Merge branch 'aw/git-p4-deletion'
[git/kirr.git] / builtin / apply.c
blob23c18c573b9147b7aaa88a8d1d2f26748e803501
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;
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, NULL, 10);
768 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
769 if (timestamp[m[3].rm_so] == '-')
770 zoneoffset = -zoneoffset;
773 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
774 * (west of GMT) or 1970-01-01 (east of GMT)
776 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
777 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
778 return 0;
780 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
781 strtol(timestamp + 14, NULL, 10) -
782 zoneoffset);
784 return ((zoneoffset < 0 && hourminute == 1440) ||
785 (0 <= zoneoffset && !hourminute));
789 * Get the name etc info from the ---/+++ lines of a traditional patch header
791 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
792 * files, we can happily check the index for a match, but for creating a
793 * new file we should try to match whatever "patch" does. I have no idea.
795 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
797 char *name;
799 first += 4; /* skip "--- " */
800 second += 4; /* skip "+++ " */
801 if (!p_value_known) {
802 int p, q;
803 p = guess_p_value(first);
804 q = guess_p_value(second);
805 if (p < 0) p = q;
806 if (0 <= p && p == q) {
807 p_value = p;
808 p_value_known = 1;
811 if (is_dev_null(first)) {
812 patch->is_new = 1;
813 patch->is_delete = 0;
814 name = find_name_traditional(second, NULL, p_value);
815 patch->new_name = name;
816 } else if (is_dev_null(second)) {
817 patch->is_new = 0;
818 patch->is_delete = 1;
819 name = find_name_traditional(first, NULL, p_value);
820 patch->old_name = name;
821 } else {
822 name = find_name_traditional(first, NULL, p_value);
823 name = find_name_traditional(second, name, p_value);
824 if (has_epoch_timestamp(first)) {
825 patch->is_new = 1;
826 patch->is_delete = 0;
827 patch->new_name = name;
828 } else if (has_epoch_timestamp(second)) {
829 patch->is_new = 0;
830 patch->is_delete = 1;
831 patch->old_name = name;
832 } else {
833 patch->old_name = patch->new_name = name;
836 if (!name)
837 die("unable to find filename in patch at line %d", linenr);
840 static int gitdiff_hdrend(const char *line, struct patch *patch)
842 return -1;
846 * We're anal about diff header consistency, to make
847 * sure that we don't end up having strange ambiguous
848 * patches floating around.
850 * As a result, gitdiff_{old|new}name() will check
851 * their names against any previous information, just
852 * to make sure..
854 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
856 if (!orig_name && !isnull)
857 return find_name(line, NULL, p_value, TERM_TAB);
859 if (orig_name) {
860 int len;
861 const char *name;
862 char *another;
863 name = orig_name;
864 len = strlen(name);
865 if (isnull)
866 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
867 another = find_name(line, NULL, p_value, TERM_TAB);
868 if (!another || memcmp(another, name, len + 1))
869 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
870 free(another);
871 return orig_name;
873 else {
874 /* expect "/dev/null" */
875 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
876 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
877 return NULL;
881 static int gitdiff_oldname(const char *line, struct patch *patch)
883 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
884 return 0;
887 static int gitdiff_newname(const char *line, struct patch *patch)
889 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
890 return 0;
893 static int gitdiff_oldmode(const char *line, struct patch *patch)
895 patch->old_mode = strtoul(line, NULL, 8);
896 return 0;
899 static int gitdiff_newmode(const char *line, struct patch *patch)
901 patch->new_mode = strtoul(line, NULL, 8);
902 return 0;
905 static int gitdiff_delete(const char *line, struct patch *patch)
907 patch->is_delete = 1;
908 patch->old_name = patch->def_name;
909 return gitdiff_oldmode(line, patch);
912 static int gitdiff_newfile(const char *line, struct patch *patch)
914 patch->is_new = 1;
915 patch->new_name = patch->def_name;
916 return gitdiff_newmode(line, patch);
919 static int gitdiff_copysrc(const char *line, struct patch *patch)
921 patch->is_copy = 1;
922 patch->old_name = find_name(line, NULL, 0, 0);
923 return 0;
926 static int gitdiff_copydst(const char *line, struct patch *patch)
928 patch->is_copy = 1;
929 patch->new_name = find_name(line, NULL, 0, 0);
930 return 0;
933 static int gitdiff_renamesrc(const char *line, struct patch *patch)
935 patch->is_rename = 1;
936 patch->old_name = find_name(line, NULL, 0, 0);
937 return 0;
940 static int gitdiff_renamedst(const char *line, struct patch *patch)
942 patch->is_rename = 1;
943 patch->new_name = find_name(line, NULL, 0, 0);
944 return 0;
947 static int gitdiff_similarity(const char *line, struct patch *patch)
949 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
950 patch->score = 0;
951 return 0;
954 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
956 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
957 patch->score = 0;
958 return 0;
961 static int gitdiff_index(const char *line, struct patch *patch)
964 * index line is N hexadecimal, "..", N hexadecimal,
965 * and optional space with octal mode.
967 const char *ptr, *eol;
968 int len;
970 ptr = strchr(line, '.');
971 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
972 return 0;
973 len = ptr - line;
974 memcpy(patch->old_sha1_prefix, line, len);
975 patch->old_sha1_prefix[len] = 0;
977 line = ptr + 2;
978 ptr = strchr(line, ' ');
979 eol = strchr(line, '\n');
981 if (!ptr || eol < ptr)
982 ptr = eol;
983 len = ptr - line;
985 if (40 < len)
986 return 0;
987 memcpy(patch->new_sha1_prefix, line, len);
988 patch->new_sha1_prefix[len] = 0;
989 if (*ptr == ' ')
990 patch->old_mode = strtoul(ptr+1, NULL, 8);
991 return 0;
995 * This is normal for a diff that doesn't change anything: we'll fall through
996 * into the next diff. Tell the parser to break out.
998 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1000 return -1;
1003 static const char *stop_at_slash(const char *line, int llen)
1005 int nslash = p_value;
1006 int i;
1008 for (i = 0; i < llen; i++) {
1009 int ch = line[i];
1010 if (ch == '/' && --nslash <= 0)
1011 return &line[i];
1013 return NULL;
1017 * This is to extract the same name that appears on "diff --git"
1018 * line. We do not find and return anything if it is a rename
1019 * patch, and it is OK because we will find the name elsewhere.
1020 * We need to reliably find name only when it is mode-change only,
1021 * creation or deletion of an empty file. In any of these cases,
1022 * both sides are the same name under a/ and b/ respectively.
1024 static char *git_header_name(char *line, int llen)
1026 const char *name;
1027 const char *second = NULL;
1028 size_t len;
1030 line += strlen("diff --git ");
1031 llen -= strlen("diff --git ");
1033 if (*line == '"') {
1034 const char *cp;
1035 struct strbuf first = STRBUF_INIT;
1036 struct strbuf sp = STRBUF_INIT;
1038 if (unquote_c_style(&first, line, &second))
1039 goto free_and_fail1;
1041 /* advance to the first slash */
1042 cp = stop_at_slash(first.buf, first.len);
1043 /* we do not accept absolute paths */
1044 if (!cp || cp == first.buf)
1045 goto free_and_fail1;
1046 strbuf_remove(&first, 0, cp + 1 - first.buf);
1049 * second points at one past closing dq of name.
1050 * find the second name.
1052 while ((second < line + llen) && isspace(*second))
1053 second++;
1055 if (line + llen <= second)
1056 goto free_and_fail1;
1057 if (*second == '"') {
1058 if (unquote_c_style(&sp, second, NULL))
1059 goto free_and_fail1;
1060 cp = stop_at_slash(sp.buf, sp.len);
1061 if (!cp || cp == sp.buf)
1062 goto free_and_fail1;
1063 /* They must match, otherwise ignore */
1064 if (strcmp(cp + 1, first.buf))
1065 goto free_and_fail1;
1066 strbuf_release(&sp);
1067 return strbuf_detach(&first, NULL);
1070 /* unquoted second */
1071 cp = stop_at_slash(second, line + llen - second);
1072 if (!cp || cp == second)
1073 goto free_and_fail1;
1074 cp++;
1075 if (line + llen - cp != first.len + 1 ||
1076 memcmp(first.buf, cp, first.len))
1077 goto free_and_fail1;
1078 return strbuf_detach(&first, NULL);
1080 free_and_fail1:
1081 strbuf_release(&first);
1082 strbuf_release(&sp);
1083 return NULL;
1086 /* unquoted first name */
1087 name = stop_at_slash(line, llen);
1088 if (!name || name == line)
1089 return NULL;
1090 name++;
1093 * since the first name is unquoted, a dq if exists must be
1094 * the beginning of the second name.
1096 for (second = name; second < line + llen; second++) {
1097 if (*second == '"') {
1098 struct strbuf sp = STRBUF_INIT;
1099 const char *np;
1101 if (unquote_c_style(&sp, second, NULL))
1102 goto free_and_fail2;
1104 np = stop_at_slash(sp.buf, sp.len);
1105 if (!np || np == sp.buf)
1106 goto free_and_fail2;
1107 np++;
1109 len = sp.buf + sp.len - np;
1110 if (len < second - name &&
1111 !strncmp(np, name, len) &&
1112 isspace(name[len])) {
1113 /* Good */
1114 strbuf_remove(&sp, 0, np - sp.buf);
1115 return strbuf_detach(&sp, NULL);
1118 free_and_fail2:
1119 strbuf_release(&sp);
1120 return NULL;
1125 * Accept a name only if it shows up twice, exactly the same
1126 * form.
1128 for (len = 0 ; ; len++) {
1129 switch (name[len]) {
1130 default:
1131 continue;
1132 case '\n':
1133 return NULL;
1134 case '\t': case ' ':
1135 second = name+len;
1136 for (;;) {
1137 char c = *second++;
1138 if (c == '\n')
1139 return NULL;
1140 if (c == '/')
1141 break;
1143 if (second[len] == '\n' && !memcmp(name, second, len)) {
1144 return xmemdupz(name, len);
1150 /* Verify that we recognize the lines following a git header */
1151 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
1153 unsigned long offset;
1155 /* A git diff has explicit new/delete information, so we don't guess */
1156 patch->is_new = 0;
1157 patch->is_delete = 0;
1160 * Some things may not have the old name in the
1161 * rest of the headers anywhere (pure mode changes,
1162 * or removing or adding empty files), so we get
1163 * the default name from the header.
