mingw: do not bother to test funny file names
[git/raj.git] / builtin / apply.c
blobdeb1364fa81452d236a224109a99b28a034f9a9a
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;
935 const char *name;
936 char *another;
937 name = orig_name;
938 len = strlen(name);
939 if (isnull)
940 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
941 another = find_name(line, NULL, p_value, TERM_TAB);
942 if (!another || memcmp(another, name, len + 1))
943 die((side == DIFF_NEW_NAME) ?
944 _("git apply: bad git-diff - inconsistent new filename on line %d") :
945 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
946 free(another);
947 return orig_name;
949 else {
950 /* expect "/dev/null" */
951 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
952 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
953 return NULL;
957 static int gitdiff_oldname(const char *line, struct patch *patch)
959 char *orig = patch->old_name;
960 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
961 DIFF_OLD_NAME);
962 if (orig != patch->old_name)
963 free(orig);
964 return 0;
967 static int gitdiff_newname(const char *line, struct patch *patch)
969 char *orig = patch->new_name;
970 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
971 DIFF_NEW_NAME);
972 if (orig != patch->new_name)
973 free(orig);
974 return 0;
977 static int gitdiff_oldmode(const char *line, struct patch *patch)
979 patch->old_mode = strtoul(line, NULL, 8);
980 return 0;
983 static int gitdiff_newmode(const char *line, struct patch *patch)
985 patch->new_mode = strtoul(line, NULL, 8);
986 return 0;
989 static int gitdiff_delete(const char *line, struct patch *patch)
991 patch->is_delete = 1;
992 free(patch->old_name);
993 patch->old_name = xstrdup_or_null(patch->def_name);
994 return gitdiff_oldmode(line, patch);
997 static int gitdiff_newfile(const char *line, struct patch *patch)
999 patch->is_new = 1;
1000 free(patch->new_name);
1001 patch->new_name = xstrdup_or_null(patch->def_name);
1002 return gitdiff_newmode(line, patch);
1005 static int gitdiff_copysrc(const char *line, struct patch *patch)
1007 patch->is_copy = 1;
1008 free(patch->old_name);
1009 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1010 return 0;
1013 static int gitdiff_copydst(const char *line, struct patch *patch)
1015 patch->is_copy = 1;
1016 free(patch->new_name);
1017 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1018 return 0;
1021 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1023 patch->is_rename = 1;
1024 free(patch->old_name);
1025 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1026 return 0;
1029 static int gitdiff_renamedst(const char *line, struct patch *patch)
1031 patch->is_rename = 1;
1032 free(patch->new_name);
1033 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1034 return 0;
1037 static int gitdiff_similarity(const char *line, struct patch *patch)
1039 unsigned long val = strtoul(line, NULL, 10);
1040 if (val <= 100)
1041 patch->score = val;
1042 return 0;
1045 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1047 unsigned long val = strtoul(line, NULL, 10);
1048 if (val <= 100)
1049 patch->score = val;
1050 return 0;
1053 static int gitdiff_index(const char *line, struct patch *patch)
1056 * index line is N hexadecimal, "..", N hexadecimal,
1057 * and optional space with octal mode.
1059 const char *ptr, *eol;
1060 int len;
1062 ptr = strchr(line, '.');
1063 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1064 return 0;
1065 len = ptr - line;
1066 memcpy(patch->old_sha1_prefix, line, len);
1067 patch->old_sha1_prefix[len] = 0;
1069 line = ptr + 2;
1070 ptr = strchr(line, ' ');
1071 eol = strchrnul(line, '\n');
1073 if (!ptr || eol < ptr)
1074 ptr = eol;
1075 len = ptr - line;
1077 if (40 < len)
1078 return 0;
1079 memcpy(patch->new_sha1_prefix, line, len);
1080 patch->new_sha1_prefix[len] = 0;
1081 if (*ptr == ' ')
1082 patch->old_mode = strtoul(ptr+1, NULL, 8);
1083 return 0;
1087 * This is normal for a diff that doesn't change anything: we'll fall through
1088 * into the next diff. Tell the parser to break out.
1090 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1092 return -1;
1096 * Skip p_value leading components from "line"; as we do not accept
1097 * absolute paths, return NULL in that case.
1099 static const char *skip_tree_prefix(const char *line, int llen)
1101 int nslash;
1102 int i;
1104 if (!p_value)
1105 return (llen && line[0] == '/') ? NULL : line;
1107 nslash = p_value;
1108 for (i = 0; i < llen; i++) {
1109 int ch = line[i];
1110 if (ch == '/' && --nslash <= 0)
1111 return (i == 0) ? NULL : &line[i + 1];
1113 return NULL;
1117 * This is to extract the same name that appears on "diff --git"
1118 * line. We do not find and return anything if it is a rename
1119 * patch, and it is OK because we will find the name elsewhere.
1120 * We need to reliably find name only when it is mode-change only,
1121 * creation or deletion of an empty file. In any of these cases,
1122 * both sides are the same name under a/ and b/ respectively.
1124 static char *git_header_name(const char *line, int llen)
1126 const char *name;
1127 const char *second = NULL;
1128 size_t len, line_len;
1130 line += strlen("diff --git ");
1131 llen -= strlen("diff --git ");
1133 if (*line == '"') {
1134 const char *cp;
1135 struct strbuf first = STRBUF_INIT;
1136 struct strbuf sp = STRBUF_INIT;
1138 if (unquote_c_style(&first, line, &second))
1139 goto free_and_fail1;
1141 /* strip the a/b prefix including trailing slash */
1142 cp = skip_tree_prefix(first.buf, first.len);
1143 if (!cp)
1144 goto free_and_fail1;
1145 strbuf_remove(&first, 0, cp - first.buf);
1148 * second points at one past closing dq of name.
1149 * find the second name.
1151 while ((second < line + llen) && isspace(*second))
1152 second++;
1154 if (line + llen <= second)
1155 goto free_and_fail1;
1156 if (*second == '"') {
1157 if (unquote_c_style(&sp, second, NULL))
1158 goto free_and_fail1;
1159 cp = skip_tree_prefix(sp.buf, sp.len);
1160 if (!cp)
1161 goto free_and_fail1;
1162 /* They must match, otherwise ignore */
1163 if (strcmp(cp, first.buf))
1164 goto free_and_fail1;
1165 strbuf_release(&sp);
1166 return strbuf_detach(&first, NULL);
1169 /* unquoted second */
1170 cp = skip_tree_prefix(second, line + llen - second);
1171 if (!cp)
1172 goto free_and_fail1;
1173 if (line + llen - cp != first.len ||
1174 memcmp(first.buf, cp, first.len))
1175 goto free_and_fail1;
1176 return strbuf_detach(&first, NULL);
1178 free_and_fail1:
1179 strbuf_release(&first);
1180 strbuf_release(&sp);
1181 return NULL;
1184 /* unquoted first name */
1185 name = skip_tree_prefix(line, llen);
1186 if (!name)
1187 return NULL;
1190 * since the first name is unquoted, a dq if exists must be
1191 * the beginning of the second name.
1193 for (second = name; second < line + llen; second++) {
1194 if (*second == '"') {
1195 struct strbuf sp = STRBUF_INIT;
1196 const char *np;
1198 if (unquote_c_style(&sp, second, NULL))
1199 goto free_and_fail2;
1201 np = skip_tree_prefix(sp.buf, sp.len);
1202 if (!np)
1203 goto free_and_fail2;
1205 len = sp.buf + sp.len - np;
1206 if (len < second - name &&
1207 !strncmp(np, name, len) &&
1208 isspace(name[len])) {
1209 /* Good */
1210 strbuf_remove(&sp, 0, np - sp.buf);
1211 return strbuf_detach(&sp, NULL);
1214 free_and_fail2:
1215 strbuf_release(&sp);
1216 return NULL;
1221 * Accept a name only if it shows up twice, exactly the same
1222 * form.
1224 second = strchr(name, '\n');
1225 if (!second)
1226 return NULL;
1227 line_len = second - name;
1228 for (len = 0 ; ; len++) {
1229 switch (name[len]) {
1230 default:
1231 continue;
1232 case '\n':
1233 return NULL;
1234 case '\t': case ' ':
1236 * Is this the separator between the preimage
1237 * and the postimage pathname? Again, we are
1238 * only interested in the case where there is
1239 * no rename, as this is only to set def_name
1240 * and a rename patch has the names elsewhere
1241 * in an unambiguous form.
1243 if (!name[len + 1])
1244 return NULL; /* no postimage name */
1245 second = skip_tree_prefix(name + len + 1,
1246 line_len - (len + 1));
1247 if (!second)
1248 return NULL;
1250 * Does len bytes starting at "name" and "second"
1251 * (that are separated by one HT or SP we just
1252 * found) exactly match?
1254 if (second[len] == '\n' && !strncmp(name, second, len))
1255 return xmemdupz(name, len);
1260 /* Verify that we recognize the lines following a git header */
1261 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1263 unsigned long offset;
1265 /* A git diff has explicit new/delete information, so we don't guess */
1266 patch->is_new = 0;
1267 patch->is_delete = 0;
1270 * Some things may not have the old name in the
1271 * rest of the headers anywhere (pure mode changes,
1272 * or removing or adding empty files), so we get
1273 * the default name from the header.
1275 patch->def_name = git_header_name(line, len);
1276 if (patch->def_name && root.len) {
1277 char *s = xstrfmt("%s%s", root.buf, patch->def_name);
1278 free(patch->def_name);
1279 patch->def_name = s;
1282 line += len;
1283 size -= len;
1284 linenr++;
1285 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1286 static const struct opentry {
1287 const char *str;
1288 int (*fn)(const char *, struct patch *);
1289 } optable[] = {
1290 { "@@ -", gitdiff_hdrend },
1291 { "--- ", gitdiff_oldname },
1292 { "+++ ", gitdiff_newname },
1293 { "old mode ", gitdiff_oldmode },
1294 { "new mode ", gitdiff_newmode },
1295 { "deleted file mode ", gitdiff_delete },
1296 { "new file mode ", gitdiff_newfile },
1297 { "copy from ", gitdiff_copysrc },
1298 { "copy to ", gitdiff_copydst },
1299 { "rename old ", gitdiff_renamesrc },
1300 { "rename new ", gitdiff_renamedst },
1301 { "rename from ", gitdiff_renamesrc },
1302 { "rename to ", gitdiff_renamedst },
1303 { "similarity index ", gitdiff_similarity },
1304 { "dissimilarity index ", gitdiff_dissimilarity },
1305 { "index ", gitdiff_index },
1306 { "", gitdiff_unrecognized },
1308 int i;
1310 len = linelen(line, size);
1311 if (!len || line[len-1] != '\n')
1312 break;
1313 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1314 const struct opentry *p = optable + i;
1315 int oplen = strlen(p->str);
1316 if (len < oplen || memcmp(p->str, line, oplen))
1317 continue;
1318 if (p->fn(line + oplen, patch) < 0)
1319 return offset;
1320 break;
1324 return offset;
1327 static int parse_num(const char *line, unsigned long *p)
1329 char *ptr;
1331 if (!isdigit(*line))
1332 return 0;
1333 *p = strtoul(line, &ptr, 10);
1334 return ptr - line;
1337 static int parse_range(const char *line, int len, int offset, const char *expect,
1338 unsigned long *p1, unsigned long *p2)
1340 int digits, ex;
1342 if (offset < 0 || offset >= len)
1343 return -1;
1344 line += offset;
1345 len -= offset;
1347 digits = parse_num(line, p1);
1348 if (!digits)
1349 return -1;
1351 offset += digits;
1352 line += digits;
1353 len -= digits;
1355 *p2 = 1;
1356 if (*line == ',') {
1357 digits = parse_num(line+1, p2);
1358 if (!digits)
1359 return -1;
1361 offset += digits+1;
1362 line += digits+1;
1363 len -= digits+1;
1366 ex = strlen(expect);
1367 if (ex > len)
1368 return -1;
1369 if (memcmp(line, expect, ex))
1370 return -1;
1372 return offset + ex;
1375 static void recount_diff(const char *line, int size, struct fragment *fragment)
1377 int oldlines = 0, newlines = 0, ret = 0;
1379 if (size < 1) {
1380 warning("recount: ignore empty hunk");
1381 return;
1384 for (;;) {
1385 int len = linelen(line, size);
1386 size -= len;
1387 line += len;
1389 if (size < 1)
1390 break;
1392 switch (*line) {
1393 case ' ': case '\n':
1394 newlines++;
1395 /* fall through */
1396 case '-':
1397 oldlines++;
1398 continue;
1399 case '+':
1400 newlines++;
1401 continue;
1402 case '\\':
1403 continue;
1404 case '@':
1405 ret = size < 3 || !starts_with(line, "@@ ");
1406 break;
1407 case 'd':
1408 ret = size < 5 || !starts_with(line, "diff ");
1409 break;
1410 default:
1411 ret = -1;
1412 break;
1414 if (ret) {
1415 warning(_("recount: unexpected line: %.*s"),
1416 (int)linelen(line, size), line);
1417 return;
1419 break;
1421 fragment->oldlines = oldlines;
1422 fragment->newlines = newlines;
1426 * Parse a unified diff fragment header of the
1427 * form "@@ -a,b +c,d @@"
1429 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1431 int offset;
1433 if (!len || line[len-1] != '\n')
1434 return -1;
1436 /* Figure out the number of lines in a fragment */
1437 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1438 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1440 return offset;
1443 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1445 unsigned long offset, len;
1447 patch->is_toplevel_relative = 0;
1448 patch->is_rename = patch->is_copy = 0;
1449 patch->is_new = patch->is_delete = -1;
1450 patch->old_mode = patch->new_mode = 0;
1451 patch->old_name = patch->new_name = NULL;
1452 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1453 unsigned long nextlen;
1455 len = linelen(line, size);
1456 if (!len)
1457 break;
1459 /* Testing this early allows us to take a few shortcuts.. */
1460 if (len < 6)
1461 continue;
1464 * Make sure we don't find any unconnected patch fragments.
