apply: do not touch a file beyond a symbolic link
[git.git] / builtin / apply.c
blobc2c6fda08c561f0b1b72aae7b252bc50ed211ae9
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 int unsafe_paths;
54 static const char *fake_ancestor;
55 static int line_termination = '\n';
56 static unsigned int p_context = UINT_MAX;
57 static const char * const apply_usage[] = {
58 N_("git apply [options] [<patch>...]"),
59 NULL
62 static enum ws_error_action {
63 nowarn_ws_error,
64 warn_on_ws_error,
65 die_on_ws_error,
66 correct_ws_error
67 } ws_error_action = warn_on_ws_error;
68 static int whitespace_error;
69 static int squelch_whitespace_errors = 5;
70 static int applied_after_fixing_ws;
72 static enum ws_ignore {
73 ignore_ws_none,
74 ignore_ws_change
75 } ws_ignore_action = ignore_ws_none;
78 static const char *patch_input_file;
79 static const char *root;
80 static int root_len;
81 static int read_stdin = 1;
82 static int options;
84 static void parse_whitespace_option(const char *option)
86 if (!option) {
87 ws_error_action = warn_on_ws_error;
88 return;
90 if (!strcmp(option, "warn")) {
91 ws_error_action = warn_on_ws_error;
92 return;
94 if (!strcmp(option, "nowarn")) {
95 ws_error_action = nowarn_ws_error;
96 return;
98 if (!strcmp(option, "error")) {
99 ws_error_action = die_on_ws_error;
100 return;
102 if (!strcmp(option, "error-all")) {
103 ws_error_action = die_on_ws_error;
104 squelch_whitespace_errors = 0;
105 return;
107 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
108 ws_error_action = correct_ws_error;
109 return;
111 die(_("unrecognized whitespace option '%s'"), option);
114 static void parse_ignorewhitespace_option(const char *option)
116 if (!option || !strcmp(option, "no") ||
117 !strcmp(option, "false") || !strcmp(option, "never") ||
118 !strcmp(option, "none")) {
119 ws_ignore_action = ignore_ws_none;
120 return;
122 if (!strcmp(option, "change")) {
123 ws_ignore_action = ignore_ws_change;
124 return;
126 die(_("unrecognized whitespace ignore option '%s'"), option);
129 static void set_default_whitespace_mode(const char *whitespace_option)
131 if (!whitespace_option && !apply_default_whitespace)
132 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
136 * For "diff-stat" like behaviour, we keep track of the biggest change
137 * we've seen, and the longest filename. That allows us to do simple
138 * scaling.
140 static int max_change, max_len;
143 * Various "current state", notably line numbers and what
144 * file (and how) we're patching right now.. The "is_xxxx"
145 * things are flags, where -1 means "don't know yet".
147 static int linenr = 1;
150 * This represents one "hunk" from a patch, starting with
151 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
152 * patch text is pointed at by patch, and its byte length
153 * is stored in size. leading and trailing are the number
154 * of context lines.
156 struct fragment {
157 unsigned long leading, trailing;
158 unsigned long oldpos, oldlines;
159 unsigned long newpos, newlines;
161 * 'patch' is usually borrowed from buf in apply_patch(),
162 * but some codepaths store an allocated buffer.
164 const char *patch;
165 unsigned free_patch:1,
166 rejected:1;
167 int size;
168 int linenr;
169 struct fragment *next;
173 * When dealing with a binary patch, we reuse "leading" field
174 * to store the type of the binary hunk, either deflated "delta"
175 * or deflated "literal".
177 #define binary_patch_method leading
178 #define BINARY_DELTA_DEFLATED 1
179 #define BINARY_LITERAL_DEFLATED 2
182 * This represents a "patch" to a file, both metainfo changes
183 * such as creation/deletion, filemode and content changes represented
184 * as a series of fragments.
186 struct patch {
187 char *new_name, *old_name, *def_name;
188 unsigned int old_mode, new_mode;
189 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
190 int rejected;
191 unsigned ws_rule;
192 int lines_added, lines_deleted;
193 int score;
194 unsigned int is_toplevel_relative:1;
195 unsigned int inaccurate_eof:1;
196 unsigned int is_binary:1;
197 unsigned int is_copy:1;
198 unsigned int is_rename:1;
199 unsigned int recount:1;
200 unsigned int conflicted_threeway:1;
201 unsigned int direct_to_threeway:1;
202 struct fragment *fragments;
203 char *result;
204 size_t resultsize;
205 char old_sha1_prefix[41];
206 char new_sha1_prefix[41];
207 struct patch *next;
209 /* three-way fallback result */
210 unsigned char threeway_stage[3][20];
213 static void free_fragment_list(struct fragment *list)
215 while (list) {
216 struct fragment *next = list->next;
217 if (list->free_patch)
218 free((char *)list->patch);
219 free(list);
220 list = next;
224 static void free_patch(struct patch *patch)
226 free_fragment_list(patch->fragments);
227 free(patch->def_name);
228 free(patch->old_name);
229 free(patch->new_name);
230 free(patch->result);
231 free(patch);
234 static void free_patch_list(struct patch *list)
236 while (list) {
237 struct patch *next = list->next;
238 free_patch(list);
239 list = next;
244 * A line in a file, len-bytes long (includes the terminating LF,
245 * except for an incomplete line at the end if the file ends with
246 * one), and its contents hashes to 'hash'.
248 struct line {
249 size_t len;
250 unsigned hash : 24;
251 unsigned flag : 8;
252 #define LINE_COMMON 1
253 #define LINE_PATCHED 2
257 * This represents a "file", which is an array of "lines".
259 struct image {
260 char *buf;
261 size_t len;
262 size_t nr;
263 size_t alloc;
264 struct line *line_allocated;
265 struct line *line;
269 * Records filenames that have been touched, in order to handle
270 * the case where more than one patches touch the same file.
273 static struct string_list fn_table;
275 static uint32_t hash_line(const char *cp, size_t len)
277 size_t i;
278 uint32_t h;
279 for (i = 0, h = 0; i < len; i++) {
280 if (!isspace(cp[i])) {
281 h = h * 3 + (cp[i] & 0xff);
284 return h;
288 * Compare lines s1 of length n1 and s2 of length n2, ignoring
289 * whitespace difference. Returns 1 if they match, 0 otherwise
291 static int fuzzy_matchlines(const char *s1, size_t n1,
292 const char *s2, size_t n2)
294 const char *last1 = s1 + n1 - 1;
295 const char *last2 = s2 + n2 - 1;
296 int result = 0;
298 /* ignore line endings */
299 while ((*last1 == '\r') || (*last1 == '\n'))
300 last1--;
301 while ((*last2 == '\r') || (*last2 == '\n'))
302 last2--;
304 /* skip leading whitespace */
305 while (isspace(*s1) && (s1 <= last1))
306 s1++;
307 while (isspace(*s2) && (s2 <= last2))
308 s2++;
309 /* early return if both lines are empty */
310 if ((s1 > last1) && (s2 > last2))
311 return 1;
312 while (!result) {
313 result = *s1++ - *s2++;
315 * Skip whitespace inside. We check for whitespace on
316 * both buffers because we don't want "a b" to match
317 * "ab"
319 if (isspace(*s1) && isspace(*s2)) {
320 while (isspace(*s1) && s1 <= last1)
321 s1++;
322 while (isspace(*s2) && s2 <= last2)
323 s2++;
326 * If we reached the end on one side only,
327 * lines don't match
329 if (
330 ((s2 > last2) && (s1 <= last1)) ||
331 ((s1 > last1) && (s2 <= last2)))
332 return 0;
333 if ((s1 > last1) && (s2 > last2))
334 break;
337 return !result;
340 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
342 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
343 img->line_allocated[img->nr].len = len;
344 img->line_allocated[img->nr].hash = hash_line(bol, len);
345 img->line_allocated[img->nr].flag = flag;
346 img->nr++;
350 * "buf" has the file contents to be patched (read from various sources).
351 * attach it to "image" and add line-based index to it.
352 * "image" now owns the "buf".
354 static void prepare_image(struct image *image, char *buf, size_t len,
355 int prepare_linetable)
357 const char *cp, *ep;
359 memset(image, 0, sizeof(*image));
360 image->buf = buf;
361 image->len = len;
363 if (!prepare_linetable)
364 return;
366 ep = image->buf + image->len;
367 cp = image->buf;
368 while (cp < ep) {
369 const char *next;
370 for (next = cp; next < ep && *next != '\n'; next++)
372 if (next < ep)
373 next++;
374 add_line_info(image, cp, next - cp, 0);
375 cp = next;
377 image->line = image->line_allocated;
380 static void clear_image(struct image *image)
382 free(image->buf);
383 free(image->line_allocated);
384 memset(image, 0, sizeof(*image));
387 /* fmt must contain _one_ %s and no other substitution */
388 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
390 struct strbuf sb = STRBUF_INIT;
392 if (patch->old_name && patch->new_name &&
393 strcmp(patch->old_name, patch->new_name)) {
394 quote_c_style(patch->old_name, &sb, NULL, 0);
395 strbuf_addstr(&sb, " => ");
396 quote_c_style(patch->new_name, &sb, NULL, 0);
397 } else {
398 const char *n = patch->new_name;
399 if (!n)
400 n = patch->old_name;
401 quote_c_style(n, &sb, NULL, 0);
403 fprintf(output, fmt, sb.buf);
404 fputc('\n', output);
405 strbuf_release(&sb);
408 #define SLOP (16)
410 static void read_patch_file(struct strbuf *sb, int fd)
412 if (strbuf_read(sb, fd, 0) < 0)
413 die_errno("git apply: failed to read");
416 * Make sure that we have some slop in the buffer
417 * so that we can do speculative "memcmp" etc, and
418 * see to it that it is NUL-filled.
420 strbuf_grow(sb, SLOP);
421 memset(sb->buf + sb->len, 0, SLOP);
424 static unsigned long linelen(const char *buffer, unsigned long size)
426 unsigned long len = 0;
427 while (size--) {
428 len++;
429 if (*buffer++ == '\n')
430 break;
432 return len;
435 static int is_dev_null(const char *str)
437 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
440 #define TERM_SPACE 1
441 #define TERM_TAB 2
443 static int name_terminate(const char *name, int namelen, int c, int terminate)
445 if (c == ' ' && !(terminate & TERM_SPACE))
446 return 0;
447 if (c == '\t' && !(terminate & TERM_TAB))
448 return 0;
450 return 1;
453 /* remove double slashes to make --index work with such filenames */
454 static char *squash_slash(char *name)
456 int i = 0, j = 0;
458 if (!name)
459 return NULL;
461 while (name[i]) {
462 if ((name[j++] = name[i++]) == '/')
463 while (name[i] == '/')
464 i++;
466 name[j] = '\0';
467 return name;
470 static char *find_name_gnu(const char *line, const char *def, int p_value)
472 struct strbuf name = STRBUF_INIT;
473 char *cp;
476 * Proposed "new-style" GNU patch/diff format; see
477 * http://marc.info/?l=git&m=112927316408690&w=2
479 if (unquote_c_style(&name, line, NULL)) {
480 strbuf_release(&name);
481 return NULL;
484 for (cp = name.buf; p_value; p_value--) {
485 cp = strchr(cp, '/');
486 if (!cp) {
487 strbuf_release(&name);
488 return NULL;
490 cp++;
493 strbuf_remove(&name, 0, cp - name.buf);
494 if (root)
495 strbuf_insert(&name, 0, root, root_len);
496 return squash_slash(strbuf_detach(&name, NULL));
499 static size_t sane_tz_len(const char *line, size_t len)
501 const char *tz, *p;
503 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
504 return 0;
505 tz = line + len - strlen(" +0500");
507 if (tz[1] != '+' && tz[1] != '-')
508 return 0;
510 for (p = tz + 2; p != line + len; p++)
511 if (!isdigit(*p))
512 return 0;
514 return line + len - tz;
517 static size_t tz_with_colon_len(const char *line, size_t len)
519 const char *tz, *p;
521 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
522 return 0;
523 tz = line + len - strlen(" +08:00");
525 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
526 return 0;
527 p = tz + 2;
528 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
529 !isdigit(*p++) || !isdigit(*p++))
530 return 0;
532 return line + len - tz;
535 static size_t date_len(const char *line, size_t len)
537 const char *date, *p;
539 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
540 return 0;
541 p = date = line + len - strlen("72-02-05");
543 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
544 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
545 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
546 return 0;
548 if (date - line >= strlen("19") &&
549 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
550 date -= strlen("19");
552 return line + len - date;
555 static size_t short_time_len(const char *line, size_t len)
557 const char *time, *p;
559 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
560 return 0;
561 p = time = line + len - strlen(" 07:01:32");
563 /* Permit 1-digit hours? */
564 if (*p++ != ' ' ||
565 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
566 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
567 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
568 return 0;
570 return line + len - time;
573 static size_t fractional_time_len(const char *line, size_t len)
575 const char *p;
576 size_t n;
578 /* Expected format: 19:41:17.620000023 */
579 if (!len || !isdigit(line[len - 1]))
580 return 0;
581 p = line + len - 1;
583 /* Fractional seconds. */
584 while (p > line && isdigit(*p))
585 p--;
586 if (*p != '.')
587 return 0;
589 /* Hours, minutes, and whole seconds. */
590 n = short_time_len(line, p - line);
591 if (!n)
592 return 0;
594 return line + len - p + n;
597 static size_t trailing_spaces_len(const char *line, size_t len)
599 const char *p;
601 /* Expected format: ' ' x (1 or more) */
602 if (!len || line[len - 1] != ' ')
603 return 0;
605 p = line + len;
606 while (p != line) {
607 p--;
608 if (*p != ' ')
609 return line + len - (p + 1);
612 /* All spaces! */
613 return len;
616 static size_t diff_timestamp_len(const char *line, size_t len)
618 const char *end = line + len;
619 size_t n;
622 * Posix: 2010-07-05 19:41:17
623 * GNU: 2010-07-05 19:41:17.620000023 -0500
626 if (!isdigit(end[-1]))
627 return 0;
629 n = sane_tz_len(line, end - line);
630 if (!n)
631 n = tz_with_colon_len(line, end - line);
632 end -= n;
634 n = short_time_len(line, end - line);
635 if (!n)
636 n = fractional_time_len(line, end - line);
637 end -= n;
639 n = date_len(line, end - line);
640 if (!n) /* No date. Too bad. */
641 return 0;
642 end -= n;
644 if (end == line) /* No space before date. */
645 return 0;
646 if (end[-1] == '\t') { /* Success! */
647 end--;
648 return line + len - end;
650 if (end[-1] != ' ') /* No space before date. */
651 return 0;
653 /* Whitespace damage. */
654 end -= trailing_spaces_len(line, end - line);
655 return line + len - end;
658 static char *null_strdup(const char *s)
660 return s ? xstrdup(s) : NULL;
663 static char *find_name_common(const char *line, const char *def,
664 int p_value, const char *end, int terminate)
666 int len;
667 const char *start = NULL;
669 if (p_value == 0)
670 start = line;
671 while (line != end) {
672 char c = *line;
674 if (!end && isspace(c)) {
675 if (c == '\n')
676 break;
677 if (name_terminate(start, line-start, c, terminate))
678 break;
680 line++;
681 if (c == '/' && !--p_value)
682 start = line;
684 if (!start)
685 return squash_slash(null_strdup(def));
686 len = line - start;
687 if (!len)
688 return squash_slash(null_strdup(def));
691 * Generally we prefer the shorter name, especially
692 * if the other one is just a variation of that with
693 * something else tacked on to the end (ie "file.orig"
694 * or "file~").
