apply: Recognize epoch timestamps with : in the timezone
[git/dscho.git] / builtin / apply.c
blob000d3e529722f73321f87084a457d878de037704
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 no_add;
47 static const char *fake_ancestor;
48 static int line_termination = '\n';
49 static unsigned int p_context = UINT_MAX;
50 static const char * const apply_usage[] = {
51 "git apply [options] [<patch>...]",
52 NULL
55 static enum ws_error_action {
56 nowarn_ws_error,
57 warn_on_ws_error,
58 die_on_ws_error,
59 correct_ws_error
60 } ws_error_action = warn_on_ws_error;
61 static int whitespace_error;
62 static int squelch_whitespace_errors = 5;
63 static int applied_after_fixing_ws;
65 static enum ws_ignore {
66 ignore_ws_none,
67 ignore_ws_change
68 } ws_ignore_action = ignore_ws_none;
71 static const char *patch_input_file;
72 static const char *root;
73 static int root_len;
74 static int read_stdin = 1;
75 static int options;
77 static void parse_whitespace_option(const char *option)
79 if (!option) {
80 ws_error_action = warn_on_ws_error;
81 return;
83 if (!strcmp(option, "warn")) {
84 ws_error_action = warn_on_ws_error;
85 return;
87 if (!strcmp(option, "nowarn")) {
88 ws_error_action = nowarn_ws_error;
89 return;
91 if (!strcmp(option, "error")) {
92 ws_error_action = die_on_ws_error;
93 return;
95 if (!strcmp(option, "error-all")) {
96 ws_error_action = die_on_ws_error;
97 squelch_whitespace_errors = 0;
98 return;
100 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
101 ws_error_action = correct_ws_error;
102 return;
104 die("unrecognized whitespace option '%s'", option);
107 static void parse_ignorewhitespace_option(const char *option)
109 if (!option || !strcmp(option, "no") ||
110 !strcmp(option, "false") || !strcmp(option, "never") ||
111 !strcmp(option, "none")) {
112 ws_ignore_action = ignore_ws_none;
113 return;
115 if (!strcmp(option, "change")) {
116 ws_ignore_action = ignore_ws_change;
117 return;
119 die("unrecognized whitespace ignore option '%s'", option);
122 static void set_default_whitespace_mode(const char *whitespace_option)
124 if (!whitespace_option && !apply_default_whitespace)
125 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
129 * For "diff-stat" like behaviour, we keep track of the biggest change
130 * we've seen, and the longest filename. That allows us to do simple
131 * scaling.
133 static int max_change, max_len;
136 * Various "current state", notably line numbers and what
137 * file (and how) we're patching right now.. The "is_xxxx"
138 * things are flags, where -1 means "don't know yet".
140 static int linenr = 1;
143 * This represents one "hunk" from a patch, starting with
144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
145 * patch text is pointed at by patch, and its byte length
146 * is stored in size. leading and trailing are the number
147 * of context lines.
149 struct fragment {
150 unsigned long leading, trailing;
151 unsigned long oldpos, oldlines;
152 unsigned long newpos, newlines;
153 const char *patch;
154 int size;
155 int rejected;
156 int linenr;
157 struct fragment *next;
161 * When dealing with a binary patch, we reuse "leading" field
162 * to store the type of the binary hunk, either deflated "delta"
163 * or deflated "literal".
165 #define binary_patch_method leading
166 #define BINARY_DELTA_DEFLATED 1
167 #define BINARY_LITERAL_DEFLATED 2
170 * This represents a "patch" to a file, both metainfo changes
171 * such as creation/deletion, filemode and content changes represented
172 * as a series of fragments.
174 struct patch {
175 char *new_name, *old_name, *def_name;
176 unsigned int old_mode, new_mode;
177 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
178 int rejected;
179 unsigned ws_rule;
180 unsigned long deflate_origlen;
181 int lines_added, lines_deleted;
182 int score;
183 unsigned int is_toplevel_relative:1;
184 unsigned int inaccurate_eof:1;
185 unsigned int is_binary:1;
186 unsigned int is_copy:1;
187 unsigned int is_rename:1;
188 unsigned int recount:1;
189 struct fragment *fragments;
190 char *result;
191 size_t resultsize;
192 char old_sha1_prefix[41];
193 char new_sha1_prefix[41];
194 struct patch *next;
198 * A line in a file, len-bytes long (includes the terminating LF,
199 * except for an incomplete line at the end if the file ends with
200 * one), and its contents hashes to 'hash'.
202 struct line {
203 size_t len;
204 unsigned hash : 24;
205 unsigned flag : 8;
206 #define LINE_COMMON 1
210 * This represents a "file", which is an array of "lines".
212 struct image {
213 char *buf;
214 size_t len;
215 size_t nr;
216 size_t alloc;
217 struct line *line_allocated;
218 struct line *line;
222 * Records filenames that have been touched, in order to handle
223 * the case where more than one patches touch the same file.
226 static struct string_list fn_table;
228 static uint32_t hash_line(const char *cp, size_t len)
230 size_t i;
231 uint32_t h;
232 for (i = 0, h = 0; i < len; i++) {
233 if (!isspace(cp[i])) {
234 h = h * 3 + (cp[i] & 0xff);
237 return h;
241 * Compare lines s1 of length n1 and s2 of length n2, ignoring
242 * whitespace difference. Returns 1 if they match, 0 otherwise
244 static int fuzzy_matchlines(const char *s1, size_t n1,
245 const char *s2, size_t n2)
247 const char *last1 = s1 + n1 - 1;
248 const char *last2 = s2 + n2 - 1;
249 int result = 0;
251 if (n1 < 0 || n2 < 0)
252 return 0;
254 /* ignore line endings */
255 while ((*last1 == '\r') || (*last1 == '\n'))
256 last1--;
257 while ((*last2 == '\r') || (*last2 == '\n'))
258 last2--;
260 /* skip leading whitespace */
261 while (isspace(*s1) && (s1 <= last1))
262 s1++;
263 while (isspace(*s2) && (s2 <= last2))
264 s2++;
265 /* early return if both lines are empty */
266 if ((s1 > last1) && (s2 > last2))
267 return 1;
268 while (!result) {
269 result = *s1++ - *s2++;
271 * Skip whitespace inside. We check for whitespace on
272 * both buffers because we don't want "a b" to match
273 * "ab"
275 if (isspace(*s1) && isspace(*s2)) {
276 while (isspace(*s1) && s1 <= last1)
277 s1++;
278 while (isspace(*s2) && s2 <= last2)
279 s2++;
282 * If we reached the end on one side only,
283 * lines don't match
285 if (
286 ((s2 > last2) && (s1 <= last1)) ||
287 ((s1 > last1) && (s2 <= last2)))
288 return 0;
289 if ((s1 > last1) && (s2 > last2))
290 break;
293 return !result;
296 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
298 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
299 img->line_allocated[img->nr].len = len;
300 img->line_allocated[img->nr].hash = hash_line(bol, len);
301 img->line_allocated[img->nr].flag = flag;
302 img->nr++;
305 static void prepare_image(struct image *image, char *buf, size_t len,
306 int prepare_linetable)
308 const char *cp, *ep;
310 memset(image, 0, sizeof(*image));
311 image->buf = buf;
312 image->len = len;
314 if (!prepare_linetable)
315 return;
317 ep = image->buf + image->len;
318 cp = image->buf;
319 while (cp < ep) {
320 const char *next;
321 for (next = cp; next < ep && *next != '\n'; next++)
323 if (next < ep)
324 next++;
325 add_line_info(image, cp, next - cp, 0);
326 cp = next;
328 image->line = image->line_allocated;
331 static void clear_image(struct image *image)
333 free(image->buf);
334 image->buf = NULL;
335 image->len = 0;
338 static void say_patch_name(FILE *output, const char *pre,
339 struct patch *patch, const char *post)
341 fputs(pre, output);
342 if (patch->old_name && patch->new_name &&
343 strcmp(patch->old_name, patch->new_name)) {
344 quote_c_style(patch->old_name, NULL, output, 0);
345 fputs(" => ", output);
346 quote_c_style(patch->new_name, NULL, output, 0);
347 } else {
348 const char *n = patch->new_name;
349 if (!n)
350 n = patch->old_name;
351 quote_c_style(n, NULL, output, 0);
353 fputs(post, output);
356 #define CHUNKSIZE (8192)
357 #define SLOP (16)
359 static void read_patch_file(struct strbuf *sb, int fd)
361 if (strbuf_read(sb, fd, 0) < 0)
362 die_errno("git apply: failed to read");
365 * Make sure that we have some slop in the buffer
366 * so that we can do speculative "memcmp" etc, and
367 * see to it that it is NUL-filled.
369 strbuf_grow(sb, SLOP);
370 memset(sb->buf + sb->len, 0, SLOP);
373 static unsigned long linelen(const char *buffer, unsigned long size)
375 unsigned long len = 0;
376 while (size--) {
377 len++;
378 if (*buffer++ == '\n')
379 break;
381 return len;
384 static int is_dev_null(const char *str)
386 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
389 #define TERM_SPACE 1
390 #define TERM_TAB 2
392 static int name_terminate(const char *name, int namelen, int c, int terminate)
394 if (c == ' ' && !(terminate & TERM_SPACE))
395 return 0;
396 if (c == '\t' && !(terminate & TERM_TAB))
397 return 0;
399 return 1;
402 /* remove double slashes to make --index work with such filenames */
403 static char *squash_slash(char *name)
405 int i = 0, j = 0;
407 if (!name)
408 return NULL;
410 while (name[i]) {
411 if ((name[j++] = name[i++]) == '/')
412 while (name[i] == '/')
413 i++;
415 name[j] = '\0';
416 return name;
419 static char *find_name_gnu(const char *line, char *def, int p_value)
421 struct strbuf name = STRBUF_INIT;
422 char *cp;
425 * Proposed "new-style" GNU patch/diff format; see
426 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
428 if (unquote_c_style(&name, line, NULL)) {
429 strbuf_release(&name);
430 return NULL;
433 for (cp = name.buf; p_value; p_value--) {
434 cp = strchr(cp, '/');
435 if (!cp) {
436 strbuf_release(&name);
437 return NULL;
439 cp++;
442 /* name can later be freed, so we need
443 * to memmove, not just return cp
445 strbuf_remove(&name, 0, cp - name.buf);
446 free(def);
447 if (root)
448 strbuf_insert(&name, 0, root, root_len);
449 return squash_slash(strbuf_detach(&name, NULL));
452 static size_t tz_len(const char *line, size_t len)
454 const char *tz, *p;
456 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
457 return 0;
458 tz = line + len - strlen(" +0500");
460 if (tz[1] != '+' && tz[1] != '-')
461 return 0;
463 for (p = tz + 2; p != line + len; p++)
464 if (!isdigit(*p))
465 return 0;
467 return line + len - tz;
470 static size_t date_len(const char *line, size_t len)
472 const char *date, *p;
474 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
475 return 0;
476 p = date = line + len - strlen("72-02-05");
478 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
479 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
480 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
481 return 0;
483 if (date - line >= strlen("19") &&
484 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
485 date -= strlen("19");
487 return line + len - date;
490 static size_t short_time_len(const char *line, size_t len)
492 const char *time, *p;
494 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
495 return 0;
496 p = time = line + len - strlen(" 07:01:32");
498 /* Permit 1-digit hours? */
499 if (*p++ != ' ' ||
500 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
501 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
502 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
503 return 0;
505 return line + len - time;
508 static size_t fractional_time_len(const char *line, size_t len)
510 const char *p;
511 size_t n;
513 /* Expected format: 19:41:17.620000023 */
514 if (!len || !isdigit(line[len - 1]))
515 return 0;
516 p = line + len - 1;
518 /* Fractional seconds. */
519 while (p > line && isdigit(*p))
520 p--;
521 if (*p != '.')
522 return 0;
524 /* Hours, minutes, and whole seconds. */
525 n = short_time_len(line, p - line);
526 if (!n)
527 return 0;
529 return line + len - p + n;
532 static size_t trailing_spaces_len(const char *line, size_t len)
534 const char *p;
536 /* Expected format: ' ' x (1 or more) */
537 if (!len || line[len - 1] != ' ')
538 return 0;
540 p = line + len;
541 while (p != line) {
542 p--;
543 if (*p != ' ')
544 return line + len - (p + 1);
547 /* All spaces! */
548 return len;
551 static size_t diff_timestamp_len(const char *line, size_t len)
553 const char *end = line + len;
554 size_t n;
557 * Posix: 2010-07-05 19:41:17
558 * GNU: 2010-07-05 19:41:17.620000023 -0500
561 if (!isdigit(end[-1]))
562 return 0;
564 n = tz_len(line, end - line);
565 end -= n;
567 n = short_time_len(line, end - line);
568 if (!n)
569 n = fractional_time_len(line, end - line);
570 end -= n;
572 n = date_len(line, end - line);
573 if (!n) /* No date. Too bad. */
574 return 0;
575 end -= n;
577 if (end == line) /* No space before date. */
578 return 0;
579 if (end[-1] == '\t') { /* Success! */
580 end--;
581 return line + len - end;
583 if (end[-1] != ' ') /* No space before date. */
584 return 0;
586 /* Whitespace damage. */
587 end -= trailing_spaces_len(line, end - line);
588 return line + len - end;
591 static char *find_name_common(const char *line, char *def, int p_value,
592 const char *end, int terminate)
594 int len;
595 const char *start = NULL;
597 if (p_value == 0)
598 start = line;
599 while (line != end) {
600 char c = *line;
602 if (!end && isspace(c)) {
603 if (c == '\n')
604 break;
605 if (name_terminate(start, line-start, c, terminate))
606 break;
608 line++;
609 if (c == '/' && !--p_value)
610 start = line;
612 if (!start)
613 return squash_slash(def);
614 len = line - start;
615 if (!len)
616 return squash_slash(def);
619 * Generally we prefer the shorter name, especially
620 * if the other one is just a variation of that with
621 * something else tacked on to the end (ie "file.orig"
622 * or "file~").