1465 * That's a sign that we didn't find a header, and that a
1466 * patch has become corrupted/broken up.
1468 if (!memcmp("@@ -", line, 4)) {
1469 struct fragment dummy;
1470 if (parse_fragment_header(line, len, &dummy) < 0)
1471 continue;
1472 die(_("patch fragment without header at line %d: %.*s"),
1473 linenr, (int)len-1, line);
1476 if (size < len + 6)
1477 break;
1480 * Git patch? It might not have a real patch, just a rename
1481 * or mode change, so we handle that specially
1483 if (!memcmp("diff --git ", line, 11)) {
1484 int git_hdr_len = parse_git_header(line, len, size, patch);
1485 if (git_hdr_len <= len)
1486 continue;
1487 if (!patch->old_name && !patch->new_name) {
1488 if (!patch->def_name)
1489 die(Q_("git diff header lacks filename information when removing "
1490 "%d leading pathname component (line %d)",
1491 "git diff header lacks filename information when removing "
1492 "%d leading pathname components (line %d)",
1493 p_value),
1494 p_value, linenr);
1495 patch->old_name = xstrdup(patch->def_name);
1496 patch->new_name = xstrdup(patch->def_name);
1498 if (!patch->is_delete && !patch->new_name)
1499 die("git diff header lacks filename information "
1500 "(line %d)", linenr);
1501 patch->is_toplevel_relative = 1;
1502 *hdrsize = git_hdr_len;
1503 return offset;
1506 /* --- followed by +++ ? */
1507 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1508 continue;
1511 * We only accept unified patches, so we want it to
1512 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1513 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1515 nextlen = linelen(line + len, size - len);
1516 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1517 continue;
1519 /* Ok, we'll consider it a patch */
1520 parse_traditional_patch(line, line+len, patch);
1521 *hdrsize = len + nextlen;
1522 linenr += 2;
1523 return offset;
1525 return -1;
1528 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1530 char *err;
1532 if (!result)
1533 return;
1535 whitespace_error++;
1536 if (squelch_whitespace_errors &&
1537 squelch_whitespace_errors < whitespace_error)
1538 return;
1540 err = whitespace_error_string(result);
1541 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1542 patch_input_file, linenr, err, len, line);
1543 free(err);
1546 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1548 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1550 record_ws_error(result, line + 1, len - 2, linenr);
1554 * Parse a unified diff. Note that this really needs to parse each
1555 * fragment separately, since the only way to know the difference
1556 * between a "---" that is part of a patch, and a "---" that starts
1557 * the next patch is to look at the line counts..
1559 static int parse_fragment(const char *line, unsigned long size,
1560 struct patch *patch, struct fragment *fragment)
1562 int added, deleted;
1563 int len = linelen(line, size), offset;
1564 unsigned long oldlines, newlines;
1565 unsigned long leading, trailing;
1567 offset = parse_fragment_header(line, len, fragment);
1568 if (offset < 0)
1569 return -1;
1570 if (offset > 0 && patch->recount)
1571 recount_diff(line + offset, size - offset, fragment);
1572 oldlines = fragment->oldlines;
1573 newlines = fragment->newlines;
1574 leading = 0;
1575 trailing = 0;
1577 /* Parse the thing.. */
1578 line += len;
1579 size -= len;
1580 linenr++;
1581 added = deleted = 0;
1582 for (offset = len;
1583 0 < size;
1584 offset += len, size -= len, line += len, linenr++) {
1585 if (!oldlines && !newlines)
1586 break;
1587 len = linelen(line, size);
1588 if (!len || line[len-1] != '\n')
1589 return -1;
1590 switch (*line) {
1591 default:
1592 return -1;
1593 case '\n': /* newer GNU diff, an empty context line */
1594 case ' ':
1595 oldlines--;
1596 newlines--;
1597 if (!deleted && !added)
1598 leading++;
1599 trailing++;
1600 if (!apply_in_reverse &&
1601 ws_error_action == correct_ws_error)
1602 check_whitespace(line, len, patch->ws_rule);
1603 break;
1604 case '-':
1605 if (apply_in_reverse &&
1606 ws_error_action != nowarn_ws_error)
1607 check_whitespace(line, len, patch->ws_rule);
1608 deleted++;
1609 oldlines--;
1610 trailing = 0;
1611 break;
1612 case '+':
1613 if (!apply_in_reverse &&
1614 ws_error_action != nowarn_ws_error)
1615 check_whitespace(line, len, patch->ws_rule);
1616 added++;
1617 newlines--;
1618 trailing = 0;
1619 break;
1622 * We allow "\ No newline at end of file". Depending
1623 * on locale settings when the patch was produced we
1624 * don't know what this line looks like. The only
1625 * thing we do know is that it begins with "\ ".
1626 * Checking for 12 is just for sanity check -- any
1627 * l10n of "\ No newline..." is at least that long.
1629 case '\\':
1630 if (len < 12 || memcmp(line, "\\ ", 2))
1631 return -1;
1632 break;
1635 if (oldlines || newlines)
1636 return -1;
1637 if (!deleted && !added)
1638 return -1;
1640 fragment->leading = leading;
1641 fragment->trailing = trailing;
1644 * If a fragment ends with an incomplete line, we failed to include
1645 * it in the above loop because we hit oldlines == newlines == 0
1646 * before seeing it.
1648 if (12 < size && !memcmp(line, "\\ ", 2))
1649 offset += linelen(line, size);
1651 patch->lines_added += added;
1652 patch->lines_deleted += deleted;
1654 if (0 < patch->is_new && oldlines)
1655 return error(_("new file depends on old contents"));
1656 if (0 < patch->is_delete && newlines)
1657 return error(_("deleted file still has contents"));
1658 return offset;
1662 * We have seen "diff --git a/... b/..." header (or a traditional patch
1663 * header). Read hunks that belong to this patch into fragments and hang
1664 * them to the given patch structure.
1666 * The (fragment->patch, fragment->size) pair points into the memory given
1667 * by the caller, not a copy, when we return.
1669 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1671 unsigned long offset = 0;
1672 unsigned long oldlines = 0, newlines = 0, context = 0;
1673 struct fragment **fragp = &patch->fragments;
1675 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1676 struct fragment *fragment;
1677 int len;
1679 fragment = xcalloc(1, sizeof(*fragment));
1680 fragment->linenr = linenr;
1681 len = parse_fragment(line, size, patch, fragment);
1682 if (len <= 0)
1683 die(_("corrupt patch at line %d"), linenr);
1684 fragment->patch = line;
1685 fragment->size = len;
1686 oldlines += fragment->oldlines;
1687 newlines += fragment->newlines;
1688 context += fragment->leading + fragment->trailing;
1690 *fragp = fragment;
1691 fragp = &fragment->next;
1693 offset += len;
1694 line += len;
1695 size -= len;
1699 * If something was removed (i.e. we have old-lines) it cannot
1700 * be creation, and if something was added it cannot be
1701 * deletion. However, the reverse is not true; --unified=0
1702 * patches that only add are not necessarily creation even
1703 * though they do not have any old lines, and ones that only
1704 * delete are not necessarily deletion.
1706 * Unfortunately, a real creation/deletion patch do _not_ have
1707 * any context line by definition, so we cannot safely tell it
1708 * apart with --unified=0 insanity. At least if the patch has
1709 * more than one hunk it is not creation or deletion.
1711 if (patch->is_new < 0 &&
1712 (oldlines || (patch->fragments && patch->fragments->next)))
1713 patch->is_new = 0;
1714 if (patch->is_delete < 0 &&
1715 (newlines || (patch->fragments && patch->fragments->next)))
1716 patch->is_delete = 0;
1718 if (0 < patch->is_new && oldlines)
1719 die(_("new file %s depends on old contents"), patch->new_name);
1720 if (0 < patch->is_delete && newlines)
1721 die(_("deleted file %s still has contents"), patch->old_name);
1722 if (!patch->is_delete && !newlines && context)
1723 fprintf_ln(stderr,
1724 _("** warning: "
1725 "file %s becomes empty but is not deleted"),
1726 patch->new_name);
1728 return offset;
1731 static inline int metadata_changes(struct patch *patch)
1733 return patch->is_rename > 0 ||
1734 patch->is_copy > 0 ||
1735 patch->is_new > 0 ||
1736 patch->is_delete ||
1737 (patch->old_mode && patch->new_mode &&
1738 patch->old_mode != patch->new_mode);
1741 static char *inflate_it(const void *data, unsigned long size,
1742 unsigned long inflated_size)
1744 git_zstream stream;
1745 void *out;
1746 int st;
1748 memset(&stream, 0, sizeof(stream));
1750 stream.next_in = (unsigned char *)data;
1751 stream.avail_in = size;
1752 stream.next_out = out = xmalloc(inflated_size);
1753 stream.avail_out = inflated_size;
1754 git_inflate_init(&stream);
1755 st = git_inflate(&stream, Z_FINISH);
1756 git_inflate_end(&stream);
1757 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1758 free(out);
1759 return NULL;
1761 return out;
1765 * Read a binary hunk and return a new fragment; fragment->patch
1766 * points at an allocated memory that the caller must free, so
1767 * it is marked as "->free_patch = 1".
1769 static struct fragment *parse_binary_hunk(char **buf_p,
1770 unsigned long *sz_p,
1771 int *status_p,
1772 int *used_p)
1775 * Expect a line that begins with binary patch method ("literal"
1776 * or "delta"), followed by the length of data before deflating.
1777 * a sequence of 'length-byte' followed by base-85 encoded data
1778 * should follow, terminated by a newline.
1780 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1781 * and we would limit the patch line to 66 characters,
1782 * so one line can fit up to 13 groups that would decode
1783 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1784 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1786 int llen, used;
1787 unsigned long size = *sz_p;
1788 char *buffer = *buf_p;
1789 int patch_method;
1790 unsigned long origlen;
1791 char *data = NULL;
1792 int hunk_size = 0;
1793 struct fragment *frag;
1795 llen = linelen(buffer, size);
1796 used = llen;
1798 *status_p = 0;
1800 if (starts_with(buffer, "delta ")) {
1801 patch_method = BINARY_DELTA_DEFLATED;
1802 origlen = strtoul(buffer + 6, NULL, 10);
1804 else if (starts_with(buffer, "literal ")) {
1805 patch_method = BINARY_LITERAL_DEFLATED;
1806 origlen = strtoul(buffer + 8, NULL, 10);
1808 else
1809 return NULL;
1811 linenr++;
1812 buffer += llen;
1813 while (1) {
1814 int byte_length, max_byte_length, newsize;
1815 llen = linelen(buffer, size);
1816 used += llen;
1817 linenr++;
1818 if (llen == 1) {
1819 /* consume the blank line */
1820 buffer++;
1821 size--;
1822 break;
1825 * Minimum line is "A00000\n" which is 7-byte long,
1826 * and the line length must be multiple of 5 plus 2.
1828 if ((llen < 7) || (llen-2) % 5)
1829 goto corrupt;
1830 max_byte_length = (llen - 2) / 5 * 4;
1831 byte_length = *buffer;
1832 if ('A' <= byte_length && byte_length <= 'Z')
1833 byte_length = byte_length - 'A' + 1;
1834 else if ('a' <= byte_length && byte_length <= 'z')
1835 byte_length = byte_length - 'a' + 27;
1836 else
1837 goto corrupt;
1838 /* if the input length was not multiple of 4, we would
1839 * have filler at the end but the filler should never
1840 * exceed 3 bytes
1842 if (max_byte_length < byte_length ||
1843 byte_length <= max_byte_length - 4)
1844 goto corrupt;
1845 newsize = hunk_size + byte_length;
1846 data = xrealloc(data, newsize);
1847 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1848 goto corrupt;
1849 hunk_size = newsize;
1850 buffer += llen;
1851 size -= llen;
1854 frag = xcalloc(1, sizeof(*frag));
1855 frag->patch = inflate_it(data, hunk_size, origlen);
1856 frag->free_patch = 1;
1857 if (!frag->patch)
1858 goto corrupt;
1859 free(data);
1860 frag->size = origlen;
1861 *buf_p = buffer;
1862 *sz_p = size;
1863 *used_p = used;
1864 frag->binary_patch_method = patch_method;
1865 return frag;
1867 corrupt:
1868 free(data);
1869 *status_p = -1;
1870 error(_("corrupt binary patch at line %d: %.*s"),
1871 linenr-1, llen-1, buffer);
1872 return NULL;
1875 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1878 * We have read "GIT binary patch\n"; what follows is a line
1879 * that says the patch method (currently, either "literal" or
1880 * "delta") and the length of data before deflating; a
1881 * sequence of 'length-byte' followed by base-85 encoded data
1882 * follows.