696 if (def) {
697 int deflen = strlen(def);
698 if (deflen < len && !strncmp(start, def, deflen))
699 return squash_slash(xstrdup(def));
702 if (root) {
703 char *ret = xmalloc(root_len + len + 1);
704 strcpy(ret, root);
705 memcpy(ret + root_len, start, len);
706 ret[root_len + len] = '\0';
707 return squash_slash(ret);
710 return squash_slash(xmemdupz(start, len));
713 static char *find_name(const char *line, char *def, int p_value, int terminate)
715 if (*line == '"') {
716 char *name = find_name_gnu(line, def, p_value);
717 if (name)
718 return name;
721 return find_name_common(line, def, p_value, NULL, terminate);
724 static char *find_name_traditional(const char *line, char *def, int p_value)
726 size_t len;
727 size_t date_len;
729 if (*line == '"') {
730 char *name = find_name_gnu(line, def, p_value);
731 if (name)
732 return name;
735 len = strchrnul(line, '\n') - line;
736 date_len = diff_timestamp_len(line, len);
737 if (!date_len)
738 return find_name_common(line, def, p_value, NULL, TERM_TAB);
739 len -= date_len;
741 return find_name_common(line, def, p_value, line + len, 0);
744 static int count_slashes(const char *cp)
746 int cnt = 0;
747 char ch;
749 while ((ch = *cp++))
750 if (ch == '/')
751 cnt++;
752 return cnt;
756 * Given the string after "--- " or "+++ ", guess the appropriate
757 * p_value for the given patch.
759 static int guess_p_value(const char *nameline)
761 char *name, *cp;
762 int val = -1;
764 if (is_dev_null(nameline))
765 return -1;
766 name = find_name_traditional(nameline, NULL, 0);
767 if (!name)
768 return -1;
769 cp = strchr(name, '/');
770 if (!cp)
771 val = 0;
772 else if (prefix) {
774 * Does it begin with "a/$our-prefix" and such? Then this is
775 * very likely to apply to our directory.
777 if (!strncmp(name, prefix, prefix_length))
778 val = count_slashes(prefix);
779 else {
780 cp++;
781 if (!strncmp(cp, prefix, prefix_length))
782 val = count_slashes(prefix) + 1;
785 free(name);
786 return val;
790 * Does the ---/+++ line has the POSIX timestamp after the last HT?
791 * GNU diff puts epoch there to signal a creation/deletion event. Is
792 * this such a timestamp?
794 static int has_epoch_timestamp(const char *nameline)
797 * We are only interested in epoch timestamp; any non-zero
798 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
799 * For the same reason, the date must be either 1969-12-31 or
800 * 1970-01-01, and the seconds part must be "00".
802 const char stamp_regexp[] =
803 "^(1969-12-31|1970-01-01)"
805 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
807 "([-+][0-2][0-9]:?[0-5][0-9])\n";
808 const char *timestamp = NULL, *cp, *colon;
809 static regex_t *stamp;
810 regmatch_t m[10];
811 int zoneoffset;
812 int hourminute;
813 int status;
815 for (cp = nameline; *cp != '\n'; cp++) {
816 if (*cp == '\t')
817 timestamp = cp + 1;
819 if (!timestamp)
820 return 0;
821 if (!stamp) {
822 stamp = xmalloc(sizeof(*stamp));
823 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
824 warning(_("Cannot prepare timestamp regexp %s"),
825 stamp_regexp);
826 return 0;
830 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
831 if (status) {
832 if (status != REG_NOMATCH)
833 warning(_("regexec returned %d for input: %s"),
834 status, timestamp);
835 return 0;
838 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
839 if (*colon == ':')
840 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
841 else
842 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
843 if (timestamp[m[3].rm_so] == '-')
844 zoneoffset = -zoneoffset;
847 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
848 * (west of GMT) or 1970-01-01 (east of GMT)
850 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
851 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
852 return 0;
854 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
855 strtol(timestamp + 14, NULL, 10) -
856 zoneoffset);
858 return ((zoneoffset < 0 && hourminute == 1440) ||
859 (0 <= zoneoffset && !hourminute));
863 * Get the name etc info from the ---/+++ lines of a traditional patch header
865 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
866 * files, we can happily check the index for a match, but for creating a
867 * new file we should try to match whatever "patch" does. I have no idea.
869 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
871 char *name;
873 first += 4; /* skip "--- " */
874 second += 4; /* skip "+++ " */
875 if (!p_value_known) {
876 int p, q;
877 p = guess_p_value(first);
878 q = guess_p_value(second);
879 if (p < 0) p = q;
880 if (0 <= p && p == q) {
881 p_value = p;
882 p_value_known = 1;
885 if (is_dev_null(first)) {
886 patch->is_new = 1;
887 patch->is_delete = 0;
888 name = find_name_traditional(second, NULL, p_value);
889 patch->new_name = name;
890 } else if (is_dev_null(second)) {
891 patch->is_new = 0;
892 patch->is_delete = 1;
893 name = find_name_traditional(first, NULL, p_value);
894 patch->old_name = name;
895 } else {
896 char *first_name;
897 first_name = find_name_traditional(first, NULL, p_value);
898 name = find_name_traditional(second, first_name, p_value);
899 free(first_name);
900 if (has_epoch_timestamp(first)) {
901 patch->is_new = 1;
902 patch->is_delete = 0;
903 patch->new_name = name;
904 } else if (has_epoch_timestamp(second)) {
905 patch->is_new = 0;
906 patch->is_delete = 1;
907 patch->old_name = name;
908 } else {
909 patch->old_name = name;
910 patch->new_name = null_strdup(name);
913 if (!name)
914 die(_("unable to find filename in patch at line %d"), linenr);
917 static int gitdiff_hdrend(const char *line, struct patch *patch)
919 return -1;
923 * We're anal about diff header consistency, to make
924 * sure that we don't end up having strange ambiguous
925 * patches floating around.
927 * As a result, gitdiff_{old|new}name() will check
928 * their names against any previous information, just
929 * to make sure..
931 #define DIFF_OLD_NAME 0
932 #define DIFF_NEW_NAME 1
934 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
936 if (!orig_name && !isnull)
937 return find_name(line, NULL, p_value, TERM_TAB);
939 if (orig_name) {
940 int len;
941 const char *name;
942 char *another;
943 name = orig_name;
944 len = strlen(name);
945 if (isnull)
946 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
947 another = find_name(line, NULL, p_value, TERM_TAB);
948 if (!another || memcmp(another, name, len + 1))
949 die((side == DIFF_NEW_NAME) ?
950 _("git apply: bad git-diff - inconsistent new filename on line %d") :
951 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
952 free(another);
953 return orig_name;
955 else {
956 /* expect "/dev/null" */
957 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
958 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
959 return NULL;
963 static int gitdiff_oldname(const char *line, struct patch *patch)
965 char *orig = patch->old_name;
966 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
967 DIFF_OLD_NAME);
968 if (orig != patch->old_name)
969 free(orig);
970 return 0;
973 static int gitdiff_newname(const char *line, struct patch *patch)
975 char *orig = patch->new_name;
976 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
977 DIFF_NEW_NAME);
978 if (orig != patch->new_name)
979 free(orig);
980 return 0;
983 static int gitdiff_oldmode(const char *line, struct patch *patch)
985 patch->old_mode = strtoul(line, NULL, 8);
986 return 0;
989 static int gitdiff_newmode(const char *line, struct patch *patch)
991 patch->new_mode = strtoul(line, NULL, 8);
992 return 0;
995 static int gitdiff_delete(const char *line, struct patch *patch)
997 patch->is_delete = 1;
998 free(patch->old_name);
999 patch->old_name = null_strdup(patch->def_name);
1000 return gitdiff_oldmode(line, patch);
1003 static int gitdiff_newfile(const char *line, struct patch *patch)
1005 patch->is_new = 1;
1006 free(patch->new_name);
1007 patch->new_name = null_strdup(patch->def_name);
1008 return gitdiff_newmode(line, patch);
1011 static int gitdiff_copysrc(const char *line, struct patch *patch)
1013 patch->is_copy = 1;
1014 free(patch->old_name);
1015 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1016 return 0;
1019 static int gitdiff_copydst(const char *line, struct patch *patch)
1021 patch->is_copy = 1;
1022 free(patch->new_name);
1023 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1024 return 0;
1027 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1029 patch->is_rename = 1;
1030 free(patch->old_name);
1031 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1032 return 0;
1035 static int gitdiff_renamedst(const char *line, struct patch *patch)
1037 patch->is_rename = 1;
1038 free(patch->new_name);
1039 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1040 return 0;
1043 static int gitdiff_similarity(const char *line, struct patch *patch)
1045 unsigned long val = strtoul(line, NULL, 10);
1046 if (val <= 100)
1047 patch->score = val;
1048 return 0;
1051 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1053 unsigned long val = strtoul(line, NULL, 10);
1054 if (val <= 100)
1055 patch->score = val;
1056 return 0;
1059 static int gitdiff_index(const char *line, struct patch *patch)
1062 * index line is N hexadecimal, "..", N hexadecimal,
1063 * and optional space with octal mode.
1065 const char *ptr, *eol;
1066 int len;
1068 ptr = strchr(line, '.');
1069 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1070 return 0;
1071 len = ptr - line;
1072 memcpy(patch->old_sha1_prefix, line, len);
1073 patch->old_sha1_prefix[len] = 0;
1075 line = ptr + 2;
1076 ptr = strchr(line, ' ');
1077 eol = strchr(line, '\n');
1079 if (!ptr || eol < ptr)
1080 ptr = eol;
1081 len = ptr - line;
1083 if (40 < len)
1084 return 0;
1085 memcpy(patch->new_sha1_prefix, line, len);
1086 patch->new_sha1_prefix[len] = 0;
1087 if (*ptr == ' ')
1088 patch->old_mode = strtoul(ptr+1, NULL, 8);
1089 return 0;
1093 * This is normal for a diff that doesn't change anything: we'll fall through
1094 * into the next diff. Tell the parser to break out.
1096 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1098 return -1;
1102 * Skip p_value leading components from "line"; as we do not accept
1103 * absolute paths, return NULL in that case.
1105 static const char *skip_tree_prefix(const char *line, int llen)
1107 int nslash;
1108 int i;
1110 if (!p_value)
1111 return (llen && line[0] == '/') ? NULL : line;
1113 nslash = p_value;
1114 for (i = 0; i < llen; i++) {
1115 int ch = line[i];
1116 if (ch == '/' && --nslash <= 0)
1117 return (i == 0) ? NULL : &line[i + 1];
1119 return NULL;
1123 * This is to extract the same name that appears on "diff --git"
1124 * line. We do not find and return anything if it is a rename
1125 * patch, and it is OK because we will find the name elsewhere.
1126 * We need to reliably find name only when it is mode-change only,
1127 * creation or deletion of an empty file. In any of these cases,
1128 * both sides are the same name under a/ and b/ respectively.
1130 static char *git_header_name(const char *line, int llen)
1132 const char *name;
1133 const char *second = NULL;
1134 size_t len, line_len;
1136 line += strlen("diff --git ");
1137 llen -= strlen("diff --git ");
1139 if (*line == '"') {
1140 const char *cp;
1141 struct strbuf first = STRBUF_INIT;
1142 struct strbuf sp = STRBUF_INIT;
1144 if (unquote_c_style(&first, line, &second))
1145 goto free_and_fail1;
1147 /* strip the a/b prefix including trailing slash */
1148 cp = skip_tree_prefix(first.buf, first.len);
1149 if (!cp)
1150 goto free_and_fail1;
1151 strbuf_remove(&first, 0, cp - first.buf);
1154 * second points at one past closing dq of name.
1155 * find the second name.
1157 while ((second < line + llen) && isspace(*second))
1158 second++;
1160 if (line + llen <= second)
1161 goto free_and_fail1;
1162 if (*second == '"') {
1163 if (unquote_c_style(&sp, second, NULL))
1164 goto free_and_fail1;
1165 cp = skip_tree_prefix(sp.buf, sp.len);
1166 if (!cp)
1167 goto free_and_fail1;
1168 /* They must match, otherwise ignore */
1169 if (strcmp(cp, first.buf))
1170 goto free_and_fail1;
1171 strbuf_release(&sp);
1172 return strbuf_detach(&first, NULL);
1175 /* unquoted second */
1176 cp = skip_tree_prefix(second, line + llen - second);
1177 if (!cp)
1178 goto free_and_fail1;
1179 if (line + llen - cp != first.len ||
1180 memcmp(first.buf, cp, first.len))
1181 goto free_and_fail1;
1182 return strbuf_detach(&first, NULL);
1184 free_and_fail1:
1185 strbuf_release(&first);
1186 strbuf_release(&sp);
1187 return NULL;
1190 /* unquoted first name */
1191 name = skip_tree_prefix(line, llen);
1192 if (!name)
1193 return NULL;
1196 * since the first name is unquoted, a dq if exists must be
1197 * the beginning of the second name.
1199 for (second = name; second < line + llen; second++) {
1200 if (*second == '"') {
1201 struct strbuf sp = STRBUF_INIT;
1202 const char *np;
1204 if (unquote_c_style(&sp, second, NULL))
1205 goto free_and_fail2;
1207 np = skip_tree_prefix(sp.buf, sp.len);
1208 if (!np)
1209 goto free_and_fail2;
1211 len = sp.buf + sp.len - np;
1212 if (len < second - name &&
1213 !strncmp(np, name, len) &&
1214 isspace(name[len])) {
1215 /* Good */
1216 strbuf_remove(&sp, 0, np - sp.buf);
1217 return strbuf_detach(&sp, NULL);
1220 free_and_fail2:
1221 strbuf_release(&sp);
1222 return NULL;
1227 * Accept a name only if it shows up twice, exactly the same
1228 * form.
1230 second = strchr(name, '\n');
1231 if (!second)
1232 return NULL;
1233 line_len = second - name;
1234 for (len = 0 ; ; len++) {
1235 switch (name[len]) {
1236 default:
1237 continue;
1238 case '\n':
1239 return NULL;
1240 case '\t': case ' ':
1242 * Is this the separator between the preimage
1243 * and the postimage pathname? Again, we are
1244 * only interested in the case where there is
1245 * no rename, as this is only to set def_name
1246 * and a rename patch has the names elsewhere
1247 * in an unambiguous form.
1249 if (!name[len + 1])
1250 return NULL; /* no postimage name */
1251 second = skip_tree_prefix(name + len + 1,
1252 line_len - (len + 1));
1253 if (!second)
1254 return NULL;
1256 * Does len bytes starting at "name" and "second"
1257 * (that are separated by one HT or SP we just
1258 * found) exactly match?
1260 if (second[len] == '\n' && !strncmp(name, second, len))
1261 return xmemdupz(name, len);
1266 /* Verify that we recognize the lines following a git header */
1267 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1269 unsigned long offset;
1271 /* A git diff has explicit new/delete information, so we don't guess */
1272 patch->is_new = 0;
1273 patch->is_delete = 0;
1276 * Some things may not have the old name in the
1277 * rest of the headers anywhere (pure mode changes,
1278 * or removing or adding empty files), so we get
1279 * the default name from the header.