624 if (def) {
625 int deflen = strlen(def);
626 if (deflen < len && !strncmp(start, def, deflen))
627 return squash_slash(def);
628 free(def);
631 if (root) {
632 char *ret = xmalloc(root_len + len + 1);
633 strcpy(ret, root);
634 memcpy(ret + root_len, start, len);
635 ret[root_len + len] = '\0';
636 return squash_slash(ret);
639 return squash_slash(xmemdupz(start, len));
642 static char *find_name(const char *line, char *def, int p_value, int terminate)
644 if (*line == '"') {
645 char *name = find_name_gnu(line, def, p_value);
646 if (name)
647 return name;
650 return find_name_common(line, def, p_value, NULL, terminate);
653 static char *find_name_traditional(const char *line, char *def, int p_value)
655 size_t len = strlen(line);
656 size_t date_len;
658 if (*line == '"') {
659 char *name = find_name_gnu(line, def, p_value);
660 if (name)
661 return name;
664 len = strchrnul(line, '\n') - line;
665 date_len = diff_timestamp_len(line, len);
666 if (!date_len)
667 return find_name_common(line, def, p_value, NULL, TERM_TAB);
668 len -= date_len;
670 return find_name_common(line, def, p_value, line + len, 0);
673 static int count_slashes(const char *cp)
675 int cnt = 0;
676 char ch;
678 while ((ch = *cp++))
679 if (ch == '/')
680 cnt++;
681 return cnt;
685 * Given the string after "--- " or "+++ ", guess the appropriate
686 * p_value for the given patch.
688 static int guess_p_value(const char *nameline)
690 char *name, *cp;
691 int val = -1;
693 if (is_dev_null(nameline))
694 return -1;
695 name = find_name_traditional(nameline, NULL, 0);
696 if (!name)
697 return -1;
698 cp = strchr(name, '/');
699 if (!cp)
700 val = 0;
701 else if (prefix) {
703 * Does it begin with "a/$our-prefix" and such? Then this is
704 * very likely to apply to our directory.
706 if (!strncmp(name, prefix, prefix_length))
707 val = count_slashes(prefix);
708 else {
709 cp++;
710 if (!strncmp(cp, prefix, prefix_length))
711 val = count_slashes(prefix) + 1;
714 free(name);
715 return val;
719 * Does the ---/+++ line has the POSIX timestamp after the last HT?
720 * GNU diff puts epoch there to signal a creation/deletion event. Is
721 * this such a timestamp?
723 static int has_epoch_timestamp(const char *nameline)
726 * We are only interested in epoch timestamp; any non-zero
727 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
728 * For the same reason, the date must be either 1969-12-31 or
729 * 1970-01-01, and the seconds part must be "00".
731 const char stamp_regexp[] =
732 "^(1969-12-31|1970-01-01)"
734 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
736 "([-+][0-2][0-9]:?[0-5][0-9])\n";
737 const char *timestamp = NULL, *cp, *colon;
738 static regex_t *stamp;
739 regmatch_t m[10];
740 int zoneoffset;
741 int hourminute;
742 int status;
744 for (cp = nameline; *cp != '\n'; cp++) {
745 if (*cp == '\t')
746 timestamp = cp + 1;
748 if (!timestamp)
749 return 0;
750 if (!stamp) {
751 stamp = xmalloc(sizeof(*stamp));
752 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
753 warning("Cannot prepare timestamp regexp %s",
754 stamp_regexp);
755 return 0;
759 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
760 if (status) {
761 if (status != REG_NOMATCH)
762 warning("regexec returned %d for input: %s",
763 status, timestamp);
764 return 0;
767 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
768 if (*colon == ':')
769 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
770 else
771 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
772 if (timestamp[m[3].rm_so] == '-')
773 zoneoffset = -zoneoffset;
776 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
777 * (west of GMT) or 1970-01-01 (east of GMT)
779 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
780 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
781 return 0;
783 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
784 strtol(timestamp + 14, NULL, 10) -
785 zoneoffset);
787 return ((zoneoffset < 0 && hourminute == 1440) ||
788 (0 <= zoneoffset && !hourminute));
792 * Get the name etc info from the ---/+++ lines of a traditional patch header
794 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
795 * files, we can happily check the index for a match, but for creating a
796 * new file we should try to match whatever "patch" does. I have no idea.
798 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
800 char *name;
802 first += 4; /* skip "--- " */
803 second += 4; /* skip "+++ " */
804 if (!p_value_known) {
805 int p, q;
806 p = guess_p_value(first);
807 q = guess_p_value(second);
808 if (p < 0) p = q;
809 if (0 <= p && p == q) {
810 p_value = p;
811 p_value_known = 1;
814 if (is_dev_null(first)) {
815 patch->is_new = 1;
816 patch->is_delete = 0;
817 name = find_name_traditional(second, NULL, p_value);
818 patch->new_name = name;
819 } else if (is_dev_null(second)) {
820 patch->is_new = 0;
821 patch->is_delete = 1;
822 name = find_name_traditional(first, NULL, p_value);
823 patch->old_name = name;
824 } else {
825 name = find_name_traditional(first, NULL, p_value);
826 name = find_name_traditional(second, name, p_value);
827 if (has_epoch_timestamp(first)) {
828 patch->is_new = 1;
829 patch->is_delete = 0;
830 patch->new_name = name;
831 } else if (has_epoch_timestamp(second)) {
832 patch->is_new = 0;
833 patch->is_delete = 1;
834 patch->old_name = name;
835 } else {
836 patch->old_name = patch->new_name = name;
839 if (!name)
840 die("unable to find filename in patch at line %d", linenr);
843 static int gitdiff_hdrend(const char *line, struct patch *patch)
845 return -1;
849 * We're anal about diff header consistency, to make
850 * sure that we don't end up having strange ambiguous
851 * patches floating around.
853 * As a result, gitdiff_{old|new}name() will check
854 * their names against any previous information, just
855 * to make sure..
857 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
859 if (!orig_name && !isnull)
860 return find_name(line, NULL, p_value, TERM_TAB);
862 if (orig_name) {
863 int len;
864 const char *name;
865 char *another;
866 name = orig_name;
867 len = strlen(name);
868 if (isnull)
869 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
870 another = find_name(line, NULL, p_value, TERM_TAB);
871 if (!another || memcmp(another, name, len + 1))
872 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
873 free(another);
874 return orig_name;
876 else {
877 /* expect "/dev/null" */
878 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
879 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
880 return NULL;
884 static int gitdiff_oldname(const char *line, struct patch *patch)
886 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
887 return 0;
890 static int gitdiff_newname(const char *line, struct patch *patch)
892 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
893 return 0;
896 static int gitdiff_oldmode(const char *line, struct patch *patch)
898 patch->old_mode = strtoul(line, NULL, 8);
899 return 0;
902 static int gitdiff_newmode(const char *line, struct patch *patch)
904 patch->new_mode = strtoul(line, NULL, 8);
905 return 0;
908 static int gitdiff_delete(const char *line, struct patch *patch)
910 patch->is_delete = 1;
911 patch->old_name = patch->def_name;
912 return gitdiff_oldmode(line, patch);
915 static int gitdiff_newfile(const char *line, struct patch *patch)
917 patch->is_new = 1;
918 patch->new_name = patch->def_name;
919 return gitdiff_newmode(line, patch);
922 static int gitdiff_copysrc(const char *line, struct patch *patch)
924 patch->is_copy = 1;
925 patch->old_name = find_name(line, NULL, 0, 0);
926 return 0;
929 static int gitdiff_copydst(const char *line, struct patch *patch)
931 patch->is_copy = 1;
932 patch->new_name = find_name(line, NULL, 0, 0);
933 return 0;
936 static int gitdiff_renamesrc(const char *line, struct patch *patch)
938 patch->is_rename = 1;
939 patch->old_name = find_name(line, NULL, 0, 0);
940 return 0;
943 static int gitdiff_renamedst(const char *line, struct patch *patch)
945 patch->is_rename = 1;
946 patch->new_name = find_name(line, NULL, 0, 0);
947 return 0;
950 static int gitdiff_similarity(const char *line, struct patch *patch)
952 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
953 patch->score = 0;
954 return 0;
957 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
959 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
960 patch->score = 0;
961 return 0;
964 static int gitdiff_index(const char *line, struct patch *patch)
967 * index line is N hexadecimal, "..", N hexadecimal,
968 * and optional space with octal mode.
970 const char *ptr, *eol;
971 int len;
973 ptr = strchr(line, '.');
974 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
975 return 0;
976 len = ptr - line;
977 memcpy(patch->old_sha1_prefix, line, len);
978 patch->old_sha1_prefix[len] = 0;
980 line = ptr + 2;
981 ptr = strchr(line, ' ');
982 eol = strchr(line, '\n');
984 if (!ptr || eol < ptr)
985 ptr = eol;
986 len = ptr - line;
988 if (40 < len)
989 return 0;
990 memcpy(patch->new_sha1_prefix, line, len);
991 patch->new_sha1_prefix[len] = 0;
992 if (*ptr == ' ')
993 patch->old_mode = strtoul(ptr+1, NULL, 8);
994 return 0;
998 * This is normal for a diff that doesn't change anything: we'll fall through
999 * into the next diff. Tell the parser to break out.
1001 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1003 return -1;
1006 static const char *stop_at_slash(const char *line, int llen)
1008 int nslash = p_value;
1009 int i;
1011 for (i = 0; i < llen; i++) {
1012 int ch = line[i];
1013 if (ch == '/' && --nslash <= 0)
1014 return &line[i];
1016 return NULL;
1020 * This is to extract the same name that appears on "diff --git"
1021 * line. We do not find and return anything if it is a rename
1022 * patch, and it is OK because we will find the name elsewhere.
1023 * We need to reliably find name only when it is mode-change only,
1024 * creation or deletion of an empty file. In any of these cases,
1025 * both sides are the same name under a/ and b/ respectively.
1027 static char *git_header_name(char *line, int llen)
1029 const char *name;
1030 const char *second = NULL;
1031 size_t len;
1033 line += strlen("diff --git ");
1034 llen -= strlen("diff --git ");
1036 if (*line == '"') {
1037 const char *cp;
1038 struct strbuf first = STRBUF_INIT;
1039 struct strbuf sp = STRBUF_INIT;
1041 if (unquote_c_style(&first, line, &second))
1042 goto free_and_fail1;
1044 /* advance to the first slash */
1045 cp = stop_at_slash(first.buf, first.len);
1046 /* we do not accept absolute paths */
1047 if (!cp || cp == first.buf)
1048 goto free_and_fail1;
1049 strbuf_remove(&first, 0, cp + 1 - first.buf);
1052 * second points at one past closing dq of name.
1053 * find the second name.
1055 while ((second < line + llen) && isspace(*second))
1056 second++;
1058 if (line + llen <= second)
1059 goto free_and_fail1;
1060 if (*second == '"') {
1061 if (unquote_c_style(&sp, second, NULL))
1062 goto free_and_fail1;
1063 cp = stop_at_slash(sp.buf, sp.len);
1064 if (!cp || cp == sp.buf)
1065 goto free_and_fail1;
1066 /* They must match, otherwise ignore */
1067 if (strcmp(cp + 1, first.buf))
1068 goto free_and_fail1;
1069 strbuf_release(&sp);
1070 return strbuf_detach(&first, NULL);
1073 /* unquoted second */
1074 cp = stop_at_slash(second, line + llen - second);
1075 if (!cp || cp == second)
1076 goto free_and_fail1;
1077 cp++;
1078 if (line + llen - cp != first.len + 1 ||
1079 memcmp(first.buf, cp, first.len))
1080 goto free_and_fail1;
1081 return strbuf_detach(&first, NULL);
1083 free_and_fail1:
1084 strbuf_release(&first);
1085 strbuf_release(&sp);
1086 return NULL;
1089 /* unquoted first name */
1090 name = stop_at_slash(line, llen);
1091 if (!name || name == line)
1092 return NULL;
1093 name++;
1096 * since the first name is unquoted, a dq if exists must be
1097 * the beginning of the second name.
1099 for (second = name; second < line + llen; second++) {
1100 if (*second == '"') {
1101 struct strbuf sp = STRBUF_INIT;
1102 const char *np;
1104 if (unquote_c_style(&sp, second, NULL))
1105 goto free_and_fail2;
1107 np = stop_at_slash(sp.buf, sp.len);
1108 if (!np || np == sp.buf)
1109 goto free_and_fail2;
1110 np++;
1112 len = sp.buf + sp.len - np;
1113 if (len < second - name &&
1114 !strncmp(np, name, len) &&
1115 isspace(name[len])) {
1116 /* Good */
1117 strbuf_remove(&sp, 0, np - sp.buf);
1118 return strbuf_detach(&sp, NULL);
1121 free_and_fail2:
1122 strbuf_release(&sp);
1123 return NULL;
1128 * Accept a name only if it shows up twice, exactly the same
1129 * form.