1884 * When a binary patch is reversible, there is another binary
1885 * hunk in the same format, starting with patch method (either
1886 * "literal" or "delta") with the length of data, and a sequence
1887 * of length-byte + base-85 encoded data, terminated with another
1888 * empty line. This data, when applied to the postimage, produces
1889 * the preimage.
1891 struct fragment *forward;
1892 struct fragment *reverse;
1893 int status;
1894 int used, used_1;
1896 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1897 if (!forward && !status)
1898 /* there has to be one hunk (forward hunk) */
1899 return error(_("unrecognized binary patch at line %d"), linenr-1);
1900 if (status)
1901 /* otherwise we already gave an error message */
1902 return status;
1904 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1905 if (reverse)
1906 used += used_1;
1907 else if (status) {
1909 * Not having reverse hunk is not an error, but having
1910 * a corrupt reverse hunk is.
1912 free((void*) forward->patch);
1913 free(forward);
1914 return status;
1916 forward->next = reverse;
1917 patch->fragments = forward;
1918 patch->is_binary = 1;
1919 return used;
1922 static void prefix_one(char **name)
1924 char *old_name = *name;
1925 if (!old_name)
1926 return;
1927 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
1928 free(old_name);
1931 static void prefix_patch(struct patch *p)
1933 if (!prefix || p->is_toplevel_relative)
1934 return;
1935 prefix_one(&p->new_name);
1936 prefix_one(&p->old_name);
1940 * include/exclude
1943 static struct string_list limit_by_name;
1944 static int has_include;
1945 static void add_name_limit(const char *name, int exclude)
1947 struct string_list_item *it;
1949 it = string_list_append(&limit_by_name, name);
1950 it->util = exclude ? NULL : (void *) 1;
1953 static int use_patch(struct patch *p)
1955 const char *pathname = p->new_name ? p->new_name : p->old_name;
1956 int i;
1958 /* Paths outside are not touched regardless of "--include" */
1959 if (0 < prefix_length) {
1960 int pathlen = strlen(pathname);
1961 if (pathlen <= prefix_length ||
1962 memcmp(prefix, pathname, prefix_length))
1963 return 0;
1966 /* See if it matches any of exclude/include rule */
1967 for (i = 0; i < limit_by_name.nr; i++) {
1968 struct string_list_item *it = &limit_by_name.items[i];
1969 if (!wildmatch(it->string, pathname, 0, NULL))
1970 return (it->util != NULL);
1974 * If we had any include, a path that does not match any rule is
1975 * not used. Otherwise, we saw bunch of exclude rules (or none)
1976 * and such a path is used.
1978 return !has_include;
1983 * Read the patch text in "buffer" that extends for "size" bytes; stop
1984 * reading after seeing a single patch (i.e. changes to a single file).
1985 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1986 * Return the number of bytes consumed, so that the caller can call us
1987 * again for the next patch.
1989 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1991 int hdrsize, patchsize;
1992 int offset = find_header(buffer, size, &hdrsize, patch);
1994 if (offset < 0)
1995 return offset;
1997 prefix_patch(patch);
1999 if (!use_patch(patch))
2000 patch->ws_rule = 0;
2001 else
2002 patch->ws_rule = whitespace_rule(patch->new_name
2003 ? patch->new_name
2004 : patch->old_name);
2006 patchsize = parse_single_patch(buffer + offset + hdrsize,
2007 size - offset - hdrsize, patch);
2009 if (!patchsize) {
2010 static const char git_binary[] = "GIT binary patch\n";
2011 int hd = hdrsize + offset;
2012 unsigned long llen = linelen(buffer + hd, size - hd);
2014 if (llen == sizeof(git_binary) - 1 &&
2015 !memcmp(git_binary, buffer + hd, llen)) {
2016 int used;
2017 linenr++;
2018 used = parse_binary(buffer + hd + llen,
2019 size - hd - llen, patch);
2020 if (used)
2021 patchsize = used + llen;
2022 else
2023 patchsize = 0;
2025 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2026 static const char *binhdr[] = {
2027 "Binary files ",
2028 "Files ",
2029 NULL,
2031 int i;
2032 for (i = 0; binhdr[i]; i++) {
2033 int len = strlen(binhdr[i]);
2034 if (len < size - hd &&
2035 !memcmp(binhdr[i], buffer + hd, len)) {
2036 linenr++;
2037 patch->is_binary = 1;
2038 patchsize = llen;
2039 break;
2044 /* Empty patch cannot be applied if it is a text patch
2045 * without metadata change. A binary patch appears
2046 * empty to us here.
2048 if ((apply || check) &&
2049 (!patch->is_binary && !metadata_changes(patch)))
2050 die(_("patch with only garbage at line %d"), linenr);
2053 return offset + hdrsize + patchsize;
2056 #define swap(a,b) myswap((a),(b),sizeof(a))
2058 #define myswap(a, b, size) do { \
2059 unsigned char mytmp[size]; \
2060 memcpy(mytmp, &a, size); \
2061 memcpy(&a, &b, size); \
2062 memcpy(&b, mytmp, size); \
2063 } while (0)
2065 static void reverse_patches(struct patch *p)
2067 for (; p; p = p->next) {
2068 struct fragment *frag = p->fragments;
2070 swap(p->new_name, p->old_name);
2071 swap(p->new_mode, p->old_mode);
2072 swap(p->is_new, p->is_delete);
2073 swap(p->lines_added, p->lines_deleted);
2074 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2076 for (; frag; frag = frag->next) {
2077 swap(frag->newpos, frag->oldpos);
2078 swap(frag->newlines, frag->oldlines);
2083 static const char pluses[] =
2084 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2085 static const char minuses[]=
2086 "----------------------------------------------------------------------";
2088 static void show_stats(struct patch *patch)
2090 struct strbuf qname = STRBUF_INIT;
2091 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2092 int max, add, del;
2094 quote_c_style(cp, &qname, NULL, 0);
2097 * "scale" the filename
2099 max = max_len;
2100 if (max > 50)
2101 max = 50;
2103 if (qname.len > max) {
2104 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2105 if (!cp)
2106 cp = qname.buf + qname.len + 3 - max;
2107 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2110 if (patch->is_binary) {
2111 printf(" %-*s | Bin\n", max, qname.buf);
2112 strbuf_release(&qname);
2113 return;
2116 printf(" %-*s |", max, qname.buf);
2117 strbuf_release(&qname);
2120 * scale the add/delete
2122 max = max + max_change > 70 ? 70 - max : max_change;
2123 add = patch->lines_added;
2124 del = patch->lines_deleted;
2126 if (max_change > 0) {
2127 int total = ((add + del) * max + max_change / 2) / max_change;
2128 add = (add * max + max_change / 2) / max_change;
2129 del = total - add;
2131 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2132 add, pluses, del, minuses);
2135 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2137 switch (st->st_mode & S_IFMT) {
2138 case S_IFLNK:
2139 if (strbuf_readlink(buf, path, st->st_size) < 0)
2140 return error(_("unable to read symlink %s"), path);
2141 return 0;
2142 case S_IFREG:
2143 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2144 return error(_("unable to open or read %s"), path);
2145 convert_to_git(path, buf->buf, buf->len, buf, 0);
2146 return 0;
2147 default:
2148 return -1;
2153 * Update the preimage, and the common lines in postimage,
2154 * from buffer buf of length len. If postlen is 0 the postimage
2155 * is updated in place, otherwise it's updated on a new buffer
2156 * of length postlen
2159 static void update_pre_post_images(struct image *preimage,
2160 struct image *postimage,
2161 char *buf,
2162 size_t len, size_t postlen)
2164 int i, ctx, reduced;
2165 char *new, *old, *fixed;
2166 struct image fixed_preimage;
2169 * Update the preimage with whitespace fixes. Note that we
2170 * are not losing preimage->buf -- apply_one_fragment() will
2171 * free "oldlines".
2173 prepare_image(&fixed_preimage, buf, len, 1);
2174 assert(postlen
2175 ? fixed_preimage.nr == preimage->nr
2176 : fixed_preimage.nr <= preimage->nr);
2177 for (i = 0; i < fixed_preimage.nr; i++)
2178 fixed_preimage.line[i].flag = preimage->line[i].flag;
2179 free(preimage->line_allocated);
2180 *preimage = fixed_preimage;
2183 * Adjust the common context lines in postimage. This can be
2184 * done in-place when we are shrinking it with whitespace
2185 * fixing, but needs a new buffer when ignoring whitespace or
2186 * expanding leading tabs to spaces.
2188 * We trust the caller to tell us if the update can be done
2189 * in place (postlen==0) or not.
2191 old = postimage->buf;
2192 if (postlen)
2193 new = postimage->buf = xmalloc(postlen);
2194 else
2195 new = old;
2196 fixed = preimage->buf;
2198 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2199 size_t len = postimage->line[i].len;
2200 if (!(postimage->line[i].flag & LINE_COMMON)) {
2201 /* an added line -- no counterparts in preimage */
2202 memmove(new, old, len);
2203 old += len;
2204 new += len;
2205 continue;
2208 /* a common context -- skip it in the original postimage */
2209 old += len;
2211 /* and find the corresponding one in the fixed preimage */
2212 while (ctx < preimage->nr &&
2213 !(preimage->line[ctx].flag & LINE_COMMON)) {
2214 fixed += preimage->line[ctx].len;
2215 ctx++;
2219 * preimage is expected to run out, if the caller
2220 * fixed addition of trailing blank lines.
2222 if (preimage->nr <= ctx) {
2223 reduced++;
2224 continue;
2227 /* and copy it in, while fixing the line length */
2228 len = preimage->line[ctx].len;
2229 memcpy(new, fixed, len);
2230 new += len;
2231 fixed += len;
2232 postimage->line[i].len = len;
2233 ctx++;
2236 if (postlen
2237 ? postlen < new - postimage->buf
2238 : postimage->len < new - postimage->buf)
2239 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2240 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2242 /* Fix the length of the whole thing */
2243 postimage->len = new - postimage->buf;
2244 postimage->nr -= reduced;
2247 static int match_fragment(struct image *img,
2248 struct image *preimage,
2249 struct image *postimage,
2250 unsigned long try,
2251 int try_lno,
2252 unsigned ws_rule,
2253 int match_beginning, int match_end)
2255 int i;
2256 char *fixed_buf, *buf, *orig, *target;
2257 struct strbuf fixed;
2258 size_t fixed_len, postlen;
2259 int preimage_limit;
2261 if (preimage->nr + try_lno <= img->nr) {
2263 * The hunk falls within the boundaries of img.
2265 preimage_limit = preimage->nr;
2266 if (match_end && (preimage->nr + try_lno != img->nr))
2267 return 0;
2268 } else if (ws_error_action == correct_ws_error &&
2269 (ws_rule & WS_BLANK_AT_EOF)) {
2271 * This hunk extends beyond the end of img, and we are
2272 * removing blank lines at the end of the file. This
2273 * many lines from the beginning of the preimage must
2274 * match with img, and the remainder of the preimage
2275 * must be blank.
2277 preimage_limit = img->nr - try_lno;
2278 } else {
2280 * The hunk extends beyond the end of the img and
2281 * we are not removing blanks at the end, so we
2282 * should reject the hunk at this position.
2284 return 0;
2287 if (match_beginning && try_lno)
2288 return 0;
2290 /* Quick hash check */
2291 for (i = 0; i < preimage_limit; i++)
2292 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2293 (preimage->line[i].hash != img->line[try_lno + i].hash))
2294 return 0;
2296 if (preimage_limit == preimage->nr) {
2298 * Do we have an exact match? If we were told to match
2299 * at the end, size must be exactly at try+fragsize,
2300 * otherwise try+fragsize must be still within the preimage,
2301 * and either case, the old piece should match the preimage
2302 * exactly.
2304 if ((match_end
2305 ? (try + preimage->len == img->len)
2306 : (try + preimage->len <= img->len)) &&
2307 !memcmp(img->buf + try, preimage->buf, preimage->len))
2308 return 1;
2309 } else {
2311 * The preimage extends beyond the end of img, so
2312 * there cannot be an exact match.
2314 * There must be one non-blank context line that match
2315 * a line before the end of img.
2317 char *buf_end;
2319 buf = preimage->buf;
2320 buf_end = buf;
2321 for (i = 0; i < preimage_limit; i++)
2322 buf_end += preimage->line[i].len;
2324 for ( ; buf < buf_end; buf++)
2325 if (!isspace(*buf))
2326 break;
2327 if (buf == buf_end)
2328 return 0;
2332 * No exact match. If we are ignoring whitespace, run a line-by-line
2333 * fuzzy matching. We collect all the line length information because
2334 * we need it to adjust whitespace if we match.
2336 if (ws_ignore_action == ignore_ws_change) {
2337 size_t imgoff = 0;
2338 size_t preoff = 0;
2339 size_t postlen = postimage->len;
2340 size_t extra_chars;
2341 char *preimage_eof;
2342 char *preimage_end;
2343 for (i = 0; i < preimage_limit; i++) {
2344 size_t prelen = preimage->line[i].len;
2345 size_t imglen = img->line[try_lno+i].len;
2347 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2348 preimage->buf + preoff, prelen))
2349 return 0;
2350 if (preimage->line[i].flag & LINE_COMMON)
2351 postlen += imglen - prelen;
2352 imgoff += imglen;
2353 preoff += prelen;
2357 * Ok, the preimage matches with whitespace fuzz.
2359 * imgoff now holds the true length of the target that
2360 * matches the preimage before the end of the file.
