apply: free patch->{def,old,new}_name fields
[git/mjg.git] / builtin / apply.c
blob04cd8d38b02e2f4771108141da1831220523387b
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "parse-options.h"
20 * --check turns on checking that the working tree matches the
21 * files that are being modified, but doesn't apply the patch
22 * --stat does just a diffstat, and doesn't actually apply
23 * --numstat does numeric diffstat, and doesn't actually apply
24 * --index-info shows the old and new index info for paths if available.
25 * --index updates the cache as well.
26 * --cached updates only the cache without ever touching the working tree.
28 static const char *prefix;
29 static int prefix_length = -1;
30 static int newfd = -1;
32 static int unidiff_zero;
33 static int p_value = 1;
34 static int p_value_known;
35 static int check_index;
36 static int update_index;
37 static int cached;
38 static int diffstat;
39 static int numstat;
40 static int summary;
41 static int check;
42 static int apply = 1;
43 static int apply_in_reverse;
44 static int apply_with_reject;
45 static int apply_verbosely;
46 static int allow_overlap;
47 static int no_add;
48 static const char *fake_ancestor;
49 static int line_termination = '\n';
50 static unsigned int p_context = UINT_MAX;
51 static const char * const apply_usage[] = {
52 "git apply [options] [<patch>...]",
53 NULL
56 static enum ws_error_action {
57 nowarn_ws_error,
58 warn_on_ws_error,
59 die_on_ws_error,
60 correct_ws_error
61 } ws_error_action = warn_on_ws_error;
62 static int whitespace_error;
63 static int squelch_whitespace_errors = 5;
64 static int applied_after_fixing_ws;
66 static enum ws_ignore {
67 ignore_ws_none,
68 ignore_ws_change
69 } ws_ignore_action = ignore_ws_none;
72 static const char *patch_input_file;
73 static const char *root;
74 static int root_len;
75 static int read_stdin = 1;
76 static int options;
78 static void parse_whitespace_option(const char *option)
80 if (!option) {
81 ws_error_action = warn_on_ws_error;
82 return;
84 if (!strcmp(option, "warn")) {
85 ws_error_action = warn_on_ws_error;
86 return;
88 if (!strcmp(option, "nowarn")) {
89 ws_error_action = nowarn_ws_error;
90 return;
92 if (!strcmp(option, "error")) {
93 ws_error_action = die_on_ws_error;
94 return;
96 if (!strcmp(option, "error-all")) {
97 ws_error_action = die_on_ws_error;
98 squelch_whitespace_errors = 0;
99 return;
101 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
102 ws_error_action = correct_ws_error;
103 return;
105 die("unrecognized whitespace option '%s'", option);
108 static void parse_ignorewhitespace_option(const char *option)
110 if (!option || !strcmp(option, "no") ||
111 !strcmp(option, "false") || !strcmp(option, "never") ||
112 !strcmp(option, "none")) {
113 ws_ignore_action = ignore_ws_none;
114 return;
116 if (!strcmp(option, "change")) {
117 ws_ignore_action = ignore_ws_change;
118 return;
120 die("unrecognized whitespace ignore option '%s'", option);
123 static void set_default_whitespace_mode(const char *whitespace_option)
125 if (!whitespace_option && !apply_default_whitespace)
126 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
130 * For "diff-stat" like behaviour, we keep track of the biggest change
131 * we've seen, and the longest filename. That allows us to do simple
132 * scaling.
134 static int max_change, max_len;
137 * Various "current state", notably line numbers and what
138 * file (and how) we're patching right now.. The "is_xxxx"
139 * things are flags, where -1 means "don't know yet".
141 static int linenr = 1;
144 * This represents one "hunk" from a patch, starting with
145 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
146 * patch text is pointed at by patch, and its byte length
147 * is stored in size. leading and trailing are the number
148 * of context lines.
150 struct fragment {
151 unsigned long leading, trailing;
152 unsigned long oldpos, oldlines;
153 unsigned long newpos, newlines;
154 const char *patch;
155 unsigned free_patch:1,
156 rejected:1;
157 int size;
158 int linenr;
159 struct fragment *next;
163 * When dealing with a binary patch, we reuse "leading" field
164 * to store the type of the binary hunk, either deflated "delta"
165 * or deflated "literal".
167 #define binary_patch_method leading
168 #define BINARY_DELTA_DEFLATED 1
169 #define BINARY_LITERAL_DEFLATED 2
172 * This represents a "patch" to a file, both metainfo changes
173 * such as creation/deletion, filemode and content changes represented
174 * as a series of fragments.
176 struct patch {
177 char *new_name, *old_name, *def_name;
178 unsigned int old_mode, new_mode;
179 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
180 int rejected;
181 unsigned ws_rule;
182 unsigned long deflate_origlen;
183 int lines_added, lines_deleted;
184 int score;
185 unsigned int is_toplevel_relative:1;
186 unsigned int inaccurate_eof:1;
187 unsigned int is_binary:1;
188 unsigned int is_copy:1;
189 unsigned int is_rename:1;
190 unsigned int recount:1;
191 struct fragment *fragments;
192 char *result;
193 size_t resultsize;
194 char old_sha1_prefix[41];
195 char new_sha1_prefix[41];
196 struct patch *next;
199 static void free_patch(struct patch *patch)
201 struct fragment *fragment = patch->fragments;
203 while (fragment) {
204 struct fragment *fragment_next = fragment->next;
205 if (fragment->patch != NULL && fragment->free_patch)
206 free((char *)fragment->patch);
207 free(fragment);
208 fragment = fragment_next;
210 free(patch->def_name);
211 free(patch->old_name);
212 free(patch->new_name);
213 free(patch);
216 static void free_patch_list(struct patch *list)
218 while (list) {
219 struct patch *next = list->next;
220 free_patch(list);
221 list = next;
226 * A line in a file, len-bytes long (includes the terminating LF,
227 * except for an incomplete line at the end if the file ends with
228 * one), and its contents hashes to 'hash'.
230 struct line {
231 size_t len;
232 unsigned hash : 24;
233 unsigned flag : 8;
234 #define LINE_COMMON 1
235 #define LINE_PATCHED 2
239 * This represents a "file", which is an array of "lines".
241 struct image {
242 char *buf;
243 size_t len;
244 size_t nr;
245 size_t alloc;
246 struct line *line_allocated;
247 struct line *line;
251 * Records filenames that have been touched, in order to handle
252 * the case where more than one patches touch the same file.
255 static struct string_list fn_table;
257 static uint32_t hash_line(const char *cp, size_t len)
259 size_t i;
260 uint32_t h;
261 for (i = 0, h = 0; i < len; i++) {
262 if (!isspace(cp[i])) {
263 h = h * 3 + (cp[i] & 0xff);
266 return h;
270 * Compare lines s1 of length n1 and s2 of length n2, ignoring
271 * whitespace difference. Returns 1 if they match, 0 otherwise
273 static int fuzzy_matchlines(const char *s1, size_t n1,
274 const char *s2, size_t n2)
276 const char *last1 = s1 + n1 - 1;
277 const char *last2 = s2 + n2 - 1;
278 int result = 0;
280 /* ignore line endings */
281 while ((*last1 == '\r') || (*last1 == '\n'))
282 last1--;
283 while ((*last2 == '\r') || (*last2 == '\n'))
284 last2--;
286 /* skip leading whitespace */
287 while (isspace(*s1) && (s1 <= last1))
288 s1++;
289 while (isspace(*s2) && (s2 <= last2))
290 s2++;
291 /* early return if both lines are empty */
292 if ((s1 > last1) && (s2 > last2))
293 return 1;
294 while (!result) {
295 result = *s1++ - *s2++;
297 * Skip whitespace inside. We check for whitespace on
298 * both buffers because we don't want "a b" to match
299 * "ab"
301 if (isspace(*s1) && isspace(*s2)) {
302 while (isspace(*s1) && s1 <= last1)
303 s1++;
304 while (isspace(*s2) && s2 <= last2)
305 s2++;
308 * If we reached the end on one side only,
309 * lines don't match
311 if (
312 ((s2 > last2) && (s1 <= last1)) ||
313 ((s1 > last1) && (s2 <= last2)))
314 return 0;
315 if ((s1 > last1) && (s2 > last2))
316 break;
319 return !result;
322 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
324 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
325 img->line_allocated[img->nr].len = len;
326 img->line_allocated[img->nr].hash = hash_line(bol, len);
327 img->line_allocated[img->nr].flag = flag;
328 img->nr++;
331 static void prepare_image(struct image *image, char *buf, size_t len,
332 int prepare_linetable)
334 const char *cp, *ep;
336 memset(image, 0, sizeof(*image));
337 image->buf = buf;
338 image->len = len;
340 if (!prepare_linetable)
341 return;
343 ep = image->buf + image->len;
344 cp = image->buf;
345 while (cp < ep) {
346 const char *next;
347 for (next = cp; next < ep && *next != '\n'; next++)
349 if (next < ep)
350 next++;
351 add_line_info(image, cp, next - cp, 0);
352 cp = next;
354 image->line = image->line_allocated;
357 static void clear_image(struct image *image)
359 free(image->buf);
360 image->buf = NULL;
361 image->len = 0;
364 static void say_patch_name(FILE *output, const char *pre,
365 struct patch *patch, const char *post)
367 fputs(pre, output);
368 if (patch->old_name && patch->new_name &&
369 strcmp(patch->old_name, patch->new_name)) {
370 quote_c_style(patch->old_name, NULL, output, 0);
371 fputs(" => ", output);
372 quote_c_style(patch->new_name, NULL, output, 0);
373 } else {
374 const char *n = patch->new_name;
375 if (!n)
376 n = patch->old_name;
377 quote_c_style(n, NULL, output, 0);
379 fputs(post, output);
382 #define CHUNKSIZE (8192)
383 #define SLOP (16)
385 static void read_patch_file(struct strbuf *sb, int fd)
387 if (strbuf_read(sb, fd, 0) < 0)
388 die_errno("git apply: failed to read");
391 * Make sure that we have some slop in the buffer
392 * so that we can do speculative "memcmp" etc, and
393 * see to it that it is NUL-filled.
395 strbuf_grow(sb, SLOP);
396 memset(sb->buf + sb->len, 0, SLOP);
399 static unsigned long linelen(const char *buffer, unsigned long size)
401 unsigned long len = 0;
402 while (size--) {
403 len++;
404 if (*buffer++ == '\n')
405 break;
407 return len;
410 static int is_dev_null(const char *str)
412 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
415 #define TERM_SPACE 1
416 #define TERM_TAB 2
418 static int name_terminate(const char *name, int namelen, int c, int terminate)
420 if (c == ' ' && !(terminate & TERM_SPACE))
421 return 0;
422 if (c == '\t' && !(terminate & TERM_TAB))
423 return 0;
425 return 1;
428 /* remove double slashes to make --index work with such filenames */
429 static char *squash_slash(char *name)
431 int i = 0, j = 0;
433 if (!name)
434 return NULL;
436 while (name[i]) {
437 if ((name[j++] = name[i++]) == '/')
438 while (name[i] == '/')
439 i++;
441 name[j] = '\0';
442 return name;
445 static char *find_name_gnu(const char *line, const char *def, int p_value)
447 struct strbuf name = STRBUF_INIT;
448 char *cp;
451 * Proposed "new-style" GNU patch/diff format; see
452 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
454 if (unquote_c_style(&name, line, NULL)) {
455 strbuf_release(&name);
456 return NULL;
459 for (cp = name.buf; p_value; p_value--) {
460 cp = strchr(cp, '/');
461 if (!cp) {
462 strbuf_release(&name);
463 return NULL;
465 cp++;
468 strbuf_remove(&name, 0, cp - name.buf);
469 if (root)
470 strbuf_insert(&name, 0, root, root_len);
471 return squash_slash(strbuf_detach(&name, NULL));
474 static size_t sane_tz_len(const char *line, size_t len)
476 const char *tz, *p;
478 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
479 return 0;
480 tz = line + len - strlen(" +0500");
482 if (tz[1] != '+' && tz[1] != '-')
483 return 0;
485 for (p = tz + 2; p != line + len; p++)
486 if (!isdigit(*p))
487 return 0;
489 return line + len - tz;
492 static size_t tz_with_colon_len(const char *line, size_t len)
494 const char *tz, *p;
496 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
497 return 0;
498 tz = line + len - strlen(" +08:00");
500 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
501 return 0;
502 p = tz + 2;
503 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
504 !isdigit(*p++) || !isdigit(*p++))
505 return 0;
507 return line + len - tz;
510 static size_t date_len(const char *line, size_t len)
512 const char *date, *p;
514 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
515 return 0;
516 p = date = line + len - strlen("72-02-05");
518 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
519 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
520 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
521 return 0;
523 if (date - line >= strlen("19") &&
524 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
525 date -= strlen("19");
527 return line + len - date;
530 static size_t short_time_len(const char *line, size_t len)
532 const char *time, *p;
534 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
535 return 0;
536 p = time = line + len - strlen(" 07:01:32");
538 /* Permit 1-digit hours? */
539 if (*p++ != ' ' ||
540 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
541 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
542 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
543 return 0;
545 return line + len - time;
548 static size_t fractional_time_len(const char *line, size_t len)
550 const char *p;
551 size_t n;
553 /* Expected format: 19:41:17.620000023 */
554 if (!len || !isdigit(line[len - 1]))
555 return 0;
556 p = line + len - 1;
558 /* Fractional seconds. */
559 while (p > line && isdigit(*p))
560 p--;
561 if (*p != '.')