1131 for (len = 0 ; ; len++) {
1132 switch (name[len]) {
1133 default:
1134 continue;
1135 case '\n':
1136 return NULL;
1137 case '\t': case ' ':
1138 second = name+len;
1139 for (;;) {
1140 char c = *second++;
1141 if (c == '\n')
1142 return NULL;
1143 if (c == '/')
1144 break;
1146 if (second[len] == '\n' && !memcmp(name, second, len)) {
1147 return xmemdupz(name, len);
1153 /* Verify that we recognize the lines following a git header */
1154 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
1156 unsigned long offset;
1158 /* A git diff has explicit new/delete information, so we don't guess */
1159 patch->is_new = 0;
1160 patch->is_delete = 0;
1163 * Some things may not have the old name in the
1164 * rest of the headers anywhere (pure mode changes,
1165 * or removing or adding empty files), so we get
1166 * the default name from the header.
1168 patch->def_name = git_header_name(line, len);
1169 if (patch->def_name && root) {
1170 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1171 strcpy(s, root);
1172 strcpy(s + root_len, patch->def_name);
1173 free(patch->def_name);
1174 patch->def_name = s;
1177 line += len;
1178 size -= len;
1179 linenr++;
1180 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1181 static const struct opentry {
1182 const char *str;
1183 int (*fn)(const char *, struct patch *);
1184 } optable[] = {
1185 { "@@ -", gitdiff_hdrend },
1186 { "--- ", gitdiff_oldname },
1187 { "+++ ", gitdiff_newname },
1188 { "old mode ", gitdiff_oldmode },
1189 { "new mode ", gitdiff_newmode },
1190 { "deleted file mode ", gitdiff_delete },
1191 { "new file mode ", gitdiff_newfile },
1192 { "copy from ", gitdiff_copysrc },
1193 { "copy to ", gitdiff_copydst },
1194 { "rename old ", gitdiff_renamesrc },
1195 { "rename new ", gitdiff_renamedst },
1196 { "rename from ", gitdiff_renamesrc },
1197 { "rename to ", gitdiff_renamedst },
1198 { "similarity index ", gitdiff_similarity },
1199 { "dissimilarity index ", gitdiff_dissimilarity },
1200 { "index ", gitdiff_index },
1201 { "", gitdiff_unrecognized },
1203 int i;
1205 len = linelen(line, size);
1206 if (!len || line[len-1] != '\n')
1207 break;
1208 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1209 const struct opentry *p = optable + i;
1210 int oplen = strlen(p->str);
1211 if (len < oplen || memcmp(p->str, line, oplen))
1212 continue;
1213 if (p->fn(line + oplen, patch) < 0)
1214 return offset;
1215 break;
1219 return offset;
1222 static int parse_num(const char *line, unsigned long *p)
1224 char *ptr;
1226 if (!isdigit(*line))
1227 return 0;
1228 *p = strtoul(line, &ptr, 10);
1229 return ptr - line;
1232 static int parse_range(const char *line, int len, int offset, const char *expect,
1233 unsigned long *p1, unsigned long *p2)
1235 int digits, ex;
1237 if (offset < 0 || offset >= len)
1238 return -1;
1239 line += offset;
1240 len -= offset;
1242 digits = parse_num(line, p1);
1243 if (!digits)
1244 return -1;
1246 offset += digits;
1247 line += digits;
1248 len -= digits;
1250 *p2 = 1;
1251 if (*line == ',') {
1252 digits = parse_num(line+1, p2);
1253 if (!digits)
1254 return -1;
1256 offset += digits+1;
1257 line += digits+1;
1258 len -= digits+1;
1261 ex = strlen(expect);
1262 if (ex > len)
1263 return -1;
1264 if (memcmp(line, expect, ex))
1265 return -1;
1267 return offset + ex;
1270 static void recount_diff(char *line, int size, struct fragment *fragment)
1272 int oldlines = 0, newlines = 0, ret = 0;
1274 if (size < 1) {
1275 warning("recount: ignore empty hunk");
1276 return;
1279 for (;;) {
1280 int len = linelen(line, size);
1281 size -= len;
1282 line += len;
1284 if (size < 1)
1285 break;
1287 switch (*line) {
1288 case ' ': case '\n':
1289 newlines++;
1290 /* fall through */
1291 case '-':
1292 oldlines++;
1293 continue;
1294 case '+':
1295 newlines++;
1296 continue;
1297 case '\\':
1298 continue;
1299 case '@':
1300 ret = size < 3 || prefixcmp(line, "@@ ");
1301 break;
1302 case 'd':
1303 ret = size < 5 || prefixcmp(line, "diff ");
1304 break;
1305 default:
1306 ret = -1;
1307 break;
1309 if (ret) {
1310 warning("recount: unexpected line: %.*s",
1311 (int)linelen(line, size), line);
1312 return;
1314 break;
1316 fragment->oldlines = oldlines;
1317 fragment->newlines = newlines;
1321 * Parse a unified diff fragment header of the
1322 * form "@@ -a,b +c,d @@"
1324 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1326 int offset;
1328 if (!len || line[len-1] != '\n')
1329 return -1;
1331 /* Figure out the number of lines in a fragment */
1332 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1333 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1335 return offset;
1338 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1340 unsigned long offset, len;
1342 patch->is_toplevel_relative = 0;
1343 patch->is_rename = patch->is_copy = 0;
1344 patch->is_new = patch->is_delete = -1;
1345 patch->old_mode = patch->new_mode = 0;
1346 patch->old_name = patch->new_name = NULL;
1347 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1348 unsigned long nextlen;
1350 len = linelen(line, size);
1351 if (!len)
1352 break;
1354 /* Testing this early allows us to take a few shortcuts.. */
1355 if (len < 6)
1356 continue;
1359 * Make sure we don't find any unconnected patch fragments.
1360 * That's a sign that we didn't find a header, and that a
1361 * patch has become corrupted/broken up.
1363 if (!memcmp("@@ -", line, 4)) {
1364 struct fragment dummy;
1365 if (parse_fragment_header(line, len, &dummy) < 0)
1366 continue;
1367 die("patch fragment without header at line %d: %.*s",
1368 linenr, (int)len-1, line);
1371 if (size < len + 6)
1372 break;
1375 * Git patch? It might not have a real patch, just a rename
1376 * or mode change, so we handle that specially
1378 if (!memcmp("diff --git ", line, 11)) {
1379 int git_hdr_len = parse_git_header(line, len, size, patch);
1380 if (git_hdr_len <= len)
1381 continue;
1382 if (!patch->old_name && !patch->new_name) {
1383 if (!patch->def_name)
1384 die("git diff header lacks filename information when removing "
1385 "%d leading pathname components (line %d)" , p_value, linenr);
1386 patch->old_name = patch->new_name = patch->def_name;
1388 patch->is_toplevel_relative = 1;
1389 *hdrsize = git_hdr_len;
1390 return offset;
1393 /* --- followed by +++ ? */
1394 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1395 continue;
1398 * We only accept unified patches, so we want it to
1399 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1400 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1402 nextlen = linelen(line + len, size - len);
1403 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1404 continue;
1406 /* Ok, we'll consider it a patch */
1407 parse_traditional_patch(line, line+len, patch);
1408 *hdrsize = len + nextlen;
1409 linenr += 2;
1410 return offset;
1412 return -1;
1415 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1417 char *err;
1419 if (!result)
1420 return;
1422 whitespace_error++;
1423 if (squelch_whitespace_errors &&
1424 squelch_whitespace_errors < whitespace_error)
1425 return;
1427 err = whitespace_error_string(result);
1428 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1429 patch_input_file, linenr, err, len, line);
1430 free(err);
1433 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1435 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1437 record_ws_error(result, line + 1, len - 2, linenr);
1441 * Parse a unified diff. Note that this really needs to parse each
1442 * fragment separately, since the only way to know the difference
1443 * between a "---" that is part of a patch, and a "---" that starts
1444 * the next patch is to look at the line counts..
1446 static int parse_fragment(char *line, unsigned long size,
1447 struct patch *patch, struct fragment *fragment)
1449 int added, deleted;
1450 int len = linelen(line, size), offset;
1451 unsigned long oldlines, newlines;
1452 unsigned long leading, trailing;
1454 offset = parse_fragment_header(line, len, fragment);
1455 if (offset < 0)
1456 return -1;
1457 if (offset > 0 && patch->recount)
1458 recount_diff(line + offset, size - offset, fragment);
1459 oldlines = fragment->oldlines;
1460 newlines = fragment->newlines;
1461 leading = 0;
1462 trailing = 0;
1464 /* Parse the thing.. */
1465 line += len;
1466 size -= len;
1467 linenr++;
1468 added = deleted = 0;
1469 for (offset = len;
1470 0 < size;
1471 offset += len, size -= len, line += len, linenr++) {
1472 if (!oldlines && !newlines)
1473 break;
1474 len = linelen(line, size);
1475 if (!len || line[len-1] != '\n')
1476 return -1;
1477 switch (*line) {
1478 default:
1479 return -1;
1480 case '\n': /* newer GNU diff, an empty context line */
1481 case ' ':
1482 oldlines--;
1483 newlines--;
1484 if (!deleted && !added)
1485 leading++;
1486 trailing++;
1487 break;
1488 case '-':
1489 if (apply_in_reverse &&
1490 ws_error_action != nowarn_ws_error)
1491 check_whitespace(line, len, patch->ws_rule);
1492 deleted++;
1493 oldlines--;
1494 trailing = 0;
1495 break;
1496 case '+':
1497 if (!apply_in_reverse &&
1498 ws_error_action != nowarn_ws_error)
1499 check_whitespace(line, len, patch->ws_rule);
1500 added++;
1501 newlines--;
1502 trailing = 0;
1503 break;
1506 * We allow "\ No newline at end of file". Depending
1507 * on locale settings when the patch was produced we
1508 * don't know what this line looks like. The only
1509 * thing we do know is that it begins with "\ ".
1510 * Checking for 12 is just for sanity check -- any
1511 * l10n of "\ No newline..." is at least that long.
1513 case '\\':
1514 if (len < 12 || memcmp(line, "\\ ", 2))
1515 return -1;
1516 break;
1519 if (oldlines || newlines)
1520 return -1;
1521 fragment->leading = leading;
1522 fragment->trailing = trailing;
1525 * If a fragment ends with an incomplete line, we failed to include
1526 * it in the above loop because we hit oldlines == newlines == 0
1527 * before seeing it.
1529 if (12 < size && !memcmp(line, "\\ ", 2))
1530 offset += linelen(line, size);
1532 patch->lines_added += added;
1533 patch->lines_deleted += deleted;
1535 if (0 < patch->is_new && oldlines)
1536 return error("new file depends on old contents");
1537 if (0 < patch->is_delete && newlines)
1538 return error("deleted file still has contents");
1539 return offset;
1542 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1544 unsigned long offset = 0;
1545 unsigned long oldlines = 0, newlines = 0, context = 0;
1546 struct fragment **fragp = &patch->fragments;
1548 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1549 struct fragment *fragment;
1550 int len;
1552 fragment = xcalloc(1, sizeof(*fragment));
1553 fragment->linenr = linenr;
1554 len = parse_fragment(line, size, patch, fragment);
1555 if (len <= 0)
1556 die("corrupt patch at line %d", linenr);
1557 fragment->patch = line;
1558 fragment->size = len;
1559 oldlines += fragment->oldlines;
1560 newlines += fragment->newlines;
1561 context += fragment->leading + fragment->trailing;
1563 *fragp = fragment;
1564 fragp = &fragment->next;
1566 offset += len;
1567 line += len;
1568 size -= len;
1572 * If something was removed (i.e. we have old-lines) it cannot
1573 * be creation, and if something was added it cannot be
1574 * deletion. However, the reverse is not true; --unified=0
1575 * patches that only add are not necessarily creation even
1576 * though they do not have any old lines, and ones that only
1577 * delete are not necessarily deletion.
1579 * Unfortunately, a real creation/deletion patch do _not_ have
1580 * any context line by definition, so we cannot safely tell it
1581 * apart with --unified=0 insanity. At least if the patch has
1582 * more than one hunk it is not creation or deletion.
1584 if (patch->is_new < 0 &&
1585 (oldlines || (patch->fragments && patch->fragments->next)))
1586 patch->is_new = 0;
1587 if (patch->is_delete < 0 &&
1588 (newlines || (patch->fragments && patch->fragments->next)))
1589 patch->is_delete = 0;
1591 if (0 < patch->is_new && oldlines)
1592 die("new file %s depends on old contents", patch->new_name);
1593 if (0 < patch->is_delete && newlines)
1594 die("deleted file %s still has contents", patch->old_name);
1595 if (!patch->is_delete && !newlines && context)
1596 fprintf(stderr, "** warning: file %s becomes empty but "
1597 "is not deleted\n", patch->new_name);
1599 return offset;
1602 static inline int metadata_changes(struct patch *patch)
1604 return patch->is_rename > 0 ||
1605 patch->is_copy > 0 ||
1606 patch->is_new > 0 ||
1607 patch->is_delete ||
1608 (patch->old_mode && patch->new_mode &&
1609 patch->old_mode != patch->new_mode);
1612 static char *inflate_it(const void *data, unsigned long size,
1613 unsigned long inflated_size)
1615 z_stream stream;
1616 void *out;
1617 int st;
1619 memset(&stream, 0, sizeof(stream));
1621 stream.next_in = (unsigned char *)data;
1622 stream.avail_in = size;
1623 stream.next_out = out = xmalloc(inflated_size);
1624 stream.avail_out = inflated_size;
1625 git_inflate_init(&stream);
1626 st = git_inflate(&stream, Z_FINISH);
1627 git_inflate_end(&stream);
1628 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1629 free(out);
1630 return NULL;
1632 return out;
1635 static struct fragment *parse_binary_hunk(char **buf_p,
1636 unsigned long *sz_p,
1637 int *status_p,
1638 int *used_p)
1641 * Expect a line that begins with binary patch method ("literal"
1642 * or "delta"), followed by the length of data before deflating.