2362 * Count the number of characters in the preimage that fall
2363 * beyond the end of the file and make sure that all of them
2364 * are whitespace characters. (This can only happen if
2365 * we are removing blank lines at the end of the file.)
2367 buf = preimage_eof = preimage->buf + preoff;
2368 for ( ; i < preimage->nr; i++)
2369 preoff += preimage->line[i].len;
2370 preimage_end = preimage->buf + preoff;
2371 for ( ; buf < preimage_end; buf++)
2372 if (!isspace(*buf))
2373 return 0;
2376 * Update the preimage and the common postimage context
2377 * lines to use the same whitespace as the target.
2378 * If whitespace is missing in the target (i.e.
2379 * if the preimage extends beyond the end of the file),
2380 * use the whitespace from the preimage.
2382 extra_chars = preimage_end - preimage_eof;
2383 strbuf_init(&fixed, imgoff + extra_chars);
2384 strbuf_add(&fixed, img->buf + try, imgoff);
2385 strbuf_add(&fixed, preimage_eof, extra_chars);
2386 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2387 update_pre_post_images(preimage, postimage,
2388 fixed_buf, fixed_len, postlen);
2389 return 1;
2392 if (ws_error_action != correct_ws_error)
2393 return 0;
2396 * The hunk does not apply byte-by-byte, but the hash says
2397 * it might with whitespace fuzz. We weren't asked to
2398 * ignore whitespace, we were asked to correct whitespace
2399 * errors, so let's try matching after whitespace correction.
2401 * While checking the preimage against the target, whitespace
2402 * errors in both fixed, we count how large the corresponding
2403 * postimage needs to be. The postimage prepared by
2404 * apply_one_fragment() has whitespace errors fixed on added
2405 * lines already, but the common lines were propagated as-is,
2406 * which may become longer when their whitespace errors are
2407 * fixed.
2410 /* First count added lines in postimage */
2411 postlen = 0;
2412 for (i = 0; i < postimage->nr; i++) {
2413 if (!(postimage->line[i].flag & LINE_COMMON))
2414 postlen += postimage->line[i].len;
2418 * The preimage may extend beyond the end of the file,
2419 * but in this loop we will only handle the part of the
2420 * preimage that falls within the file.
2422 strbuf_init(&fixed, preimage->len + 1);
2423 orig = preimage->buf;
2424 target = img->buf + try;
2425 for (i = 0; i < preimage_limit; i++) {
2426 size_t oldlen = preimage->line[i].len;
2427 size_t tgtlen = img->line[try_lno + i].len;
2428 size_t fixstart = fixed.len;
2429 struct strbuf tgtfix;
2430 int match;
2432 /* Try fixing the line in the preimage */
2433 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2435 /* Try fixing the line in the target */
2436 strbuf_init(&tgtfix, tgtlen);
2437 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2440 * If they match, either the preimage was based on
2441 * a version before our tree fixed whitespace breakage,
2442 * or we are lacking a whitespace-fix patch the tree
2443 * the preimage was based on already had (i.e. target
2444 * has whitespace breakage, the preimage doesn't).
2445 * In either case, we are fixing the whitespace breakages
2446 * so we might as well take the fix together with their
2447 * real change.
2449 match = (tgtfix.len == fixed.len - fixstart &&
2450 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2451 fixed.len - fixstart));
2453 /* Add the length if this is common with the postimage */
2454 if (preimage->line[i].flag & LINE_COMMON)
2455 postlen += tgtfix.len;
2457 strbuf_release(&tgtfix);
2458 if (!match)
2459 goto unmatch_exit;
2461 orig += oldlen;
2462 target += tgtlen;
2467 * Now handle the lines in the preimage that falls beyond the
2468 * end of the file (if any). They will only match if they are
2469 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2470 * false).
2472 for ( ; i < preimage->nr; i++) {
2473 size_t fixstart = fixed.len; /* start of the fixed preimage */
2474 size_t oldlen = preimage->line[i].len;
2475 int j;
2477 /* Try fixing the line in the preimage */
2478 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2480 for (j = fixstart; j < fixed.len; j++)
2481 if (!isspace(fixed.buf[j]))
2482 goto unmatch_exit;
2484 orig += oldlen;
2488 * Yes, the preimage is based on an older version that still
2489 * has whitespace breakages unfixed, and fixing them makes the
2490 * hunk match. Update the context lines in the postimage.
2492 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2493 if (postlen < postimage->len)
2494 postlen = 0;
2495 update_pre_post_images(preimage, postimage,
2496 fixed_buf, fixed_len, postlen);
2497 return 1;
2499 unmatch_exit:
2500 strbuf_release(&fixed);
2501 return 0;
2504 static int find_pos(struct image *img,
2505 struct image *preimage,
2506 struct image *postimage,
2507 int line,
2508 unsigned ws_rule,
2509 int match_beginning, int match_end)
2511 int i;
2512 unsigned long backwards, forwards, try;
2513 int backwards_lno, forwards_lno, try_lno;
2516 * If match_beginning or match_end is specified, there is no
2517 * point starting from a wrong line that will never match and
2518 * wander around and wait for a match at the specified end.
2520 if (match_beginning)
2521 line = 0;
2522 else if (match_end)
2523 line = img->nr - preimage->nr;
2526 * Because the comparison is unsigned, the following test
2527 * will also take care of a negative line number that can
2528 * result when match_end and preimage is larger than the target.
2530 if ((size_t) line > img->nr)
2531 line = img->nr;
2533 try = 0;
2534 for (i = 0; i < line; i++)
2535 try += img->line[i].len;
2538 * There's probably some smart way to do this, but I'll leave
2539 * that to the smart and beautiful people. I'm simple and stupid.
2541 backwards = try;
2542 backwards_lno = line;
2543 forwards = try;
2544 forwards_lno = line;
2545 try_lno = line;
2547 for (i = 0; ; i++) {
2548 if (match_fragment(img, preimage, postimage,
2549 try, try_lno, ws_rule,
2550 match_beginning, match_end))
2551 return try_lno;
2553 again:
2554 if (backwards_lno == 0 && forwards_lno == img->nr)
2555 break;
2557 if (i & 1) {
2558 if (backwards_lno == 0) {
2559 i++;
2560 goto again;
2562 backwards_lno--;
2563 backwards -= img->line[backwards_lno].len;
2564 try = backwards;
2565 try_lno = backwards_lno;
2566 } else {
2567 if (forwards_lno == img->nr) {
2568 i++;
2569 goto again;
2571 forwards += img->line[forwards_lno].len;
2572 forwards_lno++;
2573 try = forwards;
2574 try_lno = forwards_lno;
2578 return -1;
2581 static void remove_first_line(struct image *img)
2583 img->buf += img->line[0].len;
2584 img->len -= img->line[0].len;
2585 img->line++;
2586 img->nr--;
2589 static void remove_last_line(struct image *img)
2591 img->len -= img->line[--img->nr].len;
2595 * The change from "preimage" and "postimage" has been found to
2596 * apply at applied_pos (counts in line numbers) in "img".
2597 * Update "img" to remove "preimage" and replace it with "postimage".
2599 static void update_image(struct image *img,
2600 int applied_pos,
2601 struct image *preimage,
2602 struct image *postimage)
2605 * remove the copy of preimage at offset in img
2606 * and replace it with postimage
2608 int i, nr;
2609 size_t remove_count, insert_count, applied_at = 0;
2610 char *result;
2611 int preimage_limit;
2614 * If we are removing blank lines at the end of img,
2615 * the preimage may extend beyond the end.
2616 * If that is the case, we must be careful only to
2617 * remove the part of the preimage that falls within
2618 * the boundaries of img. Initialize preimage_limit
2619 * to the number of lines in the preimage that falls
2620 * within the boundaries.
2622 preimage_limit = preimage->nr;
2623 if (preimage_limit > img->nr - applied_pos)
2624 preimage_limit = img->nr - applied_pos;
2626 for (i = 0; i < applied_pos; i++)
2627 applied_at += img->line[i].len;
2629 remove_count = 0;
2630 for (i = 0; i < preimage_limit; i++)
2631 remove_count += img->line[applied_pos + i].len;
2632 insert_count = postimage->len;
2634 /* Adjust the contents */
2635 result = xmalloc(img->len + insert_count - remove_count + 1);
2636 memcpy(result, img->buf, applied_at);
2637 memcpy(result + applied_at, postimage->buf, postimage->len);
2638 memcpy(result + applied_at + postimage->len,
2639 img->buf + (applied_at + remove_count),
2640 img->len - (applied_at + remove_count));
2641 free(img->buf);
2642 img->buf = result;
2643 img->len += insert_count - remove_count;
2644 result[img->len] = '\0';
2646 /* Adjust the line table */
2647 nr = img->nr + postimage->nr - preimage_limit;
2648 if (preimage_limit < postimage->nr) {
2650 * NOTE: this knows that we never call remove_first_line()
2651 * on anything other than pre/post image.
2653 REALLOC_ARRAY(img->line, nr);
2654 img->line_allocated = img->line;
2656 if (preimage_limit != postimage->nr)
2657 memmove(img->line + applied_pos + postimage->nr,
2658 img->line + applied_pos + preimage_limit,
2659 (img->nr - (applied_pos + preimage_limit)) *
2660 sizeof(*img->line));
2661 memcpy(img->line + applied_pos,
2662 postimage->line,
2663 postimage->nr * sizeof(*img->line));
2664 if (!allow_overlap)
2665 for (i = 0; i < postimage->nr; i++)
2666 img->line[applied_pos + i].flag |= LINE_PATCHED;
2667 img->nr = nr;
2671 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2672 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2673 * replace the part of "img" with "postimage" text.
2675 static int apply_one_fragment(struct image *img, struct fragment *frag,
2676 int inaccurate_eof, unsigned ws_rule,
2677 int nth_fragment)
2679 int match_beginning, match_end;
2680 const char *patch = frag->patch;
2681 int size = frag->size;
2682 char *old, *oldlines;
2683 struct strbuf newlines;
2684 int new_blank_lines_at_end = 0;
2685 int found_new_blank_lines_at_end = 0;
2686 int hunk_linenr = frag->linenr;
2687 unsigned long leading, trailing;
2688 int pos, applied_pos;
2689 struct image preimage;
2690 struct image postimage;
2692 memset(&preimage, 0, sizeof(preimage));
2693 memset(&postimage, 0, sizeof(postimage));
2694 oldlines = xmalloc(size);
2695 strbuf_init(&newlines, size);
2697 old = oldlines;
2698 while (size > 0) {
2699 char first;
2700 int len = linelen(patch, size);
2701 int plen;
2702 int added_blank_line = 0;
2703 int is_blank_context = 0;
2704 size_t start;
2706 if (!len)
2707 break;
2710 * "plen" is how much of the line we should use for
2711 * the actual patch data. Normally we just remove the
2712 * first character on the line, but if the line is
2713 * followed by "\ No newline", then we also remove the
2714 * last one (which is the newline, of course).
2716 plen = len - 1;
2717 if (len < size && patch[len] == '\\')
2718 plen--;
2719 first = *patch;
2720 if (apply_in_reverse) {
2721 if (first == '-')
2722 first = '+';
2723 else if (first == '+')
2724 first = '-';
2727 switch (first) {
2728 case '\n':
2729 /* Newer GNU diff, empty context line */
2730 if (plen < 0)
2731 /* ... followed by '\No newline'; nothing */
2732 break;
2733 *old++ = '\n';
2734 strbuf_addch(&newlines, '\n');
2735 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2736 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2737 is_blank_context = 1;
2738 break;
2739 case ' ':
2740 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2741 ws_blank_line(patch + 1, plen, ws_rule))
2742 is_blank_context = 1;
2743 case '-':
2744 memcpy(old, patch + 1, plen);
2745 add_line_info(&preimage, old, plen,
2746 (first == ' ' ? LINE_COMMON : 0));
2747 old += plen;
2748 if (first == '-')
2749 break;
2750 /* Fall-through for ' ' */
2751 case '+':
2752 /* --no-add does not add new lines */
2753 if (first == '+' && no_add)
2754 break;
2756 start = newlines.len;
2757 if (first != '+' ||
2758 !whitespace_error ||
2759 ws_error_action != correct_ws_error) {
2760 strbuf_add(&newlines, patch + 1, plen);
2762 else {
2763 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2765 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2766 (first == '+' ? 0 : LINE_COMMON));
2767 if (first == '+' &&
2768 (ws_rule & WS_BLANK_AT_EOF) &&
2769 ws_blank_line(patch + 1, plen, ws_rule))
2770 added_blank_line = 1;
2771 break;
2772 case '@': case '\\':
2773 /* Ignore it, we already handled it */
2774 break;
2775 default:
2776 if (apply_verbosely)
2777 error(_("invalid start of line: '%c'"), first);
2778 applied_pos = -1;
2779 goto out;
2781 if (added_blank_line) {
2782 if (!new_blank_lines_at_end)
2783 found_new_blank_lines_at_end = hunk_linenr;
2784 new_blank_lines_at_end++;
2786 else if (is_blank_context)
2788 else
2789 new_blank_lines_at_end = 0;
2790 patch += len;
2791 size -= len;
2792 hunk_linenr++;
2794 if (inaccurate_eof &&
2795 old > oldlines && old[-1] == '\n' &&
2796 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2797 old--;
2798 strbuf_setlen(&newlines, newlines.len - 1);
2801 leading = frag->leading;
2802 trailing = frag->trailing;
2805 * A hunk to change lines at the beginning would begin with
2806 * @@ -1,L +N,M @@
2807 * but we need to be careful. -U0 that inserts before the second
2808 * line also has this pattern.
