apply: compute patch->def_name correctly under -p0
[git/jnareb-git.git] / builtin / apply.c
blob2ad8c4820720944f19207c4c09e87e9946116cf0
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "parse-options.h"
20 * --check turns on checking that the working tree matches the
21 * files that are being modified, but doesn't apply the patch
22 * --stat does just a diffstat, and doesn't actually apply
23 * --numstat does numeric diffstat, and doesn't actually apply
24 * --index-info shows the old and new index info for paths if available.
25 * --index updates the cache as well.
26 * --cached updates only the cache without ever touching the working tree.
28 static const char *prefix;
29 static int prefix_length = -1;
30 static int newfd = -1;
32 static int unidiff_zero;
33 static int p_value = 1;
34 static int p_value_known;
35 static int check_index;
36 static int update_index;
37 static int cached;
38 static int diffstat;
39 static int numstat;
40 static int summary;
41 static int check;
42 static int apply = 1;
43 static int apply_in_reverse;
44 static int apply_with_reject;
45 static int apply_verbosely;
46 static int allow_overlap;
47 static int no_add;
48 static const char *fake_ancestor;
49 static int line_termination = '\n';
50 static unsigned int p_context = UINT_MAX;
51 static const char * const apply_usage[] = {
52 "git apply [options] [<patch>...]",
53 NULL
56 static enum ws_error_action {
57 nowarn_ws_error,
58 warn_on_ws_error,
59 die_on_ws_error,
60 correct_ws_error
61 } ws_error_action = warn_on_ws_error;
62 static int whitespace_error;
63 static int squelch_whitespace_errors = 5;
64 static int applied_after_fixing_ws;
66 static enum ws_ignore {
67 ignore_ws_none,
68 ignore_ws_change
69 } ws_ignore_action = ignore_ws_none;
72 static const char *patch_input_file;
73 static const char *root;
74 static int root_len;
75 static int read_stdin = 1;
76 static int options;
78 static void parse_whitespace_option(const char *option)
80 if (!option) {
81 ws_error_action = warn_on_ws_error;
82 return;
84 if (!strcmp(option, "warn")) {
85 ws_error_action = warn_on_ws_error;
86 return;
88 if (!strcmp(option, "nowarn")) {
89 ws_error_action = nowarn_ws_error;
90 return;
92 if (!strcmp(option, "error")) {
93 ws_error_action = die_on_ws_error;
94 return;
96 if (!strcmp(option, "error-all")) {
97 ws_error_action = die_on_ws_error;
98 squelch_whitespace_errors = 0;
99 return;
101 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
102 ws_error_action = correct_ws_error;
103 return;
105 die("unrecognized whitespace option '%s'", option);
108 static void parse_ignorewhitespace_option(const char *option)
110 if (!option || !strcmp(option, "no") ||
111 !strcmp(option, "false") || !strcmp(option, "never") ||
112 !strcmp(option, "none")) {
113 ws_ignore_action = ignore_ws_none;
114 return;
116 if (!strcmp(option, "change")) {
117 ws_ignore_action = ignore_ws_change;
118 return;
120 die("unrecognized whitespace ignore option '%s'", option);
123 static void set_default_whitespace_mode(const char *whitespace_option)
125 if (!whitespace_option && !apply_default_whitespace)
126 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
130 * For "diff-stat" like behaviour, we keep track of the biggest change
131 * we've seen, and the longest filename. That allows us to do simple
132 * scaling.
134 static int max_change, max_len;
137 * Various "current state", notably line numbers and what
138 * file (and how) we're patching right now.. The "is_xxxx"
139 * things are flags, where -1 means "don't know yet".
141 static int linenr = 1;
144 * This represents one "hunk" from a patch, starting with
145 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
146 * patch text is pointed at by patch, and its byte length
147 * is stored in size. leading and trailing are the number
148 * of context lines.
150 struct fragment {
151 unsigned long leading, trailing;
152 unsigned long oldpos, oldlines;
153 unsigned long newpos, newlines;
154 const char *patch;
155 int size;
156 int rejected;
157 int linenr;
158 struct fragment *next;
162 * When dealing with a binary patch, we reuse "leading" field
163 * to store the type of the binary hunk, either deflated "delta"
164 * or deflated "literal".
166 #define binary_patch_method leading
167 #define BINARY_DELTA_DEFLATED 1
168 #define BINARY_LITERAL_DEFLATED 2
171 * This represents a "patch" to a file, both metainfo changes
172 * such as creation/deletion, filemode and content changes represented
173 * as a series of fragments.
175 struct patch {
176 char *new_name, *old_name, *def_name;
177 unsigned int old_mode, new_mode;
178 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
179 int rejected;
180 unsigned ws_rule;
181 unsigned long deflate_origlen;
182 int lines_added, lines_deleted;
183 int score;
184 unsigned int is_toplevel_relative:1;
185 unsigned int inaccurate_eof:1;
186 unsigned int is_binary:1;
187 unsigned int is_copy:1;
188 unsigned int is_rename:1;
189 unsigned int recount:1;
190 struct fragment *fragments;
191 char *result;
192 size_t resultsize;
193 char old_sha1_prefix[41];
194 char new_sha1_prefix[41];
195 struct patch *next;
199 * A line in a file, len-bytes long (includes the terminating LF,
200 * except for an incomplete line at the end if the file ends with
201 * one), and its contents hashes to 'hash'.
203 struct line {
204 size_t len;
205 unsigned hash : 24;
206 unsigned flag : 8;
207 #define LINE_COMMON 1
208 #define LINE_PATCHED 2
212 * This represents a "file", which is an array of "lines".
214 struct image {
215 char *buf;
216 size_t len;
217 size_t nr;
218 size_t alloc;
219 struct line *line_allocated;
220 struct line *line;
224 * Records filenames that have been touched, in order to handle
225 * the case where more than one patches touch the same file.
228 static struct string_list fn_table;
230 static uint32_t hash_line(const char *cp, size_t len)
232 size_t i;
233 uint32_t h;
234 for (i = 0, h = 0; i < len; i++) {
235 if (!isspace(cp[i])) {
236 h = h * 3 + (cp[i] & 0xff);
239 return h;
243 * Compare lines s1 of length n1 and s2 of length n2, ignoring
244 * whitespace difference. Returns 1 if they match, 0 otherwise
246 static int fuzzy_matchlines(const char *s1, size_t n1,
247 const char *s2, size_t n2)
249 const char *last1 = s1 + n1 - 1;
250 const char *last2 = s2 + n2 - 1;
251 int result = 0;
253 /* ignore line endings */
254 while ((*last1 == '\r') || (*last1 == '\n'))
255 last1--;
256 while ((*last2 == '\r') || (*last2 == '\n'))
257 last2--;
259 /* skip leading whitespace */
260 while (isspace(*s1) && (s1 <= last1))
261 s1++;
262 while (isspace(*s2) && (s2 <= last2))
263 s2++;
264 /* early return if both lines are empty */
265 if ((s1 > last1) && (s2 > last2))
266 return 1;
267 while (!result) {
268 result = *s1++ - *s2++;
270 * Skip whitespace inside. We check for whitespace on
271 * both buffers because we don't want "a b" to match
272 * "ab"
274 if (isspace(*s1) && isspace(*s2)) {
275 while (isspace(*s1) && s1 <= last1)
276 s1++;
277 while (isspace(*s2) && s2 <= last2)
278 s2++;
281 * If we reached the end on one side only,
282 * lines don't match
284 if (
285 ((s2 > last2) && (s1 <= last1)) ||
286 ((s1 > last1) && (s2 <= last2)))
287 return 0;
288 if ((s1 > last1) && (s2 > last2))
289 break;
292 return !result;
295 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
297 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
298 img->line_allocated[img->nr].len = len;
299 img->line_allocated[img->nr].hash = hash_line(bol, len);
300 img->line_allocated[img->nr].flag = flag;
301 img->nr++;
304 static void prepare_image(struct image *image, char *buf, size_t len,
305 int prepare_linetable)
307 const char *cp, *ep;
309 memset(image, 0, sizeof(*image));
310 image->buf = buf;
311 image->len = len;
313 if (!prepare_linetable)
314 return;
316 ep = image->buf + image->len;
317 cp = image->buf;
318 while (cp < ep) {
319 const char *next;
320 for (next = cp; next < ep && *next != '\n'; next++)
322 if (next < ep)
323 next++;
324 add_line_info(image, cp, next - cp, 0);
325 cp = next;
327 image->line = image->line_allocated;
330 static void clear_image(struct image *image)
332 free(image->buf);
333 image->buf = NULL;
334 image->len = 0;
337 static void say_patch_name(FILE *output, const char *pre,
338 struct patch *patch, const char *post)
340 fputs(pre, output);
341 if (patch->old_name && patch->new_name &&
342 strcmp(patch->old_name, patch->new_name)) {
343 quote_c_style(patch->old_name, NULL, output, 0);
344 fputs(" => ", output);
345 quote_c_style(patch->new_name, NULL, output, 0);
346 } else {
347 const char *n = patch->new_name;
348 if (!n)
349 n = patch->old_name;
350 quote_c_style(n, NULL, output, 0);
352 fputs(post, output);
355 #define CHUNKSIZE (8192)
356 #define SLOP (16)
358 static void read_patch_file(struct strbuf *sb, int fd)
360 if (strbuf_read(sb, fd, 0) < 0)
361 die_errno("git apply: failed to read");
364 * Make sure that we have some slop in the buffer
365 * so that we can do speculative "memcmp" etc, and
366 * see to it that it is NUL-filled.
368 strbuf_grow(sb, SLOP);
369 memset(sb->buf + sb->len, 0, SLOP);
372 static unsigned long linelen(const char *buffer, unsigned long size)
374 unsigned long len = 0;
375 while (size--) {
376 len++;
377 if (*buffer++ == '\n')
378 break;
380 return len;
383 static int is_dev_null(const char *str)
385 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
388 #define TERM_SPACE 1
389 #define TERM_TAB 2
391 static int name_terminate(const char *name, int namelen, int c, int terminate)
393 if (c == ' ' && !(terminate & TERM_SPACE))
394 return 0;
395 if (c == '\t' && !(terminate & TERM_TAB))
396 return 0;
398 return 1;
401 /* remove double slashes to make --index work with such filenames */
402 static char *squash_slash(char *name)
404 int i = 0, j = 0;
406 if (!name)
407 return NULL;
409 while (name[i]) {
410 if ((name[j++] = name[i++]) == '/')
411 while (name[i] == '/')
412 i++;
414 name[j] = '\0';
415 return name;
418 static char *find_name_gnu(const char *line, char *def, int p_value)
420 struct strbuf name = STRBUF_INIT;
421 char *cp;
424 * Proposed "new-style" GNU patch/diff format; see
425 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
427 if (unquote_c_style(&name, line, NULL)) {
428 strbuf_release(&name);
429 return NULL;
432 for (cp = name.buf; p_value; p_value--) {
433 cp = strchr(cp, '/');
434 if (!cp) {
435 strbuf_release(&name);
436 return NULL;
438 cp++;
441 /* name can later be freed, so we need
442 * to memmove, not just return cp
444 strbuf_remove(&name, 0, cp - name.buf);
445 free(def);
446 if (root)
447 strbuf_insert(&name, 0, root, root_len);
448 return squash_slash(strbuf_detach(&name, NULL));
451 static size_t sane_tz_len(const char *line, size_t len)
453 const char *tz, *p;
455 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
456 return 0;
457 tz = line + len - strlen(" +0500");
459 if (tz[1] != '+' && tz[1] != '-')
460 return 0;
462 for (p = tz + 2; p != line + len; p++)
463 if (!isdigit(*p))
464 return 0;
466 return line + len - tz;
469 static size_t tz_with_colon_len(const char *line, size_t len)
471 const char *tz, *p;
473 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
474 return 0;
475 tz = line + len - strlen(" +08:00");
477 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
478 return 0;
479 p = tz + 2;
480 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
481 !isdigit(*p++) || !isdigit(*p++))
482 return 0;
484 return line + len - tz;
487 static size_t date_len(const char *line, size_t len)
489 const char *date, *p;
491 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
492 return 0;
493 p = date = line + len - strlen("72-02-05");
495 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
496 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
497 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
498 return 0;
500 if (date - line >= strlen("19") &&
501 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
502 date -= strlen("19");
504 return line + len - date;
507 static size_t short_time_len(const char *line, size_t len)
509 const char *time, *p;
511 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
512 return 0;
513 p = time = line + len - strlen(" 07:01:32");
515 /* Permit 1-digit hours? */
516 if (*p++ != ' ' ||
517 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
518 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
519 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
520 return 0;
522 return line + len - time;
525 static size_t fractional_time_len(const char *line, size_t len)
527 const char *p;
528 size_t n;
530 /* Expected format: 19:41:17.620000023 */
531 if (!len || !isdigit(line[len - 1]))
532 return 0;
533 p = line + len - 1;
535 /* Fractional seconds. */
536 while (p > line && isdigit(*p))
537 p--;
538 if (*p != '.')