1643 * a sequence of 'length-byte' followed by base-85 encoded data
1644 * should follow, terminated by a newline.
1646 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1647 * and we would limit the patch line to 66 characters,
1648 * so one line can fit up to 13 groups that would decode
1649 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1650 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1652 int llen, used;
1653 unsigned long size = *sz_p;
1654 char *buffer = *buf_p;
1655 int patch_method;
1656 unsigned long origlen;
1657 char *data = NULL;
1658 int hunk_size = 0;
1659 struct fragment *frag;
1661 llen = linelen(buffer, size);
1662 used = llen;
1664 *status_p = 0;
1666 if (!prefixcmp(buffer, "delta ")) {
1667 patch_method = BINARY_DELTA_DEFLATED;
1668 origlen = strtoul(buffer + 6, NULL, 10);
1670 else if (!prefixcmp(buffer, "literal ")) {
1671 patch_method = BINARY_LITERAL_DEFLATED;
1672 origlen = strtoul(buffer + 8, NULL, 10);
1674 else
1675 return NULL;
1677 linenr++;
1678 buffer += llen;
1679 while (1) {
1680 int byte_length, max_byte_length, newsize;
1681 llen = linelen(buffer, size);
1682 used += llen;
1683 linenr++;
1684 if (llen == 1) {
1685 /* consume the blank line */
1686 buffer++;
1687 size--;
1688 break;
1691 * Minimum line is "A00000\n" which is 7-byte long,
1692 * and the line length must be multiple of 5 plus 2.
1694 if ((llen < 7) || (llen-2) % 5)
1695 goto corrupt;
1696 max_byte_length = (llen - 2) / 5 * 4;
1697 byte_length = *buffer;
1698 if ('A' <= byte_length && byte_length <= 'Z')
1699 byte_length = byte_length - 'A' + 1;
1700 else if ('a' <= byte_length && byte_length <= 'z')
1701 byte_length = byte_length - 'a' + 27;
1702 else
1703 goto corrupt;
1704 /* if the input length was not multiple of 4, we would
1705 * have filler at the end but the filler should never
1706 * exceed 3 bytes
1708 if (max_byte_length < byte_length ||
1709 byte_length <= max_byte_length - 4)
1710 goto corrupt;
1711 newsize = hunk_size + byte_length;
1712 data = xrealloc(data, newsize);
1713 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1714 goto corrupt;
1715 hunk_size = newsize;
1716 buffer += llen;
1717 size -= llen;
1720 frag = xcalloc(1, sizeof(*frag));
1721 frag->patch = inflate_it(data, hunk_size, origlen);
1722 if (!frag->patch)
1723 goto corrupt;
1724 free(data);
1725 frag->size = origlen;
1726 *buf_p = buffer;
1727 *sz_p = size;
1728 *used_p = used;
1729 frag->binary_patch_method = patch_method;
1730 return frag;
1732 corrupt:
1733 free(data);
1734 *status_p = -1;
1735 error("corrupt binary patch at line %d: %.*s",
1736 linenr-1, llen-1, buffer);
1737 return NULL;
1740 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1743 * We have read "GIT binary patch\n"; what follows is a line
1744 * that says the patch method (currently, either "literal" or
1745 * "delta") and the length of data before deflating; a
1746 * sequence of 'length-byte' followed by base-85 encoded data
1747 * follows.
1749 * When a binary patch is reversible, there is another binary
1750 * hunk in the same format, starting with patch method (either
1751 * "literal" or "delta") with the length of data, and a sequence
1752 * of length-byte + base-85 encoded data, terminated with another
1753 * empty line. This data, when applied to the postimage, produces
1754 * the preimage.
1756 struct fragment *forward;
1757 struct fragment *reverse;
1758 int status;
1759 int used, used_1;
1761 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1762 if (!forward && !status)
1763 /* there has to be one hunk (forward hunk) */
1764 return error("unrecognized binary patch at line %d", linenr-1);
1765 if (status)
1766 /* otherwise we already gave an error message */
1767 return status;
1769 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1770 if (reverse)
1771 used += used_1;
1772 else if (status) {
1774 * Not having reverse hunk is not an error, but having
1775 * a corrupt reverse hunk is.
1777 free((void*) forward->patch);
1778 free(forward);
1779 return status;
1781 forward->next = reverse;
1782 patch->fragments = forward;
1783 patch->is_binary = 1;
1784 return used;
1787 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1789 int hdrsize, patchsize;
1790 int offset = find_header(buffer, size, &hdrsize, patch);
1792 if (offset < 0)
1793 return offset;
1795 patch->ws_rule = whitespace_rule(patch->new_name
1796 ? patch->new_name
1797 : patch->old_name);
1799 patchsize = parse_single_patch(buffer + offset + hdrsize,
1800 size - offset - hdrsize, patch);
1802 if (!patchsize) {
1803 static const char *binhdr[] = {
1804 "Binary files ",
1805 "Files ",
1806 NULL,
1808 static const char git_binary[] = "GIT binary patch\n";
1809 int i;
1810 int hd = hdrsize + offset;
1811 unsigned long llen = linelen(buffer + hd, size - hd);
1813 if (llen == sizeof(git_binary) - 1 &&
1814 !memcmp(git_binary, buffer + hd, llen)) {
1815 int used;
1816 linenr++;
1817 used = parse_binary(buffer + hd + llen,
1818 size - hd - llen, patch);
1819 if (used)
1820 patchsize = used + llen;
1821 else
1822 patchsize = 0;
1824 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1825 for (i = 0; binhdr[i]; i++) {
1826 int len = strlen(binhdr[i]);
1827 if (len < size - hd &&
1828 !memcmp(binhdr[i], buffer + hd, len)) {
1829 linenr++;
1830 patch->is_binary = 1;
1831 patchsize = llen;
1832 break;
1837 /* Empty patch cannot be applied if it is a text patch
1838 * without metadata change. A binary patch appears
1839 * empty to us here.
1841 if ((apply || check) &&
1842 (!patch->is_binary && !metadata_changes(patch)))
1843 die("patch with only garbage at line %d", linenr);
1846 return offset + hdrsize + patchsize;
1849 #define swap(a,b) myswap((a),(b),sizeof(a))
1851 #define myswap(a, b, size) do { \
1852 unsigned char mytmp[size]; \
1853 memcpy(mytmp, &a, size); \
1854 memcpy(&a, &b, size); \
1855 memcpy(&b, mytmp, size); \
1856 } while (0)
1858 static void reverse_patches(struct patch *p)
1860 for (; p; p = p->next) {
1861 struct fragment *frag = p->fragments;
1863 swap(p->new_name, p->old_name);
1864 swap(p->new_mode, p->old_mode);
1865 swap(p->is_new, p->is_delete);
1866 swap(p->lines_added, p->lines_deleted);
1867 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1869 for (; frag; frag = frag->next) {
1870 swap(frag->newpos, frag->oldpos);
1871 swap(frag->newlines, frag->oldlines);
1876 static const char pluses[] =
1877 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1878 static const char minuses[]=
1879 "----------------------------------------------------------------------";
1881 static void show_stats(struct patch *patch)
1883 struct strbuf qname = STRBUF_INIT;
1884 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1885 int max, add, del;
1887 quote_c_style(cp, &qname, NULL, 0);
1890 * "scale" the filename
1892 max = max_len;
1893 if (max > 50)
1894 max = 50;
1896 if (qname.len > max) {
1897 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1898 if (!cp)
1899 cp = qname.buf + qname.len + 3 - max;
1900 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1903 if (patch->is_binary) {
1904 printf(" %-*s | Bin\n", max, qname.buf);
1905 strbuf_release(&qname);
1906 return;
1909 printf(" %-*s |", max, qname.buf);
1910 strbuf_release(&qname);
1913 * scale the add/delete
1915 max = max + max_change > 70 ? 70 - max : max_change;
1916 add = patch->lines_added;
1917 del = patch->lines_deleted;
1919 if (max_change > 0) {
1920 int total = ((add + del) * max + max_change / 2) / max_change;
1921 add = (add * max + max_change / 2) / max_change;
1922 del = total - add;
1924 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1925 add, pluses, del, minuses);
1928 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1930 switch (st->st_mode & S_IFMT) {
1931 case S_IFLNK:
1932 if (strbuf_readlink(buf, path, st->st_size) < 0)
1933 return error("unable to read symlink %s", path);
1934 return 0;
1935 case S_IFREG:
1936 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1937 return error("unable to open or read %s", path);
1938 convert_to_git(path, buf->buf, buf->len, buf, 0);
1939 return 0;
1940 default:
1941 return -1;
1946 * Update the preimage, and the common lines in postimage,
1947 * from buffer buf of length len. If postlen is 0 the postimage
1948 * is updated in place, otherwise it's updated on a new buffer
1949 * of length postlen
1952 static void update_pre_post_images(struct image *preimage,
1953 struct image *postimage,
1954 char *buf,
1955 size_t len, size_t postlen)
1957 int i, ctx;
1958 char *new, *old, *fixed;
1959 struct image fixed_preimage;
1962 * Update the preimage with whitespace fixes. Note that we
1963 * are not losing preimage->buf -- apply_one_fragment() will
1964 * free "oldlines".
1966 prepare_image(&fixed_preimage, buf, len, 1);
1967 assert(fixed_preimage.nr == preimage->nr);
1968 for (i = 0; i < preimage->nr; i++)
1969 fixed_preimage.line[i].flag = preimage->line[i].flag;
1970 free(preimage->line_allocated);
1971 *preimage = fixed_preimage;
1974 * Adjust the common context lines in postimage. This can be
1975 * done in-place when we are just doing whitespace fixing,
1976 * which does not make the string grow, but needs a new buffer
1977 * when ignoring whitespace causes the update, since in this case
1978 * we could have e.g. tabs converted to multiple spaces.
1979 * We trust the caller to tell us if the update can be done
1980 * in place (postlen==0) or not.
1982 old = postimage->buf;
1983 if (postlen)
1984 new = postimage->buf = xmalloc(postlen);
1985 else
1986 new = old;
1987 fixed = preimage->buf;
1988 for (i = ctx = 0; i < postimage->nr; i++) {
1989 size_t len = postimage->line[i].len;
1990 if (!(postimage->line[i].flag & LINE_COMMON)) {
1991 /* an added line -- no counterparts in preimage */
1992 memmove(new, old, len);
1993 old += len;
1994 new += len;
1995 continue;
1998 /* a common context -- skip it in the original postimage */
1999 old += len;
2001 /* and find the corresponding one in the fixed preimage */
2002 while (ctx < preimage->nr &&
2003 !(preimage->line[ctx].flag & LINE_COMMON)) {
2004 fixed += preimage->line[ctx].len;
2005 ctx++;
2007 if (preimage->nr <= ctx)
2008 die("oops");
2010 /* and copy it in, while fixing the line length */
2011 len = preimage->line[ctx].len;
2012 memcpy(new, fixed, len);
2013 new += len;
2014 fixed += len;
2015 postimage->line[i].len = len;
2016 ctx++;
2019 /* Fix the length of the whole thing */
2020 postimage->len = new - postimage->buf;
2023 static int match_fragment(struct image *img,
2024 struct image *preimage,
2025 struct image *postimage,
2026 unsigned long try,
2027 int try_lno,
2028 unsigned ws_rule,
2029 int match_beginning, int match_end)
2031 int i;
2032 char *fixed_buf, *buf, *orig, *target;
2033 struct strbuf fixed;
2034 size_t fixed_len;
2035 int preimage_limit;
2037 if (preimage->nr + try_lno <= img->nr) {
2039 * The hunk falls within the boundaries of img.
2041 preimage_limit = preimage->nr;
2042 if (match_end && (preimage->nr + try_lno != img->nr))
2043 return 0;
2044 } else if (ws_error_action == correct_ws_error &&
2045 (ws_rule & WS_BLANK_AT_EOF)) {
2047 * This hunk extends beyond the end of img, and we are
2048 * removing blank lines at the end of the file. This
2049 * many lines from the beginning of the preimage must
2050 * match with img, and the remainder of the preimage
2051 * must be blank.
2053 preimage_limit = img->nr - try_lno;
2054 } else {
2056 * The hunk extends beyond the end of the img and
2057 * we are not removing blanks at the end, so we
2058 * should reject the hunk at this position.
2060 return 0;
2063 if (match_beginning && try_lno)
2064 return 0;
2066 /* Quick hash check */
2067 for (i = 0; i < preimage_limit; i++)
2068 if (preimage->line[i].hash != img->line[try_lno + i].hash)
2069 return 0;
2071 if (preimage_limit == preimage->nr) {
2073 * Do we have an exact match? If we were told to match
2074 * at the end, size must be exactly at try+fragsize,
2075 * otherwise try+fragsize must be still within the preimage,
2076 * and either case, the old piece should match the preimage
2077 * exactly.
