag & LINE_COMMON)
2279 postlen += imglen - prelen;
2280 imgoff += imglen;
2281 preoff += prelen;
2285 * Ok, the preimage matches with whitespace fuzz.
2287 * imgoff now holds the true length of the target that
2288 * matches the preimage before the end of the file.
2290 * Count the number of characters in the preimage that fall
2291 * beyond the end of the file and make sure that all of them
2292 * are whitespace characters. (This can only happen if
2293 * we are removing blank lines at the end of the file.)
2295 buf = preimage_eof = preimage->buf + preoff;
2296 for ( ; i < preimage->nr; i++)
2297 preoff += preimage->line[i].len;
2298 preimage_end = preimage->buf + preoff;
2299 for ( ; buf < preimage_end; buf++)
2300 if (!isspace(*buf))
2301 return 0;
2304 * Update the preimage and the common postimage context
2305 * lines to use the same whitespace as the target.
2306 * If whitespace is missing in the target (i.e.
2307 * if the preimage extends beyond the end of the file),
2308 * use the whitespace from the preimage.
2310 extra_chars = preimage_end - preimage_eof;
2311 strbuf_init(&fixed, imgoff + extra_chars);
2312 strbuf_add(&fixed, img->buf + try, imgoff);
2313 strbuf_add(&fixed, preimage_eof, extra_chars);
2314 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2315 update_pre_post_images(preimage, postimage,
2316 fixed_buf, fixed_len, postlen);
2317 return 1;
2320 if (ws_error_action != correct_ws_error)
2321 return 0;
2324 * The hunk does not apply byte-by-byte, but the hash says
2325 * it might with whitespace fuzz. We haven't been asked to
2326 * ignore whitespace, we were asked to correct whitespace
2327 * errors, so let's try matching after whitespace correction.
2329 * The preimage may extend beyond the end of the file,
2330 * but in this loop we will only handle the part of the
2331 * preimage that falls within the file.
2333 strbuf_init(&fixed, preimage->len + 1);
2334 orig = preimage->buf;
2335 target = img->buf + try;
2336 for (i = 0; i < preimage_limit; i++) {
2337 size_t oldlen = preimage->line[i].len;
2338 size_t tgtlen = img->line[try_lno + i].len;
2339 size_t fixstart = fixed.len;
2340 struct strbuf tgtfix;
2341 int match;
2343 /* Try fixing the line in the preimage */
2344 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2346 /* Try fixing the line in the target */
2347 strbuf_init(&tgtfix, tgtlen);
2348 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2351 * If they match, either the preimage was based on
2352 * a version before our tree fixed whitespace breakage,
2353 * or we are lacking a whitespace-fix patch the tree
2354 * the preimage was based on already had (i.e. target
2355 * has whitespace breakage, the preimage doesn't).
2356 * In either case, we are fixing the whitespace breakages
2357 * so we might as well take the fix together with their
2358 * real change.
2360 match = (tgtfix.len == fixed.len - fixstart &&
2361 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2362 fixed.len - fixstart));
2364 strbuf_release(&tgtfix);
2365 if (!match)
2366 goto unmatch_exit;
2368 orig += oldlen;
2369 target += tgtlen;
2374 * Now handle the lines in the preimage that falls beyond the
2375 * end of the file (if any). They will only match if they are
2376 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2377 * false).
2379 for ( ; i < preimage->nr; i++) {
2380 size_t fixstart = fixed.len; /* start of the fixed preimage */
2381 size_t oldlen = preimage->line[i].len;
2382 int j;
2384 /* Try fixing the line in the preimage */
2385 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2387 for (j = fixstart; j < fixed.len; j++)
2388 if (!isspace(fixed.buf[j]))
2389 goto unmatch_exit;
2391 orig += oldlen;
2395 * Yes, the preimage is based on an older version that still
2396 * has whitespace breakages unfixed, and fixing them makes the
2397 * hunk match. Update the context lines in the postimage.
2399 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2400 update_pre_post_images(preimage, postimage,
2401 fixed_buf, fixed_len, 0);
2402 return 1;
2404 unmatch_exit:
2405 strbuf_release(&fixed);
2406 return 0;
2409 static int find_pos(struct image *img,
2410 struct image *preimage,
2411 struct image *postimage,
2412 int line,
2413 unsigned ws_rule,
2414 int match_beginning, int match_end)
2416 int i;
2417 unsigned long backwards, forwards, try;
2418 int backwards_lno, forwards_lno, try_lno;
2421 * If match_beginning or match_end is specified, there is no
2422 * point starting from a wrong line that will never match and
2423 * wander around and wait for a match at the specified end.
2425 if (match_beginning)
2426 line = 0;
2427 else if (match_end)
2428 line = img->nr - preimage->nr;
2431 * Because the comparison is unsigned, the following test
2432 * will also take care of a negative line number that can
2433 * result when match_end and preimage is larger than the target.
2435 if ((size_t) line > img->nr)
2436 line = img->nr;
2438 try = 0;
2439 for (i = 0; i < line; i++)
2440 try += img->line[i].len;
2443 * There's probably some smart way to do this, but I'll leave
2444 * that to the smart and beautiful people. I'm simple and stupid.
2446 backwards = try;
2447 backwards_lno = line;
2448 forwards = try;
2449 forwards_lno = line;
2450 try_lno = line;
2452 for (i = 0; ; i++) {
2453 if (match_fragment(img, preimage, postimage,
2454 try, try_lno, ws_rule,
2455 match_beginning, match_end))
2456 return try_lno;
2458 again:
2459 if (backwards_lno == 0 && forwards_lno == img->nr)
2460 break;
2462 if (i & 1) {
2463 if (backwards_lno == 0) {
2464 i++;
2465 goto again;
2467 backwards_lno--;
2468 backwards -= img->line[backwards_lno].len;
2469 try = backwards;
2470 try_lno = backwards_lno;
2471 } else {
2472 if (forwards_lno == img->nr) {
2473 i++;
2474 goto again;
2476 forwards += img->line[forwards_lno].len;
2477 forwards_lno++;
2478 try = forwards;
2479 try_lno = forwards_lno;
2483 return -1;
2486 static void remove_first_line(struct image *img)
2488 img->buf += img->line[0].len;
2489 img->len -= img->line[0].len;
2490 img->line++;
2491 img->nr--;
2494 static void remove_last_line(struct image *img)
2496 img->len -= img->line[--img->nr].len;
2500 * The change from "preimage" and "postimage" has been found to
2501 * apply at applied_pos (counts in line numbers) in "img".
2502 * Update "img" to remove "preimage" and replace it with "postimage".
2504 static void update_image(struct image *img,
2505 int applied_pos,
2506 struct image *preimage,
2507 struct image *postimage)
2510 * remove the copy of preimage at offset in img
2511 * and replace it with postimage
2513 int i, nr;
2514 size_t remove_count, insert_count, applied_at = 0;
2515 char *result;
2516 int preimage_limit;
2519 * If we are removing blank lines at the end of img,
2520 * the preimage may extend beyond the end.
2521 * If that is the case, we must be careful only to
2522 * remove the part of the preimage that falls within
2523 * the boundaries of img. Initialize preimage_limit
2524 * to the number of lines in the preimage that falls
2525 * within the boundaries.
2527 preimage_limit = preimage->nr;
2528 if (preimage_limit > img->nr - applied_pos)
2529 preimage_limit = img->nr - applied_pos;
2531 for (i = 0; i < applied_pos; i++)
2532 applied_at += img->line[i].len;
2534 remove_count = 0;
2535 for (i = 0; i < preimage_limit; i++)
2536 remove_count += img->line[applied_pos + i].len;
2537 insert_count = postimage->len;
2539 /* Adjust the contents */
2540 result = xmalloc(img->len + insert_count - remove_count + 1);
2541 memcpy(result, img->buf, applied_at);
2542 memcpy(result + applied_at, postimage->buf, postimage->len);
2543 memcpy(result + applied_at + postimage->len,
2544 img->buf + (applied_at + remove_count),
2545 img->len - (applied_at + remove_count));
2546 free(img->buf);
2547 img->buf = result;
2548 img->len += insert_count - remove_count;
2549 result[img->len] = '\0';
2551 /* Adjust the line table */
2552 nr = img->nr + postimage->nr - preimage_limit;
2553 if (preimage_limit < postimage->nr) {
2555 * NOTE: this knows that we never call remove_first_line()
2556 * on anything other than pre/post image.
