read-cache: introduce chmod_index_entry
[git.git] / builtin / apply.c
blob8e4da2e1bdaf02590289f54750a9602281f4a365
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 "lockfile.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "builtin.h"
16 #include "string-list.h"
17 #include "dir.h"
18 #include "diff.h"
19 #include "parse-options.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "rerere.h"
25 * --check turns on checking that the working tree matches the
26 * files that are being modified, but doesn't apply the patch
27 * --stat does just a diffstat, and doesn't actually apply
28 * --numstat does numeric diffstat, and doesn't actually apply
29 * --index-info shows the old and new index info for paths if available.
30 * --index updates the cache as well.
31 * --cached updates only the cache without ever touching the working tree.
33 static const char *prefix;
34 static int prefix_length = -1;
35 static int newfd = -1;
37 static int unidiff_zero;
38 static int p_value = 1;
39 static int p_value_known;
40 static int check_index;
41 static int update_index;
42 static int cached;
43 static int diffstat;
44 static int numstat;
45 static int summary;
46 static int check;
47 static int apply = 1;
48 static int apply_in_reverse;
49 static int apply_with_reject;
50 static int apply_verbosely;
51 static int allow_overlap;
52 static int no_add;
53 static int threeway;
54 static int unsafe_paths;
55 static const char *fake_ancestor;
56 static int line_termination = '\n';
57 static unsigned int p_context = UINT_MAX;
58 static const char * const apply_usage[] = {
59 N_("git apply [<options>] [<patch>...]"),
60 NULL
63 static enum ws_error_action {
64 nowarn_ws_error,
65 warn_on_ws_error,
66 die_on_ws_error,
67 correct_ws_error
68 } ws_error_action = warn_on_ws_error;
69 static int whitespace_error;
70 static int squelch_whitespace_errors = 5;
71 static int applied_after_fixing_ws;
73 static enum ws_ignore {
74 ignore_ws_none,
75 ignore_ws_change
76 } ws_ignore_action = ignore_ws_none;
79 static const char *patch_input_file;
80 static struct strbuf root = STRBUF_INIT;
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 struct object_id threeway_stage[3];
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 whitespaces, if both begin with whitespace */
305 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
306 while (isspace(*s1) && (s1 <= last1))
307 s1++;
308 while (isspace(*s2) && (s2 <= last2))
309 s2++;
311 /* early return if both lines are empty */
312 if ((s1 > last1) && (s2 > last2))
313 return 1;
314 while (!result) {
315 result = *s1++ - *s2++;
317 * Skip whitespace inside. We check for whitespace on
318 * both buffers because we don't want "a b" to match
319 * "ab"
321 if (isspace(*s1) && isspace(*s2)) {
322 while (isspace(*s1) && s1 <= last1)
323 s1++;
324 while (isspace(*s2) && s2 <= last2)
325 s2++;
328 * If we reached the end on one side only,
329 * lines don't match
331 if (
332 ((s2 > last2) && (s1 <= last1)) ||
333 ((s1 > last1) && (s2 <= last2)))
334 return 0;
335 if ((s1 > last1) && (s2 > last2))
336 break;
339 return !result;
342 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
344 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
345 img->line_allocated[img->nr].len = len;
346 img->line_allocated[img->nr].hash = hash_line(bol, len);
347 img->line_allocated[img->nr].flag = flag;
348 img->nr++;
352 * "buf" has the file contents to be patched (read from various sources).
353 * attach it to "image" and add line-based index to it.
354 * "image" now owns the "buf".
356 static void prepare_image(struct image *image, char *buf, size_t len,
357 int prepare_linetable)
359 const char *cp, *ep;
361 memset(image, 0, sizeof(*image));
362 image->buf = buf;
363 image->len = len;
365 if (!prepare_linetable)
366 return;
368 ep = image->buf + image->len;
369 cp = image->buf;
370 while (cp < ep) {
371 const char *next;
372 for (next = cp; next < ep && *next != '\n'; next++)
374 if (next < ep)
375 next++;
376 add_line_info(image, cp, next - cp, 0);
377 cp = next;
379 image->line = image->line_allocated;
382 static void clear_image(struct image *image)
384 free(image->buf);
385 free(image->line_allocated);
386 memset(image, 0, sizeof(*image));
389 /* fmt must contain _one_ %s and no other substitution */
390 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
392 struct strbuf sb = STRBUF_INIT;
394 if (patch->old_name && patch->new_name &&
395 strcmp(patch->old_name, patch->new_name)) {
396 quote_c_style(patch->old_name, &sb, NULL, 0);
397 strbuf_addstr(&sb, " => ");
398 quote_c_style(patch->new_name, &sb, NULL, 0);
399 } else {
400 const char *n = patch->new_name;
401 if (!n)
402 n = patch->old_name;
403 quote_c_style(n, &sb, NULL, 0);
405 fprintf(output, fmt, sb.buf);
406 fputc('\n', output);
407 strbuf_release(&sb);
410 #define SLOP (16)
412 static void read_patch_file(struct strbuf *sb, int fd)
414 if (strbuf_read(sb, fd, 0) < 0)
415 die_errno("git apply: failed to read");
418 * Make sure that we have some slop in the buffer
419 * so that we can do speculative "memcmp" etc, and
420 * see to it that it is NUL-filled.
422 strbuf_grow(sb, SLOP);
423 memset(sb->buf + sb->len, 0, SLOP);
426 static unsigned long linelen(const char *buffer, unsigned long size)
428 unsigned long len = 0;
429 while (size--) {
430 len++;
431 if (*buffer++ == '\n')
432 break;
434 return len;
437 static int is_dev_null(const char *str)
439 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
442 #define TERM_SPACE 1
443 #define TERM_TAB 2
445 static int name_terminate(const char *name, int namelen, int c, int terminate)
447 if (c == ' ' && !(terminate & TERM_SPACE))
448 return 0;
449 if (c == '\t' && !(terminate & TERM_TAB))
450 return 0;
452 return 1;
455 /* remove double slashes to make --index work with such filenames */
456 static char *squash_slash(char *name)
458 int i = 0, j = 0;
460 if (!name)
461 return NULL;
463 while (name[i]) {
464 if ((name[j++] = name[i++]) == '/')
465 while (name[i] == '/')
466 i++;
468 name[j] = '\0';
469 return name;
472 static char *find_name_gnu(const char *line, const char *def, int p_value)
474 struct strbuf name = STRBUF_INIT;
475 char *cp;
478 * Proposed "new-style" GNU patch/diff format; see
479 * http://marc.info/?l=git&m=112927316408690&w=2
481 if (unquote_c_style(&name, line, NULL)) {
482 strbuf_release(&name);
483 return NULL;
486 for (cp = name.buf; p_value; p_value--) {
487 cp = strchr(cp, '/');
488 if (!cp) {
489 strbuf_release(&name);
490 return NULL;
492 cp++;
495 strbuf_remove(&name, 0, cp - name.buf);
496 if (root.len)
497 strbuf_insert(&name, 0, root.buf, root.len);
498 return squash_slash(strbuf_detach(&name, NULL));
501 static size_t sane_tz_len(const char *line, size_t len)
503 const char *tz, *p;
505 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
506 return 0;
507 tz = line + len - strlen(" +0500");
509 if (tz[1] != '+' && tz[1] != '-')
510 return 0;
512 for (p = tz + 2; p != line + len; p++)
513 if (!isdigit(*p))
514 return 0;
516 return line + len - tz;
519 static size_t tz_with_colon_len(const char *line, size_t len)
521 const char *tz, *p;
523 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
524 return 0;
525 tz = line + len - strlen(" +08:00");
527 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
528 return 0;
529 p = tz + 2;
530 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
531 !isdigit(*p++) || !isdigit(*p++))
532 return 0;
534 return line + len - tz;
537 static size_t date_len(const char *line, size_t len)
539 const char *date, *p;
541 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
542 return 0;
543 p = date = line + len - strlen("72-02-05");
545 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
546 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
547 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
548 return 0;
550 if (date - line >= strlen("19") &&
551 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
552 date -= strlen("19");
554 return line + len - date;
557 static size_t short_time_len(const char *line, size_t len)
559 const char *time, *p;
561 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
562 return 0;
563 p = time = line + len - strlen(" 07:01:32");
565 /* Permit 1-digit hours? */
566 if (*p++ != ' ' ||
567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
568 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
569 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
570 return 0;
572 return line + len - time;
575 static size_t fractional_time_len(const char *line, size_t len)
577 const char *p;
578 size_t n;
580 /* Expected format: 19:41:17.620000023 */
581 if (!len || !isdigit(line[len - 1]))
582 return 0;
583 p = line + len - 1;
585 /* Fractional seconds. */
586 while (p > line && isdigit(*p))
587 p--;
588 if (*p != '.')
589 return 0;
591 /* Hours, minutes, and whole seconds. */
592 n = short_time_len(line, p - line);
593 if (!n)
594 return 0;
596 return line + len - p + n;
599 static size_t trailing_spaces_len(const char *line, size_t len)
601 const char *p;
603 /* Expected format: ' ' x (1 or more) */
604 if (!len || line[len - 1] != ' ')
605 return 0;
607 p = line + len;
608 while (p != line) {
609 p--;
610 if (*p != ' ')
611 return line + len - (p + 1);
614 /* All spaces! */
615 return len;
618 static size_t diff_timestamp_len(const char *line, size_t len)
620 const char *end = line + len;
621 size_t n;
624 * Posix: 2010-07-05 19:41:17
625 * GNU: 2010-07-05 19:41:17.620000023 -0500
628 if (!isdigit(end[-1]))
629 return 0;
631 n = sane_tz_len(line, end - line);
632 if (!n)
633 n = tz_with_colon_len(line, end - line);
634 end -= n;
636 n = short_time_len(line, end - line);
637 if (!n)
638 n = fractional_time_len(line, end - line);
639 end -= n;
641 n = date_len(line, end - line);
642 if (!n) /* No date. Too bad. */
643 return 0;
644 end -= n;
646 if (end == line) /* No space before date. */
647 return 0;
648 if (end[-1] == '\t') { /* Success! */
649 end--;
650 return line + len - end;
652 if (end[-1] != ' ') /* No space before date. */
653 return 0;
655 /* Whitespace damage. */
656 end -= trailing_spaces_len(line, end - line);
657 return line + len - end;
660 static char *find_name_common(const char *line, const char *def,
661 int p_value, const char *end, int terminate)
663 int len;
664 const char *start = NULL;
666 if (p_value == 0)
667 start = line;
668 while (line != end) {
669 char c = *line;
671 if (!end && isspace(c)) {
672 if (c == '\n')
673 break;
674 if (name_terminate(start, line-start, c, terminate))
675 break;
677 line++;
678 if (c == '/' && !--p_value)
679 start = line;
681 if (!start)
682 return squash_slash(xstrdup_or_null(def));
683 len = line - start;
684 if (!len)
685 return squash_slash(xstrdup_or_null(def));
688 * Generally we prefer the shorter name, especially
689 * if the other one is just a variation of that with
690 * something else tacked on to the end (ie "file.orig"
691 * or "file~").
693 if (def) {
694 int deflen = strlen(def);
695 if (deflen < len && !strncmp(start, def, deflen))
696 return squash_slash(xstrdup(def));
699 if (root.len) {
700 char *ret = xstrfmt("%s%.*s", root.buf, len, start);
701 return squash_slash(ret);
704 return squash_slash(xmemdupz(start, len));
707 static char *find_name(const char *line, char *def, int p_value, int terminate)
709 if (*line == '"') {
710 char *name = find_name_gnu(line, def, p_value);
711 if (name)
712 return name;
715 return find_name_common(line, def, p_value, NULL, terminate);
718 static char *find_name_traditional(const char *line, char *def, int p_value)
720 size_t len;
721 size_t date_len;
723 if (*line == '"') {
724 char *name = find_name_gnu(line, def, p_value);
725 if (name)
726 return name;
729 len = strchrnul(line, '\n') - line;
730 date_len = diff_timestamp_len(line, len);
731 if (!date_len)
732 return find_name_common(line, def, p_value, NULL, TERM_TAB);
733 len -= date_len;
735 return find_name_common(line, def, p_value, line + len, 0);
738 static int count_slashes(const char *cp)
740 int cnt = 0;
741 char ch;
743 while ((ch = *cp++))
744 if (ch == '/')
745 cnt++;
746 return cnt;
750 * Given the string after "--- " or "+++ ", guess the appropriate
751 * p_value for the given patch.
753 static int guess_p_value(const char *nameline)
755 char *name, *cp;
756 int val = -1;
758 if (is_dev_null(nameline))
759 return -1;
760 name = find_name_traditional(nameline, NULL, 0);
761 if (!name)
762 return -1;
763 cp = strchr(name, '/');
764 if (!cp)
765 val = 0;
766 else if (prefix) {
768 * Does it begin with "a/$our-prefix" and such? Then this is
769 * very likely to apply to our directory.
771 if (!strncmp(name, prefix, prefix_length))
772 val = count_slashes(prefix);
773 else {
774 cp++;
775 if (!strncmp(cp, prefix, prefix_length))
776 val = count_slashes(prefix) + 1;
779 free(name);
780 return val;
784 * Does the ---/+++ line have the POSIX timestamp after the last HT?
785 * GNU diff puts epoch there to signal a creation/deletion event. Is
786 * this such a timestamp?
788 static int has_epoch_timestamp(const char *nameline)
791 * We are only interested in epoch timestamp; any non-zero
792 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
793 * For the same reason, the date must be either 1969-12-31 or
794 * 1970-01-01, and the seconds part must be "00".
796 const char stamp_regexp[] =
797 "^(1969-12-31|1970-01-01)"
799 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
801 "([-+][0-2][0-9]:?[0-5][0-9])\n";
802 const char *timestamp = NULL, *cp, *colon;
803 static regex_t *stamp;
804 regmatch_t m[10];
805 int zoneoffset;
806 int hourminute;
807 int status;
809 for (cp = nameline; *cp != '\n'; cp++) {
810 if (*cp == '\t')
811 timestamp = cp + 1;
813 if (!timestamp)
814 return 0;
815 if (!stamp) {
816 stamp = xmalloc(sizeof(*stamp));
817 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
818 warning(_("Cannot prepare timestamp regexp %s"),
819 stamp_regexp);
820 return 0;
824 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
825 if (status) {
826 if (status != REG_NOMATCH)
827 warning(_("regexec returned %d for input: %s"),
828 status, timestamp);
829 return 0;
832 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
833 if (*colon == ':')
834 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
835 else
836 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
837 if (timestamp[m[3].rm_so] == '-')
838 zoneoffset = -zoneoffset;
841 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
842 * (west of GMT) or 1970-01-01 (east of GMT)
844 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
845 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
846 return 0;
848 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
849 strtol(timestamp + 14, NULL, 10) -
850 zoneoffset);
852 return ((zoneoffset < 0 && hourminute == 1440) ||
853 (0 <= zoneoffset && !hourminute));
857 * Get the name etc info from the ---/+++ lines of a traditional patch header
859 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
860 * files, we can happily check the index for a match, but for creating a
861 * new file we should try to match whatever "patch" does. I have no idea.