2810 * And a hunk to add to an empty file would begin with
2811 * @@ -0,0 +N,M @@
2813 * In other words, a hunk that is (frag->oldpos <= 1) with or
2814 * without leading context must match at the beginning.
2816 match_beginning = (!frag->oldpos ||
2817 (frag->oldpos == 1 && !unidiff_zero));
2820 * A hunk without trailing lines must match at the end.
2821 * However, we simply cannot tell if a hunk must match end
2822 * from the lack of trailing lines if the patch was generated
2823 * with unidiff without any context.
2825 match_end = !unidiff_zero && !trailing;
2827 pos = frag->newpos ? (frag->newpos - 1) : 0;
2828 preimage.buf = oldlines;
2829 preimage.len = old - oldlines;
2830 postimage.buf = newlines.buf;
2831 postimage.len = newlines.len;
2832 preimage.line = preimage.line_allocated;
2833 postimage.line = postimage.line_allocated;
2835 for (;;) {
2837 applied_pos = find_pos(img, &preimage, &postimage, pos,
2838 ws_rule, match_beginning, match_end);
2840 if (applied_pos >= 0)
2841 break;
2843 /* Am I at my context limits? */
2844 if ((leading <= p_context) && (trailing <= p_context))
2845 break;
2846 if (match_beginning || match_end) {
2847 match_beginning = match_end = 0;
2848 continue;
2852 * Reduce the number of context lines; reduce both
2853 * leading and trailing if they are equal otherwise
2854 * just reduce the larger context.
2856 if (leading >= trailing) {
2857 remove_first_line(&preimage);
2858 remove_first_line(&postimage);
2859 pos--;
2860 leading--;
2862 if (trailing > leading) {
2863 remove_last_line(&preimage);
2864 remove_last_line(&postimage);
2865 trailing--;
2869 if (applied_pos >= 0) {
2870 if (new_blank_lines_at_end &&
2871 preimage.nr + applied_pos >= img->nr &&
2872 (ws_rule & WS_BLANK_AT_EOF) &&
2873 ws_error_action != nowarn_ws_error) {
2874 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2875 found_new_blank_lines_at_end);
2876 if (ws_error_action == correct_ws_error) {
2877 while (new_blank_lines_at_end--)
2878 remove_last_line(&postimage);
2881 * We would want to prevent write_out_results()
2882 * from taking place in apply_patch() that follows
2883 * the callchain led us here, which is:
2884 * apply_patch->check_patch_list->check_patch->
2885 * apply_data->apply_fragments->apply_one_fragment
2887 if (ws_error_action == die_on_ws_error)
2888 apply = 0;
2891 if (apply_verbosely && applied_pos != pos) {
2892 int offset = applied_pos - pos;
2893 if (apply_in_reverse)
2894 offset = 0 - offset;
2895 fprintf_ln(stderr,
2896 Q_("Hunk #%d succeeded at %d (offset %d line).",
2897 "Hunk #%d succeeded at %d (offset %d lines).",
2898 offset),
2899 nth_fragment, applied_pos + 1, offset);
2903 * Warn if it was necessary to reduce the number
2904 * of context lines.
2906 if ((leading != frag->leading) ||
2907 (trailing != frag->trailing))
2908 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2909 " to apply fragment at %d"),
2910 leading, trailing, applied_pos+1);
2911 update_image(img, applied_pos, &preimage, &postimage);
2912 } else {
2913 if (apply_verbosely)
2914 error(_("while searching for:\n%.*s"),
2915 (int)(old - oldlines), oldlines);
2918 out:
2919 free(oldlines);
2920 strbuf_release(&newlines);
2921 free(preimage.line_allocated);
2922 free(postimage.line_allocated);
2924 return (applied_pos < 0);
2927 static int apply_binary_fragment(struct image *img, struct patch *patch)
2929 struct fragment *fragment = patch->fragments;
2930 unsigned long len;
2931 void *dst;
2933 if (!fragment)
2934 return error(_("missing binary patch data for '%s'"),
2935 patch->new_name ?
2936 patch->new_name :
2937 patch->old_name);
2939 /* Binary patch is irreversible without the optional second hunk */
2940 if (apply_in_reverse) {
2941 if (!fragment->next)
2942 return error("cannot reverse-apply a binary patch "
2943 "without the reverse hunk to '%s'",
2944 patch->new_name
2945 ? patch->new_name : patch->old_name);
2946 fragment = fragment->next;
2948 switch (fragment->binary_patch_method) {
2949 case BINARY_DELTA_DEFLATED:
2950 dst = patch_delta(img->buf, img->len, fragment->patch,
2951 fragment->size, &len);
2952 if (!dst)
2953 return -1;
2954 clear_image(img);
2955 img->buf = dst;
2956 img->len = len;
2957 return 0;
2958 case BINARY_LITERAL_DEFLATED:
2959 clear_image(img);
2960 img->len = fragment->size;
2961 img->buf = xmemdupz(fragment->patch, img->len);
2962 return 0;
2964 return -1;
2968 * Replace "img" with the result of applying the binary patch.
2969 * The binary patch data itself in patch->fragment is still kept
2970 * but the preimage prepared by the caller in "img" is freed here
2971 * or in the helper function apply_binary_fragment() this calls.
2973 static int apply_binary(struct image *img, struct patch *patch)
2975 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2976 unsigned char sha1[20];
2979 * For safety, we require patch index line to contain
2980 * full 40-byte textual SHA1 for old and new, at least for now.
2982 if (strlen(patch->old_sha1_prefix) != 40 ||
2983 strlen(patch->new_sha1_prefix) != 40 ||
2984 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2985 get_sha1_hex(patch->new_sha1_prefix, sha1))
2986 return error("cannot apply binary patch to '%s' "
2987 "without full index line", name);
2989 if (patch->old_name) {
2991 * See if the old one matches what the patch
2992 * applies to.
2994 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2995 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2996 return error("the patch applies to '%s' (%s), "
2997 "which does not match the "
2998 "current contents.",
2999 name, sha1_to_hex(sha1));
3001 else {
3002 /* Otherwise, the old one must be empty. */
3003 if (img->len)
3004 return error("the patch applies to an empty "
3005 "'%s' but it is not empty", name);
3008 get_sha1_hex(patch->new_sha1_prefix, sha1);
3009 if (is_null_sha1(sha1)) {
3010 clear_image(img);
3011 return 0; /* deletion patch */
3014 if (has_sha1_file(sha1)) {
3015 /* We already have the postimage */
3016 enum object_type type;
3017 unsigned long size;
3018 char *result;
3020 result = read_sha1_file(sha1, &type, &size);
3021 if (!result)
3022 return error("the necessary postimage %s for "
3023 "'%s' cannot be read",
3024 patch->new_sha1_prefix, name);
3025 clear_image(img);
3026 img->buf = result;
3027 img->len = size;
3028 } else {
3030 * We have verified buf matches the preimage;
3031 * apply the patch data to it, which is stored
3032 * in the patch->fragments->{patch,size}.
3034 if (apply_binary_fragment(img, patch))
3035 return error(_("binary patch does not apply to '%s'"),
3036 name);
3038 /* verify that the result matches */
3039 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3040 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3041 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3042 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3045 return 0;
3048 static int apply_fragments(struct image *img, struct patch *patch)
3050 struct fragment *frag = patch->fragments;
3051 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3052 unsigned ws_rule = patch->ws_rule;
3053 unsigned inaccurate_eof = patch->inaccurate_eof;
3054 int nth = 0;
3056 if (patch->is_binary)
3057 return apply_binary(img, patch);
3059 while (frag) {
3060 nth++;
3061 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3062 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3063 if (!apply_with_reject)
3064 return -1;
3065 frag->rejected = 1;
3067 frag = frag->next;
3069 return 0;
3072 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3074 if (S_ISGITLINK(mode)) {
3075 strbuf_grow(buf, 100);
3076 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3077 } else {
3078 enum object_type type;
3079 unsigned long sz;
3080 char *result;
3082 result = read_sha1_file(sha1, &type, &sz);
3083 if (!result)
3084 return -1;
3085 /* XXX read_sha1_file NUL-terminates */
3086 strbuf_attach(buf, result, sz, sz + 1);
3088 return 0;
3091 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3093 if (!ce)
3094 return 0;
3095 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3098 static struct patch *in_fn_table(const char *name)
3100 struct string_list_item *item;
3102 if (name == NULL)
3103 return NULL;
3105 item = string_list_lookup(&fn_table, name);
3106 if (item != NULL)
3107 return (struct patch *)item->util;
3109 return NULL;
3113 * item->util in the filename table records the status of the path.
3114 * Usually it points at a patch (whose result records the contents
3115 * of it after applying it), but it could be PATH_WAS_DELETED for a
3116 * path that a previously applied patch has already removed, or
3117 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3119 * The latter is needed to deal with a case where two paths A and B
3120 * are swapped by first renaming A to B and then renaming B to A;
3121 * moving A to B should not be prevented due to presence of B as we
3122 * will remove it in a later patch.
3124 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3125 #define PATH_WAS_DELETED ((struct patch *) -1)
3127 static int to_be_deleted(struct patch *patch)
3129 return patch == PATH_TO_BE_DELETED;
3132 static int was_deleted(struct patch *patch)
3134 return patch == PATH_WAS_DELETED;
3137 static void add_to_fn_table(struct patch *patch)
3139 struct string_list_item *item;
3142 * Always add new_name unless patch is a deletion
3143 * This should cover the cases for normal diffs,
3144 * file creations and copies
3146 if (patch->new_name != NULL) {
3147 item = string_list_insert(&fn_table, patch->new_name);
3148 item->util = patch;
3152 * store a failure on rename/deletion cases because
3153 * later chunks shouldn't patch old names
3155 if ((patch->new_name == NULL) || (patch->is_rename)) {
3156 item = string_list_insert(&fn_table, patch->old_name);
3157 item->util = PATH_WAS_DELETED;
3161 static void prepare_fn_table(struct patch *patch)
3164 * store information about incoming file deletion
3166 while (patch) {
3167 if ((patch->new_name == NULL) || (patch->is_rename)) {
3168 struct string_list_item *item;
3169 item = string_list_insert(&fn_table, patch->old_name);
3170 item->util = PATH_TO_BE_DELETED;
3172 patch = patch->next;
3176 static int checkout_target(struct index_state *istate,
3177 struct cache_entry *ce, struct stat *st)
3179 struct checkout costate;
3181 memset(&costate, 0, sizeof(costate));
3182 costate.base_dir = "";
3183 costate.refresh_cache = 1;
3184 costate.istate = istate;
3185 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3186 return error(_("cannot checkout %s"), ce->name);
3187 return 0;
3190 static struct patch *previous_patch(struct patch *patch, int *gone)
3192 struct patch *previous;
3194 *gone = 0;
3195 if (patch->is_copy || patch->is_rename)
3196 return NULL; /* "git" patches do not depend on the order */
3198 previous = in_fn_table(patch->old_name);
3199 if (!previous)
3200 return NULL;
3202 if (to_be_deleted(previous))
3203 return NULL; /* the deletion hasn't happened yet */
3205 if (was_deleted(previous))
3206 *gone = 1;
3208 return previous;
3211 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3213 if (S_ISGITLINK(ce->ce_mode)) {
3214 if (!S_ISDIR(st->st_mode))
3215 return -1;
3216 return 0;
3218 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3221 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3223 static int load_patch_target(struct strbuf *buf,
3224 const struct cache_entry *ce,
3225 struct stat *st,
3226 const char *name,
3227 unsigned expected_mode)
3229 if (cached || check_index) {
3230 if (read_file_or_gitlink(ce, buf))
3231 return error(_("read of %s failed"), name);
3232 } else if (name) {
3233 if (S_ISGITLINK(expected_mode)) {
3234 if (ce)
3235 return read_file_or_gitlink(ce, buf);
3236 else
3237 return SUBMODULE_PATCH_WITHOUT_INDEX;
3238 } else if (has_symlink_leading_path(name, strlen(name))) {
3239 return error(_("reading from '%s' beyond a symbolic link"), name);
3240 } else {
3241 if (read_old_data(st, name, buf))
3242 return error(_("read of %s failed"), name);
3245 return 0;
3249 * We are about to apply "patch"; populate the "image" with the
3250 * current version we have, from the working tree or from the index,
3251 * depending on the situation e.g. --cached/--index. If we are
3252 * applying a non-git patch that incrementally updates the tree,
3253 * we read from the result of a previous diff.
3255 static int load_preimage(struct image *image,
3256 struct patch *patch, struct stat *st,
3257 const struct cache_entry *ce)
3259 struct strbuf buf = STRBUF_INIT;
3260 size_t len;
3261 char *img;
3262 struct patch *previous;
3263 int status;
3265 previous = previous_patch(patch, &status);
3266 if (status)
3267 return error(_("path %s has been renamed/deleted"),
3268 patch->old_name);
3269 if (previous) {
3270 /* We have a patched copy in memory; use that. */
3271 strbuf_add(&buf, previous->result, previous->resultsize);
3272 } else {
3273 status = load_patch_target(&buf, ce, st,
3274 patch->old_name, patch->old_mode);
3275 if (status < 0)
3276 return status;
3277 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3279 * There is no way to apply subproject
3280 * patch without looking at the index.