562 return 0;
564 /* Hours, minutes, and whole seconds. */
565 n = short_time_len(line, p - line);
566 if (!n)
567 return 0;
569 return line + len - p + n;
572 static size_t trailing_spaces_len(const char *line, size_t len)
574 const char *p;
576 /* Expected format: ' ' x (1 or more) */
577 if (!len || line[len - 1] != ' ')
578 return 0;
580 p = line + len;
581 while (p != line) {
582 p--;
583 if (*p != ' ')
584 return line + len - (p + 1);
587 /* All spaces! */
588 return len;
591 static size_t diff_timestamp_len(const char *line, size_t len)
593 const char *end = line + len;
594 size_t n;
597 * Posix: 2010-07-05 19:41:17
598 * GNU: 2010-07-05 19:41:17.620000023 -0500
601 if (!isdigit(end[-1]))
602 return 0;
604 n = sane_tz_len(line, end - line);
605 if (!n)
606 n = tz_with_colon_len(line, end - line);
607 end -= n;
609 n = short_time_len(line, end - line);
610 if (!n)
611 n = fractional_time_len(line, end - line);
612 end -= n;
614 n = date_len(line, end - line);
615 if (!n) /* No date. Too bad. */
616 return 0;
617 end -= n;
619 if (end == line) /* No space before date. */
620 return 0;
621 if (end[-1] == '\t') { /* Success! */
622 end--;
623 return line + len - end;
625 if (end[-1] != ' ') /* No space before date. */
626 return 0;
628 /* Whitespace damage. */
629 end -= trailing_spaces_len(line, end - line);
630 return line + len - end;
633 static char *null_strdup(const char *s)
635 return s ? xstrdup(s) : NULL;
638 static char *find_name_common(const char *line, const char *def,
639 int p_value, const char *end, int terminate)
641 int len;
642 const char *start = NULL;
644 if (p_value == 0)
645 start = line;
646 while (line != end) {
647 char c = *line;
649 if (!end && isspace(c)) {
650 if (c == '\n')
651 break;
652 if (name_terminate(start, line-start, c, terminate))
653 break;
655 line++;
656 if (c == '/' && !--p_value)
657 start = line;
659 if (!start)
660 return squash_slash(null_strdup(def));
661 len = line - start;
662 if (!len)
663 return squash_slash(null_strdup(def));
666 * Generally we prefer the shorter name, especially
667 * if the other one is just a variation of that with
668 * something else tacked on to the end (ie "file.orig"
669 * or "file~").
671 if (def) {
672 int deflen = strlen(def);
673 if (deflen < len && !strncmp(start, def, deflen))
674 return squash_slash(xstrdup(def));
677 if (root) {
678 char *ret = xmalloc(root_len + len + 1);
679 strcpy(ret, root);
680 memcpy(ret + root_len, start, len);
681 ret[root_len + len] = '\0';
682 return squash_slash(ret);
685 return squash_slash(xmemdupz(start, len));
688 static char *find_name(const char *line, char *def, int p_value, int terminate)
690 if (*line == '"') {
691 char *name = find_name_gnu(line, def, p_value);
692 if (name)
693 return name;
696 return find_name_common(line, def, p_value, NULL, terminate);
699 static char *find_name_traditional(const char *line, char *def, int p_value)
701 size_t len = strlen(line);
702 size_t date_len;
704 if (*line == '"') {
705 char *name = find_name_gnu(line, def, p_value);
706 if (name)
707 return name;
710 len = strchrnul(line, '\n') - line;
711 date_len = diff_timestamp_len(line, len);
712 if (!date_len)
713 return find_name_common(line, def, p_value, NULL, TERM_TAB);
714 len -= date_len;
716 return find_name_common(line, def, p_value, line + len, 0);
719 static int count_slashes(const char *cp)
721 int cnt = 0;
722 char ch;
724 while ((ch = *cp++))
725 if (ch == '/')
726 cnt++;
727 return cnt;
731 * Given the string after "--- " or "+++ ", guess the appropriate
732 * p_value for the given patch.
734 static int guess_p_value(const char *nameline)
736 char *name, *cp;
737 int val = -1;
739 if (is_dev_null(nameline))
740 return -1;
741 name = find_name_traditional(nameline, NULL, 0);
742 if (!name)
743 return -1;
744 cp = strchr(name, '/');
745 if (!cp)
746 val = 0;
747 else if (prefix) {
749 * Does it begin with "a/$our-prefix" and such? Then this is
750 * very likely to apply to our directory.
752 if (!strncmp(name, prefix, prefix_length))
753 val = count_slashes(prefix);
754 else {
755 cp++;
756 if (!strncmp(cp, prefix, prefix_length))
757 val = count_slashes(prefix) + 1;
760 free(name);
761 return val;
765 * Does the ---/+++ line has the POSIX timestamp after the last HT?
766 * GNU diff puts epoch there to signal a creation/deletion event. Is
767 * this such a timestamp?
769 static int has_epoch_timestamp(const char *nameline)
772 * We are only interested in epoch timestamp; any non-zero
773 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
774 * For the same reason, the date must be either 1969-12-31 or
775 * 1970-01-01, and the seconds part must be "00".
777 const char stamp_regexp[] =
778 "^(1969-12-31|1970-01-01)"
780 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
782 "([-+][0-2][0-9]:?[0-5][0-9])\n";
783 const char *timestamp = NULL, *cp, *colon;
784 static regex_t *stamp;
785 regmatch_t m[10];
786 int zoneoffset;
787 int hourminute;
788 int status;
790 for (cp = nameline; *cp != '\n'; cp++) {
791 if (*cp == '\t')
792 timestamp = cp + 1;
794 if (!timestamp)
795 return 0;
796 if (!stamp) {
797 stamp = xmalloc(sizeof(*stamp));
798 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
799 warning("Cannot prepare timestamp regexp %s",
800 stamp_regexp);
801 return 0;
805 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
806 if (status) {
807 if (status != REG_NOMATCH)
808 warning("regexec returned %d for input: %s",
809 status, timestamp);
810 return 0;
813 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
814 if (*colon == ':')
815 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
816 else
817 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
818 if (timestamp[m[3].rm_so] == '-')
819 zoneoffset = -zoneoffset;
822 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
823 * (west of GMT) or 1970-01-01 (east of GMT)
825 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
826 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
827 return 0;
829 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
830 strtol(timestamp + 14, NULL, 10) -
831 zoneoffset);
833 return ((zoneoffset < 0 && hourminute == 1440) ||
834 (0 <= zoneoffset && !hourminute));
838 * Get the name etc info from the ---/+++ lines of a traditional patch header
840 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
841 * files, we can happily check the index for a match, but for creating a
842 * new file we should try to match whatever "patch" does. I have no idea.
844 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
846 char *name;
848 first += 4; /* skip "--- " */
849 second += 4; /* skip "+++ " */
850 if (!p_value_known) {
851 int p, q;
852 p = guess_p_value(first);
853 q = guess_p_value(second);
854 if (p < 0) p = q;
855 if (0 <= p && p == q) {
856 p_value = p;
857 p_value_known = 1;
860 if (is_dev_null(first)) {
861 patch->is_new = 1;
862 patch->is_delete = 0;
863 name = find_name_traditional(second, NULL, p_value);
864 patch->new_name = name;
865 } else if (is_dev_null(second)) {
866 patch->is_new = 0;
867 patch->is_delete = 1;
868 name = find_name_traditional(first, NULL, p_value);
869 patch->old_name = name;
870 } else {
871 char *first_name;
872 first_name = find_name_traditional(first, NULL, p_value);
873 name = find_name_traditional(second, first_name, p_value);
874 free(first_name);
875 if (has_epoch_timestamp(first)) {
876 patch->is_new = 1;
877 patch->is_delete = 0;
878 patch->new_name = name;
879 } else if (has_epoch_timestamp(second)) {
880 patch->is_new = 0;
881 patch->is_delete = 1;
882 patch->old_name = name;
883 } else {
884 patch->old_name = name;
885 patch->new_name = xstrdup(name);
888 if (!name)
889 die("unable to find filename in patch at line %d", linenr);
892 static int gitdiff_hdrend(const char *line, struct patch *patch)
894 return -1;
898 * We're anal about diff header consistency, to make
899 * sure that we don't end up having strange ambiguous
900 * patches floating around.
902 * As a result, gitdiff_{old|new}name() will check
903 * their names against any previous information, just
904 * to make sure..
906 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
908 if (!orig_name && !isnull)
909 return find_name(line, NULL, p_value, TERM_TAB);
911 if (orig_name) {
912 int len;
913 const char *name;
914 char *another;
915 name = orig_name;
916 len = strlen(name);
917 if (isnull)
918 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
919 another = find_name(line, NULL, p_value, TERM_TAB);
920 if (!another || memcmp(another, name, len + 1))
921 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
922 free(another);
923 return orig_name;
925 else {
926 /* expect "/dev/null" */
927 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
928 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
929 return NULL;
933 static int gitdiff_oldname(const char *line, struct patch *patch)
935 char *orig = patch->old_name;
936 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
937 if (orig != patch->old_name)
938 free(orig);
939 return 0;
942 static int gitdiff_newname(const char *line, struct patch *patch)
944 char *orig = patch->new_name;
945 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
946 if (orig != patch->new_name)
947 free(orig);
948 return 0;
951 static int gitdiff_oldmode(const char *line, struct patch *patch)
953 patch->old_mode = strtoul(line, NULL, 8);
954 return 0;
957 static int gitdiff_newmode(const char *line, struct patch *patch)
959 patch->new_mode = strtoul(line, NULL, 8);
960 return 0;
963 static int gitdiff_delete(const char *line, struct patch *patch)
965 patch->is_delete = 1;
966 free(patch->old_name);
967 patch->old_name = null_strdup(patch->def_name);
968 return gitdiff_oldmode(line, patch);
971 static int gitdiff_newfile(const char *line, struct patch *patch)
973 patch->is_new = 1;
974 free(patch->new_name);
975 patch->new_name = null_strdup(patch->def_name);
976 return gitdiff_newmode(line, patch);
979 static int gitdiff_copysrc(const char *line, struct patch *patch)
981 patch->is_copy = 1;
982 free(patch->old_name);
983 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
984 return 0;
987 static int gitdiff_copydst(const char *line, struct patch *patch)
989 patch->is_copy = 1;
990 free(patch->new_name);
991 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
992 return 0;
995 static int gitdiff_renamesrc(const char *line, struct patch *patch)
997 patch->is_rename = 1;
998 free(patch->old_name);
999 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1000 return 0;
1003 static int gitdiff_renamedst(const char *line, struct patch *patch)
1005 patch->is_rename = 1;
1006 free(patch->new_name);
1007 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1008 return 0;
1011 static int gitdiff_similarity(const char *line, struct patch *patch)
1013 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1014 patch->score = 0;
1015 return 0;
1018 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1020 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
1021 patch->score = 0;
1022 return 0;
1025 static int gitdiff_index(const char *line, struct patch *patch)
1028 * index line is N hexadecimal, "..", N hexadecimal,
1029 * and optional space with octal mode.
1031 const char *ptr, *eol;
1032 int len;
1034 ptr = strchr(line, '.');
1035 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1036 return 0;
1037 len = ptr - line;
1038 memcpy(patch->old_sha1_prefix, line, len);
1039 patch->old_sha1_prefix[len] = 0;
1041 line = ptr + 2;
1042 ptr = strchr(line, ' ');
1043 eol = strchr(line, '\n');
1045 if (!ptr || eol < ptr)
1046 ptr = eol;
1047 len = ptr - line;
1049 if (40 < len)
1050 return 0;
1051 memcpy(patch->new_sha1_prefix, line, len);
1052 patch->new_sha1_prefix[len] = 0;
1053 if (*ptr == ' ')
1054 patch->old_mode = strtoul(ptr+1, NULL, 8);
1055 return 0;
1059 * This is normal for a diff that doesn't change anything: we'll fall through
1060 * into the next diff. Tell the parser to break out.
1062 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1064 return -1;
1067 static const char *stop_at_slash(const char *line, int llen)
1069 int nslash = p_value;
1070 int i;
1072 for (i = 0; i < llen; i++) {
1073 int ch = line[i];
1074 if (ch == '/' && --nslash <= 0)
1075 return &line[i];
1077 return NULL;
1081 * This is to extract the same name that appears on "diff --git"
1082 * line. We do not find and return anything if it is a rename
1083 * patch, and it is OK because we will find the name elsewhere.
1084 * We need to reliably find name only when it is mode-change only,
1085 * creation or deletion of an empty file. In any of these cases,
1086 * both sides are the same name under a/ and b/ respectively.
1088 static char *git_header_name(char *line, int llen)
1090 const char *name;
1091 const char *second = NULL;
1092 size_t len, line_len;
1094 line += strlen("diff --git ");
1095 llen -= strlen("diff --git ");
1097 if (*line == '"') {
1098 const char *cp;
1099 struct strbuf first = STRBUF_INIT;
1100 struct strbuf sp = STRBUF_INIT;
1102 if (unquote_c_style(&first, line, &second))
1103 goto free_and_fail1;
1105 /* advance to the first slash */
1106 cp = stop_at_slash(first.buf, first.len);
1107 /* we do not accept absolute paths */
1108 if (!cp || cp == first.buf)
1109 goto free_and_fail1;
1110 strbuf_remove(&first, 0, cp + 1 - first.buf);
1113 * second points at one past closing dq of name.
1114 * find the second name.
1116 while ((second < line + llen) && isspace(*second))
1117 second++;
1119 if (line + llen <= second)
1120 goto free_and_fail1;
1121 if (*second == '"') {
1122 if (unquote_c_style(&sp, second, NULL))
1123 goto free_and_fail1;
1124 cp = stop_at_slash(sp.buf, sp.len);
1125 if (!cp || cp == sp.buf)
1126 goto free_and_fail1;
1127 /* They must match, otherwise ignore */
1128 if (strcmp(cp + 1, first.buf))
1129 goto free_and_fail1;
1130 strbuf_release(&sp);
1131 return strbuf_detach(&first, NULL);
1134 /* unquoted second */
1135 cp = stop_at_slash(second, line + llen - second);
1136 if (!cp || cp == second)
1137 goto free_and_fail1;
1138 cp++;
1139 if (line + llen - cp != first.len + 1 ||
1140 memcmp(first.buf, cp, first.len))
1141 goto free_and_fail1;
1142 return strbuf_detach(&first, NULL);
1144 free_and_fail1:
1145 strbuf_release(&first);
1146 strbuf_release(&sp);
1147 return NULL;
1150 /* unquoted first name */
1151 name = stop_at_slash(line, llen);
1152 if (!name || name == line)
1153 return NULL;
1154 name++;
1157 * since the first name is unquoted, a dq if exists must be
1158 * the beginning of the second name.