2079 if ((match_end
2080 ? (try + preimage->len == img->len)
2081 : (try + preimage->len <= img->len)) &&
2082 !memcmp(img->buf + try, preimage->buf, preimage->len))
2083 return 1;
2084 } else {
2086 * The preimage extends beyond the end of img, so
2087 * there cannot be an exact match.
2089 * There must be one non-blank context line that match
2090 * a line before the end of img.
2092 char *buf_end;
2094 buf = preimage->buf;
2095 buf_end = buf;
2096 for (i = 0; i < preimage_limit; i++)
2097 buf_end += preimage->line[i].len;
2099 for ( ; buf < buf_end; buf++)
2100 if (!isspace(*buf))
2101 break;
2102 if (buf == buf_end)
2103 return 0;
2107 * No exact match. If we are ignoring whitespace, run a line-by-line
2108 * fuzzy matching. We collect all the line length information because
2109 * we need it to adjust whitespace if we match.
2111 if (ws_ignore_action == ignore_ws_change) {
2112 size_t imgoff = 0;
2113 size_t preoff = 0;
2114 size_t postlen = postimage->len;
2115 size_t extra_chars;
2116 char *preimage_eof;
2117 char *preimage_end;
2118 for (i = 0; i < preimage_limit; i++) {
2119 size_t prelen = preimage->line[i].len;
2120 size_t imglen = img->line[try_lno+i].len;
2122 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2123 preimage->buf + preoff, prelen))
2124 return 0;
2125 if (preimage->line[i].flag & LINE_COMMON)
2126 postlen += imglen - prelen;
2127 imgoff += imglen;
2128 preoff += prelen;
2132 * Ok, the preimage matches with whitespace fuzz.
2134 * imgoff now holds the true length of the target that
2135 * matches the preimage before the end of the file.
2137 * Count the number of characters in the preimage that fall
2138 * beyond the end of the file and make sure that all of them
2139 * are whitespace characters. (This can only happen if
2140 * we are removing blank lines at the end of the file.)
2142 buf = preimage_eof = preimage->buf + preoff;
2143 for ( ; i < preimage->nr; i++)
2144 preoff += preimage->line[i].len;
2145 preimage_end = preimage->buf + preoff;
2146 for ( ; buf < preimage_end; buf++)
2147 if (!isspace(*buf))
2148 return 0;
2151 * Update the preimage and the common postimage context
2152 * lines to use the same whitespace as the target.
2153 * If whitespace is missing in the target (i.e.
2154 * if the preimage extends beyond the end of the file),
2155 * use the whitespace from the preimage.
2157 extra_chars = preimage_end - preimage_eof;
2158 strbuf_init(&fixed, imgoff + extra_chars);
2159 strbuf_add(&fixed, img->buf + try, imgoff);
2160 strbuf_add(&fixed, preimage_eof, extra_chars);
2161 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2162 update_pre_post_images(preimage, postimage,
2163 fixed_buf, fixed_len, postlen);
2164 return 1;
2167 if (ws_error_action != correct_ws_error)
2168 return 0;
2171 * The hunk does not apply byte-by-byte, but the hash says
2172 * it might with whitespace fuzz. We haven't been asked to
2173 * ignore whitespace, we were asked to correct whitespace
2174 * errors, so let's try matching after whitespace correction.
2176 * The preimage may extend beyond the end of the file,
2177 * but in this loop we will only handle the part of the
2178 * preimage that falls within the file.
2180 strbuf_init(&fixed, preimage->len + 1);
2181 orig = preimage->buf;
2182 target = img->buf + try;
2183 for (i = 0; i < preimage_limit; i++) {
2184 size_t oldlen = preimage->line[i].len;
2185 size_t tgtlen = img->line[try_lno + i].len;
2186 size_t fixstart = fixed.len;
2187 struct strbuf tgtfix;
2188 int match;
2190 /* Try fixing the line in the preimage */
2191 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2193 /* Try fixing the line in the target */
2194 strbuf_init(&tgtfix, tgtlen);
2195 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2198 * If they match, either the preimage was based on
2199 * a version before our tree fixed whitespace breakage,
2200 * or we are lacking a whitespace-fix patch the tree
2201 * the preimage was based on already had (i.e. target
2202 * has whitespace breakage, the preimage doesn't).
2203 * In either case, we are fixing the whitespace breakages
2204 * so we might as well take the fix together with their
2205 * real change.
2207 match = (tgtfix.len == fixed.len - fixstart &&
2208 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2209 fixed.len - fixstart));
2211 strbuf_release(&tgtfix);
2212 if (!match)
2213 goto unmatch_exit;
2215 orig += oldlen;
2216 target += tgtlen;
2221 * Now handle the lines in the preimage that falls beyond the
2222 * end of the file (if any). They will only match if they are
2223 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2224 * false).
2226 for ( ; i < preimage->nr; i++) {
2227 size_t fixstart = fixed.len; /* start of the fixed preimage */
2228 size_t oldlen = preimage->line[i].len;
2229 int j;
2231 /* Try fixing the line in the preimage */
2232 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2234 for (j = fixstart; j < fixed.len; j++)
2235 if (!isspace(fixed.buf[j]))
2236 goto unmatch_exit;
2238 orig += oldlen;
2242 * Yes, the preimage is based on an older version that still
2243 * has whitespace breakages unfixed, and fixing them makes the
2244 * hunk match. Update the context lines in the postimage.
2246 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2247 update_pre_post_images(preimage, postimage,
2248 fixed_buf, fixed_len, 0);
2249 return 1;
2251 unmatch_exit:
2252 strbuf_release(&fixed);
2253 return 0;
2256 static int find_pos(struct image *img,
2257 struct image *preimage,
2258 struct image *postimage,
2259 int line,
2260 unsigned ws_rule,
2261 int match_beginning, int match_end)
2263 int i;
2264 unsigned long backwards, forwards, try;
2265 int backwards_lno, forwards_lno, try_lno;
2268 * If match_beginning or match_end is specified, there is no
2269 * point starting from a wrong line that will never match and
2270 * wander around and wait for a match at the specified end.
2272 if (match_beginning)
2273 line = 0;
2274 else if (match_end)
2275 line = img->nr - preimage->nr;
2278 * Because the comparison is unsigned, the following test
2279 * will also take care of a negative line number that can
2280 * result when match_end and preimage is larger than the target.
2282 if ((size_t) line > img->nr)
2283 line = img->nr;
2285 try = 0;
2286 for (i = 0; i < line; i++)
2287 try += img->line[i].len;
2290 * There's probably some smart way to do this, but I'll leave
2291 * that to the smart and beautiful people. I'm simple and stupid.
2293 backwards = try;
2294 backwards_lno = line;
2295 forwards = try;
2296 forwards_lno = line;
2297 try_lno = line;
2299 for (i = 0; ; i++) {
2300 if (match_fragment(img, preimage, postimage,
2301 try, try_lno, ws_rule,
2302 match_beginning, match_end))
2303 return try_lno;
2305 again:
2306 if (backwards_lno == 0 && forwards_lno == img->nr)
2307 break;
2309 if (i & 1) {
2310 if (backwards_lno == 0) {
2311 i++;
2312 goto again;
2314 backwards_lno--;
2315 backwards -= img->line[backwards_lno].len;
2316 try = backwards;
2317 try_lno = backwards_lno;
2318 } else {
2319 if (forwards_lno == img->nr) {
2320 i++;
2321 goto again;
2323 forwards += img->line[forwards_lno].len;
2324 forwards_lno++;
2325 try = forwards;
2326 try_lno = forwards_lno;
2330 return -1;
2333 static void remove_first_line(struct image *img)
2335 img->buf += img->line[0].len;
2336 img->len -= img->line[0].len;
2337 img->line++;
2338 img->nr--;
2341 static void remove_last_line(struct image *img)
2343 img->len -= img->line[--img->nr].len;
2346 static void update_image(struct image *img,
2347 int applied_pos,
2348 struct image *preimage,
2349 struct image *postimage)
2352 * remove the copy of preimage at offset in img
2353 * and replace it with postimage
2355 int i, nr;
2356 size_t remove_count, insert_count, applied_at = 0;
2357 char *result;
2358 int preimage_limit;
2361 * If we are removing blank lines at the end of img,
2362 * the preimage may extend beyond the end.
2363 * If that is the case, we must be careful only to
2364 * remove the part of the preimage that falls within
2365 * the boundaries of img. Initialize preimage_limit
2366 * to the number of lines in the preimage that falls
2367 * within the boundaries.
2369 preimage_limit = preimage->nr;
2370 if (preimage_limit > img->nr - applied_pos)
2371 preimage_limit = img->nr - applied_pos;
2373 for (i = 0; i < applied_pos; i++)
2374 applied_at += img->line[i].len;
2376 remove_count = 0;
2377 for (i = 0; i < preimage_limit; i++)
2378 remove_count += img->line[applied_pos + i].len;
2379 insert_count = postimage->len;
2381 /* Adjust the contents */
2382 result = xmalloc(img->len + insert_count - remove_count + 1);
2383 memcpy(result, img->buf, applied_at);
2384 memcpy(result + applied_at, postimage->buf, postimage->len);
2385 memcpy(result + applied_at + postimage->len,
2386 img->buf + (applied_at + remove_count),
2387 img->len - (applied_at + remove_count));
2388 free(img->buf);
2389 img->buf = result;
2390 img->len += insert_count - remove_count;
2391 result[img->len] = '\0';
2393 /* Adjust the line table */
2394 nr = img->nr + postimage->nr - preimage_limit;
2395 if (preimage_limit < postimage->nr) {
2397 * NOTE: this knows that we never call remove_first_line()
2398 * on anything other than pre/post image.
2400 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2401 img->line_allocated = img->line;
2403 if (preimage_limit != postimage->nr)
2404 memmove(img->line + applied_pos + postimage->nr,
2405 img->line + applied_pos + preimage_limit,
2406 (img->nr - (applied_pos + preimage_limit)) *
2407 sizeof(*img->line));
2408 memcpy(img->line + applied_pos,
2409 postimage->line,
2410 postimage->nr * sizeof(*img->line));
2411 img->nr = nr;
2414 static int apply_one_fragment(struct image *img, struct fragment *frag,
2415 int inaccurate_eof, unsigned ws_rule)
2417 int match_beginning, match_end;
2418 const char *patch = frag->patch;
2419 int size = frag->size;
2420 char *old, *oldlines;
2421 struct strbuf newlines;
2422 int new_blank_lines_at_end = 0;
2423 unsigned long leading, trailing;
2424 int pos, applied_pos;
2425 struct image preimage;
2426 struct image postimage;
2428 memset(&preimage, 0, sizeof(preimage));
2429 memset(&postimage, 0, sizeof(postimage));
2430 oldlines = xmalloc(size);
2431 strbuf_init(&newlines, size);
2433 old = oldlines;
2434 while (size > 0) {
2435 char first;
2436 int len = linelen(patch, size);
2437 int plen;
2438 int added_blank_line = 0;
2439 int is_blank_context = 0;
2440 size_t start;
2442 if (!len)
2443 break;
2446 * "plen" is how much of the line we should use for
2447 * the actual patch data. Normally we just remove the
2448 * first character on the line, but if the line is
2449 * followed by "\ No newline", then we also remove the
2450 * last one (which is the newline, of course).
2452 plen = len - 1;
2453 if (len < size && patch[len] == '\\')
2454 plen--;
2455 first = *patch;
2456 if (apply_in_reverse) {
2457 if (first == '-')
2458 first = '+';
2459 else if (first == '+')
2460 first = '-';
2463 switch (first) {
2464 case '\n':
2465 /* Newer GNU diff, empty context line */
2466 if (plen < 0)
2467 /* ... followed by '\No newline'; nothing */
2468 break;
2469 *old++ = '\n';
2470 strbuf_addch(&newlines, '\n');
2471 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2472 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2473 is_blank_context = 1;
2474 break;
2475 case ' ':
2476 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2477 ws_blank_line(patch + 1, plen, ws_rule))
2478 is_blank_context = 1;
2479 case '-':
2480 memcpy(old, patch + 1, plen);
2481 add_line_info(&preimage, old, plen,
2482 (first == ' ' ? LINE_COMMON : 0));
2483 old += plen;
2484 if (first == '-')
2485 break;
2486 /* Fall-through for ' ' */
2487 case '+':
2488 /* --no-add does not add new lines */
2489 if (first == '+' && no_add)
2490 break;
2492 start = newlines.len;
2493 if (first != '+' ||
2494 !whitespace_error ||
2495 ws_error_action != correct_ws_error) {
2496 strbuf_add(&newlines, patch + 1, plen);
2498 else {
2499 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2501 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2502 (first == '+' ? 0 : LINE_COMMON));
2503 if (first == '+' &&
2504 (ws_rule & WS_BLANK_AT_EOF) &&
2505 ws_blank_line(patch + 1, plen, ws_rule))
2506 added_blank_line = 1;
2507 break;
2508 case '@': case '\\':
2509 /* Ignore it, we already handled it */
2510 break;
2511 default:
2512 if (apply_verbosely)
2513 error("invalid start of line: '%c'", first);
2514 return -1;
2516 if (added_blank_line)
2517 new_blank_lines_at_end++;
2518 else if (is_blank_context)
2520 else
2521 new_blank_lines_at_end = 0;
2522 patch += len;
2523 size -= len;
2525 if (inaccurate_eof &&
2526 old > oldlines && old[-1] == '\n' &&
2527 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2528 old--;
2529 strbuf_setlen(&newlines, newlines.len - 1);
2532 leading = frag->leading;
2533 trailing = frag->trailing;
2536 * A hunk to change lines at the beginning would begin with
2537 * @@ -1,L +N,M @@
2538 * but we need to be careful. -U0 that inserts before the second
2539 * line also has this pattern.