2558 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2559 img->line_allocated = img->line;
2561 if (preimage_limit != postimage->nr)
2562 memmove(img->line + applied_pos + postimage->nr,
2563 img->line + applied_pos + preimage_limit,
2564 (img->nr - (applied_pos + preimage_limit)) *
2565 sizeof(*img->line));
2566 memcpy(img->line + applied_pos,
2567 postimage->line,
2568 postimage->nr * sizeof(*img->line));
2569 if (!allow_overlap)
2570 for (i = 0; i < postimage->nr; i++)
2571 img->line[applied_pos + i].flag |= LINE_PATCHED;
2572 img->nr = nr;
2576 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2577 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2578 * replace the part of "img" with "postimage" text.
2580 static int apply_one_fragment(struct image *img, struct fragment *frag,
2581 int inaccurate_eof, unsigned ws_rule,
2582 int nth_fragment)
2584 int match_beginning, match_end;
2585 const char *patch = frag->patch;
2586 int size = frag->size;
2587 char *old, *oldlines;
2588 struct strbuf newlines;
2589 int new_blank_lines_at_end = 0;
2590 int found_new_blank_lines_at_end = 0;
2591 int hunk_linenr = frag->linenr;
2592 unsigned long leading, trailing;
2593 int pos, applied_pos;
2594 struct image preimage;
2595 struct image postimage;
2597 memset(&preimage, 0, sizeof(preimage));
2598 memset(&postimage, 0, sizeof(postimage));
2599 oldlines = xmalloc(size);
2600 strbuf_init(&newlines, size);
2602 old = oldlines;
2603 while (size > 0) {
2604 char first;
2605 int len = linelen(patch, size);
2606 int plen;
2607 int added_blank_line = 0;
2608 int is_blank_context = 0;
2609 size_t start;
2611 if (!len)
2612 break;
2615 * "plen" is how much of the line we should use for
2616 * the actual patch data. Normally we just remove the
2617 * first character on the line, but if the line is
2618 * followed by "\ No newline", then we also remove the
2619 * last one (which is the newline, of course).
2621 plen = len - 1;
2622 if (len < size && patch[len] == '\\')
2623 plen--;
2624 first = *patch;
2625 if (apply_in_reverse) {
2626 if (first == '-')
2627 first = '+';
2628 else if (first == '+')
2629 first = '-';
2632 switch (first) {
2633 case '\n':
2634 /* Newer GNU diff, empty context line */
2635 if (plen < 0)
2636 /* ... followed by '\No newline'; nothing */
2637 break;
2638 *old++ = '\n';
2639 strbuf_addch(&newlines, '\n');
2640 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2641 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2642 is_blank_context = 1;
2643 break;
2644 case ' ':
2645 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2646 ws_blank_line(patch + 1, plen, ws_rule))
2647 is_blank_context = 1;
2648 case '-':
2649 memcpy(old, patch + 1, plen);
2650 add_line_info(&preimage, old, plen,
2651 (first == ' ' ? LINE_COMMON : 0));
2652 old += plen;
2653 if (first == '-')
2654 break;
2655 /* Fall-through for ' ' */
2656 case '+':
2657 /* --no-add does not add new lines */
2658 if (first == '+' && no_add)
2659 break;
2661 start = newlines.len;
2662 if (first != '+' ||
2663 !whitespace_error ||
2664 ws_error_action != correct_ws_error) {
2665 strbuf_add(&newlines, patch + 1, plen);
2667 else {
2668 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2670 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2671 (first == '+' ? 0 : LINE_COMMON));
2672 if (first == '+' &&
2673 (ws_rule & WS_BLANK_AT_EOF) &&
2674 ws_blank_line(patch + 1, plen, ws_rule))
2675 added_blank_line = 1;
2676 break;
2677 case '@': case '\\':
2678 /* Ignore it, we already handled it */
2679 break;
2680 default:
2681 if (apply_verbosely)
2682 error(_("invalid start of line: '%c'"), first);
2683 return -1;
2685 if (added_blank_line) {
2686 if (!new_blank_lines_at_end)
2687 found_new_blank_lines_at_end = hunk_linenr;
2688 new_blank_lines_at_end++;
2690 else if (is_blank_context)
2692 else
2693 new_blank_lines_at_end = 0;
2694 patch += len;
2695 size -= len;
2696 hunk_linenr++;
2698 if (inaccurate_eof &&
2699 old > oldlines && old[-1] == '\n' &&
2700 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2701 old--;
2702 strbuf_setlen(&newlines, newlines.len - 1);
2705 leading = frag->leading;
2706 trailing = frag->trailing;
2709 * A hunk to change lines at the beginning would begin with
2710 * @@ -1,L +N,M @@
2711 * but we need to be careful. -U0 that inserts before the second
2712 * line also has this pattern.
2714 * And a hunk to add to an empty file would begin with
2715 * @@ -0,0 +N,M @@
2717 * In other words, a hunk that is (frag->oldpos <= 1) with or
2718 * without leading context must match at the beginning.
2720 match_beginning = (!frag->oldpos ||
2721 (frag->oldpos == 1 && !unidiff_zero));
2724 * A hunk without trailing lines must match at the end.
2725 * However, we simply cannot tell if a hunk must match end
2726 * from the lack of trailing lines if the patch was generated
2727 * with unidiff without any context.
2729 match_end = !unidiff_zero && !trailing;
2731 pos = frag->newpos ? (frag->newpos - 1) : 0;
2732 preimage.buf = oldlines;
2733 preimage.len = old - oldlines;
2734 postimage.buf = newlines.buf;
2735 postimage.len = newlines.len;
2736 preimage.line = preimage.line_allocated;
2737 postimage.line = postimage.line_allocated;
2739 for (;;) {
2741 applied_pos = find_pos(img, &preimage, &postimage, pos,
2742 ws_rule, match_beginning, match_end);
2744 if (applied_pos >= 0)
2745 break;
2747 /* Am I at my context limits? */
2748 if ((leading <= p_context) && (trailing <= p_context))
2749 break;
2750 if (match_beginning || match_end) {
2751 match_beginning = match_end = 0;
2752 continue;
2756 * Reduce the number of context lines; reduce both
2757 * leading and trailing if they are equal otherwise
2758 * just reduce the larger context.
2760 if (leading >= trailing) {
2761 remove_first_line(&preimage);
2762 remove_first_line(&postimage);
2763 pos--;
2764 leading--;
2766 if (trailing > leading) {
2767 remove_last_line(&preimage);
2768 remove_last_line(&postimage);
2769 trailing--;
2773 if (applied_pos >= 0) {
2774 if (new_blank_lines_at_end &&
2775 preimage.nr + applied_pos >= img->nr &&
2776 (ws_rule & WS_BLANK_AT_EOF) &&
2777 ws_error_action != nowarn_ws_error) {
2778 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2779 found_new_blank_lines_at_end);
2780 if (ws_error_action == correct_ws_error) {
2781 while (new_blank_lines_at_end--)
2782 remove_last_line(&postimage);
2785 * We would want to prevent write_out_results()
2786 * from taking place in apply_patch() that follows
2787 * the callchain led us here, which is:
2788 * apply_patch->check_patch_list->check_patch->
2789 * apply_data->apply_fragments->apply_one_fragment
2791 if (ws_error_action == die_on_ws_error)
2792 apply = 0;
2795 if (apply_verbosely && applied_pos != pos) {
2796 int offset = applied_pos - pos;
2797 if (apply_in_reverse)
2798 offset = 0 - offset;
2799 fprintf_ln(stderr,
2800 Q_("Hunk #%d succeeded at %d (offset %d line).",
2801 "Hunk #%d succeeded at %d (offset %d lines).",
2802 offset),
2803 nth_fragment, applied_pos + 1, offset);
2807 * Warn if it was necessary to reduce the number
2808 * of context lines.