1281 patch->def_name = git_header_name(line, len);
1282 if (patch->def_name && root) {
1283 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1284 strcpy(s, root);
1285 strcpy(s + root_len, patch->def_name);
1286 free(patch->def_name);
1287 patch->def_name = s;
1290 line += len;
1291 size -= len;
1292 linenr++;
1293 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1294 static const struct opentry {
1295 const char *str;
1296 int (*fn)(const char *, struct patch *);
1297 } optable[] = {
1298 { "@@ -", gitdiff_hdrend },
1299 { "--- ", gitdiff_oldname },
1300 { "+++ ", gitdiff_newname },
1301 { "old mode ", gitdiff_oldmode },
1302 { "new mode ", gitdiff_newmode },
1303 { "deleted file mode ", gitdiff_delete },
1304 { "new file mode ", gitdiff_newfile },
1305 { "copy from ", gitdiff_copysrc },
1306 { "copy to ", gitdiff_copydst },
1307 { "rename old ", gitdiff_renamesrc },
1308 { "rename new ", gitdiff_renamedst },
1309 { "rename from ", gitdiff_renamesrc },
1310 { "rename to ", gitdiff_renamedst },
1311 { "similarity index ", gitdiff_similarity },
1312 { "dissimilarity index ", gitdiff_dissimilarity },
1313 { "index ", gitdiff_index },
1314 { "", gitdiff_unrecognized },
1316 int i;
1318 len = linelen(line, size);
1319 if (!len || line[len-1] != '\n')
1320 break;
1321 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1322 const struct opentry *p = optable + i;
1323 int oplen = strlen(p->str);
1324 if (len < oplen || memcmp(p->str, line, oplen))
1325 continue;
1326 if (p->fn(line + oplen, patch) < 0)
1327 return offset;
1328 break;
1332 return offset;
1335 static int parse_num(const char *line, unsigned long *p)
1337 char *ptr;
1339 if (!isdigit(*line))
1340 return 0;
1341 *p = strtoul(line, &ptr, 10);
1342 return ptr - line;
1345 static int parse_range(const char *line, int len, int offset, const char *expect,
1346 unsigned long *p1, unsigned long *p2)
1348 int digits, ex;
1350 if (offset < 0 || offset >= len)
1351 return -1;
1352 line += offset;
1353 len -= offset;
1355 digits = parse_num(line, p1);
1356 if (!digits)
1357 return -1;
1359 offset += digits;
1360 line += digits;
1361 len -= digits;
1363 *p2 = 1;
1364 if (*line == ',') {
1365 digits = parse_num(line+1, p2);
1366 if (!digits)
1367 return -1;
1369 offset += digits+1;
1370 line += digits+1;
1371 len -= digits+1;
1374 ex = strlen(expect);
1375 if (ex > len)
1376 return -1;
1377 if (memcmp(line, expect, ex))
1378 return -1;
1380 return offset + ex;
1383 static void recount_diff(const char *line, int size, struct fragment *fragment)
1385 int oldlines = 0, newlines = 0, ret = 0;
1387 if (size < 1) {
1388 warning("recount: ignore empty hunk");
1389 return;
1392 for (;;) {
1393 int len = linelen(line, size);
1394 size -= len;
1395 line += len;
1397 if (size < 1)
1398 break;
1400 switch (*line) {
1401 case ' ': case '\n':
1402 newlines++;
1403 /* fall through */
1404 case '-':
1405 oldlines++;
1406 continue;
1407 case '+':
1408 newlines++;
1409 continue;
1410 case '\\':
1411 continue;
1412 case '@':
1413 ret = size < 3 || prefixcmp(line, "@@ ");
1414 break;
1415 case 'd':
1416 ret = size < 5 || prefixcmp(line, "diff ");
1417 break;
1418 default:
1419 ret = -1;
1420 break;
1422 if (ret) {
1423 warning(_("recount: unexpected line: %.*s"),
1424 (int)linelen(line, size), line);
1425 return;
1427 break;
1429 fragment->oldlines = oldlines;
1430 fragment->newlines = newlines;
1434 * Parse a unified diff fragment header of the
1435 * form "@@ -a,b +c,d @@"
1437 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1439 int offset;
1441 if (!len || line[len-1] != '\n')
1442 return -1;
1444 /* Figure out the number of lines in a fragment */
1445 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1446 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1448 return offset;
1451 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1453 unsigned long offset, len;
1455 patch->is_toplevel_relative = 0;
1456 patch->is_rename = patch->is_copy = 0;
1457 patch->is_new = patch->is_delete = -1;
1458 patch->old_mode = patch->new_mode = 0;
1459 patch->old_name = patch->new_name = NULL;
1460 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1461 unsigned long nextlen;
1463 len = linelen(line, size);
1464 if (!len)
1465 break;
1467 /* Testing this early allows us to take a few shortcuts.. */
1468 if (len < 6)
1469 continue;
1472 * Make sure we don't find any unconnected patch fragments.
1473 * That's a sign that we didn't find a header, and that a
1474 * patch has become corrupted/broken up.
1476 if (!memcmp("@@ -", line, 4)) {
1477 struct fragment dummy;
1478 if (parse_fragment_header(line, len, &dummy) < 0)
1479 continue;
1480 die(_("patch fragment without header at line %d: %.*s"),
1481 linenr, (int)len-1, line);
1484 if (size < len + 6)
1485 break;
1488 * Git patch? It might not have a real patch, just a rename
1489 * or mode change, so we handle that specially
1491 if (!memcmp("diff --git ", line, 11)) {
1492 int git_hdr_len = parse_git_header(line, len, size, patch);
1493 if (git_hdr_len <= len)
1494 continue;
1495 if (!patch->old_name && !patch->new_name) {
1496 if (!patch->def_name)
1497 die(Q_("git diff header lacks filename information when removing "
1498 "%d leading pathname component (line %d)",
1499 "git diff header lacks filename information when removing "
1500 "%d leading pathname components (line %d)",
1501 p_value),
1502 p_value, linenr);
1503 patch->old_name = xstrdup(patch->def_name);
1504 patch->new_name = xstrdup(patch->def_name);
1506 if (!patch->is_delete && !patch->new_name)
1507 die("git diff header lacks filename information "
1508 "(line %d)", linenr);
1509 patch->is_toplevel_relative = 1;
1510 *hdrsize = git_hdr_len;
1511 return offset;
1514 /* --- followed by +++ ? */
1515 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1516 continue;
1519 * We only accept unified patches, so we want it to
1520 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1521 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1523 nextlen = linelen(line + len, size - len);
1524 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1525 continue;
1527 /* Ok, we'll consider it a patch */
1528 parse_traditional_patch(line, line+len, patch);
1529 *hdrsize = len + nextlen;
1530 linenr += 2;
1531 return offset;
1533 return -1;
1536 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1538 char *err;
1540 if (!result)
1541 return;
1543 whitespace_error++;
1544 if (squelch_whitespace_errors &&
1545 squelch_whitespace_errors < whitespace_error)
1546 return;
1548 err = whitespace_error_string(result);
1549 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1550 patch_input_file, linenr, err, len, line);
1551 free(err);
1554 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1556 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1558 record_ws_error(result, line + 1, len - 2, linenr);
1562 * Parse a unified diff. Note that this really needs to parse each
1563 * fragment separately, since the only way to know the difference
1564 * between a "---" that is part of a patch, and a "---" that starts
1565 * the next patch is to look at the line counts..
1567 static int parse_fragment(const char *line, unsigned long size,
1568 struct patch *patch, struct fragment *fragment)
1570 int added, deleted;
1571 int len = linelen(line, size), offset;
1572 unsigned long oldlines, newlines;
1573 unsigned long leading, trailing;
1575 offset = parse_fragment_header(line, len, fragment);
1576 if (offset < 0)
1577 return -1;
1578 if (offset > 0 && patch->recount)
1579 recount_diff(line + offset, size - offset, fragment);
1580 oldlines = fragment->oldlines;
1581 newlines = fragment->newlines;
1582 leading = 0;
1583 trailing = 0;
1585 /* Parse the thing.. */
1586 line += len;
1587 size -= len;
1588 linenr++;
1589 added = deleted = 0;
1590 for (offset = len;
1591 0 < size;
1592 offset += len, size -= len, line += len, linenr++) {
1593 if (!oldlines && !newlines)
1594 break;
1595 len = linelen(line, size);
1596 if (!len || line[len-1] != '\n')
1597 return -1;
1598 switch (*line) {
1599 default:
1600 return -1;
1601 case '\n': /* newer GNU diff, an empty context line */
1602 case ' ':
1603 oldlines--;
1604 newlines--;
1605 if (!deleted && !added)
1606 leading++;
1607 trailing++;
1608 break;
1609 case '-':
1610 if (apply_in_reverse &&
1611 ws_error_action != nowarn_ws_error)
1612 check_whitespace(line, len, patch->ws_rule);
1613 deleted++;
1614 oldlines--;
1615 trailing = 0;
1616 break;
1617 case '+':
1618 if (!apply_in_reverse &&
1619 ws_error_action != nowarn_ws_error)
1620 check_whitespace(line, len, patch->ws_rule);
1621 added++;
1622 newlines--;
1623 trailing = 0;
1624 break;
1627 * We allow "\ No newline at end of file". Depending
1628 * on locale settings when the patch was produced we
1629 * don't know what this line looks like. The only
1630 * thing we do know is that it begins with "\ ".
1631 * Checking for 12 is just for sanity check -- any
1632 * l10n of "\ No newline..." is at least that long.
1634 case '\\':
1635 if (len < 12 || memcmp(line, "\\ ", 2))
1636 return -1;
1637 break;
1640 if (oldlines || newlines)
1641 return -1;
1642 fragment->leading = leading;
1643 fragment->trailing = trailing;
1646 * If a fragment ends with an incomplete line, we failed to include
1647 * it in the above loop because we hit oldlines == newlines == 0
1648 * before seeing it.
1650 if (12 < size && !memcmp(line, "\\ ", 2))
1651 offset += linelen(line, size);
1653 patch->lines_added += added;
1654 patch->lines_deleted += deleted;
1656 if (0 < patch->is_new && oldlines)
1657 return error(_("new file depends on old contents"));
1658 if (0 < patch->is_delete && newlines)
1659 return error(_("deleted file still has contents"));
1660 return offset;
1664 * We have seen "diff --git a/... b/..." header (or a traditional patch
1665 * header). Read hunks that belong to this patch into fragments and hang
1666 * them to the given patch structure.
1668 * The (fragment->patch, fragment->size) pair points into the memory given
1669 * by the caller, not a copy, when we return.
1671 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1673 unsigned long offset = 0;
1674 unsigned long oldlines = 0, newlines = 0, context = 0;
1675 struct fragment **fragp = &patch->fragments;
1677 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1678 struct fragment *fragment;
1679 int len;
1681 fragment = xcalloc(1, sizeof(*fragment));
1682 fragment->linenr = linenr;
1683 len = parse_fragment(line, size, patch, fragment);
1684 if (len <= 0)
1685 die(_("corrupt patch at line %d"), linenr);
1686 fragment->patch = line;
1687 fragment->size = len;
1688 oldlines += fragment->oldlines;
1689 newlines += fragment->newlines;
1690 context += fragment->leading + fragment->trailing;
1692 *fragp = fragment;
1693 fragp = &fragment->next;
1695 offset += len;
1696 line += len;
1697 size -= len;
1701 * If something was removed (i.e. we have old-lines) it cannot
1702 * be creation, and if something was added it cannot be
1703 * deletion. However, the reverse is not true; --unified=0
1704 * patches that only add are not necessarily creation even
1705 * though they do not have any old lines, and ones that only
1706 * delete are not necessarily deletion.
1708 * Unfortunately, a real creation/deletion patch do _not_ have
1709 * any context line by definition, so we cannot safely tell it
1710 * apart with --unified=0 insanity. At least if the patch has
1711 * more than one hunk it is not creation or deletion.
1713 if (patch->is_new < 0 &&
1714 (oldlines || (patch->fragments && patch->fragments->next)))
1715 patch->is_new = 0;
1716 if (patch->is_delete < 0 &&
1717 (newlines || (patch->fragments && patch->fragments->next)))
1718 patch->is_delete = 0;
1720 if (0 < patch->is_new && oldlines)
1721 die(_("new file %s depends on old contents"), patch->new_name);
1722 if (0 < patch->is_delete && newlines)
1723 die(_("deleted file %s still has contents"), patch->old_name);
1724 if (!patch->is_delete && !newlines && context)
1725 fprintf_ln(stderr,
1726 _("** warning: "
1727 "file %s becomes empty but is not deleted"),
1728 patch->new_name);
1730 return offset;
1733 static inline int metadata_changes(struct patch *patch)
1735 return patch->is_rename > 0 ||
1736 patch->is_copy > 0 ||
1737 patch->is_new > 0 ||
1738 patch->is_delete ||
1739 (patch->old_mode && patch->new_mode &&
1740 patch->old_mode != patch->new_mode);
1743 static char *inflate_it(const void *data, unsigned long size,
1744 unsigned long inflated_size)
1746 git_zstream stream;
1747 void *out;
1748 int st;
1750 memset(&stream, 0, sizeof(stream));
1752 stream.next_in = (unsigned char *)data;
1753 stream.avail_in = size;
1754 stream.next_out = out = xmalloc(inflated_size);
1755 stream.avail_out = inflated_size;
1756 git_inflate_init(&stream);
1757 st = git_inflate(&stream, Z_FINISH);
1758 git_inflate_end(&stream);
1759 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1760 free(out);
1761 return NULL;
1763 return out;
1767 * Read a binary hunk and return a new fragment; fragment->patch
1768 * points at an allocated memory that the caller must free, so
1769 * it is marked as "->free_patch = 1".
1771 static struct fragment *parse_binary_hunk(char **buf_p,
1772 unsigned long *sz_p,
1773 int *status_p,
1774 int *used_p)
1777 * Expect a line that begins with binary patch method ("literal"
1778 * or "delta"), followed by the length of data before deflating.
1779 * a sequence of 'length-byte' followed by base-85 encoded data
1780 * should follow, terminated by a newline.
1782 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1783 * and we would limit the patch line to 66 characters,
1784 * so one line can fit up to 13 groups that would decode
1785 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1786 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1788 int llen, used;
1789 unsigned long size = *sz_p;
1790 char *buffer = *buf_p;
1791 int patch_method;
1792 unsigned long origlen;
1793 char *data = NULL;
1794 int hunk_size = 0;
1795 struct fragment *frag;
1797 llen = linelen(buffer, size);
1798 used = llen;
1800 *status_p = 0;
1802 if (!prefixcmp(buffer, "delta ")) {
1803 patch_method = BINARY_DELTA_DEFLATED;
1804 origlen = strtoul(buffer + 6, NULL, 10);
1806 else if (!prefixcmp(buffer, "literal ")) {
1807 patch_method = BINARY_LITERAL_DEFLATED;
1808 origlen = strtoul(buffer + 8, NULL, 10);
1810 else
1811 return NULL;
1813 linenr++;
1814 buffer += llen;
1815 while (1) {
1816 int byte_length, max_byte_length, newsize;
1817 llen = linelen(buffer, size);
1818 used += llen;
1819 linenr++;
1820 if (llen == 1) {
1821 /* consume the blank line */
1822 buffer++;
1823 size--;
1824 break;
1827 * Minimum line is "A00000\n" which is 7-byte long,
1828 * and the line length must be multiple of 5 plus 2.
1830 if ((llen < 7) || (llen-2) % 5)
1831 goto corrupt;
1832 max_byte_length = (llen - 2) / 5 * 4;
1833 byte_length = *buffer;
1834 if ('A' <= byte_length && byte_length <= 'Z')
1835 byte_length = byte_length - 'A' + 1;
1836 else if ('a' <= byte_length && byte_length <= 'z')
1837 byte_length = byte_length - 'a' + 27;
1838 else
1839 goto corrupt;
1840 /* if the input length was not multiple of 4, we would
1841 * have filler at the end but the filler should never
1842 * exceed 3 bytes
1844 if (max_byte_length < byte_length ||
1845 byte_length <= max_byte_length - 4)
1846 goto corrupt;
1847 newsize = hunk_size + byte_length;
1848 data = xrealloc(data, newsize);
1849 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1850 goto corrupt;
1851 hunk_size = newsize;
1852 buffer += llen;
1853 size -= llen;
1856 frag = xcalloc(1, sizeof(*frag));
1857 frag->patch = inflate_it(data, hunk_size, origlen);
1858 frag->free_patch = 1;
1859 if (!frag->patch)
1860 goto corrupt;
1861 free(data);
1862 frag->size = origlen;
1863 *buf_p = buffer;
1864 *sz_p = size;
1865 *used_p = used;
1866 frag->binary_patch_method = patch_method;
1867 return frag;
1869 corrupt:
1870 free(data);
1871 *status_p = -1;
1872 error(_("corrupt binary patch at line %d: %.*s"),
1873 linenr-1, llen-1, buffer);
1874 return NULL;
1877 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1880 * We have read "GIT binary patch\n"; what follows is a line
1881 * that says the patch method (currently, either "literal" or
1882 * "delta") and the length of data before deflating; a
1883 * sequence of 'length-byte' followed by base-85 encoded data
1884 * follows.