1165 patch->def_name = git_header_name(line, len);
1166 if (patch->def_name && root) {
1167 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1168 strcpy(s, root);
1169 strcpy(s + root_len, patch->def_name);
1170 free(patch->def_name);
1171 patch->def_name = s;
1174 line += len;
1175 size -= len;
1176 linenr++;
1177 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1178 static const struct opentry {
1179 const char *str;
1180 int (*fn)(const char *, struct patch *);
1181 } optable[] = {
1182 { "@@ -", gitdiff_hdrend },
1183 { "--- ", gitdiff_oldname },
1184 { "+++ ", gitdiff_newname },
1185 { "old mode ", gitdiff_oldmode },
1186 { "new mode ", gitdiff_newmode },
1187 { "deleted file mode ", gitdiff_delete },
1188 { "new file mode ", gitdiff_newfile },
1189 { "copy from ", gitdiff_copysrc },
1190 { "copy to ", gitdiff_copydst },
1191 { "rename old ", gitdiff_renamesrc },
1192 { "rename new ", gitdiff_renamedst },
1193 { "rename from ", gitdiff_renamesrc },
1194 { "rename to ", gitdiff_renamedst },
1195 { "similarity index ", gitdiff_similarity },
1196 { "dissimilarity index ", gitdiff_dissimilarity },
1197 { "index ", gitdiff_index },
1198 { "", gitdiff_unrecognized },
1200 int i;
1202 len = linelen(line, size);
1203 if (!len || line[len-1] != '\n')
1204 break;
1205 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1206 const struct opentry *p = optable + i;
1207 int oplen = strlen(p->str);
1208 if (len < oplen || memcmp(p->str, line, oplen))
1209 continue;
1210 if (p->fn(line + oplen, patch) < 0)
1211 return offset;
1212 break;
1216 return offset;
1219 static int parse_num(const char *line, unsigned long *p)
1221 char *ptr;
1223 if (!isdigit(*line))
1224 return 0;
1225 *p = strtoul(line, &ptr, 10);
1226 return ptr - line;
1229 static int parse_range(const char *line, int len, int offset, const char *expect,
1230 unsigned long *p1, unsigned long *p2)
1232 int digits, ex;
1234 if (offset < 0 || offset >= len)
1235 return -1;
1236 line += offset;
1237 len -= offset;
1239 digits = parse_num(line, p1);
1240 if (!digits)
1241 return -1;
1243 offset += digits;
1244 line += digits;
1245 len -= digits;
1247 *p2 = 1;
1248 if (*line == ',') {
1249 digits = parse_num(line+1, p2);
1250 if (!digits)
1251 return -1;
1253 offset += digits+1;
1254 line += digits+1;
1255 len -= digits+1;
1258 ex = strlen(expect);
1259 if (ex > len)
1260 return -1;
1261 if (memcmp(line, expect, ex))
1262 return -1;
1264 return offset + ex;
1267 static void recount_diff(char *line, int size, struct fragment *fragment)
1269 int oldlines = 0, newlines = 0, ret = 0;
1271 if (size < 1) {
1272 warning("recount: ignore empty hunk");
1273 return;
1276 for (;;) {
1277 int len = linelen(line, size);
1278 size -= len;
1279 line += len;
1281 if (size < 1)
1282 break;
1284 switch (*line) {
1285 case ' ': case '\n':
1286 newlines++;
1287 /* fall through */
1288 case '-':
1289 oldlines++;
1290 continue;
1291 case '+':
1292 newlines++;
1293 continue;
1294 case '\\':
1295 continue;
1296 case '@':
1297 ret = size < 3 || prefixcmp(line, "@@ ");
1298 break;
1299 case 'd':
1300 ret = size < 5 || prefixcmp(line, "diff ");
1301 break;
1302 default:
1303 ret = -1;
1304 break;
1306 if (ret) {
1307 warning("recount: unexpected line: %.*s",
1308 (int)linelen(line, size), line);
1309 return;
1311 break;
1313 fragment->oldlines = oldlines;
1314 fragment->newlines = newlines;
1318 * Parse a unified diff fragment header of the
1319 * form "@@ -a,b +c,d @@"
1321 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1323 int offset;
1325 if (!len || line[len-1] != '\n')
1326 return -1;
1328 /* Figure out the number of lines in a fragment */
1329 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1330 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1332 return offset;
1335 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1337 unsigned long offset, len;
1339 patch->is_toplevel_relative = 0;
1340 patch->is_rename = patch->is_copy = 0;
1341 patch->is_new = patch->is_delete = -1;
1342 patch->old_mode = patch->new_mode = 0;
1343 patch->old_name = patch->new_name = NULL;
1344 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1345 unsigned long nextlen;
1347 len = linelen(line, size);
1348 if (!len)
1349 break;
1351 /* Testing this early allows us to take a few shortcuts.. */
1352 if (len < 6)
1353 continue;
1356 * Make sure we don't find any unconnected patch fragments.
1357 * That's a sign that we didn't find a header, and that a
1358 * patch has become corrupted/broken up.
1360 if (!memcmp("@@ -", line, 4)) {
1361 struct fragment dummy;
1362 if (parse_fragment_header(line, len, &dummy) < 0)
1363 continue;
1364 die("patch fragment without header at line %d: %.*s",
1365 linenr, (int)len-1, line);
1368 if (size < len + 6)
1369 break;
1372 * Git patch? It might not have a real patch, just a rename
1373 * or mode change, so we handle that specially
1375 if (!memcmp("diff --git ", line, 11)) {
1376 int git_hdr_len = parse_git_header(line, len, size, patch);
1377 if (git_hdr_len <= len)
1378 continue;
1379 if (!patch->old_name && !patch->new_name) {
1380 if (!patch->def_name)
1381 die("git diff header lacks filename information when removing "
1382 "%d leading pathname components (line %d)" , p_value, linenr);
1383 patch->old_name = patch->new_name = patch->def_name;
1385 patch->is_toplevel_relative = 1;
1386 *hdrsize = git_hdr_len;
1387 return offset;
1390 /* --- followed by +++ ? */
1391 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1392 continue;
1395 * We only accept unified patches, so we want it to
1396 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1397 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1399 nextlen = linelen(line + len, size - len);
1400 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1401 continue;
1403 /* Ok, we'll consider it a patch */
1404 parse_traditional_patch(line, line+len, patch);
1405 *hdrsize = len + nextlen;
1406 linenr += 2;
1407 return offset;
1409 return -1;
1412 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1414 char *err;
1416 if (!result)
1417 return;
1419 whitespace_error++;
1420 if (squelch_whitespace_errors &&
1421 squelch_whitespace_errors < whitespace_error)
1422 return;
1424 err = whitespace_error_string(result);
1425 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1426 patch_input_file, linenr, err, len, line);
1427 free(err);
1430 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1432 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1434 record_ws_error(result, line + 1, len - 2, linenr);
1438 * Parse a unified diff. Note that this really needs to parse each
1439 * fragment separately, since the only way to know the difference
1440 * between a "---" that is part of a patch, and a "---" that starts
1441 * the next patch is to look at the line counts..
1443 static int parse_fragment(char *line, unsigned long size,
1444 struct patch *patch, struct fragment *fragment)
1446 int added, deleted;
1447 int len = linelen(line, size), offset;
1448 unsigned long oldlines, newlines;
1449 unsigned long leading, trailing;
1451 offset = parse_fragment_header(line, len, fragment);
1452 if (offset < 0)
1453 return -1;
1454 if (offset > 0 && patch->recount)
1455 recount_diff(line + offset, size - offset, fragment);
1456 oldlines = fragment->oldlines;
1457 newlines = fragment->newlines;
1458 leading = 0;
1459 trailing = 0;
1461 /* Parse the thing.. */
1462 line += len;
1463 size -= len;
1464 linenr++;
1465 added = deleted = 0;
1466 for (offset = len;
1467 0 < size;
1468 offset += len, size -= len, line += len, linenr++) {
1469 if (!oldlines && !newlines)
1470 break;
1471 len = linelen(line, size);
1472 if (!len || line[len-1] != '\n')
1473 return -1;
1474 switch (*line) {
1475 default:
1476 return -1;
1477 case '\n': /* newer GNU diff, an empty context line */
1478 case ' ':
1479 oldlines--;
1480 newlines--;
1481 if (!deleted && !added)
1482 leading++;
1483 trailing++;
1484 break;
1485 case '-':
1486 if (apply_in_reverse &&
1487 ws_error_action != nowarn_ws_error)
1488 check_whitespace(line, len, patch->ws_rule);
1489 deleted++;
1490 oldlines--;
1491 trailing = 0;
1492 break;
1493 case '+':
1494 if (!apply_in_reverse &&
1495 ws_error_action != nowarn_ws_error)
1496 check_whitespace(line, len, patch->ws_rule);
1497 added++;
1498 newlines--;
1499 trailing = 0;
1500 break;
1503 * We allow "\ No newline at end of file". Depending
1504 * on locale settings when the patch was produced we
1505 * don't know what this line looks like. The only
1506 * thing we do know is that it begins with "\ ".
1507 * Checking for 12 is just for sanity check -- any
1508 * l10n of "\ No newline..." is at least that long.
1510 case '\\':
1511 if (len < 12 || memcmp(line, "\\ ", 2))
1512 return -1;
1513 break;
1516 if (oldlines || newlines)
1517 return -1;
1518 fragment->leading = leading;
1519 fragment->trailing = trailing;
1522 * If a fragment ends with an incomplete line, we failed to include
1523 * it in the above loop because we hit oldlines == newlines == 0
1524 * before seeing it.
1526 if (12 < size && !memcmp(line, "\\ ", 2))
1527 offset += linelen(line, size);
1529 patch->lines_added += added;
1530 patch->lines_deleted += deleted;
1532 if (0 < patch->is_new && oldlines)
1533 return error("new file depends on old contents");
1534 if (0 < patch->is_delete && newlines)
1535 return error("deleted file still has contents");
1536 return offset;
1539 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1541 unsigned long offset = 0;
1542 unsigned long oldlines = 0, newlines = 0, context = 0;
1543 struct fragment **fragp = &patch->fragments;
1545 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1546 struct fragment *fragment;
1547 int len;
1549 fragment = xcalloc(1, sizeof(*fragment));
1550 fragment->linenr = linenr;
1551 len = parse_fragment(line, size, patch, fragment);
1552 if (len <= 0)
1553 die("corrupt patch at line %d", linenr);
1554 fragment->patch = line;
1555 fragment->size = len;
1556 oldlines += fragment->oldlines;
1557 newlines += fragment->newlines;
1558 context += fragment->leading + fragment->trailing;
1560 *fragp = fragment;
1561 fragp = &fragment->next;
1563 offset += len;
1564 line += len;
1565 size -= len;
1569 * If something was removed (i.e. we have old-lines) it cannot
1570 * be creation, and if something was added it cannot be
1571 * deletion. However, the reverse is not true; --unified=0
1572 * patches that only add are not necessarily creation even
1573 * though they do not have any old lines, and ones that only
1574 * delete are not necessarily deletion.
1576 * Unfortunately, a real creation/deletion patch do _not_ have
1577 * any context line by definition, so we cannot safely tell it
1578 * apart with --unified=0 insanity. At least if the patch has
1579 * more than one hunk it is not creation or deletion.