539 return 0;
541 /* Hours, minutes, and whole seconds. */
542 n = short_time_len(line, p - line);
543 if (!n)
544 return 0;
546 return line + len - p + n;
549 static size_t trailing_spaces_len(const char *line, size_t len)
551 const char *p;
553 /* Expected format: ' ' x (1 or more) */
554 if (!len || line[len - 1] != ' ')
555 return 0;
557 p = line + len;
558 while (p != line) {
559 p--;
560 if (*p != ' ')
561 return line + len - (p + 1);
564 /* All spaces! */
565 return len;
568 static size_t diff_timestamp_len(const char *line, size_t len)
570 const char *end = line + len;
571 size_t n;
574 * Posix: 2010-07-05 19:41:17
575 * GNU: 2010-07-05 19:41:17.620000023 -0500
578 if (!isdigit(end[-1]))
579 return 0;
581 n = sane_tz_len(line, end - line);
582 if (!n)
583 n = tz_with_colon_len(line, end - line);
584 end -= n;
586 n = short_time_len(line, end - line);
587 if (!n)
588 n = fractional_time_len(line, end - line);
589 end -= n;
591 n = date_len(line, end - line);
592 if (!n) /* No date. Too bad. */
593 return 0;
594 end -= n;
596 if (end == line) /* No space before date. */
597 return 0;
598 if (end[-1] == '\t') { /* Success! */
599 end--;
600 return line + len - end;
602 if (end[-1] != ' ') /* No space before date. */
603 return 0;
605 /* Whitespace damage. */
606 end -= trailing_spaces_len(line, end - line);
607 return line + len - end;
610 static char *find_name_common(const char *line, char *def, int p_value,
611 const char *end, int terminate)
613 int len;
614 const char *start = NULL;
616 if (p_value == 0)
617 start = line;
618 while (line != end) {
619 char c = *line;
621 if (!end && isspace(c)) {
622 if (c == '\n')
623 break;
624 if (name_terminate(start, line-start, c, terminate))
625 break;
627 line++;
628 if (c == '/' && !--p_value)
629 start = line;
631 if (!start)
632 return squash_slash(def);
633 len = line - start;
634 if (!len)
635 return squash_slash(def);
638 * Generally we prefer the shorter name, especially
639 * if the other one is just a variation of that with
640 * something else tacked on to the end (ie "file.orig"
641 * or "file~").
643 if (def) {
644 int deflen = strlen(def);
645 if (deflen < len && !strncmp(start, def, deflen))
646 return squash_slash(def);
647 free(def);
650 if (root) {
651 char *ret = xmalloc(root_len + len + 1);
652 strcpy(ret, root);
653 memcpy(ret + root_len, start, len);
654 ret[root_len + len] = '\0';
655 return squash_slash(ret);
658 return squash_slash(xmemdupz(start, len));
661 static char *find_name(const char *line, char *def, int p_value, int terminate)
663 if (*line == '"') {
664 char *name = find_name_gnu(line, def, p_value);
665 if (name)
666 return name;
669 return find_name_common(line, def, p_value, NULL, terminate);
672 static char *find_name_traditional(const char *line, char *def, int p_value)
674 size_t len = strlen(line);
675 size_t date_len;
677 if (*line == '"') {
678 char *name = find_name_gnu(line, def, p_value);
679 if (name)
680 return name;
683 len = strchrnul(line, '\n') - line;
684 date_len = diff_timestamp_len(line, len);
685 if (!date_len)
686 return find_name_common(line, def, p_value, NULL, TERM_TAB);
687 len -= date_len;
689 return find_name_common(line, def, p_value, line + len, 0);
692 static int count_slashes(const char *cp)
694 int cnt = 0;
695 char ch;
697 while ((ch = *cp++))
698 if (ch == '/')
699 cnt++;
700 return cnt;
704 * Given the string after "--- " or "+++ ", guess the appropriate
705 * p_value for the given patch.
707 static int guess_p_value(const char *nameline)
709 char *name, *cp;
710 int val = -1;
712 if (is_dev_null(nameline))
713 return -1;
714 name = find_name_traditional(nameline, NULL, 0);
715 if (!name)
716 return -1;
717 cp = strchr(name, '/');
718 if (!cp)
719 val = 0;
720 else if (prefix) {
722 * Does it begin with "a/$our-prefix" and such? Then this is
723 * very likely to apply to our directory.
725 if (!strncmp(name, prefix, prefix_length))
726 val = count_slashes(prefix);
727 else {
728 cp++;
729 if (!strncmp(cp, prefix, prefix_length))
730 val = count_slashes(prefix) + 1;
733 free(name);
734 return val;
738 * Does the ---/+++ line has the POSIX timestamp after the last HT?
739 * GNU diff puts epoch there to signal a creation/deletion event. Is
740 * this such a timestamp?
742 static int has_epoch_timestamp(const char *nameline)
745 * We are only interested in epoch timestamp; any non-zero
746 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
747 * For the same reason, the date must be either 1969-12-31 or
748 * 1970-01-01, and the seconds part must be "00".
750 const char stamp_regexp[] =
751 "^(1969-12-31|1970-01-01)"
753 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
755 "([-+][0-2][0-9]:?[0-5][0-9])\n";
756 const char *timestamp = NULL, *cp, *colon;
757 static regex_t *stamp;
758 regmatch_t m[10];
759 int zoneoffset;
760 int hourminute;
761 int status;
763 for (cp = nameline; *cp != '\n'; cp++) {
764 if (*cp == '\t')
765 timestamp = cp + 1;
767 if (!timestamp)
768 return 0;
769 if (!stamp) {
770 stamp = xmalloc(sizeof(*stamp));
771 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
772 warning("Cannot prepare timestamp regexp %s",
773 stamp_regexp);
774 return 0;
778 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
779 if (status) {
780 if (status != REG_NOMATCH)
781 warning("regexec returned %d for input: %s",
782 status, timestamp);
783 return 0;
786 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
787 if (*colon == ':')
788 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
789 else
790 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
791 if (timestamp[m[3].rm_so] == '-')
792 zoneoffset = -zoneoffset;
795 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
796 * (west of GMT) or 1970-01-01 (east of GMT)
798 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
799 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
800 return 0;
802 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
803 strtol(timestamp + 14, NULL, 10) -
804 zoneoffset);
806 return ((zoneoffset < 0 && hourminute == 1440) ||
807 (0 <= zoneoffset && !hourminute));
811 * Get the name etc info from the ---/+++ lines of a traditional patch header
813 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
814 * files, we can happily check the index for a match, but for creating a
815 * new file we should try to match whatever "patch" does. I have no idea.
817 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
819 char *name;
821 first += 4; /* skip "--- " */
822 second += 4; /* skip "+++ " */
823 if (!p_value_known) {
824 int p, q;
825 p = guess_p_value(first);
826 q = guess_p_value(second);
827 if (p < 0) p = q;
828 if (0 <= p && p == q) {
829 p_value = p;
830 p_value_known = 1;
833 if (is_dev_null(first)) {
834 patch->is_new = 1;
835 patch->is_delete = 0;
836 name = find_name_traditional(second, NULL, p_value);
837 patch->new_name = name;
838 } else if (is_dev_null(second)) {
839 patch->is_new = 0;
840 patch->is_delete = 1;
841 name = find_name_traditional(first, NULL, p_value);
842 patch->old_name = name;
843 } else {
844 name = find_name_traditional(first, NULL, p_value);
845 name = find_name_traditional(second, name, p_value);
846 if (has_epoch_timestamp(first)) {
847 patch->is_new = 1;
848 patch->is_delete = 0;
849 patch->new_name = name;
850 } else if (has_epoch_timestamp(second)) {
851 patch->is_new = 0;
852 patch->is_delete = 1;
853 patch->old_name = name;
854 } else {
855 patch->old_name = patch->new_name = name;
858 if (!name)
859 die("unable to find filename in patch at line %d", linenr);
862 static int gitdiff_hdrend(const char *line, struct patch *patch)
864 return -1;
868 * We're anal about diff header consistency, to make
869 * sure that we don't end up having strange ambiguous
870 * patches floating around.
872 * As a result, gitdiff_{old|new}name() will check
873 * their names against any previous information, just
874 * to make sure..
876 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
878 if (!orig_name && !isnull)
879 return find_name(line, NULL, p_value, TERM_TAB);
881 if (orig_name) {
882 int len;
883 const char *name;
884 char *another;
885 name = orig_name;
886 len = strlen(name);
887 if (isnull)
888 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
889 another = find_name(line, NULL, p_value, TERM_TAB);
890 if (!another || memcmp(another, name, len + 1))
891 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
892 free(another);
893 return orig_name;
895 else {
896 /* expect "/dev/null" */
897 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
898 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
899 return NULL;
903 static int gitdiff_oldname(const char *line, struct patch *patch)
905 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
906 return 0;
909 static int gitdiff_newname(const char *line, struct patch *patch)
911 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
912 return 0;
915 static int gitdiff_oldmode(const char *line, struct patch *patch)
917 patch->old_mode = strtoul(line, NULL, 8);
918 return 0;
921 static int gitdiff_newmode(const char *line, struct patch *patch)
923 patch->new_mode = strtoul(line, NULL, 8);
924 return 0;
927 static int gitdiff_delete(const char *line, struct patch *patch)
929 patch->is_delete = 1;
930 patch->old_name = patch->def_name;
931 return gitdiff_oldmode(line, patch);
934 static int gitdiff_newfile(const char *line, struct patch *patch)
936 patch->is_new = 1;
937 patch->new_name = patch->def_name;
938 return gitdiff_newmode(line, patch);
941 static int gitdiff_copysrc(const char *line, struct patch *patch)
943 patch->is_copy = 1;
944 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
945 return 0;
948 static int gitdiff_copydst(const char *line, struct patch *patch)
950 patch->is_copy = 1;
951 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
952 return 0;
955 static int gitdiff_renamesrc(const char *line, struct patch *patch)
957 patch->is_rename = 1;
958 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
959 return 0;
962 static int gitdiff_renamedst(const char *line, struct patch *patch)
964 patch->is_rename = 1;
965 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
966 return 0;
969 static int gitdiff_similarity(const char *line, struct patch *patch)
971 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
972 patch->score = 0;
973 return 0;
976 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
978 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
979 patch->score = 0;
980 return 0;
983 static int gitdiff_index(const char *line, struct patch *patch)
986 * index line is N hexadecimal, "..", N hexadecimal,
987 * and optional space with octal mode.