1886 * When a binary patch is reversible, there is another binary
1887 * hunk in the same format, starting with patch method (either
1888 * "literal" or "delta") with the length of data, and a sequence
1889 * of length-byte + base-85 encoded data, terminated with another
1890 * empty line. This data, when applied to the postimage, produces
1891 * the preimage.
1893 struct fragment *forward;
1894 struct fragment *reverse;
1895 int status;
1896 int used, used_1;
1898 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1899 if (!forward && !status)
1900 /* there has to be one hunk (forward hunk) */
1901 return error(_("unrecognized binary patch at line %d"), linenr-1);
1902 if (status)
1903 /* otherwise we already gave an error message */
1904 return status;
1906 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1907 if (reverse)
1908 used += used_1;
1909 else if (status) {
1911 * Not having reverse hunk is not an error, but having
1912 * a corrupt reverse hunk is.
1914 free((void*) forward->patch);
1915 free(forward);
1916 return status;
1918 forward->next = reverse;
1919 patch->fragments = forward;
1920 patch->is_binary = 1;
1921 return used;
1925 * Read the patch text in "buffer" that extends for "size" bytes; stop
1926 * reading after seeing a single patch (i.e. changes to a single file).
1927 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1928 * Return the number of bytes consumed, so that the caller can call us
1929 * again for the next patch.
1931 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1933 int hdrsize, patchsize;
1934 int offset = find_header(buffer, size, &hdrsize, patch);
1936 if (offset < 0)
1937 return offset;
1939 patch->ws_rule = whitespace_rule(patch->new_name
1940 ? patch->new_name
1941 : patch->old_name);
1943 patchsize = parse_single_patch(buffer + offset + hdrsize,
1944 size - offset - hdrsize, patch);
1946 if (!patchsize) {
1947 static const char *binhdr[] = {
1948 "Binary files ",
1949 "Files ",
1950 NULL,
1952 static const char git_binary[] = "GIT binary patch\n";
1953 int i;
1954 int hd = hdrsize + offset;
1955 unsigned long llen = linelen(buffer + hd, size - hd);
1957 if (llen == sizeof(git_binary) - 1 &&
1958 !memcmp(git_binary, buffer + hd, llen)) {
1959 int used;
1960 linenr++;
1961 used = parse_binary(buffer + hd + llen,
1962 size - hd - llen, patch);
1963 if (used)
1964 patchsize = used + llen;
1965 else
1966 patchsize = 0;
1968 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1969 for (i = 0; binhdr[i]; i++) {
1970 int len = strlen(binhdr[i]);
1971 if (len < size - hd &&
1972 !memcmp(binhdr[i], buffer + hd, len)) {
1973 linenr++;
1974 patch->is_binary = 1;
1975 patchsize = llen;
1976 break;
1981 /* Empty patch cannot be applied if it is a text patch
1982 * without metadata change. A binary patch appears
1983 * empty to us here.
1985 if ((apply || check) &&
1986 (!patch->is_binary && !metadata_changes(patch)))
1987 die(_("patch with only garbage at line %d"), linenr);
1990 return offset + hdrsize + patchsize;
1993 #define swap(a,b) myswap((a),(b),sizeof(a))
1995 #define myswap(a, b, size) do { \
1996 unsigned char mytmp[size]; \
1997 memcpy(mytmp, &a, size); \
1998 memcpy(&a, &b, size); \
1999 memcpy(&b, mytmp, size); \
2000 } while (0)
2002 static void reverse_patches(struct patch *p)
2004 for (; p; p = p->next) {
2005 struct fragment *frag = p->fragments;
2007 swap(p->new_name, p->old_name);
2008 swap(p->new_mode, p->old_mode);
2009 swap(p->is_new, p->is_delete);
2010 swap(p->lines_added, p->lines_deleted);
2011 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2013 for (; frag; frag = frag->next) {
2014 swap(frag->newpos, frag->oldpos);
2015 swap(frag->newlines, frag->oldlines);
2020 static const char pluses[] =
2021 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2022 static const char minuses[]=
2023 "----------------------------------------------------------------------";
2025 static void show_stats(struct patch *patch)
2027 struct strbuf qname = STRBUF_INIT;
2028 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2029 int max, add, del;
2031 quote_c_style(cp, &qname, NULL, 0);
2034 * "scale" the filename
2036 max = max_len;
2037 if (max > 50)
2038 max = 50;
2040 if (qname.len > max) {
2041 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2042 if (!cp)
2043 cp = qname.buf + qname.len + 3 - max;
2044 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2047 if (patch->is_binary) {
2048 printf(" %-*s | Bin\n", max, qname.buf);
2049 strbuf_release(&qname);
2050 return;
2053 printf(" %-*s |", max, qname.buf);
2054 strbuf_release(&qname);
2057 * scale the add/delete
2059 max = max + max_change > 70 ? 70 - max : max_change;
2060 add = patch->lines_added;
2061 del = patch->lines_deleted;
2063 if (max_change > 0) {
2064 int total = ((add + del) * max + max_change / 2) / max_change;
2065 add = (add * max + max_change / 2) / max_change;
2066 del = total - add;
2068 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2069 add, pluses, del, minuses);
2072 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2074 switch (st->st_mode & S_IFMT) {
2075 case S_IFLNK:
2076 if (strbuf_readlink(buf, path, st->st_size) < 0)
2077 return error(_("unable to read symlink %s"), path);
2078 return 0;
2079 case S_IFREG:
2080 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2081 return error(_("unable to open or read %s"), path);
2082 convert_to_git(path, buf->buf, buf->len, buf, 0);
2083 return 0;
2084 default:
2085 return -1;
2090 * Update the preimage, and the common lines in postimage,
2091 * from buffer buf of length len. If postlen is 0 the postimage
2092 * is updated in place, otherwise it's updated on a new buffer
2093 * of length postlen
2096 static void update_pre_post_images(struct image *preimage,
2097 struct image *postimage,
2098 char *buf,
2099 size_t len, size_t postlen)
2101 int i, ctx, reduced;
2102 char *new, *old, *fixed;
2103 struct image fixed_preimage;
2106 * Update the preimage with whitespace fixes. Note that we
2107 * are not losing preimage->buf -- apply_one_fragment() will
2108 * free "oldlines".
2110 prepare_image(&fixed_preimage, buf, len, 1);
2111 assert(postlen
2112 ? fixed_preimage.nr == preimage->nr
2113 : fixed_preimage.nr <= preimage->nr);
2114 for (i = 0; i < fixed_preimage.nr; i++)
2115 fixed_preimage.line[i].flag = preimage->line[i].flag;
2116 free(preimage->line_allocated);
2117 *preimage = fixed_preimage;
2120 * Adjust the common context lines in postimage. This can be
2121 * done in-place when we are shrinking it with whitespace
2122 * fixing, but needs a new buffer when ignoring whitespace or
2123 * expanding leading tabs to spaces.
2125 * We trust the caller to tell us if the update can be done
2126 * in place (postlen==0) or not.
2128 old = postimage->buf;
2129 if (postlen)
2130 new = postimage->buf = xmalloc(postlen);
2131 else
2132 new = old;
2133 fixed = preimage->buf;
2135 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2136 size_t len = postimage->line[i].len;
2137 if (!(postimage->line[i].flag & LINE_COMMON)) {
2138 /* an added line -- no counterparts in preimage */
2139 memmove(new, old, len);
2140 old += len;
2141 new += len;
2142 continue;
2145 /* a common context -- skip it in the original postimage */
2146 old += len;
2148 /* and find the corresponding one in the fixed preimage */
2149 while (ctx < preimage->nr &&
2150 !(preimage->line[ctx].flag & LINE_COMMON)) {
2151 fixed += preimage->line[ctx].len;
2152 ctx++;
2156 * preimage is expected to run out, if the caller
2157 * fixed addition of trailing blank lines.
2159 if (preimage->nr <= ctx) {
2160 reduced++;
2161 continue;
2164 /* and copy it in, while fixing the line length */
2165 len = preimage->line[ctx].len;
2166 memcpy(new, fixed, len);
2167 new += len;
2168 fixed += len;
2169 postimage->line[i].len = len;
2170 ctx++;
2173 /* Fix the length of the whole thing */
2174 postimage->len = new - postimage->buf;
2175 postimage->nr -= reduced;
2178 static int match_fragment(struct image *img,
2179 struct image *preimage,
2180 struct image *postimage,
2181 unsigned long try,
2182 int try_lno,
2183 unsigned ws_rule,
2184 int match_beginning, int match_end)
2186 int i;
2187 char *fixed_buf, *buf, *orig, *target;
2188 struct strbuf fixed;
2189 size_t fixed_len, postlen;
2190 int preimage_limit;
2192 if (preimage->nr + try_lno <= img->nr) {
2194 * The hunk falls within the boundaries of img.
2196 preimage_limit = preimage->nr;
2197 if (match_end && (preimage->nr + try_lno != img->nr))
2198 return 0;
2199 } else if (ws_error_action == correct_ws_error &&
2200 (ws_rule & WS_BLANK_AT_EOF)) {
2202 * This hunk extends beyond the end of img, and we are
2203 * removing blank lines at the end of the file. This
2204 * many lines from the beginning of the preimage must
2205 * match with img, and the remainder of the preimage
2206 * must be blank.
2208 preimage_limit = img->nr - try_lno;
2209 } else {
2211 * The hunk extends beyond the end of the img and
2212 * we are not removing blanks at the end, so we
2213 * should reject the hunk at this position.
2215 return 0;
2218 if (match_beginning && try_lno)
2219 return 0;
2221 /* Quick hash check */
2222 for (i = 0; i < preimage_limit; i++)
2223 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2224 (preimage->line[i].hash != img->line[try_lno + i].hash))
2225 return 0;
2227 if (preimage_limit == preimage->nr) {
2229 * Do we have an exact match? If we were told to match
2230 * at the end, size must be exactly at try+fragsize,
2231 * otherwise try+fragsize must be still within the preimage,
2232 * and either case, the old piece should match the preimage
2233 * exactly.
2235 if ((match_end
2236 ? (try + preimage->len == img->len)
2237 : (try + preimage->len <= img->len)) &&
2238 !memcmp(img->buf + try, preimage->buf, preimage->len))
2239 return 1;
2240 } else {
2242 * The preimage extends beyond the end of img, so
2243 * there cannot be an exact match.
2245 * There must be one non-blank context line that match
2246 * a line before the end of img.
2248 char *buf_end;
2250 buf = preimage->buf;
2251 buf_end = buf;
2252 for (i = 0; i < preimage_limit; i++)
2253 buf_end += preimage->line[i].len;
2255 for ( ; buf < buf_end; buf++)
2256 if (!isspace(*buf))
2257 break;
2258 if (buf == buf_end)
2259 return 0;
2263 * No exact match. If we are ignoring whitespace, run a line-by-line
2264 * fuzzy matching. We collect all the line length information because
2265 * we need it to adjust whitespace if we match.
2267 if (ws_ignore_action == ignore_ws_change) {
2268 size_t imgoff = 0;
2269 size_t preoff = 0;
2270 size_t postlen = postimage->len;
2271 size_t extra_chars;
2272 char *preimage_eof;
2273 char *preimage_end;
2274 for (i = 0; i < preimage_limit; i++) {
2275 size_t prelen = preimage->line[i].len;
2276 size_t imglen = img->line[try_lno+i].len;
2278 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2279 preimage->buf + preoff, prelen))
2280 return 0;
2281 if (preimage->line[i].flag & LINE_COMMON)
2282 postlen += imglen - prelen;
2283 imgoff += imglen;
2284 preoff += prelen;
2288 * Ok, the preimage matches with whitespace fuzz.
2290 * imgoff now holds the true length of the target that
2291 * matches the preimage before the end of the file.
2293 * Count the number of characters in the preimage that fall
2294 * beyond the end of the file and make sure that all of them
2295 * are whitespace characters. (This can only happen if
2296 * we are removing blank lines at the end of the file.)
2298 buf = preimage_eof = preimage->buf + preoff;
2299 for ( ; i < preimage->nr; i++)
2300 preoff += preimage->line[i].len;
2301 preimage_end = preimage->buf + preoff;
2302 for ( ; buf < preimage_end; buf++)
2303 if (!isspace(*buf))
2304 return 0;
2307 * Update the preimage and the common postimage context
2308 * lines to use the same whitespace as the target.
2309 * If whitespace is missing in the target (i.e.
2310 * if the preimage extends beyond the end of the file),
2311 * use the whitespace from the preimage.
2313 extra_chars = preimage_end - preimage_eof;
2314 strbuf_init(&fixed, imgoff + extra_chars);
2315 strbuf_add(&fixed, img->buf + try, imgoff);
2316 strbuf_add(&fixed, preimage_eof, extra_chars);
2317 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2318 update_pre_post_images(preimage, postimage,
2319 fixed_buf, fixed_len, postlen);
2320 return 1;
2323 if (ws_error_action != correct_ws_error)
2324 return 0;
2327 * The hunk does not apply byte-by-byte, but the hash says
2328 * it might with whitespace fuzz. We haven't been asked to
2329 * ignore whitespace, we were asked to correct whitespace
2330 * errors, so let's try matching after whitespace correction.
2332 * The preimage may extend beyond the end of the file,
2333 * but in this loop we will only handle the part of the
2334 * preimage that falls within the file.
2336 strbuf_init(&fixed, preimage->len + 1);
2337 orig = preimage->buf;
2338 target = img->buf + try;
2339 postlen = 0;
2340 for (i = 0; i < preimage_limit; i++) {
2341 size_t oldlen = preimage->line[i].len;
2342 size_t tgtlen = img->line[try_lno + i].len;
2343 size_t fixstart = fixed.len;
2344 struct strbuf tgtfix;
2345 int match;
2347 /* Try fixing the line in the preimage */
2348 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2350 /* Try fixing the line in the target */
2351 strbuf_init(&tgtfix, tgtlen);
2352 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2355 * If they match, either the preimage was based on
2356 * a version before our tree fixed whitespace breakage,
2357 * or we are lacking a whitespace-fix patch the tree
2358 * the preimage was based on already had (i.e. target
2359 * has whitespace breakage, the preimage doesn't).
2360 * In either case, we are fixing the whitespace breakages
2361 * so we might as well take the fix together with their
2362 * real change.
2364 match = (tgtfix.len == fixed.len - fixstart &&
2365 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2366 fixed.len - fixstart));
2367 postlen += tgtfix.len;
2369 strbuf_release(&tgtfix);
2370 if (!match)
2371 goto unmatch_exit;
2373 orig += oldlen;
2374 target += tgtlen;
2379 * Now handle the lines in the preimage that falls beyond the
2380 * end of the file (if any). They will only match if they are
2381 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2382 * false).
2384 for ( ; i < preimage->nr; i++) {
2385 size_t fixstart = fixed.len; /* start of the fixed preimage */
2386 size_t oldlen = preimage->line[i].len;
2387 int j;
2389 /* Try fixing the line in the preimage */
2390 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2392 for (j = fixstart; j < fixed.len; j++)
2393 if (!isspace(fixed.buf[j]))
2394 goto unmatch_exit;
2396 orig += oldlen;
2400 * Yes, the preimage is based on an older version that still
2401 * has whitespace breakages unfixed, and fixing them makes the
2402 * hunk match. Update the context lines in the postimage.
2404 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2405 if (postlen < postimage->len)
2406 postlen = 0;
2407 update_pre_post_images(preimage, postimage,
2408 fixed_buf, fixed_len, postlen);
2409 return 1;
2411 unmatch_exit:
2412 strbuf_release(&fixed);
2413 return 0;
2416 static int find_pos(struct image *img,
2417 struct image *preimage,
2418 struct image *postimage,
2419 int line,
2420 unsigned ws_rule,
2421 int match_beginning, int match_end)
2423 int i;
2424 unsigned long backwards, forwards, try;
2425 int backwards_lno, forwards_lno, try_lno;
2428 * If match_beginning or match_end is specified, there is no
2429 * point starting from a wrong line that will never match and
2430 * wander around and wait for a match at the specified end.