2541 * And a hunk to add to an empty file would begin with
2542 * @@ -0,0 +N,M @@
2544 * In other words, a hunk that is (frag->oldpos <= 1) with or
2545 * without leading context must match at the beginning.
2547 match_beginning = (!frag->oldpos ||
2548 (frag->oldpos == 1 && !unidiff_zero));
2551 * A hunk without trailing lines must match at the end.
2552 * However, we simply cannot tell if a hunk must match end
2553 * from the lack of trailing lines if the patch was generated
2554 * with unidiff without any context.
2556 match_end = !unidiff_zero && !trailing;
2558 pos = frag->newpos ? (frag->newpos - 1) : 0;
2559 preimage.buf = oldlines;
2560 preimage.len = old - oldlines;
2561 postimage.buf = newlines.buf;
2562 postimage.len = newlines.len;
2563 preimage.line = preimage.line_allocated;
2564 postimage.line = postimage.line_allocated;
2566 for (;;) {
2568 applied_pos = find_pos(img, &preimage, &postimage, pos,
2569 ws_rule, match_beginning, match_end);
2571 if (applied_pos >= 0)
2572 break;
2574 /* Am I at my context limits? */
2575 if ((leading <= p_context) && (trailing <= p_context))
2576 break;
2577 if (match_beginning || match_end) {
2578 match_beginning = match_end = 0;
2579 continue;
2583 * Reduce the number of context lines; reduce both
2584 * leading and trailing if they are equal otherwise
2585 * just reduce the larger context.
2587 if (leading >= trailing) {
2588 remove_first_line(&preimage);
2589 remove_first_line(&postimage);
2590 pos--;
2591 leading--;
2593 if (trailing > leading) {
2594 remove_last_line(&preimage);
2595 remove_last_line(&postimage);
2596 trailing--;
2600 if (applied_pos >= 0) {
2601 if (new_blank_lines_at_end &&
2602 preimage.nr + applied_pos >= img->nr &&
2603 (ws_rule & WS_BLANK_AT_EOF) &&
2604 ws_error_action != nowarn_ws_error) {
2605 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2606 if (ws_error_action == correct_ws_error) {
2607 while (new_blank_lines_at_end--)
2608 remove_last_line(&postimage);
2611 * We would want to prevent write_out_results()
2612 * from taking place in apply_patch() that follows
2613 * the callchain led us here, which is:
2614 * apply_patch->check_patch_list->check_patch->
2615 * apply_data->apply_fragments->apply_one_fragment
2617 if (ws_error_action == die_on_ws_error)
2618 apply = 0;
2622 * Warn if it was necessary to reduce the number
2623 * of context lines.
2625 if ((leading != frag->leading) ||
2626 (trailing != frag->trailing))
2627 fprintf(stderr, "Context reduced to (%ld/%ld)"
2628 " to apply fragment at %d\n",
2629 leading, trailing, applied_pos+1);
2630 update_image(img, applied_pos, &preimage, &postimage);
2631 } else {
2632 if (apply_verbosely)
2633 error("while searching for:\n%.*s",
2634 (int)(old - oldlines), oldlines);
2637 free(oldlines);
2638 strbuf_release(&newlines);
2639 free(preimage.line_allocated);
2640 free(postimage.line_allocated);
2642 return (applied_pos < 0);
2645 static int apply_binary_fragment(struct image *img, struct patch *patch)
2647 struct fragment *fragment = patch->fragments;
2648 unsigned long len;
2649 void *dst;
2651 /* Binary patch is irreversible without the optional second hunk */
2652 if (apply_in_reverse) {
2653 if (!fragment->next)
2654 return error("cannot reverse-apply a binary patch "
2655 "without the reverse hunk to '%s'",
2656 patch->new_name
2657 ? patch->new_name : patch->old_name);
2658 fragment = fragment->next;
2660 switch (fragment->binary_patch_method) {
2661 case BINARY_DELTA_DEFLATED:
2662 dst = patch_delta(img->buf, img->len, fragment->patch,
2663 fragment->size, &len);
2664 if (!dst)
2665 return -1;
2666 clear_image(img);
2667 img->buf = dst;
2668 img->len = len;
2669 return 0;
2670 case BINARY_LITERAL_DEFLATED:
2671 clear_image(img);
2672 img->len = fragment->size;
2673 img->buf = xmalloc(img->len+1);
2674 memcpy(img->buf, fragment->patch, img->len);
2675 img->buf[img->len] = '\0';
2676 return 0;
2678 return -1;
2681 static int apply_binary(struct image *img, struct patch *patch)
2683 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2684 unsigned char sha1[20];
2687 * For safety, we require patch index line to contain
2688 * full 40-byte textual SHA1 for old and new, at least for now.
2690 if (strlen(patch->old_sha1_prefix) != 40 ||
2691 strlen(patch->new_sha1_prefix) != 40 ||
2692 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2693 get_sha1_hex(patch->new_sha1_prefix, sha1))
2694 return error("cannot apply binary patch to '%s' "
2695 "without full index line", name);
2697 if (patch->old_name) {
2699 * See if the old one matches what the patch
2700 * applies to.
2702 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2703 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2704 return error("the patch applies to '%s' (%s), "
2705 "which does not match the "
2706 "current contents.",
2707 name, sha1_to_hex(sha1));
2709 else {
2710 /* Otherwise, the old one must be empty. */
2711 if (img->len)
2712 return error("the patch applies to an empty "
2713 "'%s' but it is not empty", name);
2716 get_sha1_hex(patch->new_sha1_prefix, sha1);
2717 if (is_null_sha1(sha1)) {
2718 clear_image(img);
2719 return 0; /* deletion patch */
2722 if (has_sha1_file(sha1)) {
2723 /* We already have the postimage */
2724 enum object_type type;
2725 unsigned long size;
2726 char *result;
2728 result = read_sha1_file(sha1, &type, &size);
2729 if (!result)
2730 return error("the necessary postimage %s for "
2731 "'%s' cannot be read",
2732 patch->new_sha1_prefix, name);
2733 clear_image(img);
2734 img->buf = result;
2735 img->len = size;
2736 } else {
2738 * We have verified buf matches the preimage;
2739 * apply the patch data to it, which is stored
2740 * in the patch->fragments->{patch,size}.
2742 if (apply_binary_fragment(img, patch))
2743 return error("binary patch does not apply to '%s'",
2744 name);
2746 /* verify that the result matches */
2747 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2748 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2749 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2750 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2753 return 0;
2756 static int apply_fragments(struct image *img, struct patch *patch)
2758 struct fragment *frag = patch->fragments;
2759 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2760 unsigned ws_rule = patch->ws_rule;
2761 unsigned inaccurate_eof = patch->inaccurate_eof;
2763 if (patch->is_binary)
2764 return apply_binary(img, patch);
2766 while (frag) {
2767 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2768 error("patch failed: %s:%ld", name, frag->oldpos);
2769 if (!apply_with_reject)
2770 return -1;
2771 frag->rejected = 1;
2773 frag = frag->next;
2775 return 0;
2778 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2780 if (!ce)
2781 return 0;
2783 if (S_ISGITLINK(ce->ce_mode)) {
2784 strbuf_grow(buf, 100);
2785 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2786 } else {
2787 enum object_type type;
2788 unsigned long sz;
2789 char *result;
2791 result = read_sha1_file(ce->sha1, &type, &sz);
2792 if (!result)
2793 return -1;
2794 /* XXX read_sha1_file NUL-terminates */
2795 strbuf_attach(buf, result, sz, sz + 1);
2797 return 0;
2800 static struct patch *in_fn_table(const char *name)
2802 struct string_list_item *item;
2804 if (name == NULL)
2805 return NULL;
2807 item = string_list_lookup(&fn_table, name);
2808 if (item != NULL)
2809 return (struct patch *)item->util;
2811 return NULL;
2815 * item->util in the filename table records the status of the path.
2816 * Usually it points at a patch (whose result records the contents
2817 * of it after applying it), but it could be PATH_WAS_DELETED for a
2818 * path that a previously applied patch has already removed.
2820 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2821 #define PATH_WAS_DELETED ((struct patch *) -1)
2823 static int to_be_deleted(struct patch *patch)
2825 return patch == PATH_TO_BE_DELETED;
2828 static int was_deleted(struct patch *patch)
2830 return patch == PATH_WAS_DELETED;
2833 static void add_to_fn_table(struct patch *patch)
2835 struct string_list_item *item;
2838 * Always add new_name unless patch is a deletion
2839 * This should cover the cases for normal diffs,
2840 * file creations and copies
2842 if (patch->new_name != NULL) {
2843 item = string_list_insert(&fn_table, patch->new_name);
2844 item->util = patch;
2848 * store a failure on rename/deletion cases because
2849 * later chunks shouldn't patch old names
2851 if ((patch->new_name == NULL) || (patch->is_rename)) {
2852 item = string_list_insert(&fn_table, patch->old_name);
2853 item->util = PATH_WAS_DELETED;
2857 static void prepare_fn_table(struct patch *patch)
2860 * store information about incoming file deletion
2862 while (patch) {
2863 if ((patch->new_name == NULL) || (patch->is_rename)) {
2864 struct string_list_item *item;
2865 item = string_list_insert(&fn_table, patch->old_name);
2866 item->util = PATH_TO_BE_DELETED;
2868 patch = patch->next;
2872 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2874 struct strbuf buf = STRBUF_INIT;
2875 struct image image;
2876 size_t len;
2877 char *img;
2878 struct patch *tpatch;
2880 if (!(patch->is_copy || patch->is_rename) &&
2881 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2882 if (was_deleted(tpatch)) {
2883 return error("patch %s has been renamed/deleted",
2884 patch->old_name);
2886 /* We have a patched copy in memory use that */
2887 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2888 } else if (cached) {
2889 if (read_file_or_gitlink(ce, &buf))
2890 return error("read of %s failed", patch->old_name);
2891 } else if (patch->old_name) {
2892 if (S_ISGITLINK(patch->old_mode)) {
2893 if (ce) {
2894 read_file_or_gitlink(ce, &buf);
2895 } else {
2897 * There is no way to apply subproject
2898 * patch without looking at the index.
2900 patch->fragments = NULL;
2902 } else {
2903 if (read_old_data(st, patch->old_name, &buf))
2904 return error("read of %s failed", patch->old_name);
2908 img = strbuf_detach(&buf, &len);
2909 prepare_image(&image, img, len, !patch->is_binary);
2911 if (apply_fragments(&image, patch) < 0)
2912 return -1; /* note with --reject this succeeds. */
2913 patch->result = image.buf;
2914 patch->resultsize = image.len;
2915 add_to_fn_table(patch);
2916 free(image.line_allocated);
2918 if (0 < patch->is_delete && patch->resultsize)
2919 return error("removal patch leaves file contents");
2921 return 0;
2924 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2926 struct stat nst;
2927 if (!lstat(new_name, &nst)) {
2928 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2929 return 0;
2931 * A leading component of new_name might be a symlink
2932 * that is going to be removed with this patch, but
2933 * still pointing at somewhere that has the path.
2934 * In such a case, path "new_name" does not exist as
2935 * far as git is concerned.
2937 if (has_symlink_leading_path(new_name, strlen(new_name)))
2938 return 0;
2940 return error("%s: already exists in working directory", new_name);
2942 else if ((errno != ENOENT) && (errno != ENOTDIR))
2943 return error("%s: %s", new_name, strerror(errno));
2944 return 0;
2947 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2949 if (S_ISGITLINK(ce->ce_mode)) {
2950 if (!S_ISDIR(st->st_mode))
2951 return -1;
2952 return 0;
2954 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2957 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2959 const char *old_name = patch->old_name;
2960 struct patch *tpatch = NULL;
2961 int stat_ret = 0;
2962 unsigned st_mode = 0;
2965 * Make sure that we do not have local modifications from the
2966 * index when we are looking at the index. Also make sure
2967 * we have the preimage file to be patched in the work tree,
2968 * unless --cached, which tells git to apply only in the index.