863 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
865 char *name;
867 first += 4; /* skip "--- " */
868 second += 4; /* skip "+++ " */
869 if (!p_value_known) {
870 int p, q;
871 p = guess_p_value(first);
872 q = guess_p_value(second);
873 if (p < 0) p = q;
874 if (0 <= p && p == q) {
875 p_value = p;
876 p_value_known = 1;
879 if (is_dev_null(first)) {
880 patch->is_new = 1;
881 patch->is_delete = 0;
882 name = find_name_traditional(second, NULL, p_value);
883 patch->new_name = name;
884 } else if (is_dev_null(second)) {
885 patch->is_new = 0;
886 patch->is_delete = 1;
887 name = find_name_traditional(first, NULL, p_value);
888 patch->old_name = name;
889 } else {
890 char *first_name;
891 first_name = find_name_traditional(first, NULL, p_value);
892 name = find_name_traditional(second, first_name, p_value);
893 free(first_name);
894 if (has_epoch_timestamp(first)) {
895 patch->is_new = 1;
896 patch->is_delete = 0;
897 patch->new_name = name;
898 } else if (has_epoch_timestamp(second)) {
899 patch->is_new = 0;
900 patch->is_delete = 1;
901 patch->old_name = name;
902 } else {
903 patch->old_name = name;
904 patch->new_name = xstrdup_or_null(name);
907 if (!name)
908 die(_("unable to find filename in patch at line %d"), linenr);
911 static int gitdiff_hdrend(const char *line, struct patch *patch)
913 return -1;
917 * We're anal about diff header consistency, to make
918 * sure that we don't end up having strange ambiguous
919 * patches floating around.
921 * As a result, gitdiff_{old|new}name() will check
922 * their names against any previous information, just
923 * to make sure..
925 #define DIFF_OLD_NAME 0
926 #define DIFF_NEW_NAME 1
928 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
930 if (!orig_name && !isnull)
931 return find_name(line, NULL, p_value, TERM_TAB);
933 if (orig_name) {
934 int len = strlen(orig_name);
935 char *another;
936 if (isnull)
937 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
938 orig_name, linenr);
939 another = find_name(line, NULL, p_value, TERM_TAB);
940 if (!another || memcmp(another, orig_name, len + 1))
941 die((side == DIFF_NEW_NAME) ?
942 _("git apply: bad git-diff - inconsistent new filename on line %d") :
943 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
944 free(another);
945 return orig_name;
946 } else {
947 /* expect "/dev/null" */
948 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
949 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
950 return NULL;
954 static int gitdiff_oldname(const char *line, struct patch *patch)
956 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
957 DIFF_OLD_NAME);
958 return 0;
961 static int gitdiff_newname(const char *line, struct patch *patch)
963 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
964 DIFF_NEW_NAME);
965 return 0;
968 static int gitdiff_oldmode(const char *line, struct patch *patch)
970 patch->old_mode = strtoul(line, NULL, 8);
971 return 0;
974 static int gitdiff_newmode(const char *line, struct patch *patch)
976 patch->new_mode = strtoul(line, NULL, 8);
977 return 0;
980 static int gitdiff_delete(const char *line, struct patch *patch)
982 patch->is_delete = 1;
983 free(patch->old_name);
984 patch->old_name = xstrdup_or_null(patch->def_name);
985 return gitdiff_oldmode(line, patch);
988 static int gitdiff_newfile(const char *line, struct patch *patch)
990 patch->is_new = 1;
991 free(patch->new_name);
992 patch->new_name = xstrdup_or_null(patch->def_name);
993 return gitdiff_newmode(line, patch);
996 static int gitdiff_copysrc(const char *line, struct patch *patch)
998 patch->is_copy = 1;
999 free(patch->old_name);
1000 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1001 return 0;
1004 static int gitdiff_copydst(const char *line, struct patch *patch)
1006 patch->is_copy = 1;
1007 free(patch->new_name);
1008 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1009 return 0;
1012 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1014 patch->is_rename = 1;
1015 free(patch->old_name);
1016 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1017 return 0;
1020 static int gitdiff_renamedst(const char *line, struct patch *patch)
1022 patch->is_rename = 1;
1023 free(patch->new_name);
1024 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1025 return 0;
1028 static int gitdiff_similarity(const char *line, struct patch *patch)
1030 unsigned long val = strtoul(line, NULL, 10);
1031 if (val <= 100)
1032 patch->score = val;
1033 return 0;
1036 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1038 unsigned long val = strtoul(line, NULL, 10);
1039 if (val <= 100)
1040 patch->score = val;
1041 return 0;
1044 static int gitdiff_index(const char *line, struct patch *patch)
1047 * index line is N hexadecimal, "..", N hexadecimal,
1048 * and optional space with octal mode.
1050 const char *ptr, *eol;
1051 int len;
1053 ptr = strchr(line, '.');
1054 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1055 return 0;
1056 len = ptr - line;
1057 memcpy(patch->old_sha1_prefix, line, len);
1058 patch->old_sha1_prefix[len] = 0;
1060 line = ptr + 2;
1061 ptr = strchr(line, ' ');
1062 eol = strchrnul(line, '\n');
1064 if (!ptr || eol < ptr)
1065 ptr = eol;
1066 len = ptr - line;
1068 if (40 < len)
1069 return 0;
1070 memcpy(patch->new_sha1_prefix, line, len);
1071 patch->new_sha1_prefix[len] = 0;
1072 if (*ptr == ' ')
1073 patch->old_mode = strtoul(ptr+1, NULL, 8);
1074 return 0;
1078 * This is normal for a diff that doesn't change anything: we'll fall through
1079 * into the next diff. Tell the parser to break out.
1081 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1083 return -1;
1087 * Skip p_value leading components from "line"; as we do not accept
1088 * absolute paths, return NULL in that case.
1090 static const char *skip_tree_prefix(const char *line, int llen)
1092 int nslash;
1093 int i;
1095 if (!p_value)
1096 return (llen && line[0] == '/') ? NULL : line;
1098 nslash = p_value;
1099 for (i = 0; i < llen; i++) {
1100 int ch = line[i];
1101 if (ch == '/' && --nslash <= 0)
1102 return (i == 0) ? NULL : &line[i + 1];
1104 return NULL;
1108 * This is to extract the same name that appears on "diff --git"
1109 * line. We do not find and return anything if it is a rename
1110 * patch, and it is OK because we will find the name elsewhere.
1111 * We need to reliably find name only when it is mode-change only,
1112 * creation or deletion of an empty file. In any of these cases,
1113 * both sides are the same name under a/ and b/ respectively.
1115 static char *git_header_name(const char *line, int llen)
1117 const char *name;
1118 const char *second = NULL;
1119 size_t len, line_len;
1121 line += strlen("diff --git ");
1122 llen -= strlen("diff --git ");
1124 if (*line == '"') {
1125 const char *cp;
1126 struct strbuf first = STRBUF_INIT;
1127 struct strbuf sp = STRBUF_INIT;
1129 if (unquote_c_style(&first, line, &second))
1130 goto free_and_fail1;
1132 /* strip the a/b prefix including trailing slash */
1133 cp = skip_tree_prefix(first.buf, first.len);
1134 if (!cp)
1135 goto free_and_fail1;
1136 strbuf_remove(&first, 0, cp - first.buf);
1139 * second points at one past closing dq of name.
1140 * find the second name.
1142 while ((second < line + llen) && isspace(*second))
1143 second++;
1145 if (line + llen <= second)
1146 goto free_and_fail1;
1147 if (*second == '"') {
1148 if (unquote_c_style(&sp, second, NULL))
1149 goto free_and_fail1;
1150 cp = skip_tree_prefix(sp.buf, sp.len);
1151 if (!cp)
1152 goto free_and_fail1;
1153 /* They must match, otherwise ignore */
1154 if (strcmp(cp, first.buf))
1155 goto free_and_fail1;
1156 strbuf_release(&sp);
1157 return strbuf_detach(&first, NULL);
1160 /* unquoted second */
1161 cp = skip_tree_prefix(second, line + llen - second);
1162 if (!cp)
1163 goto free_and_fail1;
1164 if (line + llen - cp != first.len ||
1165 memcmp(first.buf, cp, first.len))
1166 goto free_and_fail1;
1167 return strbuf_detach(&first, NULL);
1169 free_and_fail1:
1170 strbuf_release(&first);
1171 strbuf_release(&sp);
1172 return NULL;
1175 /* unquoted first name */
1176 name = skip_tree_prefix(line, llen);
1177 if (!name)
1178 return NULL;
1181 * since the first name is unquoted, a dq if exists must be
1182 * the beginning of the second name.
1184 for (second = name; second < line + llen; second++) {
1185 if (*second == '"') {
1186 struct strbuf sp = STRBUF_INIT;
1187 const char *np;
1189 if (unquote_c_style(&sp, second, NULL))
1190 goto free_and_fail2;
1192 np = skip_tree_prefix(sp.buf, sp.len);
1193 if (!np)
1194 goto free_and_fail2;
1196 len = sp.buf + sp.len - np;
1197 if (len < second - name &&
1198 !strncmp(np, name, len) &&
1199 isspace(name[len])) {
1200 /* Good */
1201 strbuf_remove(&sp, 0, np - sp.buf);
1202 return strbuf_detach(&sp, NULL);
1205 free_and_fail2:
1206 strbuf_release(&sp);
1207 return NULL;
1212 * Accept a name only if it shows up twice, exactly the same
1213 * form.
1215 second = strchr(name, '\n');
1216 if (!second)
1217 return NULL;
1218 line_len = second - name;
1219 for (len = 0 ; ; len++) {
1220 switch (name[len]) {
1221 default:
1222 continue;
1223 case '\n':
1224 return NULL;
1225 case '\t': case ' ':
1227 * Is this the separator between the preimage
1228 * and the postimage pathname? Again, we are
1229 * only interested in the case where there is
1230 * no rename, as this is only to set def_name
1231 * and a rename patch has the names elsewhere
1232 * in an unambiguous form.
1234 if (!name[len + 1])
1235 return NULL; /* no postimage name */
1236 second = skip_tree_prefix(name + len + 1,
1237 line_len - (len + 1));
1238 if (!second)
1239 return NULL;
1241 * Does len bytes starting at "name" and "second"
1242 * (that are separated by one HT or SP we just
1243 * found) exactly match?
1245 if (second[len] == '\n' && !strncmp(name, second, len))
1246 return xmemdupz(name, len);
1251 /* Verify that we recognize the lines following a git header */
1252 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1254 unsigned long offset;
1256 /* A git diff has explicit new/delete information, so we don't guess */
1257 patch->is_new = 0;
1258 patch->is_delete = 0;
1261 * Some things may not have the old name in the
1262 * rest of the headers anywhere (pure mode changes,
1263 * or removing or adding empty files), so we get
1264 * the default name from the header.
1266 patch->def_name = git_header_name(line, len);
1267 if (patch->def_name && root.len) {
1268 char *s = xstrfmt("%s%s", root.buf, patch->def_name);
1269 free(patch->def_name);
1270 patch->def_name = s;
1273 line += len;
1274 size -= len;
1275 linenr++;
1276 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1277 static const struct opentry {
1278 const char *str;
1279 int (*fn)(const char *, struct patch *);
1280 } optable[] = {
1281 { "@@ -", gitdiff_hdrend },
1282 { "--- ", gitdiff_oldname },
1283 { "+++ ", gitdiff_newname },
1284 { "old mode ", gitdiff_oldmode },
1285 { "new mode ", gitdiff_newmode },
1286 { "deleted file mode ", gitdiff_delete },
1287 { "new file mode ", gitdiff_newfile },
1288 { "copy from ", gitdiff_copysrc },
1289 { "copy to ", gitdiff_copydst },
1290 { "rename old ", gitdiff_renamesrc },
1291 { "rename new ", gitdiff_renamedst },
1292 { "rename from ", gitdiff_renamesrc },
1293 { "rename to ", gitdiff_renamedst },
1294 { "similarity index ", gitdiff_similarity },
1295 { "dissimilarity index ", gitdiff_dissimilarity },
1296 { "index ", gitdiff_index },
1297 { "", gitdiff_unrecognized },
1299 int i;
1301 len = linelen(line, size);
1302 if (!len || line[len-1] != '\n')
1303 break;
1304 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1305 const struct opentry *p = optable + i;
1306 int oplen = strlen(p->str);
1307 if (len < oplen || memcmp(p->str, line, oplen))
1308 continue;
1309 if (p->fn(line + oplen, patch) < 0)
1310 return offset;
1311 break;
1315 return offset;
1318 static int parse_num(const char *line, unsigned long *p)
1320 char *ptr;
1322 if (!isdigit(*line))
1323 return 0;
1324 *p = strtoul(line, &ptr, 10);
1325 return ptr - line;
1328 static int parse_range(const char *line, int len, int offset, const char *expect,
1329 unsigned long *p1, unsigned long *p2)
1331 int digits, ex;
1333 if (offset < 0 || offset >= len)
1334 return -1;
1335 line += offset;
1336 len -= offset;
1338 digits = parse_num(line, p1);
1339 if (!digits)
1340 return -1;
1342 offset += digits;
1343 line += digits;
1344 len -= digits;
1346 *p2 = 1;
1347 if (*line == ',') {
1348 digits = parse_num(line+1, p2);
1349 if (!digits)
1350 return -1;
1352 offset += digits+1;
1353 line += digits+1;
1354 len -= digits+1;
1357 ex = strlen(expect);
1358 if (ex > len)
1359 return -1;
1360 if (memcmp(line, expect, ex))
1361 return -1;
1363 return offset + ex;
1366 static void recount_diff(const char *line, int size, struct fragment *fragment)
1368 int oldlines = 0, newlines = 0, ret = 0;
1370 if (size < 1) {
1371 warning("recount: ignore empty hunk");
1372 return;
1375 for (;;) {
1376 int len = linelen(line, size);
1377 size -= len;
1378 line += len;
1380 if (size < 1)
1381 break;
1383 switch (*line) {
1384 case ' ': case '\n':
1385 newlines++;
1386 /* fall through */
1387 case '-':
1388 oldlines++;
1389 continue;
1390 case '+':
1391 newlines++;
1392 continue;
1393 case '\\':
1394 continue;
1395 case '@':
1396 ret = size < 3 || !starts_with(line, "@@ ");
1397 break;
1398 case 'd':
1399 ret = size < 5 || !starts_with(line, "diff ");
1400 break;
1401 default:
1402 ret = -1;
1403 break;
1405 if (ret) {
1406 warning(_("recount: unexpected line: %.*s"),
1407 (int)linelen(line, size), line);
1408 return;
1410 break;
1412 fragment->oldlines = oldlines;
1413 fragment->newlines = newlines;
1417 * Parse a unified diff fragment header of the
1418 * form "@@ -a,b +c,d @@"
1420 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1422 int offset;
1424 if (!len || line[len-1] != '\n')
1425 return -1;
1427 /* Figure out the number of lines in a fragment */
1428 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1429 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1431 return offset;
1434 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1436 unsigned long offset, len;
1438 patch->is_toplevel_relative = 0;
1439 patch->is_rename = patch->is_copy = 0;
1440 patch->is_new = patch->is_delete = -1;
1441 patch->old_mode = patch->new_mode = 0;
1442 patch->old_name = patch->new_name = NULL;
1443 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1444 unsigned long nextlen;
1446 len = linelen(line, size);
1447 if (!len)
1448 break;
1450 /* Testing this early allows us to take a few shortcuts.. */
1451 if (len < 6)
1452 continue;
1455 * Make sure we don't find any unconnected patch fragments.
1456 * That's a sign that we didn't find a header, and that a
1457 * patch has become corrupted/broken up.