2810 if ((leading != frag->leading) ||
2811 (trailing != frag->trailing))
2812 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2813 " to apply fragment at %d"),
2814 leading, trailing, applied_pos+1);
2815 update_image(img, applied_pos, &preimage, &postimage);
2816 } else {
2817 if (apply_verbosely)
2818 error(_("while searching for:\n%.*s"),
2819 (int)(old - oldlines), oldlines);
2822 free(oldlines);
2823 strbuf_release(&newlines);
2824 free(preimage.line_allocated);
2825 free(postimage.line_allocated);
2827 return (applied_pos < 0);
2830 static int apply_binary_fragment(struct image *img, struct patch *patch)
2832 struct fragment *fragment = patch->fragments;
2833 unsigned long len;
2834 void *dst;
2836 if (!fragment)
2837 return error(_("missing binary patch data for '%s'"),
2838 patch->new_name ?
2839 patch->new_name :
2840 patch->old_name);
2842 /* Binary patch is irreversible without the optional second hunk */
2843 if (apply_in_reverse) {
2844 if (!fragment->next)
2845 return error("cannot reverse-apply a binary patch "
2846 "without the reverse hunk to '%s'",
2847 patch->new_name
2848 ? patch->new_name : patch->old_name);
2849 fragment = fragment->next;
2851 switch (fragment->binary_patch_method) {
2852 case BINARY_DELTA_DEFLATED:
2853 dst = patch_delta(img->buf, img->len, fragment->patch,
2854 fragment->size, &len);
2855 if (!dst)
2856 return -1;
2857 clear_image(img);
2858 img->buf = dst;
2859 img->len = len;
2860 return 0;
2861 case BINARY_LITERAL_DEFLATED:
2862 clear_image(img);
2863 img->len = fragment->size;
2864 img->buf = xmalloc(img->len+1);
2865 memcpy(img->buf, fragment->patch, img->len);
2866 img->buf[img->len] = '\0';
2867 return 0;
2869 return -1;
2873 * Replace "img" with the result of applying the binary patch.
2874 * The binary patch data itself in patch->fragment is still kept
2875 * but the preimage prepared by the caller in "img" is freed here
2876 * or in the helper function apply_binary_fragment() this calls.
2878 static int apply_binary(struct image *img, struct patch *patch)
2880 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2881 unsigned char sha1[20];
2884 * For safety, we require patch index line to contain
2885 * full 40-byte textual SHA1 for old and new, at least for now.
2887 if (strlen(patch->old_sha1_prefix) != 40 ||
2888 strlen(patch->new_sha1_prefix) != 40 ||
2889 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2890 get_sha1_hex(patch->new_sha1_prefix, sha1))
2891 return error("cannot apply binary patch to '%s' "
2892 "without full index line", name);
2894 if (patch->old_name) {
2896 * See if the old one matches what the patch
2897 * applies to.
2899 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2900 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2901 return error("the patch applies to '%s' (%s), "
2902 "which does not match the "
2903 "current contents.",
2904 name, sha1_to_hex(sha1));
2906 else {
2907 /* Otherwise, the old one must be empty. */
2908 if (img->len)
2909 return error("the patch applies to an empty "
2910 "'%s' but it is not empty", name);
2913 get_sha1_hex(patch->new_sha1_prefix, sha1);
2914 if (is_null_sha1(sha1)) {
2915 clear_image(img);
2916 return 0; /* deletion patch */
2919 if (has_sha1_file(sha1)) {
2920 /* We already have the postimage */
2921 enum object_type type;
2922 unsigned long size;
2923 char *result;
2925 result = read_sha1_file(sha1, &type, &size);
2926 if (!result)
2927 return error("the necessary postimage %s for "
2928 "'%s' cannot be read",
2929 patch->new_sha1_prefix, name);
2930 clear_image(img);
2931 img->buf = result;
2932 img->len = size;
2933 } else {
2935 * We have verified buf matches the preimage;
2936 * apply the patch data to it, which is stored
2937 * in the patch->fragments->{patch,size}.
2939 if (apply_binary_fragment(img, patch))
2940 return error(_("binary patch does not apply to '%s'"),
2941 name);
2943 /* verify that the result matches */
2944 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2945 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2946 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2947 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2950 return 0;
2953 static int apply_fragments(struct image *img, struct patch *patch)
2955 struct fragment *frag = patch->fragments;
2956 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2957 unsigned ws_rule = patch->ws_rule;
2958 unsigned inaccurate_eof = patch->inaccurate_eof;
2959 int nth = 0;
2961 if (patch->is_binary)
2962 return apply_binary(img, patch);
2964 while (frag) {
2965 nth++;
2966 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2967 error(_("patch failed: %s:%ld"), name, frag->oldpos);
2968 if (!apply_with_reject)
2969 return -1;
2970 frag->rejected = 1;
2972 frag = frag->next;
2974 return 0;
2977 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2979 if (S_ISGITLINK(mode)) {
2980 strbuf_grow(buf, 100);
2981 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2982 } else {
2983 enum object_type type;
2984 unsigned long sz;
2985 char *result;
2987 result = read_sha1_file(sha1, &type, &sz);
2988 if (!result)
2989 return -1;
2990 /* XXX read_sha1_file NUL-terminates */
2991 strbuf_attach(buf, result, sz, sz + 1);
2993 return 0;
2996 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2998 if (!ce)
2999 return 0;
3000 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3003 static struct patch *in_fn_table(const char *name)
3005 struct string_list_item *item;
3007 if (name == NULL)
3008 return NULL;
3010 item = string_list_lookup(&fn_table, name);
3011 if (item != NULL)
3012 return (struct patch *)item->util;
3014 return NULL;
3018 * item->util in the filename table records the status of the path.
3019 * Usually it points at a patch (whose result records the contents
3020 * of it after applying it), but it could be PATH_WAS_DELETED for a
3021 * path that a previously applied patch has already removed, or
3022 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3024 * The latter is needed to deal with a case where two paths A and B
3025 * are swapped by first renaming A to B and then renaming B to A;
3026 * moving A to B should not be prevented due to presense of B as we
3027 * will remove it in a later patch.
3029 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3030 #define PATH_WAS_DELETED ((struct patch *) -1)
3032 static int to_be_deleted(struct patch *patch)
3034 return patch == PATH_TO_BE_DELETED;
3037 static int was_deleted(struct patch *patch)
3039 return patch == PATH_WAS_DELETED;
3042 static void add_to_fn_table(struct patch *patch)
3044 struct string_list_item *item;
3047 * Always add new_name unless patch is a deletion
3048 * This should cover the cases for normal diffs,
3049 * file creations and copies
3051 if (patch->new_name != NULL) {
3052 item = string_list_insert(&fn_table, patch->new_name);
3053 item->util = patch;
3057 * store a failure on rename/deletion cases because
3058 * later chunks shouldn't patch old names
3060 if ((patch->new_name == NULL) || (patch->is_rename)) {
3061 item = string_list_insert(&fn_table, patch->old_name);
3062 item->util = PATH_WAS_DELETED;
3066 static void prepare_fn_table(struct patch *patch)
3069 * store information about incoming file deletion
3071 while (patch) {
3072 if ((patch->new_name == NULL) || (patch->is_rename)) {
3073 struct string_list_item *item;
3074 item = string_list_insert(&fn_table, patch->old_name);
3075 item->util = PATH_TO_BE_DELETED;
3077 patch = patch->next;
3081 static int checkout_target(struct cache_entry *ce, struct stat *st)
3083 struct checkout costate;
3085 memset(&costate, 0, sizeof(costate));
3086 costate.base_dir = "";
3087 costate.refresh_cache = 1;
3088 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3089 return error(_("cannot checkout %s"), ce->name);
3090 return 0;
3093 static struct patch *previous_patch(struct patch *patch, int *gone)
3095 struct patch *previous;
3097 *gone = 0;
3098 if (patch->is_copy || patch->is_rename)
3099 return NULL; /* "git" patches do not depend on the order */
3101 previous = in_fn_table(patch->old_name);
3102 if (!previous)
3103 return NULL;
3105 if (to_be_deleted(previous))
3106 return NULL; /* the deletion hasn't happened yet */
3108 if (was_deleted(previous))
3109 *gone = 1;
3111 return previous;
3114 static int verify_index_match(struct cache_entry *ce, struct stat *st)
3116 if (S_ISGITLINK(ce->ce_mode)) {
3117 if (!S_ISDIR(st->st_mode))
3118 return -1;
3119 return 0;
3121 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3124 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3126 static int load_patch_target(struct strbuf *buf,
3127 struct cache_entry *ce,
3128 struct stat *st,
3129 const char *name,
3130 unsigned expected_mode)
3132 if (cached) {
3133 if (read_file_or_gitlink(ce, buf))
3134 return error(_("read of %s failed"), name);
3135 } else if (name) {
3136 if (S_ISGITLINK(expected_mode)) {
3137 if (ce)
3138 return read_file_or_gitlink(ce, buf);
3139 else
3140 return SUBMODULE_PATCH_WITHOUT_INDEX;
3141 } else {
3142 if (read_old_data(st, name, buf))
3143 return error(_("read of %s failed"), name);
3146 return 0;
3150 * We are about to apply "patch"; populate the "image" with the
3151 * current version we have, from the working tree or from the index,
3152 * depending on the situation e.g. --cached/--index. If we are
3153 * applying a non-git patch that incrementally updates the tree,
3154 * we read from the result of a previous diff.