989 const char *ptr, *eol;
990 int len;
992 ptr = strchr(line, '.');
993 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
994 return 0;
995 len = ptr - line;
996 memcpy(patch->old_sha1_prefix, line, len);
997 patch->old_sha1_prefix[len] = 0;
999 line = ptr + 2;
1000 ptr = strchr(line, ' ');
1001 eol = strchr(line, '\n');
1003 if (!ptr || eol < ptr)
1004 ptr = eol;
1005 len = ptr - line;
1007 if (40 < len)
1008 return 0;
1009 memcpy(patch->new_sha1_prefix, line, len);
1010 patch->new_sha1_prefix[len] = 0;
1011 if (*ptr == ' ')
1012 patch->old_mode = strtoul(ptr+1, NULL, 8);
1013 return 0;
1017 * This is normal for a diff that doesn't change anything: we'll fall through
1018 * into the next diff. Tell the parser to break out.
1020 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1022 return -1;
1026 * Skip p_value leading components from "line"; as we do not accept
1027 * absolute paths, return NULL in that case.
1029 static const char *skip_tree_prefix(const char *line, int llen)
1031 int nslash;
1032 int i;
1034 if (!p_value)
1035 return (llen && line[0] == '/') ? NULL : line;
1037 nslash = p_value;
1038 for (i = 0; i < llen; i++) {
1039 int ch = line[i];
1040 if (ch == '/' && --nslash <= 0)
1041 return (i == 0) ? NULL : &line[i + 1];
1043 return NULL;
1047 * This is to extract the same name that appears on "diff --git"
1048 * line. We do not find and return anything if it is a rename
1049 * patch, and it is OK because we will find the name elsewhere.
1050 * We need to reliably find name only when it is mode-change only,
1051 * creation or deletion of an empty file. In any of these cases,
1052 * both sides are the same name under a/ and b/ respectively.
1054 static char *git_header_name(char *line, int llen)
1056 const char *name;
1057 const char *second = NULL;
1058 size_t len, line_len;
1060 line += strlen("diff --git ");
1061 llen -= strlen("diff --git ");
1063 if (*line == '"') {
1064 const char *cp;
1065 struct strbuf first = STRBUF_INIT;
1066 struct strbuf sp = STRBUF_INIT;
1068 if (unquote_c_style(&first, line, &second))
1069 goto free_and_fail1;
1071 /* strip the a/b prefix including trailing slash */
1072 cp = skip_tree_prefix(first.buf, first.len);
1073 if (!cp)
1074 goto free_and_fail1;
1075 strbuf_remove(&first, 0, cp - first.buf);
1078 * second points at one past closing dq of name.
1079 * find the second name.
1081 while ((second < line + llen) && isspace(*second))
1082 second++;
1084 if (line + llen <= second)
1085 goto free_and_fail1;
1086 if (*second == '"') {
1087 if (unquote_c_style(&sp, second, NULL))
1088 goto free_and_fail1;
1089 cp = skip_tree_prefix(sp.buf, sp.len);
1090 if (!cp)
1091 goto free_and_fail1;
1092 /* They must match, otherwise ignore */
1093 if (strcmp(cp, first.buf))
1094 goto free_and_fail1;
1095 strbuf_release(&sp);
1096 return strbuf_detach(&first, NULL);
1099 /* unquoted second */
1100 cp = skip_tree_prefix(second, line + llen - second);
1101 if (!cp)
1102 goto free_and_fail1;
1103 if (line + llen - cp != first.len ||
1104 memcmp(first.buf, cp, first.len))
1105 goto free_and_fail1;
1106 return strbuf_detach(&first, NULL);
1108 free_and_fail1:
1109 strbuf_release(&first);
1110 strbuf_release(&sp);
1111 return NULL;
1114 /* unquoted first name */
1115 name = skip_tree_prefix(line, llen);
1116 if (!name)
1117 return NULL;
1120 * since the first name is unquoted, a dq if exists must be
1121 * the beginning of the second name.
1123 for (second = name; second < line + llen; second++) {
1124 if (*second == '"') {
1125 struct strbuf sp = STRBUF_INIT;
1126 const char *np;
1128 if (unquote_c_style(&sp, second, NULL))
1129 goto free_and_fail2;
1131 np = skip_tree_prefix(sp.buf, sp.len);
1132 if (!np)
1133 goto free_and_fail2;
1135 len = sp.buf + sp.len - np;
1136 if (len < second - name &&
1137 !strncmp(np, name, len) &&
1138 isspace(name[len])) {
1139 /* Good */
1140 strbuf_remove(&sp, 0, np - sp.buf);
1141 return strbuf_detach(&sp, NULL);
1144 free_and_fail2:
1145 strbuf_release(&sp);
1146 return NULL;
1151 * Accept a name only if it shows up twice, exactly the same
1152 * form.
1154 second = strchr(name, '\n');
1155 if (!second)
1156 return NULL;
1157 line_len = second - name;
1158 for (len = 0 ; ; len++) {
1159 switch (name[len]) {
1160 default:
1161 continue;
1162 case '\n':
1163 return NULL;
1164 case '\t': case ' ':
1166 * Is this the separator between the preimage
1167 * and the postimage pathname? Again, we are
1168 * only interested in the case where there is
1169 * no rename, as this is only to set def_name
1170 * and a rename patch has the names elsewhere
1171 * in an unambiguous form.
1173 if (!name[len + 1])
1174 return NULL; /* no postimage name */
1175 second = skip_tree_prefix(name + len + 1,
1176 line_len - (len + 1));
1177 if (!second)
1178 return NULL;
1180 * Does len bytes starting at "name" and "second"
1181 * (that are separated by one HT or SP we just
1182 * found) exactly match?
1184 if (second[len] == '\n' && !strncmp(name, second, len))
1185 return xmemdupz(name, len);
1190 /* Verify that we recognize the lines following a git header */
1191 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
1193 unsigned long offset;
1195 /* A git diff has explicit new/delete information, so we don't guess */
1196 patch->is_new = 0;
1197 patch->is_delete = 0;
1200 * Some things may not have the old name in the
1201 * rest of the headers anywhere (pure mode changes,
1202 * or removing or adding empty files), so we get
1203 * the default name from the header.
1205 patch->def_name = git_header_name(line, len);
1206 if (patch->def_name && root) {
1207 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
1208 strcpy(s, root);
1209 strcpy(s + root_len, patch->def_name);
1210 free(patch->def_name);
1211 patch->def_name = s;
1214 line += len;
1215 size -= len;
1216 linenr++;
1217 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1218 static const struct opentry {
1219 const char *str;
1220 int (*fn)(const char *, struct patch *);
1221 } optable[] = {
1222 { "@@ -", gitdiff_hdrend },
1223 { "--- ", gitdiff_oldname },
1224 { "+++ ", gitdiff_newname },
1225 { "old mode ", gitdiff_oldmode },
1226 { "new mode ", gitdiff_newmode },
1227 { "deleted file mode ", gitdiff_delete },
1228 { "new file mode ", gitdiff_newfile },
1229 { "copy from ", gitdiff_copysrc },
1230 { "copy to ", gitdiff_copydst },
1231 { "rename old ", gitdiff_renamesrc },
1232 { "rename new ", gitdiff_renamedst },
1233 { "rename from ", gitdiff_renamesrc },
1234 { "rename to ", gitdiff_renamedst },
1235 { "similarity index ", gitdiff_similarity },
1236 { "dissimilarity index ", gitdiff_dissimilarity },
1237 { "index ", gitdiff_index },
1238 { "", gitdiff_unrecognized },
1240 int i;
1242 len = linelen(line, size);
1243 if (!len || line[len-1] != '\n')
1244 break;
1245 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1246 const struct opentry *p = optable + i;
1247 int oplen = strlen(p->str);
1248 if (len < oplen || memcmp(p->str, line, oplen))
1249 continue;
1250 if (p->fn(line + oplen, patch) < 0)
1251 return offset;
1252 break;
1256 return offset;
1259 static int parse_num(const char *line, unsigned long *p)
1261 char *ptr;
1263 if (!isdigit(*line))
1264 return 0;
1265 *p = strtoul(line, &ptr, 10);
1266 return ptr - line;
1269 static int parse_range(const char *line, int len, int offset, const char *expect,
1270 unsigned long *p1, unsigned long *p2)
1272 int digits, ex;
1274 if (offset < 0 || offset >= len)
1275 return -1;
1276 line += offset;
1277 len -= offset;
1279 digits = parse_num(line, p1);
1280 if (!digits)
1281 return -1;
1283 offset += digits;
1284 line += digits;
1285 len -= digits;
1287 *p2 = 1;
1288 if (*line == ',') {
1289 digits = parse_num(line+1, p2);
1290 if (!digits)
1291 return -1;
1293 offset += digits+1;
1294 line += digits+1;
1295 len -= digits+1;
1298 ex = strlen(expect);
1299 if (ex > len)
1300 return -1;
1301 if (memcmp(line, expect, ex))
1302 return -1;
1304 return offset + ex;
1307 static void recount_diff(char *line, int size, struct fragment *fragment)
1309 int oldlines = 0, newlines = 0, ret = 0;
1311 if (size < 1) {
1312 warning("recount: ignore empty hunk");
1313 return;
1316 for (;;) {
1317 int len = linelen(line, size);
1318 size -= len;
1319 line += len;
1321 if (size < 1)
1322 break;
1324 switch (*line) {
1325 case ' ': case '\n':
1326 newlines++;
1327 /* fall through */
1328 case '-':
1329 oldlines++;
1330 continue;
1331 case '+':
1332 newlines++;
1333 continue;
1334 case '\\':
1335 continue;
1336 case '@':
1337 ret = size < 3 || prefixcmp(line, "@@ ");
1338 break;
1339 case 'd':
1340 ret = size < 5 || prefixcmp(line, "diff ");
1341 break;
1342 default:
1343 ret = -1;
1344 break;
1346 if (ret) {
1347 warning("recount: unexpected line: %.*s",
1348 (int)linelen(line, size), line);
1349 return;
1351 break;
1353 fragment->oldlines = oldlines;
1354 fragment->newlines = newlines;
1358 * Parse a unified diff fragment header of the
1359 * form "@@ -a,b +c,d @@"
1361 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1363 int offset;
1365 if (!len || line[len-1] != '\n')
1366 return -1;
1368 /* Figure out the number of lines in a fragment */
1369 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1370 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1372 return offset;
1375 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1377 unsigned long offset, len;
1379 patch->is_toplevel_relative = 0;
1380 patch->is_rename = patch->is_copy = 0;
1381 patch->is_new = patch->is_delete = -1;
1382 patch->old_mode = patch->new_mode = 0;
1383 patch->old_name = patch->new_name = NULL;
1384 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1385 unsigned long nextlen;
1387 len = linelen(line, size);
1388 if (!len)
1389 break;
1391 /* Testing this early allows us to take a few shortcuts.. */
1392 if (len < 6)
1393 continue;
1396 * Make sure we don't find any unconnected patch fragments.
1397 * That's a sign that we didn't find a header, and that a
1398 * patch has become corrupted/broken up.