1459 if (!memcmp("@@ -", line, 4)) {
1460 struct fragment dummy;
1461 if (parse_fragment_header(line, len, &dummy) < 0)
1462 continue;
1463 die(_("patch fragment without header at line %d: %.*s"),
1464 linenr, (int)len-1, line);
1467 if (size < len + 6)
1468 break;
1471 * Git patch? It might not have a real patch, just a rename
1472 * or mode change, so we handle that specially
1474 if (!memcmp("diff --git ", line, 11)) {
1475 int git_hdr_len = parse_git_header(line, len, size, patch);
1476 if (git_hdr_len <= len)
1477 continue;
1478 if (!patch->old_name && !patch->new_name) {
1479 if (!patch->def_name)
1480 die(Q_("git diff header lacks filename information when removing "
1481 "%d leading pathname component (line %d)",
1482 "git diff header lacks filename information when removing "
1483 "%d leading pathname components (line %d)",
1484 p_value),
1485 p_value, linenr);
1486 patch->old_name = xstrdup(patch->def_name);
1487 patch->new_name = xstrdup(patch->def_name);
1489 if (!patch->is_delete && !patch->new_name)
1490 die("git diff header lacks filename information "
1491 "(line %d)", linenr);
1492 patch->is_toplevel_relative = 1;
1493 *hdrsize = git_hdr_len;
1494 return offset;
1497 /* --- followed by +++ ? */
1498 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1499 continue;
1502 * We only accept unified patches, so we want it to
1503 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1504 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1506 nextlen = linelen(line + len, size - len);
1507 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1508 continue;
1510 /* Ok, we'll consider it a patch */
1511 parse_traditional_patch(line, line+len, patch);
1512 *hdrsize = len + nextlen;
1513 linenr += 2;
1514 return offset;
1516 return -1;
1519 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1521 char *err;
1523 if (!result)
1524 return;
1526 whitespace_error++;
1527 if (squelch_whitespace_errors &&
1528 squelch_whitespace_errors < whitespace_error)
1529 return;
1531 err = whitespace_error_string(result);
1532 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1533 patch_input_file, linenr, err, len, line);
1534 free(err);
1537 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1539 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1541 record_ws_error(result, line + 1, len - 2, linenr);
1545 * Parse a unified diff. Note that this really needs to parse each
1546 * fragment separately, since the only way to know the difference
1547 * between a "---" that is part of a patch, and a "---" that starts
1548 * the next patch is to look at the line counts..
1550 static int parse_fragment(const char *line, unsigned long size,
1551 struct patch *patch, struct fragment *fragment)
1553 int added, deleted;
1554 int len = linelen(line, size), offset;
1555 unsigned long oldlines, newlines;
1556 unsigned long leading, trailing;
1558 offset = parse_fragment_header(line, len, fragment);
1559 if (offset < 0)
1560 return -1;
1561 if (offset > 0 && patch->recount)
1562 recount_diff(line + offset, size - offset, fragment);
1563 oldlines = fragment->oldlines;
1564 newlines = fragment->newlines;
1565 leading = 0;
1566 trailing = 0;
1568 /* Parse the thing.. */
1569 line += len;
1570 size -= len;
1571 linenr++;
1572 added = deleted = 0;
1573 for (offset = len;
1574 0 < size;
1575 offset += len, size -= len, line += len, linenr++) {
1576 if (!oldlines && !newlines)
1577 break;
1578 len = linelen(line, size);
1579 if (!len || line[len-1] != '\n')
1580 return -1;
1581 switch (*line) {
1582 default:
1583 return -1;
1584 case '\n': /* newer GNU diff, an empty context line */
1585 case ' ':
1586 oldlines--;
1587 newlines--;
1588 if (!deleted && !added)
1589 leading++;
1590 trailing++;
1591 if (!apply_in_reverse &&
1592 ws_error_action == correct_ws_error)
1593 check_whitespace(line, len, patch->ws_rule);
1594 break;
1595 case '-':
1596 if (apply_in_reverse &&
1597 ws_error_action != nowarn_ws_error)
1598 check_whitespace(line, len, patch->ws_rule);
1599 deleted++;
1600 oldlines--;
1601 trailing = 0;
1602 break;
1603 case '+':
1604 if (!apply_in_reverse &&
1605 ws_error_action != nowarn_ws_error)
1606 check_whitespace(line, len, patch->ws_rule);
1607 added++;
1608 newlines--;
1609 trailing = 0;
1610 break;
1613 * We allow "\ No newline at end of file". Depending
1614 * on locale settings when the patch was produced we
1615 * don't know what this line looks like. The only
1616 * thing we do know is that it begins with "\ ".
1617 * Checking for 12 is just for sanity check -- any
1618 * l10n of "\ No newline..." is at least that long.
1620 case '\\':
1621 if (len < 12 || memcmp(line, "\\ ", 2))
1622 return -1;
1623 break;
1626 if (oldlines || newlines)
1627 return -1;
1628 if (!deleted && !added)
1629 return -1;
1631 fragment->leading = leading;
1632 fragment->trailing = trailing;
1635 * If a fragment ends with an incomplete line, we failed to include
1636 * it in the above loop because we hit oldlines == newlines == 0
1637 * before seeing it.
1639 if (12 < size && !memcmp(line, "\\ ", 2))
1640 offset += linelen(line, size);
1642 patch->lines_added += added;
1643 patch->lines_deleted += deleted;
1645 if (0 < patch->is_new && oldlines)
1646 return error(_("new file depends on old contents"));
1647 if (0 < patch->is_delete && newlines)
1648 return error(_("deleted file still has contents"));
1649 return offset;
1653 * We have seen "diff --git a/... b/..." header (or a traditional patch
1654 * header). Read hunks that belong to this patch into fragments and hang
1655 * them to the given patch structure.
1657 * The (fragment->patch, fragment->size) pair points into the memory given
1658 * by the caller, not a copy, when we return.
1660 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1662 unsigned long offset = 0;
1663 unsigned long oldlines = 0, newlines = 0, context = 0;
1664 struct fragment **fragp = &patch->fragments;
1666 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1667 struct fragment *fragment;
1668 int len;
1670 fragment = xcalloc(1, sizeof(*fragment));
1671 fragment->linenr = linenr;
1672 len = parse_fragment(line, size, patch, fragment);
1673 if (len <= 0)
1674 die(_("corrupt patch at line %d"), linenr);
1675 fragment->patch = line;
1676 fragment->size = len;
1677 oldlines += fragment->oldlines;
1678 newlines += fragment->newlines;
1679 context += fragment->leading + fragment->trailing;
1681 *fragp = fragment;
1682 fragp = &fragment->next;
1684 offset += len;
1685 line += len;
1686 size -= len;
1690 * If something was removed (i.e. we have old-lines) it cannot
1691 * be creation, and if something was added it cannot be
1692 * deletion. However, the reverse is not true; --unified=0
1693 * patches that only add are not necessarily creation even
1694 * though they do not have any old lines, and ones that only
1695 * delete are not necessarily deletion.
1697 * Unfortunately, a real creation/deletion patch do _not_ have
1698 * any context line by definition, so we cannot safely tell it
1699 * apart with --unified=0 insanity. At least if the patch has
1700 * more than one hunk it is not creation or deletion.
1702 if (patch->is_new < 0 &&
1703 (oldlines || (patch->fragments && patch->fragments->next)))
1704 patch->is_new = 0;
1705 if (patch->is_delete < 0 &&
1706 (newlines || (patch->fragments && patch->fragments->next)))
1707 patch->is_delete = 0;
1709 if (0 < patch->is_new && oldlines)
1710 die(_("new file %s depends on old contents"), patch->new_name);
1711 if (0 < patch->is_delete && newlines)
1712 die(_("deleted file %s still has contents"), patch->old_name);
1713 if (!patch->is_delete && !newlines && context)
1714 fprintf_ln(stderr,
1715 _("** warning: "
1716 "file %s becomes empty but is not deleted"),
1717 patch->new_name);
1719 return offset;
1722 static inline int metadata_changes(struct patch *patch)
1724 return patch->is_rename > 0 ||
1725 patch->is_copy > 0 ||
1726 patch->is_new > 0 ||
1727 patch->is_delete ||
1728 (patch->old_mode && patch->new_mode &&
1729 patch->old_mode != patch->new_mode);
1732 static char *inflate_it(const void *data, unsigned long size,
1733 unsigned long inflated_size)
1735 git_zstream stream;
1736 void *out;
1737 int st;
1739 memset(&stream, 0, sizeof(stream));
1741 stream.next_in = (unsigned char *)data;
1742 stream.avail_in = size;
1743 stream.next_out = out = xmalloc(inflated_size);
1744 stream.avail_out = inflated_size;
1745 git_inflate_init(&stream);
1746 st = git_inflate(&stream, Z_FINISH);
1747 git_inflate_end(&stream);
1748 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1749 free(out);
1750 return NULL;
1752 return out;
1756 * Read a binary hunk and return a new fragment; fragment->patch
1757 * points at an allocated memory that the caller must free, so
1758 * it is marked as "->free_patch = 1".
1760 static struct fragment *parse_binary_hunk(char **buf_p,
1761 unsigned long *sz_p,
1762 int *status_p,
1763 int *used_p)
1766 * Expect a line that begins with binary patch method ("literal"
1767 * or "delta"), followed by the length of data before deflating.
1768 * a sequence of 'length-byte' followed by base-85 encoded data
1769 * should follow, terminated by a newline.
1771 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1772 * and we would limit the patch line to 66 characters,
1773 * so one line can fit up to 13 groups that would decode
1774 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1775 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1777 int llen, used;
1778 unsigned long size = *sz_p;
1779 char *buffer = *buf_p;
1780 int patch_method;
1781 unsigned long origlen;
1782 char *data = NULL;
1783 int hunk_size = 0;
1784 struct fragment *frag;
1786 llen = linelen(buffer, size);
1787 used = llen;
1789 *status_p = 0;
1791 if (starts_with(buffer, "delta ")) {
1792 patch_method = BINARY_DELTA_DEFLATED;
1793 origlen = strtoul(buffer + 6, NULL, 10);
1795 else if (starts_with(buffer, "literal ")) {
1796 patch_method = BINARY_LITERAL_DEFLATED;
1797 origlen = strtoul(buffer + 8, NULL, 10);
1799 else
1800 return NULL;
1802 linenr++;
1803 buffer += llen;
1804 while (1) {
1805 int byte_length, max_byte_length, newsize;
1806 llen = linelen(buffer, size);
1807 used += llen;
1808 linenr++;
1809 if (llen == 1) {
1810 /* consume the blank line */
1811 buffer++;
1812 size--;
1813 break;
1816 * Minimum line is "A00000\n" which is 7-byte long,
1817 * and the line length must be multiple of 5 plus 2.
1819 if ((llen < 7) || (llen-2) % 5)
1820 goto corrupt;
1821 max_byte_length = (llen - 2) / 5 * 4;
1822 byte_length = *buffer;
1823 if ('A' <= byte_length && byte_length <= 'Z')
1824 byte_length = byte_length - 'A' + 1;
1825 else if ('a' <= byte_length && byte_length <= 'z')
1826 byte_length = byte_length - 'a' + 27;
1827 else
1828 goto corrupt;
1829 /* if the input length was not multiple of 4, we would
1830 * have filler at the end but the filler should never
1831 * exceed 3 bytes
1833 if (max_byte_length < byte_length ||
1834 byte_length <= max_byte_length - 4)
1835 goto corrupt;
1836 newsize = hunk_size + byte_length;
1837 data = xrealloc(data, newsize);
1838 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1839 goto corrupt;
1840 hunk_size = newsize;
1841 buffer += llen;
1842 size -= llen;
1845 frag = xcalloc(1, sizeof(*frag));
1846 frag->patch = inflate_it(data, hunk_size, origlen);
1847 frag->free_patch = 1;
1848 if (!frag->patch)
1849 goto corrupt;
1850 free(data);
1851 frag->size = origlen;
1852 *buf_p = buffer;
1853 *sz_p = size;
1854 *used_p = used;
1855 frag->binary_patch_method = patch_method;
1856 return frag;
1858 corrupt:
1859 free(data);
1860 *status_p = -1;
1861 error(_("corrupt binary patch at line %d: %.*s"),
1862 linenr-1, llen-1, buffer);
1863 return NULL;
1867 * Returns:
1868 * -1 in case of error,
1869 * the length of the parsed binary patch otherwise
1871 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1874 * We have read "GIT binary patch\n"; what follows is a line
1875 * that says the patch method (currently, either "literal" or
1876 * "delta") and the length of data before deflating; a
1877 * sequence of 'length-byte' followed by base-85 encoded data
1878 * follows.
1880 * When a binary patch is reversible, there is another binary
1881 * hunk in the same format, starting with patch method (either
1882 * "literal" or "delta") with the length of data, and a sequence
1883 * of length-byte + base-85 encoded data, terminated with another
1884 * empty line. This data, when applied to the postimage, produces
1885 * the preimage.
1887 struct fragment *forward;
1888 struct fragment *reverse;
1889 int status;
1890 int used, used_1;
1892 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1893 if (!forward && !status)
1894 /* there has to be one hunk (forward hunk) */
1895 return error(_("unrecognized binary patch at line %d"), linenr-1);
1896 if (status)
1897 /* otherwise we already gave an error message */
1898 return status;
1900 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1901 if (reverse)
1902 used += used_1;
1903 else if (status) {
1905 * Not having reverse hunk is not an error, but having
1906 * a corrupt reverse hunk is.
1908 free((void*) forward->patch);
1909 free(forward);
1910 return status;
1912 forward->next = reverse;
1913 patch->fragments = forward;
1914 patch->is_binary = 1;
1915 return used;
1918 static void prefix_one(char **name)
1920 char *old_name = *name;
1921 if (!old_name)
1922 return;
1923 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
1924 free(old_name);
1927 static void prefix_patch(struct patch *p)
1929 if (!prefix || p->is_toplevel_relative)
1930 return;
1931 prefix_one(&p->new_name);
1932 prefix_one(&p->old_name);
1936 * include/exclude
1939 static struct string_list limit_by_name;
1940 static int has_include;
1941 static void add_name_limit(const char *name, int exclude)
1943 struct string_list_item *it;
1945 it = string_list_append(&limit_by_name, name);
1946 it->util = exclude ? NULL : (void *) 1;
1949 static int use_patch(struct patch *p)
1951 const char *pathname = p->new_name ? p->new_name : p->old_name;
1952 int i;
1954 /* Paths outside are not touched regardless of "--include" */
1955 if (0 < prefix_length) {
1956 int pathlen = strlen(pathname);
1957 if (pathlen <= prefix_length ||
1958 memcmp(prefix, pathname, prefix_length))
1959 return 0;
1962 /* See if it matches any of exclude/include rule */
1963 for (i = 0; i < limit_by_name.nr; i++) {
1964 struct string_list_item *it = &limit_by_name.items[i];
1965 if (!wildmatch(it->string, pathname, 0, NULL))
1966 return (it->util != NULL);
1970 * If we had any include, a path that does not match any rule is
1971 * not used. Otherwise, we saw bunch of exclude rules (or none)
1972 * and such a path is used.
1974 return !has_include;
1979 * Read the patch text in "buffer" that extends for "size" bytes; stop
1980 * reading after seeing a single patch (i.e. changes to a single file).
1981 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1982 * Return the number of bytes consumed, so that the caller can call us
1983 * again for the next patch.