1581 if (patch->is_new < 0 &&
1582 (oldlines || (patch->fragments && patch->fragments->next)))
1583 patch->is_new = 0;
1584 if (patch->is_delete < 0 &&
1585 (newlines || (patch->fragments && patch->fragments->next)))
1586 patch->is_delete = 0;
1588 if (0 < patch->is_new && oldlines)
1589 die("new file %s depends on old contents", patch->new_name);
1590 if (0 < patch->is_delete && newlines)
1591 die("deleted file %s still has contents", patch->old_name);
1592 if (!patch->is_delete && !newlines && context)
1593 fprintf(stderr, "** warning: file %s becomes empty but "
1594 "is not deleted\n", patch->new_name);
1596 return offset;
1599 static inline int metadata_changes(struct patch *patch)
1601 return patch->is_rename > 0 ||
1602 patch->is_copy > 0 ||
1603 patch->is_new > 0 ||
1604 patch->is_delete ||
1605 (patch->old_mode && patch->new_mode &&
1606 patch->old_mode != patch->new_mode);
1609 static char *inflate_it(const void *data, unsigned long size,
1610 unsigned long inflated_size)
1612 z_stream stream;
1613 void *out;
1614 int st;
1616 memset(&stream, 0, sizeof(stream));
1618 stream.next_in = (unsigned char *)data;
1619 stream.avail_in = size;
1620 stream.next_out = out = xmalloc(inflated_size);
1621 stream.avail_out = inflated_size;
1622 git_inflate_init(&stream);
1623 st = git_inflate(&stream, Z_FINISH);
1624 git_inflate_end(&stream);
1625 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1626 free(out);
1627 return NULL;
1629 return out;
1632 static struct fragment *parse_binary_hunk(char **buf_p,
1633 unsigned long *sz_p,
1634 int *status_p,
1635 int *used_p)
1638 * Expect a line that begins with binary patch method ("literal"
1639 * or "delta"), followed by the length of data before deflating.
1640 * a sequence of 'length-byte' followed by base-85 encoded data
1641 * should follow, terminated by a newline.
1643 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1644 * and we would limit the patch line to 66 characters,
1645 * so one line can fit up to 13 groups that would decode
1646 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1647 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1649 int llen, used;
1650 unsigned long size = *sz_p;
1651 char *buffer = *buf_p;
1652 int patch_method;
1653 unsigned long origlen;
1654 char *data = NULL;
1655 int hunk_size = 0;
1656 struct fragment *frag;
1658 llen = linelen(buffer, size);
1659 used = llen;
1661 *status_p = 0;
1663 if (!prefixcmp(buffer, "delta ")) {
1664 patch_method = BINARY_DELTA_DEFLATED;
1665 origlen = strtoul(buffer + 6, NULL, 10);
1667 else if (!prefixcmp(buffer, "literal ")) {
1668 patch_method = BINARY_LITERAL_DEFLATED;
1669 origlen = strtoul(buffer + 8, NULL, 10);
1671 else
1672 return NULL;
1674 linenr++;
1675 buffer += llen;
1676 while (1) {
1677 int byte_length, max_byte_length, newsize;
1678 llen = linelen(buffer, size);
1679 used += llen;
1680 linenr++;
1681 if (llen == 1) {
1682 /* consume the blank line */
1683 buffer++;
1684 size--;
1685 break;
1688 * Minimum line is "A00000\n" which is 7-byte long,
1689 * and the line length must be multiple of 5 plus 2.
1691 if ((llen < 7) || (llen-2) % 5)
1692 goto corrupt;
1693 max_byte_length = (llen - 2) / 5 * 4;
1694 byte_length = *buffer;
1695 if ('A' <= byte_length && byte_length <= 'Z')
1696 byte_length = byte_length - 'A' + 1;
1697 else if ('a' <= byte_length && byte_length <= 'z')
1698 byte_length = byte_length - 'a' + 27;
1699 else
1700 goto corrupt;
1701 /* if the input length was not multiple of 4, we would
1702 * have filler at the end but the filler should never
1703 * exceed 3 bytes
1705 if (max_byte_length < byte_length ||
1706 byte_length <= max_byte_length - 4)
1707 goto corrupt;
1708 newsize = hunk_size + byte_length;
1709 data = xrealloc(data, newsize);
1710 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1711 goto corrupt;
1712 hunk_size = newsize;
1713 buffer += llen;
1714 size -= llen;
1717 frag = xcalloc(1, sizeof(*frag));
1718 frag->patch = inflate_it(data, hunk_size, origlen);
1719 if (!frag->patch)
1720 goto corrupt;
1721 free(data);
1722 frag->size = origlen;
1723 *buf_p = buffer;
1724 *sz_p = size;
1725 *used_p = used;
1726 frag->binary_patch_method = patch_method;
1727 return frag;
1729 corrupt:
1730 free(data);
1731 *status_p = -1;
1732 error("corrupt binary patch at line %d: %.*s",
1733 linenr-1, llen-1, buffer);
1734 return NULL;
1737 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1740 * We have read "GIT binary patch\n"; what follows is a line
1741 * that says the patch method (currently, either "literal" or
1742 * "delta") and the length of data before deflating; a
1743 * sequence of 'length-byte' followed by base-85 encoded data
1744 * follows.
1746 * When a binary patch is reversible, there is another binary
1747 * hunk in the same format, starting with patch method (either
1748 * "literal" or "delta") with the length of data, and a sequence
1749 * of length-byte + base-85 encoded data, terminated with another
1750 * empty line. This data, when applied to the postimage, produces
1751 * the preimage.
1753 struct fragment *forward;
1754 struct fragment *reverse;
1755 int status;
1756 int used, used_1;
1758 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1759 if (!forward && !status)
1760 /* there has to be one hunk (forward hunk) */
1761 return error("unrecognized binary patch at line %d", linenr-1);
1762 if (status)
1763 /* otherwise we already gave an error message */
1764 return status;
1766 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1767 if (reverse)
1768 used += used_1;
1769 else if (status) {
1771 * Not having reverse hunk is not an error, but having
1772 * a corrupt reverse hunk is.
1774 free((void*) forward->patch);
1775 free(forward);
1776 return status;
1778 forward->next = reverse;
1779 patch->fragments = forward;
1780 patch->is_binary = 1;
1781 return used;
1784 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1786 int hdrsize, patchsize;
1787 int offset = find_header(buffer, size, &hdrsize, patch);
1789 if (offset < 0)
1790 return offset;
1792 patch->ws_rule = whitespace_rule(patch->new_name
1793 ? patch->new_name
1794 : patch->old_name);
1796 patchsize = parse_single_patch(buffer + offset + hdrsize,
1797 size - offset - hdrsize, patch);
1799 if (!patchsize) {
1800 static const char *binhdr[] = {
1801 "Binary files ",
1802 "Files ",
1803 NULL,
1805 static const char git_binary[] = "GIT binary patch\n";
1806 int i;
1807 int hd = hdrsize + offset;
1808 unsigned long llen = linelen(buffer + hd, size - hd);
1810 if (llen == sizeof(git_binary) - 1 &&
1811 !memcmp(git_binary, buffer + hd, llen)) {
1812 int used;
1813 linenr++;
1814 used = parse_binary(buffer + hd + llen,
1815 size - hd - llen, patch);
1816 if (used)
1817 patchsize = used + llen;
1818 else
1819 patchsize = 0;
1821 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1822 for (i = 0; binhdr[i]; i++) {
1823 int len = strlen(binhdr[i]);
1824 if (len < size - hd &&
1825 !memcmp(binhdr[i], buffer + hd, len)) {
1826 linenr++;
1827 patch->is_binary = 1;
1828 patchsize = llen;
1829 break;
1834 /* Empty patch cannot be applied if it is a text patch
1835 * without metadata change. A binary patch appears
1836 * empty to us here.
1838 if ((apply || check) &&
1839 (!patch->is_binary && !metadata_changes(patch)))
1840 die("patch with only garbage at line %d", linenr);
1843 return offset + hdrsize + patchsize;
1846 #define swap(a,b) myswap((a),(b),sizeof(a))
1848 #define myswap(a, b, size) do { \
1849 unsigned char mytmp[size]; \
1850 memcpy(mytmp, &a, size); \
1851 memcpy(&a, &b, size); \
1852 memcpy(&b, mytmp, size); \
1853 } while (0)
1855 static void reverse_patches(struct patch *p)
1857 for (; p; p = p->next) {
1858 struct fragment *frag = p->fragments;
1860 swap(p->new_name, p->old_name);
1861 swap(p->new_mode, p->old_mode);
1862 swap(p->is_new, p->is_delete);
1863 swap(p->lines_added, p->lines_deleted);
1864 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1866 for (; frag; frag = frag->next) {
1867 swap(frag->newpos, frag->oldpos);
1868 swap(frag->newlines, frag->oldlines);
1873 static const char pluses[] =
1874 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1875 static const char minuses[]=
1876 "----------------------------------------------------------------------";
1878 static void show_stats(struct patch *patch)
1880 struct strbuf qname = STRBUF_INIT;
1881 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1882 int max, add, del;
1884 quote_c_style(cp, &qname, NULL, 0);
1887 * "scale" the filename
1889 max = max_len;
1890 if (max > 50)
1891 max = 50;
1893 if (qname.len > max) {
1894 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1895 if (!cp)
1896 cp = qname.buf + qname.len + 3 - max;
1897 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1900 if (patch->is_binary) {
1901 printf(" %-*s | Bin\n", max, qname.buf);
1902 strbuf_release(&qname);
1903 return;
1906 printf(" %-*s |", max, qname.buf);
1907 strbuf_release(&qname);
1910 * scale the add/delete
1912 max = max + max_change > 70 ? 70 - max : max_change;
1913 add = patch->lines_added;
1914 del = patch->lines_deleted;
1916 if (max_change > 0) {
1917 int total = ((add + del) * max + max_change / 2) / max_change;
1918 add = (add * max + max_change / 2) / max_change;
1919 del = total - add;
1921 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1922 add, pluses, del, minuses);
1925 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1927 switch (st->st_mode & S_IFMT) {
1928 case S_IFLNK:
1929 if (strbuf_readlink(buf, path, st->st_size) < 0)
1930 return error("unable to read symlink %s", path);
1931 return 0;
1932 case S_IFREG:
1933 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1934 return error("unable to open or read %s", path);
1935 convert_to_git(path, buf->buf, buf->len, buf, 0);
1936 return 0;
1937 default:
1938 return -1;
1943 * Update the preimage, and the common lines in postimage,
1944 * from buffer buf of length len. If postlen is 0 the postimage
1945 * is updated in place, otherwise it's updated on a new buffer
1946 * of length postlen
1949 static void update_pre_post_images(struct image *preimage,
1950 struct image *postimage,
1951 char *buf,
1952 size_t len, size_t postlen)
1954 int i, ctx;
1955 char *new, *old, *fixed;
1956 struct image fixed_preimage;
1959 * Update the preimage with whitespace fixes. Note that we
1960 * are not losing preimage->buf -- apply_one_fragment() will
1961 * free "oldlines".
1963 prepare_image(&fixed_preimage, buf, len, 1);
1964 assert(fixed_preimage.nr == preimage->nr);
1965 for (i = 0; i < preimage->nr; i++)
1966 fixed_preimage.line[i].flag = preimage->line[i].flag;
1967 free(preimage->line_allocated);
1968 *preimage = fixed_preimage;
1971 * Adjust the common context lines in postimage. This can be
1972 * done in-place when we are just doing whitespace fixing,
1973 * which does not make the string grow, but needs a new buffer
1974 * when ignoring whitespace causes the update, since in this case
1975 * we could have e.g. tabs converted to multiple spaces.
1976 * We trust the caller to tell us if the update can be done
1977 * in place (postlen==0) or not.