1400 if (!memcmp("@@ -", line, 4)) {
1401 struct fragment dummy;
1402 if (parse_fragment_header(line, len, &dummy) < 0)
1403 continue;
1404 die("patch fragment without header at line %d: %.*s",
1405 linenr, (int)len-1, line);
1408 if (size < len + 6)
1409 break;
1412 * Git patch? It might not have a real patch, just a rename
1413 * or mode change, so we handle that specially
1415 if (!memcmp("diff --git ", line, 11)) {
1416 int git_hdr_len = parse_git_header(line, len, size, patch);
1417 if (git_hdr_len <= len)
1418 continue;
1419 if (!patch->old_name && !patch->new_name) {
1420 if (!patch->def_name)
1421 die("git diff header lacks filename information when removing "
1422 "%d leading pathname components (line %d)" , p_value, linenr);
1423 patch->old_name = patch->new_name = patch->def_name;
1425 if (!patch->is_delete && !patch->new_name)
1426 die("git diff header lacks filename information "
1427 "(line %d)", linenr);
1428 patch->is_toplevel_relative = 1;
1429 *hdrsize = git_hdr_len;
1430 return offset;
1433 /* --- followed by +++ ? */
1434 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1435 continue;
1438 * We only accept unified patches, so we want it to
1439 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1440 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1442 nextlen = linelen(line + len, size - len);
1443 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1444 continue;
1446 /* Ok, we'll consider it a patch */
1447 parse_traditional_patch(line, line+len, patch);
1448 *hdrsize = len + nextlen;
1449 linenr += 2;
1450 return offset;
1452 return -1;
1455 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1457 char *err;
1459 if (!result)
1460 return;
1462 whitespace_error++;
1463 if (squelch_whitespace_errors &&
1464 squelch_whitespace_errors < whitespace_error)
1465 return;
1467 err = whitespace_error_string(result);
1468 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1469 patch_input_file, linenr, err, len, line);
1470 free(err);
1473 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1475 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1477 record_ws_error(result, line + 1, len - 2, linenr);
1481 * Parse a unified diff. Note that this really needs to parse each
1482 * fragment separately, since the only way to know the difference
1483 * between a "---" that is part of a patch, and a "---" that starts
1484 * the next patch is to look at the line counts..
1486 static int parse_fragment(char *line, unsigned long size,
1487 struct patch *patch, struct fragment *fragment)
1489 int added, deleted;
1490 int len = linelen(line, size), offset;
1491 unsigned long oldlines, newlines;
1492 unsigned long leading, trailing;
1494 offset = parse_fragment_header(line, len, fragment);
1495 if (offset < 0)
1496 return -1;
1497 if (offset > 0 && patch->recount)
1498 recount_diff(line + offset, size - offset, fragment);
1499 oldlines = fragment->oldlines;
1500 newlines = fragment->newlines;
1501 leading = 0;
1502 trailing = 0;
1504 /* Parse the thing.. */
1505 line += len;
1506 size -= len;
1507 linenr++;
1508 added = deleted = 0;
1509 for (offset = len;
1510 0 < size;
1511 offset += len, size -= len, line += len, linenr++) {
1512 if (!oldlines && !newlines)
1513 break;
1514 len = linelen(line, size);
1515 if (!len || line[len-1] != '\n')
1516 return -1;
1517 switch (*line) {
1518 default:
1519 return -1;
1520 case '\n': /* newer GNU diff, an empty context line */
1521 case ' ':
1522 oldlines--;
1523 newlines--;
1524 if (!deleted && !added)
1525 leading++;
1526 trailing++;
1527 break;
1528 case '-':
1529 if (apply_in_reverse &&
1530 ws_error_action != nowarn_ws_error)
1531 check_whitespace(line, len, patch->ws_rule);
1532 deleted++;
1533 oldlines--;
1534 trailing = 0;
1535 break;
1536 case '+':
1537 if (!apply_in_reverse &&
1538 ws_error_action != nowarn_ws_error)
1539 check_whitespace(line, len, patch->ws_rule);
1540 added++;
1541 newlines--;
1542 trailing = 0;
1543 break;
1546 * We allow "\ No newline at end of file". Depending
1547 * on locale settings when the patch was produced we
1548 * don't know what this line looks like. The only
1549 * thing we do know is that it begins with "\ ".
1550 * Checking for 12 is just for sanity check -- any
1551 * l10n of "\ No newline..." is at least that long.
1553 case '\\':
1554 if (len < 12 || memcmp(line, "\\ ", 2))
1555 return -1;
1556 break;
1559 if (oldlines || newlines)
1560 return -1;
1561 fragment->leading = leading;
1562 fragment->trailing = trailing;
1565 * If a fragment ends with an incomplete line, we failed to include
1566 * it in the above loop because we hit oldlines == newlines == 0
1567 * before seeing it.
1569 if (12 < size && !memcmp(line, "\\ ", 2))
1570 offset += linelen(line, size);
1572 patch->lines_added += added;
1573 patch->lines_deleted += deleted;
1575 if (0 < patch->is_new && oldlines)
1576 return error("new file depends on old contents");
1577 if (0 < patch->is_delete && newlines)
1578 return error("deleted file still has contents");
1579 return offset;
1582 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1584 unsigned long offset = 0;
1585 unsigned long oldlines = 0, newlines = 0, context = 0;
1586 struct fragment **fragp = &patch->fragments;
1588 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1589 struct fragment *fragment;
1590 int len;
1592 fragment = xcalloc(1, sizeof(*fragment));
1593 fragment->linenr = linenr;
1594 len = parse_fragment(line, size, patch, fragment);
1595 if (len <= 0)
1596 die("corrupt patch at line %d", linenr);
1597 fragment->patch = line;
1598 fragment->size = len;
1599 oldlines += fragment->oldlines;
1600 newlines += fragment->newlines;
1601 context += fragment->leading + fragment->trailing;
1603 *fragp = fragment;
1604 fragp = &fragment->next;
1606 offset += len;
1607 line += len;
1608 size -= len;
1612 * If something was removed (i.e. we have old-lines) it cannot
1613 * be creation, and if something was added it cannot be
1614 * deletion. However, the reverse is not true; --unified=0
1615 * patches that only add are not necessarily creation even
1616 * though they do not have any old lines, and ones that only
1617 * delete are not necessarily deletion.
1619 * Unfortunately, a real creation/deletion patch do _not_ have
1620 * any context line by definition, so we cannot safely tell it
1621 * apart with --unified=0 insanity. At least if the patch has
1622 * more than one hunk it is not creation or deletion.
1624 if (patch->is_new < 0 &&
1625 (oldlines || (patch->fragments && patch->fragments->next)))
1626 patch->is_new = 0;
1627 if (patch->is_delete < 0 &&
1628 (newlines || (patch->fragments && patch->fragments->next)))
1629 patch->is_delete = 0;
1631 if (0 < patch->is_new && oldlines)
1632 die("new file %s depends on old contents", patch->new_name);
1633 if (0 < patch->is_delete && newlines)
1634 die("deleted file %s still has contents", patch->old_name);
1635 if (!patch->is_delete && !newlines && context)
1636 fprintf(stderr, "** warning: file %s becomes empty but "
1637 "is not deleted\n", patch->new_name);
1639 return offset;
1642 static inline int metadata_changes(struct patch *patch)
1644 return patch->is_rename > 0 ||
1645 patch->is_copy > 0 ||
1646 patch->is_new > 0 ||
1647 patch->is_delete ||
1648 (patch->old_mode && patch->new_mode &&
1649 patch->old_mode != patch->new_mode);
1652 static char *inflate_it(const void *data, unsigned long size,
1653 unsigned long inflated_size)
1655 git_zstream stream;
1656 void *out;
1657 int st;
1659 memset(&stream, 0, sizeof(stream));
1661 stream.next_in = (unsigned char *)data;
1662 stream.avail_in = size;
1663 stream.next_out = out = xmalloc(inflated_size);
1664 stream.avail_out = inflated_size;
1665 git_inflate_init(&stream);
1666 st = git_inflate(&stream, Z_FINISH);
1667 git_inflate_end(&stream);
1668 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1669 free(out);
1670 return NULL;
1672 return out;
1675 static struct fragment *parse_binary_hunk(char **buf_p,
1676 unsigned long *sz_p,
1677 int *status_p,
1678 int *used_p)
1681 * Expect a line that begins with binary patch method ("literal"
1682 * or "delta"), followed by the length of data before deflating.
1683 * a sequence of 'length-byte' followed by base-85 encoded data
1684 * should follow, terminated by a newline.
1686 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1687 * and we would limit the patch line to 66 characters,
1688 * so one line can fit up to 13 groups that would decode
1689 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1690 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1692 int llen, used;
1693 unsigned long size = *sz_p;
1694 char *buffer = *buf_p;
1695 int patch_method;
1696 unsigned long origlen;
1697 char *data = NULL;
1698 int hunk_size = 0;
1699 struct fragment *frag;
1701 llen = linelen(buffer, size);
1702 used = llen;
1704 *status_p = 0;
1706 if (!prefixcmp(buffer, "delta ")) {
1707 patch_method = BINARY_DELTA_DEFLATED;
1708 origlen = strtoul(buffer + 6, NULL, 10);
1710 else if (!prefixcmp(buffer, "literal ")) {
1711 patch_method = BINARY_LITERAL_DEFLATED;
1712 origlen = strtoul(buffer + 8, NULL, 10);
1714 else
1715 return NULL;
1717 linenr++;
1718 buffer += llen;
1719 while (1) {
1720 int byte_length, max_byte_length, newsize;
1721 llen = linelen(buffer, size);
1722 used += llen;
1723 linenr++;
1724 if (llen == 1) {
1725 /* consume the blank line */
1726 buffer++;
1727 size--;
1728 break;
1731 * Minimum line is "A00000\n" which is 7-byte long,
1732 * and the line length must be multiple of 5 plus 2.
1734 if ((llen < 7) || (llen-2) % 5)
1735 goto corrupt;
1736 max_byte_length = (llen - 2) / 5 * 4;
1737 byte_length = *buffer;
1738 if ('A' <= byte_length && byte_length <= 'Z')
1739 byte_length = byte_length - 'A' + 1;
1740 else if ('a' <= byte_length && byte_length <= 'z')
1741 byte_length = byte_length - 'a' + 27;
1742 else
1743 goto corrupt;
1744 /* if the input length was not multiple of 4, we would
1745 * have filler at the end but the filler should never
1746 * exceed 3 bytes
1748 if (max_byte_length < byte_length ||
1749 byte_length <= max_byte_length - 4)
1750 goto corrupt;
1751 newsize = hunk_size + byte_length;
1752 data = xrealloc(data, newsize);
1753 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1754 goto corrupt;
1755 hunk_size = newsize;
1756 buffer += llen;
1757 size -= llen;
1760 frag = xcalloc(1, sizeof(*frag));
1761 frag->patch = inflate_it(data, hunk_size, origlen);
1762 if (!frag->patch)
1763 goto corrupt;
1764 free(data);
1765 frag->size = origlen;
1766 *buf_p = buffer;
1767 *sz_p = size;
1768 *used_p = used;
1769 frag->binary_patch_method = patch_method;
1770 return frag;
1772 corrupt:
1773 free(data);
1774 *status_p = -1;
1775 error("corrupt binary patch at line %d: %.*s",
1776 linenr-1, llen-1, buffer);
1777 return NULL;
1780 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1783 * We have read "GIT binary patch\n"; what follows is a line
1784 * that says the patch method (currently, either "literal" or
1785 * "delta") and the length of data before deflating; a
1786 * sequence of 'length-byte' followed by base-85 encoded data
1787 * follows.
1789 * When a binary patch is reversible, there is another binary
1790 * hunk in the same format, starting with patch method (either
1791 * "literal" or "delta") with the length of data, and a sequence
1792 * of length-byte + base-85 encoded data, terminated with another
1793 * empty line. This data, when applied to the postimage, produces
1794 * the preimage.
1796 struct fragment *forward;
1797 struct fragment *reverse;
1798 int status;
1799 int used, used_1;
1801 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1802 if (!forward && !status)
1803 /* there has to be one hunk (forward hunk) */
1804 return error("unrecognized binary patch at line %d", linenr-1);
1805 if (status)
1806 /* otherwise we already gave an error message */
1807 return status;
1809 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1810 if (reverse)
1811 used += used_1;
1812 else if (status) {
1814 * Not having reverse hunk is not an error, but having
1815 * a corrupt reverse hunk is.