2970 if (!old_name)
2971 return 0;
2973 assert(patch->is_new <= 0);
2975 if (!(patch->is_copy || patch->is_rename) &&
2976 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2977 if (was_deleted(tpatch))
2978 return error("%s: has been deleted/renamed", old_name);
2979 st_mode = tpatch->new_mode;
2980 } else if (!cached) {
2981 stat_ret = lstat(old_name, st);
2982 if (stat_ret && errno != ENOENT)
2983 return error("%s: %s", old_name, strerror(errno));
2986 if (to_be_deleted(tpatch))
2987 tpatch = NULL;
2989 if (check_index && !tpatch) {
2990 int pos = cache_name_pos(old_name, strlen(old_name));
2991 if (pos < 0) {
2992 if (patch->is_new < 0)
2993 goto is_new;
2994 return error("%s: does not exist in index", old_name);
2996 *ce = active_cache[pos];
2997 if (stat_ret < 0) {
2998 struct checkout costate;
2999 /* checkout */
3000 memset(&costate, 0, sizeof(costate));
3001 costate.base_dir = "";
3002 costate.refresh_cache = 1;
3003 if (checkout_entry(*ce, &costate, NULL) ||
3004 lstat(old_name, st))
3005 return -1;
3007 if (!cached && verify_index_match(*ce, st))
3008 return error("%s: does not match index", old_name);
3009 if (cached)
3010 st_mode = (*ce)->ce_mode;
3011 } else if (stat_ret < 0) {
3012 if (patch->is_new < 0)
3013 goto is_new;
3014 return error("%s: %s", old_name, strerror(errno));
3017 if (!cached && !tpatch)
3018 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3020 if (patch->is_new < 0)
3021 patch->is_new = 0;
3022 if (!patch->old_mode)
3023 patch->old_mode = st_mode;
3024 if ((st_mode ^ patch->old_mode) & S_IFMT)
3025 return error("%s: wrong type", old_name);
3026 if (st_mode != patch->old_mode)
3027 warning("%s has type %o, expected %o",
3028 old_name, st_mode, patch->old_mode);
3029 if (!patch->new_mode && !patch->is_delete)
3030 patch->new_mode = st_mode;
3031 return 0;
3033 is_new:
3034 patch->is_new = 1;
3035 patch->is_delete = 0;
3036 patch->old_name = NULL;
3037 return 0;
3040 static int check_patch(struct patch *patch)
3042 struct stat st;
3043 const char *old_name = patch->old_name;
3044 const char *new_name = patch->new_name;
3045 const char *name = old_name ? old_name : new_name;
3046 struct cache_entry *ce = NULL;
3047 struct patch *tpatch;
3048 int ok_if_exists;
3049 int status;
3051 patch->rejected = 1; /* we will drop this after we succeed */
3053 status = check_preimage(patch, &ce, &st);
3054 if (status)
3055 return status;
3056 old_name = patch->old_name;
3058 if ((tpatch = in_fn_table(new_name)) &&
3059 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3061 * A type-change diff is always split into a patch to
3062 * delete old, immediately followed by a patch to
3063 * create new (see diff.c::run_diff()); in such a case
3064 * it is Ok that the entry to be deleted by the
3065 * previous patch is still in the working tree and in
3066 * the index.
3068 ok_if_exists = 1;
3069 else
3070 ok_if_exists = 0;
3072 if (new_name &&
3073 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3074 if (check_index &&
3075 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3076 !ok_if_exists)
3077 return error("%s: already exists in index", new_name);
3078 if (!cached) {
3079 int err = check_to_create_blob(new_name, ok_if_exists);
3080 if (err)
3081 return err;
3083 if (!patch->new_mode) {
3084 if (0 < patch->is_new)
3085 patch->new_mode = S_IFREG | 0644;
3086 else
3087 patch->new_mode = patch->old_mode;
3091 if (new_name && old_name) {
3092 int same = !strcmp(old_name, new_name);
3093 if (!patch->new_mode)
3094 patch->new_mode = patch->old_mode;
3095 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3096 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
3097 patch->new_mode, new_name, patch->old_mode,
3098 same ? "" : " of ", same ? "" : old_name);
3101 if (apply_data(patch, &st, ce) < 0)
3102 return error("%s: patch does not apply", name);
3103 patch->rejected = 0;
3104 return 0;
3107 static int check_patch_list(struct patch *patch)
3109 int err = 0;
3111 prepare_fn_table(patch);
3112 while (patch) {
3113 if (apply_verbosely)
3114 say_patch_name(stderr,
3115 "Checking patch ", patch, "...\n");
3116 err |= check_patch(patch);
3117 patch = patch->next;
3119 return err;
3122 /* This function tries to read the sha1 from the current index */
3123 static int get_current_sha1(const char *path, unsigned char *sha1)
3125 int pos;
3127 if (read_cache() < 0)
3128 return -1;
3129 pos = cache_name_pos(path, strlen(path));
3130 if (pos < 0)
3131 return -1;
3132 hashcpy(sha1, active_cache[pos]->sha1);
3133 return 0;
3136 /* Build an index that contains the just the files needed for a 3way merge */
3137 static void build_fake_ancestor(struct patch *list, const char *filename)
3139 struct patch *patch;
3140 struct index_state result = { NULL };
3141 int fd;
3143 /* Once we start supporting the reverse patch, it may be
3144 * worth showing the new sha1 prefix, but until then...
3146 for (patch = list; patch; patch = patch->next) {
3147 const unsigned char *sha1_ptr;
3148 unsigned char sha1[20];
3149 struct cache_entry *ce;
3150 const char *name;
3152 name = patch->old_name ? patch->old_name : patch->new_name;
3153 if (0 < patch->is_new)
3154 continue;
3155 else if (get_sha1(patch->old_sha1_prefix, sha1))
3156 /* git diff has no index line for mode/type changes */
3157 if (!patch->lines_added && !patch->lines_deleted) {
3158 if (get_current_sha1(patch->old_name, sha1))
3159 die("mode change for %s, which is not "
3160 "in current HEAD", name);
3161 sha1_ptr = sha1;
3162 } else
3163 die("sha1 information is lacking or useless "
3164 "(%s).", name);
3165 else
3166 sha1_ptr = sha1;
3168 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3169 if (!ce)
3170 die("make_cache_entry failed for path '%s'", name);
3171 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3172 die ("Could not add %s to temporary index", name);
3175 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3176 if (fd < 0 || write_index(&result, fd) || close(fd))
3177 die ("Could not write temporary index to %s", filename);
3179 discard_index(&result);
3182 static void stat_patch_list(struct patch *patch)
3184 int files, adds, dels;
3186 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3187 files++;
3188 adds += patch->lines_added;
3189 dels += patch->lines_deleted;
3190 show_stats(patch);
3193 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3196 static void numstat_patch_list(struct patch *patch)
3198 for ( ; patch; patch = patch->next) {
3199 const char *name;
3200 name = patch->new_name ? patch->new_name : patch->old_name;
3201 if (patch->is_binary)
3202 printf("-\t-\t");
3203 else
3204 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3205 write_name_quoted(name, stdout, line_termination);
3209 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3211 if (mode)
3212 printf(" %s mode %06o %s\n", newdelete, mode, name);
3213 else
3214 printf(" %s %s\n", newdelete, name);
3217 static void show_mode_change(struct patch *p, int show_name)
3219 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3220 if (show_name)
3221 printf(" mode change %06o => %06o %s\n",
3222 p->old_mode, p->new_mode, p->new_name);
3223 else
3224 printf(" mode change %06o => %06o\n",
3225 p->old_mode, p->new_mode);
3229 static void show_rename_copy(struct patch *p)
3231 const char *renamecopy = p->is_rename ? "rename" : "copy";
3232 const char *old, *new;
3234 /* Find common prefix */
3235 old = p->old_name;
3236 new = p->new_name;
3237 while (1) {
3238 const char *slash_old, *slash_new;
3239 slash_old = strchr(old, '/');
3240 slash_new = strchr(new, '/');
3241 if (!slash_old ||
3242 !slash_new ||
3243 slash_old - old != slash_new - new ||
3244 memcmp(old, new, slash_new - new))
3245 break;
3246 old = slash_old + 1;
3247 new = slash_new + 1;
3249 /* p->old_name thru old is the common prefix, and old and new
3250 * through the end of names are renames
3252 if (old != p->old_name)
3253 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3254 (int)(old - p->old_name), p->old_name,
3255 old, new, p->score);
3256 else
3257 printf(" %s %s => %s (%d%%)\n", renamecopy,
3258 p->old_name, p->new_name, p->score);
3259 show_mode_change(p, 0);
3262 static void summary_patch_list(struct patch *patch)
3264 struct patch *p;
3266 for (p = patch; p; p = p->next) {
3267 if (p->is_new)
3268 show_file_mode_name("create", p->new_mode, p->new_name);
3269 else if (p->is_delete)
3270 show_file_mode_name("delete", p->old_mode, p->old_name);
3271 else {
3272 if (p->is_rename || p->is_copy)
3273 show_rename_copy(p);
3274 else {
3275 if (p->score) {
3276 printf(" rewrite %s (%d%%)\n",
3277 p->new_name, p->score);
3278 show_mode_change(p, 0);
3280 else
3281 show_mode_change(p, 1);
3287 static void patch_stats(struct patch *patch)
3289 int lines = patch->lines_added + patch->lines_deleted;
3291 if (lines > max_change)
3292 max_change = lines;
3293 if (patch->old_name) {
3294 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3295 if (!len)
3296 len = strlen(patch->old_name);
3297 if (len > max_len)
3298 max_len = len;
3300 if (patch->new_name) {
3301 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3302 if (!len)
3303 len = strlen(patch->new_name);
3304 if (len > max_len)
3305 max_len = len;
3309 static void remove_file(struct patch *patch, int rmdir_empty)
3311 if (update_index) {
3312 if (remove_file_from_cache(patch->old_name) < 0)
3313 die("unable to remove %s from index", patch->old_name);
3315 if (!cached) {
3316 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3317 remove_path(patch->old_name);
3322 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3324 struct stat st;
3325 struct cache_entry *ce;
3326 int namelen = strlen(path);
3327 unsigned ce_size = cache_entry_size(namelen);
3329 if (!update_index)
3330 return;
3332 ce = xcalloc(1, ce_size);
3333 memcpy(ce->name, path, namelen);
3334 ce->ce_mode = create_ce_mode(mode);
3335 ce->ce_flags = namelen;
3336 if (S_ISGITLINK(mode)) {
3337 const char *s = buf;
3339 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3340 die("corrupt patch for subproject %s", path);
3341 } else {
3342 if (!cached) {
3343 if (lstat(path, &st) < 0)
3344 die_errno("unable to stat newly created file '%s'",
3345 path);
3346 fill_stat_cache_info(ce, &st);
3348 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3349 die("unable to create backing store for newly created file %s", path);
3351 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3352 die("unable to add cache entry for %s", path);
3355 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3357 int fd;
3358 struct strbuf nbuf = STRBUF_INIT;
3360 if (S_ISGITLINK(mode)) {
3361 struct stat st;
3362 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3363 return 0;
3364 return mkdir(path, 0777);
3367 if (has_symlinks && S_ISLNK(mode))
3368 /* Although buf:size is counted string, it also is NUL
3369 * terminated.
3371 return symlink(buf, path);
3373 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3374 if (fd < 0)
3375 return -1;
3377 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3378 size = nbuf.len;
3379 buf = nbuf.buf;
3381 write_or_die(fd, buf, size);
3382 strbuf_release(&nbuf);
3384 if (close(fd) < 0)
3385 die_errno("closing file '%s'", path);
3386 return 0;
3390 * We optimistically assume that the directories exist,
3391 * which is true 99% of the time anyway. If they don't,
3392 * we create them and try again.
3394 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3396 if (cached)
3397 return;
3398 if (!try_create_file(path, mode, buf, size))
3399 return;
3401 if (errno == ENOENT) {
3402 if (safe_create_leading_directories(path))
3403 return;
3404 if (!try_create_file(path, mode, buf, size))
3405 return;
3408 if (errno == EEXIST || errno == EACCES) {
3409 /* We may be trying to create a file where a directory
3410 * used to be.
3412 struct stat st;
3413 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3414 errno = EEXIST;
3417 if (errno == EEXIST) {
3418 unsigned int nr = getpid();
3420 for (;;) {
3421 char newpath[PATH_MAX];
3422 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3423 if (!try_create_file(newpath, mode, buf, size)) {
3424 if (!rename(newpath, path))
3425 return;
3426 unlink_or_warn(newpath);
3427 break;
3429 if (errno != EEXIST)
3430 break;
3431 ++nr;
3434 die_errno("unable to write file '%s' mode %o", path, mode);
3437 static void create_file(struct patch *patch)
3439 char *path = patch->new_name;
3440 unsigned mode = patch->new_mode;
3441 unsigned long size = patch->resultsize;
3442 char *buf = patch->result;
3444 if (!mode)
3445 mode = S_IFREG | 0644;
3446 create_one_file(path, mode, buf, size);
3447 add_index_file(path, mode, buf, size);
3450 /* phase zero is to remove, phase one is to create */
3451 static void write_out_one_result(struct patch *patch, int phase)
3453 if (patch->is_delete > 0) {
3454 if (phase == 0)
3455 remove_file(patch, 1);
3456 return;
3458 if (patch->is_new > 0 || patch->is_copy) {
3459 if (phase == 1)
3460 create_file(patch);
3461 return;
3464 * Rename or modification boils down to the same
3465 * thing: remove the old, write the new
3467 if (phase == 0)
3468 remove_file(patch, patch->is_rename);
3469 if (phase == 1)
3470 create_file(patch);
3473 static int write_out_one_reject(struct patch *patch)
3475 FILE *rej;
3476 char namebuf[PATH_MAX];
3477 struct fragment *frag;
3478 int cnt = 0;
3480 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3481 if (!frag->rejected)
3482 continue;
3483 cnt++;
3486 if (!cnt) {
3487 if (apply_verbosely)
3488 say_patch_name(stderr,
3489 "Applied patch ", patch, " cleanly.\n");
3490 return 0;
3493 /* This should not happen, because a removal patch that leaves
3494 * contents are marked "rejected" at the patch level.