3156 static int load_preimage(struct image *image,
3157 struct patch *patch, struct stat *st, struct cache_entry *ce)
3159 struct strbuf buf = STRBUF_INIT;
3160 size_t len;
3161 char *img;
3162 struct patch *previous;
3163 int status;
3165 previous = previous_patch(patch, &status);
3166 if (status)
3167 return error(_("path %s has been renamed/deleted"),
3168 patch->old_name);
3169 if (previous) {
3170 /* We have a patched copy in memory; use that. */
3171 strbuf_add(&buf, previous->result, previous->resultsize);
3172 } else {
3173 status = load_patch_target(&buf, ce, st,
3174 patch->old_name, patch->old_mode);
3175 if (status < 0)
3176 return status;
3177 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3179 * There is no way to apply subproject
3180 * patch without looking at the index.
3181 * NEEDSWORK: shouldn't this be flagged
3182 * as an error???
3184 free_fragment_list(patch->fragments);
3185 patch->fragments = NULL;
3186 } else if (status) {
3187 return error(_("read of %s failed"), patch->old_name);
3191 img = strbuf_detach(&buf, &len);
3192 prepare_image(image, img, len, !patch->is_binary);
3193 return 0;
3196 static int three_way_merge(struct image *image,
3197 char *path,
3198 const unsigned char *base,
3199 const unsigned char *ours,
3200 const unsigned char *theirs)
3202 mmfile_t base_file, our_file, their_file;
3203 mmbuffer_t result = { NULL };
3204 int status;
3206 read_mmblob(&base_file, base);
3207 read_mmblob(&our_file, ours);
3208 read_mmblob(&their_file, theirs);
3209 status = ll_merge(&result, path,
3210 &base_file, "base",
3211 &our_file, "ours",
3212 &their_file, "theirs", NULL);
3213 free(base_file.ptr);
3214 free(our_file.ptr);
3215 free(their_file.ptr);
3216 if (status < 0 || !result.ptr) {
3217 free(result.ptr);
3218 return -1;
3220 clear_image(image);
3221 image->buf = result.ptr;
3222 image->len = result.size;
3224 return status;
3228 * When directly falling back to add/add three-way merge, we read from
3229 * the current contents of the new_name. In no cases other than that
3230 * this function will be called.
3232 static int load_current(struct image *image, struct patch *patch)
3234 struct strbuf buf = STRBUF_INIT;
3235 int status, pos;
3236 size_t len;
3237 char *img;
3238 struct stat st;
3239 struct cache_entry *ce;
3240 char *name = patch->new_name;
3241 unsigned mode = patch->new_mode;
3243 if (!patch->is_new)
3244 die("BUG: patch to %s is not a creation", patch->old_name);
3246 pos = cache_name_pos(name, strlen(name));
3247 if (pos < 0)
3248 return error(_("%s: does not exist in index"), name);
3249 ce = active_cache[pos];
3250 if (lstat(name, &st)) {
3251 if (errno != ENOENT)
3252 return error(_("%s: %s"), name, strerror(errno));
3253 if (checkout_target(ce, &st))
3254 return -1;
3256 if (verify_index_match(ce, &st))
3257 return error(_("%s: does not match index"), name);
3259 status = load_patch_target(&buf, ce, &st, name, mode);
3260 if (status < 0)
3261 return status;
3262 else if (status)
3263 return -1;
3264 img = strbuf_detach(&buf, &len);
3265 prepare_image(image, img, len, !patch->is_binary);
3266 return 0;
3269 static int try_threeway(struct image *image, struct patch *patch,
3270 struct stat *st, struct cache_entry *ce)
3272 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3273 struct strbuf buf = STRBUF_INIT;
3274 size_t len;
3275 int status;
3276 char *img;
3277 struct image tmp_image;
3279 /* No point falling back to 3-way merge in these cases */
3280 if (patch->is_delete ||
3281 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3282 return -1;
3284 /* Preimage the patch was prepared for */
3285 if (patch->is_new)
3286 write_sha1_file("", 0, blob_type, pre_sha1);
3287 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3288 read_blob_object(&buf, pre_sha1, patch->old_mode))
3289 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3291 fprintf(stderr, "Falling back to three-way merge...\n");
3293 img = strbuf_detach(&buf, &len);
3294 prepare_image(&tmp_image, img, len, 1);
3295 /* Apply the patch to get the post image */
3296 if (apply_fragments(&tmp_image, patch) < 0) {
3297 clear_image(&tmp_image);
3298 return -1;
3300 /* post_sha1[] is theirs */
3301 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3302 clear_image(&tmp_image);
3304 /* our_sha1[] is ours */
3305 if (patch->is_new) {
3306 if (load_current(&tmp_image, patch))
3307 return error("cannot read the current contents of '%s'",
3308 patch->new_name);
3309 } else {
3310 if (load_preimage(&tmp_image, patch, st, ce))
3311 return error("cannot read the current contents of '%s'",
3312 patch->old_name);
3314 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3315 clear_image(&tmp_image);
3317 /* in-core three-way merge between post and our using pre as base */
3318 status = three_way_merge(image, patch->new_name,
3319 pre_sha1, our_sha1, post_sha1);
3320 if (status < 0) {
3321 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3322 return status;
3325 if (status) {
3326 patch->conflicted_threeway = 1;
3327 if (patch->is_new)
3328 hashclr(patch->threeway_stage[0]);
3329 else
3330 hashcpy(patch->threeway_stage[0], pre_sha1);
3331 hashcpy(patch->threeway_stage[1], our_sha1);
3332 hashcpy(patch->threeway_stage[2], post_sha1);
3333 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3334 } else {
3335 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3337 return 0;
3340 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
3342 struct image image;
3344 if (load_preimage(&image, patch, st, ce) < 0)
3345 return -1;
3347 if (patch->direct_to_threeway ||
3348 apply_fragments(&image, patch) < 0) {
3349 /* Note: with --reject, apply_fragments() returns 0 */
3350 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3351 return -1;
3353 patch->result = image.buf;
3354 patch->resultsize = image.len;
3355 add_to_fn_table(patch);
3356 free(image.line_allocated);
3358 if (0 < patch->is_delete && patch->resultsize)
3359 return error(_("removal patch leaves file contents"));
3361 return 0;
3365 * If "patch" that we are looking at modifies or deletes what we have,
3366 * we would want it not to lose any local modification we have, either
3367 * in the working tree or in the index.