1817 free((void*) forward->patch);
1818 free(forward);
1819 return status;
1821 forward->next = reverse;
1822 patch->fragments = forward;
1823 patch->is_binary = 1;
1824 return used;
1827 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1829 int hdrsize, patchsize;
1830 int offset = find_header(buffer, size, &hdrsize, patch);
1832 if (offset < 0)
1833 return offset;
1835 patch->ws_rule = whitespace_rule(patch->new_name
1836 ? patch->new_name
1837 : patch->old_name);
1839 patchsize = parse_single_patch(buffer + offset + hdrsize,
1840 size - offset - hdrsize, patch);
1842 if (!patchsize) {
1843 static const char *binhdr[] = {
1844 "Binary files ",
1845 "Files ",
1846 NULL,
1848 static const char git_binary[] = "GIT binary patch\n";
1849 int i;
1850 int hd = hdrsize + offset;
1851 unsigned long llen = linelen(buffer + hd, size - hd);
1853 if (llen == sizeof(git_binary) - 1 &&
1854 !memcmp(git_binary, buffer + hd, llen)) {
1855 int used;
1856 linenr++;
1857 used = parse_binary(buffer + hd + llen,
1858 size - hd - llen, patch);
1859 if (used)
1860 patchsize = used + llen;
1861 else
1862 patchsize = 0;
1864 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1865 for (i = 0; binhdr[i]; i++) {
1866 int len = strlen(binhdr[i]);
1867 if (len < size - hd &&
1868 !memcmp(binhdr[i], buffer + hd, len)) {
1869 linenr++;
1870 patch->is_binary = 1;
1871 patchsize = llen;
1872 break;
1877 /* Empty patch cannot be applied if it is a text patch
1878 * without metadata change. A binary patch appears
1879 * empty to us here.
1881 if ((apply || check) &&
1882 (!patch->is_binary && !metadata_changes(patch)))
1883 die("patch with only garbage at line %d", linenr);
1886 return offset + hdrsize + patchsize;
1889 #define swap(a,b) myswap((a),(b),sizeof(a))
1891 #define myswap(a, b, size) do { \
1892 unsigned char mytmp[size]; \
1893 memcpy(mytmp, &a, size); \
1894 memcpy(&a, &b, size); \
1895 memcpy(&b, mytmp, size); \
1896 } while (0)
1898 static void reverse_patches(struct patch *p)
1900 for (; p; p = p->next) {
1901 struct fragment *frag = p->fragments;
1903 swap(p->new_name, p->old_name);
1904 swap(p->new_mode, p->old_mode);
1905 swap(p->is_new, p->is_delete);
1906 swap(p->lines_added, p->lines_deleted);
1907 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1909 for (; frag; frag = frag->next) {
1910 swap(frag->newpos, frag->oldpos);
1911 swap(frag->newlines, frag->oldlines);
1916 static const char pluses[] =
1917 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1918 static const char minuses[]=
1919 "----------------------------------------------------------------------";
1921 static void show_stats(struct patch *patch)
1923 struct strbuf qname = STRBUF_INIT;
1924 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1925 int max, add, del;
1927 quote_c_style(cp, &qname, NULL, 0);
1930 * "scale" the filename
1932 max = max_len;
1933 if (max > 50)
1934 max = 50;
1936 if (qname.len > max) {
1937 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1938 if (!cp)
1939 cp = qname.buf + qname.len + 3 - max;
1940 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1943 if (patch->is_binary) {
1944 printf(" %-*s | Bin\n", max, qname.buf);
1945 strbuf_release(&qname);
1946 return;
1949 printf(" %-*s |", max, qname.buf);
1950 strbuf_release(&qname);
1953 * scale the add/delete
1955 max = max + max_change > 70 ? 70 - max : max_change;
1956 add = patch->lines_added;
1957 del = patch->lines_deleted;
1959 if (max_change > 0) {
1960 int total = ((add + del) * max + max_change / 2) / max_change;
1961 add = (add * max + max_change / 2) / max_change;
1962 del = total - add;
1964 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1965 add, pluses, del, minuses);
1968 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1970 switch (st->st_mode & S_IFMT) {
1971 case S_IFLNK:
1972 if (strbuf_readlink(buf, path, st->st_size) < 0)
1973 return error("unable to read symlink %s", path);
1974 return 0;
1975 case S_IFREG:
1976 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1977 return error("unable to open or read %s", path);
1978 convert_to_git(path, buf->buf, buf->len, buf, 0);
1979 return 0;
1980 default:
1981 return -1;
1986 * Update the preimage, and the common lines in postimage,
1987 * from buffer buf of length len. If postlen is 0 the postimage
1988 * is updated in place, otherwise it's updated on a new buffer
1989 * of length postlen
1992 static void update_pre_post_images(struct image *preimage,
1993 struct image *postimage,
1994 char *buf,
1995 size_t len, size_t postlen)
1997 int i, ctx;
1998 char *new, *old, *fixed;
1999 struct image fixed_preimage;
2002 * Update the preimage with whitespace fixes. Note that we
2003 * are not losing preimage->buf -- apply_one_fragment() will
2004 * free "oldlines".
2006 prepare_image(&fixed_preimage, buf, len, 1);
2007 assert(fixed_preimage.nr == preimage->nr);
2008 for (i = 0; i < preimage->nr; i++)
2009 fixed_preimage.line[i].flag = preimage->line[i].flag;
2010 free(preimage->line_allocated);
2011 *preimage = fixed_preimage;
2014 * Adjust the common context lines in postimage. This can be
2015 * done in-place when we are just doing whitespace fixing,
2016 * which does not make the string grow, but needs a new buffer
2017 * when ignoring whitespace causes the update, since in this case
2018 * we could have e.g. tabs converted to multiple spaces.
2019 * We trust the caller to tell us if the update can be done
2020 * in place (postlen==0) or not.
2022 old = postimage->buf;
2023 if (postlen)
2024 new = postimage->buf = xmalloc(postlen);
2025 else
2026 new = old;
2027 fixed = preimage->buf;
2028 for (i = ctx = 0; i < postimage->nr; i++) {
2029 size_t len = postimage->line[i].len;
2030 if (!(postimage->line[i].flag & LINE_COMMON)) {
2031 /* an added line -- no counterparts in preimage */
2032 memmove(new, old, len);
2033 old += len;
2034 new += len;
2035 continue;
2038 /* a common context -- skip it in the original postimage */
2039 old += len;
2041 /* and find the corresponding one in the fixed preimage */
2042 while (ctx < preimage->nr &&
2043 !(preimage->line[ctx].flag & LINE_COMMON)) {
2044 fixed += preimage->line[ctx].len;
2045 ctx++;
2047 if (preimage->nr <= ctx)
2048 die("oops");
2050 /* and copy it in, while fixing the line length */
2051 len = preimage->line[ctx].len;
2052 memcpy(new, fixed, len);
2053 new += len;
2054 fixed += len;
2055 postimage->line[i].len = len;
2056 ctx++;
2059 /* Fix the length of the whole thing */
2060 postimage->len = new - postimage->buf;
2063 static int match_fragment(struct image *img,
2064 struct image *preimage,
2065 struct image *postimage,
2066 unsigned long try,
2067 int try_lno,
2068 unsigned ws_rule,
2069 int match_beginning, int match_end)
2071 int i;
2072 char *fixed_buf, *buf, *orig, *target;
2073 struct strbuf fixed;
2074 size_t fixed_len;
2075 int preimage_limit;
2077 if (preimage->nr + try_lno <= img->nr) {
2079 * The hunk falls within the boundaries of img.
2081 preimage_limit = preimage->nr;
2082 if (match_end && (preimage->nr + try_lno != img->nr))
2083 return 0;
2084 } else if (ws_error_action == correct_ws_error &&
2085 (ws_rule & WS_BLANK_AT_EOF)) {
2087 * This hunk extends beyond the end of img, and we are
2088 * removing blank lines at the end of the file. This
2089 * many lines from the beginning of the preimage must
2090 * match with img, and the remainder of the preimage
2091 * must be blank.
2093 preimage_limit = img->nr - try_lno;
2094 } else {
2096 * The hunk extends beyond the end of the img and
2097 * we are not removing blanks at the end, so we
2098 * should reject the hunk at this position.
2100 return 0;
2103 if (match_beginning && try_lno)
2104 return 0;
2106 /* Quick hash check */
2107 for (i = 0; i < preimage_limit; i++)
2108 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2109 (preimage->line[i].hash != img->line[try_lno + i].hash))
2110 return 0;
2112 if (preimage_limit == preimage->nr) {
2114 * Do we have an exact match? If we were told to match
2115 * at the end, size must be exactly at try+fragsize,
2116 * otherwise try+fragsize must be still within the preimage,
2117 * and either case, the old piece should match the preimage
2118 * exactly.
2120 if ((match_end
2121 ? (try + preimage->len == img->len)
2122 : (try + preimage->len <= img->len)) &&
2123 !memcmp(img->buf + try, preimage->buf, preimage->len))
2124 return 1;
2125 } else {
2127 * The preimage extends beyond the end of img, so
2128 * there cannot be an exact match.
2130 * There must be one non-blank context line that match
2131 * a line before the end of img.
2133 char *buf_end;
2135 buf = preimage->buf;
2136 buf_end = buf;
2137 for (i = 0; i < preimage_limit; i++)
2138 buf_end += preimage->line[i].len;
2140 for ( ; buf < buf_end; buf++)
2141 if (!isspace(*buf))
2142 break;
2143 if (buf == buf_end)
2144 return 0;
2148 * No exact match. If we are ignoring whitespace, run a line-by-line
2149 * fuzzy matching. We collect all the line length information because
2150 * we need it to adjust whitespace if we match.
2152 if (ws_ignore_action == ignore_ws_change) {
2153 size_t imgoff = 0;
2154 size_t preoff = 0;
2155 size_t postlen = postimage->len;
2156 size_t extra_chars;
2157 char *preimage_eof;
2158 char *preimage_end;
2159 for (i = 0; i < preimage_limit; i++) {
2160 size_t prelen = preimage->line[i].len;
2161 size_t imglen = img->line[try_lno+i].len;
2163 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2164 preimage->buf + preoff, prelen))
2165 return 0;
2166 if (preimage->line[i].flag & LINE_COMMON)
2167 postlen += imglen - prelen;
2168 imgoff += imglen;
2169 preoff += prelen;
2173 * Ok, the preimage matches with whitespace fuzz.
2175 * imgoff now holds the true length of the target that
2176 * matches the preimage before the end of the file.
2178 * Count the number of characters in the preimage that fall
2179 * beyond the end of the file and make sure that all of them
2180 * are whitespace characters. (This can only happen if
2181 * we are removing blank lines at the end of the file.)
2183 buf = preimage_eof = preimage->buf + preoff;
2184 for ( ; i < preimage->nr; i++)
2185 preoff += preimage->line[i].len;
2186 preimage_end = preimage->buf + preoff;
2187 for ( ; buf < preimage_end; buf++)
2188 if (!isspace(*buf))
2189 return 0;
2192 * Update the preimage and the common postimage context
2193 * lines to use the same whitespace as the target.
2194 * If whitespace is missing in the target (i.e.
2195 * if the preimage extends beyond the end of the file),
2196 * use the whitespace from the preimage.
2198 extra_chars = preimage_end - preimage_eof;
2199 strbuf_init(&fixed, imgoff + extra_chars);
2200 strbuf_add(&fixed, img->buf + try, imgoff);
2201 strbuf_add(&fixed, preimage_eof, extra_chars);
2202 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2203 update_pre_post_images(preimage, postimage,
2204 fixed_buf, fixed_len, postlen);
2205 return 1;
2208 if (ws_error_action != correct_ws_error)
2209 return 0;
2212 * The hunk does not apply byte-by-byte, but the hash says
2213 * it might with whitespace fuzz. We haven't been asked to
2214 * ignore whitespace, we were asked to correct whitespace
2215 * errors, so let's try matching after whitespace correction.