1985 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1987 int hdrsize, patchsize;
1988 int offset = find_header(buffer, size, &hdrsize, patch);
1990 if (offset < 0)
1991 return offset;
1993 prefix_patch(patch);
1995 if (!use_patch(patch))
1996 patch->ws_rule = 0;
1997 else
1998 patch->ws_rule = whitespace_rule(patch->new_name
1999 ? patch->new_name
2000 : patch->old_name);
2002 patchsize = parse_single_patch(buffer + offset + hdrsize,
2003 size - offset - hdrsize, patch);
2005 if (!patchsize) {
2006 static const char git_binary[] = "GIT binary patch\n";
2007 int hd = hdrsize + offset;
2008 unsigned long llen = linelen(buffer + hd, size - hd);
2010 if (llen == sizeof(git_binary) - 1 &&
2011 !memcmp(git_binary, buffer + hd, llen)) {
2012 int used;
2013 linenr++;
2014 used = parse_binary(buffer + hd + llen,
2015 size - hd - llen, patch);
2016 if (used < 0)
2017 return -1;
2018 if (used)
2019 patchsize = used + llen;
2020 else
2021 patchsize = 0;
2023 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2024 static const char *binhdr[] = {
2025 "Binary files ",
2026 "Files ",
2027 NULL,
2029 int i;
2030 for (i = 0; binhdr[i]; i++) {
2031 int len = strlen(binhdr[i]);
2032 if (len < size - hd &&
2033 !memcmp(binhdr[i], buffer + hd, len)) {
2034 linenr++;
2035 patch->is_binary = 1;
2036 patchsize = llen;
2037 break;
2042 /* Empty patch cannot be applied if it is a text patch
2043 * without metadata change. A binary patch appears
2044 * empty to us here.
2046 if ((apply || check) &&
2047 (!patch->is_binary && !metadata_changes(patch)))
2048 die(_("patch with only garbage at line %d"), linenr);
2051 return offset + hdrsize + patchsize;
2054 #define swap(a,b) myswap((a),(b),sizeof(a))
2056 #define myswap(a, b, size) do { \
2057 unsigned char mytmp[size]; \
2058 memcpy(mytmp, &a, size); \
2059 memcpy(&a, &b, size); \
2060 memcpy(&b, mytmp, size); \
2061 } while (0)
2063 static void reverse_patches(struct patch *p)
2065 for (; p; p = p->next) {
2066 struct fragment *frag = p->fragments;
2068 swap(p->new_name, p->old_name);
2069 swap(p->new_mode, p->old_mode);
2070 swap(p->is_new, p->is_delete);
2071 swap(p->lines_added, p->lines_deleted);
2072 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2074 for (; frag; frag = frag->next) {
2075 swap(frag->newpos, frag->oldpos);
2076 swap(frag->newlines, frag->oldlines);
2081 static const char pluses[] =
2082 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2083 static const char minuses[]=
2084 "----------------------------------------------------------------------";
2086 static void show_stats(struct patch *patch)
2088 struct strbuf qname = STRBUF_INIT;
2089 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2090 int max, add, del;
2092 quote_c_style(cp, &qname, NULL, 0);
2095 * "scale" the filename
2097 max = max_len;
2098 if (max > 50)
2099 max = 50;
2101 if (qname.len > max) {
2102 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2103 if (!cp)
2104 cp = qname.buf + qname.len + 3 - max;
2105 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2108 if (patch->is_binary) {
2109 printf(" %-*s | Bin\n", max, qname.buf);
2110 strbuf_release(&qname);
2111 return;
2114 printf(" %-*s |", max, qname.buf);
2115 strbuf_release(&qname);
2118 * scale the add/delete
2120 max = max + max_change > 70 ? 70 - max : max_change;
2121 add = patch->lines_added;
2122 del = patch->lines_deleted;
2124 if (max_change > 0) {
2125 int total = ((add + del) * max + max_change / 2) / max_change;
2126 add = (add * max + max_change / 2) / max_change;
2127 del = total - add;
2129 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2130 add, pluses, del, minuses);
2133 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2135 switch (st->st_mode & S_IFMT) {
2136 case S_IFLNK:
2137 if (strbuf_readlink(buf, path, st->st_size) < 0)
2138 return error(_("unable to read symlink %s"), path);
2139 return 0;
2140 case S_IFREG:
2141 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2142 return error(_("unable to open or read %s"), path);
2143 convert_to_git(path, buf->buf, buf->len, buf, 0);
2144 return 0;
2145 default:
2146 return -1;
2151 * Update the preimage, and the common lines in postimage,
2152 * from buffer buf of length len. If postlen is 0 the postimage
2153 * is updated in place, otherwise it's updated on a new buffer
2154 * of length postlen
2157 static void update_pre_post_images(struct image *preimage,
2158 struct image *postimage,
2159 char *buf,
2160 size_t len, size_t postlen)
2162 int i, ctx, reduced;
2163 char *new, *old, *fixed;
2164 struct image fixed_preimage;
2167 * Update the preimage with whitespace fixes. Note that we
2168 * are not losing preimage->buf -- apply_one_fragment() will
2169 * free "oldlines".
2171 prepare_image(&fixed_preimage, buf, len, 1);
2172 assert(postlen
2173 ? fixed_preimage.nr == preimage->nr
2174 : fixed_preimage.nr <= preimage->nr);
2175 for (i = 0; i < fixed_preimage.nr; i++)
2176 fixed_preimage.line[i].flag = preimage->line[i].flag;
2177 free(preimage->line_allocated);
2178 *preimage = fixed_preimage;
2181 * Adjust the common context lines in postimage. This can be
2182 * done in-place when we are shrinking it with whitespace
2183 * fixing, but needs a new buffer when ignoring whitespace or
2184 * expanding leading tabs to spaces.
2186 * We trust the caller to tell us if the update can be done
2187 * in place (postlen==0) or not.
2189 old = postimage->buf;
2190 if (postlen)
2191 new = postimage->buf = xmalloc(postlen);
2192 else
2193 new = old;
2194 fixed = preimage->buf;
2196 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2197 size_t len = postimage->line[i].len;
2198 if (!(postimage->line[i].flag & LINE_COMMON)) {
2199 /* an added line -- no counterparts in preimage */
2200 memmove(new, old, len);
2201 old += len;
2202 new += len;
2203 continue;
2206 /* a common context -- skip it in the original postimage */
2207 old += len;
2209 /* and find the corresponding one in the fixed preimage */
2210 while (ctx < preimage->nr &&
2211 !(preimage->line[ctx].flag & LINE_COMMON)) {
2212 fixed += preimage->line[ctx].len;
2213 ctx++;
2217 * preimage is expected to run out, if the caller
2218 * fixed addition of trailing blank lines.
2220 if (preimage->nr <= ctx) {
2221 reduced++;
2222 continue;
2225 /* and copy it in, while fixing the line length */
2226 len = preimage->line[ctx].len;
2227 memcpy(new, fixed, len);
2228 new += len;
2229 fixed += len;
2230 postimage->line[i].len = len;
2231 ctx++;
2234 if (postlen
2235 ? postlen < new - postimage->buf
2236 : postimage->len < new - postimage->buf)
2237 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2238 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2240 /* Fix the length of the whole thing */
2241 postimage->len = new - postimage->buf;
2242 postimage->nr -= reduced;
2245 static int match_fragment(struct image *img,
2246 struct image *preimage,
2247 struct image *postimage,
2248 unsigned long try,
2249 int try_lno,
2250 unsigned ws_rule,
2251 int match_beginning, int match_end)
2253 int i;
2254 char *fixed_buf, *buf, *orig, *target;
2255 struct strbuf fixed;
2256 size_t fixed_len, postlen;
2257 int preimage_limit;
2259 if (preimage->nr + try_lno <= img->nr) {
2261 * The hunk falls within the boundaries of img.
2263 preimage_limit = preimage->nr;
2264 if (match_end && (preimage->nr + try_lno != img->nr))
2265 return 0;
2266 } else if (ws_error_action == correct_ws_error &&
2267 (ws_rule & WS_BLANK_AT_EOF)) {
2269 * This hunk extends beyond the end of img, and we are
2270 * removing blank lines at the end of the file. This
2271 * many lines from the beginning of the preimage must
2272 * match with img, and the remainder of the preimage
2273 * must be blank.
2275 preimage_limit = img->nr - try_lno;
2276 } else {
2278 * The hunk extends beyond the end of the img and
2279 * we are not removing blanks at the end, so we
2280 * should reject the hunk at this position.
2282 return 0;
2285 if (match_beginning && try_lno)
2286 return 0;
2288 /* Quick hash check */
2289 for (i = 0; i < preimage_limit; i++)
2290 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2291 (preimage->line[i].hash != img->line[try_lno + i].hash))
2292 return 0;
2294 if (preimage_limit == preimage->nr) {
2296 * Do we have an exact match? If we were told to match
2297 * at the end, size must be exactly at try+fragsize,
2298 * otherwise try+fragsize must be still within the preimage,
2299 * and either case, the old piece should match the preimage
2300 * exactly.
2302 if ((match_end
2303 ? (try + preimage->len == img->len)
2304 : (try + preimage->len <= img->len)) &&
2305 !memcmp(img->buf + try, preimage->buf, preimage->len))
2306 return 1;
2307 } else {
2309 * The preimage extends beyond the end of img, so
2310 * there cannot be an exact match.
2312 * There must be one non-blank context line that match
2313 * a line before the end of img.
2315 char *buf_end;
2317 buf = preimage->buf;
2318 buf_end = buf;
2319 for (i = 0; i < preimage_limit; i++)
2320 buf_end += preimage->line[i].len;
2322 for ( ; buf < buf_end; buf++)
2323 if (!isspace(*buf))
2324 break;
2325 if (buf == buf_end)
2326 return 0;
2330 * No exact match. If we are ignoring whitespace, run a line-by-line
2331 * fuzzy matching. We collect all the line length information because
2332 * we need it to adjust whitespace if we match.
2334 if (ws_ignore_action == ignore_ws_change) {
2335 size_t imgoff = 0;
2336 size_t preoff = 0;
2337 size_t postlen = postimage->len;
2338 size_t extra_chars;
2339 char *preimage_eof;
2340 char *preimage_end;
2341 for (i = 0; i < preimage_limit; i++) {
2342 size_t prelen = preimage->line[i].len;
2343 size_t imglen = img->line[try_lno+i].len;
2345 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2346 preimage->buf + preoff, prelen))
2347 return 0;
2348 if (preimage->line[i].flag & LINE_COMMON)
2349 postlen += imglen - prelen;
2350 imgoff += imglen;
2351 preoff += prelen;
2355 * Ok, the preimage matches with whitespace fuzz.
2357 * imgoff now holds the true length of the target that
2358 * matches the preimage before the end of the file.
2360 * Count the number of characters in the preimage that fall
2361 * beyond the end of the file and make sure that all of them
2362 * are whitespace characters. (This can only happen if
2363 * we are removing blank lines at the end of the file.)
2365 buf = preimage_eof = preimage->buf + preoff;
2366 for ( ; i < preimage->nr; i++)
2367 preoff += preimage->line[i].len;
2368 preimage_end = preimage->buf + preoff;
2369 for ( ; buf < preimage_end; buf++)
2370 if (!isspace(*buf))
2371 return 0;
2374 * Update the preimage and the common postimage context
2375 * lines to use the same whitespace as the target.
2376 * If whitespace is missing in the target (i.e.
2377 * if the preimage extends beyond the end of the file),
2378 * use the whitespace from the preimage.
2380 extra_chars = preimage_end - preimage_eof;
2381 strbuf_init(&fixed, imgoff + extra_chars);
2382 strbuf_add(&fixed, img->buf + try, imgoff);
2383 strbuf_add(&fixed, preimage_eof, extra_chars);
2384 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2385 update_pre_post_images(preimage, postimage,
2386 fixed_buf, fixed_len, postlen);
2387 return 1;
2390 if (ws_error_action != correct_ws_error)
2391 return 0;
2394 * The hunk does not apply byte-by-byte, but the hash says
2395 * it might with whitespace fuzz. We weren't asked to
2396 * ignore whitespace, we were asked to correct whitespace
2397 * errors, so let's try matching after whitespace correction.
2399 * While checking the preimage against the target, whitespace
2400 * errors in both fixed, we count how large the corresponding
2401 * postimage needs to be. The postimage prepared by
2402 * apply_one_fragment() has whitespace errors fixed on added
2403 * lines already, but the common lines were propagated as-is,
2404 * which may become longer when their whitespace errors are
2405 * fixed.
2408 /* First count added lines in postimage */
2409 postlen = 0;
2410 for (i = 0; i < postimage->nr; i++) {
2411 if (!(postimage->line[i].flag & LINE_COMMON))
2412 postlen += postimage->line[i].len;
2416 * The preimage may extend beyond the end of the file,
2417 * but in this loop we will only handle the part of the
2418 * preimage that falls within the file.
2420 strbuf_init(&fixed, preimage->len + 1);
2421 orig = preimage->buf;
2422 target = img->buf + try;
2423 for (i = 0; i < preimage_limit; i++) {
2424 size_t oldlen = preimage->line[i].len;
2425 size_t tgtlen = img->line[try_lno + i].len;
2426 size_t fixstart = fixed.len;
2427 struct strbuf tgtfix;
2428 int match;
2430 /* Try fixing the line in the preimage */
2431 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2433 /* Try fixing the line in the target */
2434 strbuf_init(&tgtfix, tgtlen);
2435 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2438 * If they match, either the preimage was based on
2439 * a version before our tree fixed whitespace breakage,
2440 * or we are lacking a whitespace-fix patch the tree
2441 * the preimage was based on already had (i.e. target
2442 * has whitespace breakage, the preimage doesn't).
2443 * In either case, we are fixing the whitespace breakages
2444 * so we might as well take the fix together with their
2445 * real change.
2447 match = (tgtfix.len == fixed.len - fixstart &&
2448 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2449 fixed.len - fixstart));
2451 /* Add the length if this is common with the postimage */
2452 if (preimage->line[i].flag & LINE_COMMON)
2453 postlen += tgtfix.len;
2455 strbuf_release(&tgtfix);
2456 if (!match)
2457 goto unmatch_exit;
2459 orig += oldlen;
2460 target += tgtlen;
2465 * Now handle the lines in the preimage that falls beyond the
2466 * end of the file (if any). They will only match if they are
2467 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2468 * false).
2470 for ( ; i < preimage->nr; i++) {
2471 size_t fixstart = fixed.len; /* start of the fixed preimage */
2472 size_t oldlen = preimage->line[i].len;
2473 int j;
2475 /* Try fixing the line in the preimage */
2476 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2478 for (j = fixstart; j < fixed.len; j++)
2479 if (!isspace(fixed.buf[j]))
2480 goto unmatch_exit;
2482 orig += oldlen;
2486 * Yes, the preimage is based on an older version that still
2487 * has whitespace breakages unfixed, and fixing them makes the
2488 * hunk match. Update the context lines in the postimage.
2490 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2491 if (postlen < postimage->len)
2492 postlen = 0;
2493 update_pre_post_images(preimage, postimage,
2494 fixed_buf, fixed_len, postlen);
2495 return 1;
2497 unmatch_exit:
2498 strbuf_release(&fixed);
2499 return 0;
2502 static int find_pos(struct image *img,
2503 struct image *preimage,
2504 struct image *postimage,
2505 int line,
2506 unsigned ws_rule,
2507 int match_beginning, int match_end)
2509 int i;
2510 unsigned long backwards, forwards, try;
2511 int backwards_lno, forwards_lno, try_lno;
2514 * If match_beginning or match_end is specified, there is no
2515 * point starting from a wrong line that will never match and
2516 * wander around and wait for a match at the specified end.