1979 old = postimage->buf;
1980 if (postlen)
1981 new = postimage->buf = xmalloc(postlen);
1982 else
1983 new = old;
1984 fixed = preimage->buf;
1985 for (i = ctx = 0; i < postimage->nr; i++) {
1986 size_t len = postimage->line[i].len;
1987 if (!(postimage->line[i].flag & LINE_COMMON)) {
1988 /* an added line -- no counterparts in preimage */
1989 memmove(new, old, len);
1990 old += len;
1991 new += len;
1992 continue;
1995 /* a common context -- skip it in the original postimage */
1996 old += len;
1998 /* and find the corresponding one in the fixed preimage */
1999 while (ctx < preimage->nr &&
2000 !(preimage->line[ctx].flag & LINE_COMMON)) {
2001 fixed += preimage->line[ctx].len;
2002 ctx++;
2004 if (preimage->nr <= ctx)
2005 die("oops");
2007 /* and copy it in, while fixing the line length */
2008 len = preimage->line[ctx].len;
2009 memcpy(new, fixed, len);
2010 new += len;
2011 fixed += len;
2012 postimage->line[i].len = len;
2013 ctx++;
2016 /* Fix the length of the whole thing */
2017 postimage->len = new - postimage->buf;
2020 static int match_fragment(struct image *img,
2021 struct image *preimage,
2022 struct image *postimage,
2023 unsigned long try,
2024 int try_lno,
2025 unsigned ws_rule,
2026 int match_beginning, int match_end)
2028 int i;
2029 char *fixed_buf, *buf, *orig, *target;
2030 struct strbuf fixed;
2031 size_t fixed_len;
2032 int preimage_limit;
2034 if (preimage->nr + try_lno <= img->nr) {
2036 * The hunk falls within the boundaries of img.
2038 preimage_limit = preimage->nr;
2039 if (match_end && (preimage->nr + try_lno != img->nr))
2040 return 0;
2041 } else if (ws_error_action == correct_ws_error &&
2042 (ws_rule & WS_BLANK_AT_EOF)) {
2044 * This hunk extends beyond the end of img, and we are
2045 * removing blank lines at the end of the file. This
2046 * many lines from the beginning of the preimage must
2047 * match with img, and the remainder of the preimage
2048 * must be blank.
2050 preimage_limit = img->nr - try_lno;
2051 } else {
2053 * The hunk extends beyond the end of the img and
2054 * we are not removing blanks at the end, so we
2055 * should reject the hunk at this position.
2057 return 0;
2060 if (match_beginning && try_lno)
2061 return 0;
2063 /* Quick hash check */
2064 for (i = 0; i < preimage_limit; i++)
2065 if (preimage->line[i].hash != img->line[try_lno + i].hash)
2066 return 0;
2068 if (preimage_limit == preimage->nr) {
2070 * Do we have an exact match? If we were told to match
2071 * at the end, size must be exactly at try+fragsize,
2072 * otherwise try+fragsize must be still within the preimage,
2073 * and either case, the old piece should match the preimage
2074 * exactly.
2076 if ((match_end
2077 ? (try + preimage->len == img->len)
2078 : (try + preimage->len <= img->len)) &&
2079 !memcmp(img->buf + try, preimage->buf, preimage->len))
2080 return 1;
2081 } else {
2083 * The preimage extends beyond the end of img, so
2084 * there cannot be an exact match.
2086 * There must be one non-blank context line that match
2087 * a line before the end of img.
2089 char *buf_end;
2091 buf = preimage->buf;
2092 buf_end = buf;
2093 for (i = 0; i < preimage_limit; i++)
2094 buf_end += preimage->line[i].len;
2096 for ( ; buf < buf_end; buf++)
2097 if (!isspace(*buf))
2098 break;
2099 if (buf == buf_end)
2100 return 0;
2104 * No exact match. If we are ignoring whitespace, run a line-by-line
2105 * fuzzy matching. We collect all the line length information because
2106 * we need it to adjust whitespace if we match.
2108 if (ws_ignore_action == ignore_ws_change) {
2109 size_t imgoff = 0;
2110 size_t preoff = 0;
2111 size_t postlen = postimage->len;
2112 size_t extra_chars;
2113 char *preimage_eof;
2114 char *preimage_end;
2115 for (i = 0; i < preimage_limit; i++) {
2116 size_t prelen = preimage->line[i].len;
2117 size_t imglen = img->line[try_lno+i].len;
2119 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2120 preimage->buf + preoff, prelen))
2121 return 0;
2122 if (preimage->line[i].flag & LINE_COMMON)
2123 postlen += imglen - prelen;
2124 imgoff += imglen;
2125 preoff += prelen;
2129 * Ok, the preimage matches with whitespace fuzz.
2131 * imgoff now holds the true length of the target that
2132 * matches the preimage before the end of the file.
2134 * Count the number of characters in the preimage that fall
2135 * beyond the end of the file and make sure that all of them
2136 * are whitespace characters. (This can only happen if
2137 * we are removing blank lines at the end of the file.)
2139 buf = preimage_eof = preimage->buf + preoff;
2140 for ( ; i < preimage->nr; i++)
2141 preoff += preimage->line[i].len;
2142 preimage_end = preimage->buf + preoff;
2143 for ( ; buf < preimage_end; buf++)
2144 if (!isspace(*buf))
2145 return 0;
2148 * Update the preimage and the common postimage context
2149 * lines to use the same whitespace as the target.
2150 * If whitespace is missing in the target (i.e.
2151 * if the preimage extends beyond the end of the file),
2152 * use the whitespace from the preimage.
2154 extra_chars = preimage_end - preimage_eof;
2155 strbuf_init(&fixed, imgoff + extra_chars);
2156 strbuf_add(&fixed, img->buf + try, imgoff);
2157 strbuf_add(&fixed, preimage_eof, extra_chars);
2158 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2159 update_pre_post_images(preimage, postimage,
2160 fixed_buf, fixed_len, postlen);
2161 return 1;
2164 if (ws_error_action != correct_ws_error)
2165 return 0;
2168 * The hunk does not apply byte-by-byte, but the hash says
2169 * it might with whitespace fuzz. We haven't been asked to
2170 * ignore whitespace, we were asked to correct whitespace
2171 * errors, so let's try matching after whitespace correction.
2173 * The preimage may extend beyond the end of the file,
2174 * but in this loop we will only handle the part of the
2175 * preimage that falls within the file.
2177 strbuf_init(&fixed, preimage->len + 1);
2178 orig = preimage->buf;
2179 target = img->buf + try;
2180 for (i = 0; i < preimage_limit; i++) {
2181 size_t oldlen = preimage->line[i].len;
2182 size_t tgtlen = img->line[try_lno + i].len;
2183 size_t fixstart = fixed.len;
2184 struct strbuf tgtfix;
2185 int match;
2187 /* Try fixing the line in the preimage */
2188 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2190 /* Try fixing the line in the target */
2191 strbuf_init(&tgtfix, tgtlen);
2192 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2195 * If they match, either the preimage was based on
2196 * a version before our tree fixed whitespace breakage,
2197 * or we are lacking a whitespace-fix patch the tree
2198 * the preimage was based on already had (i.e. target
2199 * has whitespace breakage, the preimage doesn't).
2200 * In either case, we are fixing the whitespace breakages
2201 * so we might as well take the fix together with their
2202 * real change.
2204 match = (tgtfix.len == fixed.len - fixstart &&
2205 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2206 fixed.len - fixstart));
2208 strbuf_release(&tgtfix);
2209 if (!match)
2210 goto unmatch_exit;
2212 orig += oldlen;
2213 target += tgtlen;
2218 * Now handle the lines in the preimage that falls beyond the
2219 * end of the file (if any). They will only match if they are
2220 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2221 * false).
2223 for ( ; i < preimage->nr; i++) {
2224 size_t fixstart = fixed.len; /* start of the fixed preimage */
2225 size_t oldlen = preimage->line[i].len;
2226 int j;
2228 /* Try fixing the line in the preimage */
2229 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2231 for (j = fixstart; j < fixed.len; j++)
2232 if (!isspace(fixed.buf[j]))
2233 goto unmatch_exit;
2235 orig += oldlen;
2239 * Yes, the preimage is based on an older version that still
2240 * has whitespace breakages unfixed, and fixing them makes the
2241 * hunk match. Update the context lines in the postimage.
2243 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2244 update_pre_post_images(preimage, postimage,
2245 fixed_buf, fixed_len, 0);
2246 return 1;
2248 unmatch_exit:
2249 strbuf_release(&fixed);
2250 return 0;
2253 static int find_pos(struct image *img,
2254 struct image *preimage,
2255 struct image *postimage,
2256 int line,
2257 unsigned ws_rule,
2258 int match_beginning, int match_end)
2260 int i;
2261 unsigned long backwards, forwards, try;
2262 int backwards_lno, forwards_lno, try_lno;
2265 * If match_beginning or match_end is specified, there is no
2266 * point starting from a wrong line that will never match and
2267 * wander around and wait for a match at the specified end.
2269 if (match_beginning)
2270 line = 0;
2271 else if (match_end)
2272 line = img->nr - preimage->nr;
2275 * Because the comparison is unsigned, the following test
2276 * will also take care of a negative line number that can
2277 * result when match_end and preimage is larger than the target.
2279 if ((size_t) line > img->nr)
2280 line = img->nr;
2282 try = 0;
2283 for (i = 0; i < line; i++)
2284 try += img->line[i].len;
2287 * There's probably some smart way to do this, but I'll leave
2288 * that to the smart and beautiful people. I'm simple and stupid.
2290 backwards = try;
2291 backwards_lno = line;
2292 forwards = try;
2293 forwards_lno = line;
2294 try_lno = line;
2296 for (i = 0; ; i++) {
2297 if (match_fragment(img, preimage, postimage,
2298 try, try_lno, ws_rule,
2299 match_beginning, match_end))
2300 return try_lno;
2302 again:
2303 if (backwards_lno == 0 && forwards_lno == img->nr)
2304 break;
2306 if (i & 1) {
2307 if (backwards_lno == 0) {
2308 i++;
2309 goto again;
2311 backwards_lno--;
2312 backwards -= img->line[backwards_lno].len;
2313 try = backwards;
2314 try_lno = backwards_lno;
2315 } else {
2316 if (forwards_lno == img->nr) {
2317 i++;
2318 goto again;
2320 forwards += img->line[forwards_lno].len;
2321 forwards_lno++;
2322 try = forwards;
2323 try_lno = forwards_lno;
2327 return -1;
2330 static void remove_first_line(struct image *img)
2332 img->buf += img->line[0].len;
2333 img->len -= img->line[0].len;
2334 img->line++;
2335 img->nr--;
2338 static void remove_last_line(struct image *img)
2340 img->len -= img->line[--img->nr].len;
2343 static void update_image(struct image *img,
2344 int applied_pos,
2345 struct image *preimage,
2346 struct image *postimage)
2349 * remove the copy of preimage at offset in img
2350 * and replace it with postimage
2352 int i, nr;
2353 size_t remove_count, insert_count, applied_at = 0;
2354 char *result;
2355 int preimage_limit;
2358 * If we are removing blank lines at the end of img,
2359 * the preimage may extend beyond the end.
2360 * If that is the case, we must be careful only to
2361 * remove the part of the preimage that falls within
2362 * the boundaries of img. Initialize preimage_limit
2363 * to the number of lines in the preimage that falls
2364 * within the boundaries.