1160 for (second = name; second < line + llen; second++) {
1161 if (*second == '"') {
1162 struct strbuf sp = STRBUF_INIT;
1163 const char *np;
1165 if (unquote_c_style(&sp, second, NULL))
1166 goto free_and_fail2;
1168 np = stop_at_slash(sp.buf, sp.len);
1169 if (!np || np == sp.buf)
1170 goto free_and_fail2;
1171 np++;
1173 len = sp.buf + sp.len - np;
1174 if (len < second - name &&
1175 !strncmp(np, name, len) &&
1176 isspace(name[len])) {
1177 /* Good */
1178 strbuf_remove(&sp, 0, np - sp.buf);
1179 return strbuf_detach(&sp, NULL);
1182 free_and_fail2:
1183 strbuf_release(&sp);
1184 return NULL;
1189 * Accept a name only if it shows up twice, exactly the same
1190 * form.
1192 second = strchr(name, '\n');
1193 if (!second)
1194 return NULL;
1195 line_len = second - name;
1196 for (len = 0 ; ; len++) {
1197 switch (name[len]) {
1198 default:
1199 continue;
1200 case '\n':
1201 return NULL;
1202 case '\t': case ' ':
1203 second = stop_at_slash(name + len, line_len - len);
1204 if (!second)
1205 return NULL;
1206 second++;
1207 if (second[len] == '\n' && !strncmp(name, second, len)) {
1208 return xmemdupz(name, len);
1214 /* Verify that we recognize the lines following a git header */
1215 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
1217 unsigned long offset;
1219 /* A git diff has explicit new/delete information, so we don't guess */
1220 patch->is_new = 0;
1221 patch->is_delete = 0;
1224 * Some things may not have the old name in the
1225 * rest of the headers anywhere (pure mode changes,
1226 * or removing or adding empty files), so we get
1227 * the default name from the header.
1229 patch->def_name = git_header_name(line, len);
1230 if (patch->def_name && root) {
1231 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1232 strcpy(s, root);
1233 strcpy(s + root_len, patch->def_name);
1234 free(patch->def_name);
1235 patch->def_name = s;
1238 line += len;
1239 size -= len;
1240 linenr++;
1241 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1242 static const struct opentry {
1243 const char *str;
1244 int (*fn)(const char *, struct patch *);
1245 } optable[] = {
1246 { "@@ -", gitdiff_hdrend },
1247 { "--- ", gitdiff_oldname },
1248 { "+++ ", gitdiff_newname },
1249 { "old mode ", gitdiff_oldmode },
1250 { "new mode ", gitdiff_newmode },
1251 { "deleted file mode ", gitdiff_delete },
1252 { "new file mode ", gitdiff_newfile },
1253 { "copy from ", gitdiff_copysrc },
1254 { "copy to ", gitdiff_copydst },
1255 { "rename old ", gitdiff_renamesrc },
1256 { "rename new ", gitdiff_renamedst },
1257 { "rename from ", gitdiff_renamesrc },
1258 { "rename to ", gitdiff_renamedst },
1259 { "similarity index ", gitdiff_similarity },
1260 { "dissimilarity index ", gitdiff_dissimilarity },
1261 { "index ", gitdiff_index },
1262 { "", gitdiff_unrecognized },
1264 int i;
1266 len = linelen(line, size);
1267 if (!len || line[len-1] != '\n')
1268 break;
1269 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1270 const struct opentry *p = optable + i;
1271 int oplen = strlen(p->str);
1272 if (len < oplen || memcmp(p->str, line, oplen))
1273 continue;
1274 if (p->fn(line + oplen, patch) < 0)
1275 return offset;
1276 break;
1280 return offset;
1283 static int parse_num(const char *line, unsigned long *p)
1285 char *ptr;
1287 if (!isdigit(*line))
1288 return 0;
1289 *p = strtoul(line, &ptr, 10);
1290 return ptr - line;
1293 static int parse_range(const char *line, int len, int offset, const char *expect,
1294 unsigned long *p1, unsigned long *p2)
1296 int digits, ex;
1298 if (offset < 0 || offset >= len)
1299 return -1;
1300 line += offset;
1301 len -= offset;
1303 digits = parse_num(line, p1);
1304 if (!digits)
1305 return -1;
1307 offset += digits;
1308 line += digits;
1309 len -= digits;
1311 *p2 = 1;
1312 if (*line == ',') {
1313 digits = parse_num(line+1, p2);
1314 if (!digits)
1315 return -1;
1317 offset += digits+1;
1318 line += digits+1;
1319 len -= digits+1;
1322 ex = strlen(expect);
1323 if (ex > len)
1324 return -1;
1325 if (memcmp(line, expect, ex))
1326 return -1;
1328 return offset + ex;
1331 static void recount_diff(char *line, int size, struct fragment *fragment)
1333 int oldlines = 0, newlines = 0, ret = 0;
1335 if (size < 1) {
1336 warning("recount: ignore empty hunk");
1337 return;
1340 for (;;) {
1341 int len = linelen(line, size);
1342 size -= len;
1343 line += len;
1345 if (size < 1)
1346 break;
1348 switch (*line) {
1349 case ' ': case '\n':
1350 newlines++;
1351 /* fall through */
1352 case '-':
1353 oldlines++;
1354 continue;
1355 case '+':
1356 newlines++;
1357 continue;
1358 case '\\':
1359 continue;
1360 case '@':
1361 ret = size < 3 || prefixcmp(line, "@@ ");
1362 break;
1363 case 'd':
1364 ret = size < 5 || prefixcmp(line, "diff ");
1365 break;
1366 default:
1367 ret = -1;
1368 break;
1370 if (ret) {
1371 warning("recount: unexpected line: %.*s",
1372 (int)linelen(line, size), line);
1373 return;
1375 break;
1377 fragment->oldlines = oldlines;
1378 fragment->newlines = newlines;
1382 * Parse a unified diff fragment header of the
1383 * form "@@ -a,b +c,d @@"
1385 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1387 int offset;
1389 if (!len || line[len-1] != '\n')
1390 return -1;
1392 /* Figure out the number of lines in a fragment */
1393 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1394 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1396 return offset;
1399 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1401 unsigned long offset, len;
1403 patch->is_toplevel_relative = 0;
1404 patch->is_rename = patch->is_copy = 0;
1405 patch->is_new = patch->is_delete = -1;
1406 patch->old_mode = patch->new_mode = 0;
1407 patch->old_name = patch->new_name = NULL;
1408 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1409 unsigned long nextlen;
1411 len = linelen(line, size);
1412 if (!len)
1413 break;
1415 /* Testing this early allows us to take a few shortcuts.. */
1416 if (len < 6)
1417 continue;
1420 * Make sure we don't find any unconnected patch fragments.
1421 * That's a sign that we didn't find a header, and that a
1422 * patch has become corrupted/broken up.
1424 if (!memcmp("@@ -", line, 4)) {
1425 struct fragment dummy;
1426 if (parse_fragment_header(line, len, &dummy) < 0)
1427 continue;
1428 die("patch fragment without header at line %d: %.*s",
1429 linenr, (int)len-1, line);
1432 if (size < len + 6)
1433 break;
1436 * Git patch? It might not have a real patch, just a rename
1437 * or mode change, so we handle that specially
1439 if (!memcmp("diff --git ", line, 11)) {
1440 int git_hdr_len = parse_git_header(line, len, size, patch);
1441 if (git_hdr_len <= len)
1442 continue;
1443 if (!patch->old_name && !patch->new_name) {
1444 if (!patch->def_name)
1445 die("git diff header lacks filename information when removing "
1446 "%d leading pathname components (line %d)" , p_value, linenr);
1447 patch->old_name = xstrdup(patch->def_name);
1448 patch->new_name = xstrdup(patch->def_name);
1450 if (!patch->is_delete && !patch->new_name)
1451 die("git diff header lacks filename information "
1452 "(line %d)", linenr);
1453 patch->is_toplevel_relative = 1;
1454 *hdrsize = git_hdr_len;
1455 return offset;
1458 /* --- followed by +++ ? */
1459 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1460 continue;
1463 * We only accept unified patches, so we want it to
1464 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1465 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1467 nextlen = linelen(line + len, size - len);
1468 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1469 continue;
1471 /* Ok, we'll consider it a patch */
1472 parse_traditional_patch(line, line+len, patch);
1473 *hdrsize = len + nextlen;
1474 linenr += 2;
1475 return offset;
1477 return -1;
1480 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1482 char *err;
1484 if (!result)
1485 return;
1487 whitespace_error++;
1488 if (squelch_whitespace_errors &&
1489 squelch_whitespace_errors < whitespace_error)
1490 return;
1492 err = whitespace_error_string(result);
1493 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1494 patch_input_file, linenr, err, len, line);
1495 free(err);
1498 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1500 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1502 record_ws_error(result, line + 1, len - 2, linenr);
1506 * Parse a unified diff. Note that this really needs to parse each
1507 * fragment separately, since the only way to know the difference
1508 * between a "---" that is part of a patch, and a "---" that starts
1509 * the next patch is to look at the line counts..
1511 static int parse_fragment(char *line, unsigned long size,
1512 struct patch *patch, struct fragment *fragment)
1514 int added, deleted;
1515 int len = linelen(line, size), offset;
1516 unsigned long oldlines, newlines;
1517 unsigned long leading, trailing;
1519 offset = parse_fragment_header(line, len, fragment);
1520 if (offset < 0)
1521 return -1;
1522 if (offset > 0 && patch->recount)
1523 recount_diff(line + offset, size - offset, fragment);
1524 oldlines = fragment->oldlines;
1525 newlines = fragment->newlines;
1526 leading = 0;
1527 trailing = 0;
1529 /* Parse the thing.. */
1530 line += len;
1531 size -= len;
1532 linenr++;
1533 added = deleted = 0;
1534 for (offset = len;
1535 0 < size;
1536 offset += len, size -= len, line += len, linenr++) {
1537 if (!oldlines && !newlines)
1538 break;
1539 len = linelen(line, size);
1540 if (!len || line[len-1] != '\n')
1541 return -1;
1542 switch (*line) {
1543 default:
1544 return -1;
1545 case '\n': /* newer GNU diff, an empty context line */
1546 case ' ':
1547 oldlines--;
1548 newlines--;
1549 if (!deleted && !added)
1550 leading++;
1551 trailing++;
1552 break;
1553 case '-':
1554 if (apply_in_reverse &&
1555 ws_error_action != nowarn_ws_error)
1556 check_whitespace(line, len, patch->ws_rule);
1557 deleted++;
1558 oldlines--;
1559 trailing = 0;
1560 break;
1561 case '+':
1562 if (!apply_in_reverse &&
1563 ws_error_action != nowarn_ws_error)
1564 check_whitespace(line, len, patch->ws_rule);
1565 added++;
1566 newlines--;
1567 trailing = 0;
1568 break;
1571 * We allow "\ No newline at end of file". Depending
1572 * on locale settings when the patch was produced we
1573 * don't know what this line looks like. The only
1574 * thing we do know is that it begins with "\ ".
1575 * Checking for 12 is just for sanity check -- any
1576 * l10n of "\ No newline..." is at least that long.
1578 case '\\':
1579 if (len < 12 || memcmp(line, "\\ ", 2))
1580 return -1;
1581 break;
1584 if (oldlines || newlines)
1585 return -1;
1586 fragment->leading = leading;
1587 fragment->trailing = trailing;
1590 * If a fragment ends with an incomplete line, we failed to include
1591 * it in the above loop because we hit oldlines == newlines == 0
1592 * before seeing it.
1594 if (12 < size && !memcmp(line, "\\ ", 2))
1595 offset += linelen(line, size);
1597 patch->lines_added += added;
1598 patch->lines_deleted += deleted;
1600 if (0 < patch->is_new && oldlines)
1601 return error("new file depends on old contents");
1602 if (0 < patch->is_delete && newlines)
1603 return error("deleted file still has contents");
1604 return offset;
1607 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1609 unsigned long offset = 0;
1610 unsigned long oldlines = 0, newlines = 0, context = 0;
1611 struct fragment **fragp = &patch->fragments;
1613 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1614 struct fragment *fragment;
1615 int len;
1617 fragment = xcalloc(1, sizeof(*fragment));
1618 fragment->linenr = linenr;
1619 len = parse_fragment(line, size, patch, fragment);
1620 if (len <= 0)
1621 die("corrupt patch at line %d", linenr);
1622 fragment->patch = line;
1623 fragment->size = len;
1624 oldlines += fragment->oldlines;
1625 newlines += fragment->newlines;
1626 context += fragment->leading + fragment->trailing;
1628 *fragp = fragment;
1629 fragp = &fragment->next;
1631 offset += len;
1632 line += len;
1633 size -= len;
1637 * If something was removed (i.e. we have old-lines) it cannot
1638 * be creation, and if something was added it cannot be
1639 * deletion. However, the reverse is not true; --unified=0
1640 * patches that only add are not necessarily creation even
1641 * though they do not have any old lines, and ones that only
1642 * delete are not necessarily deletion.
1644 * Unfortunately, a real creation/deletion patch do _not_ have
1645 * any context line by definition, so we cannot safely tell it
1646 * apart with --unified=0 insanity. At least if the patch has
1647 * more than one hunk it is not creation or deletion.