3281 * NEEDSWORK: shouldn't this be flagged
3282 * as an error???
3284 free_fragment_list(patch->fragments);
3285 patch->fragments = NULL;
3286 } else if (status) {
3287 return error(_("read of %s failed"), patch->old_name);
3291 img = strbuf_detach(&buf, &len);
3292 prepare_image(image, img, len, !patch->is_binary);
3293 return 0;
3296 static int three_way_merge(struct image *image,
3297 char *path,
3298 const unsigned char *base,
3299 const unsigned char *ours,
3300 const unsigned char *theirs)
3302 mmfile_t base_file, our_file, their_file;
3303 mmbuffer_t result = { NULL };
3304 int status;
3306 read_mmblob(&base_file, base);
3307 read_mmblob(&our_file, ours);
3308 read_mmblob(&their_file, theirs);
3309 status = ll_merge(&result, path,
3310 &base_file, "base",
3311 &our_file, "ours",
3312 &their_file, "theirs", NULL);
3313 free(base_file.ptr);
3314 free(our_file.ptr);
3315 free(their_file.ptr);
3316 if (status < 0 || !result.ptr) {
3317 free(result.ptr);
3318 return -1;
3320 clear_image(image);
3321 image->buf = result.ptr;
3322 image->len = result.size;
3324 return status;
3328 * When directly falling back to add/add three-way merge, we read from
3329 * the current contents of the new_name. In no cases other than that
3330 * this function will be called.
3332 static int load_current(struct image *image, struct patch *patch)
3334 struct strbuf buf = STRBUF_INIT;
3335 int status, pos;
3336 size_t len;
3337 char *img;
3338 struct stat st;
3339 struct cache_entry *ce;
3340 char *name = patch->new_name;
3341 unsigned mode = patch->new_mode;
3343 if (!patch->is_new)
3344 die("BUG: patch to %s is not a creation", patch->old_name);
3346 pos = cache_name_pos(name, strlen(name));
3347 if (pos < 0)
3348 return error(_("%s: does not exist in index"), name);
3349 ce = active_cache[pos];
3350 if (lstat(name, &st)) {
3351 if (errno != ENOENT)
3352 return error(_("%s: %s"), name, strerror(errno));
3353 if (checkout_target(&the_index, ce, &st))
3354 return -1;
3356 if (verify_index_match(ce, &st))
3357 return error(_("%s: does not match index"), name);
3359 status = load_patch_target(&buf, ce, &st, name, mode);
3360 if (status < 0)
3361 return status;
3362 else if (status)
3363 return -1;
3364 img = strbuf_detach(&buf, &len);
3365 prepare_image(image, img, len, !patch->is_binary);
3366 return 0;
3369 static int try_threeway(struct image *image, struct patch *patch,
3370 struct stat *st, const struct cache_entry *ce)
3372 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3373 struct strbuf buf = STRBUF_INIT;
3374 size_t len;
3375 int status;
3376 char *img;
3377 struct image tmp_image;
3379 /* No point falling back to 3-way merge in these cases */
3380 if (patch->is_delete ||
3381 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3382 return -1;
3384 /* Preimage the patch was prepared for */
3385 if (patch->is_new)
3386 write_sha1_file("", 0, blob_type, pre_sha1);
3387 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3388 read_blob_object(&buf, pre_sha1, patch->old_mode))
3389 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3391 fprintf(stderr, "Falling back to three-way merge...\n");
3393 img = strbuf_detach(&buf, &len);
3394 prepare_image(&tmp_image, img, len, 1);
3395 /* Apply the patch to get the post image */
3396 if (apply_fragments(&tmp_image, patch) < 0) {
3397 clear_image(&tmp_image);
3398 return -1;
3400 /* post_sha1[] is theirs */
3401 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3402 clear_image(&tmp_image);
3404 /* our_sha1[] is ours */
3405 if (patch->is_new) {
3406 if (load_current(&tmp_image, patch))
3407 return error("cannot read the current contents of '%s'",
3408 patch->new_name);
3409 } else {
3410 if (load_preimage(&tmp_image, patch, st, ce))
3411 return error("cannot read the current contents of '%s'",
3412 patch->old_name);
3414 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3415 clear_image(&tmp_image);
3417 /* in-core three-way merge between post and our using pre as base */
3418 status = three_way_merge(image, patch->new_name,
3419 pre_sha1, our_sha1, post_sha1);
3420 if (status < 0) {
3421 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3422 return status;
3425 if (status) {
3426 patch->conflicted_threeway = 1;
3427 if (patch->is_new)
3428 oidclr(&patch->threeway_stage[0]);
3429 else
3430 hashcpy(patch->threeway_stage[0].hash, pre_sha1);
3431 hashcpy(patch->threeway_stage[1].hash, our_sha1);
3432 hashcpy(patch->threeway_stage[2].hash, post_sha1);
3433 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3434 } else {
3435 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3437 return 0;
3440 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3442 struct image image;
3444 if (load_preimage(&image, patch, st, ce) < 0)
3445 return -1;
3447 if (patch->direct_to_threeway ||
3448 apply_fragments(&image, patch) < 0) {
3449 /* Note: with --reject, apply_fragments() returns 0 */
3450 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3451 return -1;
3453 patch->result = image.buf;
3454 patch->resultsize = image.len;
3455 add_to_fn_table(patch);
3456 free(image.line_allocated);
3458 if (0 < patch->is_delete && patch->resultsize)
3459 return error(_("removal patch leaves file contents"));
3461 return 0;
3465 * If "patch" that we are looking at modifies or deletes what we have,
3466 * we would want it not to lose any local modification we have, either
3467 * in the working tree or in the index.
3469 * This also decides if a non-git patch is a creation patch or a
3470 * modification to an existing empty file. We do not check the state
3471 * of the current tree for a creation patch in this function; the caller
3472 * check_patch() separately makes sure (and errors out otherwise) that
3473 * the path the patch creates does not exist in the current tree.
3475 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3477 const char *old_name = patch->old_name;
3478 struct patch *previous = NULL;
3479 int stat_ret = 0, status;
3480 unsigned st_mode = 0;
3482 if (!old_name)
3483 return 0;
3485 assert(patch->is_new <= 0);
3486 previous = previous_patch(patch, &status);
3488 if (status)
3489 return error(_("path %s has been renamed/deleted"), old_name);
3490 if (previous) {
3491 st_mode = previous->new_mode;
3492 } else if (!cached) {
3493 stat_ret = lstat(old_name, st);
3494 if (stat_ret && errno != ENOENT)
3495 return error(_("%s: %s"), old_name, strerror(errno));
3498 if (check_index && !previous) {
3499 int pos = cache_name_pos(old_name, strlen(old_name));
3500 if (pos < 0) {
3501 if (patch->is_new < 0)
3502 goto is_new;
3503 return error(_("%s: does not exist in index"), old_name);
3505 *ce = active_cache[pos];
3506 if (stat_ret < 0) {
3507 if (checkout_target(&the_index, *ce, st))
3508 return -1;
3510 if (!cached && verify_index_match(*ce, st))
3511 return error(_("%s: does not match index"), old_name);
3512 if (cached)
3513 st_mode = (*ce)->ce_mode;
3514 } else if (stat_ret < 0) {
3515 if (patch->is_new < 0)
3516 goto is_new;
3517 return error(_("%s: %s"), old_name, strerror(errno));
3520 if (!cached && !previous)
3521 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3523 if (patch->is_new < 0)
3524 patch->is_new = 0;
3525 if (!patch->old_mode)
3526 patch->old_mode = st_mode;
3527 if ((st_mode ^ patch->old_mode) & S_IFMT)
3528 return error(_("%s: wrong type"), old_name);
3529 if (st_mode != patch->old_mode)
3530 warning(_("%s has type %o, expected %o"),
3531 old_name, st_mode, patch->old_mode);
3532 if (!patch->new_mode && !patch->is_delete)
3533 patch->new_mode = st_mode;
3534 return 0;
3536 is_new:
3537 patch->is_new = 1;
3538 patch->is_delete = 0;
3539 free(patch->old_name);
3540 patch->old_name = NULL;
3541 return 0;
3545 #define EXISTS_IN_INDEX 1
3546 #define EXISTS_IN_WORKTREE 2
3548 static int check_to_create(const char *new_name, int ok_if_exists)
3550 struct stat nst;
3552 if (check_index &&
3553 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3554 !ok_if_exists)
3555 return EXISTS_IN_INDEX;
3556 if (cached)
3557 return 0;
3559 if (!lstat(new_name, &nst)) {
3560 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3561 return 0;
3563 * A leading component of new_name might be a symlink
3564 * that is going to be removed with this patch, but
3565 * still pointing at somewhere that has the path.
3566 * In such a case, path "new_name" does not exist as
3567 * far as git is concerned.
3569 if (has_symlink_leading_path(new_name, strlen(new_name)))
3570 return 0;
3572 return EXISTS_IN_WORKTREE;
3573 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3574 return error("%s: %s", new_name, strerror(errno));
3576 return 0;
3580 * We need to keep track of how symlinks in the preimage are
3581 * manipulated by the patches. A patch to add a/b/c where a/b
3582 * is a symlink should not be allowed to affect the directory
3583 * the symlink points at, but if the same patch removes a/b,
3584 * it is perfectly fine, as the patch removes a/b to make room
3585 * to create a directory a/b so that a/b/c can be created.
3587 static struct string_list symlink_changes;
3588 #define SYMLINK_GOES_AWAY 01
3589 #define SYMLINK_IN_RESULT 02
3591 static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3593 struct string_list_item *ent;
3595 ent = string_list_lookup(&symlink_changes, path);
3596 if (!ent) {
3597 ent = string_list_insert(&symlink_changes, path);
3598 ent->util = (void *)0;
3600 ent->util = (void *)(what | ((uintptr_t)ent->util));
3601 return (uintptr_t)ent->util;
3604 static uintptr_t check_symlink_changes(const char *path)
3606 struct string_list_item *ent;
3608 ent = string_list_lookup(&symlink_changes, path);
3609 if (!ent)
3610 return 0;
3611 return (uintptr_t)ent->util;
3614 static void prepare_symlink_changes(struct patch *patch)
3616 for ( ; patch; patch = patch->next) {
3617 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3618 (patch->is_rename || patch->is_delete))
3619 /* the symlink at patch->old_name is removed */
3620 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3622 if (patch->new_name && S_ISLNK(patch->new_mode))
3623 /* the symlink at patch->new_name is created or remains */
3624 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3628 static int path_is_beyond_symlink_1(struct strbuf *name)
3630 do {
3631 unsigned int change;
3633 while (--name->len && name->buf[name->len] != '/')
3634 ; /* scan backwards */
3635 if (!name->len)
3636 break;
3637 name->buf[name->len] = '\0';
3638 change = check_symlink_changes(name->buf);
3639 if (change & SYMLINK_IN_RESULT)
3640 return 1;
3641 if (change & SYMLINK_GOES_AWAY)
3643 * This cannot be "return 0", because we may
3644 * see a new one created at a higher level.
3646 continue;
3648 /* otherwise, check the preimage */
3649 if (check_index) {
3650 struct cache_entry *ce;
3652 ce = cache_file_exists(name->buf, name->len, ignore_case);
3653 if (ce && S_ISLNK(ce->ce_mode))
3654 return 1;
3655 } else {
3656 struct stat st;
3657 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3658 return 1;
3660 } while (1);
3661 return 0;
3664 static int path_is_beyond_symlink(const char *name_)
3666 int ret;
3667 struct strbuf name = STRBUF_INIT;
3669 assert(*name_ != '\0');
3670 strbuf_addstr(&name, name_);
3671 ret = path_is_beyond_symlink_1(&name);
3672 strbuf_release(&name);
3674 return ret;
3677 static void die_on_unsafe_path(struct patch *patch)
3679 const char *old_name = NULL;
3680 const char *new_name = NULL;
3681 if (patch->is_delete)
3682 old_name = patch->old_name;
3683 else if (!patch->is_new && !patch->is_copy)
3684 old_name = patch->old_name;
3685 if (!patch->is_delete)
3686 new_name = patch->new_name;
3688 if (old_name && !verify_path(old_name))
3689 die(_("invalid path '%s'"), old_name);
3690 if (new_name && !verify_path(new_name))
3691 die(_("invalid path '%s'"), new_name);
3695 * Check and apply the patch in-core; leave the result in patch->result
3696 * for the caller to write it out to the final destination.
3698 static int check_patch(struct patch *patch)
3700 struct stat st;
3701 const char *old_name = patch->old_name;
3702 const char *new_name = patch->new_name;
3703 const char *name = old_name ? old_name : new_name;
3704 struct cache_entry *ce = NULL;
3705 struct patch *tpatch;
3706 int ok_if_exists;
3707 int status;
3709 patch->rejected = 1; /* we will drop this after we succeed */
3711 status = check_preimage(patch, &ce, &st);
3712 if (status)
3713 return status;
3714 old_name = patch->old_name;
3717 * A type-change diff is always split into a patch to delete
3718 * old, immediately followed by a patch to create new (see
3719 * diff.c::run_diff()); in such a case it is Ok that the entry
3720 * to be deleted by the previous patch is still in the working
3721 * tree and in the index.