2366 preimage_limit = preimage->nr;
2367 if (preimage_limit > img->nr - applied_pos)
2368 preimage_limit = img->nr - applied_pos;
2370 for (i = 0; i < applied_pos; i++)
2371 applied_at += img->line[i].len;
2373 remove_count = 0;
2374 for (i = 0; i < preimage_limit; i++)
2375 remove_count += img->line[applied_pos + i].len;
2376 insert_count = postimage->len;
2378 /* Adjust the contents */
2379 result = xmalloc(img->len + insert_count - remove_count + 1);
2380 memcpy(result, img->buf, applied_at);
2381 memcpy(result + applied_at, postimage->buf, postimage->len);
2382 memcpy(result + applied_at + postimage->len,
2383 img->buf + (applied_at + remove_count),
2384 img->len - (applied_at + remove_count));
2385 free(img->buf);
2386 img->buf = result;
2387 img->len += insert_count - remove_count;
2388 result[img->len] = '\0';
2390 /* Adjust the line table */
2391 nr = img->nr + postimage->nr - preimage_limit;
2392 if (preimage_limit < postimage->nr) {
2394 * NOTE: this knows that we never call remove_first_line()
2395 * on anything other than pre/post image.
2397 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2398 img->line_allocated = img->line;
2400 if (preimage_limit != postimage->nr)
2401 memmove(img->line + applied_pos + postimage->nr,
2402 img->line + applied_pos + preimage_limit,
2403 (img->nr - (applied_pos + preimage_limit)) *
2404 sizeof(*img->line));
2405 memcpy(img->line + applied_pos,
2406 postimage->line,
2407 postimage->nr * sizeof(*img->line));
2408 img->nr = nr;
2411 static int apply_one_fragment(struct image *img, struct fragment *frag,
2412 int inaccurate_eof, unsigned ws_rule)
2414 int match_beginning, match_end;
2415 const char *patch = frag->patch;
2416 int size = frag->size;
2417 char *old, *oldlines;
2418 struct strbuf newlines;
2419 int new_blank_lines_at_end = 0;
2420 unsigned long leading, trailing;
2421 int pos, applied_pos;
2422 struct image preimage;
2423 struct image postimage;
2425 memset(&preimage, 0, sizeof(preimage));
2426 memset(&postimage, 0, sizeof(postimage));
2427 oldlines = xmalloc(size);
2428 strbuf_init(&newlines, size);
2430 old = oldlines;
2431 while (size > 0) {
2432 char first;
2433 int len = linelen(patch, size);
2434 int plen;
2435 int added_blank_line = 0;
2436 int is_blank_context = 0;
2437 size_t start;
2439 if (!len)
2440 break;
2443 * "plen" is how much of the line we should use for
2444 * the actual patch data. Normally we just remove the
2445 * first character on the line, but if the line is
2446 * followed by "\ No newline", then we also remove the
2447 * last one (which is the newline, of course).
2449 plen = len - 1;
2450 if (len < size && patch[len] == '\\')
2451 plen--;
2452 first = *patch;
2453 if (apply_in_reverse) {
2454 if (first == '-')
2455 first = '+';
2456 else if (first == '+')
2457 first = '-';
2460 switch (first) {
2461 case '\n':
2462 /* Newer GNU diff, empty context line */
2463 if (plen < 0)
2464 /* ... followed by '\No newline'; nothing */
2465 break;
2466 *old++ = '\n';
2467 strbuf_addch(&newlines, '\n');
2468 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2469 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2470 is_blank_context = 1;
2471 break;
2472 case ' ':
2473 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2474 ws_blank_line(patch + 1, plen, ws_rule))
2475 is_blank_context = 1;
2476 case '-':
2477 memcpy(old, patch + 1, plen);
2478 add_line_info(&preimage, old, plen,
2479 (first == ' ' ? LINE_COMMON : 0));
2480 old += plen;
2481 if (first == '-')
2482 break;
2483 /* Fall-through for ' ' */
2484 case '+':
2485 /* --no-add does not add new lines */
2486 if (first == '+' && no_add)
2487 break;
2489 start = newlines.len;
2490 if (first != '+' ||
2491 !whitespace_error ||
2492 ws_error_action != correct_ws_error) {
2493 strbuf_add(&newlines, patch + 1, plen);
2495 else {
2496 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2498 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2499 (first == '+' ? 0 : LINE_COMMON));
2500 if (first == '+' &&
2501 (ws_rule & WS_BLANK_AT_EOF) &&
2502 ws_blank_line(patch + 1, plen, ws_rule))
2503 added_blank_line = 1;
2504 break;
2505 case '@': case '\\':
2506 /* Ignore it, we already handled it */
2507 break;
2508 default:
2509 if (apply_verbosely)
2510 error("invalid start of line: '%c'", first);
2511 return -1;
2513 if (added_blank_line)
2514 new_blank_lines_at_end++;
2515 else if (is_blank_context)
2517 else
2518 new_blank_lines_at_end = 0;
2519 patch += len;
2520 size -= len;
2522 if (inaccurate_eof &&
2523 old > oldlines && old[-1] == '\n' &&
2524 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2525 old--;
2526 strbuf_setlen(&newlines, newlines.len - 1);
2529 leading = frag->leading;
2530 trailing = frag->trailing;
2533 * A hunk to change lines at the beginning would begin with
2534 * @@ -1,L +N,M @@
2535 * but we need to be careful. -U0 that inserts before the second
2536 * line also has this pattern.
2538 * And a hunk to add to an empty file would begin with
2539 * @@ -0,0 +N,M @@
2541 * In other words, a hunk that is (frag->oldpos <= 1) with or
2542 * without leading context must match at the beginning.
2544 match_beginning = (!frag->oldpos ||
2545 (frag->oldpos == 1 && !unidiff_zero));
2548 * A hunk without trailing lines must match at the end.
2549 * However, we simply cannot tell if a hunk must match end
2550 * from the lack of trailing lines if the patch was generated
2551 * with unidiff without any context.
2553 match_end = !unidiff_zero && !trailing;
2555 pos = frag->newpos ? (frag->newpos - 1) : 0;
2556 preimage.buf = oldlines;
2557 preimage.len = old - oldlines;
2558 postimage.buf = newlines.buf;
2559 postimage.len = newlines.len;
2560 preimage.line = preimage.line_allocated;
2561 postimage.line = postimage.line_allocated;
2563 for (;;) {
2565 applied_pos = find_pos(img, &preimage, &postimage, pos,
2566 ws_rule, match_beginning, match_end);
2568 if (applied_pos >= 0)
2569 break;
2571 /* Am I at my context limits? */
2572 if ((leading <= p_context) && (trailing <= p_context))
2573 break;
2574 if (match_beginning || match_end) {
2575 match_beginning = match_end = 0;
2576 continue;
2580 * Reduce the number of context lines; reduce both
2581 * leading and trailing if they are equal otherwise
2582 * just reduce the larger context.
2584 if (leading >= trailing) {
2585 remove_first_line(&preimage);
2586 remove_first_line(&postimage);
2587 pos--;
2588 leading--;
2590 if (trailing > leading) {
2591 remove_last_line(&preimage);
2592 remove_last_line(&postimage);
2593 trailing--;
2597 if (applied_pos >= 0) {
2598 if (new_blank_lines_at_end &&
2599 preimage.nr + applied_pos >= img->nr &&
2600 (ws_rule & WS_BLANK_AT_EOF) &&
2601 ws_error_action != nowarn_ws_error) {
2602 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2603 if (ws_error_action == correct_ws_error) {
2604 while (new_blank_lines_at_end--)
2605 remove_last_line(&postimage);
2608 * We would want to prevent write_out_results()
2609 * from taking place in apply_patch() that follows
2610 * the callchain led us here, which is:
2611 * apply_patch->check_patch_list->check_patch->
2612 * apply_data->apply_fragments->apply_one_fragment
2614 if (ws_error_action == die_on_ws_error)
2615 apply = 0;
2619 * Warn if it was necessary to reduce the number
2620 * of context lines.
2622 if ((leading != frag->leading) ||
2623 (trailing != frag->trailing))
2624 fprintf(stderr, "Context reduced to (%ld/%ld)"
2625 " to apply fragment at %d\n",
2626 leading, trailing, applied_pos+1);
2627 update_image(img, applied_pos, &preimage, &postimage);
2628 } else {
2629 if (apply_verbosely)
2630 error("while searching for:\n%.*s",
2631 (int)(old - oldlines), oldlines);
2634 free(oldlines);
2635 strbuf_release(&newlines);
2636 free(preimage.line_allocated);
2637 free(postimage.line_allocated);
2639 return (applied_pos < 0);
2642 static int apply_binary_fragment(struct image *img, struct patch *patch)
2644 struct fragment *fragment = patch->fragments;
2645 unsigned long len;
2646 void *dst;
2648 /* Binary patch is irreversible without the optional second hunk */
2649 if (apply_in_reverse) {
2650 if (!fragment->next)
2651 return error("cannot reverse-apply a binary patch "
2652 "without the reverse hunk to '%s'",
2653 patch->new_name
2654 ? patch->new_name : patch->old_name);
2655 fragment = fragment->next;
2657 switch (fragment->binary_patch_method) {
2658 case BINARY_DELTA_DEFLATED:
2659 dst = patch_delta(img->buf, img->len, fragment->patch,
2660 fragment->size, &len);
2661 if (!dst)
2662 return -1;
2663 clear_image(img);
2664 img->buf = dst;
2665 img->len = len;
2666 return 0;
2667 case BINARY_LITERAL_DEFLATED:
2668 clear_image(img);
2669 img->len = fragment->size;
2670 img->buf = xmalloc(img->len+1);
2671 memcpy(img->buf, fragment->patch, img->len);
2672 img->buf[img->len] = '\0';
2673 return 0;
2675 return -1;
2678 static int apply_binary(struct image *img, struct patch *patch)
2680 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2681 unsigned char sha1[20];
2684 * For safety, we require patch index line to contain
2685 * full 40-byte textual SHA1 for old and new, at least for now.
2687 if (strlen(patch->old_sha1_prefix) != 40 ||
2688 strlen(patch->new_sha1_prefix) != 40 ||
2689 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2690 get_sha1_hex(patch->new_sha1_prefix, sha1))
2691 return error("cannot apply binary patch to '%s' "
2692 "without full index line", name);
2694 if (patch->old_name) {
2696 * See if the old one matches what the patch
2697 * applies to.
2699 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2700 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2701 return error("the patch applies to '%s' (%s), "
2702 "which does not match the "
2703 "current contents.",
2704 name, sha1_to_hex(sha1));
2706 else {
2707 /* Otherwise, the old one must be empty. */
2708 if (img->len)
2709 return error("the patch applies to an empty "
2710 "'%s' but it is not empty", name);
2713 get_sha1_hex(patch->new_sha1_prefix, sha1);
2714 if (is_null_sha1(sha1)) {
2715 clear_image(img);
2716 return 0; /* deletion patch */
2719 if (has_sha1_file(sha1)) {
2720 /* We already have the postimage */
2721 enum object_type type;
2722 unsigned long size;
2723 char *result;
2725 result = read_sha1_file(sha1, &type, &size);
2726 if (!result)
2727 return error("the necessary postimage %s for "
2728 "'%s' cannot be read",
2729 patch->new_sha1_prefix, name);
2730 clear_image(img);
2731 img->buf = result;
2732 img->len = size;
2733 } else {
2735 * We have verified buf matches the preimage;
2736 * apply the patch data to it, which is stored
2737 * in the patch->fragments->{patch,size}.