1649 if (patch->is_new < 0 &&
1650 (oldlines || (patch->fragments && patch->fragments->next)))
1651 patch->is_new = 0;
1652 if (patch->is_delete < 0 &&
1653 (newlines || (patch->fragments && patch->fragments->next)))
1654 patch->is_delete = 0;
1656 if (0 < patch->is_new && oldlines)
1657 die("new file %s depends on old contents", patch->new_name);
1658 if (0 < patch->is_delete && newlines)
1659 die("deleted file %s still has contents", patch->old_name);
1660 if (!patch->is_delete && !newlines && context)
1661 fprintf(stderr, "** warning: file %s becomes empty but "
1662 "is not deleted\n", patch->new_name);
1664 return offset;
1667 static inline int metadata_changes(struct patch *patch)
1669 return patch->is_rename > 0 ||
1670 patch->is_copy > 0 ||
1671 patch->is_new > 0 ||
1672 patch->is_delete ||
1673 (patch->old_mode && patch->new_mode &&
1674 patch->old_mode != patch->new_mode);
1677 static char *inflate_it(const void *data, unsigned long size,
1678 unsigned long inflated_size)
1680 git_zstream stream;
1681 void *out;
1682 int st;
1684 memset(&stream, 0, sizeof(stream));
1686 stream.next_in = (unsigned char *)data;
1687 stream.avail_in = size;
1688 stream.next_out = out = xmalloc(inflated_size);
1689 stream.avail_out = inflated_size;
1690 git_inflate_init(&stream);
1691 st = git_inflate(&stream, Z_FINISH);
1692 git_inflate_end(&stream);
1693 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1694 free(out);
1695 return NULL;
1697 return out;
1700 static struct fragment *parse_binary_hunk(char **buf_p,
1701 unsigned long *sz_p,
1702 int *status_p,
1703 int *used_p)
1706 * Expect a line that begins with binary patch method ("literal"
1707 * or "delta"), followed by the length of data before deflating.
1708 * a sequence of 'length-byte' followed by base-85 encoded data
1709 * should follow, terminated by a newline.
1711 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1712 * and we would limit the patch line to 66 characters,
1713 * so one line can fit up to 13 groups that would decode
1714 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1715 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1717 int llen, used;
1718 unsigned long size = *sz_p;
1719 char *buffer = *buf_p;
1720 int patch_method;
1721 unsigned long origlen;
1722 char *data = NULL;
1723 int hunk_size = 0;
1724 struct fragment *frag;
1726 llen = linelen(buffer, size);
1727 used = llen;
1729 *status_p = 0;
1731 if (!prefixcmp(buffer, "delta ")) {
1732 patch_method = BINARY_DELTA_DEFLATED;
1733 origlen = strtoul(buffer + 6, NULL, 10);
1735 else if (!prefixcmp(buffer, "literal ")) {
1736 patch_method = BINARY_LITERAL_DEFLATED;
1737 origlen = strtoul(buffer + 8, NULL, 10);
1739 else
1740 return NULL;
1742 linenr++;
1743 buffer += llen;
1744 while (1) {
1745 int byte_length, max_byte_length, newsize;
1746 llen = linelen(buffer, size);
1747 used += llen;
1748 linenr++;
1749 if (llen == 1) {
1750 /* consume the blank line */
1751 buffer++;
1752 size--;
1753 break;
1756 * Minimum line is "A00000\n" which is 7-byte long,
1757 * and the line length must be multiple of 5 plus 2.
1759 if ((llen < 7) || (llen-2) % 5)
1760 goto corrupt;
1761 max_byte_length = (llen - 2) / 5 * 4;
1762 byte_length = *buffer;
1763 if ('A' <= byte_length && byte_length <= 'Z')
1764 byte_length = byte_length - 'A' + 1;
1765 else if ('a' <= byte_length && byte_length <= 'z')
1766 byte_length = byte_length - 'a' + 27;
1767 else
1768 goto corrupt;
1769 /* if the input length was not multiple of 4, we would
1770 * have filler at the end but the filler should never
1771 * exceed 3 bytes
1773 if (max_byte_length < byte_length ||
1774 byte_length <= max_byte_length - 4)
1775 goto corrupt;
1776 newsize = hunk_size + byte_length;
1777 data = xrealloc(data, newsize);
1778 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1779 goto corrupt;
1780 hunk_size = newsize;
1781 buffer += llen;
1782 size -= llen;
1785 frag = xcalloc(1, sizeof(*frag));
1786 frag->patch = inflate_it(data, hunk_size, origlen);
1787 frag->free_patch = 1;
1788 if (!frag->patch)
1789 goto corrupt;
1790 free(data);
1791 frag->size = origlen;
1792 *buf_p = buffer;
1793 *sz_p = size;
1794 *used_p = used;
1795 frag->binary_patch_method = patch_method;
1796 return frag;
1798 corrupt:
1799 free(data);
1800 *status_p = -1;
1801 error("corrupt binary patch at line %d: %.*s",
1802 linenr-1, llen-1, buffer);
1803 return NULL;
1806 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1809 * We have read "GIT binary patch\n"; what follows is a line
1810 * that says the patch method (currently, either "literal" or
1811 * "delta") and the length of data before deflating; a
1812 * sequence of 'length-byte' followed by base-85 encoded data
1813 * follows.
1815 * When a binary patch is reversible, there is another binary
1816 * hunk in the same format, starting with patch method (either
1817 * "literal" or "delta") with the length of data, and a sequence
1818 * of length-byte + base-85 encoded data, terminated with another
1819 * empty line. This data, when applied to the postimage, produces
1820 * the preimage.
1822 struct fragment *forward;
1823 struct fragment *reverse;
1824 int status;
1825 int used, used_1;
1827 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1828 if (!forward && !status)
1829 /* there has to be one hunk (forward hunk) */
1830 return error("unrecognized binary patch at line %d", linenr-1);
1831 if (status)
1832 /* otherwise we already gave an error message */
1833 return status;
1835 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1836 if (reverse)
1837 used += used_1;
1838 else if (status) {
1840 * Not having reverse hunk is not an error, but having
1841 * a corrupt reverse hunk is.
1843 free((void*) forward->patch);
1844 free(forward);
1845 return status;
1847 forward->next = reverse;
1848 patch->fragments = forward;
1849 patch->is_binary = 1;
1850 return used;
1853 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1855 int hdrsize, patchsize;
1856 int offset = find_header(buffer, size, &hdrsize, patch);
1858 if (offset < 0)
1859 return offset;
1861 patch->ws_rule = whitespace_rule(patch->new_name
1862 ? patch->new_name
1863 : patch->old_name);
1865 patchsize = parse_single_patch(buffer + offset + hdrsize,
1866 size - offset - hdrsize, patch);
1868 if (!patchsize) {
1869 static const char *binhdr[] = {
1870 "Binary files ",
1871 "Files ",
1872 NULL,
1874 static const char git_binary[] = "GIT binary patch\n";
1875 int i;
1876 int hd = hdrsize + offset;
1877 unsigned long llen = linelen(buffer + hd, size - hd);
1879 if (llen == sizeof(git_binary) - 1 &&
1880 !memcmp(git_binary, buffer + hd, llen)) {
1881 int used;
1882 linenr++;
1883 used = parse_binary(buffer + hd + llen,
1884 size - hd - llen, patch);
1885 if (used)
1886 patchsize = used + llen;
1887 else
1888 patchsize = 0;
1890 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1891 for (i = 0; binhdr[i]; i++) {
1892 int len = strlen(binhdr[i]);
1893 if (len < size - hd &&
1894 !memcmp(binhdr[i], buffer + hd, len)) {
1895 linenr++;
1896 patch->is_binary = 1;
1897 patchsize = llen;
1898 break;
1903 /* Empty patch cannot be applied if it is a text patch
1904 * without metadata change. A binary patch appears
1905 * empty to us here.
1907 if ((apply || check) &&
1908 (!patch->is_binary && !metadata_changes(patch)))
1909 die("patch with only garbage at line %d", linenr);
1912 return offset + hdrsize + patchsize;
1915 #define swap(a,b) myswap((a),(b),sizeof(a))
1917 #define myswap(a, b, size) do { \
1918 unsigned char mytmp[size]; \
1919 memcpy(mytmp, &a, size); \
1920 memcpy(&a, &b, size); \
1921 memcpy(&b, mytmp, size); \
1922 } while (0)
1924 static void reverse_patches(struct patch *p)
1926 for (; p; p = p->next) {
1927 struct fragment *frag = p->fragments;
1929 swap(p->new_name, p->old_name);
1930 swap(p->new_mode, p->old_mode);
1931 swap(p->is_new, p->is_delete);
1932 swap(p->lines_added, p->lines_deleted);
1933 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1935 for (; frag; frag = frag->next) {
1936 swap(frag->newpos, frag->oldpos);
1937 swap(frag->newlines, frag->oldlines);
1942 static const char pluses[] =
1943 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1944 static const char minuses[]=
1945 "----------------------------------------------------------------------";
1947 static void show_stats(struct patch *patch)
1949 struct strbuf qname = STRBUF_INIT;
1950 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1951 int max, add, del;
1953 quote_c_style(cp, &qname, NULL, 0);
1956 * "scale" the filename
1958 max = max_len;
1959 if (max > 50)
1960 max = 50;
1962 if (qname.len > max) {
1963 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1964 if (!cp)
1965 cp = qname.buf + qname.len + 3 - max;
1966 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1969 if (patch->is_binary) {
1970 printf(" %-*s | Bin\n", max, qname.buf);
1971 strbuf_release(&qname);
1972 return;
1975 printf(" %-*s |", max, qname.buf);
1976 strbuf_release(&qname);
1979 * scale the add/delete
1981 max = max + max_change > 70 ? 70 - max : max_change;
1982 add = patch->lines_added;
1983 del = patch->lines_deleted;
1985 if (max_change > 0) {
1986 int total = ((add + del) * max + max_change / 2) / max_change;
1987 add = (add * max + max_change / 2) / max_change;
1988 del = total - add;
1990 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1991 add, pluses, del, minuses);
1994 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1996 switch (st->st_mode & S_IFMT) {
1997 case S_IFLNK:
1998 if (strbuf_readlink(buf, path, st->st_size) < 0)
1999 return error("unable to read symlink %s", path);
2000 return 0;
2001 case S_IFREG:
2002 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2003 return error("unable to open or read %s", path);
2004 convert_to_git(path, buf->buf, buf->len, buf, 0);
2005 return 0;
2006 default:
2007 return -1;
2012 * Update the preimage, and the common lines in postimage,
2013 * from buffer buf of length len. If postlen is 0 the postimage
2014 * is updated in place, otherwise it's updated on a new buffer
2015 * of length postlen
2018 static void update_pre_post_images(struct image *preimage,
2019 struct image *postimage,
2020 char *buf,
2021 size_t len, size_t postlen)
2023 int i, ctx;
2024 char *new, *old, *fixed;
2025 struct image fixed_preimage;
2028 * Update the preimage with whitespace fixes. Note that we
2029 * are not losing preimage->buf -- apply_one_fragment() will
2030 * free "oldlines".
2032 prepare_image(&fixed_preimage, buf, len, 1);
2033 assert(fixed_preimage.nr == preimage->nr);
2034 for (i = 0; i < preimage->nr; i++)
2035 fixed_preimage.line[i].flag = preimage->line[i].flag;
2036 free(preimage->line_allocated);
2037 *preimage = fixed_preimage;
2040 * Adjust the common context lines in postimage. This can be
2041 * done in-place when we are just doing whitespace fixing,
2042 * which does not make the string grow, but needs a new buffer
2043 * when ignoring whitespace causes the update, since in this case
2044 * we could have e.g. tabs converted to multiple spaces.
2045 * We trust the caller to tell us if the update can be done
2046 * in place (postlen==0) or not.
2048 old = postimage->buf;
2049 if (postlen)
2050 new = postimage->buf = xmalloc(postlen);
2051 else
2052 new = old;
2053 fixed = preimage->buf;
2054 for (i = ctx = 0; i < postimage->nr; i++) {
2055 size_t len = postimage->line[i].len;
2056 if (!(postimage->line[i].flag & LINE_COMMON)) {
2057 /* an added line -- no counterparts in preimage */
2058 memmove(new, old, len);
2059 old += len;
2060 new += len;
2061 continue;
2064 /* a common context -- skip it in the original postimage */
2065 old += len;
2067 /* and find the corresponding one in the fixed preimage */
2068 while (ctx < preimage->nr &&
2069 !(preimage->line[ctx].flag & LINE_COMMON)) {
2070 fixed += preimage->line[ctx].len;
2071 ctx++;
2073 if (preimage->nr <= ctx)
2074 die("oops");
2076 /* and copy it in, while fixing the line length */
2077 len = preimage->line[ctx].len;
2078 memcpy(new, fixed, len);
2079 new += len;
2080 fixed += len;
2081 postimage->line[i].len = len;
2082 ctx++;
2085 /* Fix the length of the whole thing */
2086 postimage->len = new - postimage->buf;
2089 static int match_fragment(struct image *img,
2090 struct image *preimage,
2091 struct image *postimage,
2092 unsigned long try,
2093 int try_lno,
2094 unsigned ws_rule,
2095 int match_beginning, int match_end)
2097 int i;
2098 char *fixed_buf, *buf, *orig, *target;
2099 struct strbuf fixed;
2100 size_t fixed_len;
2101 int preimage_limit;
2103 if (preimage->nr + try_lno <= img->nr) {
2105 * The hunk falls within the boundaries of img.
2107 preimage_limit = preimage->nr;
2108 if (match_end && (preimage->nr + try_lno != img->nr))
2109 return 0;
2110 } else if (ws_error_action == correct_ws_error &&
2111 (ws_rule & WS_BLANK_AT_EOF)) {
2113 * This hunk extends beyond the end of img, and we are
2114 * removing blank lines at the end of the file. This
2115 * many lines from the beginning of the preimage must
2116 * match with img, and the remainder of the preimage
2117 * must be blank.