2217 * The preimage may extend beyond the end of the file,
2218 * but in this loop we will only handle the part of the
2219 * preimage that falls within the file.
2221 strbuf_init(&fixed, preimage->len + 1);
2222 orig = preimage->buf;
2223 target = img->buf + try;
2224 for (i = 0; i < preimage_limit; i++) {
2225 size_t oldlen = preimage->line[i].len;
2226 size_t tgtlen = img->line[try_lno + i].len;
2227 size_t fixstart = fixed.len;
2228 struct strbuf tgtfix;
2229 int match;
2231 /* Try fixing the line in the preimage */
2232 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2234 /* Try fixing the line in the target */
2235 strbuf_init(&tgtfix, tgtlen);
2236 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2239 * If they match, either the preimage was based on
2240 * a version before our tree fixed whitespace breakage,
2241 * or we are lacking a whitespace-fix patch the tree
2242 * the preimage was based on already had (i.e. target
2243 * has whitespace breakage, the preimage doesn't).
2244 * In either case, we are fixing the whitespace breakages
2245 * so we might as well take the fix together with their
2246 * real change.
2248 match = (tgtfix.len == fixed.len - fixstart &&
2249 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2250 fixed.len - fixstart));
2252 strbuf_release(&tgtfix);
2253 if (!match)
2254 goto unmatch_exit;
2256 orig += oldlen;
2257 target += tgtlen;
2262 * Now handle the lines in the preimage that falls beyond the
2263 * end of the file (if any). They will only match if they are
2264 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2265 * false).
2267 for ( ; i < preimage->nr; i++) {
2268 size_t fixstart = fixed.len; /* start of the fixed preimage */
2269 size_t oldlen = preimage->line[i].len;
2270 int j;
2272 /* Try fixing the line in the preimage */
2273 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2275 for (j = fixstart; j < fixed.len; j++)
2276 if (!isspace(fixed.buf[j]))
2277 goto unmatch_exit;
2279 orig += oldlen;
2283 * Yes, the preimage is based on an older version that still
2284 * has whitespace breakages unfixed, and fixing them makes the
2285 * hunk match. Update the context lines in the postimage.
2287 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2288 update_pre_post_images(preimage, postimage,
2289 fixed_buf, fixed_len, 0);
2290 return 1;
2292 unmatch_exit:
2293 strbuf_release(&fixed);
2294 return 0;
2297 static int find_pos(struct image *img,
2298 struct image *preimage,
2299 struct image *postimage,
2300 int line,
2301 unsigned ws_rule,
2302 int match_beginning, int match_end)
2304 int i;
2305 unsigned long backwards, forwards, try;
2306 int backwards_lno, forwards_lno, try_lno;
2309 * If match_beginning or match_end is specified, there is no
2310 * point starting from a wrong line that will never match and
2311 * wander around and wait for a match at the specified end.
2313 if (match_beginning)
2314 line = 0;
2315 else if (match_end)
2316 line = img->nr - preimage->nr;
2319 * Because the comparison is unsigned, the following test
2320 * will also take care of a negative line number that can
2321 * result when match_end and preimage is larger than the target.
2323 if ((size_t) line > img->nr)
2324 line = img->nr;
2326 try = 0;
2327 for (i = 0; i < line; i++)
2328 try += img->line[i].len;
2331 * There's probably some smart way to do this, but I'll leave
2332 * that to the smart and beautiful people. I'm simple and stupid.
2334 backwards = try;
2335 backwards_lno = line;
2336 forwards = try;
2337 forwards_lno = line;
2338 try_lno = line;
2340 for (i = 0; ; i++) {
2341 if (match_fragment(img, preimage, postimage,
2342 try, try_lno, ws_rule,
2343 match_beginning, match_end))
2344 return try_lno;
2346 again:
2347 if (backwards_lno == 0 && forwards_lno == img->nr)
2348 break;
2350 if (i & 1) {
2351 if (backwards_lno == 0) {
2352 i++;
2353 goto again;
2355 backwards_lno--;
2356 backwards -= img->line[backwards_lno].len;
2357 try = backwards;
2358 try_lno = backwards_lno;
2359 } else {
2360 if (forwards_lno == img->nr) {
2361 i++;
2362 goto again;
2364 forwards += img->line[forwards_lno].len;
2365 forwards_lno++;
2366 try = forwards;
2367 try_lno = forwards_lno;
2371 return -1;
2374 static void remove_first_line(struct image *img)
2376 img->buf += img->line[0].len;
2377 img->len -= img->line[0].len;
2378 img->line++;
2379 img->nr--;
2382 static void remove_last_line(struct image *img)
2384 img->len -= img->line[--img->nr].len;
2387 static void update_image(struct image *img,
2388 int applied_pos,
2389 struct image *preimage,
2390 struct image *postimage)
2393 * remove the copy of preimage at offset in img
2394 * and replace it with postimage
2396 int i, nr;
2397 size_t remove_count, insert_count, applied_at = 0;
2398 char *result;
2399 int preimage_limit;
2402 * If we are removing blank lines at the end of img,
2403 * the preimage may extend beyond the end.
2404 * If that is the case, we must be careful only to
2405 * remove the part of the preimage that falls within
2406 * the boundaries of img. Initialize preimage_limit
2407 * to the number of lines in the preimage that falls
2408 * within the boundaries.
2410 preimage_limit = preimage->nr;
2411 if (preimage_limit > img->nr - applied_pos)
2412 preimage_limit = img->nr - applied_pos;
2414 for (i = 0; i < applied_pos; i++)
2415 applied_at += img->line[i].len;
2417 remove_count = 0;
2418 for (i = 0; i < preimage_limit; i++)
2419 remove_count += img->line[applied_pos + i].len;
2420 insert_count = postimage->len;
2422 /* Adjust the contents */
2423 result = xmalloc(img->len + insert_count - remove_count + 1);
2424 memcpy(result, img->buf, applied_at);
2425 memcpy(result + applied_at, postimage->buf, postimage->len);
2426 memcpy(result + applied_at + postimage->len,
2427 img->buf + (applied_at + remove_count),
2428 img->len - (applied_at + remove_count));
2429 free(img->buf);
2430 img->buf = result;
2431 img->len += insert_count - remove_count;
2432 result[img->len] = '\0';
2434 /* Adjust the line table */
2435 nr = img->nr + postimage->nr - preimage_limit;
2436 if (preimage_limit < postimage->nr) {
2438 * NOTE: this knows that we never call remove_first_line()
2439 * on anything other than pre/post image.
2441 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2442 img->line_allocated = img->line;
2444 if (preimage_limit != postimage->nr)
2445 memmove(img->line + applied_pos + postimage->nr,
2446 img->line + applied_pos + preimage_limit,
2447 (img->nr - (applied_pos + preimage_limit)) *
2448 sizeof(*img->line));
2449 memcpy(img->line + applied_pos,
2450 postimage->line,
2451 postimage->nr * sizeof(*img->line));
2452 if (!allow_overlap)
2453 for (i = 0; i < postimage->nr; i++)
2454 img->line[applied_pos + i].flag |= LINE_PATCHED;
2455 img->nr = nr;
2458 static int apply_one_fragment(struct image *img, struct fragment *frag,
2459 int inaccurate_eof, unsigned ws_rule,
2460 int nth_fragment)
2462 int match_beginning, match_end;
2463 const char *patch = frag->patch;
2464 int size = frag->size;
2465 char *old, *oldlines;
2466 struct strbuf newlines;
2467 int new_blank_lines_at_end = 0;
2468 int found_new_blank_lines_at_end = 0;
2469 int hunk_linenr = frag->linenr;
2470 unsigned long leading, trailing;
2471 int pos, applied_pos;
2472 struct image preimage;
2473 struct image postimage;
2475 memset(&preimage, 0, sizeof(preimage));
2476 memset(&postimage, 0, sizeof(postimage));
2477 oldlines = xmalloc(size);
2478 strbuf_init(&newlines, size);
2480 old = oldlines;
2481 while (size > 0) {
2482 char first;
2483 int len = linelen(patch, size);
2484 int plen;
2485 int added_blank_line = 0;
2486 int is_blank_context = 0;
2487 size_t start;
2489 if (!len)
2490 break;
2493 * "plen" is how much of the line we should use for
2494 * the actual patch data. Normally we just remove the
2495 * first character on the line, but if the line is
2496 * followed by "\ No newline", then we also remove the
2497 * last one (which is the newline, of course).
2499 plen = len - 1;
2500 if (len < size && patch[len] == '\\')
2501 plen--;
2502 first = *patch;
2503 if (apply_in_reverse) {
2504 if (first == '-')
2505 first = '+';
2506 else if (first == '+')
2507 first = '-';
2510 switch (first) {
2511 case '\n':
2512 /* Newer GNU diff, empty context line */
2513 if (plen < 0)
2514 /* ... followed by '\No newline'; nothing */
2515 break;
2516 *old++ = '\n';
2517 strbuf_addch(&newlines, '\n');
2518 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2519 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2520 is_blank_context = 1;
2521 break;
2522 case ' ':
2523 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2524 ws_blank_line(patch + 1, plen, ws_rule))
2525 is_blank_context = 1;
2526 case '-':
2527 memcpy(old, patch + 1, plen);
2528 add_line_info(&preimage, old, plen,
2529 (first == ' ' ? LINE_COMMON : 0));
2530 old += plen;
2531 if (first == '-')
2532 break;
2533 /* Fall-through for ' ' */
2534 case '+':
2535 /* --no-add does not add new lines */
2536 if (first == '+' && no_add)
2537 break;
2539 start = newlines.len;
2540 if (first != '+' ||
2541 !whitespace_error ||
2542 ws_error_action != correct_ws_error) {
2543 strbuf_add(&newlines, patch + 1, plen);
2545 else {
2546 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2548 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2549 (first == '+' ? 0 : LINE_COMMON));
2550 if (first == '+' &&
2551 (ws_rule & WS_BLANK_AT_EOF) &&
2552 ws_blank_line(patch + 1, plen, ws_rule))
2553 added_blank_line = 1;
2554 break;
2555 case '@': case '\\':
2556 /* Ignore it, we already handled it */
2557 break;
2558 default:
2559 if (apply_verbosely)
2560 error("invalid start of line: '%c'", first);
2561 return -1;
2563 if (added_blank_line) {
2564 if (!new_blank_lines_at_end)
2565 found_new_blank_lines_at_end = hunk_linenr;
2566 new_blank_lines_at_end++;
2568 else if (is_blank_context)
2570 else
2571 new_blank_lines_at_end = 0;
2572 patch += len;
2573 size -= len;
2574 hunk_linenr++;
2576 if (inaccurate_eof &&
2577 old > oldlines && old[-1] == '\n' &&
2578 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2579 old--;
2580 strbuf_setlen(&newlines, newlines.len - 1);
2583 leading = frag->leading;
2584 trailing = frag->trailing;
2587 * A hunk to change lines at the beginning would begin with
2588 * @@ -1,L +N,M @@
2589 * but we need to be careful. -U0 that inserts before the second
2590 * line also has this pattern.
2592 * And a hunk to add to an empty file would begin with
2593 * @@ -0,0 +N,M @@
2595 * In other words, a hunk that is (frag->oldpos <= 1) with or
2596 * without leading context must match at the beginning.
2598 match_beginning = (!frag->oldpos ||
2599 (frag->oldpos == 1 && !unidiff_zero));
2602 * A hunk without trailing lines must match at the end.
2603 * However, we simply cannot tell if a hunk must match end
2604 * from the lack of trailing lines if the patch was generated
2605 * with unidiff without any context.