3369 * This also decides if a non-git patch is a creation patch or a
3370 * modification to an existing empty file. We do not check the state
3371 * of the current tree for a creation patch in this function; the caller
3372 * check_patch() separately makes sure (and errors out otherwise) that
3373 * the path the patch creates does not exist in the current tree.
3375 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3377 const char *old_name = patch->old_name;
3378 struct patch *previous = NULL;
3379 int stat_ret = 0, status;
3380 unsigned st_mode = 0;
3382 if (!old_name)
3383 return 0;
3385 assert(patch->is_new <= 0);
3386 previous = previous_patch(patch, &status);
3388 if (status)
3389 return error(_("path %s has been renamed/deleted"), old_name);
3390 if (previous) {
3391 st_mode = previous->new_mode;
3392 } else if (!cached) {
3393 stat_ret = lstat(old_name, st);
3394 if (stat_ret && errno != ENOENT)
3395 return error(_("%s: %s"), old_name, strerror(errno));
3398 if (check_index && !previous) {
3399 int pos = cache_name_pos(old_name, strlen(old_name));
3400 if (pos < 0) {
3401 if (patch->is_new < 0)
3402 goto is_new;
3403 return error(_("%s: does not exist in index"), old_name);
3405 *ce = active_cache[pos];
3406 if (stat_ret < 0) {
3407 if (checkout_target(*ce, st))
3408 return -1;
3410 if (!cached && verify_index_match(*ce, st))
3411 return error(_("%s: does not match index"), old_name);
3412 if (cached)
3413 st_mode = (*ce)->ce_mode;
3414 } else if (stat_ret < 0) {
3415 if (patch->is_new < 0)
3416 goto is_new;
3417 return error(_("%s: %s"), old_name, strerror(errno));
3420 if (!cached && !previous)
3421 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3423 if (patch->is_new < 0)
3424 patch->is_new = 0;
3425 if (!patch->old_mode)
3426 patch->old_mode = st_mode;
3427 if ((st_mode ^ patch->old_mode) & S_IFMT)
3428 return error(_("%s: wrong type"), old_name);
3429 if (st_mode != patch->old_mode)
3430 warning(_("%s has type %o, expected %o"),
3431 old_name, st_mode, patch->old_mode);
3432 if (!patch->new_mode && !patch->is_delete)
3433 patch->new_mode = st_mode;
3434 return 0;
3436 is_new:
3437 patch->is_new = 1;
3438 patch->is_delete = 0;
3439 free(patch->old_name);
3440 patch->old_name = NULL;
3441 return 0;
3445 #define EXISTS_IN_INDEX 1
3446 #define EXISTS_IN_WORKTREE 2
3448 static int check_to_create(const char *new_name, int ok_if_exists)
3450 struct stat nst;
3452 if (check_index &&
3453 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3454 !ok_if_exists)
3455 return EXISTS_IN_INDEX;
3456 if (cached)
3457 return 0;
3459 if (!lstat(new_name, &nst)) {
3460 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3461 return 0;
3463 * A leading component of new_name might be a symlink
3464 * that is going to be removed with this patch, but
3465 * still pointing at somewhere that has the path.
3466 * In such a case, path "new_name" does not exist as
3467 * far as git is concerned.
3469 if (has_symlink_leading_path(new_name, strlen(new_name)))
3470 return 0;
3472 return EXISTS_IN_WORKTREE;
3473 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3474 return error("%s: %s", new_name, strerror(errno));
3476 return 0;
3480 * Check and apply the patch in-core; leave the result in patch->result
3481 * for the caller to write it out to the final destination.
3483 static int check_patch(struct patch *patch)
3485 struct stat st;
3486 const char *old_name = patch->old_name;
3487 const char *new_name = patch->new_name;
3488 const char *name = old_name ? old_name : new_name;
3489 struct cache_entry *ce = NULL;
3490 struct patch *tpatch;
3491 int ok_if_exists;
3492 int status;
3494 patch->rejected = 1; /* we will drop this after we succeed */
3496 status = check_preimage(patch, &ce, &st);
3497 if (status)
3498 return status;
3499 old_name = patch->old_name;
3502 * A type-change diff is always split into a patch to delete
3503 * old, immediately followed by a patch to create new (see
3504 * diff.c::run_diff()); in such a case it is Ok that the entry
3505 * to be deleted by the previous patch is still in the working
3506 * tree and in the index.
3508 * A patch to swap-rename between A and B would first rename A
3509 * to B and then rename B to A. While applying the first one,
3510 * the presense of B should not stop A from getting renamed to
3511 * B; ask to_be_deleted() about the later rename. Removal of
3512 * B and rename from A to B is handled the same way by asking
3513 * was_deleted().
3515 if ((tpatch = in_fn_table(new_name)) &&
3516 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3517 ok_if_exists = 1;
3518 else
3519 ok_if_exists = 0;
3521 if (new_name &&
3522 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3523 int err = check_to_create(new_name, ok_if_exists);
3525 if (err && threeway) {
3526 patch->direct_to_threeway = 1;
3527 } else switch (err) {
3528 case 0:
3529 break; /* happy */
3530 case EXISTS_IN_INDEX:
3531 return error(_("%s: already exists in index"), new_name);
3532 break;
3533 case EXISTS_IN_WORKTREE:
3534 return error(_("%s: already exists in working directory"),
3535 new_name);
3536 default:
3537 return err;
3540 if (!patch->new_mode) {
3541 if (0 < patch->is_new)
3542 patch->new_mode = S_IFREG | 0644;
3543 else
3544 patch->new_mode = patch->old_mode;
3548 if (new_name && old_name) {
3549 int same = !strcmp(old_name, new_name);
3550 if (!patch->new_mode)
3551 patch->new_mode = patch->old_mode;
3552 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3553 if (same)
3554 return error(_("new mode (%o) of %s does not "
3555 "match old mode (%o)"),
3556 patch->new_mode, new_name,
3557 patch->old_mode);
3558 else
3559 return error(_("new mode (%o) of %s does not "
3560 "match old mode (%o) of %s"),
3561 patch->new_mode, new_name,
3562 patch->old_mode, old_name);
3566 if (apply_data(patch, &st, ce) < 0)
3567 return error(_("%s: patch does not apply"), name);
3568 patch->rejected = 0;
3569 return 0;
3572 static int check_patch_list(struct patch *patch)
3574 int err = 0;
3576 prepare_fn_table(patch);
3577 while (patch) {
3578 if (apply_verbosely)
3579 say_patch_name(stderr,
3580 _("Checking patch %s..."), patch);
3581 err |= check_patch(patch);
3582 patch = patch->next;
3584 return err;
3587 /* This function tries to read the sha1 from the current index */
3588 static int get_current_sha1(const char *path, unsigned char *sha1)
3590 int pos;
3592 if (read_cache() < 0)
3593 return -1;
3594 pos = cache_name_pos(path, strlen(path));
3595 if (pos < 0)
3596 return -1;
3597 hashcpy(sha1, active_cache[pos]->sha1);
3598 return 0;
3601 /* Build an index that contains the just the files needed for a 3way merge */
3602 static void build_fake_ancestor(struct patch *list, const char *filename)
3604 struct patch *patch;
3605 struct index_state result = { NULL };
3606 int fd;
3608 /* Once we start supporting the reverse patch, it may be
3609 * worth showing the new sha1 prefix, but until then...