3723 * A patch to swap-rename between A and B would first rename A
3724 * to B and then rename B to A. While applying the first one,
3725 * the presence of B should not stop A from getting renamed to
3726 * B; ask to_be_deleted() about the later rename. Removal of
3727 * B and rename from A to B is handled the same way by asking
3728 * was_deleted().
3730 if ((tpatch = in_fn_table(new_name)) &&
3731 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3732 ok_if_exists = 1;
3733 else
3734 ok_if_exists = 0;
3736 if (new_name &&
3737 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3738 int err = check_to_create(new_name, ok_if_exists);
3740 if (err && threeway) {
3741 patch->direct_to_threeway = 1;
3742 } else switch (err) {
3743 case 0:
3744 break; /* happy */
3745 case EXISTS_IN_INDEX:
3746 return error(_("%s: already exists in index"), new_name);
3747 break;
3748 case EXISTS_IN_WORKTREE:
3749 return error(_("%s: already exists in working directory"),
3750 new_name);
3751 default:
3752 return err;
3755 if (!patch->new_mode) {
3756 if (0 < patch->is_new)
3757 patch->new_mode = S_IFREG | 0644;
3758 else
3759 patch->new_mode = patch->old_mode;
3763 if (new_name && old_name) {
3764 int same = !strcmp(old_name, new_name);
3765 if (!patch->new_mode)
3766 patch->new_mode = patch->old_mode;
3767 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3768 if (same)
3769 return error(_("new mode (%o) of %s does not "
3770 "match old mode (%o)"),
3771 patch->new_mode, new_name,
3772 patch->old_mode);
3773 else
3774 return error(_("new mode (%o) of %s does not "
3775 "match old mode (%o) of %s"),
3776 patch->new_mode, new_name,
3777 patch->old_mode, old_name);
3781 if (!unsafe_paths)
3782 die_on_unsafe_path(patch);
3785 * An attempt to read from or delete a path that is beyond a
3786 * symbolic link will be prevented by load_patch_target() that
3787 * is called at the beginning of apply_data() so we do not
3788 * have to worry about a patch marked with "is_delete" bit
3789 * here. We however need to make sure that the patch result
3790 * is not deposited to a path that is beyond a symbolic link
3791 * here.
3793 if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))
3794 return error(_("affected file '%s' is beyond a symbolic link"),
3795 patch->new_name);
3797 if (apply_data(patch, &st, ce) < 0)
3798 return error(_("%s: patch does not apply"), name);
3799 patch->rejected = 0;
3800 return 0;
3803 static int check_patch_list(struct patch *patch)
3805 int err = 0;
3807 prepare_symlink_changes(patch);
3808 prepare_fn_table(patch);
3809 while (patch) {
3810 if (apply_verbosely)
3811 say_patch_name(stderr,
3812 _("Checking patch %s..."), patch);
3813 err |= check_patch(patch);
3814 patch = patch->next;
3816 return err;
3819 /* This function tries to read the sha1 from the current index */
3820 static int get_current_sha1(const char *path, unsigned char *sha1)
3822 int pos;
3824 if (read_cache() < 0)
3825 return -1;
3826 pos = cache_name_pos(path, strlen(path));
3827 if (pos < 0)
3828 return -1;
3829 hashcpy(sha1, active_cache[pos]->sha1);
3830 return 0;
3833 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3836 * A usable gitlink patch has only one fragment (hunk) that looks like:
3837 * @@ -1 +1 @@
3838 * -Subproject commit <old sha1>
3839 * +Subproject commit <new sha1>
3840 * or
3841 * @@ -1 +0,0 @@
3842 * -Subproject commit <old sha1>
3843 * for a removal patch.
3845 struct fragment *hunk = p->fragments;
3846 static const char heading[] = "-Subproject commit ";
3847 char *preimage;
3849 if (/* does the patch have only one hunk? */
3850 hunk && !hunk->next &&
3851 /* is its preimage one line? */
3852 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3853 /* does preimage begin with the heading? */
3854 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3855 starts_with(++preimage, heading) &&
3856 /* does it record full SHA-1? */
3857 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3858 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3859 /* does the abbreviated name on the index line agree with it? */
3860 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3861 return 0; /* it all looks fine */
3863 /* we may have full object name on the index line */
3864 return get_sha1_hex(p->old_sha1_prefix, sha1);
3867 /* Build an index that contains the just the files needed for a 3way merge */
3868 static void build_fake_ancestor(struct patch *list, const char *filename)
3870 struct patch *patch;
3871 struct index_state result = { NULL };
3872 static struct lock_file lock;
3874 /* Once we start supporting the reverse patch, it may be
3875 * worth showing the new sha1 prefix, but until then...
3877 for (patch = list; patch; patch = patch->next) {
3878 unsigned char sha1[20];
3879 struct cache_entry *ce;
3880 const char *name;
3882 name = patch->old_name ? patch->old_name : patch->new_name;
3883 if (0 < patch->is_new)
3884 continue;
3886 if (S_ISGITLINK(patch->old_mode)) {
3887 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3888 ; /* ok, the textual part looks sane */
3889 else
3890 die("sha1 information is lacking or useless for submodule %s",
3891 name);
3892 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3893 ; /* ok */
3894 } else if (!patch->lines_added && !patch->lines_deleted) {
3895 /* mode-only change: update the current */
3896 if (get_current_sha1(patch->old_name, sha1))
3897 die("mode change for %s, which is not "
3898 "in current HEAD", name);
3899 } else
3900 die("sha1 information is lacking or useless "
3901 "(%s).", name);
3903 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3904 if (!ce)
3905 die(_("make_cache_entry failed for path '%s'"), name);
3906 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3907 die ("Could not add %s to temporary index", name);
3910 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3911 if (write_locked_index(&result, &lock, COMMIT_LOCK))
3912 die ("Could not write temporary index to %s", filename);
3914 discard_index(&result);
3917 static void stat_patch_list(struct patch *patch)
3919 int files, adds, dels;
3921 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3922 files++;
3923 adds += patch->lines_added;
3924 dels += patch->lines_deleted;
3925 show_stats(patch);
3928 print_stat_summary(stdout, files, adds, dels);
3931 static void numstat_patch_list(struct patch *patch)
3933 for ( ; patch; patch = patch->next) {
3934 const char *name;
3935 name = patch->new_name ? patch->new_name : patch->old_name;
3936 if (patch->is_binary)
3937 printf("-\t-\t");
3938 else
3939 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3940 write_name_quoted(name, stdout, line_termination);
3944 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3946 if (mode)
3947 printf(" %s mode %06o %s\n", newdelete, mode, name);
3948 else
3949 printf(" %s %s\n", newdelete, name);
3952 static void show_mode_change(struct patch *p, int show_name)
3954 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3955 if (show_name)
3956 printf(" mode change %06o => %06o %s\n",
3957 p->old_mode, p->new_mode, p->new_name);
3958 else
3959 printf(" mode change %06o => %06o\n",
3960 p->old_mode, p->new_mode);
3964 static void show_rename_copy(struct patch *p)
3966 const char *renamecopy = p->is_rename ? "rename" : "copy";
3967 const char *old, *new;
3969 /* Find common prefix */
3970 old = p->old_name;
3971 new = p->new_name;
3972 while (1) {
3973 const char *slash_old, *slash_new;
3974 slash_old = strchr(old, '/');
3975 slash_new = strchr(new, '/');
3976 if (!slash_old ||
3977 !slash_new ||
3978 slash_old - old != slash_new - new ||
3979 memcmp(old, new, slash_new - new))
3980 break;
3981 old = slash_old + 1;
3982 new = slash_new + 1;
3984 /* p->old_name thru old is the common prefix, and old and new
3985 * through the end of names are renames
3987 if (old != p->old_name)
3988 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3989 (int)(old - p->old_name), p->old_name,
3990 old, new, p->score);
3991 else
3992 printf(" %s %s => %s (%d%%)\n", renamecopy,
3993 p->old_name, p->new_name, p->score);
3994 show_mode_change(p, 0);
3997 static void summary_patch_list(struct patch *patch)
3999 struct patch *p;
4001 for (p = patch; p; p = p->next) {
4002 if (p->is_new)
4003 show_file_mode_name("create", p->new_mode, p->new_name);
4004 else if (p->is_delete)
4005 show_file_mode_name("delete", p->old_mode, p->old_name);
4006 else {
4007 if (p->is_rename || p->is_copy)
4008 show_rename_copy(p);
4009 else {
4010 if (p->score) {
4011 printf(" rewrite %s (%d%%)\n",
4012 p->new_name, p->score);
4013 show_mode_change(p, 0);
4015 else
4016 show_mode_change(p, 1);
4022 static void patch_stats(struct patch *patch)
4024 int lines = patch->lines_added + patch->lines_deleted;
4026 if (lines > max_change)
4027 max_change = lines;
4028 if (patch->old_name) {
4029 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4030 if (!len)
4031 len = strlen(patch->old_name);
4032 if (len > max_len)
4033 max_len = len;
4035 if (patch->new_name) {
4036 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4037 if (!len)
4038 len = strlen(patch->new_name);
4039 if (len > max_len)
4040 max_len = len;
4044 static void remove_file(struct patch *patch, int rmdir_empty)
4046 if (update_index) {
4047 if (remove_file_from_cache(patch->old_name) < 0)
4048 die(_("unable to remove %s from index"), patch->old_name);
4050 if (!cached) {
4051 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4052 remove_path(patch->old_name);
4057 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
4059 struct stat st;
4060 struct cache_entry *ce;
4061 int namelen = strlen(path);
4062 unsigned ce_size = cache_entry_size(namelen);
4064 if (!update_index)
4065 return;
4067 ce = xcalloc(1, ce_size);
4068 memcpy(ce->name, path, namelen);
4069 ce->ce_mode = create_ce_mode(mode);
4070 ce->ce_flags = create_ce_flags(0);
4071 ce->ce_namelen = namelen;
4072 if (S_ISGITLINK(mode)) {
4073 const char *s;
4075 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4076 get_sha1_hex(s, ce->sha1))
4077 die(_("corrupt patch for submodule %s"), path);
4078 } else {
4079 if (!cached) {
4080 if (lstat(path, &st) < 0)
4081 die_errno(_("unable to stat newly created file '%s'"),
4082 path);
4083 fill_stat_cache_info(ce, &st);
4085 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
4086 die(_("unable to create backing store for newly created file %s"), path);
4088 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4089 die(_("unable to add cache entry for %s"), path);
4092 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4094 int fd;
4095 struct strbuf nbuf = STRBUF_INIT;
4097 if (S_ISGITLINK(mode)) {
4098 struct stat st;
4099 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4100 return 0;
4101 return mkdir(path, 0777);
4104 if (has_symlinks && S_ISLNK(mode))
4105 /* Although buf:size is counted string, it also is NUL
4106 * terminated.
4108 return symlink(buf, path);
4110 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4111 if (fd < 0)
4112 return -1;
4114 if (convert_to_working_tree(path, buf, size, &nbuf)) {
4115 size = nbuf.len;
4116 buf = nbuf.buf;
4118 write_or_die(fd, buf, size);
4119 strbuf_release(&nbuf);
4121 if (close(fd) < 0)
4122 die_errno(_("closing file '%s'"), path);
4123 return 0;
4127 * We optimistically assume that the directories exist,
4128 * which is true 99% of the time anyway. If they don't,
4129 * we create them and try again.
4131 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
4133 if (cached)
4134 return;
4135 if (!try_create_file(path, mode, buf, size))
4136 return;
4138 if (errno == ENOENT) {
4139 if (safe_create_leading_directories(path))
4140 return;
4141 if (!try_create_file(path, mode, buf, size))
4142 return;
4145 if (errno == EEXIST || errno == EACCES) {
4146 /* We may be trying to create a file where a directory
4147 * used to be.
4149 struct stat st;
4150 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4151 errno = EEXIST;
4154 if (errno == EEXIST) {
4155 unsigned int nr = getpid();
4157 for (;;) {
4158 char newpath[PATH_MAX];
4159 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4160 if (!try_create_file(newpath, mode, buf, size)) {
4161 if (!rename(newpath, path))
4162 return;
4163 unlink_or_warn(newpath);
4164 break;
4166 if (errno != EEXIST)
4167 break;
4168 ++nr;
4171 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4174 static void add_conflicted_stages_file(struct patch *patch)
4176 int stage, namelen;
4177 unsigned ce_size, mode;
4178 struct cache_entry *ce;
4180 if (!update_index)
4181 return;
4182 namelen = strlen(patch->new_name);
4183 ce_size = cache_entry_size(namelen);
4184 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4186 remove_file_from_cache(patch->new_name);
4187 for (stage = 1; stage < 4; stage++) {
4188 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4189 continue;
4190 ce = xcalloc(1, ce_size);
4191 memcpy(ce->name, patch->new_name, namelen);
4192 ce->ce_mode = create_ce_mode(mode);
4193 ce->ce_flags = create_ce_flags(stage);
4194 ce->ce_namelen = namelen;
4195 hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);
4196 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4197 die(_("unable to add cache entry for %s"), patch->new_name);
4201 static void create_file(struct patch *patch)
4203 char *path = patch->new_name;
4204 unsigned mode = patch->new_mode;
4205 unsigned long size = patch->resultsize;
4206 char *buf = patch->result;
4208 if (!mode)
4209 mode = S_IFREG | 0644;
4210 create_one_file(path, mode, buf, size);
4212 if (patch->conflicted_threeway)
4213 add_conflicted_stages_file(patch);
4214 else
4215 add_index_file(path, mode, buf, size);
4218 /* phase zero is to remove, phase one is to create */
4219 static void write_out_one_result(struct patch *patch, int phase)
4221 if (patch->is_delete > 0) {
4222 if (phase == 0)
4223 remove_file(patch, 1);
4224 return;
4226 if (patch->is_new > 0 || patch->is_copy) {
4227 if (phase == 1)
4228 create_file(patch);
4229 return;
4232 * Rename or modification boils down to the same
4233 * thing: remove the old, write the new
4235 if (phase == 0)
4236 remove_file(patch, patch->is_rename);
4237 if (phase == 1)
4238 create_file(patch);
4241 static int write_out_one_reject(struct patch *patch)
4243 FILE *rej;
4244 char namebuf[PATH_MAX];
4245 struct fragment *frag;
4246 int cnt = 0;
4247 struct strbuf sb = STRBUF_INIT;
4249 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4250 if (!frag->rejected)
4251 continue;
4252 cnt++;
4255 if (!cnt) {
4256 if (apply_verbosely)
4257 say_patch_name(stderr,
4258 _("Applied patch %s cleanly."), patch);
4259 return 0;
4262 /* This should not happen, because a removal patch that leaves
4263 * contents are marked "rejected" at the patch level.