3496 if (!patch->new_name)
3497 die("internal error");
3499 /* Say this even without --verbose */
3500 say_patch_name(stderr, "Applying patch ", patch, " with");
3501 fprintf(stderr, " %d rejects...\n", cnt);
3503 cnt = strlen(patch->new_name);
3504 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3505 cnt = ARRAY_SIZE(namebuf) - 5;
3506 warning("truncating .rej filename to %.*s.rej",
3507 cnt - 1, patch->new_name);
3509 memcpy(namebuf, patch->new_name, cnt);
3510 memcpy(namebuf + cnt, ".rej", 5);
3512 rej = fopen(namebuf, "w");
3513 if (!rej)
3514 return error("cannot open %s: %s", namebuf, strerror(errno));
3516 /* Normal git tools never deal with .rej, so do not pretend
3517 * this is a git patch by saying --git nor give extended
3518 * headers. While at it, maybe please "kompare" that wants
3519 * the trailing TAB and some garbage at the end of line ;-).
3521 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3522 patch->new_name, patch->new_name);
3523 for (cnt = 1, frag = patch->fragments;
3524 frag;
3525 cnt++, frag = frag->next) {
3526 if (!frag->rejected) {
3527 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3528 continue;
3530 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3531 fprintf(rej, "%.*s", frag->size, frag->patch);
3532 if (frag->patch[frag->size-1] != '\n')
3533 fputc('\n', rej);
3535 fclose(rej);
3536 return -1;
3539 static int write_out_results(struct patch *list, int skipped_patch)
3541 int phase;
3542 int errs = 0;
3543 struct patch *l;
3545 if (!list && !skipped_patch)
3546 return error("No changes");
3548 for (phase = 0; phase < 2; phase++) {
3549 l = list;
3550 while (l) {
3551 if (l->rejected)
3552 errs = 1;
3553 else {
3554 write_out_one_result(l, phase);
3555 if (phase == 1 && write_out_one_reject(l))
3556 errs = 1;
3558 l = l->next;
3561 return errs;
3564 static struct lock_file lock_file;
3566 static struct string_list limit_by_name;
3567 static int has_include;
3568 static void add_name_limit(const char *name, int exclude)
3570 struct string_list_item *it;
3572 it = string_list_append(&limit_by_name, name);
3573 it->util = exclude ? NULL : (void *) 1;
3576 static int use_patch(struct patch *p)
3578 const char *pathname = p->new_name ? p->new_name : p->old_name;
3579 int i;
3581 /* Paths outside are not touched regardless of "--include" */
3582 if (0 < prefix_length) {
3583 int pathlen = strlen(pathname);
3584 if (pathlen <= prefix_length ||
3585 memcmp(prefix, pathname, prefix_length))
3586 return 0;
3589 /* See if it matches any of exclude/include rule */
3590 for (i = 0; i < limit_by_name.nr; i++) {
3591 struct string_list_item *it = &limit_by_name.items[i];
3592 if (!fnmatch(it->string, pathname, 0))
3593 return (it->util != NULL);
3597 * If we had any include, a path that does not match any rule is
3598 * not used. Otherwise, we saw bunch of exclude rules (or none)
3599 * and such a path is used.
3601 return !has_include;
3605 static void prefix_one(char **name)
3607 char *old_name = *name;
3608 if (!old_name)
3609 return;
3610 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3611 free(old_name);
3614 static void prefix_patches(struct patch *p)
3616 if (!prefix || p->is_toplevel_relative)
3617 return;
3618 for ( ; p; p = p->next) {
3619 if (p->new_name == p->old_name) {
3620 char *prefixed = p->new_name;
3621 prefix_one(&prefixed);
3622 p->new_name = p->old_name = prefixed;
3624 else {
3625 prefix_one(&p->new_name);
3626 prefix_one(&p->old_name);
3631 #define INACCURATE_EOF (1<<0)
3632 #define RECOUNT (1<<1)
3634 static int apply_patch(int fd, const char *filename, int options)
3636 size_t offset;
3637 struct strbuf buf = STRBUF_INIT;
3638 struct patch *list = NULL, **listp = &list;
3639 int skipped_patch = 0;
3641 /* FIXME - memory leak when using multiple patch files as inputs */
3642 memset(&fn_table, 0, sizeof(struct string_list));
3643 patch_input_file = filename;
3644 read_patch_file(&buf, fd);
3645 offset = 0;
3646 while (offset < buf.len) {
3647 struct patch *patch;
3648 int nr;
3650 patch = xcalloc(1, sizeof(*patch));
3651 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3652 patch->recount = !!(options & RECOUNT);
3653 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3654 if (nr < 0)
3655 break;
3656 if (apply_in_reverse)
3657 reverse_patches(patch);
3658 if (prefix)
3659 prefix_patches(patch);
3660 if (use_patch(patch)) {
3661 patch_stats(patch);
3662 *listp = patch;
3663 listp = &patch->next;
3665 else {
3666 /* perhaps free it a bit better? */
3667 free(patch);
3668 skipped_patch++;
3670 offset += nr;
3673 if (whitespace_error && (ws_error_action == die_on_ws_error))
3674 apply = 0;
3676 update_index = check_index && apply;
3677 if (update_index && newfd < 0)
3678 newfd = hold_locked_index(&lock_file, 1);
3680 if (check_index) {
3681 if (read_cache() < 0)
3682 die("unable to read index file");
3685 if ((check || apply) &&
3686 check_patch_list(list) < 0 &&
3687 !apply_with_reject)
3688 exit(1);
3690 if (apply && write_out_results(list, skipped_patch))
3691 exit(1);
3693 if (fake_ancestor)
3694 build_fake_ancestor(list, fake_ancestor);
3696 if (diffstat)
3697 stat_patch_list(list);
3699 if (numstat)
3700 numstat_patch_list(list);
3702 if (summary)
3703 summary_patch_list(list);
3705 strbuf_release(&buf);
3706 return 0;
3709 static int git_apply_config(const char *var, const char *value, void *cb)
3711 if (!strcmp(var, "apply.whitespace"))
3712 return git_config_string(&apply_default_whitespace, var, value);
3713 else if (!strcmp(var, "apply.ignorewhitespace"))
3714 return git_config_string(&apply_default_ignorewhitespace, var, value);
3715 return git_default_config(var, value, cb);
3718 static int option_parse_exclude(const struct option *opt,
3719 const char *arg, int unset)
3721 add_name_limit(arg, 1);
3722 return 0;
3725 static int option_parse_include(const struct option *opt,
3726 const char *arg, int unset)
3728 add_name_limit(arg, 0);
3729 has_include = 1;
3730 return 0;
3733 static int option_parse_p(const struct option *opt,
3734 const char *arg, int unset)
3736 p_value = atoi(arg);
3737 p_value_known = 1;
3738 return 0;
3741 static int option_parse_z(const struct option *opt,
3742 const char *arg, int unset)
3744 if (unset)
3745 line_termination = '\n';
3746 else
3747 line_termination = 0;
3748 return 0;
3751 static int option_parse_space_change(const struct option *opt,
3752 const char *arg, int unset)
3754 if (unset)
3755 ws_ignore_action = ignore_ws_none;
3756 else
3757 ws_ignore_action = ignore_ws_change;
3758 return 0;
3761 static int option_parse_whitespace(const struct option *opt,
3762 const char *arg, int unset)
3764 const char **whitespace_option = opt->value;
3766 *whitespace_option = arg;
3767 parse_whitespace_option(arg);
3768 return 0;
3771 static int option_parse_directory(const struct option *opt,
3772 const char *arg, int unset)
3774 root_len = strlen(arg);
3775 if (root_len && arg[root_len - 1] != '/') {
3776 char *new_root;
3777 root = new_root = xmalloc(root_len + 2);
3778 strcpy(new_root, arg);
3779 strcpy(new_root + root_len++, "/");
3780 } else
3781 root = arg;
3782 return 0;
3785 int cmd_apply(int argc, const char **argv, const char *prefix_)
3787 int i;
3788 int errs = 0;
3789 int is_not_gitdir = !startup_info->have_repository;
3790 int binary;
3791 int force_apply = 0;
3793 const char *whitespace_option = NULL;
3795 struct option builtin_apply_options[] = {
3796 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3797 "don't apply changes matching the given path",
3798 0, option_parse_exclude },
3799 { OPTION_CALLBACK, 0, "include", NULL, "path",
3800 "apply changes matching the given path",
3801 0, option_parse_include },
3802 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3803 "remove <num> leading slashes from traditional diff paths",
3804 0, option_parse_p },
3805 OPT_BOOLEAN(0, "no-add", &no_add,
3806 "ignore additions made by the patch"),
3807 OPT_BOOLEAN(0, "stat", &diffstat,
3808 "instead of applying the patch, output diffstat for the input"),
3809 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3810 NULL, "old option, now no-op",
3811 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3812 { OPTION_BOOLEAN, 0, "binary", &binary,
3813 NULL, "old option, now no-op",
3814 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3815 OPT_BOOLEAN(0, "numstat", &numstat,
3816 "shows number of added and deleted lines in decimal notation"),
3817 OPT_BOOLEAN(0, "summary", &summary,
3818 "instead of applying the patch, output a summary for the input"),
3819 OPT_BOOLEAN(0, "check", &check,
3820 "instead of applying the patch, see if the patch is applicable"),
3821 OPT_BOOLEAN(0, "index", &check_index,
3822 "make sure the patch is applicable to the current index"),
3823 OPT_BOOLEAN(0, "cached", &cached,
3824 "apply a patch without touching the working tree"),
3825 OPT_BOOLEAN(0, "apply", &force_apply,
3826 "also apply the patch (use with --stat/--summary/--check)"),
3827 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3828 "build a temporary index based on embedded index information"),
3829 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3830 "paths are separated with NUL character",
3831 PARSE_OPT_NOARG, option_parse_z },
3832 OPT_INTEGER('C', NULL, &p_context,
3833 "ensure at least <n> lines of context match"),
3834 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3835 "detect new or modified lines that have whitespace errors",
3836 0, option_parse_whitespace },
3837 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3838 "ignore changes in whitespace when finding context",
3839 PARSE_OPT_NOARG, option_parse_space_change },
3840 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3841 "ignore changes in whitespace when finding context",
3842 PARSE_OPT_NOARG, option_parse_space_change },
3843 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3844 "apply the patch in reverse"),
3845 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3846 "don't expect at least one line of context"),
3847 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3848 "leave the rejected hunks in corresponding *.rej files"),
3849 OPT__VERBOSE(&apply_verbosely),
3850 OPT_BIT(0, "inaccurate-eof", &options,
3851 "tolerate incorrectly detected missing new-line at the end of file",
3852 INACCURATE_EOF),
3853 OPT_BIT(0, "recount", &options,
3854 "do not trust the line counts in the hunk headers",
3855 RECOUNT),
3856 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3857 "prepend <root> to all filenames",
3858 0, option_parse_directory },
3859 OPT_END()
3862 prefix = prefix_;
3863 prefix_length = prefix ? strlen(prefix) : 0;
3864 git_config(git_apply_config, NULL);
3865 if (apply_default_whitespace)
3866 parse_whitespace_option(apply_default_whitespace);
3867 if (apply_default_ignorewhitespace)
3868 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3870 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3871 apply_usage, 0);
3873 if (apply_with_reject)
3874 apply = apply_verbosely = 1;
3875 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3876 apply = 0;
3877 if (check_index && is_not_gitdir)
3878 die("--index outside a repository");
3879 if (cached) {
3880 if (is_not_gitdir)
3881 die("--cached outside a repository");
3882 check_index = 1;
3884 for (i = 0; i < argc; i++) {
3885 const char *arg = argv[i];
3886 int fd;
3888 if (!strcmp(arg, "-")) {
3889 errs |= apply_patch(0, "<stdin>", options);
3890 read_stdin = 0;
3891 continue;
3892 } else if (0 < prefix_length)
3893 arg = prefix_filename(prefix, prefix_length, arg);
3895 fd = open(arg, O_RDONLY);
3896 if (fd < 0)
3897 die_errno("can't open patch '%s'", arg);
3898 read_stdin = 0;
3899 set_default_whitespace_mode(whitespace_option);
3900 errs |= apply_patch(fd, arg, options);
3901 close(fd);
3903 set_default_whitespace_mode(whitespace_option);
3904 if (read_stdin)
3905 errs |= apply_patch(0, "<stdin>", options);
3906 if (whitespace_error) {
3907 if (squelch_whitespace_errors &&
3908 squelch_whitespace_errors < whitespace_error) {
3909 int squelched =
3910 whitespace_error - squelch_whitespace_errors;
3911 warning("squelched %d "
3912 "whitespace error%s",
3913 squelched,
3914 squelched == 1 ? "" : "s");
3916 if (ws_error_action == die_on_ws_error)
3917 die("%d line%s add%s whitespace errors.",
3918 whitespace_error,
3919 whitespace_error == 1 ? "" : "s",
3920 whitespace_error == 1 ? "s" : "");
3921 if (applied_after_fixing_ws && apply)
3922 warning("%d line%s applied after"
3923 " fixing whitespace errors.",
3924 applied_after_fixing_ws,
3925 applied_after_fixing_ws == 1 ? "" : "s");
3926 else if (whitespace_error)
3927 warning("%d line%s add%s whitespace errors.",
3928 whitespace_error,
3929 whitespace_error == 1 ? "" : "s",
3930 whitespace_error == 1 ? "s" : "");
3933 if (update_index) {
3934 if (write_cache(newfd, active_cache, active_nr) ||
3935 commit_locked_index(&lock_file))
3936 die("Unable to write new index file");
3939 return !!errs;