2518 if (match_beginning)
2519 line = 0;
2520 else if (match_end)
2521 line = img->nr - preimage->nr;
2524 * Because the comparison is unsigned, the following test
2525 * will also take care of a negative line number that can
2526 * result when match_end and preimage is larger than the target.
2528 if ((size_t) line > img->nr)
2529 line = img->nr;
2531 try = 0;
2532 for (i = 0; i < line; i++)
2533 try += img->line[i].len;
2536 * There's probably some smart way to do this, but I'll leave
2537 * that to the smart and beautiful people. I'm simple and stupid.
2539 backwards = try;
2540 backwards_lno = line;
2541 forwards = try;
2542 forwards_lno = line;
2543 try_lno = line;
2545 for (i = 0; ; i++) {
2546 if (match_fragment(img, preimage, postimage,
2547 try, try_lno, ws_rule,
2548 match_beginning, match_end))
2549 return try_lno;
2551 again:
2552 if (backwards_lno == 0 && forwards_lno == img->nr)
2553 break;
2555 if (i & 1) {
2556 if (backwards_lno == 0) {
2557 i++;
2558 goto again;
2560 backwards_lno--;
2561 backwards -= img->line[backwards_lno].len;
2562 try = backwards;
2563 try_lno = backwards_lno;
2564 } else {
2565 if (forwards_lno == img->nr) {
2566 i++;
2567 goto again;
2569 forwards += img->line[forwards_lno].len;
2570 forwards_lno++;
2571 try = forwards;
2572 try_lno = forwards_lno;
2576 return -1;
2579 static void remove_first_line(struct image *img)
2581 img->buf += img->line[0].len;
2582 img->len -= img->line[0].len;
2583 img->line++;
2584 img->nr--;
2587 static void remove_last_line(struct image *img)
2589 img->len -= img->line[--img->nr].len;
2593 * The change from "preimage" and "postimage" has been found to
2594 * apply at applied_pos (counts in line numbers) in "img".
2595 * Update "img" to remove "preimage" and replace it with "postimage".
2597 static void update_image(struct image *img,
2598 int applied_pos,
2599 struct image *preimage,
2600 struct image *postimage)
2603 * remove the copy of preimage at offset in img
2604 * and replace it with postimage
2606 int i, nr;
2607 size_t remove_count, insert_count, applied_at = 0;
2608 char *result;
2609 int preimage_limit;
2612 * If we are removing blank lines at the end of img,
2613 * the preimage may extend beyond the end.
2614 * If that is the case, we must be careful only to
2615 * remove the part of the preimage that falls within
2616 * the boundaries of img. Initialize preimage_limit
2617 * to the number of lines in the preimage that falls
2618 * within the boundaries.
2620 preimage_limit = preimage->nr;
2621 if (preimage_limit > img->nr - applied_pos)
2622 preimage_limit = img->nr - applied_pos;
2624 for (i = 0; i < applied_pos; i++)
2625 applied_at += img->line[i].len;
2627 remove_count = 0;
2628 for (i = 0; i < preimage_limit; i++)
2629 remove_count += img->line[applied_pos + i].len;
2630 insert_count = postimage->len;
2632 /* Adjust the contents */
2633 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2634 memcpy(result, img->buf, applied_at);
2635 memcpy(result + applied_at, postimage->buf, postimage->len);
2636 memcpy(result + applied_at + postimage->len,
2637 img->buf + (applied_at + remove_count),
2638 img->len - (applied_at + remove_count));
2639 free(img->buf);
2640 img->buf = result;
2641 img->len += insert_count - remove_count;
2642 result[img->len] = '\0';
2644 /* Adjust the line table */
2645 nr = img->nr + postimage->nr - preimage_limit;
2646 if (preimage_limit < postimage->nr) {
2648 * NOTE: this knows that we never call remove_first_line()
2649 * on anything other than pre/post image.
2651 REALLOC_ARRAY(img->line, nr);
2652 img->line_allocated = img->line;
2654 if (preimage_limit != postimage->nr)
2655 memmove(img->line + applied_pos + postimage->nr,
2656 img->line + applied_pos + preimage_limit,
2657 (img->nr - (applied_pos + preimage_limit)) *
2658 sizeof(*img->line));
2659 memcpy(img->line + applied_pos,
2660 postimage->line,
2661 postimage->nr * sizeof(*img->line));
2662 if (!allow_overlap)
2663 for (i = 0; i < postimage->nr; i++)
2664 img->line[applied_pos + i].flag |= LINE_PATCHED;
2665 img->nr = nr;
2669 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2670 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2671 * replace the part of "img" with "postimage" text.
2673 static int apply_one_fragment(struct image *img, struct fragment *frag,
2674 int inaccurate_eof, unsigned ws_rule,
2675 int nth_fragment)
2677 int match_beginning, match_end;
2678 const char *patch = frag->patch;
2679 int size = frag->size;
2680 char *old, *oldlines;
2681 struct strbuf newlines;
2682 int new_blank_lines_at_end = 0;
2683 int found_new_blank_lines_at_end = 0;
2684 int hunk_linenr = frag->linenr;
2685 unsigned long leading, trailing;
2686 int pos, applied_pos;
2687 struct image preimage;
2688 struct image postimage;
2690 memset(&preimage, 0, sizeof(preimage));
2691 memset(&postimage, 0, sizeof(postimage));
2692 oldlines = xmalloc(size);
2693 strbuf_init(&newlines, size);
2695 old = oldlines;
2696 while (size > 0) {
2697 char first;
2698 int len = linelen(patch, size);
2699 int plen;
2700 int added_blank_line = 0;
2701 int is_blank_context = 0;
2702 size_t start;
2704 if (!len)
2705 break;
2708 * "plen" is how much of the line we should use for
2709 * the actual patch data. Normally we just remove the
2710 * first character on the line, but if the line is
2711 * followed by "\ No newline", then we also remove the
2712 * last one (which is the newline, of course).
2714 plen = len - 1;
2715 if (len < size && patch[len] == '\\')
2716 plen--;
2717 first = *patch;
2718 if (apply_in_reverse) {
2719 if (first == '-')
2720 first = '+';
2721 else if (first == '+')
2722 first = '-';
2725 switch (first) {
2726 case '\n':
2727 /* Newer GNU diff, empty context line */
2728 if (plen < 0)
2729 /* ... followed by '\No newline'; nothing */
2730 break;
2731 *old++ = '\n';
2732 strbuf_addch(&newlines, '\n');
2733 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2734 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2735 is_blank_context = 1;
2736 break;
2737 case ' ':
2738 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2739 ws_blank_line(patch + 1, plen, ws_rule))
2740 is_blank_context = 1;
2741 case '-':
2742 memcpy(old, patch + 1, plen);
2743 add_line_info(&preimage, old, plen,
2744 (first == ' ' ? LINE_COMMON : 0));
2745 old += plen;
2746 if (first == '-')
2747 break;
2748 /* Fall-through for ' ' */
2749 case '+':
2750 /* --no-add does not add new lines */
2751 if (first == '+' && no_add)
2752 break;
2754 start = newlines.len;
2755 if (first != '+' ||
2756 !whitespace_error ||
2757 ws_error_action != correct_ws_error) {
2758 strbuf_add(&newlines, patch + 1, plen);
2760 else {
2761 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2763 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2764 (first == '+' ? 0 : LINE_COMMON));
2765 if (first == '+' &&
2766 (ws_rule & WS_BLANK_AT_EOF) &&
2767 ws_blank_line(patch + 1, plen, ws_rule))
2768 added_blank_line = 1;
2769 break;
2770 case '@': case '\\':
2771 /* Ignore it, we already handled it */
2772 break;
2773 default:
2774 if (apply_verbosely)
2775 error(_("invalid start of line: '%c'"), first);
2776 applied_pos = -1;
2777 goto out;
2779 if (added_blank_line) {
2780 if (!new_blank_lines_at_end)
2781 found_new_blank_lines_at_end = hunk_linenr;
2782 new_blank_lines_at_end++;
2784 else if (is_blank_context)
2786 else
2787 new_blank_lines_at_end = 0;
2788 patch += len;
2789 size -= len;
2790 hunk_linenr++;
2792 if (inaccurate_eof &&
2793 old > oldlines && old[-1] == '\n' &&
2794 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2795 old--;
2796 strbuf_setlen(&newlines, newlines.len - 1);
2799 leading = frag->leading;
2800 trailing = frag->trailing;
2803 * A hunk to change lines at the beginning would begin with
2804 * @@ -1,L +N,M @@
2805 * but we need to be careful. -U0 that inserts before the second
2806 * line also has this pattern.
2808 * And a hunk to add to an empty file would begin with
2809 * @@ -0,0 +N,M @@
2811 * In other words, a hunk that is (frag->oldpos <= 1) with or
2812 * without leading context must match at the beginning.
2814 match_beginning = (!frag->oldpos ||
2815 (frag->oldpos == 1 && !unidiff_zero));
2818 * A hunk without trailing lines must match at the end.
2819 * However, we simply cannot tell if a hunk must match end
2820 * from the lack of trailing lines if the patch was generated
2821 * with unidiff without any context.
2823 match_end = !unidiff_zero && !trailing;
2825 pos = frag->newpos ? (frag->newpos - 1) : 0;
2826 preimage.buf = oldlines;
2827 preimage.len = old - oldlines;
2828 postimage.buf = newlines.buf;
2829 postimage.len = newlines.len;
2830 preimage.line = preimage.line_allocated;
2831 postimage.line = postimage.line_allocated;
2833 for (;;) {
2835 applied_pos = find_pos(img, &preimage, &postimage, pos,
2836 ws_rule, match_beginning, match_end);
2838 if (applied_pos >= 0)
2839 break;
2841 /* Am I at my context limits? */
2842 if ((leading <= p_context) && (trailing <= p_context))
2843 break;
2844 if (match_beginning || match_end) {
2845 match_beginning = match_end = 0;
2846 continue;
2850 * Reduce the number of context lines; reduce both
2851 * leading and trailing if they are equal otherwise
2852 * just reduce the larger context.
2854 if (leading >= trailing) {
2855 remove_first_line(&preimage);
2856 remove_first_line(&postimage);
2857 pos--;
2858 leading--;
2860 if (trailing > leading) {
2861 remove_last_line(&preimage);
2862 remove_last_line(&postimage);
2863 trailing--;
2867 if (applied_pos >= 0) {
2868 if (new_blank_lines_at_end &&
2869 preimage.nr + applied_pos >= img->nr &&
2870 (ws_rule & WS_BLANK_AT_EOF) &&
2871 ws_error_action != nowarn_ws_error) {
2872 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2873 found_new_blank_lines_at_end);
2874 if (ws_error_action == correct_ws_error) {
2875 while (new_blank_lines_at_end--)
2876 remove_last_line(&postimage);
2879 * We would want to prevent write_out_results()
2880 * from taking place in apply_patch() that follows
2881 * the callchain led us here, which is:
2882 * apply_patch->check_patch_list->check_patch->
2883 * apply_data->apply_fragments->apply_one_fragment
2885 if (ws_error_action == die_on_ws_error)
2886 apply = 0;
2889 if (apply_verbosely && applied_pos != pos) {
2890 int offset = applied_pos - pos;
2891 if (apply_in_reverse)
2892 offset = 0 - offset;
2893 fprintf_ln(stderr,
2894 Q_("Hunk #%d succeeded at %d (offset %d line).",
2895 "Hunk #%d succeeded at %d (offset %d lines).",
2896 offset),
2897 nth_fragment, applied_pos + 1, offset);
2901 * Warn if it was necessary to reduce the number
2902 * of context lines.
2904 if ((leading != frag->leading) ||
2905 (trailing != frag->trailing))
2906 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2907 " to apply fragment at %d"),
2908 leading, trailing, applied_pos+1);
2909 update_image(img, applied_pos, &preimage, &postimage);
2910 } else {
2911 if (apply_verbosely)
2912 error(_("while searching for:\n%.*s"),
2913 (int)(old - oldlines), oldlines);
2916 out:
2917 free(oldlines);
2918 strbuf_release(&newlines);
2919 free(preimage.line_allocated);
2920 free(postimage.line_allocated);
2922 return (applied_pos < 0);
2925 static int apply_binary_fragment(struct image *img, struct patch *patch)
2927 struct fragment *fragment = patch->fragments;
2928 unsigned long len;
2929 void *dst;
2931 if (!fragment)
2932 return error(_("missing binary patch data for '%s'"),
2933 patch->new_name ?
2934 patch->new_name :
2935 patch->old_name);
2937 /* Binary patch is irreversible without the optional second hunk */
2938 if (apply_in_reverse) {
2939 if (!fragment->next)
2940 return error("cannot reverse-apply a binary patch "
2941 "without the reverse hunk to '%s'",
2942 patch->new_name
2943 ? patch->new_name : patch->old_name);
2944 fragment = fragment->next;
2946 switch (fragment->binary_patch_method) {
2947 case BINARY_DELTA_DEFLATED:
2948 dst = patch_delta(img->buf, img->len, fragment->patch,
2949 fragment->size, &len);
2950 if (!dst)
2951 return -1;
2952 clear_image(img);
2953 img->buf = dst;
2954 img->len = len;
2955 return 0;
2956 case BINARY_LITERAL_DEFLATED:
2957 clear_image(img);
2958 img->len = fragment->size;
2959 img->buf = xmemdupz(fragment->patch, img->len);
2960 return 0;
2962 return -1;
2966 * Replace "img" with the result of applying the binary patch.
2967 * The binary patch data itself in patch->fragment is still kept
2968 * but the preimage prepared by the caller in "img" is freed here
2969 * or in the helper function apply_binary_fragment() this calls.
2971 static int apply_binary(struct image *img, struct patch *patch)
2973 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2974 unsigned char sha1[20];
2977 * For safety, we require patch index line to contain
2978 * full 40-byte textual SHA1 for old and new, at least for now.
2980 if (strlen(patch->old_sha1_prefix) != 40 ||
2981 strlen(patch->new_sha1_prefix) != 40 ||
2982 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2983 get_sha1_hex(patch->new_sha1_prefix, sha1))
2984 return error("cannot apply binary patch to '%s' "
2985 "without full index line", name);
2987 if (patch->old_name) {
2989 * See if the old one matches what the patch
2990 * applies to.
2992 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2993 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2994 return error("the patch applies to '%s' (%s), "
2995 "which does not match the "
2996 "current contents.",
2997 name, sha1_to_hex(sha1));
2999 else {
3000 /* Otherwise, the old one must be empty. */
3001 if (img->len)
3002 return error("the patch applies to an empty "
3003 "'%s' but it is not empty", name);
3006 get_sha1_hex(patch->new_sha1_prefix, sha1);
3007 if (is_null_sha1(sha1)) {
3008 clear_image(img);
3009 return 0; /* deletion patch */
3012 if (has_sha1_file(sha1)) {
3013 /* We already have the postimage */
3014 enum object_type type;
3015 unsigned long size;
3016 char *result;
3018 result = read_sha1_file(sha1, &type, &size);
3019 if (!result)
3020 return error("the necessary postimage %s for "
3021 "'%s' cannot be read",
3022 patch->new_sha1_prefix, name);
3023 clear_image(img);
3024 img->buf = result;
3025 img->len = size;
3026 } else {
3028 * We have verified buf matches the preimage;
3029 * apply the patch data to it, which is stored
3030 * in the patch->fragments->{patch,size}.