2119 preimage_limit = img->nr - try_lno;
2120 } else {
2122 * The hunk extends beyond the end of the img and
2123 * we are not removing blanks at the end, so we
2124 * should reject the hunk at this position.
2126 return 0;
2129 if (match_beginning && try_lno)
2130 return 0;
2132 /* Quick hash check */
2133 for (i = 0; i < preimage_limit; i++)
2134 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2135 (preimage->line[i].hash != img->line[try_lno + i].hash))
2136 return 0;
2138 if (preimage_limit == preimage->nr) {
2140 * Do we have an exact match? If we were told to match
2141 * at the end, size must be exactly at try+fragsize,
2142 * otherwise try+fragsize must be still within the preimage,
2143 * and either case, the old piece should match the preimage
2144 * exactly.
2146 if ((match_end
2147 ? (try + preimage->len == img->len)
2148 : (try + preimage->len <= img->len)) &&
2149 !memcmp(img->buf + try, preimage->buf, preimage->len))
2150 return 1;
2151 } else {
2153 * The preimage extends beyond the end of img, so
2154 * there cannot be an exact match.
2156 * There must be one non-blank context line that match
2157 * a line before the end of img.
2159 char *buf_end;
2161 buf = preimage->buf;
2162 buf_end = buf;
2163 for (i = 0; i < preimage_limit; i++)
2164 buf_end += preimage->line[i].len;
2166 for ( ; buf < buf_end; buf++)
2167 if (!isspace(*buf))
2168 break;
2169 if (buf == buf_end)
2170 return 0;
2174 * No exact match. If we are ignoring whitespace, run a line-by-line
2175 * fuzzy matching. We collect all the line length information because
2176 * we need it to adjust whitespace if we match.
2178 if (ws_ignore_action == ignore_ws_change) {
2179 size_t imgoff = 0;
2180 size_t preoff = 0;
2181 size_t postlen = postimage->len;
2182 size_t extra_chars;
2183 char *preimage_eof;
2184 char *preimage_end;
2185 for (i = 0; i < preimage_limit; i++) {
2186 size_t prelen = preimage->line[i].len;
2187 size_t imglen = img->line[try_lno+i].len;
2189 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2190 preimage->buf + preoff, prelen))
2191 return 0;
2192 if (preimage->line[i].flag & LINE_COMMON)
2193 postlen += imglen - prelen;
2194 imgoff += imglen;
2195 preoff += prelen;
2199 * Ok, the preimage matches with whitespace fuzz.
2201 * imgoff now holds the true length of the target that
2202 * matches the preimage before the end of the file.
2204 * Count the number of characters in the preimage that fall
2205 * beyond the end of the file and make sure that all of them
2206 * are whitespace characters. (This can only happen if
2207 * we are removing blank lines at the end of the file.)
2209 buf = preimage_eof = preimage->buf + preoff;
2210 for ( ; i < preimage->nr; i++)
2211 preoff += preimage->line[i].len;
2212 preimage_end = preimage->buf + preoff;
2213 for ( ; buf < preimage_end; buf++)
2214 if (!isspace(*buf))
2215 return 0;
2218 * Update the preimage and the common postimage context
2219 * lines to use the same whitespace as the target.
2220 * If whitespace is missing in the target (i.e.
2221 * if the preimage extends beyond the end of the file),
2222 * use the whitespace from the preimage.
2224 extra_chars = preimage_end - preimage_eof;
2225 strbuf_init(&fixed, imgoff + extra_chars);
2226 strbuf_add(&fixed, img->buf + try, imgoff);
2227 strbuf_add(&fixed, preimage_eof, extra_chars);
2228 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2229 update_pre_post_images(preimage, postimage,
2230 fixed_buf, fixed_len, postlen);
2231 return 1;
2234 if (ws_error_action != correct_ws_error)
2235 return 0;
2238 * The hunk does not apply byte-by-byte, but the hash says
2239 * it might with whitespace fuzz. We haven't been asked to
2240 * ignore whitespace, we were asked to correct whitespace
2241 * errors, so let's try matching after whitespace correction.
2243 * The preimage may extend beyond the end of the file,
2244 * but in this loop we will only handle the part of the
2245 * preimage that falls within the file.
2247 strbuf_init(&fixed, preimage->len + 1);
2248 orig = preimage->buf;
2249 target = img->buf + try;
2250 for (i = 0; i < preimage_limit; i++) {
2251 size_t oldlen = preimage->line[i].len;
2252 size_t tgtlen = img->line[try_lno + i].len;
2253 size_t fixstart = fixed.len;
2254 struct strbuf tgtfix;
2255 int match;
2257 /* Try fixing the line in the preimage */
2258 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2260 /* Try fixing the line in the target */
2261 strbuf_init(&tgtfix, tgtlen);
2262 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2265 * If they match, either the preimage was based on
2266 * a version before our tree fixed whitespace breakage,
2267 * or we are lacking a whitespace-fix patch the tree
2268 * the preimage was based on already had (i.e. target
2269 * has whitespace breakage, the preimage doesn't).
2270 * In either case, we are fixing the whitespace breakages
2271 * so we might as well take the fix together with their
2272 * real change.
2274 match = (tgtfix.len == fixed.len - fixstart &&
2275 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2276 fixed.len - fixstart));
2278 strbuf_release(&tgtfix);
2279 if (!match)
2280 goto unmatch_exit;
2282 orig += oldlen;
2283 target += tgtlen;
2288 * Now handle the lines in the preimage that falls beyond the
2289 * end of the file (if any). They will only match if they are
2290 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2291 * false).
2293 for ( ; i < preimage->nr; i++) {
2294 size_t fixstart = fixed.len; /* start of the fixed preimage */
2295 size_t oldlen = preimage->line[i].len;
2296 int j;
2298 /* Try fixing the line in the preimage */
2299 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2301 for (j = fixstart; j < fixed.len; j++)
2302 if (!isspace(fixed.buf[j]))
2303 goto unmatch_exit;
2305 orig += oldlen;
2309 * Yes, the preimage is based on an older version that still
2310 * has whitespace breakages unfixed, and fixing them makes the
2311 * hunk match. Update the context lines in the postimage.
2313 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2314 update_pre_post_images(preimage, postimage,
2315 fixed_buf, fixed_len, 0);
2316 return 1;
2318 unmatch_exit:
2319 strbuf_release(&fixed);
2320 return 0;
2323 static int find_pos(struct image *img,
2324 struct image *preimage,
2325 struct image *postimage,
2326 int line,
2327 unsigned ws_rule,
2328 int match_beginning, int match_end)
2330 int i;
2331 unsigned long backwards, forwards, try;
2332 int backwards_lno, forwards_lno, try_lno;
2335 * If match_beginning or match_end is specified, there is no
2336 * point starting from a wrong line that will never match and
2337 * wander around and wait for a match at the specified end.
2339 if (match_beginning)
2340 line = 0;
2341 else if (match_end)
2342 line = img->nr - preimage->nr;
2345 * Because the comparison is unsigned, the following test
2346 * will also take care of a negative line number that can
2347 * result when match_end and preimage is larger than the target.
2349 if ((size_t) line > img->nr)
2350 line = img->nr;
2352 try = 0;
2353 for (i = 0; i < line; i++)
2354 try += img->line[i].len;
2357 * There's probably some smart way to do this, but I'll leave
2358 * that to the smart and beautiful people. I'm simple and stupid.
2360 backwards = try;
2361 backwards_lno = line;
2362 forwards = try;
2363 forwards_lno = line;
2364 try_lno = line;
2366 for (i = 0; ; i++) {
2367 if (match_fragment(img, preimage, postimage,
2368 try, try_lno, ws_rule,
2369 match_beginning, match_end))
2370 return try_lno;
2372 again:
2373 if (backwards_lno == 0 && forwards_lno == img->nr)
2374 break;
2376 if (i & 1) {
2377 if (backwards_lno == 0) {
2378 i++;
2379 goto again;
2381 backwards_lno--;
2382 backwards -= img->line[backwards_lno].len;
2383 try = backwards;
2384 try_lno = backwards_lno;
2385 } else {
2386 if (forwards_lno == img->nr) {
2387 i++;
2388 goto again;
2390 forwards += img->line[forwards_lno].len;
2391 forwards_lno++;
2392 try = forwards;
2393 try_lno = forwards_lno;
2397 return -1;
2400 static void remove_first_line(struct image *img)
2402 img->buf += img->line[0].len;
2403 img->len -= img->line[0].len;
2404 img->line++;
2405 img->nr--;
2408 static void remove_last_line(struct image *img)
2410 img->len -= img->line[--img->nr].len;
2413 static void update_image(struct image *img,
2414 int applied_pos,
2415 struct image *preimage,
2416 struct image *postimage)
2419 * remove the copy of preimage at offset in img
2420 * and replace it with postimage
2422 int i, nr;
2423 size_t remove_count, insert_count, applied_at = 0;
2424 char *result;
2425 int preimage_limit;
2428 * If we are removing blank lines at the end of img,
2429 * the preimage may extend beyond the end.
2430 * If that is the case, we must be careful only to
2431 * remove the part of the preimage that falls within
2432 * the boundaries of img. Initialize preimage_limit
2433 * to the number of lines in the preimage that falls
2434 * within the boundaries.
2436 preimage_limit = preimage->nr;
2437 if (preimage_limit > img->nr - applied_pos)
2438 preimage_limit = img->nr - applied_pos;
2440 for (i = 0; i < applied_pos; i++)
2441 applied_at += img->line[i].len;
2443 remove_count = 0;
2444 for (i = 0; i < preimage_limit; i++)
2445 remove_count += img->line[applied_pos + i].len;
2446 insert_count = postimage->len;
2448 /* Adjust the contents */
2449 result = xmalloc(img->len + insert_count - remove_count + 1);
2450 memcpy(result, img->buf, applied_at);
2451 memcpy(result + applied_at, postimage->buf, postimage->len);
2452 memcpy(result + applied_at + postimage->len,
2453 img->buf + (applied_at + remove_count),
2454 img->len - (applied_at + remove_count));
2455 free(img->buf);
2456 img->buf = result;
2457 img->len += insert_count - remove_count;
2458 result[img->len] = '\0';
2460 /* Adjust the line table */
2461 nr = img->nr + postimage->nr - preimage_limit;
2462 if (preimage_limit < postimage->nr) {
2464 * NOTE: this knows that we never call remove_first_line()
2465 * on anything other than pre/post image.
2467 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2468 img->line_allocated = img->line;
2470 if (preimage_limit != postimage->nr)
2471 memmove(img->line + applied_pos + postimage->nr,
2472 img->line + applied_pos + preimage_limit,
2473 (img->nr - (applied_pos + preimage_limit)) *
2474 sizeof(*img->line));
2475 memcpy(img->line + applied_pos,
2476 postimage->line,
2477 postimage->nr * sizeof(*img->line));
2478 if (!allow_overlap)
2479 for (i = 0; i < postimage->nr; i++)
2480 img->line[applied_pos + i].flag |= LINE_PATCHED;
2481 img->nr = nr;
2484 static int apply_one_fragment(struct image *img, struct fragment *frag,
2485 int inaccurate_eof, unsigned ws_rule,
2486 int nth_fragment)
2488 int match_beginning, match_end;
2489 const char *patch = frag->patch;
2490 int size = frag->size;
2491 char *old, *oldlines;
2492 struct strbuf newlines;
2493 int new_blank_lines_at_end = 0;
2494 int found_new_blank_lines_at_end = 0;
2495 int hunk_linenr = frag->linenr;
2496 unsigned long leading, trailing;
2497 int pos, applied_pos;
2498 struct image preimage;
2499 struct image postimage;
2501 memset(&preimage, 0, sizeof(preimage));
2502 memset(&postimage, 0, sizeof(postimage));
2503 oldlines = xmalloc(size);
2504 strbuf_init(&newlines, size);
2506 old = oldlines;
2507 while (size > 0) {
2508 char first;
2509 int len = linelen(patch, size);
2510 int plen;
2511 int added_blank_line = 0;
2512 int is_blank_context = 0;
2513 size_t start;
2515 if (!len)
2516 break;
2519 * "plen" is how much of the line we should use for
2520 * the actual patch data. Normally we just remove the
2521 * first character on the line, but if the line is
2522 * followed by "\ No newline", then we also remove the
2523 * last one (which is the newline, of course).
2525 plen = len - 1;
2526 if (len < size && patch[len] == '\\')
2527 plen--;
2528 first = *patch;
2529 if (apply_in_reverse) {
2530 if (first == '-')
2531 first = '+';
2532 else if (first == '+')
2533 first = '-';
2536 switch (first) {
2537 case '\n':
2538 /* Newer GNU diff, empty context line */
2539 if (plen < 0)
2540 /* ... followed by '\No newline'; nothing */
2541 break;
2542 *old++ = '\n';
2543 strbuf_addch(&newlines, '\n');
2544 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2545 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2546 is_blank_context = 1;
2547 break;
2548 case ' ':
2549 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2550 ws_blank_line(patch + 1, plen, ws_rule))
2551 is_blank_context = 1;
2552 case '-':
2553 memcpy(old, patch + 1, plen);
2554 add_line_info(&preimage, old, plen,
2555 (first == ' ' ? LINE_COMMON : 0));
2556 old += plen;
2557 if (first == '-')
2558 break;
2559 /* Fall-through for ' ' */
2560 case '+':
2561 /* --no-add does not add new lines */
2562 if (first == '+' && no_add)
2563 break;
2565 start = newlines.len;
2566 if (first != '+' ||
2567 !whitespace_error ||
2568 ws_error_action != correct_ws_error) {
2569 strbuf_add(&newlines, patch + 1, plen);
2571 else {
2572 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2574 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2575 (first == '+' ? 0 : LINE_COMMON));
2576 if (first == '+' &&
2577 (ws_rule & WS_BLANK_AT_EOF) &&
2578 ws_blank_line(patch + 1, plen, ws_rule))
2579 added_blank_line = 1;
2580 break;
2581 case '@': case '\\':
2582 /* Ignore it, we already handled it */
2583 break;
2584 default:
2585 if (apply_verbosely)
2586 error("invalid start of line: '%c'", first);
2587 return -1;
2589 if (added_blank_line) {
2590 if (!new_blank_lines_at_end)
2591 found_new_blank_lines_at_end = hunk_linenr;
2592 new_blank_lines_at_end++;
2594 else if (is_blank_context)
2596 else
2597 new_blank_lines_at_end = 0;
2598 patch += len;
2599 size -= len;
2600 hunk_linenr++;
2602 if (inaccurate_eof &&
2603 old > oldlines && old[-1] == '\n' &&
2604 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2605 old--;
2606 strbuf_setlen(&newlines, newlines.len - 1);
2609 leading = frag->leading;
2610 trailing = frag->trailing;
2613 * A hunk to change lines at the beginning would begin with
2614 * @@ -1,L +N,M @@
2615 * but we need to be careful. -U0 that inserts before the second
2616 * line also has this pattern.