2432 if (match_beginning)
2433 line = 0;
2434 else if (match_end)
2435 line = img->nr - preimage->nr;
2438 * Because the comparison is unsigned, the following test
2439 * will also take care of a negative line number that can
2440 * result when match_end and preimage is larger than the target.
2442 if ((size_t) line > img->nr)
2443 line = img->nr;
2445 try = 0;
2446 for (i = 0; i < line; i++)
2447 try += img->line[i].len;
2450 * There's probably some smart way to do this, but I'll leave
2451 * that to the smart and beautiful people. I'm simple and stupid.
2453 backwards = try;
2454 backwards_lno = line;
2455 forwards = try;
2456 forwards_lno = line;
2457 try_lno = line;
2459 for (i = 0; ; i++) {
2460 if (match_fragment(img, preimage, postimage,
2461 try, try_lno, ws_rule,
2462 match_beginning, match_end))
2463 return try_lno;
2465 again:
2466 if (backwards_lno == 0 && forwards_lno == img->nr)
2467 break;
2469 if (i & 1) {
2470 if (backwards_lno == 0) {
2471 i++;
2472 goto again;
2474 backwards_lno--;
2475 backwards -= img->line[backwards_lno].len;
2476 try = backwards;
2477 try_lno = backwards_lno;
2478 } else {
2479 if (forwards_lno == img->nr) {
2480 i++;
2481 goto again;
2483 forwards += img->line[forwards_lno].len;
2484 forwards_lno++;
2485 try = forwards;
2486 try_lno = forwards_lno;
2490 return -1;
2493 static void remove_first_line(struct image *img)
2495 img->buf += img->line[0].len;
2496 img->len -= img->line[0].len;
2497 img->line++;
2498 img->nr--;
2501 static void remove_last_line(struct image *img)
2503 img->len -= img->line[--img->nr].len;
2507 * The change from "preimage" and "postimage" has been found to
2508 * apply at applied_pos (counts in line numbers) in "img".
2509 * Update "img" to remove "preimage" and replace it with "postimage".
2511 static void update_image(struct image *img,
2512 int applied_pos,
2513 struct image *preimage,
2514 struct image *postimage)
2517 * remove the copy of preimage at offset in img
2518 * and replace it with postimage
2520 int i, nr;
2521 size_t remove_count, insert_count, applied_at = 0;
2522 char *result;
2523 int preimage_limit;
2526 * If we are removing blank lines at the end of img,
2527 * the preimage may extend beyond the end.
2528 * If that is the case, we must be careful only to
2529 * remove the part of the preimage that falls within
2530 * the boundaries of img. Initialize preimage_limit
2531 * to the number of lines in the preimage that falls
2532 * within the boundaries.
2534 preimage_limit = preimage->nr;
2535 if (preimage_limit > img->nr - applied_pos)
2536 preimage_limit = img->nr - applied_pos;
2538 for (i = 0; i < applied_pos; i++)
2539 applied_at += img->line[i].len;
2541 remove_count = 0;
2542 for (i = 0; i < preimage_limit; i++)
2543 remove_count += img->line[applied_pos + i].len;
2544 insert_count = postimage->len;
2546 /* Adjust the contents */
2547 result = xmalloc(img->len + insert_count - remove_count + 1);
2548 memcpy(result, img->buf, applied_at);
2549 memcpy(result + applied_at, postimage->buf, postimage->len);
2550 memcpy(result + applied_at + postimage->len,
2551 img->buf + (applied_at + remove_count),
2552 img->len - (applied_at + remove_count));
2553 free(img->buf);
2554 img->buf = result;
2555 img->len += insert_count - remove_count;
2556 result[img->len] = '\0';
2558 /* Adjust the line table */
2559 nr = img->nr + postimage->nr - preimage_limit;
2560 if (preimage_limit < postimage->nr) {
2562 * NOTE: this knows that we never call remove_first_line()
2563 * on anything other than pre/post image.
2565 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2566 img->line_allocated = img->line;
2568 if (preimage_limit != postimage->nr)
2569 memmove(img->line + applied_pos + postimage->nr,
2570 img->line + applied_pos + preimage_limit,
2571 (img->nr - (applied_pos + preimage_limit)) *
2572 sizeof(*img->line));
2573 memcpy(img->line + applied_pos,
2574 postimage->line,
2575 postimage->nr * sizeof(*img->line));
2576 if (!allow_overlap)
2577 for (i = 0; i < postimage->nr; i++)
2578 img->line[applied_pos + i].flag |= LINE_PATCHED;
2579 img->nr = nr;
2583 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2584 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2585 * replace the part of "img" with "postimage" text.
2587 static int apply_one_fragment(struct image *img, struct fragment *frag,
2588 int inaccurate_eof, unsigned ws_rule,
2589 int nth_fragment)
2591 int match_beginning, match_end;
2592 const char *patch = frag->patch;
2593 int size = frag->size;
2594 char *old, *oldlines;
2595 struct strbuf newlines;
2596 int new_blank_lines_at_end = 0;
2597 int found_new_blank_lines_at_end = 0;
2598 int hunk_linenr = frag->linenr;
2599 unsigned long leading, trailing;
2600 int pos, applied_pos;
2601 struct image preimage;
2602 struct image postimage;
2604 memset(&preimage, 0, sizeof(preimage));
2605 memset(&postimage, 0, sizeof(postimage));
2606 oldlines = xmalloc(size);
2607 strbuf_init(&newlines, size);
2609 old = oldlines;
2610 while (size > 0) {
2611 char first;
2612 int len = linelen(patch, size);
2613 int plen;
2614 int added_blank_line = 0;
2615 int is_blank_context = 0;
2616 size_t start;
2618 if (!len)
2619 break;
2622 * "plen" is how much of the line we should use for
2623 * the actual patch data. Normally we just remove the
2624 * first character on the line, but if the line is
2625 * followed by "\ No newline", then we also remove the
2626 * last one (which is the newline, of course).
2628 plen = len - 1;
2629 if (len < size && patch[len] == '\\')
2630 plen--;
2631 first = *patch;
2632 if (apply_in_reverse) {
2633 if (first == '-')
2634 first = '+';
2635 else if (first == '+')
2636 first = '-';
2639 switch (first) {
2640 case '\n':
2641 /* Newer GNU diff, empty context line */
2642 if (plen < 0)
2643 /* ... followed by '\No newline'; nothing */
2644 break;
2645 *old++ = '\n';
2646 strbuf_addch(&newlines, '\n');
2647 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2648 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2649 is_blank_context = 1;
2650 break;
2651 case ' ':
2652 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2653 ws_blank_line(patch + 1, plen, ws_rule))
2654 is_blank_context = 1;
2655 case '-':
2656 memcpy(old, patch + 1, plen);
2657 add_line_info(&preimage, old, plen,
2658 (first == ' ' ? LINE_COMMON : 0));
2659 old += plen;
2660 if (first == '-')
2661 break;
2662 /* Fall-through for ' ' */
2663 case '+':
2664 /* --no-add does not add new lines */
2665 if (first == '+' && no_add)
2666 break;
2668 start = newlines.len;
2669 if (first != '+' ||
2670 !whitespace_error ||
2671 ws_error_action != correct_ws_error) {
2672 strbuf_add(&newlines, patch + 1, plen);
2674 else {
2675 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2677 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2678 (first == '+' ? 0 : LINE_COMMON));
2679 if (first == '+' &&
2680 (ws_rule & WS_BLANK_AT_EOF) &&
2681 ws_blank_line(patch + 1, plen, ws_rule))
2682 added_blank_line = 1;
2683 break;
2684 case '@': case '\\':
2685 /* Ignore it, we already handled it */
2686 break;
2687 default:
2688 if (apply_verbosely)
2689 error(_("invalid start of line: '%c'"), first);
2690 return -1;
2692 if (added_blank_line) {
2693 if (!new_blank_lines_at_end)
2694 found_new_blank_lines_at_end = hunk_linenr;
2695 new_blank_lines_at_end++;
2697 else if (is_blank_context)
2699 else
2700 new_blank_lines_at_end = 0;
2701 patch += len;
2702 size -= len;
2703 hunk_linenr++;
2705 if (inaccurate_eof &&
2706 old > oldlines && old[-1] == '\n' &&
2707 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2708 old--;
2709 strbuf_setlen(&newlines, newlines.len - 1);
2712 leading = frag->leading;
2713 trailing = frag->trailing;
2716 * A hunk to change lines at the beginning would begin with
2717 * @@ -1,L +N,M @@
2718 * but we need to be careful. -U0 that inserts before the second
2719 * line also has this pattern.
2721 * And a hunk to add to an empty file would begin with
2722 * @@ -0,0 +N,M @@
2724 * In other words, a hunk that is (frag->oldpos <= 1) with or
2725 * without leading context must match at the beginning.
2727 match_beginning = (!frag->oldpos ||
2728 (frag->oldpos == 1 && !unidiff_zero));
2731 * A hunk without trailing lines must match at the end.
2732 * However, we simply cannot tell if a hunk must match end
2733 * from the lack of trailing lines if the patch was generated
2734 * with unidiff without any context.
2736 match_end = !unidiff_zero && !trailing;
2738 pos = frag->newpos ? (frag->newpos - 1) : 0;
2739 preimage.buf = oldlines;
2740 preimage.len = old - oldlines;
2741 postimage.buf = newlines.buf;
2742 postimage.len = newlines.len;
2743 preimage.line = preimage.line_allocated;
2744 postimage.line = postimage.line_allocated;
2746 for (;;) {
2748 applied_pos = find_pos(img, &preimage, &postimage, pos,
2749 ws_rule, match_beginning, match_end);
2751 if (applied_pos >= 0)
2752 break;
2754 /* Am I at my context limits? */
2755 if ((leading <= p_context) && (trailing <= p_context))
2756 break;
2757 if (match_beginning || match_end) {
2758 match_beginning = match_end = 0;
2759 continue;
2763 * Reduce the number of context lines; reduce both
2764 * leading and trailing if they are equal otherwise
2765 * just reduce the larger context.
2767 if (leading >= trailing) {
2768 remove_first_line(&preimage);
2769 remove_first_line(&postimage);
2770 pos--;
2771 leading--;
2773 if (trailing > leading) {
2774 remove_last_line(&preimage);
2775 remove_last_line(&postimage);
2776 trailing--;
2780 if (applied_pos >= 0) {
2781 if (new_blank_lines_at_end &&
2782 preimage.nr + applied_pos >= img->nr &&
2783 (ws_rule & WS_BLANK_AT_EOF) &&
2784 ws_error_action != nowarn_ws_error) {
2785 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2786 found_new_blank_lines_at_end);
2787 if (ws_error_action == correct_ws_error) {
2788 while (new_blank_lines_at_end--)
2789 remove_last_line(&postimage);
2792 * We would want to prevent write_out_results()
2793 * from taking place in apply_patch() that follows
2794 * the callchain led us here, which is:
2795 * apply_patch->check_patch_list->check_patch->
2796 * apply_data->apply_fragments->apply_one_fragment
2798 if (ws_error_action == die_on_ws_error)
2799 apply = 0;
2802 if (apply_verbosely && applied_pos != pos) {
2803 int offset = applied_pos - pos;
2804 if (apply_in_reverse)
2805 offset = 0 - offset;
2806 fprintf_ln(stderr,
2807 Q_("Hunk #%d succeeded at %d (offset %d line).",
2808 "Hunk #%d succeeded at %d (offset %d lines).",
2809 offset),
2810 nth_fragment, applied_pos + 1, offset);
2814 * Warn if it was necessary to reduce the number
2815 * of context lines.
2817 if ((leading != frag->leading) ||
2818 (trailing != frag->trailing))
2819 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2820 " to apply fragment at %d"),
2821 leading, trailing, applied_pos+1);
2822 update_image(img, applied_pos, &preimage, &postimage);
2823 } else {
2824 if (apply_verbosely)
2825 error(_("while searching for:\n%.*s"),
2826 (int)(old - oldlines), oldlines);
2829 free(oldlines);
2830 strbuf_release(&newlines);
2831 free(preimage.line_allocated);
2832 free(postimage.line_allocated);
2834 return (applied_pos < 0);
2837 static int apply_binary_fragment(struct image *img, struct patch *patch)
2839 struct fragment *fragment = patch->fragments;
2840 unsigned long len;
2841 void *dst;
2843 if (!fragment)
2844 return error(_("missing binary patch data for '%s'"),
2845 patch->new_name ?
2846 patch->new_name :
2847 patch->old_name);
2849 /* Binary patch is irreversible without the optional second hunk */
2850 if (apply_in_reverse) {
2851 if (!fragment->next)
2852 return error("cannot reverse-apply a binary patch "
2853 "without the reverse hunk to '%s'",
2854 patch->new_name
2855 ? patch->new_name : patch->old_name);
2856 fragment = fragment->next;
2858 switch (fragment->binary_patch_method) {
2859 case BINARY_DELTA_DEFLATED:
2860 dst = patch_delta(img->buf, img->len, fragment->patch,
2861 fragment->size, &len);
2862 if (!dst)
2863 return -1;
2864 clear_image(img);
2865 img->buf = dst;
2866 img->len = len;
2867 return 0;
2868 case BINARY_LITERAL_DEFLATED:
2869 clear_image(img);
2870 img->len = fragment->size;
2871 img->buf = xmalloc(img->len+1);
2872 memcpy(img->buf, fragment->patch, img->len);
2873 img->buf[img->len] = '\0';
2874 return 0;
2876 return -1;
2880 * Replace "img" with the result of applying the binary patch.
2881 * The binary patch data itself in patch->fragment is still kept
2882 * but the preimage prepared by the caller in "img" is freed here
2883 * or in the helper function apply_binary_fragment() this calls.
2885 static int apply_binary(struct image *img, struct patch *patch)
2887 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2888 unsigned char sha1[20];
2891 * For safety, we require patch index line to contain
2892 * full 40-byte textual SHA1 for old and new, at least for now.
2894 if (strlen(patch->old_sha1_prefix) != 40 ||
2895 strlen(patch->new_sha1_prefix) != 40 ||
2896 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2897 get_sha1_hex(patch->new_sha1_prefix, sha1))
2898 return error("cannot apply binary patch to '%s' "
2899 "without full index line", name);
2901 if (patch->old_name) {
2903 * See if the old one matches what the patch
2904 * applies to.
2906 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2907 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2908 return error("the patch applies to '%s' (%s), "
2909 "which does not match the "
2910 "current contents.",
2911 name, sha1_to_hex(sha1));
2913 else {
2914 /* Otherwise, the old one must be empty. */
2915 if (img->len)
2916 return error("the patch applies to an empty "
2917 "'%s' but it is not empty", name);
2920 get_sha1_hex(patch->new_sha1_prefix, sha1);
2921 if (is_null_sha1(sha1)) {
2922 clear_image(img);
2923 return 0; /* deletion patch */
2926 if (has_sha1_file(sha1)) {
2927 /* We already have the postimage */
2928 enum object_type type;
2929 unsigned long size;
2930 char *result;
2932 result = read_sha1_file(sha1, &type, &size);
2933 if (!result)
2934 return error("the necessary postimage %s for "
2935 "'%s' cannot be read",
2936 patch->new_sha1_prefix, name);
2937 clear_image(img);
2938 img->buf = result;
2939 img->len = size;
2940 } else {
2942 * We have verified buf matches the preimage;
2943 * apply the patch data to it, which is stored
2944 * in the patch->fragments->{patch,size}.