2739 if (apply_binary_fragment(img, patch))
2740 return error("binary patch does not apply to '%s'",
2741 name);
2743 /* verify that the result matches */
2744 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2745 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2746 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2747 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2750 return 0;
2753 static int apply_fragments(struct image *img, struct patch *patch)
2755 struct fragment *frag = patch->fragments;
2756 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2757 unsigned ws_rule = patch->ws_rule;
2758 unsigned inaccurate_eof = patch->inaccurate_eof;
2760 if (patch->is_binary)
2761 return apply_binary(img, patch);
2763 while (frag) {
2764 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2765 error("patch failed: %s:%ld", name, frag->oldpos);
2766 if (!apply_with_reject)
2767 return -1;
2768 frag->rejected = 1;
2770 frag = frag->next;
2772 return 0;
2775 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2777 if (!ce)
2778 return 0;
2780 if (S_ISGITLINK(ce->ce_mode)) {
2781 strbuf_grow(buf, 100);
2782 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2783 } else {
2784 enum object_type type;
2785 unsigned long sz;
2786 char *result;
2788 result = read_sha1_file(ce->sha1, &type, &sz);
2789 if (!result)
2790 return -1;
2791 /* XXX read_sha1_file NUL-terminates */
2792 strbuf_attach(buf, result, sz, sz + 1);
2794 return 0;
2797 static struct patch *in_fn_table(const char *name)
2799 struct string_list_item *item;
2801 if (name == NULL)
2802 return NULL;
2804 item = string_list_lookup(&fn_table, name);
2805 if (item != NULL)
2806 return (struct patch *)item->util;
2808 return NULL;
2812 * item->util in the filename table records the status of the path.
2813 * Usually it points at a patch (whose result records the contents
2814 * of it after applying it), but it could be PATH_WAS_DELETED for a
2815 * path that a previously applied patch has already removed.
2817 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2818 #define PATH_WAS_DELETED ((struct patch *) -1)
2820 static int to_be_deleted(struct patch *patch)
2822 return patch == PATH_TO_BE_DELETED;
2825 static int was_deleted(struct patch *patch)
2827 return patch == PATH_WAS_DELETED;
2830 static void add_to_fn_table(struct patch *patch)
2832 struct string_list_item *item;
2835 * Always add new_name unless patch is a deletion
2836 * This should cover the cases for normal diffs,
2837 * file creations and copies
2839 if (patch->new_name != NULL) {
2840 item = string_list_insert(&fn_table, patch->new_name);
2841 item->util = patch;
2845 * store a failure on rename/deletion cases because
2846 * later chunks shouldn't patch old names
2848 if ((patch->new_name == NULL) || (patch->is_rename)) {
2849 item = string_list_insert(&fn_table, patch->old_name);
2850 item->util = PATH_WAS_DELETED;
2854 static void prepare_fn_table(struct patch *patch)
2857 * store information about incoming file deletion
2859 while (patch) {
2860 if ((patch->new_name == NULL) || (patch->is_rename)) {
2861 struct string_list_item *item;
2862 item = string_list_insert(&fn_table, patch->old_name);
2863 item->util = PATH_TO_BE_DELETED;
2865 patch = patch->next;
2869 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2871 struct strbuf buf = STRBUF_INIT;
2872 struct image image;
2873 size_t len;
2874 char *img;
2875 struct patch *tpatch;
2877 if (!(patch->is_copy || patch->is_rename) &&
2878 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2879 if (was_deleted(tpatch)) {
2880 return error("patch %s has been renamed/deleted",
2881 patch->old_name);
2883 /* We have a patched copy in memory use that */
2884 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2885 } else if (cached) {
2886 if (read_file_or_gitlink(ce, &buf))
2887 return error("read of %s failed", patch->old_name);
2888 } else if (patch->old_name) {
2889 if (S_ISGITLINK(patch->old_mode)) {
2890 if (ce) {
2891 read_file_or_gitlink(ce, &buf);
2892 } else {
2894 * There is no way to apply subproject
2895 * patch without looking at the index.
2897 patch->fragments = NULL;
2899 } else {
2900 if (read_old_data(st, patch->old_name, &buf))
2901 return error("read of %s failed", patch->old_name);
2905 img = strbuf_detach(&buf, &len);
2906 prepare_image(&image, img, len, !patch->is_binary);
2908 if (apply_fragments(&image, patch) < 0)
2909 return -1; /* note with --reject this succeeds. */
2910 patch->result = image.buf;
2911 patch->resultsize = image.len;
2912 add_to_fn_table(patch);
2913 free(image.line_allocated);
2915 if (0 < patch->is_delete && patch->resultsize)
2916 return error("removal patch leaves file contents");
2918 return 0;
2921 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2923 struct stat nst;
2924 if (!lstat(new_name, &nst)) {
2925 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2926 return 0;
2928 * A leading component of new_name might be a symlink
2929 * that is going to be removed with this patch, but
2930 * still pointing at somewhere that has the path.
2931 * In such a case, path "new_name" does not exist as
2932 * far as git is concerned.
2934 if (has_symlink_leading_path(new_name, strlen(new_name)))
2935 return 0;
2937 return error("%s: already exists in working directory", new_name);
2939 else if ((errno != ENOENT) && (errno != ENOTDIR))
2940 return error("%s: %s", new_name, strerror(errno));
2941 return 0;
2944 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2946 if (S_ISGITLINK(ce->ce_mode)) {
2947 if (!S_ISDIR(st->st_mode))
2948 return -1;
2949 return 0;
2951 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
2954 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2956 const char *old_name = patch->old_name;
2957 struct patch *tpatch = NULL;
2958 int stat_ret = 0;
2959 unsigned st_mode = 0;
2962 * Make sure that we do not have local modifications from the
2963 * index when we are looking at the index. Also make sure
2964 * we have the preimage file to be patched in the work tree,
2965 * unless --cached, which tells git to apply only in the index.
2967 if (!old_name)
2968 return 0;
2970 assert(patch->is_new <= 0);
2972 if (!(patch->is_copy || patch->is_rename) &&
2973 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2974 if (was_deleted(tpatch))
2975 return error("%s: has been deleted/renamed", old_name);
2976 st_mode = tpatch->new_mode;
2977 } else if (!cached) {
2978 stat_ret = lstat(old_name, st);
2979 if (stat_ret && errno != ENOENT)
2980 return error("%s: %s", old_name, strerror(errno));
2983 if (to_be_deleted(tpatch))
2984 tpatch = NULL;
2986 if (check_index && !tpatch) {
2987 int pos = cache_name_pos(old_name, strlen(old_name));
2988 if (pos < 0) {
2989 if (patch->is_new < 0)
2990 goto is_new;
2991 return error("%s: does not exist in index", old_name);
2993 *ce = active_cache[pos];
2994 if (stat_ret < 0) {
2995 struct checkout costate;
2996 /* checkout */
2997 memset(&costate, 0, sizeof(costate));
2998 costate.base_dir = "";
2999 costate.refresh_cache = 1;
3000 if (checkout_entry(*ce, &costate, NULL) ||
3001 lstat(old_name, st))
3002 return -1;
3004 if (!cached && verify_index_match(*ce, st))
3005 return error("%s: does not match index", old_name);
3006 if (cached)
3007 st_mode = (*ce)->ce_mode;
3008 } else if (stat_ret < 0) {
3009 if (patch->is_new < 0)
3010 goto is_new;
3011 return error("%s: %s", old_name, strerror(errno));
3014 if (!cached && !tpatch)
3015 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3017 if (patch->is_new < 0)
3018 patch->is_new = 0;
3019 if (!patch->old_mode)
3020 patch->old_mode = st_mode;
3021 if ((st_mode ^ patch->old_mode) & S_IFMT)
3022 return error("%s: wrong type", old_name);
3023 if (st_mode != patch->old_mode)
3024 warning("%s has type %o, expected %o",
3025 old_name, st_mode, patch->old_mode);
3026 if (!patch->new_mode && !patch->is_delete)
3027 patch->new_mode = st_mode;
3028 return 0;
3030 is_new:
3031 patch->is_new = 1;
3032 patch->is_delete = 0;
3033 patch->old_name = NULL;
3034 return 0;
3037 static int check_patch(struct patch *patch)
3039 struct stat st;
3040 const char *old_name = patch->old_name;
3041 const char *new_name = patch->new_name;
3042 const char *name = old_name ? old_name : new_name;
3043 struct cache_entry *ce = NULL;
3044 struct patch *tpatch;
3045 int ok_if_exists;
3046 int status;
3048 patch->rejected = 1; /* we will drop this after we succeed */
3050 status = check_preimage(patch, &ce, &st);
3051 if (status)
3052 return status;
3053 old_name = patch->old_name;
3055 if ((tpatch = in_fn_table(new_name)) &&
3056 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3058 * A type-change diff is always split into a patch to
3059 * delete old, immediately followed by a patch to
3060 * create new (see diff.c::run_diff()); in such a case
3061 * it is Ok that the entry to be deleted by the
3062 * previous patch is still in the working tree and in
3063 * the index.
3065 ok_if_exists = 1;
3066 else
3067 ok_if_exists = 0;
3069 if (new_name &&
3070 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3071 if (check_index &&
3072 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3073 !ok_if_exists)
3074 return error("%s: already exists in index", new_name);
3075 if (!cached) {
3076 int err = check_to_create_blob(new_name, ok_if_exists);
3077 if (err)
3078 return err;
3080 if (!patch->new_mode) {
3081 if (0 < patch->is_new)
3082 patch->new_mode = S_IFREG | 0644;
3083 else
3084 patch->new_mode = patch->old_mode;
3088 if (new_name && old_name) {
3089 int same = !strcmp(old_name, new_name);
3090 if (!patch->new_mode)
3091 patch->new_mode = patch->old_mode;
3092 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3093 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
3094 patch->new_mode, new_name, patch->old_mode,
3095 same ? "" : " of ", same ? "" : old_name);
3098 if (apply_data(patch, &st, ce) < 0)
3099 return error("%s: patch does not apply", name);
3100 patch->rejected = 0;
3101 return 0;
3104 static int check_patch_list(struct patch *patch)
3106 int err = 0;
3108 prepare_fn_table(patch);
3109 while (patch) {
3110 if (apply_verbosely)
3111 say_patch_name(stderr,
3112 "Checking patch ", patch, "...\n");
3113 err |= check_patch(patch);
3114 patch = patch->next;
3116 return err;
3119 /* This function tries to read the sha1 from the current index */
3120 static int get_current_sha1(const char *path, unsigned char *sha1)
3122 int pos;
3124 if (read_cache() < 0)
3125 return -1;
3126 pos = cache_name_pos(path, strlen(path));
3127 if (pos < 0)
3128 return -1;
3129 hashcpy(sha1, active_cache[pos]->sha1);
3130 return 0;
3133 /* Build an index that contains the just the files needed for a 3way merge */
3134 static void build_fake_ancestor(struct patch *list, const char *filename)
3136 struct patch *patch;
3137 struct index_state result = { NULL };
3138 int fd;
3140 /* Once we start supporting the reverse patch, it may be
3141 * worth showing the new sha1 prefix, but until then...