2618 * And a hunk to add to an empty file would begin with
2619 * @@ -0,0 +N,M @@
2621 * In other words, a hunk that is (frag->oldpos <= 1) with or
2622 * without leading context must match at the beginning.
2624 match_beginning = (!frag->oldpos ||
2625 (frag->oldpos == 1 && !unidiff_zero));
2628 * A hunk without trailing lines must match at the end.
2629 * However, we simply cannot tell if a hunk must match end
2630 * from the lack of trailing lines if the patch was generated
2631 * with unidiff without any context.
2633 match_end = !unidiff_zero && !trailing;
2635 pos = frag->newpos ? (frag->newpos - 1) : 0;
2636 preimage.buf = oldlines;
2637 preimage.len = old - oldlines;
2638 postimage.buf = newlines.buf;
2639 postimage.len = newlines.len;
2640 preimage.line = preimage.line_allocated;
2641 postimage.line = postimage.line_allocated;
2643 for (;;) {
2645 applied_pos = find_pos(img, &preimage, &postimage, pos,
2646 ws_rule, match_beginning, match_end);
2648 if (applied_pos >= 0)
2649 break;
2651 /* Am I at my context limits? */
2652 if ((leading <= p_context) && (trailing <= p_context))
2653 break;
2654 if (match_beginning || match_end) {
2655 match_beginning = match_end = 0;
2656 continue;
2660 * Reduce the number of context lines; reduce both
2661 * leading and trailing if they are equal otherwise
2662 * just reduce the larger context.
2664 if (leading >= trailing) {
2665 remove_first_line(&preimage);
2666 remove_first_line(&postimage);
2667 pos--;
2668 leading--;
2670 if (trailing > leading) {
2671 remove_last_line(&preimage);
2672 remove_last_line(&postimage);
2673 trailing--;
2677 if (applied_pos >= 0) {
2678 if (new_blank_lines_at_end &&
2679 preimage.nr + applied_pos >= img->nr &&
2680 (ws_rule & WS_BLANK_AT_EOF) &&
2681 ws_error_action != nowarn_ws_error) {
2682 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2683 found_new_blank_lines_at_end);
2684 if (ws_error_action == correct_ws_error) {
2685 while (new_blank_lines_at_end--)
2686 remove_last_line(&postimage);
2689 * We would want to prevent write_out_results()
2690 * from taking place in apply_patch() that follows
2691 * the callchain led us here, which is:
2692 * apply_patch->check_patch_list->check_patch->
2693 * apply_data->apply_fragments->apply_one_fragment
2695 if (ws_error_action == die_on_ws_error)
2696 apply = 0;
2699 if (apply_verbosely && applied_pos != pos) {
2700 int offset = applied_pos - pos;
2701 if (apply_in_reverse)
2702 offset = 0 - offset;
2703 fprintf(stderr,
2704 "Hunk #%d succeeded at %d (offset %d lines).\n",
2705 nth_fragment, applied_pos + 1, offset);
2709 * Warn if it was necessary to reduce the number
2710 * of context lines.
2712 if ((leading != frag->leading) ||
2713 (trailing != frag->trailing))
2714 fprintf(stderr, "Context reduced to (%ld/%ld)"
2715 " to apply fragment at %d\n",
2716 leading, trailing, applied_pos+1);
2717 update_image(img, applied_pos, &preimage, &postimage);
2718 } else {
2719 if (apply_verbosely)
2720 error("while searching for:\n%.*s",
2721 (int)(old - oldlines), oldlines);
2724 free(oldlines);
2725 strbuf_release(&newlines);
2726 free(preimage.line_allocated);
2727 free(postimage.line_allocated);
2729 return (applied_pos < 0);
2732 static int apply_binary_fragment(struct image *img, struct patch *patch)
2734 struct fragment *fragment = patch->fragments;
2735 unsigned long len;
2736 void *dst;
2738 if (!fragment)
2739 return error("missing binary patch data for '%s'",
2740 patch->new_name ?
2741 patch->new_name :
2742 patch->old_name);
2744 /* Binary patch is irreversible without the optional second hunk */
2745 if (apply_in_reverse) {
2746 if (!fragment->next)
2747 return error("cannot reverse-apply a binary patch "
2748 "without the reverse hunk to '%s'",
2749 patch->new_name
2750 ? patch->new_name : patch->old_name);
2751 fragment = fragment->next;
2753 switch (fragment->binary_patch_method) {
2754 case BINARY_DELTA_DEFLATED:
2755 dst = patch_delta(img->buf, img->len, fragment->patch,
2756 fragment->size, &len);
2757 if (!dst)
2758 return -1;
2759 clear_image(img);
2760 img->buf = dst;
2761 img->len = len;
2762 return 0;
2763 case BINARY_LITERAL_DEFLATED:
2764 clear_image(img);
2765 img->len = fragment->size;
2766 img->buf = xmalloc(img->len+1);
2767 memcpy(img->buf, fragment->patch, img->len);
2768 img->buf[img->len] = '\0';
2769 return 0;
2771 return -1;
2774 static int apply_binary(struct image *img, struct patch *patch)
2776 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2777 unsigned char sha1[20];
2780 * For safety, we require patch index line to contain
2781 * full 40-byte textual SHA1 for old and new, at least for now.
2783 if (strlen(patch->old_sha1_prefix) != 40 ||
2784 strlen(patch->new_sha1_prefix) != 40 ||
2785 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2786 get_sha1_hex(patch->new_sha1_prefix, sha1))
2787 return error("cannot apply binary patch to '%s' "
2788 "without full index line", name);
2790 if (patch->old_name) {
2792 * See if the old one matches what the patch
2793 * applies to.
2795 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2796 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2797 return error("the patch applies to '%s' (%s), "
2798 "which does not match the "
2799 "current contents.",
2800 name, sha1_to_hex(sha1));
2802 else {
2803 /* Otherwise, the old one must be empty. */
2804 if (img->len)
2805 return error("the patch applies to an empty "
2806 "'%s' but it is not empty", name);
2809 get_sha1_hex(patch->new_sha1_prefix, sha1);
2810 if (is_null_sha1(sha1)) {
2811 clear_image(img);
2812 return 0; /* deletion patch */
2815 if (has_sha1_file(sha1)) {
2816 /* We already have the postimage */
2817 enum object_type type;
2818 unsigned long size;
2819 char *result;
2821 result = read_sha1_file(sha1, &type, &size);
2822 if (!result)
2823 return error("the necessary postimage %s for "
2824 "'%s' cannot be read",
2825 patch->new_sha1_prefix, name);
2826 clear_image(img);
2827 img->buf = result;
2828 img->len = size;
2829 } else {
2831 * We have verified buf matches the preimage;
2832 * apply the patch data to it, which is stored
2833 * in the patch->fragments->{patch,size}.
2835 if (apply_binary_fragment(img, patch))
2836 return error("binary patch does not apply to '%s'",
2837 name);
2839 /* verify that the result matches */
2840 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2841 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2842 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2843 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2846 return 0;
2849 static int apply_fragments(struct image *img, struct patch *patch)
2851 struct fragment *frag = patch->fragments;
2852 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2853 unsigned ws_rule = patch->ws_rule;
2854 unsigned inaccurate_eof = patch->inaccurate_eof;
2855 int nth = 0;
2857 if (patch->is_binary)
2858 return apply_binary(img, patch);
2860 while (frag) {
2861 nth++;
2862 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2863 error("patch failed: %s:%ld", name, frag->oldpos);
2864 if (!apply_with_reject)
2865 return -1;
2866 frag->rejected = 1;
2868 frag = frag->next;
2870 return 0;
2873 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2875 if (!ce)
2876 return 0;
2878 if (S_ISGITLINK(ce->ce_mode)) {
2879 strbuf_grow(buf, 100);
2880 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2881 } else {
2882 enum object_type type;
2883 unsigned long sz;
2884 char *result;
2886 result = read_sha1_file(ce->sha1, &type, &sz);
2887 if (!result)
2888 return -1;
2889 /* XXX read_sha1_file NUL-terminates */
2890 strbuf_attach(buf, result, sz, sz + 1);
2892 return 0;
2895 static struct patch *in_fn_table(const char *name)
2897 struct string_list_item *item;
2899 if (name == NULL)
2900 return NULL;
2902 item = string_list_lookup(&fn_table, name);
2903 if (item != NULL)
2904 return (struct patch *)item->util;
2906 return NULL;
2910 * item->util in the filename table records the status of the path.
2911 * Usually it points at a patch (whose result records the contents
2912 * of it after applying it), but it could be PATH_WAS_DELETED for a
2913 * path that a previously applied patch has already removed.
2915 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2916 #define PATH_WAS_DELETED ((struct patch *) -1)
2918 static int to_be_deleted(struct patch *patch)
2920 return patch == PATH_TO_BE_DELETED;
2923 static int was_deleted(struct patch *patch)
2925 return patch == PATH_WAS_DELETED;
2928 static void add_to_fn_table(struct patch *patch)
2930 struct string_list_item *item;
2933 * Always add new_name unless patch is a deletion
2934 * This should cover the cases for normal diffs,
2935 * file creations and copies
2937 if (patch->new_name != NULL) {
2938 item = string_list_insert(&fn_table, patch->new_name);
2939 item->util = patch;
2943 * store a failure on rename/deletion cases because
2944 * later chunks shouldn't patch old names
2946 if ((patch->new_name == NULL) || (patch->is_rename)) {
2947 item = string_list_insert(&fn_table, patch->old_name);
2948 item->util = PATH_WAS_DELETED;
2952 static void prepare_fn_table(struct patch *patch)
2955 * store information about incoming file deletion
2957 while (patch) {
2958 if ((patch->new_name == NULL) || (patch->is_rename)) {
2959 struct string_list_item *item;
2960 item = string_list_insert(&fn_table, patch->old_name);
2961 item->util = PATH_TO_BE_DELETED;
2963 patch = patch->next;
2967 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2969 struct strbuf buf = STRBUF_INIT;
2970 struct image image;
2971 size_t len;
2972 char *img;
2973 struct patch *tpatch;
2975 if (!(patch->is_copy || patch->is_rename) &&
2976 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2977 if (was_deleted(tpatch)) {
2978 return error("patch %s has been renamed/deleted",
2979 patch->old_name);
2981 /* We have a patched copy in memory use that */
2982 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2983 } else if (cached) {
2984 if (read_file_or_gitlink(ce, &buf))
2985 return error("read of %s failed", patch->old_name);
2986 } else if (patch->old_name) {
2987 if (S_ISGITLINK(patch->old_mode)) {
2988 if (ce) {
2989 read_file_or_gitlink(ce, &buf);
2990 } else {
2992 * There is no way to apply subproject
2993 * patch without looking at the index.
2995 patch->fragments = NULL;
2997 } else {
2998 if (read_old_data(st, patch->old_name, &buf))
2999 return error("read of %s failed", patch->old_name);
3003 img = strbuf_detach(&buf, &len);
3004 prepare_image(&image, img, len, !patch->is_binary);
3006 if (apply_fragments(&image, patch) < 0)
3007 return -1; /* note with --reject this succeeds. */
3008 patch->result = image.buf;
3009 patch->resultsize = image.len;
3010 add_to_fn_table(patch);
3011 free(image.line_allocated);
3013 if (0 < patch->is_delete && patch->resultsize)
3014 return error("removal patch leaves file contents");
3016 return 0;
3019 static int check_to_create_blob(const char *new_name, int ok_if_exists)
3021 struct stat nst;
3022 if (!lstat(new_name, &nst)) {
3023 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3024 return 0;
3026 * A leading component of new_name might be a symlink
3027 * that is going to be removed with this patch, but
3028 * still pointing at somewhere that has the path.
3029 * In such a case, path "new_name" does not exist as
3030 * far as git is concerned.
3032 if (has_symlink_leading_path(new_name, strlen(new_name)))
3033 return 0;
3035 return error("%s: already exists in working directory", new_name);
3037 else if ((errno != ENOENT) && (errno != ENOTDIR))
3038 return error("%s: %s", new_name, strerror(errno));
3039 return 0;
3042 static int verify_index_match(struct cache_entry *ce, struct stat *st)
3044 if (S_ISGITLINK(ce->ce_mode)) {
3045 if (!S_ISDIR(st->st_mode))
3046 return -1;
3047 return 0;
3049 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3052 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3054 const char *old_name = patch->old_name;
3055 struct patch *tpatch = NULL;
3056 int stat_ret = 0;
3057 unsigned st_mode = 0;
3060 * Make sure that we do not have local modifications from the
3061 * index when we are looking at the index. Also make sure
3062 * we have the preimage file to be patched in the work tree,
3063 * unless --cached, which tells git to apply only in the index.