3611 for (patch = list; patch; patch = patch->next) {
3612 const unsigned char *sha1_ptr;
3613 unsigned char sha1[20];
3614 struct cache_entry *ce;
3615 const char *name;
3617 name = patch->old_name ? patch->old_name : patch->new_name;
3618 if (0 < patch->is_new)
3619 continue;
3620 else if (get_sha1_blob(patch->old_sha1_prefix, sha1))
3621 /* git diff has no index line for mode/type changes */
3622 if (!patch->lines_added && !patch->lines_deleted) {
3623 if (get_current_sha1(patch->old_name, sha1))
3624 die("mode change for %s, which is not "
3625 "in current HEAD", name);
3626 sha1_ptr = sha1;
3627 } else
3628 die("sha1 information is lacking or useless "
3629 "(%s).", name);
3630 else
3631 sha1_ptr = sha1;
3633 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3634 if (!ce)
3635 die(_("make_cache_entry failed for path '%s'"), name);
3636 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3637 die ("Could not add %s to temporary index", name);
3640 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3641 if (fd < 0 || write_index(&result, fd) || close(fd))
3642 die ("Could not write temporary index to %s", filename);
3644 discard_index(&result);
3647 static void stat_patch_list(struct patch *patch)
3649 int files, adds, dels;
3651 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3652 files++;
3653 adds += patch->lines_added;
3654 dels += patch->lines_deleted;
3655 show_stats(patch);
3658 print_stat_summary(stdout, files, adds, dels);
3661 static void numstat_patch_list(struct patch *patch)
3663 for ( ; patch; patch = patch->next) {
3664 const char *name;
3665 name = patch->new_name ? patch->new_name : patch->old_name;
3666 if (patch->is_binary)
3667 printf("-\t-\t");
3668 else
3669 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3670 write_name_quoted(name, stdout, line_termination);
3674 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3676 if (mode)
3677 printf(" %s mode %06o %s\n", newdelete, mode, name);
3678 else
3679 printf(" %s %s\n", newdelete, name);
3682 static void show_mode_change(struct patch *p, int show_name)
3684 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3685 if (show_name)
3686 printf(" mode change %06o => %06o %s\n",
3687 p->old_mode, p->new_mode, p->new_name);
3688 else
3689 printf(" mode change %06o => %06o\n",
3690 p->old_mode, p->new_mode);
3694 static void show_rename_copy(struct patch *p)
3696 const char *renamecopy = p->is_rename ? "rename" : "copy";
3697 const char *old, *new;
3699 /* Find common prefix */
3700 old = p->old_name;
3701 new = p->new_name;
3702 while (1) {
3703 const char *slash_old, *slash_new;
3704 slash_old = strchr(old, '/');
3705 slash_new = strchr(new, '/');
3706 if (!slash_old ||
3707 !slash_new ||
3708 slash_old - old != slash_new - new ||
3709 memcmp(old, new, slash_new - new))
3710 break;
3711 old = slash_old + 1;
3712 new = slash_new + 1;
3714 /* p->old_name thru old is the common prefix, and old and new
3715 * through the end of names are renames
3717 if (old != p->old_name)
3718 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3719 (int)(old - p->old_name), p->old_name,
3720 old, new, p->score);
3721 else
3722 printf(" %s %s => %s (%d%%)\n", renamecopy,
3723 p->old_name, p->new_name, p->score);
3724 show_mode_change(p, 0);
3727 static void summary_patch_list(struct patch *patch)
3729 struct patch *p;
3731 for (p = patch; p; p = p->next) {
3732 if (p->is_new)
3733 show_file_mode_name("create", p->new_mode, p->new_name);
3734 else if (p->is_delete)
3735 show_file_mode_name("delete", p->old_mode, p->old_name);
3736 else {
3737 if (p->is_rename || p->is_copy)
3738 show_rename_copy(p);
3739 else {
3740 if (p->score) {
3741 printf(" rewrite %s (%d%%)\n",
3742 p->new_name, p->score);
3743 show_mode_change(p, 0);
3745 else
3746 show_mode_change(p, 1);
3752 static void patch_stats(struct patch *patch)
3754 int lines = patch->lines_added + patch->lines_deleted;
3756 if (lines > max_change)
3757 max_change = lines;
3758 if (patch->old_name) {
3759 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3760 if (!len)
3761 len = strlen(patch->old_name);
3762 if (len > max_len)
3763 max_len = len;
3765 if (patch->new_name) {
3766 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3767 if (!len)
3768 len = strlen(patch->new_name);
3769 if (len > max_len)
3770 max_len = len;
3774 static void remove_file(struct patch *patch, int rmdir_empty)
3776 if (update_index) {
3777 if (remove_file_from_cache(patch->old_name) < 0)
3778 die(_("unable to remove %s from index"), patch->old_name);
3780 if (!cached) {
3781 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3782 remove_path(patch->old_name);
3787 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3789 struct stat st;
3790 struct cache_entry *ce;
3791 int namelen = strlen(path);
3792 unsigned ce_size = cache_entry_size(namelen);
3794 if (!update_index)
3795 return;
3797 ce = xcalloc(1, ce_size);
3798 memcpy(ce->name, path, namelen);
3799 ce->ce_mode = create_ce_mode(mode);
3800 ce->ce_flags = create_ce_flags(0);
3801 ce->ce_namelen = namelen;
3802 if (S_ISGITLINK(mode)) {
3803 const char *s = buf;
3805 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3806 die(_("corrupt patch for subproject %s"), path);
3807 } else {
3808 if (!cached) {
3809 if (lstat(path, &st) < 0)
3810 die_errno(_("unable to stat newly created file '%s'"),
3811 path);
3812 fill_stat_cache_info(ce, &st);
3814 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3815 die(_("unable to create backing store for newly created file %s"), path);
3817 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3818 die(_("unable to add cache entry for %s"), path);
3821 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3823 int fd;
3824 struct strbuf nbuf = STRBUF_INIT;
3826 if (S_ISGITLINK(mode)) {
3827 struct stat st;
3828 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3829 return 0;
3830 return mkdir(path, 0777);
3833 if (has_symlinks && S_ISLNK(mode))
3834 /* Although buf:size is counted string, it also is NUL
3835 * terminated.
3837 return symlink(buf, path);
3839 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3840 if (fd < 0)
3841 return -1;
3843 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3844 size = nbuf.len;
3845 buf = nbuf.buf;
3847 write_or_die(fd, buf, size);
3848 strbuf_release(&nbuf);
3850 if (close(fd) < 0)
3851 die_errno(_("closing file '%s'"), path);
3852 return 0;
3856 * We optimistically assume that the directories exist,
3857 * which is true 99% of the time anyway. If they don't,
3858 * we create them and try again.
3860 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3862 if (cached)
3863 return;
3864 if (!try_create_file(path, mode, buf, size))
3865 return;
3867 if (errno == ENOENT) {
3868 if (safe_create_leading_directories(path))
3869 return;
3870 if (!try_create_file(path, mode, buf, size))
3871 return;
3874 if (errno == EEXIST || errno == EACCES) {
3875 /* We may be trying to create a file where a directory
3876 * used to be.
3878 struct stat st;
3879 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3880 errno = EEXIST;
3883 if (errno == EEXIST) {
3884 unsigned int nr = getpid();
3886 for (;;) {
3887 char newpath[PATH_MAX];
3888 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3889 if (!try_create_file(newpath, mode, buf, size)) {
3890 if (!rename(newpath, path))
3891 return;
3892 unlink_or_warn(newpath);
3893 break;
3895 if (errno != EEXIST)
3896 break;
3897 ++nr;
3900 die_errno(_("unable to write file '%s' mode %o"), path, mode);
3903 static void add_conflicted_stages_file(struct patch *patch)
3905 int stage, namelen;
3906 unsigned ce_size, mode;
3907 struct cache_entry *ce;
3909 if (!update_index)
3910 return;
3911 namelen = strlen(patch->new_name);
3912 ce_size = cache_entry_size(namelen);
3913 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
3915 remove_file_from_cache(patch->new_name);
3916 for (stage = 1; stage < 4; stage++) {
3917 if (is_null_sha1(patch->threeway_stage[stage - 1]))
3918 continue;
3919 ce = xcalloc(1, ce_size);
3920 memcpy(ce->name, patch->new_name, namelen);
3921 ce->ce_mode = create_ce_mode(mode);
3922 ce->ce_flags = create_ce_flags(stage);
3923 ce->ce_namelen = namelen;
3924 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
3925 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3926 die(_("unable to add cache entry for %s"), patch->new_name);
3930 static void create_file(struct patch *patch)
3932 char *path = patch->new_name;
3933 unsigned mode = patch->new_mode;
3934 unsigned long size = patch->resultsize;
3935 char *buf = patch->result;
3937 if (!mode)
3938 mode = S_IFREG | 0644;
3939 create_one_file(path, mode, buf, size);
3941 if (patch->conflicted_threeway)
3942 add_conflicted_stages_file(patch);
3943 else
3944 add_index_file(path, mode, buf, size);
3947 /* phase zero is to remove, phase one is to create */
3948 static void write_out_one_result(struct patch *patch, int phase)
3950 if (patch->is_delete > 0) {
3951 if (phase == 0)
3952 remove_file(patch, 1);
3953 return;
3955 if (patch->is_new > 0 || patch->is_copy) {
3956 if (phase == 1)
3957 create_file(patch);
3958 return;
3961 * Rename or modification boils down to the same
3962 * thing: remove the old, write the new
3964 if (phase == 0)
3965 remove_file(patch, patch->is_rename);
3966 if (phase == 1)
3967 create_file(patch);
3970 static int write_out_one_reject(struct patch *patch)
3972 FILE *rej;
3973 char namebuf[PATH_MAX];
3974 struct fragment *frag;
3975 int cnt = 0;
3976 struct strbuf sb = STRBUF_INIT;
3978 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3979 if (!frag->rejected)
3980 continue;
3981 cnt++;
3984 if (!cnt) {
3985 if (apply_verbosely)
3986 say_patch_name(stderr,
3987 _("Applied patch %s cleanly."), patch);
3988 return 0;
3991 /* This should not happen, because a removal patch that leaves
3992 * contents are marked "rejected" at the patch level.