3143 for (patch = list; patch; patch = patch->next) {
3144 const unsigned char *sha1_ptr;
3145 unsigned char sha1[20];
3146 struct cache_entry *ce;
3147 const char *name;
3149 name = patch->old_name ? patch->old_name : patch->new_name;
3150 if (0 < patch->is_new)
3151 continue;
3152 else if (get_sha1(patch->old_sha1_prefix, sha1))
3153 /* git diff has no index line for mode/type changes */
3154 if (!patch->lines_added && !patch->lines_deleted) {
3155 if (get_current_sha1(patch->old_name, sha1))
3156 die("mode change for %s, which is not "
3157 "in current HEAD", name);
3158 sha1_ptr = sha1;
3159 } else
3160 die("sha1 information is lacking or useless "
3161 "(%s).", name);
3162 else
3163 sha1_ptr = sha1;
3165 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3166 if (!ce)
3167 die("make_cache_entry failed for path '%s'", name);
3168 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3169 die ("Could not add %s to temporary index", name);
3172 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3173 if (fd < 0 || write_index(&result, fd) || close(fd))
3174 die ("Could not write temporary index to %s", filename);
3176 discard_index(&result);
3179 static void stat_patch_list(struct patch *patch)
3181 int files, adds, dels;
3183 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3184 files++;
3185 adds += patch->lines_added;
3186 dels += patch->lines_deleted;
3187 show_stats(patch);
3190 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3193 static void numstat_patch_list(struct patch *patch)
3195 for ( ; patch; patch = patch->next) {
3196 const char *name;
3197 name = patch->new_name ? patch->new_name : patch->old_name;
3198 if (patch->is_binary)
3199 printf("-\t-\t");
3200 else
3201 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3202 write_name_quoted(name, stdout, line_termination);
3206 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3208 if (mode)
3209 printf(" %s mode %06o %s\n", newdelete, mode, name);
3210 else
3211 printf(" %s %s\n", newdelete, name);
3214 static void show_mode_change(struct patch *p, int show_name)
3216 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3217 if (show_name)
3218 printf(" mode change %06o => %06o %s\n",
3219 p->old_mode, p->new_mode, p->new_name);
3220 else
3221 printf(" mode change %06o => %06o\n",
3222 p->old_mode, p->new_mode);
3226 static void show_rename_copy(struct patch *p)
3228 const char *renamecopy = p->is_rename ? "rename" : "copy";
3229 const char *old, *new;
3231 /* Find common prefix */
3232 old = p->old_name;
3233 new = p->new_name;
3234 while (1) {
3235 const char *slash_old, *slash_new;
3236 slash_old = strchr(old, '/');
3237 slash_new = strchr(new, '/');
3238 if (!slash_old ||
3239 !slash_new ||
3240 slash_old - old != slash_new - new ||
3241 memcmp(old, new, slash_new - new))
3242 break;
3243 old = slash_old + 1;
3244 new = slash_new + 1;
3246 /* p->old_name thru old is the common prefix, and old and new
3247 * through the end of names are renames
3249 if (old != p->old_name)
3250 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3251 (int)(old - p->old_name), p->old_name,
3252 old, new, p->score);
3253 else
3254 printf(" %s %s => %s (%d%%)\n", renamecopy,
3255 p->old_name, p->new_name, p->score);
3256 show_mode_change(p, 0);
3259 static void summary_patch_list(struct patch *patch)
3261 struct patch *p;
3263 for (p = patch; p; p = p->next) {
3264 if (p->is_new)
3265 show_file_mode_name("create", p->new_mode, p->new_name);
3266 else if (p->is_delete)
3267 show_file_mode_name("delete", p->old_mode, p->old_name);
3268 else {
3269 if (p->is_rename || p->is_copy)
3270 show_rename_copy(p);
3271 else {
3272 if (p->score) {
3273 printf(" rewrite %s (%d%%)\n",
3274 p->new_name, p->score);
3275 show_mode_change(p, 0);
3277 else
3278 show_mode_change(p, 1);
3284 static void patch_stats(struct patch *patch)
3286 int lines = patch->lines_added + patch->lines_deleted;
3288 if (lines > max_change)
3289 max_change = lines;
3290 if (patch->old_name) {
3291 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3292 if (!len)
3293 len = strlen(patch->old_name);
3294 if (len > max_len)
3295 max_len = len;
3297 if (patch->new_name) {
3298 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3299 if (!len)
3300 len = strlen(patch->new_name);
3301 if (len > max_len)
3302 max_len = len;
3306 static void remove_file(struct patch *patch, int rmdir_empty)
3308 if (update_index) {
3309 if (remove_file_from_cache(patch->old_name) < 0)
3310 die("unable to remove %s from index", patch->old_name);
3312 if (!cached) {
3313 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3314 remove_path(patch->old_name);
3319 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3321 struct stat st;
3322 struct cache_entry *ce;
3323 int namelen = strlen(path);
3324 unsigned ce_size = cache_entry_size(namelen);
3326 if (!update_index)
3327 return;
3329 ce = xcalloc(1, ce_size);
3330 memcpy(ce->name, path, namelen);
3331 ce->ce_mode = create_ce_mode(mode);
3332 ce->ce_flags = namelen;
3333 if (S_ISGITLINK(mode)) {
3334 const char *s = buf;
3336 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3337 die("corrupt patch for subproject %s", path);
3338 } else {
3339 if (!cached) {
3340 if (lstat(path, &st) < 0)
3341 die_errno("unable to stat newly created file '%s'",
3342 path);
3343 fill_stat_cache_info(ce, &st);
3345 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3346 die("unable to create backing store for newly created file %s", path);
3348 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3349 die("unable to add cache entry for %s", path);
3352 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3354 int fd;
3355 struct strbuf nbuf = STRBUF_INIT;
3357 if (S_ISGITLINK(mode)) {
3358 struct stat st;
3359 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3360 return 0;
3361 return mkdir(path, 0777);
3364 if (has_symlinks && S_ISLNK(mode))
3365 /* Although buf:size is counted string, it also is NUL
3366 * terminated.
3368 return symlink(buf, path);
3370 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3371 if (fd < 0)
3372 return -1;
3374 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3375 size = nbuf.len;
3376 buf = nbuf.buf;
3378 write_or_die(fd, buf, size);
3379 strbuf_release(&nbuf);
3381 if (close(fd) < 0)
3382 die_errno("closing file '%s'", path);
3383 return 0;
3387 * We optimistically assume that the directories exist,
3388 * which is true 99% of the time anyway. If they don't,
3389 * we create them and try again.
3391 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3393 if (cached)
3394 return;
3395 if (!try_create_file(path, mode, buf, size))
3396 return;
3398 if (errno == ENOENT) {
3399 if (safe_create_leading_directories(path))
3400 return;
3401 if (!try_create_file(path, mode, buf, size))
3402 return;
3405 if (errno == EEXIST || errno == EACCES) {
3406 /* We may be trying to create a file where a directory
3407 * used to be.
3409 struct stat st;
3410 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3411 errno = EEXIST;
3414 if (errno == EEXIST) {
3415 unsigned int nr = getpid();
3417 for (;;) {
3418 char newpath[PATH_MAX];
3419 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3420 if (!try_create_file(newpath, mode, buf, size)) {
3421 if (!rename(newpath, path))
3422 return;
3423 unlink_or_warn(newpath);
3424 break;
3426 if (errno != EEXIST)
3427 break;
3428 ++nr;
3431 die_errno("unable to write file '%s' mode %o", path, mode);
3434 static void create_file(struct patch *patch)
3436 char *path = patch->new_name;
3437 unsigned mode = patch->new_mode;
3438 unsigned long size = patch->resultsize;
3439 char *buf = patch->result;
3441 if (!mode)
3442 mode = S_IFREG | 0644;
3443 create_one_file(path, mode, buf, size);
3444 add_index_file(path, mode, buf, size);
3447 /* phase zero is to remove, phase one is to create */
3448 static void write_out_one_result(struct patch *patch, int phase)
3450 if (patch->is_delete > 0) {
3451 if (phase == 0)
3452 remove_file(patch, 1);
3453 return;
3455 if (patch->is_new > 0 || patch->is_copy) {
3456 if (phase == 1)
3457 create_file(patch);
3458 return;
3461 * Rename or modification boils down to the same
3462 * thing: remove the old, write the new
3464 if (phase == 0)
3465 remove_file(patch, patch->is_rename);
3466 if (phase == 1)
3467 create_file(patch);
3470 static int write_out_one_reject(struct patch *patch)
3472 FILE *rej;
3473 char namebuf[PATH_MAX];
3474 struct fragment *frag;
3475 int cnt = 0;
3477 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3478 if (!frag->rejected)
3479 continue;
3480 cnt++;
3483 if (!cnt) {
3484 if (apply_verbosely)
3485 say_patch_name(stderr,
3486 "Applied patch ", patch, " cleanly.\n");
3487 return 0;
3490 /* This should not happen, because a removal patch that leaves
3491 * contents are marked "rejected" at the patch level.
3493 if (!patch->new_name)
3494 die("internal error");
3496 /* Say this even without --verbose */
3497 say_patch_name(stderr, "Applying patch ", patch, " with");
3498 fprintf(stderr, " %d rejects...\n", cnt);
3500 cnt = strlen(patch->new_name);
3501 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3502 cnt = ARRAY_SIZE(namebuf) - 5;
3503 warning("truncating .rej filename to %.*s.rej",
3504 cnt - 1, patch->new_name);
3506 memcpy(namebuf, patch->new_name, cnt);
3507 memcpy(namebuf + cnt, ".rej", 5);
3509 rej = fopen(namebuf, "w");
3510 if (!rej)
3511 return error("cannot open %s: %s", namebuf, strerror(errno));
3513 /* Normal git tools never deal with .rej, so do not pretend
3514 * this is a git patch by saying --git nor give extended
3515 * headers. While at it, maybe please "kompare" that wants
3516 * the trailing TAB and some garbage at the end of line ;-).
3518 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3519 patch->new_name, patch->new_name);
3520 for (cnt = 1, frag = patch->fragments;
3521 frag;
3522 cnt++, frag = frag->next) {
3523 if (!frag->rejected) {
3524 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3525 continue;
3527 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3528 fprintf(rej, "%.*s", frag->size, frag->patch);
3529 if (frag->patch[frag->size-1] != '\n')
3530 fputc('\n', rej);
3532 fclose(rej);
3533 return -1;
3536 static int write_out_results(struct patch *list, int skipped_patch)
3538 int phase;
3539 int errs = 0;
3540 struct patch *l;
3542 if (!list && !skipped_patch)
3543 return error("No changes");
3545 for (phase = 0; phase < 2; phase++) {
3546 l = list;
3547 while (l) {
3548 if (l->rejected)
3549 errs = 1;
3550 else {
3551 write_out_one_result(l, phase);
3552 if (phase == 1 && write_out_one_reject(l))
3553 errs = 1;
3555 l = l->next;
3558 return errs;
3561 static struct lock_file lock_file;
3563 static struct string_list limit_by_name;
3564 static int has_include;
3565 static void add_name_limit(const char *name, int exclude)
3567 struct string_list_item *it;
3569 it = string_list_append(&limit_by_name, name);
3570 it->util = exclude ? NULL : (void *) 1;
3573 static int use_patch(struct patch *p)
3575 const char *pathname = p->new_name ? p->new_name : p->old_name;
3576 int i;
3578 /* Paths outside are not touched regardless of "--include" */
3579 if (0 < prefix_length) {
3580 int pathlen = strlen(pathname);
3581 if (pathlen <= prefix_length ||
3582 memcmp(prefix, pathname, prefix_length))
3583 return 0;
3586 /* See if it matches any of exclude/include rule */
3587 for (i = 0; i < limit_by_name.nr; i++) {
3588 struct string_list_item *it = &limit_by_name.items[i];
3589 if (!fnmatch(it->string, pathname, 0))
3590 return (it->util != NULL);
3594 * If we had any include, a path that does not match any rule is
3595 * not used. Otherwise, we saw bunch of exclude rules (or none)
3596 * and such a path is used.