4265 if (!patch->new_name)
4266 die(_("internal error"));
4268 /* Say this even without --verbose */
4269 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4270 "Applying patch %%s with %d rejects...",
4271 cnt),
4272 cnt);
4273 say_patch_name(stderr, sb.buf, patch);
4274 strbuf_release(&sb);
4276 cnt = strlen(patch->new_name);
4277 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4278 cnt = ARRAY_SIZE(namebuf) - 5;
4279 warning(_("truncating .rej filename to %.*s.rej"),
4280 cnt - 1, patch->new_name);
4282 memcpy(namebuf, patch->new_name, cnt);
4283 memcpy(namebuf + cnt, ".rej", 5);
4285 rej = fopen(namebuf, "w");
4286 if (!rej)
4287 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4289 /* Normal git tools never deal with .rej, so do not pretend
4290 * this is a git patch by saying --git or giving extended
4291 * headers. While at it, maybe please "kompare" that wants
4292 * the trailing TAB and some garbage at the end of line ;-).
4294 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4295 patch->new_name, patch->new_name);
4296 for (cnt = 1, frag = patch->fragments;
4297 frag;
4298 cnt++, frag = frag->next) {
4299 if (!frag->rejected) {
4300 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4301 continue;
4303 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4304 fprintf(rej, "%.*s", frag->size, frag->patch);
4305 if (frag->patch[frag->size-1] != '\n')
4306 fputc('\n', rej);
4308 fclose(rej);
4309 return -1;
4312 static int write_out_results(struct patch *list)
4314 int phase;
4315 int errs = 0;
4316 struct patch *l;
4317 struct string_list cpath = STRING_LIST_INIT_DUP;
4319 for (phase = 0; phase < 2; phase++) {
4320 l = list;
4321 while (l) {
4322 if (l->rejected)
4323 errs = 1;
4324 else {
4325 write_out_one_result(l, phase);
4326 if (phase == 1) {
4327 if (write_out_one_reject(l))
4328 errs = 1;
4329 if (l->conflicted_threeway) {
4330 string_list_append(&cpath, l->new_name);
4331 errs = 1;
4335 l = l->next;
4339 if (cpath.nr) {
4340 struct string_list_item *item;
4342 string_list_sort(&cpath);
4343 for_each_string_list_item(item, &cpath)
4344 fprintf(stderr, "U %s\n", item->string);
4345 string_list_clear(&cpath, 0);
4347 rerere(0);
4350 return errs;
4353 static struct lock_file lock_file;
4355 #define INACCURATE_EOF (1<<0)
4356 #define RECOUNT (1<<1)
4358 static int apply_patch(int fd, const char *filename, int options)
4360 size_t offset;
4361 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4362 struct patch *list = NULL, **listp = &list;
4363 int skipped_patch = 0;
4365 patch_input_file = filename;
4366 read_patch_file(&buf, fd);
4367 offset = 0;
4368 while (offset < buf.len) {
4369 struct patch *patch;
4370 int nr;
4372 patch = xcalloc(1, sizeof(*patch));
4373 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4374 patch->recount = !!(options & RECOUNT);
4375 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4376 if (nr < 0)
4377 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 free_patch(patch);
4387 skipped_patch++;
4389 offset += nr;
4392 if (!list && !skipped_patch)
4393 die(_("unrecognized input"));
4395 if (whitespace_error && (ws_error_action == die_on_ws_error))
4396 apply = 0;
4398 update_index = check_index && apply;
4399 if (update_index && newfd < 0)
4400 newfd = hold_locked_index(&lock_file, 1);
4402 if (check_index) {
4403 if (read_cache() < 0)
4404 die(_("unable to read index file"));
4407 if ((check || apply) &&
4408 check_patch_list(list) < 0 &&
4409 !apply_with_reject)
4410 exit(1);
4412 if (apply && write_out_results(list)) {
4413 if (apply_with_reject)
4414 exit(1);
4415 /* with --3way, we still need to write the index out */
4416 return 1;
4419 if (fake_ancestor)
4420 build_fake_ancestor(list, fake_ancestor);
4422 if (diffstat)
4423 stat_patch_list(list);
4425 if (numstat)
4426 numstat_patch_list(list);
4428 if (summary)
4429 summary_patch_list(list);
4431 free_patch_list(list);
4432 strbuf_release(&buf);
4433 string_list_clear(&fn_table, 0);
4434 return 0;
4437 static void git_apply_config(void)
4439 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4440 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4441 git_config(git_default_config, NULL);
4444 static int option_parse_exclude(const struct option *opt,
4445 const char *arg, int unset)
4447 add_name_limit(arg, 1);
4448 return 0;
4451 static int option_parse_include(const struct option *opt,
4452 const char *arg, int unset)
4454 add_name_limit(arg, 0);
4455 has_include = 1;
4456 return 0;
4459 static int option_parse_p(const struct option *opt,
4460 const char *arg, int unset)
4462 p_value = atoi(arg);
4463 p_value_known = 1;
4464 return 0;
4467 static int option_parse_z(const struct option *opt,
4468 const char *arg, int unset)
4470 if (unset)
4471 line_termination = '\n';
4472 else
4473 line_termination = 0;
4474 return 0;
4477 static int option_parse_space_change(const struct option *opt,
4478 const char *arg, int unset)
4480 if (unset)
4481 ws_ignore_action = ignore_ws_none;
4482 else
4483 ws_ignore_action = ignore_ws_change;
4484 return 0;
4487 static int option_parse_whitespace(const struct option *opt,
4488 const char *arg, int unset)
4490 const char **whitespace_option = opt->value;
4492 *whitespace_option = arg;
4493 parse_whitespace_option(arg);
4494 return 0;
4497 static int option_parse_directory(const struct option *opt,
4498 const char *arg, int unset)
4500 strbuf_reset(&root);
4501 strbuf_addstr(&root, arg);
4502 strbuf_complete(&root, '/');
4503 return 0;
4506 int cmd_apply(int argc, const char **argv, const char *prefix_)
4508 int i;
4509 int errs = 0;
4510 int is_not_gitdir = !startup_info->have_repository;
4511 int force_apply = 0;
4513 const char *whitespace_option = NULL;
4515 struct option builtin_apply_options[] = {
4516 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4517 N_("don't apply changes matching the given path"),
4518 0, option_parse_exclude },
4519 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4520 N_("apply changes matching the given path"),
4521 0, option_parse_include },
4522 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4523 N_("remove <num> leading slashes from traditional diff paths"),
4524 0, option_parse_p },
4525 OPT_BOOL(0, "no-add", &no_add,
4526 N_("ignore additions made by the patch")),
4527 OPT_BOOL(0, "stat", &diffstat,
4528 N_("instead of applying the patch, output diffstat for the input")),
4529 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4530 OPT_NOOP_NOARG(0, "binary"),
4531 OPT_BOOL(0, "numstat", &numstat,
4532 N_("show number of added and deleted lines in decimal notation")),
4533 OPT_BOOL(0, "summary", &summary,
4534 N_("instead of applying the patch, output a summary for the input")),
4535 OPT_BOOL(0, "check", &check,
4536 N_("instead of applying the patch, see if the patch is applicable")),
4537 OPT_BOOL(0, "index", &check_index,
4538 N_("make sure the patch is applicable to the current index")),
4539 OPT_BOOL(0, "cached", &cached,
4540 N_("apply a patch without touching the working tree")),
4541 OPT_BOOL(0, "unsafe-paths", &unsafe_paths,
4542 N_("accept a patch that touches outside the working area")),
4543 OPT_BOOL(0, "apply", &force_apply,
4544 N_("also apply the patch (use with --stat/--summary/--check)")),
4545 OPT_BOOL('3', "3way", &threeway,
4546 N_( "attempt three-way merge if a patch does not apply")),
4547 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4548 N_("build a temporary index based on embedded index information")),
4549 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4550 N_("paths are separated with NUL character"),
4551 PARSE_OPT_NOARG, option_parse_z },
4552 OPT_INTEGER('C', NULL, &p_context,
4553 N_("ensure at least <n> lines of context match")),
4554 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4555 N_("detect new or modified lines that have whitespace errors"),
4556 0, option_parse_whitespace },
4557 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4558 N_("ignore changes in whitespace when finding context"),
4559 PARSE_OPT_NOARG, option_parse_space_change },
4560 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4561 N_("ignore changes in whitespace when finding context"),
4562 PARSE_OPT_NOARG, option_parse_space_change },
4563 OPT_BOOL('R', "reverse", &apply_in_reverse,
4564 N_("apply the patch in reverse")),
4565 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4566 N_("don't expect at least one line of context")),
4567 OPT_BOOL(0, "reject", &apply_with_reject,
4568 N_("leave the rejected hunks in corresponding *.rej files")),
4569 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4570 N_("allow overlapping hunks")),
4571 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4572 OPT_BIT(0, "inaccurate-eof", &options,
4573 N_("tolerate incorrectly detected missing new-line at the end of file"),
4574 INACCURATE_EOF),
4575 OPT_BIT(0, "recount", &options,
4576 N_("do not trust the line counts in the hunk headers"),
4577 RECOUNT),
4578 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4579 N_("prepend <root> to all filenames"),
4580 0, option_parse_directory },
4581 OPT_END()
4584 prefix = prefix_;
4585 prefix_length = prefix ? strlen(prefix) : 0;
4586 git_apply_config();
4587 if (apply_default_whitespace)
4588 parse_whitespace_option(apply_default_whitespace);
4589 if (apply_default_ignorewhitespace)
4590 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4592 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4593 apply_usage, 0);
4595 if (apply_with_reject && threeway)
4596 die("--reject and --3way cannot be used together.");
4597 if (cached && threeway)
4598 die("--cached and --3way cannot be used together.");
4599 if (threeway) {
4600 if (is_not_gitdir)
4601 die(_("--3way outside a repository"));
4602 check_index = 1;
4604 if (apply_with_reject)
4605 apply = apply_verbosely = 1;
4606 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4607 apply = 0;
4608 if (check_index && is_not_gitdir)
4609 die(_("--index outside a repository"));
4610 if (cached) {
4611 if (is_not_gitdir)
4612 die(_("--cached outside a repository"));
4613 check_index = 1;
4615 if (check_index)
4616 unsafe_paths = 0;
4618 for (i = 0; i < argc; i++) {
4619 const char *arg = argv[i];
4620 int fd;
4622 if (!strcmp(arg, "-")) {
4623 errs |= apply_patch(0, "<stdin>", options);
4624 read_stdin = 0;
4625 continue;
4626 } else if (0 < prefix_length)
4627 arg = prefix_filename(prefix, prefix_length, arg);
4629 fd = open(arg, O_RDONLY);
4630 if (fd < 0)
4631 die_errno(_("can't open patch '%s'"), arg);
4632 read_stdin = 0;
4633 set_default_whitespace_mode(whitespace_option);
4634 errs |= apply_patch(fd, arg, options);
4635 close(fd);
4637 set_default_whitespace_mode(whitespace_option);
4638 if (read_stdin)
4639 errs |= apply_patch(0, "<stdin>", options);
4640 if (whitespace_error) {
4641 if (squelch_whitespace_errors &&
4642 squelch_whitespace_errors < whitespace_error) {
4643 int squelched =
4644 whitespace_error - squelch_whitespace_errors;
4645 warning(Q_("squelched %d whitespace error",
4646 "squelched %d whitespace errors",
4647 squelched),
4648 squelched);
4650 if (ws_error_action == die_on_ws_error)
4651 die(Q_("%d line adds whitespace errors.",
4652 "%d lines add whitespace errors.",
4653 whitespace_error),
4654 whitespace_error);
4655 if (applied_after_fixing_ws && apply)
4656 warning("%d line%s applied after"
4657 " fixing whitespace errors.",
4658 applied_after_fixing_ws,
4659 applied_after_fixing_ws == 1 ? "" : "s");
4660 else if (whitespace_error)
4661 warning(Q_("%d line adds whitespace errors.",
4662 "%d lines add whitespace errors.",
4663 whitespace_error),
4664 whitespace_error);
4667 if (update_index) {
4668 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4669 die(_("Unable to write new index file"));
4672 return !!errs;