2946 if (apply_binary_fragment(img, patch))
2947 return error(_("binary patch does not apply to '%s'"),
2948 name);
2950 /* verify that the result matches */
2951 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2952 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2953 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
2954 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2957 return 0;
2960 static int apply_fragments(struct image *img, struct patch *patch)
2962 struct fragment *frag = patch->fragments;
2963 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2964 unsigned ws_rule = patch->ws_rule;
2965 unsigned inaccurate_eof = patch->inaccurate_eof;
2966 int nth = 0;
2968 if (patch->is_binary)
2969 return apply_binary(img, patch);
2971 while (frag) {
2972 nth++;
2973 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2974 error(_("patch failed: %s:%ld"), name, frag->oldpos);
2975 if (!apply_with_reject)
2976 return -1;
2977 frag->rejected = 1;
2979 frag = frag->next;
2981 return 0;
2984 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
2986 if (S_ISGITLINK(mode)) {
2987 strbuf_grow(buf, 100);
2988 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
2989 } else {
2990 enum object_type type;
2991 unsigned long sz;
2992 char *result;
2994 result = read_sha1_file(sha1, &type, &sz);
2995 if (!result)
2996 return -1;
2997 /* XXX read_sha1_file NUL-terminates */
2998 strbuf_attach(buf, result, sz, sz + 1);
3000 return 0;
3003 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3005 if (!ce)
3006 return 0;
3007 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3010 static struct patch *in_fn_table(const char *name)
3012 struct string_list_item *item;
3014 if (name == NULL)
3015 return NULL;
3017 item = string_list_lookup(&fn_table, name);
3018 if (item != NULL)
3019 return (struct patch *)item->util;
3021 return NULL;
3025 * item->util in the filename table records the status of the path.
3026 * Usually it points at a patch (whose result records the contents
3027 * of it after applying it), but it could be PATH_WAS_DELETED for a
3028 * path that a previously applied patch has already removed, or
3029 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3031 * The latter is needed to deal with a case where two paths A and B
3032 * are swapped by first renaming A to B and then renaming B to A;
3033 * moving A to B should not be prevented due to presence of B as we
3034 * will remove it in a later patch.
3036 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3037 #define PATH_WAS_DELETED ((struct patch *) -1)
3039 static int to_be_deleted(struct patch *patch)
3041 return patch == PATH_TO_BE_DELETED;
3044 static int was_deleted(struct patch *patch)
3046 return patch == PATH_WAS_DELETED;
3049 static void add_to_fn_table(struct patch *patch)
3051 struct string_list_item *item;
3054 * Always add new_name unless patch is a deletion
3055 * This should cover the cases for normal diffs,
3056 * file creations and copies
3058 if (patch->new_name != NULL) {
3059 item = string_list_insert(&fn_table, patch->new_name);
3060 item->util = patch;
3064 * store a failure on rename/deletion cases because
3065 * later chunks shouldn't patch old names
3067 if ((patch->new_name == NULL) || (patch->is_rename)) {
3068 item = string_list_insert(&fn_table, patch->old_name);
3069 item->util = PATH_WAS_DELETED;
3073 static void prepare_fn_table(struct patch *patch)
3076 * store information about incoming file deletion
3078 while (patch) {
3079 if ((patch->new_name == NULL) || (patch->is_rename)) {
3080 struct string_list_item *item;
3081 item = string_list_insert(&fn_table, patch->old_name);
3082 item->util = PATH_TO_BE_DELETED;
3084 patch = patch->next;
3088 static int checkout_target(struct cache_entry *ce, struct stat *st)
3090 struct checkout costate;
3092 memset(&costate, 0, sizeof(costate));
3093 costate.base_dir = "";
3094 costate.refresh_cache = 1;
3095 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3096 return error(_("cannot checkout %s"), ce->name);
3097 return 0;
3100 static struct patch *previous_patch(struct patch *patch, int *gone)
3102 struct patch *previous;
3104 *gone = 0;
3105 if (patch->is_copy || patch->is_rename)
3106 return NULL; /* "git" patches do not depend on the order */
3108 previous = in_fn_table(patch->old_name);
3109 if (!previous)
3110 return NULL;
3112 if (to_be_deleted(previous))
3113 return NULL; /* the deletion hasn't happened yet */
3115 if (was_deleted(previous))
3116 *gone = 1;
3118 return previous;
3121 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3123 if (S_ISGITLINK(ce->ce_mode)) {
3124 if (!S_ISDIR(st->st_mode))
3125 return -1;
3126 return 0;
3128 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3131 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3133 static int load_patch_target(struct strbuf *buf,
3134 const struct cache_entry *ce,
3135 struct stat *st,
3136 const char *name,
3137 unsigned expected_mode)
3139 if (cached || check_index) {
3140 if (read_file_or_gitlink(ce, buf))
3141 return error(_("read of %s failed"), name);
3142 } else if (name) {
3143 if (S_ISGITLINK(expected_mode)) {
3144 if (ce)
3145 return read_file_or_gitlink(ce, buf);
3146 else
3147 return SUBMODULE_PATCH_WITHOUT_INDEX;
3148 } else if (has_symlink_leading_path(name, strlen(name))) {
3149 return error(_("reading from '%s' beyond a symbolic link"), name);
3150 } else {
3151 if (read_old_data(st, name, buf))
3152 return error(_("read of %s failed"), name);
3155 return 0;
3159 * We are about to apply "patch"; populate the "image" with the
3160 * current version we have, from the working tree or from the index,
3161 * depending on the situation e.g. --cached/--index. If we are
3162 * applying a non-git patch that incrementally updates the tree,
3163 * we read from the result of a previous diff.
3165 static int load_preimage(struct image *image,
3166 struct patch *patch, struct stat *st,
3167 const struct cache_entry *ce)
3169 struct strbuf buf = STRBUF_INIT;
3170 size_t len;
3171 char *img;
3172 struct patch *previous;
3173 int status;
3175 previous = previous_patch(patch, &status);
3176 if (status)
3177 return error(_("path %s has been renamed/deleted"),
3178 patch->old_name);
3179 if (previous) {
3180 /* We have a patched copy in memory; use that. */
3181 strbuf_add(&buf, previous->result, previous->resultsize);
3182 } else {
3183 status = load_patch_target(&buf, ce, st,
3184 patch->old_name, patch->old_mode);
3185 if (status < 0)
3186 return status;
3187 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3189 * There is no way to apply subproject
3190 * patch without looking at the index.
3191 * NEEDSWORK: shouldn't this be flagged
3192 * as an error???
3194 free_fragment_list(patch->fragments);
3195 patch->fragments = NULL;
3196 } else if (status) {
3197 return error(_("read of %s failed"), patch->old_name);
3201 img = strbuf_detach(&buf, &len);
3202 prepare_image(image, img, len, !patch->is_binary);
3203 return 0;
3206 static int three_way_merge(struct image *image,
3207 char *path,
3208 const unsigned char *base,
3209 const unsigned char *ours,
3210 const unsigned char *theirs)
3212 mmfile_t base_file, our_file, their_file;
3213 mmbuffer_t result = { NULL };
3214 int status;
3216 read_mmblob(&base_file, base);
3217 read_mmblob(&our_file, ours);
3218 read_mmblob(&their_file, theirs);
3219 status = ll_merge(&result, path,
3220 &base_file, "base",
3221 &our_file, "ours",
3222 &their_file, "theirs", NULL);
3223 free(base_file.ptr);
3224 free(our_file.ptr);
3225 free(their_file.ptr);
3226 if (status < 0 || !result.ptr) {
3227 free(result.ptr);
3228 return -1;
3230 clear_image(image);
3231 image->buf = result.ptr;
3232 image->len = result.size;
3234 return status;
3238 * When directly falling back to add/add three-way merge, we read from
3239 * the current contents of the new_name. In no cases other than that
3240 * this function will be called.
3242 static int load_current(struct image *image, struct patch *patch)
3244 struct strbuf buf = STRBUF_INIT;
3245 int status, pos;
3246 size_t len;
3247 char *img;
3248 struct stat st;
3249 struct cache_entry *ce;
3250 char *name = patch->new_name;
3251 unsigned mode = patch->new_mode;
3253 if (!patch->is_new)
3254 die("BUG: patch to %s is not a creation", patch->old_name);
3256 pos = cache_name_pos(name, strlen(name));
3257 if (pos < 0)
3258 return error(_("%s: does not exist in index"), name);
3259 ce = active_cache[pos];
3260 if (lstat(name, &st)) {
3261 if (errno != ENOENT)
3262 return error(_("%s: %s"), name, strerror(errno));
3263 if (checkout_target(ce, &st))
3264 return -1;
3266 if (verify_index_match(ce, &st))
3267 return error(_("%s: does not match index"), name);
3269 status = load_patch_target(&buf, ce, &st, name, mode);
3270 if (status < 0)
3271 return status;
3272 else if (status)
3273 return -1;
3274 img = strbuf_detach(&buf, &len);
3275 prepare_image(image, img, len, !patch->is_binary);
3276 return 0;
3279 static int try_threeway(struct image *image, struct patch *patch,
3280 struct stat *st, const struct cache_entry *ce)
3282 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3283 struct strbuf buf = STRBUF_INIT;
3284 size_t len;
3285 int status;
3286 char *img;
3287 struct image tmp_image;
3289 /* No point falling back to 3-way merge in these cases */
3290 if (patch->is_delete ||
3291 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3292 return -1;
3294 /* Preimage the patch was prepared for */
3295 if (patch->is_new)
3296 write_sha1_file("", 0, blob_type, pre_sha1);
3297 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3298 read_blob_object(&buf, pre_sha1, patch->old_mode))
3299 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3301 fprintf(stderr, "Falling back to three-way merge...\n");
3303 img = strbuf_detach(&buf, &len);
3304 prepare_image(&tmp_image, img, len, 1);
3305 /* Apply the patch to get the post image */
3306 if (apply_fragments(&tmp_image, patch) < 0) {
3307 clear_image(&tmp_image);
3308 return -1;
3310 /* post_sha1[] is theirs */
3311 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3312 clear_image(&tmp_image);
3314 /* our_sha1[] is ours */
3315 if (patch->is_new) {
3316 if (load_current(&tmp_image, patch))
3317 return error("cannot read the current contents of '%s'",
3318 patch->new_name);
3319 } else {
3320 if (load_preimage(&tmp_image, patch, st, ce))
3321 return error("cannot read the current contents of '%s'",
3322 patch->old_name);
3324 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3325 clear_image(&tmp_image);
3327 /* in-core three-way merge between post and our using pre as base */
3328 status = three_way_merge(image, patch->new_name,
3329 pre_sha1, our_sha1, post_sha1);
3330 if (status < 0) {
3331 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3332 return status;
3335 if (status) {
3336 patch->conflicted_threeway = 1;
3337 if (patch->is_new)
3338 hashclr(patch->threeway_stage[0]);
3339 else
3340 hashcpy(patch->threeway_stage[0], pre_sha1);
3341 hashcpy(patch->threeway_stage[1], our_sha1);
3342 hashcpy(patch->threeway_stage[2], post_sha1);
3343 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3344 } else {
3345 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3347 return 0;
3350 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3352 struct image image;
3354 if (load_preimage(&image, patch, st, ce) < 0)
3355 return -1;
3357 if (patch->direct_to_threeway ||
3358 apply_fragments(&image, patch) < 0) {
3359 /* Note: with --reject, apply_fragments() returns 0 */
3360 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3361 return -1;
3363 patch->result = image.buf;
3364 patch->resultsize = image.len;
3365 add_to_fn_table(patch);
3366 free(image.line_allocated);
3368 if (0 < patch->is_delete && patch->resultsize)
3369 return error(_("removal patch leaves file contents"));
3371 return 0;
3375 * If "patch" that we are looking at modifies or deletes what we have,
3376 * we would want it not to lose any local modification we have, either
3377 * in the working tree or in the index.
3379 * This also decides if a non-git patch is a creation patch or a
3380 * modification to an existing empty file. We do not check the state
3381 * of the current tree for a creation patch in this function; the caller
3382 * check_patch() separately makes sure (and errors out otherwise) that
3383 * the path the patch creates does not exist in the current tree.
3385 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3387 const char *old_name = patch->old_name;
3388 struct patch *previous = NULL;
3389 int stat_ret = 0, status;
3390 unsigned st_mode = 0;
3392 if (!old_name)
3393 return 0;
3395 assert(patch->is_new <= 0);
3396 previous = previous_patch(patch, &status);
3398 if (status)
3399 return error(_("path %s has been renamed/deleted"), old_name);
3400 if (previous) {
3401 st_mode = previous->new_mode;
3402 } else if (!cached) {
3403 stat_ret = lstat(old_name, st);
3404 if (stat_ret && errno != ENOENT)
3405 return error(_("%s: %s"), old_name, strerror(errno));
3408 if (check_index && !previous) {
3409 int pos = cache_name_pos(old_name, strlen(old_name));
3410 if (pos < 0) {
3411 if (patch->is_new < 0)
3412 goto is_new;
3413 return error(_("%s: does not exist in index"), old_name);
3415 *ce = active_cache[pos];
3416 if (stat_ret < 0) {
3417 if (checkout_target(*ce, st))
3418 return -1;
3420 if (!cached && verify_index_match(*ce, st))
3421 return error(_("%s: does not match index"), old_name);
3422 if (cached)
3423 st_mode = (*ce)->ce_mode;
3424 } else if (stat_ret < 0) {
3425 if (patch->is_new < 0)
3426 goto is_new;
3427 return error(_("%s: %s"), old_name, strerror(errno));
3430 if (!cached && !previous)
3431 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3433 if (patch->is_new < 0)
3434 patch->is_new = 0;
3435 if (!patch->old_mode)
3436 patch->old_mode = st_mode;
3437 if ((st_mode ^ patch->old_mode) & S_IFMT)
3438 return error(_("%s: wrong type"), old_name);
3439 if (st_mode != patch->old_mode)
3440 warning(_("%s has type %o, expected %o"),
3441 old_name, st_mode, patch->old_mode);
3442 if (!patch->new_mode && !patch->is_delete)
3443 patch->new_mode = st_mode;
3444 return 0;
3446 is_new:
3447 patch->is_new = 1;
3448 patch->is_delete = 0;
3449 free(patch->old_name);
3450 patch->old_name = NULL;
3451 return 0;
3455 #define EXISTS_IN_INDEX 1
3456 #define EXISTS_IN_WORKTREE 2
3458 static int check_to_create(const char *new_name, int ok_if_exists)
3460 struct stat nst;
3462 if (check_index &&
3463 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3464 !ok_if_exists)
3465 return EXISTS_IN_INDEX;
3466 if (cached)
3467 return 0;
3469 if (!lstat(new_name, &nst)) {
3470 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3471 return 0;
3473 * A leading component of new_name might be a symlink
3474 * that is going to be removed with this patch, but
3475 * still pointing at somewhere that has the path.
3476 * In such a case, path "new_name" does not exist as
3477 * far as git is concerned.
3479 if (has_symlink_leading_path(new_name, strlen(new_name)))
3480 return 0;
3482 return EXISTS_IN_WORKTREE;
3483 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3484 return error("%s: %s", new_name, strerror(errno));
3486 return 0;
3490 * We need to keep track of how symlinks in the preimage are
3491 * manipulated by the patches. A patch to add a/b/c where a/b
3492 * is a symlink should not be allowed to affect the directory
3493 * the symlink points at, but if the same patch removes a/b,
3494 * it is perfectly fine, as the patch removes a/b to make room
3495 * to create a directory a/b so that a/b/c can be created.
3497 static struct string_list symlink_changes;
3498 #define SYMLINK_GOES_AWAY 01
3499 #define SYMLINK_IN_RESULT 02
3501 static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3503 struct string_list_item *ent;
3505 ent = string_list_lookup(&symlink_changes, path);
3506 if (!ent) {
3507 ent = string_list_insert(&symlink_changes, path);
3508 ent->util = (void *)0;
3510 ent->util = (void *)(what | ((uintptr_t)ent->util));
3511 return (uintptr_t)ent->util;
3514 static uintptr_t check_symlink_changes(const char *path)
3516 struct string_list_item *ent;
3518 ent = string_list_lookup(&symlink_changes, path);
3519 if (!ent)
3520 return 0;
3521 return (uintptr_t)ent->util;
3524 static void prepare_symlink_changes(struct patch *patch)
3526 for ( ; patch; patch = patch->next) {
3527 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3528 (patch->is_rename || patch->is_delete))
3529 /* the symlink at patch->old_name is removed */
3530 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3532 if (patch->new_name && S_ISLNK(patch->new_mode))
3533 /* the symlink at patch->new_name is created or remains */
3534 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3538 static int path_is_beyond_symlink_1(struct strbuf *name)
3540 do {
3541 unsigned int change;
3543 while (--name->len && name->buf[name->len] != '/')
3544 ; /* scan backwards */
3545 if (!name->len)
3546 break;
3547 name->buf[name->len] = '\0';
3548 change = check_symlink_changes(name->buf);
3549 if (change & SYMLINK_IN_RESULT)
3550 return 1;
3551 if (change & SYMLINK_GOES_AWAY)
3553 * This cannot be "return 0", because we may
3554 * see a new one created at a higher level.