2607 match_end = !unidiff_zero && !trailing;
2609 pos = frag->newpos ? (frag->newpos - 1) : 0;
2610 preimage.buf = oldlines;
2611 preimage.len = old - oldlines;
2612 postimage.buf = newlines.buf;
2613 postimage.len = newlines.len;
2614 preimage.line = preimage.line_allocated;
2615 postimage.line = postimage.line_allocated;
2617 for (;;) {
2619 applied_pos = find_pos(img, &preimage, &postimage, pos,
2620 ws_rule, match_beginning, match_end);
2622 if (applied_pos >= 0)
2623 break;
2625 /* Am I at my context limits? */
2626 if ((leading <= p_context) && (trailing <= p_context))
2627 break;
2628 if (match_beginning || match_end) {
2629 match_beginning = match_end = 0;
2630 continue;
2634 * Reduce the number of context lines; reduce both
2635 * leading and trailing if they are equal otherwise
2636 * just reduce the larger context.
2638 if (leading >= trailing) {
2639 remove_first_line(&preimage);
2640 remove_first_line(&postimage);
2641 pos--;
2642 leading--;
2644 if (trailing > leading) {
2645 remove_last_line(&preimage);
2646 remove_last_line(&postimage);
2647 trailing--;
2651 if (applied_pos >= 0) {
2652 if (new_blank_lines_at_end &&
2653 preimage.nr + applied_pos >= img->nr &&
2654 (ws_rule & WS_BLANK_AT_EOF) &&
2655 ws_error_action != nowarn_ws_error) {
2656 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2657 found_new_blank_lines_at_end);
2658 if (ws_error_action == correct_ws_error) {
2659 while (new_blank_lines_at_end--)
2660 remove_last_line(&postimage);
2663 * We would want to prevent write_out_results()
2664 * from taking place in apply_patch() that follows
2665 * the callchain led us here, which is:
2666 * apply_patch->check_patch_list->check_patch->
2667 * apply_data->apply_fragments->apply_one_fragment
2669 if (ws_error_action == die_on_ws_error)
2670 apply = 0;
2673 if (apply_verbosely && applied_pos != pos) {
2674 int offset = applied_pos - pos;
2675 if (apply_in_reverse)
2676 offset = 0 - offset;
2677 fprintf(stderr,
2678 "Hunk #%d succeeded at %d (offset %d lines).\n",
2679 nth_fragment, applied_pos + 1, offset);
2683 * Warn if it was necessary to reduce the number
2684 * of context lines.
2686 if ((leading != frag->leading) ||
2687 (trailing != frag->trailing))
2688 fprintf(stderr, "Context reduced to (%ld/%ld)"
2689 " to apply fragment at %d\n",
2690 leading, trailing, applied_pos+1);
2691 update_image(img, applied_pos, &preimage, &postimage);
2692 } else {
2693 if (apply_verbosely)
2694 error("while searching for:\n%.*s",
2695 (int)(old - oldlines), oldlines);
2698 free(oldlines);
2699 strbuf_release(&newlines);
2700 free(preimage.line_allocated);
2701 free(postimage.line_allocated);
2703 return (applied_pos < 0);
2706 static int apply_binary_fragment(struct image *img, struct patch *patch)
2708 struct fragment *fragment = patch->fragments;
2709 unsigned long len;
2710 void *dst;
2712 if (!fragment)
2713 return error("missing binary patch data for '%s'",
2714 patch->new_name ?
2715 patch->new_name :
2716 patch->old_name);
2718 /* Binary patch is irreversible without the optional second hunk */
2719 if (apply_in_reverse) {
2720 if (!fragment->next)
2721 return error("cannot reverse-apply a binary patch "
2722 "without the reverse hunk to '%s'",
2723 patch->new_name
2724 ? patch->new_name : patch->old_name);
2725 fragment = fragment->next;
2727 switch (fragment->binary_patch_method) {
2728 case BINARY_DELTA_DEFLATED:
2729 dst = patch_delta(img->buf, img->len, fragment->patch,
2730 fragment->size, &len);
2731 if (!dst)
2732 return -1;
2733 clear_image(img);
2734 img->buf = dst;
2735 img->len = len;
2736 return 0;
2737 case BINARY_LITERAL_DEFLATED:
2738 clear_image(img);
2739 img->len = fragment->size;
2740 img->buf = xmalloc(img->len+1);
2741 memcpy(img->buf, fragment->patch, img->len);
2742 img->buf[img->len] = '\0';
2743 return 0;
2745 return -1;
2748 static int apply_binary(struct image *img, struct patch *patch)
2750 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2751 unsigned char sha1[20];
2754 * For safety, we require patch index line to contain
2755 * full 40-byte textual SHA1 for old and new, at least for now.
2757 if (strlen(patch->old_sha1_prefix) != 40 ||
2758 strlen(patch->new_sha1_prefix) != 40 ||
2759 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2760 get_sha1_hex(patch->new_sha1_prefix, sha1))
2761 return error("cannot apply binary patch to '%s' "
2762 "without full index line", name);
2764 if (patch->old_name) {
2766 * See if the old one matches what the patch
2767 * applies to.
2769 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2770 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2771 return error("the patch applies to '%s' (%s), "
2772 "which does not match the "
2773 "current contents.",
2774 name, sha1_to_hex(sha1));
2776 else {
2777 /* Otherwise, the old one must be empty. */
2778 if (img->len)
2779 return error("the patch applies to an empty "
2780 "'%s' but it is not empty", name);
2783 get_sha1_hex(patch->new_sha1_prefix, sha1);
2784 if (is_null_sha1(sha1)) {
2785 clear_image(img);
2786 return 0; /* deletion patch */
2789 if (has_sha1_file(sha1)) {
2790 /* We already have the postimage */
2791 enum object_type type;
2792 unsigned long size;
2793 char *result;
2795 result = read_sha1_file(sha1, &type, &size);
2796 if (!result)
2797 return error("the necessary postimage %s for "
2798 "'%s' cannot be read",
2799 patch->new_sha1_prefix, name);
2800 clear_image(img);
2801 img->buf = result;
2802 img->len = size;
2803 } else {
2805 * We have verified buf matches the preimage;
2806 * apply the patch data to it, which is stored
2807 * in the patch->fragments->{patch,size}.
2809 if (apply_binary_fragment(img, patch))
2810 return error("binary patch does not apply to '%s'",
2811 name);
2813 /* verify that the result matches */
2814 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2815 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2816 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2817 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2820 return 0;
2823 static int apply_fragments(struct image *img, struct patch *patch)
2825 struct fragment *frag = patch->fragments;
2826 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2827 unsigned ws_rule = patch->ws_rule;
2828 unsigned inaccurate_eof = patch->inaccurate_eof;
2829 int nth = 0;
2831 if (patch->is_binary)
2832 return apply_binary(img, patch);
2834 while (frag) {
2835 nth++;
2836 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
2837 error("patch failed: %s:%ld", name, frag->oldpos);
2838 if (!apply_with_reject)
2839 return -1;
2840 frag->rejected = 1;
2842 frag = frag->next;
2844 return 0;
2847 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2849 if (!ce)
2850 return 0;
2852 if (S_ISGITLINK(ce->ce_mode)) {
2853 strbuf_grow(buf, 100);
2854 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2855 } else {
2856 enum object_type type;
2857 unsigned long sz;
2858 char *result;
2860 result = read_sha1_file(ce->sha1, &type, &sz);
2861 if (!result)
2862 return -1;
2863 /* XXX read_sha1_file NUL-terminates */
2864 strbuf_attach(buf, result, sz, sz + 1);
2866 return 0;
2869 static struct patch *in_fn_table(const char *name)
2871 struct string_list_item *item;
2873 if (name == NULL)
2874 return NULL;
2876 item = string_list_lookup(&fn_table, name);
2877 if (item != NULL)
2878 return (struct patch *)item->util;
2880 return NULL;
2884 * item->util in the filename table records the status of the path.
2885 * Usually it points at a patch (whose result records the contents
2886 * of it after applying it), but it could be PATH_WAS_DELETED for a
2887 * path that a previously applied patch has already removed.
2889 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2890 #define PATH_WAS_DELETED ((struct patch *) -1)
2892 static int to_be_deleted(struct patch *patch)
2894 return patch == PATH_TO_BE_DELETED;
2897 static int was_deleted(struct patch *patch)
2899 return patch == PATH_WAS_DELETED;
2902 static void add_to_fn_table(struct patch *patch)
2904 struct string_list_item *item;
2907 * Always add new_name unless patch is a deletion
2908 * This should cover the cases for normal diffs,
2909 * file creations and copies
2911 if (patch->new_name != NULL) {
2912 item = string_list_insert(&fn_table, patch->new_name);
2913 item->util = patch;
2917 * store a failure on rename/deletion cases because
2918 * later chunks shouldn't patch old names
2920 if ((patch->new_name == NULL) || (patch->is_rename)) {
2921 item = string_list_insert(&fn_table, patch->old_name);
2922 item->util = PATH_WAS_DELETED;
2926 static void prepare_fn_table(struct patch *patch)
2929 * store information about incoming file deletion
2931 while (patch) {
2932 if ((patch->new_name == NULL) || (patch->is_rename)) {
2933 struct string_list_item *item;
2934 item = string_list_insert(&fn_table, patch->old_name);
2935 item->util = PATH_TO_BE_DELETED;
2937 patch = patch->next;
2941 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2943 struct strbuf buf = STRBUF_INIT;
2944 struct image image;
2945 size_t len;
2946 char *img;
2947 struct patch *tpatch;
2949 if (!(patch->is_copy || patch->is_rename) &&
2950 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2951 if (was_deleted(tpatch)) {
2952 return error("patch %s has been renamed/deleted",
2953 patch->old_name);
2955 /* We have a patched copy in memory use that */
2956 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2957 } else if (cached) {
2958 if (read_file_or_gitlink(ce, &buf))
2959 return error("read of %s failed", patch->old_name);
2960 } else if (patch->old_name) {
2961 if (S_ISGITLINK(patch->old_mode)) {
2962 if (ce) {
2963 read_file_or_gitlink(ce, &buf);
2964 } else {
2966 * There is no way to apply subproject
2967 * patch without looking at the index.
2969 patch->fragments = NULL;
2971 } else {
2972 if (read_old_data(st, patch->old_name, &buf))
2973 return error("read of %s failed", patch->old_name);
2977 img = strbuf_detach(&buf, &len);
2978 prepare_image(&image, img, len, !patch->is_binary);
2980 if (apply_fragments(&image, patch) < 0)
2981 return -1; /* note with --reject this succeeds. */
2982 patch->result = image.buf;
2983 patch->resultsize = image.len;
2984 add_to_fn_table(patch);
2985 free(image.line_allocated);
2987 if (0 < patch->is_delete && patch->resultsize)
2988 return error("removal patch leaves file contents");
2990 return 0;
2993 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2995 struct stat nst;
2996 if (!lstat(new_name, &nst)) {
2997 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2998 return 0;
3000 * A leading component of new_name might be a symlink
3001 * that is going to be removed with this patch, but
3002 * still pointing at somewhere that has the path.
3003 * In such a case, path "new_name" does not exist as
3004 * far as git is concerned.
3006 if (has_symlink_leading_path(new_name, strlen(new_name)))
3007 return 0;
3009 return error("%s: already exists in working directory", new_name);
3011 else if ((errno != ENOENT) && (errno != ENOTDIR))
3012 return error("%s: %s", new_name, strerror(errno));
3013 return 0;
3016 static int verify_index_match(struct cache_entry *ce, struct stat *st)
3018 if (S_ISGITLINK(ce->ce_mode)) {
3019 if (!S_ISDIR(st->st_mode))
3020 return -1;
3021 return 0;
3023 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3026 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3028 const char *old_name = patch->old_name;
3029 struct patch *tpatch = NULL;
3030 int stat_ret = 0;
3031 unsigned st_mode = 0;
3034 * Make sure that we do not have local modifications from the
3035 * index when we are looking at the index. Also make sure
3036 * we have the preimage file to be patched in the work tree,
3037 * unless --cached, which tells git to apply only in the index.