3994 if (!patch->new_name)
3995 die(_("internal error"));
3997 /* Say this even without --verbose */
3998 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
3999 "Applying patch %%s with %d rejects...",
4000 cnt),
4001 cnt);
4002 say_patch_name(stderr, sb.buf, patch);
4003 strbuf_release(&sb);
4005 cnt = strlen(patch->new_name);
4006 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4007 cnt = ARRAY_SIZE(namebuf) - 5;
4008 warning(_("truncating .rej filename to %.*s.rej"),
4009 cnt - 1, patch->new_name);
4011 memcpy(namebuf, patch->new_name, cnt);
4012 memcpy(namebuf + cnt, ".rej", 5);
4014 rej = fopen(namebuf, "w");
4015 if (!rej)
4016 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4018 /* Normal git tools never deal with .rej, so do not pretend
4019 * this is a git patch by saying --git nor give extended
4020 * headers. While at it, maybe please "kompare" that wants
4021 * the trailing TAB and some garbage at the end of line ;-).
4023 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4024 patch->new_name, patch->new_name);
4025 for (cnt = 1, frag = patch->fragments;
4026 frag;
4027 cnt++, frag = frag->next) {
4028 if (!frag->rejected) {
4029 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4030 continue;
4032 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4033 fprintf(rej, "%.*s", frag->size, frag->patch);
4034 if (frag->patch[frag->size-1] != '\n')
4035 fputc('\n', rej);
4037 fclose(rej);
4038 return -1;
4041 static int write_out_results(struct patch *list)
4043 int phase;
4044 int errs = 0;
4045 struct patch *l;
4046 struct string_list cpath = STRING_LIST_INIT_DUP;
4048 for (phase = 0; phase < 2; phase++) {
4049 l = list;
4050 while (l) {
4051 if (l->rejected)
4052 errs = 1;
4053 else {
4054 write_out_one_result(l, phase);
4055 if (phase == 1) {
4056 if (write_out_one_reject(l))
4057 errs = 1;
4058 if (l->conflicted_threeway) {
4059 string_list_append(&cpath, l->new_name);
4060 errs = 1;
4064 l = l->next;
4068 if (cpath.nr) {
4069 struct string_list_item *item;
4071 sort_string_list(&cpath);
4072 for_each_string_list_item(item, &cpath)
4073 fprintf(stderr, "U %s\n", item->string);
4074 string_list_clear(&cpath, 0);
4076 rerere(0);
4079 return errs;
4082 static struct lock_file lock_file;
4084 static struct string_list limit_by_name;
4085 static int has_include;
4086 static void add_name_limit(const char *name, int exclude)
4088 struct string_list_item *it;
4090 it = string_list_append(&limit_by_name, name);
4091 it->util = exclude ? NULL : (void *) 1;
4094 static int use_patch(struct patch *p)
4096 const char *pathname = p->new_name ? p->new_name : p->old_name;
4097 int i;
4099 /* Paths outside are not touched regardless of "--include" */
4100 if (0 < prefix_length) {
4101 int pathlen = strlen(pathname);
4102 if (pathlen <= prefix_length ||
4103 memcmp(prefix, pathname, prefix_length))
4104 return 0;
4107 /* See if it matches any of exclude/include rule */
4108 for (i = 0; i < limit_by_name.nr; i++) {
4109 struct string_list_item *it = &limit_by_name.items[i];
4110 if (!fnmatch(it->string, pathname, 0))
4111 return (it->util != NULL);
4115 * If we had any include, a path that does not match any rule is
4116 * not used. Otherwise, we saw bunch of exclude rules (or none)
4117 * and such a path is used.
4119 return !has_include;
4123 static void prefix_one(char **name)
4125 char *old_name = *name;
4126 if (!old_name)
4127 return;
4128 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
4129 free(old_name);
4132 static void prefix_patches(struct patch *p)
4134 if (!prefix || p->is_toplevel_relative)
4135 return;
4136 for ( ; p; p = p->next) {
4137 prefix_one(&p->new_name);
4138 prefix_one(&p->old_name);
4142 #define INACCURATE_EOF (1<<0)
4143 #define RECOUNT (1<<1)
4145 static int apply_patch(int fd, const char *filename, int options)
4147 size_t offset;
4148 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4149 struct patch *list = NULL, **listp = &list;
4150 int skipped_patch = 0;
4152 patch_input_file = filename;
4153 read_patch_file(&buf, fd);
4154 offset = 0;
4155 while (offset < buf.len) {
4156 struct patch *patch;
4157 int nr;
4159 patch = xcalloc(1, sizeof(*patch));
4160 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4161 patch->recount = !!(options & RECOUNT);
4162 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4163 if (nr < 0)
4164 break;
4165 if (apply_in_reverse)
4166 reverse_patches(patch);
4167 if (prefix)
4168 prefix_patches(patch);
4169 if (use_patch(patch)) {
4170 patch_stats(patch);
4171 *listp = patch;
4172 listp = &patch->next;
4174 else {
4175 free_patch(patch);
4176 skipped_patch++;
4178 offset += nr;
4181 if (!list && !skipped_patch)
4182 die(_("unrecognized input"));
4184 if (whitespace_error && (ws_error_action == die_on_ws_error))
4185 apply = 0;
4187 update_index = check_index && apply;
4188 if (update_index && newfd < 0)
4189 newfd = hold_locked_index(&lock_file, 1);
4191 if (check_index) {
4192 if (read_cache() < 0)
4193 die(_("unable to read index file"));
4196 if ((check || apply) &&
4197 check_patch_list(list) < 0 &&
4198 !apply_with_reject)
4199 exit(1);
4201 if (apply && write_out_results(list)) {
4202 if (apply_with_reject)
4203 exit(1);
4204 /* with --3way, we still need to write the index out */
4205 return 1;
4208 if (fake_ancestor)
4209 build_fake_ancestor(list, fake_ancestor);
4211 if (diffstat)
4212 stat_patch_list(list);
4214 if (numstat)
4215 numstat_patch_list(list);
4217 if (summary)
4218 summary_patch_list(list);
4220 free_patch_list(list);
4221 strbuf_release(&buf);
4222 string_list_clear(&fn_table, 0);
4223 return 0;
4226 static int git_apply_config(const char *var, const char *value, void *cb)
4228 if (!strcmp(var, "apply.whitespace"))
4229 return git_config_string(&apply_default_whitespace, var, value);
4230 else if (!strcmp(var, "apply.ignorewhitespace"))
4231 return git_config_string(&apply_default_ignorewhitespace, var, value);
4232 return git_default_config(var, value, cb);
4235 static int option_parse_exclude(const struct option *opt,
4236 const char *arg, int unset)
4238 add_name_limit(arg, 1);
4239 return 0;
4242 static int option_parse_include(const struct option *opt,
4243 const char *arg, int unset)
4245 add_name_limit(arg, 0);
4246 has_include = 1;
4247 return 0;
4250 static int option_parse_p(const struct option *opt,
4251 const char *arg, int unset)
4253 p_value = atoi(arg);
4254 p_value_known = 1;
4255 return 0;
4258 static int option_parse_z(const struct option *opt,
4259 const char *arg, int unset)
4261 if (unset)
4262 line_termination = '\n';
4263 else
4264 line_termination = 0;
4265 return 0;
4268 static int option_parse_space_change(const struct option *opt,
4269 const char *arg, int unset)
4271 if (unset)
4272 ws_ignore_action = ignore_ws_none;
4273 else
4274 ws_ignore_action = ignore_ws_change;