3556 continue;
3558 /* otherwise, check the preimage */
3559 if (check_index) {
3560 struct cache_entry *ce;
3562 ce = cache_file_exists(name->buf, name->len, ignore_case);
3563 if (ce && S_ISLNK(ce->ce_mode))
3564 return 1;
3565 } else {
3566 struct stat st;
3567 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3568 return 1;
3570 } while (1);
3571 return 0;
3574 static int path_is_beyond_symlink(const char *name_)
3576 int ret;
3577 struct strbuf name = STRBUF_INIT;
3579 assert(*name_ != '\0');
3580 strbuf_addstr(&name, name_);
3581 ret = path_is_beyond_symlink_1(&name);
3582 strbuf_release(&name);
3584 return ret;
3587 static void die_on_unsafe_path(struct patch *patch)
3589 const char *old_name = NULL;
3590 const char *new_name = NULL;
3591 if (patch->is_delete)
3592 old_name = patch->old_name;
3593 else if (!patch->is_new && !patch->is_copy)
3594 old_name = patch->old_name;
3595 if (!patch->is_delete)
3596 new_name = patch->new_name;
3598 if (old_name && !verify_path(old_name))
3599 die(_("invalid path '%s'"), old_name);
3600 if (new_name && !verify_path(new_name))
3601 die(_("invalid path '%s'"), new_name);
3605 * Check and apply the patch in-core; leave the result in patch->result
3606 * for the caller to write it out to the final destination.
3608 static int check_patch(struct patch *patch)
3610 struct stat st;
3611 const char *old_name = patch->old_name;
3612 const char *new_name = patch->new_name;
3613 const char *name = old_name ? old_name : new_name;
3614 struct cache_entry *ce = NULL;
3615 struct patch *tpatch;
3616 int ok_if_exists;
3617 int status;
3619 patch->rejected = 1; /* we will drop this after we succeed */
3621 status = check_preimage(patch, &ce, &st);
3622 if (status)
3623 return status;
3624 old_name = patch->old_name;
3627 * A type-change diff is always split into a patch to delete
3628 * old, immediately followed by a patch to create new (see
3629 * diff.c::run_diff()); in such a case it is Ok that the entry
3630 * to be deleted by the previous patch is still in the working
3631 * tree and in the index.
3633 * A patch to swap-rename between A and B would first rename A
3634 * to B and then rename B to A. While applying the first one,
3635 * the presence of B should not stop A from getting renamed to
3636 * B; ask to_be_deleted() about the later rename. Removal of
3637 * B and rename from A to B is handled the same way by asking
3638 * was_deleted().
3640 if ((tpatch = in_fn_table(new_name)) &&
3641 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3642 ok_if_exists = 1;
3643 else
3644 ok_if_exists = 0;
3646 if (new_name &&
3647 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3648 int err = check_to_create(new_name, ok_if_exists);
3650 if (err && threeway) {
3651 patch->direct_to_threeway = 1;
3652 } else switch (err) {
3653 case 0:
3654 break; /* happy */
3655 case EXISTS_IN_INDEX:
3656 return error(_("%s: already exists in index"), new_name);
3657 break;
3658 case EXISTS_IN_WORKTREE:
3659 return error(_("%s: already exists in working directory"),
3660 new_name);
3661 default:
3662 return err;
3665 if (!patch->new_mode) {
3666 if (0 < patch->is_new)
3667 patch->new_mode = S_IFREG | 0644;
3668 else
3669 patch->new_mode = patch->old_mode;
3673 if (new_name && old_name) {
3674 int same = !strcmp(old_name, new_name);
3675 if (!patch->new_mode)
3676 patch->new_mode = patch->old_mode;
3677 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3678 if (same)
3679 return error(_("new mode (%o) of %s does not "
3680 "match old mode (%o)"),
3681 patch->new_mode, new_name,
3682 patch->old_mode);
3683 else
3684 return error(_("new mode (%o) of %s does not "
3685 "match old mode (%o) of %s"),
3686 patch->new_mode, new_name,
3687 patch->old_mode, old_name);
3691 if (!unsafe_paths)
3692 die_on_unsafe_path(patch);
3695 * An attempt to read from or delete a path that is beyond a
3696 * symbolic link will be prevented by load_patch_target() that
3697 * is called at the beginning of apply_data() so we do not
3698 * have to worry about a patch marked with "is_delete" bit
3699 * here. We however need to make sure that the patch result
3700 * is not deposited to a path that is beyond a symbolic link
3701 * here.
3703 if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))
3704 return error(_("affected file '%s' is beyond a symbolic link"),
3705 patch->new_name);
3707 if (apply_data(patch, &st, ce) < 0)
3708 return error(_("%s: patch does not apply"), name);
3709 patch->rejected = 0;
3710 return 0;
3713 static int check_patch_list(struct patch *patch)
3715 int err = 0;
3717 prepare_symlink_changes(patch);
3718 prepare_fn_table(patch);
3719 while (patch) {
3720 if (apply_verbosely)
3721 say_patch_name(stderr,
3722 _("Checking patch %s..."), patch);
3723 err |= check_patch(patch);
3724 patch = patch->next;
3726 return err;
3729 /* This function tries to read the sha1 from the current index */
3730 static int get_current_sha1(const char *path, unsigned char *sha1)
3732 int pos;
3734 if (read_cache() < 0)
3735 return -1;
3736 pos = cache_name_pos(path, strlen(path));
3737 if (pos < 0)
3738 return -1;
3739 hashcpy(sha1, active_cache[pos]->sha1);
3740 return 0;
3743 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3746 * A usable gitlink patch has only one fragment (hunk) that looks like:
3747 * @@ -1 +1 @@
3748 * -Subproject commit <old sha1>
3749 * +Subproject commit <new sha1>
3750 * or
3751 * @@ -1 +0,0 @@
3752 * -Subproject commit <old sha1>
3753 * for a removal patch.
3755 struct fragment *hunk = p->fragments;
3756 static const char heading[] = "-Subproject commit ";
3757 char *preimage;
3759 if (/* does the patch have only one hunk? */
3760 hunk && !hunk->next &&
3761 /* is its preimage one line? */
3762 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3763 /* does preimage begin with the heading? */
3764 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3765 !prefixcmp(++preimage, heading) &&
3766 /* does it record full SHA-1? */
3767 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3768 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3769 /* does the abbreviated name on the index line agree with it? */
3770 !prefixcmp(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3771 return 0; /* it all looks fine */
3773 /* we may have full object name on the index line */
3774 return get_sha1_hex(p->old_sha1_prefix, sha1);
3777 /* Build an index that contains the just the files needed for a 3way merge */
3778 static void build_fake_ancestor(struct patch *list, const char *filename)
3780 struct patch *patch;
3781 struct index_state result = { NULL };
3782 int fd;
3784 /* Once we start supporting the reverse patch, it may be
3785 * worth showing the new sha1 prefix, but until then...
3787 for (patch = list; patch; patch = patch->next) {
3788 unsigned char sha1[20];
3789 struct cache_entry *ce;
3790 const char *name;
3792 name = patch->old_name ? patch->old_name : patch->new_name;
3793 if (0 < patch->is_new)
3794 continue;
3796 if (S_ISGITLINK(patch->old_mode)) {
3797 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3798 ; /* ok, the textual part looks sane */
3799 else
3800 die("sha1 information is lacking or useless for submoule %s",
3801 name);
3802 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3803 ; /* ok */
3804 } else if (!patch->lines_added && !patch->lines_deleted) {
3805 /* mode-only change: update the current */
3806 if (get_current_sha1(patch->old_name, sha1))
3807 die("mode change for %s, which is not "
3808 "in current HEAD", name);
3809 } else
3810 die("sha1 information is lacking or useless "
3811 "(%s).", name);
3813 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3814 if (!ce)
3815 die(_("make_cache_entry failed for path '%s'"), name);
3816 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3817 die ("Could not add %s to temporary index", name);
3820 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3821 if (fd < 0 || write_index(&result, fd) || close(fd))
3822 die ("Could not write temporary index to %s", filename);
3824 discard_index(&result);
3827 static void stat_patch_list(struct patch *patch)
3829 int files, adds, dels;
3831 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3832 files++;
3833 adds += patch->lines_added;
3834 dels += patch->lines_deleted;
3835 show_stats(patch);
3838 print_stat_summary(stdout, files, adds, dels);
3841 static void numstat_patch_list(struct patch *patch)
3843 for ( ; patch; patch = patch->next) {
3844 const char *name;
3845 name = patch->new_name ? patch->new_name : patch->old_name;
3846 if (patch->is_binary)
3847 printf("-\t-\t");
3848 else
3849 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3850 write_name_quoted(name, stdout, line_termination);
3854 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3856 if (mode)
3857 printf(" %s mode %06o %s\n", newdelete, mode, name);
3858 else
3859 printf(" %s %s\n", newdelete, name);
3862 static void show_mode_change(struct patch *p, int show_name)
3864 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3865 if (show_name)
3866 printf(" mode change %06o => %06o %s\n",
3867 p->old_mode, p->new_mode, p->new_name);
3868 else
3869 printf(" mode change %06o => %06o\n",
3870 p->old_mode, p->new_mode);
3874 static void show_rename_copy(struct patch *p)
3876 const char *renamecopy = p->is_rename ? "rename" : "copy";
3877 const char *old, *new;
3879 /* Find common prefix */
3880 old = p->old_name;
3881 new = p->new_name;
3882 while (1) {
3883 const char *slash_old, *slash_new;
3884 slash_old = strchr(old, '/');
3885 slash_new = strchr(new, '/');
3886 if (!slash_old ||
3887 !slash_new ||
3888 slash_old - old != slash_new - new ||
3889 memcmp(old, new, slash_new - new))
3890 break;
3891 old = slash_old + 1;
3892 new = slash_new + 1;
3894 /* p->old_name thru old is the common prefix, and old and new
3895 * through the end of names are renames
3897 if (old != p->old_name)
3898 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3899 (int)(old - p->old_name), p->old_name,
3900 old, new, p->score);
3901 else
3902 printf(" %s %s => %s (%d%%)\n", renamecopy,
3903 p->old_name, p->new_name, p->score);
3904 show_mode_change(p, 0);
3907 static void summary_patch_list(struct patch *patch)
3909 struct patch *p;
3911 for (p = patch; p; p = p->next) {
3912 if (p->is_new)
3913 show_file_mode_name("create", p->new_mode, p->new_name);
3914 else if (p->is_delete)
3915 show_file_mode_name("delete", p->old_mode, p->old_name);
3916 else {
3917 if (p->is_rename || p->is_copy)
3918 show_rename_copy(p);
3919 else {
3920 if (p->score) {
3921 printf(" rewrite %s (%d%%)\n",
3922 p->new_name, p->score);
3923 show_mode_change(p, 0);
3925 else
3926 show_mode_change(p, 1);
3932 static void patch_stats(struct patch *patch)
3934 int lines = patch->lines_added + patch->lines_deleted;
3936 if (lines > max_change)
3937 max_change = lines;
3938 if (patch->old_name) {
3939 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3940 if (!len)
3941 len = strlen(patch->old_name);
3942 if (len > max_len)
3943 max_len = len;
3945 if (patch->new_name) {
3946 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3947 if (!len)
3948 len = strlen(patch->new_name);
3949 if (len > max_len)
3950 max_len = len;
3954 static void remove_file(struct patch *patch, int rmdir_empty)
3956 if (update_index) {
3957 if (remove_file_from_cache(patch->old_name) < 0)
3958 die(_("unable to remove %s from index"), patch->old_name);
3960 if (!cached) {
3961 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3962 remove_path(patch->old_name);
3967 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3969 struct stat st;
3970 struct cache_entry *ce;
3971 int namelen = strlen(path);
3972 unsigned ce_size = cache_entry_size(namelen);
3974 if (!update_index)
3975 return;
3977 ce = xcalloc(1, ce_size);
3978 memcpy(ce->name, path, namelen);
3979 ce->ce_mode = create_ce_mode(mode);
3980 ce->ce_flags = create_ce_flags(0);
3981 ce->ce_namelen = namelen;
3982 if (S_ISGITLINK(mode)) {
3983 const char *s = buf;
3985 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3986 die(_("corrupt patch for submodule %s"), path);
3987 } else {
3988 if (!cached) {
3989 if (lstat(path, &st) < 0)
3990 die_errno(_("unable to stat newly created file '%s'"),
3991 path);
3992 fill_stat_cache_info(ce, &st);
3994 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3995 die(_("unable to create backing store for newly created file %s"), path);
3997 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3998 die(_("unable to add cache entry for %s"), path);
4001 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4003 int fd;
4004 struct strbuf nbuf = STRBUF_INIT;
4006 if (S_ISGITLINK(mode)) {
4007 struct stat st;
4008 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4009 return 0;
4010 return mkdir(path, 0777);
4013 if (has_symlinks && S_ISLNK(mode))
4014 /* Although buf:size is counted string, it also is NUL
4015 * terminated.
4017 return symlink(buf, path);
4019 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4020 if (fd < 0)
4021 return -1;
4023 if (convert_to_working_tree(path, buf, size, &nbuf)) {
4024 size = nbuf.len;
4025 buf = nbuf.buf;
4027 write_or_die(fd, buf, size);
4028 strbuf_release(&nbuf);
4030 if (close(fd) < 0)
4031 die_errno(_("closing file '%s'"), path);
4032 return 0;
4036 * We optimistically assume that the directories exist,
4037 * which is true 99% of the time anyway. If they don't,
4038 * we create them and try again.
4040 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
4042 if (cached)
4043 return;
4044 if (!try_create_file(path, mode, buf, size))
4045 return;
4047 if (errno == ENOENT) {
4048 if (safe_create_leading_directories(path))
4049 return;
4050 if (!try_create_file(path, mode, buf, size))
4051 return;
4054 if (errno == EEXIST || errno == EACCES) {
4055 /* We may be trying to create a file where a directory
4056 * used to be.