3065 if (!old_name)
3066 return 0;
3068 assert(patch->is_new <= 0);
3070 if (!(patch->is_copy || patch->is_rename) &&
3071 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
3072 if (was_deleted(tpatch))
3073 return error("%s: has been deleted/renamed", old_name);
3074 st_mode = tpatch->new_mode;
3075 } else if (!cached) {
3076 stat_ret = lstat(old_name, st);
3077 if (stat_ret && errno != ENOENT)
3078 return error("%s: %s", old_name, strerror(errno));
3081 if (to_be_deleted(tpatch))
3082 tpatch = NULL;
3084 if (check_index && !tpatch) {
3085 int pos = cache_name_pos(old_name, strlen(old_name));
3086 if (pos < 0) {
3087 if (patch->is_new < 0)
3088 goto is_new;
3089 return error("%s: does not exist in index", old_name);
3091 *ce = active_cache[pos];
3092 if (stat_ret < 0) {
3093 struct checkout costate;
3094 /* checkout */
3095 memset(&costate, 0, sizeof(costate));
3096 costate.base_dir = "";
3097 costate.refresh_cache = 1;
3098 if (checkout_entry(*ce, &costate, NULL) ||
3099 lstat(old_name, st))
3100 return -1;
3102 if (!cached && verify_index_match(*ce, st))
3103 return error("%s: does not match index", old_name);
3104 if (cached)
3105 st_mode = (*ce)->ce_mode;
3106 } else if (stat_ret < 0) {
3107 if (patch->is_new < 0)
3108 goto is_new;
3109 return error("%s: %s", old_name, strerror(errno));
3112 if (!cached && !tpatch)
3113 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3115 if (patch->is_new < 0)
3116 patch->is_new = 0;
3117 if (!patch->old_mode)
3118 patch->old_mode = st_mode;
3119 if ((st_mode ^ patch->old_mode) & S_IFMT)
3120 return error("%s: wrong type", old_name);
3121 if (st_mode != patch->old_mode)
3122 warning("%s has type %o, expected %o",
3123 old_name, st_mode, patch->old_mode);
3124 if (!patch->new_mode && !patch->is_delete)
3125 patch->new_mode = st_mode;
3126 return 0;
3128 is_new:
3129 patch->is_new = 1;
3130 patch->is_delete = 0;
3131 free(patch->old_name);
3132 patch->old_name = NULL;
3133 return 0;
3136 static int check_patch(struct patch *patch)
3138 struct stat st;
3139 const char *old_name = patch->old_name;
3140 const char *new_name = patch->new_name;
3141 const char *name = old_name ? old_name : new_name;
3142 struct cache_entry *ce = NULL;
3143 struct patch *tpatch;
3144 int ok_if_exists;
3145 int status;
3147 patch->rejected = 1; /* we will drop this after we succeed */
3149 status = check_preimage(patch, &ce, &st);
3150 if (status)
3151 return status;
3152 old_name = patch->old_name;
3154 if ((tpatch = in_fn_table(new_name)) &&
3155 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3157 * A type-change diff is always split into a patch to
3158 * delete old, immediately followed by a patch to
3159 * create new (see diff.c::run_diff()); in such a case
3160 * it is Ok that the entry to be deleted by the
3161 * previous patch is still in the working tree and in
3162 * the index.
3164 ok_if_exists = 1;
3165 else
3166 ok_if_exists = 0;
3168 if (new_name &&
3169 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3170 if (check_index &&
3171 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3172 !ok_if_exists)
3173 return error("%s: already exists in index", new_name);
3174 if (!cached) {
3175 int err = check_to_create_blob(new_name, ok_if_exists);
3176 if (err)
3177 return err;
3179 if (!patch->new_mode) {
3180 if (0 < patch->is_new)
3181 patch->new_mode = S_IFREG | 0644;
3182 else
3183 patch->new_mode = patch->old_mode;
3187 if (new_name && old_name) {
3188 int same = !strcmp(old_name, new_name);
3189 if (!patch->new_mode)
3190 patch->new_mode = patch->old_mode;
3191 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3192 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
3193 patch->new_mode, new_name, patch->old_mode,
3194 same ? "" : " of ", same ? "" : old_name);
3197 if (apply_data(patch, &st, ce) < 0)
3198 return error("%s: patch does not apply", name);
3199 patch->rejected = 0;
3200 return 0;
3203 static int check_patch_list(struct patch *patch)
3205 int err = 0;
3207 prepare_fn_table(patch);
3208 while (patch) {
3209 if (apply_verbosely)
3210 say_patch_name(stderr,
3211 "Checking patch ", patch, "...\n");
3212 err |= check_patch(patch);
3213 patch = patch->next;
3215 return err;
3218 /* This function tries to read the sha1 from the current index */
3219 static int get_current_sha1(const char *path, unsigned char *sha1)
3221 int pos;
3223 if (read_cache() < 0)
3224 return -1;
3225 pos = cache_name_pos(path, strlen(path));
3226 if (pos < 0)
3227 return -1;
3228 hashcpy(sha1, active_cache[pos]->sha1);
3229 return 0;
3232 /* Build an index that contains the just the files needed for a 3way merge */
3233 static void build_fake_ancestor(struct patch *list, const char *filename)
3235 struct patch *patch;
3236 struct index_state result = { NULL };
3237 int fd;
3239 /* Once we start supporting the reverse patch, it may be
3240 * worth showing the new sha1 prefix, but until then...
3242 for (patch = list; patch; patch = patch->next) {
3243 const unsigned char *sha1_ptr;
3244 unsigned char sha1[20];
3245 struct cache_entry *ce;
3246 const char *name;
3248 name = patch->old_name ? patch->old_name : patch->new_name;
3249 if (0 < patch->is_new)
3250 continue;
3251 else if (get_sha1(patch->old_sha1_prefix, sha1))
3252 /* git diff has no index line for mode/type changes */
3253 if (!patch->lines_added && !patch->lines_deleted) {
3254 if (get_current_sha1(patch->old_name, sha1))
3255 die("mode change for %s, which is not "
3256 "in current HEAD", name);
3257 sha1_ptr = sha1;
3258 } else
3259 die("sha1 information is lacking or useless "
3260 "(%s).", name);
3261 else
3262 sha1_ptr = sha1;
3264 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3265 if (!ce)
3266 die("make_cache_entry failed for path '%s'", name);
3267 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3268 die ("Could not add %s to temporary index", name);
3271 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3272 if (fd < 0 || write_index(&result, fd) || close(fd))
3273 die ("Could not write temporary index to %s", filename);
3275 discard_index(&result);
3278 static void stat_patch_list(struct patch *patch)
3280 int files, adds, dels;
3282 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3283 files++;
3284 adds += patch->lines_added;
3285 dels += patch->lines_deleted;
3286 show_stats(patch);
3289 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3292 static void numstat_patch_list(struct patch *patch)
3294 for ( ; patch; patch = patch->next) {
3295 const char *name;
3296 name = patch->new_name ? patch->new_name : patch->old_name;
3297 if (patch->is_binary)
3298 printf("-\t-\t");
3299 else
3300 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3301 write_name_quoted(name, stdout, line_termination);
3305 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3307 if (mode)
3308 printf(" %s mode %06o %s\n", newdelete, mode, name);
3309 else
3310 printf(" %s %s\n", newdelete, name);
3313 static void show_mode_change(struct patch *p, int show_name)
3315 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3316 if (show_name)
3317 printf(" mode change %06o => %06o %s\n",
3318 p->old_mode, p->new_mode, p->new_name);
3319 else
3320 printf(" mode change %06o => %06o\n",
3321 p->old_mode, p->new_mode);
3325 static void show_rename_copy(struct patch *p)
3327 const char *renamecopy = p->is_rename ? "rename" : "copy";
3328 const char *old, *new;
3330 /* Find common prefix */
3331 old = p->old_name;
3332 new = p->new_name;
3333 while (1) {
3334 const char *slash_old, *slash_new;
3335 slash_old = strchr(old, '/');
3336 slash_new = strchr(new, '/');
3337 if (!slash_old ||
3338 !slash_new ||
3339 slash_old - old != slash_new - new ||
3340 memcmp(old, new, slash_new - new))
3341 break;
3342 old = slash_old + 1;
3343 new = slash_new + 1;
3345 /* p->old_name thru old is the common prefix, and old and new
3346 * through the end of names are renames
3348 if (old != p->old_name)
3349 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3350 (int)(old - p->old_name), p->old_name,
3351 old, new, p->score);
3352 else
3353 printf(" %s %s => %s (%d%%)\n", renamecopy,
3354 p->old_name, p->new_name, p->score);
3355 show_mode_change(p, 0);
3358 static void summary_patch_list(struct patch *patch)
3360 struct patch *p;
3362 for (p = patch; p; p = p->next) {
3363 if (p->is_new)
3364 show_file_mode_name("create", p->new_mode, p->new_name);
3365 else if (p->is_delete)
3366 show_file_mode_name("delete", p->old_mode, p->old_name);
3367 else {
3368 if (p->is_rename || p->is_copy)
3369 show_rename_copy(p);
3370 else {
3371 if (p->score) {
3372 printf(" rewrite %s (%d%%)\n",
3373 p->new_name, p->score);
3374 show_mode_change(p, 0);
3376 else
3377 show_mode_change(p, 1);
3383 static void patch_stats(struct patch *patch)
3385 int lines = patch->lines_added + patch->lines_deleted;
3387 if (lines > max_change)
3388 max_change = lines;
3389 if (patch->old_name) {
3390 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3391 if (!len)
3392 len = strlen(patch->old_name);
3393 if (len > max_len)
3394 max_len = len;
3396 if (patch->new_name) {
3397 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3398 if (!len)
3399 len = strlen(patch->new_name);
3400 if (len > max_len)
3401 max_len = len;
3405 static void remove_file(struct patch *patch, int rmdir_empty)
3407 if (update_index) {
3408 if (remove_file_from_cache(patch->old_name) < 0)
3409 die("unable to remove %s from index", patch->old_name);
3411 if (!cached) {
3412 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3413 remove_path(patch->old_name);
3418 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3420 struct stat st;
3421 struct cache_entry *ce;
3422 int namelen = strlen(path);
3423 unsigned ce_size = cache_entry_size(namelen);
3425 if (!update_index)
3426 return;
3428 ce = xcalloc(1, ce_size);
3429 memcpy(ce->name, path, namelen);
3430 ce->ce_mode = create_ce_mode(mode);
3431 ce->ce_flags = namelen;
3432 if (S_ISGITLINK(mode)) {
3433 const char *s = buf;
3435 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3436 die("corrupt patch for subproject %s", path);
3437 } else {
3438 if (!cached) {
3439 if (lstat(path, &st) < 0)
3440 die_errno("unable to stat newly created file '%s'",
3441 path);
3442 fill_stat_cache_info(ce, &st);
3444 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3445 die("unable to create backing store for newly created file %s", path);
3447 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3448 die("unable to add cache entry for %s", path);
3451 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3453 int fd;
3454 struct strbuf nbuf = STRBUF_INIT;
3456 if (S_ISGITLINK(mode)) {
3457 struct stat st;
3458 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3459 return 0;
3460 return mkdir(path, 0777);
3463 if (has_symlinks && S_ISLNK(mode))
3464 /* Although buf:size is counted string, it also is NUL
3465 * terminated.
3467 return symlink(buf, path);
3469 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3470 if (fd < 0)
3471 return -1;
3473 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3474 size = nbuf.len;
3475 buf = nbuf.buf;
3477 write_or_die(fd, buf, size);
3478 strbuf_release(&nbuf);
3480 if (close(fd) < 0)
3481 die_errno("closing file '%s'", path);
3482 return 0;
3486 * We optimistically assume that the directories exist,
3487 * which is true 99% of the time anyway. If they don't,
3488 * we create them and try again.
3490 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3492 if (cached)
3493 return;
3494 if (!try_create_file(path, mode, buf, size))
3495 return;
3497 if (errno == ENOENT) {
3498 if (safe_create_leading_directories(path))
3499 return;
3500 if (!try_create_file(path, mode, buf, size))
3501 return;
3504 if (errno == EEXIST || errno == EACCES) {
3505 /* We may be trying to create a file where a directory
3506 * used to be.
3508 struct stat st;
3509 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3510 errno = EEXIST;
3513 if (errno == EEXIST) {
3514 unsigned int nr = getpid();
3516 for (;;) {
3517 char newpath[PATH_MAX];
3518 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3519 if (!try_create_file(newpath, mode, buf, size)) {
3520 if (!rename(newpath, path))
3521 return;
3522 unlink_or_warn(newpath);
3523 break;
3525 if (errno != EEXIST)
3526 break;
3527 ++nr;
3530 die_errno("unable to write file '%s' mode %o", path, mode);
3533 static void create_file(struct patch *patch)
3535 char *path = patch->new_name;
3536 unsigned mode = patch->new_mode;
3537 unsigned long size = patch->resultsize;
3538 char *buf = patch->result;
3540 if (!mode)
3541 mode = S_IFREG | 0644;
3542 create_one_file(path, mode, buf, size);
3543 add_index_file(path, mode, buf, size);
3546 /* phase zero is to remove, phase one is to create */
3547 static void write_out_one_result(struct patch *patch, int phase)
3549 if (patch->is_delete > 0) {
3550 if (phase == 0)
3551 remove_file(patch, 1);
3552 return;
3554 if (patch->is_new > 0 || patch->is_copy) {
3555 if (phase == 1)
3556 create_file(patch);
3557 return;
3560 * Rename or modification boils down to the same
3561 * thing: remove the old, write the new
3563 if (phase == 0)
3564 remove_file(patch, patch->is_rename);
3565 if (phase == 1)
3566 create_file(patch);
3569 static int write_out_one_reject(struct patch *patch)
3571 FILE *rej;
3572 char namebuf[PATH_MAX];
3573 struct fragment *frag;
3574 int cnt = 0;
3576 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3577 if (!frag->rejected)
3578 continue;
3579 cnt++;
3582 if (!cnt) {
3583 if (apply_verbosely)
3584 say_patch_name(stderr,
3585 "Applied patch ", patch, " cleanly.\n");
3586 return 0;
3589 /* This should not happen, because a removal patch that leaves
3590 * contents are marked "rejected" at the patch level.