3598 return !has_include;
3602 static void prefix_one(char **name)
3604 char *old_name = *name;
3605 if (!old_name)
3606 return;
3607 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3608 free(old_name);
3611 static void prefix_patches(struct patch *p)
3613 if (!prefix || p->is_toplevel_relative)
3614 return;
3615 for ( ; p; p = p->next) {
3616 if (p->new_name == p->old_name) {
3617 char *prefixed = p->new_name;
3618 prefix_one(&prefixed);
3619 p->new_name = p->old_name = prefixed;
3621 else {
3622 prefix_one(&p->new_name);
3623 prefix_one(&p->old_name);
3628 #define INACCURATE_EOF (1<<0)
3629 #define RECOUNT (1<<1)
3631 static int apply_patch(int fd, const char *filename, int options)
3633 size_t offset;
3634 struct strbuf buf = STRBUF_INIT;
3635 struct patch *list = NULL, **listp = &list;
3636 int skipped_patch = 0;
3638 /* FIXME - memory leak when using multiple patch files as inputs */
3639 memset(&fn_table, 0, sizeof(struct string_list));
3640 patch_input_file = filename;
3641 read_patch_file(&buf, fd);
3642 offset = 0;
3643 while (offset < buf.len) {
3644 struct patch *patch;
3645 int nr;
3647 patch = xcalloc(1, sizeof(*patch));
3648 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3649 patch->recount = !!(options & RECOUNT);
3650 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3651 if (nr < 0)
3652 break;
3653 if (apply_in_reverse)
3654 reverse_patches(patch);
3655 if (prefix)
3656 prefix_patches(patch);
3657 if (use_patch(patch)) {
3658 patch_stats(patch);
3659 *listp = patch;
3660 listp = &patch->next;
3662 else {
3663 /* perhaps free it a bit better? */
3664 free(patch);
3665 skipped_patch++;
3667 offset += nr;
3670 if (whitespace_error && (ws_error_action == die_on_ws_error))
3671 apply = 0;
3673 update_index = check_index && apply;
3674 if (update_index && newfd < 0)
3675 newfd = hold_locked_index(&lock_file, 1);
3677 if (check_index) {
3678 if (read_cache() < 0)
3679 die("unable to read index file");
3682 if ((check || apply) &&
3683 check_patch_list(list) < 0 &&
3684 !apply_with_reject)
3685 exit(1);
3687 if (apply && write_out_results(list, skipped_patch))
3688 exit(1);
3690 if (fake_ancestor)
3691 build_fake_ancestor(list, fake_ancestor);
3693 if (diffstat)
3694 stat_patch_list(list);
3696 if (numstat)
3697 numstat_patch_list(list);
3699 if (summary)
3700 summary_patch_list(list);
3702 strbuf_release(&buf);
3703 return 0;
3706 static int git_apply_config(const char *var, const char *value, void *cb)
3708 if (!strcmp(var, "apply.whitespace"))
3709 return git_config_string(&apply_default_whitespace, var, value);
3710 else if (!strcmp(var, "apply.ignorewhitespace"))
3711 return git_config_string(&apply_default_ignorewhitespace, var, value);
3712 return git_default_config(var, value, cb);
3715 static int option_parse_exclude(const struct option *opt,
3716 const char *arg, int unset)
3718 add_name_limit(arg, 1);
3719 return 0;
3722 static int option_parse_include(const struct option *opt,
3723 const char *arg, int unset)
3725 add_name_limit(arg, 0);
3726 has_include = 1;
3727 return 0;
3730 static int option_parse_p(const struct option *opt,
3731 const char *arg, int unset)
3733 p_value = atoi(arg);
3734 p_value_known = 1;
3735 return 0;
3738 static int option_parse_z(const struct option *opt,
3739 const char *arg, int unset)
3741 if (unset)
3742 line_termination = '\n';
3743 else
3744 line_termination = 0;
3745 return 0;
3748 static int option_parse_space_change(const struct option *opt,
3749 const char *arg, int unset)
3751 if (unset)
3752 ws_ignore_action = ignore_ws_none;
3753 else
3754 ws_ignore_action = ignore_ws_change;
3755 return 0;
3758 static int option_parse_whitespace(const struct option *opt,
3759 const char *arg, int unset)
3761 const char **whitespace_option = opt->value;
3763 *whitespace_option = arg;
3764 parse_whitespace_option(arg);
3765 return 0;
3768 static int option_parse_directory(const struct option *opt,
3769 const char *arg, int unset)
3771 root_len = strlen(arg);
3772 if (root_len && arg[root_len - 1] != '/') {
3773 char *new_root;
3774 root = new_root = xmalloc(root_len + 2);
3775 strcpy(new_root, arg);
3776 strcpy(new_root + root_len++, "/");
3777 } else
3778 root = arg;
3779 return 0;
3782 int cmd_apply(int argc, const char **argv, const char *prefix_)
3784 int i;
3785 int errs = 0;
3786 int is_not_gitdir = !startup_info->have_repository;
3787 int binary;
3788 int force_apply = 0;
3790 const char *whitespace_option = NULL;
3792 struct option builtin_apply_options[] = {
3793 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3794 "don't apply changes matching the given path",
3795 0, option_parse_exclude },
3796 { OPTION_CALLBACK, 0, "include", NULL, "path",
3797 "apply changes matching the given path",
3798 0, option_parse_include },
3799 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3800 "remove <num> leading slashes from traditional diff paths",
3801 0, option_parse_p },
3802 OPT_BOOLEAN(0, "no-add", &no_add,
3803 "ignore additions made by the patch"),
3804 OPT_BOOLEAN(0, "stat", &diffstat,
3805 "instead of applying the patch, output diffstat for the input"),
3806 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3807 NULL, "old option, now no-op",
3808 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3809 { OPTION_BOOLEAN, 0, "binary", &binary,
3810 NULL, "old option, now no-op",
3811 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3812 OPT_BOOLEAN(0, "numstat", &numstat,
3813 "shows number of added and deleted lines in decimal notation"),
3814 OPT_BOOLEAN(0, "summary", &summary,
3815 "instead of applying the patch, output a summary for the input"),
3816 OPT_BOOLEAN(0, "check", &check,
3817 "instead of applying the patch, see if the patch is applicable"),
3818 OPT_BOOLEAN(0, "index", &check_index,
3819 "make sure the patch is applicable to the current index"),
3820 OPT_BOOLEAN(0, "cached", &cached,
3821 "apply a patch without touching the working tree"),
3822 OPT_BOOLEAN(0, "apply", &force_apply,
3823 "also apply the patch (use with --stat/--summary/--check)"),
3824 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3825 "build a temporary index based on embedded index information"),
3826 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3827 "paths are separated with NUL character",
3828 PARSE_OPT_NOARG, option_parse_z },
3829 OPT_INTEGER('C', NULL, &p_context,
3830 "ensure at least <n> lines of context match"),
3831 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3832 "detect new or modified lines that have whitespace errors",
3833 0, option_parse_whitespace },
3834 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3835 "ignore changes in whitespace when finding context",
3836 PARSE_OPT_NOARG, option_parse_space_change },
3837 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3838 "ignore changes in whitespace when finding context",
3839 PARSE_OPT_NOARG, option_parse_space_change },
3840 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3841 "apply the patch in reverse"),
3842 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3843 "don't expect at least one line of context"),
3844 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3845 "leave the rejected hunks in corresponding *.rej files"),
3846 OPT__VERBOSE(&apply_verbosely),
3847 OPT_BIT(0, "inaccurate-eof", &options,
3848 "tolerate incorrectly detected missing new-line at the end of file",
3849 INACCURATE_EOF),
3850 OPT_BIT(0, "recount", &options,
3851 "do not trust the line counts in the hunk headers",
3852 RECOUNT),
3853 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3854 "prepend <root> to all filenames",
3855 0, option_parse_directory },
3856 OPT_END()
3859 prefix = prefix_;
3860 prefix_length = prefix ? strlen(prefix) : 0;
3861 git_config(git_apply_config, NULL);
3862 if (apply_default_whitespace)
3863 parse_whitespace_option(apply_default_whitespace);
3864 if (apply_default_ignorewhitespace)
3865 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3867 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3868 apply_usage, 0);
3870 if (apply_with_reject)
3871 apply = apply_verbosely = 1;
3872 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3873 apply = 0;
3874 if (check_index && is_not_gitdir)
3875 die("--index outside a repository");
3876 if (cached) {
3877 if (is_not_gitdir)
3878 die("--cached outside a repository");
3879 check_index = 1;
3881 for (i = 0; i < argc; i++) {
3882 const char *arg = argv[i];
3883 int fd;
3885 if (!strcmp(arg, "-")) {
3886 errs |= apply_patch(0, "<stdin>", options);
3887 read_stdin = 0;
3888 continue;
3889 } else if (0 < prefix_length)
3890 arg = prefix_filename(prefix, prefix_length, arg);
3892 fd = open(arg, O_RDONLY);
3893 if (fd < 0)
3894 die_errno("can't open patch '%s'", arg);
3895 read_stdin = 0;
3896 set_default_whitespace_mode(whitespace_option);
3897 errs |= apply_patch(fd, arg, options);
3898 close(fd);
3900 set_default_whitespace_mode(whitespace_option);
3901 if (read_stdin)
3902 errs |= apply_patch(0, "<stdin>", options);
3903 if (whitespace_error) {
3904 if (squelch_whitespace_errors &&
3905 squelch_whitespace_errors < whitespace_error) {
3906 int squelched =
3907 whitespace_error - squelch_whitespace_errors;
3908 warning("squelched %d "
3909 "whitespace error%s",
3910 squelched,
3911 squelched == 1 ? "" : "s");
3913 if (ws_error_action == die_on_ws_error)
3914 die("%d line%s add%s whitespace errors.",
3915 whitespace_error,
3916 whitespace_error == 1 ? "" : "s",
3917 whitespace_error == 1 ? "s" : "");
3918 if (applied_after_fixing_ws && apply)
3919 warning("%d line%s applied after"
3920 " fixing whitespace errors.",
3921 applied_after_fixing_ws,
3922 applied_after_fixing_ws == 1 ? "" : "s");
3923 else if (whitespace_error)
3924 warning("%d line%s add%s whitespace errors.",
3925 whitespace_error,
3926 whitespace_error == 1 ? "" : "s",
3927 whitespace_error == 1 ? "s" : "");
3930 if (update_index) {
3931 if (write_cache(newfd, active_cache, active_nr) ||
3932 commit_locked_index(&lock_file))
3933 die("Unable to write new index file");
3936 return !!errs;