4058 struct stat st;
4059 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4060 errno = EEXIST;
4063 if (errno == EEXIST) {
4064 unsigned int nr = getpid();
4066 for (;;) {
4067 char newpath[PATH_MAX];
4068 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4069 if (!try_create_file(newpath, mode, buf, size)) {
4070 if (!rename(newpath, path))
4071 return;
4072 unlink_or_warn(newpath);
4073 break;
4075 if (errno != EEXIST)
4076 break;
4077 ++nr;
4080 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4083 static void add_conflicted_stages_file(struct patch *patch)
4085 int stage, namelen;
4086 unsigned ce_size, mode;
4087 struct cache_entry *ce;
4089 if (!update_index)
4090 return;
4091 namelen = strlen(patch->new_name);
4092 ce_size = cache_entry_size(namelen);
4093 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4095 remove_file_from_cache(patch->new_name);
4096 for (stage = 1; stage < 4; stage++) {
4097 if (is_null_sha1(patch->threeway_stage[stage - 1]))
4098 continue;
4099 ce = xcalloc(1, ce_size);
4100 memcpy(ce->name, patch->new_name, namelen);
4101 ce->ce_mode = create_ce_mode(mode);
4102 ce->ce_flags = create_ce_flags(stage);
4103 ce->ce_namelen = namelen;
4104 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
4105 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4106 die(_("unable to add cache entry for %s"), patch->new_name);
4110 static void create_file(struct patch *patch)
4112 char *path = patch->new_name;
4113 unsigned mode = patch->new_mode;
4114 unsigned long size = patch->resultsize;
4115 char *buf = patch->result;
4117 if (!mode)
4118 mode = S_IFREG | 0644;
4119 create_one_file(path, mode, buf, size);
4121 if (patch->conflicted_threeway)
4122 add_conflicted_stages_file(patch);
4123 else
4124 add_index_file(path, mode, buf, size);
4127 /* phase zero is to remove, phase one is to create */
4128 static void write_out_one_result(struct patch *patch, int phase)
4130 if (patch->is_delete > 0) {
4131 if (phase == 0)
4132 remove_file(patch, 1);
4133 return;
4135 if (patch->is_new > 0 || patch->is_copy) {
4136 if (phase == 1)
4137 create_file(patch);
4138 return;
4141 * Rename or modification boils down to the same
4142 * thing: remove the old, write the new
4144 if (phase == 0)
4145 remove_file(patch, patch->is_rename);
4146 if (phase == 1)
4147 create_file(patch);
4150 static int write_out_one_reject(struct patch *patch)
4152 FILE *rej;
4153 char namebuf[PATH_MAX];
4154 struct fragment *frag;
4155 int cnt = 0;
4156 struct strbuf sb = STRBUF_INIT;
4158 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4159 if (!frag->rejected)
4160 continue;
4161 cnt++;
4164 if (!cnt) {
4165 if (apply_verbosely)
4166 say_patch_name(stderr,
4167 _("Applied patch %s cleanly."), patch);
4168 return 0;
4171 /* This should not happen, because a removal patch that leaves
4172 * contents are marked "rejected" at the patch level.
4174 if (!patch->new_name)
4175 die(_("internal error"));
4177 /* Say this even without --verbose */
4178 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4179 "Applying patch %%s with %d rejects...",
4180 cnt),
4181 cnt);
4182 say_patch_name(stderr, sb.buf, patch);
4183 strbuf_release(&sb);
4185 cnt = strlen(patch->new_name);
4186 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4187 cnt = ARRAY_SIZE(namebuf) - 5;
4188 warning(_("truncating .rej filename to %.*s.rej"),
4189 cnt - 1, patch->new_name);
4191 memcpy(namebuf, patch->new_name, cnt);
4192 memcpy(namebuf + cnt, ".rej", 5);
4194 rej = fopen(namebuf, "w");
4195 if (!rej)
4196 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4198 /* Normal git tools never deal with .rej, so do not pretend
4199 * this is a git patch by saying --git nor give extended
4200 * headers. While at it, maybe please "kompare" that wants
4201 * the trailing TAB and some garbage at the end of line ;-).
4203 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4204 patch->new_name, patch->new_name);
4205 for (cnt = 1, frag = patch->fragments;
4206 frag;
4207 cnt++, frag = frag->next) {
4208 if (!frag->rejected) {
4209 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4210 continue;
4212 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4213 fprintf(rej, "%.*s", frag->size, frag->patch);
4214 if (frag->patch[frag->size-1] != '\n')
4215 fputc('\n', rej);
4217 fclose(rej);
4218 return -1;
4221 static int write_out_results(struct patch *list)
4223 int phase;
4224 int errs = 0;
4225 struct patch *l;
4226 struct string_list cpath = STRING_LIST_INIT_DUP;
4228 for (phase = 0; phase < 2; phase++) {
4229 l = list;
4230 while (l) {
4231 if (l->rejected)
4232 errs = 1;
4233 else {
4234 write_out_one_result(l, phase);
4235 if (phase == 1) {
4236 if (write_out_one_reject(l))
4237 errs = 1;
4238 if (l->conflicted_threeway) {
4239 string_list_append(&cpath, l->new_name);
4240 errs = 1;
4244 l = l->next;
4248 if (cpath.nr) {
4249 struct string_list_item *item;
4251 sort_string_list(&cpath);
4252 for_each_string_list_item(item, &cpath)
4253 fprintf(stderr, "U %s\n", item->string);
4254 string_list_clear(&cpath, 0);
4256 rerere(0);
4259 return errs;
4262 static struct lock_file lock_file;
4264 static struct string_list limit_by_name;
4265 static int has_include;
4266 static void add_name_limit(const char *name, int exclude)
4268 struct string_list_item *it;
4270 it = string_list_append(&limit_by_name, name);
4271 it->util = exclude ? NULL : (void *) 1;
4274 static int use_patch(struct patch *p)
4276 const char *pathname = p->new_name ? p->new_name : p->old_name;
4277 int i;
4279 /* Paths outside are not touched regardless of "--include" */
4280 if (0 < prefix_length) {
4281 int pathlen = strlen(pathname);
4282 if (pathlen <= prefix_length ||
4283 memcmp(prefix, pathname, prefix_length))
4284 return 0;
4287 /* See if it matches any of exclude/include rule */
4288 for (i = 0; i < limit_by_name.nr; i++) {
4289 struct string_list_item *it = &limit_by_name.items[i];
4290 if (!fnmatch(it->string, pathname, 0))
4291 return (it->util != NULL);
4295 * If we had any include, a path that does not match any rule is
4296 * not used. Otherwise, we saw bunch of exclude rules (or none)
4297 * and such a path is used.
4299 return !has_include;
4303 static void prefix_one(char **name)
4305 char *old_name = *name;
4306 if (!old_name)
4307 return;
4308 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
4309 free(old_name);
4312 static void prefix_patches(struct patch *p)
4314 if (!prefix || p->is_toplevel_relative)
4315 return;
4316 for ( ; p; p = p->next) {
4317 prefix_one(&p->new_name);
4318 prefix_one(&p->old_name);
4322 #define INACCURATE_EOF (1<<0)
4323 #define RECOUNT (1<<1)
4325 static int apply_patch(int fd, const char *filename, int options)
4327 size_t offset;
4328 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4329 struct patch *list = NULL, **listp = &list;
4330 int skipped_patch = 0;
4332 patch_input_file = filename;
4333 read_patch_file(&buf, fd);
4334 offset = 0;
4335 while (offset < buf.len) {
4336 struct patch *patch;
4337 int nr;
4339 patch = xcalloc(1, sizeof(*patch));
4340 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4341 patch->recount = !!(options & RECOUNT);
4342 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4343 if (nr < 0)
4344 break;
4345 if (apply_in_reverse)
4346 reverse_patches(patch);
4347 if (prefix)
4348 prefix_patches(patch);
4349 if (use_patch(patch)) {
4350 patch_stats(patch);
4351 *listp = patch;
4352 listp = &patch->next;
4354 else {
4355 free_patch(patch);
4356 skipped_patch++;
4358 offset += nr;
4361 if (!list && !skipped_patch)
4362 die(_("unrecognized input"));
4364 if (whitespace_error && (ws_error_action == die_on_ws_error))
4365 apply = 0;
4367 update_index = check_index && apply;
4368 if (update_index && newfd < 0)
4369 newfd = hold_locked_index(&lock_file, 1);
4371 if (check_index) {
4372 if (read_cache() < 0)
4373 die(_("unable to read index file"));
4376 if ((check || apply) &&
4377 check_patch_list(list) < 0 &&
4378 !apply_with_reject)
4379 exit(1);
4381 if (apply && write_out_results(list)) {
4382 if (apply_with_reject)
4383 exit(1);
4384 /* with --3way, we still need to write the index out */
4385 return 1;
4388 if (fake_ancestor)
4389 build_fake_ancestor(list, fake_ancestor);
4391 if (diffstat)
4392 stat_patch_list(list);
4394 if (numstat)
4395 numstat_patch_list(list);
4397 if (summary)
4398 summary_patch_list(list);
4400 free_patch_list(list);
4401 strbuf_release(&buf);
4402 string_list_clear(&fn_table, 0);
4403 return 0;
4406 static int git_apply_config(const char *var, const char *value, void *cb)
4408 if (!strcmp(var, "apply.whitespace"))
4409 return git_config_string(&apply_default_whitespace, var, value);
4410 else if (!strcmp(var, "apply.ignorewhitespace"))
4411 return git_config_string(&apply_default_ignorewhitespace, var, value);
4412 return git_default_config(var, value, cb);
4415 static int option_parse_exclude(const struct option *opt,
4416 const char *arg, int unset)
4418 add_name_limit(arg, 1);
4419 return 0;
4422 static int option_parse_include(const struct option *opt,
4423 const char *arg, int unset)
4425 add_name_limit(arg, 0);
4426 has_include = 1;
4427 return 0;
4430 static int option_parse_p(const struct option *opt,
4431 const char *arg, int unset)
4433 p_value = atoi(arg);
4434 p_value_known = 1;
4435 return 0;
4438 static int option_parse_z(const struct option *opt,
4439 const char *arg, int unset)
4441 if (unset)
4442 line_termination = '\n';
4443 else
4444 line_termination = 0;
4445 return 0;
4448 static int option_parse_space_change(const struct option *opt,
4449 const char *arg, int unset)
4451 if (unset)
4452 ws_ignore_action = ignore_ws_none;
4453 else
4454 ws_ignore_action = ignore_ws_change;
4455 return 0;
4458 static int option_parse_whitespace(const struct option *opt,
4459 const char *arg, int unset)
4461 const char **whitespace_option = opt->value;
4463 *whitespace_option = arg;
4464 parse_whitespace_option(arg);
4465 return 0;
4468 static int option_parse_directory(const struct option *opt,
4469 const char *arg, int unset)
4471 root_len = strlen(arg);
4472 if (root_len && arg[root_len - 1] != '/') {
4473 char *new_root;
4474 root = new_root = xmalloc(root_len + 2);
4475 strcpy(new_root, arg);
4476 strcpy(new_root + root_len++, "/");
4477 } else
4478 root = arg;
4479 return 0;
4482 int cmd_apply(int argc, const char **argv, const char *prefix_)
4484 int i;
4485 int errs = 0;
4486 int is_not_gitdir = !startup_info->have_repository;
4487 int force_apply = 0;
4489 const char *whitespace_option = NULL;
4491 struct option builtin_apply_options[] = {
4492 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4493 N_("don't apply changes matching the given path"),
4494 0, option_parse_exclude },
4495 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4496 N_("apply changes matching the given path"),
4497 0, option_parse_include },
4498 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4499 N_("remove <num> leading slashes from traditional diff paths"),
4500 0, option_parse_p },
4501 OPT_BOOL(0, "no-add", &no_add,
4502 N_("ignore additions made by the patch")),
4503 OPT_BOOL(0, "stat", &diffstat,
4504 N_("instead of applying the patch, output diffstat for the input")),
4505 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4506 OPT_NOOP_NOARG(0, "binary"),
4507 OPT_BOOL(0, "numstat", &numstat,
4508 N_("show number of added and deleted lines in decimal notation")),
4509 OPT_BOOL(0, "summary", &summary,
4510 N_("instead of applying the patch, output a summary for the input")),
4511 OPT_BOOL(0, "check", &check,
4512 N_("instead of applying the patch, see if the patch is applicable")),
4513 OPT_BOOL(0, "index", &check_index,
4514 N_("make sure the patch is applicable to the current index")),
4515 OPT_BOOL(0, "cached", &cached,
4516 N_("apply a patch without touching the working tree")),
4517 OPT_BOOL(0, "unsafe-paths", &unsafe_paths,
4518 N_("accept a patch that touches outside the working area")),
4519 OPT_BOOL(0, "apply", &force_apply,
4520 N_("also apply the patch (use with --stat/--summary/--check)")),
4521 OPT_BOOL('3', "3way", &threeway,
4522 N_( "attempt three-way merge if a patch does not apply")),
4523 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4524 N_("build a temporary index based on embedded index information")),
4525 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4526 N_("paths are separated with NUL character"),
4527 PARSE_OPT_NOARG, option_parse_z },
4528 OPT_INTEGER('C', NULL, &p_context,
4529 N_("ensure at least <n> lines of context match")),
4530 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4531 N_("detect new or modified lines that have whitespace errors"),
4532 0, option_parse_whitespace },
4533 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4534 N_("ignore changes in whitespace when finding context"),
4535 PARSE_OPT_NOARG, option_parse_space_change },
4536 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4537 N_("ignore changes in whitespace when finding context"),
4538 PARSE_OPT_NOARG, option_parse_space_change },
4539 OPT_BOOL('R', "reverse", &apply_in_reverse,
4540 N_("apply the patch in reverse")),
4541 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4542 N_("don't expect at least one line of context")),
4543 OPT_BOOL(0, "reject", &apply_with_reject,
4544 N_("leave the rejected hunks in corresponding *.rej files")),
4545 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4546 N_("allow overlapping hunks")),
4547 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4548 OPT_BIT(0, "inaccurate-eof", &options,
4549 N_("tolerate incorrectly detected missing new-line at the end of file"),
4550 INACCURATE_EOF),
4551 OPT_BIT(0, "recount", &options,
4552 N_("do not trust the line counts in the hunk headers"),
4553 RECOUNT),
4554 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4555 N_("prepend <root> to all filenames"),
4556 0, option_parse_directory },
4557 OPT_END()
4560 prefix = prefix_;
4561 prefix_length = prefix ? strlen(prefix) : 0;
4562 git_config(git_apply_config, NULL);
4563 if (apply_default_whitespace)
4564 parse_whitespace_option(apply_default_whitespace);
4565 if (apply_default_ignorewhitespace)
4566 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4568 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4569 apply_usage, 0);
4571 if (apply_with_reject && threeway)
4572 die("--reject and --3way cannot be used together.");
4573 if (cached && threeway)
4574 die("--cached and --3way cannot be used together.");
4575 if (threeway) {
4576 if (is_not_gitdir)
4577 die(_("--3way outside a repository"));
4578 check_index = 1;
4580 if (apply_with_reject)
4581 apply = apply_verbosely = 1;
4582 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4583 apply = 0;
4584 if (check_index && is_not_gitdir)
4585 die(_("--index outside a repository"));
4586 if (cached) {
4587 if (is_not_gitdir)
4588 die(_("--cached outside a repository"));
4589 check_index = 1;
4591 if (check_index)
4592 unsafe_paths = 0;
4594 for (i = 0; i < argc; i++) {
4595 const char *arg = argv[i];
4596 int fd;
4598 if (!strcmp(arg, "-")) {
4599 errs |= apply_patch(0, "<stdin>", options);
4600 read_stdin = 0;
4601 continue;
4602 } else if (0 < prefix_length)
4603 arg = prefix_filename(prefix, prefix_length, arg);
4605 fd = open(arg, O_RDONLY);
4606 if (fd < 0)
4607 die_errno(_("can't open patch '%s'"), arg);
4608 read_stdin = 0;
4609 set_default_whitespace_mode(whitespace_option);
4610 errs |= apply_patch(fd, arg, options);
4611 close(fd);
4613 set_default_whitespace_mode(whitespace_option);
4614 if (read_stdin)
4615 errs |= apply_patch(0, "<stdin>", options);
4616 if (whitespace_error) {
4617 if (squelch_whitespace_errors &&
4618 squelch_whitespace_errors < whitespace_error) {
4619 int squelched =
4620 whitespace_error - squelch_whitespace_errors;
4621 warning(Q_("squelched %d whitespace error",
4622 "squelched %d whitespace errors",
4623 squelched),
4624 squelched);
4626 if (ws_error_action == die_on_ws_error)
4627 die(Q_("%d line adds whitespace errors.",
4628 "%d lines add whitespace errors.",
4629 whitespace_error),
4630 whitespace_error);
4631 if (applied_after_fixing_ws && apply)
4632 warning("%d line%s applied after"
4633 " fixing whitespace errors.",
4634 applied_after_fixing_ws,
4635 applied_after_fixing_ws == 1 ? "" : "s");
4636 else if (whitespace_error)
4637 warning(Q_("%d line adds whitespace errors.",
4638 "%d lines add whitespace errors.",
4639 whitespace_error),
4640 whitespace_error);
4643 if (update_index) {
4644 if (write_cache(newfd, active_cache, active_nr) ||
4645 commit_locked_index(&lock_file))
4646 die(_("Unable to write new index file"));
4649 return !!errs;