3032 if (apply_binary_fragment(img, patch))
3033 return error(_("binary patch does not apply to '%s'"),
3034 name);
3036 /* verify that the result matches */
3037 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3038 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3039 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3040 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3043 return 0;
3046 static int apply_fragments(struct image *img, struct patch *patch)
3048 struct fragment *frag = patch->fragments;
3049 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3050 unsigned ws_rule = patch->ws_rule;
3051 unsigned inaccurate_eof = patch->inaccurate_eof;
3052 int nth = 0;
3054 if (patch->is_binary)
3055 return apply_binary(img, patch);
3057 while (frag) {
3058 nth++;
3059 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3060 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3061 if (!apply_with_reject)
3062 return -1;
3063 frag->rejected = 1;
3065 frag = frag->next;
3067 return 0;
3070 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3072 if (S_ISGITLINK(mode)) {
3073 strbuf_grow(buf, 100);
3074 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3075 } else {
3076 enum object_type type;
3077 unsigned long sz;
3078 char *result;
3080 result = read_sha1_file(sha1, &type, &sz);
3081 if (!result)
3082 return -1;
3083 /* XXX read_sha1_file NUL-terminates */
3084 strbuf_attach(buf, result, sz, sz + 1);
3086 return 0;
3089 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3091 if (!ce)
3092 return 0;
3093 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3096 static struct patch *in_fn_table(const char *name)
3098 struct string_list_item *item;
3100 if (name == NULL)
3101 return NULL;
3103 item = string_list_lookup(&fn_table, name);
3104 if (item != NULL)
3105 return (struct patch *)item->util;
3107 return NULL;
3111 * item->util in the filename table records the status of the path.
3112 * Usually it points at a patch (whose result records the contents
3113 * of it after applying it), but it could be PATH_WAS_DELETED for a
3114 * path that a previously applied patch has already removed, or
3115 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3117 * The latter is needed to deal with a case where two paths A and B
3118 * are swapped by first renaming A to B and then renaming B to A;
3119 * moving A to B should not be prevented due to presence of B as we
3120 * will remove it in a later patch.
3122 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3123 #define PATH_WAS_DELETED ((struct patch *) -1)
3125 static int to_be_deleted(struct patch *patch)
3127 return patch == PATH_TO_BE_DELETED;
3130 static int was_deleted(struct patch *patch)
3132 return patch == PATH_WAS_DELETED;
3135 static void add_to_fn_table(struct patch *patch)
3137 struct string_list_item *item;
3140 * Always add new_name unless patch is a deletion
3141 * This should cover the cases for normal diffs,
3142 * file creations and copies
3144 if (patch->new_name != NULL) {
3145 item = string_list_insert(&fn_table, patch->new_name);
3146 item->util = patch;
3150 * store a failure on rename/deletion cases because
3151 * later chunks shouldn't patch old names
3153 if ((patch->new_name == NULL) || (patch->is_rename)) {
3154 item = string_list_insert(&fn_table, patch->old_name);
3155 item->util = PATH_WAS_DELETED;
3159 static void prepare_fn_table(struct patch *patch)
3162 * store information about incoming file deletion
3164 while (patch) {
3165 if ((patch->new_name == NULL) || (patch->is_rename)) {
3166 struct string_list_item *item;
3167 item = string_list_insert(&fn_table, patch->old_name);
3168 item->util = PATH_TO_BE_DELETED;
3170 patch = patch->next;
3174 static int checkout_target(struct index_state *istate,
3175 struct cache_entry *ce, struct stat *st)
3177 struct checkout costate;
3179 memset(&costate, 0, sizeof(costate));
3180 costate.base_dir = "";
3181 costate.refresh_cache = 1;
3182 costate.istate = istate;
3183 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3184 return error(_("cannot checkout %s"), ce->name);
3185 return 0;
3188 static struct patch *previous_patch(struct patch *patch, int *gone)
3190 struct patch *previous;
3192 *gone = 0;
3193 if (patch->is_copy || patch->is_rename)
3194 return NULL; /* "git" patches do not depend on the order */
3196 previous = in_fn_table(patch->old_name);
3197 if (!previous)
3198 return NULL;
3200 if (to_be_deleted(previous))
3201 return NULL; /* the deletion hasn't happened yet */
3203 if (was_deleted(previous))
3204 *gone = 1;
3206 return previous;
3209 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3211 if (S_ISGITLINK(ce->ce_mode)) {
3212 if (!S_ISDIR(st->st_mode))
3213 return -1;
3214 return 0;
3216 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3219 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3221 static int load_patch_target(struct strbuf *buf,
3222 const struct cache_entry *ce,
3223 struct stat *st,
3224 const char *name,
3225 unsigned expected_mode)
3227 if (cached || check_index) {
3228 if (read_file_or_gitlink(ce, buf))
3229 return error(_("read of %s failed"), name);
3230 } else if (name) {
3231 if (S_ISGITLINK(expected_mode)) {
3232 if (ce)
3233 return read_file_or_gitlink(ce, buf);
3234 else
3235 return SUBMODULE_PATCH_WITHOUT_INDEX;
3236 } else if (has_symlink_leading_path(name, strlen(name))) {
3237 return error(_("reading from '%s' beyond a symbolic link"), name);
3238 } else {
3239 if (read_old_data(st, name, buf))
3240 return error(_("read of %s failed"), name);
3243 return 0;
3247 * We are about to apply "patch"; populate the "image" with the
3248 * current version we have, from the working tree or from the index,
3249 * depending on the situation e.g. --cached/--index. If we are
3250 * applying a non-git patch that incrementally updates the tree,
3251 * we read from the result of a previous diff.
3253 static int load_preimage(struct image *image,
3254 struct patch *patch, struct stat *st,
3255 const struct cache_entry *ce)
3257 struct strbuf buf = STRBUF_INIT;
3258 size_t len;
3259 char *img;
3260 struct patch *previous;
3261 int status;
3263 previous = previous_patch(patch, &status);
3264 if (status)
3265 return error(_("path %s has been renamed/deleted"),
3266 patch->old_name);
3267 if (previous) {
3268 /* We have a patched copy in memory; use that. */
3269 strbuf_add(&buf, previous->result, previous->resultsize);
3270 } else {
3271 status = load_patch_target(&buf, ce, st,
3272 patch->old_name, patch->old_mode);
3273 if (status < 0)
3274 return status;
3275 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3277 * There is no way to apply subproject
3278 * patch without looking at the index.
3279 * NEEDSWORK: shouldn't this be flagged
3280 * as an error???
3282 free_fragment_list(patch->fragments);
3283 patch->fragments = NULL;
3284 } else if (status) {
3285 return error(_("read of %s failed"), patch->old_name);
3289 img = strbuf_detach(&buf, &len);
3290 prepare_image(image, img, len, !patch->is_binary);
3291 return 0;
3294 static int three_way_merge(struct image *image,
3295 char *path,
3296 const unsigned char *base,
3297 const unsigned char *ours,
3298 const unsigned char *theirs)
3300 mmfile_t base_file, our_file, their_file;
3301 mmbuffer_t result = { NULL };
3302 int status;
3304 read_mmblob(&base_file, base);
3305 read_mmblob(&our_file, ours);
3306 read_mmblob(&their_file, theirs);
3307 status = ll_merge(&result, path,
3308 &base_file, "base",
3309 &our_file, "ours",
3310 &their_file, "theirs", NULL);
3311 free(base_file.ptr);
3312 free(our_file.ptr);
3313 free(their_file.ptr);
3314 if (status < 0 || !result.ptr) {
3315 free(result.ptr);
3316 return -1;
3318 clear_image(image);
3319 image->buf = result.ptr;
3320 image->len = result.size;
3322 return status;
3326 * When directly falling back to add/add three-way merge, we read from
3327 * the current contents of the new_name. In no cases other than that
3328 * this function will be called.
3330 static int load_current(struct image *image, struct patch *patch)
3332 struct strbuf buf = STRBUF_INIT;
3333 int status, pos;
3334 size_t len;
3335 char *img;
3336 struct stat st;
3337 struct cache_entry *ce;
3338 char *name = patch->new_name;
3339 unsigned mode = patch->new_mode;
3341 if (!patch->is_new)
3342 die("BUG: patch to %s is not a creation", patch->old_name);
3344 pos = cache_name_pos(name, strlen(name));
3345 if (pos < 0)
3346 return error(_("%s: does not exist in index"), name);
3347 ce = active_cache[pos];
3348 if (lstat(name, &st)) {
3349 if (errno != ENOENT)
3350 return error(_("%s: %s"), name, strerror(errno));
3351 if (checkout_target(&the_index, ce, &st))
3352 return -1;
3354 if (verify_index_match(ce, &st))
3355 return error(_("%s: does not match index"), name);
3357 status = load_patch_target(&buf, ce, &st, name, mode);
3358 if (status < 0)
3359 return status;
3360 else if (status)
3361 return -1;
3362 img = strbuf_detach(&buf, &len);
3363 prepare_image(image, img, len, !patch->is_binary);
3364 return 0;
3367 static int try_threeway(struct image *image, struct patch *patch,
3368 struct stat *st, const struct cache_entry *ce)
3370 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3371 struct strbuf buf = STRBUF_INIT;
3372 size_t len;
3373 int status;
3374 char *img;
3375 struct image tmp_image;
3377 /* No point falling back to 3-way merge in these cases */
3378 if (patch->is_delete ||
3379 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3380 return -1;
3382 /* Preimage the patch was prepared for */
3383 if (patch->is_new)
3384 write_sha1_file("", 0, blob_type, pre_sha1);
3385 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3386 read_blob_object(&buf, pre_sha1, patch->old_mode))
3387 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3389 fprintf(stderr, "Falling back to three-way merge...\n");
3391 img = strbuf_detach(&buf, &len);
3392 prepare_image(&tmp_image, img, len, 1);
3393 /* Apply the patch to get the post image */
3394 if (apply_fragments(&tmp_image, patch) < 0) {
3395 clear_image(&tmp_image);
3396 return -1;
3398 /* post_sha1[] is theirs */
3399 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3400 clear_image(&tmp_image);
3402 /* our_sha1[] is ours */
3403 if (patch->is_new) {
3404 if (load_current(&tmp_image, patch))
3405 return error("cannot read the current contents of '%s'",
3406 patch->new_name);
3407 } else {
3408 if (load_preimage(&tmp_image, patch, st, ce))
3409 return error("cannot read the current contents of '%s'",
3410 patch->old_name);
3412 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3413 clear_image(&tmp_image);
3415 /* in-core three-way merge between post and our using pre as base */
3416 status = three_way_merge(image, patch->new_name,
3417 pre_sha1, our_sha1, post_sha1);
3418 if (status < 0) {
3419 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3420 return status;
3423 if (status) {
3424 patch->conflicted_threeway = 1;
3425 if (patch->is_new)
3426 oidclr(&patch->threeway_stage[0]);
3427 else
3428 hashcpy(patch->threeway_stage[0].hash, pre_sha1);
3429 hashcpy(patch->threeway_stage[1].hash, our_sha1);
3430 hashcpy(patch->threeway_stage[2].hash, post_sha1);
3431 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3432 } else {
3433 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3435 return 0;
3438 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3440 struct image image;
3442 if (load_preimage(&image, patch, st, ce) < 0)
3443 return -1;
3445 if (patch->direct_to_threeway ||
3446 apply_fragments(&image, patch) < 0) {
3447 /* Note: with --reject, apply_fragments() returns 0 */
3448 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3449 return -1;
3451 patch->result = image.buf;
3452 patch->resultsize = image.len;
3453 add_to_fn_table(patch);
3454 free(image.line_allocated);
3456 if (0 < patch->is_delete && patch->resultsize)
3457 return error(_("removal patch leaves file contents"));
3459 return 0;
3463 * If "patch" that we are looking at modifies or deletes what we have,
3464 * we would want it not to lose any local modification we have, either
3465 * in the working tree or in the index.
3467 * This also decides if a non-git patch is a creation patch or a
3468 * modification to an existing empty file. We do not check the state
3469 * of the current tree for a creation patch in this function; the caller
3470 * check_patch() separately makes sure (and errors out otherwise) that
3471 * the path the patch creates does not exist in the current tree.
3473 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3475 const char *old_name = patch->old_name;
3476 struct patch *previous = NULL;
3477 int stat_ret = 0, status;
3478 unsigned st_mode = 0;
3480 if (!old_name)
3481 return 0;
3483 assert(patch->is_new <= 0);
3484 previous = previous_patch(patch, &status);
3486 if (status)
3487 return error(_("path %s has been renamed/deleted"), old_name);
3488 if (previous) {
3489 st_mode = previous->new_mode;
3490 } else if (!cached) {
3491 stat_ret = lstat(old_name, st);
3492 if (stat_ret && errno != ENOENT)
3493 return error(_("%s: %s"), old_name, strerror(errno));
3496 if (check_index && !previous) {
3497 int pos = cache_name_pos(old_name, strlen(old_name));
3498 if (pos < 0) {
3499 if (patch->is_new < 0)
3500 goto is_new;
3501 return error(_("%s: does not exist in index"), old_name);
3503 *ce = active_cache[pos];
3504 if (stat_ret < 0) {
3505 if (checkout_target(&the_index, *ce, st))
3506 return -1;
3508 if (!cached && verify_index_match(*ce, st))
3509 return error(_("%s: does not match index"), old_name);
3510 if (cached)
3511 st_mode = (*ce)->ce_mode;
3512 } else if (stat_ret < 0) {
3513 if (patch->is_new < 0)
3514 goto is_new;
3515 return error(_("%s: %s"), old_name, strerror(errno));
3518 if (!cached && !previous)
3519 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3521 if (patch->is_new < 0)
3522 patch->is_new = 0;
3523 if (!patch->old_mode)
3524 patch->old_mode = st_mode;
3525 if ((st_mode ^ patch->old_mode) & S_IFMT)
3526 return error(_("%s: wrong type"), old_name);
3527 if (st_mode != patch->old_mode)
3528 warning(_("%s has type %o, expected %o"),
3529 old_name, st_mode, patch->old_mode);
3530 if (!patch->new_mode && !patch->is_delete)
3531 patch->new_mode = st_mode;
3532 return 0;
3534 is_new:
3535 patch->is_new = 1;
3536 patch->is_delete = 0;
3537 free(patch->old_name);
3538 patch->old_name = NULL;
3539 return 0;
3543 #define EXISTS_IN_INDEX 1
3544 #define EXISTS_IN_WORKTREE 2
3546 static int check_to_create(const char *new_name, int ok_if_exists)
3548 struct stat nst;
3550 if (check_index &&
3551 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3552 !ok_if_exists)
3553 return EXISTS_IN_INDEX;
3554 if (cached)
3555 return 0;
3557 if (!lstat(new_name, &nst)) {
3558 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3559 return 0;
3561 * A leading component of new_name might be a symlink
3562 * that is going to be removed with this patch, but
3563 * still pointing at somewhere that has the path.
3564 * In such a case, path "new_name" does not exist as
3565 * far as git is concerned.
3567 if (has_symlink_leading_path(new_name, strlen(new_name)))
3568 return 0;
3570 return EXISTS_IN_WORKTREE;
3571 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3572 return error("%s: %s", new_name, strerror(errno));
3574 return 0;
3578 * We need to keep track of how symlinks in the preimage are
3579 * manipulated by the patches. A patch to add a/b/c where a/b
3580 * is a symlink should not be allowed to affect the directory
3581 * the symlink points at, but if the same patch removes a/b,
3582 * it is perfectly fine, as the patch removes a/b to make room
3583 * to create a directory a/b so that a/b/c can be created.