4275 return 0;
4278 static int option_parse_whitespace(const struct option *opt,
4279 const char *arg, int unset)
4281 const char **whitespace_option = opt->value;
4283 *whitespace_option = arg;
4284 parse_whitespace_option(arg);
4285 return 0;
4288 static int option_parse_directory(const struct option *opt,
4289 const char *arg, int unset)
4291 root_len = strlen(arg);
4292 if (root_len && arg[root_len - 1] != '/') {
4293 char *new_root;
4294 root = new_root = xmalloc(root_len + 2);
4295 strcpy(new_root, arg);
4296 strcpy(new_root + root_len++, "/");
4297 } else
4298 root = arg;
4299 return 0;
4302 int cmd_apply(int argc, const char **argv, const char *prefix_)
4304 int i;
4305 int errs = 0;
4306 int is_not_gitdir = !startup_info->have_repository;
4307 int force_apply = 0;
4309 const char *whitespace_option = NULL;
4311 struct option builtin_apply_options[] = {
4312 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4313 N_("don't apply changes matching the given path"),
4314 0, option_parse_exclude },
4315 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4316 N_("apply changes matching the given path"),
4317 0, option_parse_include },
4318 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4319 N_("remove <num> leading slashes from traditional diff paths"),
4320 0, option_parse_p },
4321 OPT_BOOLEAN(0, "no-add", &no_add,
4322 N_("ignore additions made by the patch")),
4323 OPT_BOOLEAN(0, "stat", &diffstat,
4324 N_("instead of applying the patch, output diffstat for the input")),
4325 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4326 OPT_NOOP_NOARG(0, "binary"),
4327 OPT_BOOLEAN(0, "numstat", &numstat,
4328 N_("show number of added and deleted lines in decimal notation")),
4329 OPT_BOOLEAN(0, "summary", &summary,
4330 N_("instead of applying the patch, output a summary for the input")),
4331 OPT_BOOLEAN(0, "check", &check,
4332 N_("instead of applying the patch, see if the patch is applicable")),
4333 OPT_BOOLEAN(0, "index", &check_index,
4334 N_("make sure the patch is applicable to the current index")),
4335 OPT_BOOLEAN(0, "cached", &cached,
4336 N_("apply a patch without touching the working tree")),
4337 OPT_BOOLEAN(0, "apply", &force_apply,
4338 N_("also apply the patch (use with --stat/--summary/--check)")),
4339 OPT_BOOL('3', "3way", &threeway,
4340 N_( "attempt three-way merge if a patch does not apply")),
4341 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4342 N_("build a temporary index based on embedded index information")),
4343 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4344 N_("paths are separated with NUL character"),
4345 PARSE_OPT_NOARG, option_parse_z },
4346 OPT_INTEGER('C', NULL, &p_context,
4347 N_("ensure at least <n> lines of context match")),
4348 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4349 N_("detect new or modified lines that have whitespace errors"),
4350 0, option_parse_whitespace },
4351 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4352 N_("ignore changes in whitespace when finding context"),
4353 PARSE_OPT_NOARG, option_parse_space_change },
4354 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4355 N_("ignore changes in whitespace when finding context"),
4356 PARSE_OPT_NOARG, option_parse_space_change },
4357 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
4358 N_("apply the patch in reverse")),
4359 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
4360 N_("don't expect at least one line of context")),
4361 OPT_BOOLEAN(0, "reject", &apply_with_reject,
4362 N_("leave the rejected hunks in corresponding *.rej files")),
4363 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
4364 N_("allow overlapping hunks")),
4365 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4366 OPT_BIT(0, "inaccurate-eof", &options,
4367 N_("tolerate incorrectly detected missing new-line at the end of file"),
4368 INACCURATE_EOF),
4369 OPT_BIT(0, "recount", &options,
4370 N_("do not trust the line counts in the hunk headers"),
4371 RECOUNT),
4372 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4373 N_("prepend <root> to all filenames"),
4374 0, option_parse_directory },
4375 OPT_END()
4378 prefix = prefix_;
4379 prefix_length = prefix ? strlen(prefix) : 0;
4380 git_config(git_apply_config, NULL);
4381 if (apply_default_whitespace)
4382 parse_whitespace_option(apply_default_whitespace);
4383 if (apply_default_ignorewhitespace)
4384 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4386 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4387 apply_usage, 0);
4389 if (apply_with_reject && threeway)
4390 die("--reject and --3way cannot be used together.");
4391 if (cached && threeway)
4392 die("--cached and --3way cannot be used together.");
4393 if (threeway) {
4394 if (is_not_gitdir)
4395 die(_("--3way outside a repository"));
4396 check_index = 1;
4398 if (apply_with_reject)
4399 apply = apply_verbosely = 1;
4400 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4401 apply = 0;
4402 if (check_index && is_not_gitdir)
4403 die(_("--index outside a repository"));
4404 if (cached) {
4405 if (is_not_gitdir)
4406 die(_("--cached outside a repository"));
4407 check_index = 1;
4409 for (i = 0; i < argc; i++) {
4410 const char *arg = argv[i];
4411 int fd;
4413 if (!strcmp(arg, "-")) {
4414 errs |= apply_patch(0, "<stdin>", options);
4415 read_stdin = 0;
4416 continue;
4417 } else if (0 < prefix_length)
4418 arg = prefix_filename(prefix, prefix_length, arg);
4420 fd = open(arg, O_RDONLY);
4421 if (fd < 0)
4422 die_errno(_("can't open patch '%s'"), arg);
4423 read_stdin = 0;
4424 set_default_whitespace_mode(whitespace_option);
4425 errs |= apply_patch(fd, arg, options);
4426 close(fd);
4428 set_default_whitespace_mode(whitespace_option);
4429 if (read_stdin)
4430 errs |= apply_patch(0, "<stdin>", options);
4431 if (whitespace_error) {
4432 if (squelch_whitespace_errors &&
4433 squelch_whitespace_errors < whitespace_error) {
4434 int squelched =
4435 whitespace_error - squelch_whitespace_errors;
4436 warning(Q_("squelched %d whitespace error",
4437 "squelched %d whitespace errors",
4438 squelched),
4439 squelched);
4441 if (ws_error_action == die_on_ws_error)
4442 die(Q_("%d line adds whitespace errors.",
4443 "%d lines add whitespace errors.",
4444 whitespace_error),
4445 whitespace_error);
4446 if (applied_after_fixing_ws && apply)
4447 warning("%d line%s applied after"
4448 " fixing whitespace errors.",
4449 applied_after_fixing_ws,
4450 applied_after_fixing_ws == 1 ? "" : "s");
4451 else if (whitespace_error)
4452 warning(Q_("%d line adds whitespace errors.",
4453 "%d lines add whitespace errors.",
4454 whitespace_error),
4455 whitespace_error);
4458 if (update_index) {
4459 if (write_cache(newfd, active_cache, active_nr) ||
4460 commit_locked_index(&lock_file))
4461 die(_("Unable to write new index file"));
4464 return !!errs;