3592 if (!patch->new_name)
3593 die("internal error");
3595 /* Say this even without --verbose */
3596 say_patch_name(stderr, "Applying patch ", patch, " with");
3597 fprintf(stderr, " %d rejects...\n", cnt);
3599 cnt = strlen(patch->new_name);
3600 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3601 cnt = ARRAY_SIZE(namebuf) - 5;
3602 warning("truncating .rej filename to %.*s.rej",
3603 cnt - 1, patch->new_name);
3605 memcpy(namebuf, patch->new_name, cnt);
3606 memcpy(namebuf + cnt, ".rej", 5);
3608 rej = fopen(namebuf, "w");
3609 if (!rej)
3610 return error("cannot open %s: %s", namebuf, strerror(errno));
3612 /* Normal git tools never deal with .rej, so do not pretend
3613 * this is a git patch by saying --git nor give extended
3614 * headers. While at it, maybe please "kompare" that wants
3615 * the trailing TAB and some garbage at the end of line ;-).
3617 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3618 patch->new_name, patch->new_name);
3619 for (cnt = 1, frag = patch->fragments;
3620 frag;
3621 cnt++, frag = frag->next) {
3622 if (!frag->rejected) {
3623 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3624 continue;
3626 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3627 fprintf(rej, "%.*s", frag->size, frag->patch);
3628 if (frag->patch[frag->size-1] != '\n')
3629 fputc('\n', rej);
3631 fclose(rej);
3632 return -1;
3635 static int write_out_results(struct patch *list)
3637 int phase;
3638 int errs = 0;
3639 struct patch *l;
3641 for (phase = 0; phase < 2; phase++) {
3642 l = list;
3643 while (l) {
3644 if (l->rejected)
3645 errs = 1;
3646 else {
3647 write_out_one_result(l, phase);
3648 if (phase == 1 && write_out_one_reject(l))
3649 errs = 1;
3651 l = l->next;
3654 return errs;
3657 static struct lock_file lock_file;
3659 static struct string_list limit_by_name;
3660 static int has_include;
3661 static void add_name_limit(const char *name, int exclude)
3663 struct string_list_item *it;
3665 it = string_list_append(&limit_by_name, name);
3666 it->util = exclude ? NULL : (void *) 1;
3669 static int use_patch(struct patch *p)
3671 const char *pathname = p->new_name ? p->new_name : p->old_name;
3672 int i;
3674 /* Paths outside are not touched regardless of "--include" */
3675 if (0 < prefix_length) {
3676 int pathlen = strlen(pathname);
3677 if (pathlen <= prefix_length ||
3678 memcmp(prefix, pathname, prefix_length))
3679 return 0;
3682 /* See if it matches any of exclude/include rule */
3683 for (i = 0; i < limit_by_name.nr; i++) {
3684 struct string_list_item *it = &limit_by_name.items[i];
3685 if (!fnmatch(it->string, pathname, 0))
3686 return (it->util != NULL);
3690 * If we had any include, a path that does not match any rule is
3691 * not used. Otherwise, we saw bunch of exclude rules (or none)
3692 * and such a path is used.
3694 return !has_include;
3698 static void prefix_one(char **name)
3700 char *old_name = *name;
3701 if (!old_name)
3702 return;
3703 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3704 free(old_name);
3707 static void prefix_patches(struct patch *p)
3709 if (!prefix || p->is_toplevel_relative)
3710 return;
3711 for ( ; p; p = p->next) {
3712 prefix_one(&p->new_name);
3713 prefix_one(&p->old_name);
3717 #define INACCURATE_EOF (1<<0)
3718 #define RECOUNT (1<<1)
3720 static int apply_patch(int fd, const char *filename, int options)
3722 size_t offset;
3723 struct strbuf buf = STRBUF_INIT;
3724 struct patch *list = NULL, **listp = &list;
3725 int skipped_patch = 0;
3727 memset(&fn_table, 0, sizeof(struct string_list));
3728 patch_input_file = filename;
3729 read_patch_file(&buf, fd);
3730 offset = 0;
3731 while (offset < buf.len) {
3732 struct patch *patch;
3733 int nr;
3735 patch = xcalloc(1, sizeof(*patch));
3736 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3737 patch->recount = !!(options & RECOUNT);
3738 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3739 if (nr < 0)
3740 break;
3741 if (apply_in_reverse)
3742 reverse_patches(patch);
3743 if (prefix)
3744 prefix_patches(patch);
3745 if (use_patch(patch)) {
3746 patch_stats(patch);
3747 *listp = patch;
3748 listp = &patch->next;
3750 else {
3751 free_patch(patch);
3752 skipped_patch++;
3754 offset += nr;
3757 if (!list && !skipped_patch)
3758 die("unrecognized input");
3760 if (whitespace_error && (ws_error_action == die_on_ws_error))
3761 apply = 0;
3763 update_index = check_index && apply;
3764 if (update_index && newfd < 0)
3765 newfd = hold_locked_index(&lock_file, 1);
3767 if (check_index) {
3768 if (read_cache() < 0)
3769 die("unable to read index file");
3772 if ((check || apply) &&
3773 check_patch_list(list) < 0 &&
3774 !apply_with_reject)
3775 exit(1);
3777 if (apply && write_out_results(list))
3778 exit(1);
3780 if (fake_ancestor)
3781 build_fake_ancestor(list, fake_ancestor);
3783 if (diffstat)
3784 stat_patch_list(list);
3786 if (numstat)
3787 numstat_patch_list(list);
3789 if (summary)
3790 summary_patch_list(list);
3792 free_patch_list(list);
3793 strbuf_release(&buf);
3794 return 0;
3797 static int git_apply_config(const char *var, const char *value, void *cb)
3799 if (!strcmp(var, "apply.whitespace"))
3800 return git_config_string(&apply_default_whitespace, var, value);
3801 else if (!strcmp(var, "apply.ignorewhitespace"))
3802 return git_config_string(&apply_default_ignorewhitespace, var, value);
3803 return git_default_config(var, value, cb);
3806 static int option_parse_exclude(const struct option *opt,
3807 const char *arg, int unset)
3809 add_name_limit(arg, 1);
3810 return 0;
3813 static int option_parse_include(const struct option *opt,
3814 const char *arg, int unset)
3816 add_name_limit(arg, 0);
3817 has_include = 1;
3818 return 0;
3821 static int option_parse_p(const struct option *opt,
3822 const char *arg, int unset)
3824 p_value = atoi(arg);
3825 p_value_known = 1;
3826 return 0;
3829 static int option_parse_z(const struct option *opt,
3830 const char *arg, int unset)
3832 if (unset)
3833 line_termination = '\n';
3834 else
3835 line_termination = 0;
3836 return 0;
3839 static int option_parse_space_change(const struct option *opt,
3840 const char *arg, int unset)
3842 if (unset)
3843 ws_ignore_action = ignore_ws_none;
3844 else
3845 ws_ignore_action = ignore_ws_change;
3846 return 0;
3849 static int option_parse_whitespace(const struct option *opt,
3850 const char *arg, int unset)
3852 const char **whitespace_option = opt->value;
3854 *whitespace_option = arg;
3855 parse_whitespace_option(arg);
3856 return 0;
3859 static int option_parse_directory(const struct option *opt,
3860 const char *arg, int unset)
3862 root_len = strlen(arg);
3863 if (root_len && arg[root_len - 1] != '/') {
3864 char *new_root;
3865 root = new_root = xmalloc(root_len + 2);
3866 strcpy(new_root, arg);
3867 strcpy(new_root + root_len++, "/");
3868 } else
3869 root = arg;
3870 return 0;
3873 int cmd_apply(int argc, const char **argv, const char *prefix_)
3875 int i;
3876 int errs = 0;
3877 int is_not_gitdir = !startup_info->have_repository;
3878 int force_apply = 0;
3880 const char *whitespace_option = NULL;
3882 struct option builtin_apply_options[] = {
3883 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3884 "don't apply changes matching the given path",
3885 0, option_parse_exclude },
3886 { OPTION_CALLBACK, 0, "include", NULL, "path",
3887 "apply changes matching the given path",
3888 0, option_parse_include },
3889 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3890 "remove <num> leading slashes from traditional diff paths",
3891 0, option_parse_p },
3892 OPT_BOOLEAN(0, "no-add", &no_add,
3893 "ignore additions made by the patch"),
3894 OPT_BOOLEAN(0, "stat", &diffstat,
3895 "instead of applying the patch, output diffstat for the input"),
3896 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
3897 OPT_NOOP_NOARG(0, "binary"),
3898 OPT_BOOLEAN(0, "numstat", &numstat,
3899 "shows number of added and deleted lines in decimal notation"),
3900 OPT_BOOLEAN(0, "summary", &summary,
3901 "instead of applying the patch, output a summary for the input"),
3902 OPT_BOOLEAN(0, "check", &check,
3903 "instead of applying the patch, see if the patch is applicable"),
3904 OPT_BOOLEAN(0, "index", &check_index,
3905 "make sure the patch is applicable to the current index"),
3906 OPT_BOOLEAN(0, "cached", &cached,
3907 "apply a patch without touching the working tree"),
3908 OPT_BOOLEAN(0, "apply", &force_apply,
3909 "also apply the patch (use with --stat/--summary/--check)"),
3910 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3911 "build a temporary index based on embedded index information"),
3912 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3913 "paths are separated with NUL character",
3914 PARSE_OPT_NOARG, option_parse_z },
3915 OPT_INTEGER('C', NULL, &p_context,
3916 "ensure at least <n> lines of context match"),
3917 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3918 "detect new or modified lines that have whitespace errors",
3919 0, option_parse_whitespace },
3920 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3921 "ignore changes in whitespace when finding context",
3922 PARSE_OPT_NOARG, option_parse_space_change },
3923 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3924 "ignore changes in whitespace when finding context",
3925 PARSE_OPT_NOARG, option_parse_space_change },
3926 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3927 "apply the patch in reverse"),
3928 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3929 "don't expect at least one line of context"),
3930 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3931 "leave the rejected hunks in corresponding *.rej files"),
3932 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
3933 "allow overlapping hunks"),
3934 OPT__VERBOSE(&apply_verbosely, "be verbose"),
3935 OPT_BIT(0, "inaccurate-eof", &options,
3936 "tolerate incorrectly detected missing new-line at the end of file",
3937 INACCURATE_EOF),
3938 OPT_BIT(0, "recount", &options,
3939 "do not trust the line counts in the hunk headers",
3940 RECOUNT),
3941 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3942 "prepend <root> to all filenames",
3943 0, option_parse_directory },
3944 OPT_END()
3947 prefix = prefix_;
3948 prefix_length = prefix ? strlen(prefix) : 0;
3949 git_config(git_apply_config, NULL);
3950 if (apply_default_whitespace)
3951 parse_whitespace_option(apply_default_whitespace);
3952 if (apply_default_ignorewhitespace)
3953 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3955 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3956 apply_usage, 0);
3958 if (apply_with_reject)
3959 apply = apply_verbosely = 1;
3960 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3961 apply = 0;
3962 if (check_index && is_not_gitdir)
3963 die("--index outside a repository");
3964 if (cached) {
3965 if (is_not_gitdir)
3966 die("--cached outside a repository");
3967 check_index = 1;
3969 for (i = 0; i < argc; i++) {
3970 const char *arg = argv[i];
3971 int fd;
3973 if (!strcmp(arg, "-")) {
3974 errs |= apply_patch(0, "<stdin>", options);
3975 read_stdin = 0;
3976 continue;
3977 } else if (0 < prefix_length)
3978 arg = prefix_filename(prefix, prefix_length, arg);
3980 fd = open(arg, O_RDONLY);
3981 if (fd < 0)
3982 die_errno("can't open patch '%s'", arg);
3983 read_stdin = 0;
3984 set_default_whitespace_mode(whitespace_option);
3985 errs |= apply_patch(fd, arg, options);
3986 close(fd);
3988 set_default_whitespace_mode(whitespace_option);
3989 if (read_stdin)
3990 errs |= apply_patch(0, "<stdin>", options);
3991 if (whitespace_error) {
3992 if (squelch_whitespace_errors &&
3993 squelch_whitespace_errors < whitespace_error) {
3994 int squelched =
3995 whitespace_error - squelch_whitespace_errors;
3996 warning("squelched %d "
3997 "whitespace error%s",
3998 squelched,
3999 squelched == 1 ? "" : "s");
4001 if (ws_error_action == die_on_ws_error)
4002 die("%d line%s add%s whitespace errors.",
4003 whitespace_error,
4004 whitespace_error == 1 ? "" : "s",
4005 whitespace_error == 1 ? "s" : "");
4006 if (applied_after_fixing_ws && apply)
4007 warning("%d line%s applied after"
4008 " fixing whitespace errors.",
4009 applied_after_fixing_ws,
4010 applied_after_fixing_ws == 1 ? "" : "s");
4011 else if (whitespace_error)
4012 warning("%d line%s add%s whitespace errors.",
4013 whitespace_error,
4014 whitespace_error == 1 ? "" : "s",
4015 whitespace_error == 1 ? "s" : "");
4018 if (update_index) {
4019 if (write_cache(newfd, active_cache, active_nr) ||
4020 commit_locked_index(&lock_file))
4021 die("Unable to write new index file");
4024 return !!errs;