3585 static struct string_list symlink_changes;
3586 #define SYMLINK_GOES_AWAY 01
3587 #define SYMLINK_IN_RESULT 02
3589 static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3591 struct string_list_item *ent;
3593 ent = string_list_lookup(&symlink_changes, path);
3594 if (!ent) {
3595 ent = string_list_insert(&symlink_changes, path);
3596 ent->util = (void *)0;
3598 ent->util = (void *)(what | ((uintptr_t)ent->util));
3599 return (uintptr_t)ent->util;
3602 static uintptr_t check_symlink_changes(const char *path)
3604 struct string_list_item *ent;
3606 ent = string_list_lookup(&symlink_changes, path);
3607 if (!ent)
3608 return 0;
3609 return (uintptr_t)ent->util;
3612 static void prepare_symlink_changes(struct patch *patch)
3614 for ( ; patch; patch = patch->next) {
3615 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3616 (patch->is_rename || patch->is_delete))
3617 /* the symlink at patch->old_name is removed */
3618 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3620 if (patch->new_name && S_ISLNK(patch->new_mode))
3621 /* the symlink at patch->new_name is created or remains */
3622 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3626 static int path_is_beyond_symlink_1(struct strbuf *name)
3628 do {
3629 unsigned int change;
3631 while (--name->len && name->buf[name->len] != '/')
3632 ; /* scan backwards */
3633 if (!name->len)
3634 break;
3635 name->buf[name->len] = '\0';
3636 change = check_symlink_changes(name->buf);
3637 if (change & SYMLINK_IN_RESULT)
3638 return 1;
3639 if (change & SYMLINK_GOES_AWAY)
3641 * This cannot be "return 0", because we may
3642 * see a new one created at a higher level.
3644 continue;
3646 /* otherwise, check the preimage */
3647 if (check_index) {
3648 struct cache_entry *ce;
3650 ce = cache_file_exists(name->buf, name->len, ignore_case);
3651 if (ce && S_ISLNK(ce->ce_mode))
3652 return 1;
3653 } else {
3654 struct stat st;
3655 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3656 return 1;
3658 } while (1);
3659 return 0;
3662 static int path_is_beyond_symlink(const char *name_)
3664 int ret;
3665 struct strbuf name = STRBUF_INIT;
3667 assert(*name_ != '\0');
3668 strbuf_addstr(&name, name_);
3669 ret = path_is_beyond_symlink_1(&name);
3670 strbuf_release(&name);
3672 return ret;
3675 static void die_on_unsafe_path(struct patch *patch)
3677 const char *old_name = NULL;
3678 const char *new_name = NULL;
3679 if (patch->is_delete)
3680 old_name = patch->old_name;
3681 else if (!patch->is_new && !patch->is_copy)
3682 old_name = patch->old_name;
3683 if (!patch->is_delete)
3684 new_name = patch->new_name;
3686 if (old_name && !verify_path(old_name))
3687 die(_("invalid path '%s'"), old_name);
3688 if (new_name && !verify_path(new_name))
3689 die(_("invalid path '%s'"), new_name);
3693 * Check and apply the patch in-core; leave the result in patch->result
3694 * for the caller to write it out to the final destination.
3696 static int check_patch(struct patch *patch)
3698 struct stat st;
3699 const char *old_name = patch->old_name;
3700 const char *new_name = patch->new_name;
3701 const char *name = old_name ? old_name : new_name;
3702 struct cache_entry *ce = NULL;
3703 struct patch *tpatch;
3704 int ok_if_exists;
3705 int status;
3707 patch->rejected = 1; /* we will drop this after we succeed */
3709 status = check_preimage(patch, &ce, &st);
3710 if (status)
3711 return status;
3712 old_name = patch->old_name;
3715 * A type-change diff is always split into a patch to delete
3716 * old, immediately followed by a patch to create new (see
3717 * diff.c::run_diff()); in such a case it is Ok that the entry
3718 * to be deleted by the previous patch is still in the working
3719 * tree and in the index.
3721 * A patch to swap-rename between A and B would first rename A
3722 * to B and then rename B to A. While applying the first one,
3723 * the presence of B should not stop A from getting renamed to
3724 * B; ask to_be_deleted() about the later rename. Removal of
3725 * B and rename from A to B is handled the same way by asking
3726 * was_deleted().
3728 if ((tpatch = in_fn_table(new_name)) &&
3729 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3730 ok_if_exists = 1;
3731 else
3732 ok_if_exists = 0;
3734 if (new_name &&
3735 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3736 int err = check_to_create(new_name, ok_if_exists);
3738 if (err && threeway) {
3739 patch->direct_to_threeway = 1;
3740 } else switch (err) {
3741 case 0:
3742 break; /* happy */
3743 case EXISTS_IN_INDEX:
3744 return error(_("%s: already exists in index"), new_name);
3745 break;
3746 case EXISTS_IN_WORKTREE:
3747 return error(_("%s: already exists in working directory"),
3748 new_name);
3749 default:
3750 return err;
3753 if (!patch->new_mode) {
3754 if (0 < patch->is_new)
3755 patch->new_mode = S_IFREG | 0644;
3756 else
3757 patch->new_mode = patch->old_mode;
3761 if (new_name && old_name) {
3762 int same = !strcmp(old_name, new_name);
3763 if (!patch->new_mode)
3764 patch->new_mode = patch->old_mode;
3765 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3766 if (same)
3767 return error(_("new mode (%o) of %s does not "
3768 "match old mode (%o)"),
3769 patch->new_mode, new_name,
3770 patch->old_mode);
3771 else
3772 return error(_("new mode (%o) of %s does not "
3773 "match old mode (%o) of %s"),
3774 patch->new_mode, new_name,
3775 patch->old_mode, old_name);
3779 if (!unsafe_paths)
3780 die_on_unsafe_path(patch);
3783 * An attempt to read from or delete a path that is beyond a
3784 * symbolic link will be prevented by load_patch_target() that
3785 * is called at the beginning of apply_data() so we do not
3786 * have to worry about a patch marked with "is_delete" bit
3787 * here. We however need to make sure that the patch result
3788 * is not deposited to a path that is beyond a symbolic link
3789 * here.
3791 if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))
3792 return error(_("affected file '%s' is beyond a symbolic link"),
3793 patch->new_name);
3795 if (apply_data(patch, &st, ce) < 0)
3796 return error(_("%s: patch does not apply"), name);
3797 patch->rejected = 0;
3798 return 0;
3801 static int check_patch_list(struct patch *patch)
3803 int err = 0;
3805 prepare_symlink_changes(patch);
3806 prepare_fn_table(patch);
3807 while (patch) {
3808 if (apply_verbosely)
3809 say_patch_name(stderr,
3810 _("Checking patch %s..."), patch);
3811 err |= check_patch(patch);
3812 patch = patch->next;
3814 return err;
3817 /* This function tries to read the sha1 from the current index */
3818 static int get_current_sha1(const char *path, unsigned char *sha1)
3820 int pos;
3822 if (read_cache() < 0)
3823 return -1;
3824 pos = cache_name_pos(path, strlen(path));
3825 if (pos < 0)
3826 return -1;
3827 hashcpy(sha1, active_cache[pos]->sha1);
3828 return 0;
3831 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3834 * A usable gitlink patch has only one fragment (hunk) that looks like:
3835 * @@ -1 +1 @@
3836 * -Subproject commit <old sha1>
3837 * +Subproject commit <new sha1>
3838 * or
3839 * @@ -1 +0,0 @@
3840 * -Subproject commit <old sha1>
3841 * for a removal patch.
3843 struct fragment *hunk = p->fragments;
3844 static const char heading[] = "-Subproject commit ";
3845 char *preimage;
3847 if (/* does the patch have only one hunk? */
3848 hunk && !hunk->next &&
3849 /* is its preimage one line? */
3850 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3851 /* does preimage begin with the heading? */
3852 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3853 starts_with(++preimage, heading) &&
3854 /* does it record full SHA-1? */
3855 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3856 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3857 /* does the abbreviated name on the index line agree with it? */
3858 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3859 return 0; /* it all looks fine */
3861 /* we may have full object name on the index line */
3862 return get_sha1_hex(p->old_sha1_prefix, sha1);
3865 /* Build an index that contains the just the files needed for a 3way merge */
3866 static void build_fake_ancestor(struct patch *list, const char *filename)
3868 struct patch *patch;
3869 struct index_state result = { NULL };
3870 static struct lock_file lock;
3872 /* Once we start supporting the reverse patch, it may be
3873 * worth showing the new sha1 prefix, but until then...
3875 for (patch = list; patch; patch = patch->next) {
3876 unsigned char sha1[20];
3877 struct cache_entry *ce;
3878 const char *name;
3880 name = patch->old_name ? patch->old_name : patch->new_name;
3881 if (0 < patch->is_new)
3882 continue;
3884 if (S_ISGITLINK(patch->old_mode)) {
3885 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3886 ; /* ok, the textual part looks sane */
3887 else
3888 die("sha1 information is lacking or useless for submodule %s",
3889 name);
3890 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3891 ; /* ok */
3892 } else if (!patch->lines_added && !patch->lines_deleted) {
3893 /* mode-only change: update the current */
3894 if (get_current_sha1(patch->old_name, sha1))
3895 die("mode change for %s, which is not "
3896 "in current HEAD", name);
3897 } else
3898 die("sha1 information is lacking or useless "
3899 "(%s).", name);
3901 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3902 if (!ce)
3903 die(_("make_cache_entry failed for path '%s'"), name);
3904 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3905 die ("Could not add %s to temporary index", name);
3908 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3909 if (write_locked_index(&result, &lock, COMMIT_LOCK))
3910 die ("Could not write temporary index to %s", filename);
3912 discard_index(&result);
3915 static void stat_patch_list(struct patch *patch)
3917 int files, adds, dels;
3919 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3920 files++;
3921 adds += patch->lines_added;
3922 dels += patch->lines_deleted;
3923 show_stats(patch);
3926 print_stat_summary(stdout, files, adds, dels);
3929 static void numstat_patch_list(struct patch *patch)
3931 for ( ; patch; patch = patch->next) {
3932 const char *name;
3933 name = patch->new_name ? patch->new_name : patch->old_name;
3934 if (patch->is_binary)
3935 printf("-\t-\t");
3936 else
3937 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3938 write_name_quoted(name, stdout, line_termination);
3942 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3944 if (mode)
3945 printf(" %s mode %06o %s\n", newdelete, mode, name);
3946 else
3947 printf(" %s %s\n", newdelete, name);
3950 static void show_mode_change(struct patch *p, int show_name)
3952 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3953 if (show_name)
3954 printf(" mode change %06o => %06o %s\n",
3955 p->old_mode, p->new_mode, p->new_name);
3956 else
3957 printf(" mode change %06o => %06o\n",
3958 p->old_mode, p->new_mode);
3962 static void show_rename_copy(struct patch *p)
3964 const char *renamecopy = p->is_rename ? "rename" : "copy";
3965 const char *old, *new;
3967 /* Find common prefix */
3968 old = p->old_name;
3969 new = p->new_name;
3970 while (1) {
3971 const char *slash_old, *slash_new;
3972 slash_old = strchr(old, '/');
3973 slash_new = strchr(new, '/');
3974 if (!slash_old ||
3975 !slash_new ||
3976 slash_old - old != slash_new - new ||
3977 memcmp(old, new, slash_new - new))
3978 break;
3979 old = slash_old + 1;
3980 new = slash_new + 1;
3982 /* p->old_name thru old is the common prefix, and old and new
3983 * through the end of names are renames
3985 if (old != p->old_name)
3986 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3987 (int)(old - p->old_name), p->old_name,
3988 old, new, p->score);
3989 else
3990 printf(" %s %s => %s (%d%%)\n", renamecopy,
3991 p->old_name, p->new_name, p->score);
3992 show_mode_change(p, 0);
3995 static void summary_patch_list(struct patch *patch)
3997 struct patch *p;
3999 for (p = patch; p; p = p->next) {
4000 if (p->is_new)
4001 show_file_mode_name("create", p->new_mode, p->new_name);
4002 else if (p->is_delete)
4003 show_file_mode_name("delete", p->old_mode, p->old_name);
4004 else {
4005 if (p->is_rename || p->is_copy)
4006 show_rename_copy(p);
4007 else {
4008 if (p->score) {
4009 printf(" rewrite %s (%d%%)\n",
4010 p->new_name, p->score);
4011 show_mode_change(p, 0);
4013 else
4014 show_mode_change(p, 1);
4020 static void patch_stats(struct patch *patch)
4022 int lines = patch->lines_added + patch->lines_deleted;
4024 if (lines > max_change)
4025 max_change = lines;
4026 if (patch->old_name) {
4027 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4028 if (!len)
4029 len = strlen(patch->old_name);
4030 if (len > max_len)
4031 max_len = len;
4033 if (patch->new_name) {
4034 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4035 if (!len)
4036 len = strlen(patch->new_name);
4037 if (len > max_len)
4038 max_len = len;
4042 static void remove_file(struct patch *patch, int rmdir_empty)
4044 if (update_index) {
4045 if (remove_file_from_cache(patch->old_name) < 0)
4046 die(_("unable to remove %s from index"), patch->old_name);
4048 if (!cached) {
4049 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4050 remove_path(patch->old_name);
4055 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
4057 struct stat st;
4058 struct cache_entry *ce;
4059 int namelen = strlen(path);
4060 unsigned ce_size = cache_entry_size(namelen);
4062 if (!update_index)
4063 return;
4065 ce = xcalloc(1, ce_size);
4066 memcpy(ce->name, path, namelen);
4067 ce->ce_mode = create_ce_mode(mode);
4068 ce->ce_flags = create_ce_flags(0);
4069 ce->ce_namelen = namelen;
4070 if (S_ISGITLINK(mode)) {
4071 const char *s;
4073 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4074 get_sha1_hex(s, ce->sha1))
4075 die(_("corrupt patch for submodule %s"), path);
4076 } else {
4077 if (!cached) {
4078 if (lstat(path, &st) < 0)
4079 die_errno(_("unable to stat newly created file '%s'"),
4080 path);
4081 fill_stat_cache_info(ce, &st);
4083 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
4084 die(_("unable to create backing store for newly created file %s"), path);
4086 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4087 die(_("unable to add cache entry for %s"), path);
4090 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4092 int fd;
4093 struct strbuf nbuf = STRBUF_INIT;
4095 if (S_ISGITLINK(mode)) {
4096 struct stat st;
4097 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4098 return 0;
4099 return mkdir(path, 0777);
4102 if (has_symlinks && S_ISLNK(mode))
4103 /* Although buf:size is counted string, it also is NUL
4104 * terminated.
4106 return symlink(buf, path);
4108 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4109 if (fd < 0)
4110 return -1;
4112 if (convert_to_working_tree(path, buf, size, &nbuf)) {
4113 size = nbuf.len;
4114 buf = nbuf.buf;
4116 write_or_die(fd, buf, size);
4117 strbuf_release(&nbuf);
4119 if (close(fd) < 0)
4120 die_errno(_("closing file '%s'"), path);
4121 return 0;
4125 * We optimistically assume that the directories exist,
4126 * which is true 99% of the time anyway. If they don't,
4127 * we create them and try again.
4129 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
4131 if (cached)
4132 return;
4133 if (!try_create_file(path, mode, buf, size))
4134 return;
4136 if (errno == ENOENT) {
4137 if (safe_create_leading_directories(path))
4138 return;
4139 if (!try_create_file(path, mode, buf, size))
4140 return;
4143 if (errno == EEXIST || errno == EACCES) {
4144 /* We may be trying to create a file where a directory
4145 * used to be.
4147 struct stat st;
4148 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4149 errno = EEXIST;
4152 if (errno == EEXIST) {
4153 unsigned int nr = getpid();
4155 for (;;) {
4156 char newpath[PATH_MAX];
4157 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4158 if (!try_create_file(newpath, mode, buf, size)) {
4159 if (!rename(newpath, path))
4160 return;
4161 unlink_or_warn(newpath);
4162 break;
4164 if (errno != EEXIST)
4165 break;
4166 ++nr;
4169 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4172 static void add_conflicted_stages_file(struct patch *patch)
4174 int stage, namelen;
4175 unsigned ce_size, mode;
4176 struct cache_entry *ce;
4178 if (!update_index)
4179 return;
4180 namelen = strlen(patch->new_name);
4181 ce_size = cache_entry_size(namelen);
4182 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4184 remove_file_from_cache(patch->new_name);
4185 for (stage = 1; stage < 4; stage++) {
4186 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4187 continue;
4188 ce = xcalloc(1, ce_size);
4189 memcpy(ce->name, patch->new_name, namelen);
4190 ce->ce_mode = create_ce_mode(mode);
4191 ce->ce_flags = create_ce_flags(stage);
4192 ce->ce_namelen = namelen;
4193 hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);
4194 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4195 die(_("unable to add cache entry for %s"), patch->new_name);
4199 static void create_file(struct patch *patch)
4201 char *path = patch->new_name;
4202 unsigned mode = patch->new_mode;
4203 unsigned long size = patch->resultsize;
4204 char *buf = patch->result;
4206 if (!mode)
4207 mode = S_IFREG | 0644;
4208 create_one_file(path, mode, buf, size);
4210 if (patch->conflicted_threeway)
4211 add_conflicted_stages_file(patch);
4212 else
4213 add_index_file(path, mode, buf, size);
4216 /* phase zero is to remove, phase one is to create */
4217 static void write_out_one_result(struct patch *patch, int phase)
4219 if (patch->is_delete > 0) {
4220 if (phase == 0)
4221 remove_file(patch, 1);
4222 return;
4224 if (patch->is_new > 0 || patch->is_copy) {
4225 if (phase == 1)
4226 create_file(patch);
4227 return;
4230 * Rename or modification boils down to the same
4231 * thing: remove the old, write the new
4233 if (phase == 0)
4234 remove_file(patch, patch->is_rename);
4235 if (phase == 1)
4236 create_file(patch);
4239 static int write_out_one_reject(struct patch *patch)
4241 FILE *rej;
4242 char namebuf[PATH_MAX];
4243 struct fragment *frag;
4244 int cnt = 0;
4245 struct strbuf sb = STRBUF_INIT;
4247 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4248 if (!frag->rejected)
4249 continue;
4250 cnt++;
4253 if (!cnt) {
4254 if (apply_verbosely)
4255 say_patch_name(stderr,
4256 _("Applied patch %s cleanly."), patch);
4257 return 0;
4260 /* This should not happen, because a removal patch that leaves
4261 * contents are marked "rejected" at the patch level.
4263 if (!patch->new_name)
4264 die(_("internal error"));
4266 /* Say this even without --verbose */
4267 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4268 "Applying patch %%s with %d rejects...",
4269 cnt),
4270 cnt);
4271 say_patch_name(stderr, sb.buf, patch);
4272 strbuf_release(&sb);
4274 cnt = strlen(patch->new_name);
4275 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4276 cnt = ARRAY_SIZE(namebuf) - 5;
4277 warning(_("truncating .rej filename to %.*s.rej"),
4278 cnt - 1, patch->new_name);
4280 memcpy(namebuf, patch->new_name, cnt);
4281 memcpy(namebuf + cnt, ".rej", 5);
4283 rej = fopen(namebuf, "w");
4284 if (!rej)
4285 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4287 /* Normal git tools never deal with .rej, so do not pretend
4288 * this is a git patch by saying --git or giving extended
4289 * headers. While at it, maybe please "kompare" that wants
4290 * the trailing TAB and some garbage at the end of line ;-).
4292 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4293 patch->new_name, patch->new_name);
4294 for (cnt = 1, frag = patch->fragments;
4295 frag;
4296 cnt++, frag = frag->next) {
4297 if (!frag->rejected) {
4298 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4299 continue;
4301 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4302 fprintf(rej, "%.*s", frag->size, frag->patch);
4303 if (frag->patch[frag->size-1] != '\n')
4304 fputc('\n', rej);
4306 fclose(rej);
4307 return -1;
4310 static int write_out_results(struct patch *list)
4312 int phase;
4313 int errs = 0;
4314 struct patch *l;
4315 struct string_list cpath = STRING_LIST_INIT_DUP;
4317 for (phase = 0; phase < 2; phase++) {
4318 l = list;
4319 while (l) {
4320 if (l->rejected)
4321 errs = 1;
4322 else {
4323 write_out_one_result(l, phase);
4324 if (phase == 1) {
4325 if (write_out_one_reject(l))
4326 errs = 1;
4327 if (l->conflicted_threeway) {
4328 string_list_append(&cpath, l->new_name);
4329 errs = 1;
4333 l = l->next;
4337 if (cpath.nr) {
4338 struct string_list_item *item;
4340 string_list_sort(&cpath);
4341 for_each_string_list_item(item, &cpath)
4342 fprintf(stderr, "U %s\n", item->string);
4343 string_list_clear(&cpath, 0);
4345 rerere(0);
4348 return errs;
4351 static struct lock_file lock_file;
4353 #define INACCURATE_EOF (1<<0)
4354 #define RECOUNT (1<<1)
4356 static int apply_patch(int fd, const char *filename, int options)
4358 size_t offset;
4359 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4360 struct patch *list = NULL, **listp = &list;
4361 int skipped_patch = 0;
4363 patch_input_file = filename;
4364 read_patch_file(&buf, fd);
4365 offset = 0;
4366 while (offset < buf.len) {
4367 struct patch *patch;
4368 int nr;
4370 patch = xcalloc(1, sizeof(*patch));
4371 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4372 patch->recount = !!(options & RECOUNT);
4373 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4374 if (nr < 0) {
4375 free_patch(patch);
4376 break;
4378 if (apply_in_reverse)
4379 reverse_patches(patch);
4380 if (use_patch(patch)) {
4381 patch_stats(patch);
4382 *listp = patch;
4383 listp = &patch->next;
4385 else {
4386 if (apply_verbosely)
4387 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4388 free_patch(patch);
4389 skipped_patch++;
4391 offset += nr;
4394 if (!list && !skipped_patch)
4395 die(_("unrecognized input"));
4397 if (whitespace_error && (ws_error_action == die_on_ws_error))
4398 apply = 0;
4400 update_index = check_index && apply;
4401 if (update_index && newfd < 0)
4402 newfd = hold_locked_index(&lock_file, 1);
4404 if (check_index) {
4405 if (read_cache() < 0)
4406 die(_("unable to read index file"));
4409 if ((check || apply) &&
4410 check_patch_list(list) < 0 &&
4411 !apply_with_reject)
4412 exit(1);
4414 if (apply && write_out_results(list)) {
4415 if (apply_with_reject)
4416 exit(1);
4417 /* with --3way, we still need to write the index out */
4418 return 1;
4421 if (fake_ancestor)
4422 build_fake_ancestor(list, fake_ancestor);
4424 if (diffstat)
4425 stat_patch_list(list);
4427 if (numstat)
4428 numstat_patch_list(list);
4430 if (summary)
4431 summary_patch_list(list);
4433 free_patch_list(list);
4434 strbuf_release(&buf);
4435 string_list_clear(&fn_table, 0);
4436 return 0;
4439 static void git_apply_config(void)
4441 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4442 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4443 git_config(git_default_config, NULL);
4446 static int option_parse_exclude(const struct option *opt,
4447 const char *arg, int unset)
4449 add_name_limit(arg, 1);
4450 return 0;
4453 static int option_parse_include(const struct option *opt,
4454 const char *arg, int unset)
4456 add_name_limit(arg, 0);
4457 has_include = 1;
4458 return 0;
4461 static int option_parse_p(const struct option *opt,
4462 const char *arg, int unset)
4464 p_value = atoi(arg);
4465 p_value_known = 1;
4466 return 0;
4469 static int option_parse_space_change(const struct option *opt,
4470 const char *arg, int unset)
4472 if (unset)
4473 ws_ignore_action = ignore_ws_none;
4474 else
4475 ws_ignore_action = ignore_ws_change;
4476 return 0;
4479 static int option_parse_whitespace(const struct option *opt,
4480 const char *arg, int unset)
4482 const char **whitespace_option = opt->value;
4484 *whitespace_option = arg;
4485 parse_whitespace_option(arg);
4486 return 0;
4489 static int option_parse_directory(const struct option *opt,
4490 const char *arg, int unset)
4492 strbuf_reset(&root);
4493 strbuf_addstr(&root, arg);
4494 strbuf_complete(&root, '/');
4495 return 0;
4498 int cmd_apply(int argc, const char **argv, const char *prefix_)
4500 int i;
4501 int errs = 0;
4502 int is_not_gitdir = !startup_info->have_repository;
4503 int force_apply = 0;
4505 const char *whitespace_option = NULL;
4507 struct option builtin_apply_options[] = {
4508 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4509 N_("don't apply changes matching the given path"),
4510 0, option_parse_exclude },
4511 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4512 N_("apply changes matching the given path"),
4513 0, option_parse_include },
4514 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4515 N_("remove <num> leading slashes from traditional diff paths"),
4516 0, option_parse_p },
4517 OPT_BOOL(0, "no-add", &no_add,
4518 N_("ignore additions made by the patch")),
4519 OPT_BOOL(0, "stat", &diffstat,
4520 N_("instead of applying the patch, output diffstat for the input")),
4521 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4522 OPT_NOOP_NOARG(0, "binary"),
4523 OPT_BOOL(0, "numstat", &numstat,
4524 N_("show number of added and deleted lines in decimal notation")),
4525 OPT_BOOL(0, "summary", &summary,
4526 N_("instead of applying the patch, output a summary for the input")),
4527 OPT_BOOL(0, "check", &check,
4528 N_("instead of applying the patch, see if the patch is applicable")),
4529 OPT_BOOL(0, "index", &check_index,
4530 N_("make sure the patch is applicable to the current index")),
4531 OPT_BOOL(0, "cached", &cached,
4532 N_("apply a patch without touching the working tree")),
4533 OPT_BOOL(0, "unsafe-paths", &unsafe_paths,
4534 N_("accept a patch that touches outside the working area")),
4535 OPT_BOOL(0, "apply", &force_apply,
4536 N_("also apply the patch (use with --stat/--summary/--check)")),
4537 OPT_BOOL('3', "3way", &threeway,
4538 N_( "attempt three-way merge if a patch does not apply")),
4539 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4540 N_("build a temporary index based on embedded index information")),
4541 /* Think twice before adding "--nul" synonym to this */
4542 OPT_SET_INT('z', NULL, &line_termination,
4543 N_("paths are separated with NUL character"), '\0'),
4544 OPT_INTEGER('C', NULL, &p_context,
4545 N_("ensure at least <n> lines of context match")),
4546 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4547 N_("detect new or modified lines that have whitespace errors"),
4548 0, option_parse_whitespace },
4549 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4550 N_("ignore changes in whitespace when finding context"),
4551 PARSE_OPT_NOARG, option_parse_space_change },
4552 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4553 N_("ignore changes in whitespace when finding context"),
4554 PARSE_OPT_NOARG, option_parse_space_change },
4555 OPT_BOOL('R', "reverse", &apply_in_reverse,
4556 N_("apply the patch in reverse")),
4557 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4558 N_("don't expect at least one line of context")),
4559 OPT_BOOL(0, "reject", &apply_with_reject,
4560 N_("leave the rejected hunks in corresponding *.rej files")),
4561 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4562 N_("allow overlapping hunks")),
4563 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4564 OPT_BIT(0, "inaccurate-eof", &options,
4565 N_("tolerate incorrectly detected missing new-line at the end of file"),
4566 INACCURATE_EOF),
4567 OPT_BIT(0, "recount", &options,
4568 N_("do not trust the line counts in the hunk headers"),
4569 RECOUNT),
4570 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4571 N_("prepend <root> to all filenames"),
4572 0, option_parse_directory },
4573 OPT_END()
4576 prefix = prefix_;
4577 prefix_length = prefix ? strlen(prefix) : 0;
4578 git_apply_config();
4579 if (apply_default_whitespace)
4580 parse_whitespace_option(apply_default_whitespace);
4581 if (apply_default_ignorewhitespace)
4582 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4584 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4585 apply_usage, 0);
4587 if (apply_with_reject && threeway)
4588 die("--reject and --3way cannot be used together.");
4589 if (cached && threeway)
4590 die("--cached and --3way cannot be used together.");
4591 if (threeway) {
4592 if (is_not_gitdir)
4593 die(_("--3way outside a repository"));
4594 check_index = 1;
4596 if (apply_with_reject)
4597 apply = apply_verbosely = 1;
4598 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4599 apply = 0;
4600 if (check_index && is_not_gitdir)
4601 die(_("--index outside a repository"));
4602 if (cached) {
4603 if (is_not_gitdir)
4604 die(_("--cached outside a repository"));
4605 check_index = 1;
4607 if (check_index)
4608 unsafe_paths = 0;
4610 for (i = 0; i < argc; i++) {
4611 const char *arg = argv[i];
4612 int fd;
4614 if (!strcmp(arg, "-")) {
4615 errs |= apply_patch(0, "<stdin>", options);
4616 read_stdin = 0;
4617 continue;
4618 } else if (0 < prefix_length)
4619 arg = prefix_filename(prefix, prefix_length, arg);
4621 fd = open(arg, O_RDONLY);
4622 if (fd < 0)
4623 die_errno(_("can't open patch '%s'"), arg);
4624 read_stdin = 0;
4625 set_default_whitespace_mode(whitespace_option);
4626 errs |= apply_patch(fd, arg, options);
4627 close(fd);
4629 set_default_whitespace_mode(whitespace_option);
4630 if (read_stdin)
4631 errs |= apply_patch(0, "<stdin>", options);
4632 if (whitespace_error) {
4633 if (squelch_whitespace_errors &&
4634 squelch_whitespace_errors < whitespace_error) {
4635 int squelched =
4636 whitespace_error - squelch_whitespace_errors;
4637 warning(Q_("squelched %d whitespace error",
4638 "squelched %d whitespace errors",
4639 squelched),
4640 squelched);
4642 if (ws_error_action == die_on_ws_error)
4643 die(Q_("%d line adds whitespace errors.",
4644 "%d lines add whitespace errors.",
4645 whitespace_error),
4646 whitespace_error);
4647 if (applied_after_fixing_ws && apply)
4648 warning("%d line%s applied after"
4649 " fixing whitespace errors.",
4650 applied_after_fixing_ws,
4651 applied_after_fixing_ws == 1 ? "" : "s");
4652 else if (whitespace_error)
4653 warning(Q_("%d line adds whitespace errors.",
4654 "%d lines add whitespace errors.",
4655 whitespace_error),
4656 whitespace_error);
4659 if (update_index) {
4660 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4661 die(_("Unable to write new index file"));
4664 return !!errs;