3039 if (!old_name)
3040 return 0;
3042 assert(patch->is_new <= 0);
3044 if (!(patch->is_copy || patch->is_rename) &&
3045 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
3046 if (was_deleted(tpatch))
3047 return error("%s: has been deleted/renamed", old_name);
3048 st_mode = tpatch->new_mode;
3049 } else if (!cached) {
3050 stat_ret = lstat(old_name, st);
3051 if (stat_ret && errno != ENOENT)
3052 return error("%s: %s", old_name, strerror(errno));
3055 if (to_be_deleted(tpatch))
3056 tpatch = NULL;
3058 if (check_index && !tpatch) {
3059 int pos = cache_name_pos(old_name, strlen(old_name));
3060 if (pos < 0) {
3061 if (patch->is_new < 0)
3062 goto is_new;
3063 return error("%s: does not exist in index", old_name);
3065 *ce = active_cache[pos];
3066 if (stat_ret < 0) {
3067 struct checkout costate;
3068 /* checkout */
3069 memset(&costate, 0, sizeof(costate));
3070 costate.base_dir = "";
3071 costate.refresh_cache = 1;
3072 if (checkout_entry(*ce, &costate, NULL) ||
3073 lstat(old_name, st))
3074 return -1;
3076 if (!cached && verify_index_match(*ce, st))
3077 return error("%s: does not match index", old_name);
3078 if (cached)
3079 st_mode = (*ce)->ce_mode;
3080 } else if (stat_ret < 0) {
3081 if (patch->is_new < 0)
3082 goto is_new;
3083 return error("%s: %s", old_name, strerror(errno));
3086 if (!cached && !tpatch)
3087 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3089 if (patch->is_new < 0)
3090 patch->is_new = 0;
3091 if (!patch->old_mode)
3092 patch->old_mode = st_mode;
3093 if ((st_mode ^ patch->old_mode) & S_IFMT)
3094 return error("%s: wrong type", old_name);
3095 if (st_mode != patch->old_mode)
3096 warning("%s has type %o, expected %o",
3097 old_name, st_mode, patch->old_mode);
3098 if (!patch->new_mode && !patch->is_delete)
3099 patch->new_mode = st_mode;
3100 return 0;
3102 is_new:
3103 patch->is_new = 1;
3104 patch->is_delete = 0;
3105 patch->old_name = NULL;
3106 return 0;
3109 static int check_patch(struct patch *patch)
3111 struct stat st;
3112 const char *old_name = patch->old_name;
3113 const char *new_name = patch->new_name;
3114 const char *name = old_name ? old_name : new_name;
3115 struct cache_entry *ce = NULL;
3116 struct patch *tpatch;
3117 int ok_if_exists;
3118 int status;
3120 patch->rejected = 1; /* we will drop this after we succeed */
3122 status = check_preimage(patch, &ce, &st);
3123 if (status)
3124 return status;
3125 old_name = patch->old_name;
3127 if ((tpatch = in_fn_table(new_name)) &&
3128 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3130 * A type-change diff is always split into a patch to
3131 * delete old, immediately followed by a patch to
3132 * create new (see diff.c::run_diff()); in such a case
3133 * it is Ok that the entry to be deleted by the
3134 * previous patch is still in the working tree and in
3135 * the index.
3137 ok_if_exists = 1;
3138 else
3139 ok_if_exists = 0;
3141 if (new_name &&
3142 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
3143 if (check_index &&
3144 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3145 !ok_if_exists)
3146 return error("%s: already exists in index", new_name);
3147 if (!cached) {
3148 int err = check_to_create_blob(new_name, ok_if_exists);
3149 if (err)
3150 return err;
3152 if (!patch->new_mode) {
3153 if (0 < patch->is_new)
3154 patch->new_mode = S_IFREG | 0644;
3155 else
3156 patch->new_mode = patch->old_mode;
3160 if (new_name && old_name) {
3161 int same = !strcmp(old_name, new_name);
3162 if (!patch->new_mode)
3163 patch->new_mode = patch->old_mode;
3164 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
3165 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
3166 patch->new_mode, new_name, patch->old_mode,
3167 same ? "" : " of ", same ? "" : old_name);
3170 if (apply_data(patch, &st, ce) < 0)
3171 return error("%s: patch does not apply", name);
3172 patch->rejected = 0;
3173 return 0;
3176 static int check_patch_list(struct patch *patch)
3178 int err = 0;
3180 prepare_fn_table(patch);
3181 while (patch) {
3182 if (apply_verbosely)
3183 say_patch_name(stderr,
3184 "Checking patch ", patch, "...\n");
3185 err |= check_patch(patch);
3186 patch = patch->next;
3188 return err;
3191 /* This function tries to read the sha1 from the current index */
3192 static int get_current_sha1(const char *path, unsigned char *sha1)
3194 int pos;
3196 if (read_cache() < 0)
3197 return -1;
3198 pos = cache_name_pos(path, strlen(path));
3199 if (pos < 0)
3200 return -1;
3201 hashcpy(sha1, active_cache[pos]->sha1);
3202 return 0;
3205 /* Build an index that contains the just the files needed for a 3way merge */
3206 static void build_fake_ancestor(struct patch *list, const char *filename)
3208 struct patch *patch;
3209 struct index_state result = { NULL };
3210 int fd;
3212 /* Once we start supporting the reverse patch, it may be
3213 * worth showing the new sha1 prefix, but until then...
3215 for (patch = list; patch; patch = patch->next) {
3216 const unsigned char *sha1_ptr;
3217 unsigned char sha1[20];
3218 struct cache_entry *ce;
3219 const char *name;
3221 name = patch->old_name ? patch->old_name : patch->new_name;
3222 if (0 < patch->is_new)
3223 continue;
3224 else if (get_sha1(patch->old_sha1_prefix, sha1))
3225 /* git diff has no index line for mode/type changes */
3226 if (!patch->lines_added && !patch->lines_deleted) {
3227 if (get_current_sha1(patch->old_name, sha1))
3228 die("mode change for %s, which is not "
3229 "in current HEAD", name);
3230 sha1_ptr = sha1;
3231 } else
3232 die("sha1 information is lacking or useless "
3233 "(%s).", name);
3234 else
3235 sha1_ptr = sha1;
3237 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
3238 if (!ce)
3239 die("make_cache_entry failed for path '%s'", name);
3240 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3241 die ("Could not add %s to temporary index", name);
3244 fd = open(filename, O_WRONLY | O_CREAT, 0666);
3245 if (fd < 0 || write_index(&result, fd) || close(fd))
3246 die ("Could not write temporary index to %s", filename);
3248 discard_index(&result);
3251 static void stat_patch_list(struct patch *patch)
3253 int files, adds, dels;
3255 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3256 files++;
3257 adds += patch->lines_added;
3258 dels += patch->lines_deleted;
3259 show_stats(patch);
3262 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
3265 static void numstat_patch_list(struct patch *patch)
3267 for ( ; patch; patch = patch->next) {
3268 const char *name;
3269 name = patch->new_name ? patch->new_name : patch->old_name;
3270 if (patch->is_binary)
3271 printf("-\t-\t");
3272 else
3273 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3274 write_name_quoted(name, stdout, line_termination);
3278 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3280 if (mode)
3281 printf(" %s mode %06o %s\n", newdelete, mode, name);
3282 else
3283 printf(" %s %s\n", newdelete, name);
3286 static void show_mode_change(struct patch *p, int show_name)
3288 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3289 if (show_name)
3290 printf(" mode change %06o => %06o %s\n",
3291 p->old_mode, p->new_mode, p->new_name);
3292 else
3293 printf(" mode change %06o => %06o\n",
3294 p->old_mode, p->new_mode);
3298 static void show_rename_copy(struct patch *p)
3300 const char *renamecopy = p->is_rename ? "rename" : "copy";
3301 const char *old, *new;
3303 /* Find common prefix */
3304 old = p->old_name;
3305 new = p->new_name;
3306 while (1) {
3307 const char *slash_old, *slash_new;
3308 slash_old = strchr(old, '/');
3309 slash_new = strchr(new, '/');
3310 if (!slash_old ||
3311 !slash_new ||
3312 slash_old - old != slash_new - new ||
3313 memcmp(old, new, slash_new - new))
3314 break;
3315 old = slash_old + 1;
3316 new = slash_new + 1;
3318 /* p->old_name thru old is the common prefix, and old and new
3319 * through the end of names are renames
3321 if (old != p->old_name)
3322 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3323 (int)(old - p->old_name), p->old_name,
3324 old, new, p->score);
3325 else
3326 printf(" %s %s => %s (%d%%)\n", renamecopy,
3327 p->old_name, p->new_name, p->score);
3328 show_mode_change(p, 0);
3331 static void summary_patch_list(struct patch *patch)
3333 struct patch *p;
3335 for (p = patch; p; p = p->next) {
3336 if (p->is_new)
3337 show_file_mode_name("create", p->new_mode, p->new_name);
3338 else if (p->is_delete)
3339 show_file_mode_name("delete", p->old_mode, p->old_name);
3340 else {
3341 if (p->is_rename || p->is_copy)
3342 show_rename_copy(p);
3343 else {
3344 if (p->score) {
3345 printf(" rewrite %s (%d%%)\n",
3346 p->new_name, p->score);
3347 show_mode_change(p, 0);
3349 else
3350 show_mode_change(p, 1);
3356 static void patch_stats(struct patch *patch)
3358 int lines = patch->lines_added + patch->lines_deleted;
3360 if (lines > max_change)
3361 max_change = lines;
3362 if (patch->old_name) {
3363 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3364 if (!len)
3365 len = strlen(patch->old_name);
3366 if (len > max_len)
3367 max_len = len;
3369 if (patch->new_name) {
3370 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3371 if (!len)
3372 len = strlen(patch->new_name);
3373 if (len > max_len)
3374 max_len = len;
3378 static void remove_file(struct patch *patch, int rmdir_empty)
3380 if (update_index) {
3381 if (remove_file_from_cache(patch->old_name) < 0)
3382 die("unable to remove %s from index", patch->old_name);
3384 if (!cached) {
3385 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3386 remove_path(patch->old_name);
3391 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3393 struct stat st;
3394 struct cache_entry *ce;
3395 int namelen = strlen(path);
3396 unsigned ce_size = cache_entry_size(namelen);
3398 if (!update_index)
3399 return;
3401 ce = xcalloc(1, ce_size);
3402 memcpy(ce->name, path, namelen);
3403 ce->ce_mode = create_ce_mode(mode);
3404 ce->ce_flags = namelen;
3405 if (S_ISGITLINK(mode)) {
3406 const char *s = buf;
3408 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3409 die("corrupt patch for subproject %s", path);
3410 } else {
3411 if (!cached) {
3412 if (lstat(path, &st) < 0)
3413 die_errno("unable to stat newly created file '%s'",
3414 path);
3415 fill_stat_cache_info(ce, &st);
3417 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3418 die("unable to create backing store for newly created file %s", path);
3420 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3421 die("unable to add cache entry for %s", path);
3424 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3426 int fd;
3427 struct strbuf nbuf = STRBUF_INIT;
3429 if (S_ISGITLINK(mode)) {
3430 struct stat st;
3431 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3432 return 0;
3433 return mkdir(path, 0777);
3436 if (has_symlinks && S_ISLNK(mode))
3437 /* Although buf:size is counted string, it also is NUL
3438 * terminated.
3440 return symlink(buf, path);
3442 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3443 if (fd < 0)
3444 return -1;
3446 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3447 size = nbuf.len;
3448 buf = nbuf.buf;
3450 write_or_die(fd, buf, size);
3451 strbuf_release(&nbuf);
3453 if (close(fd) < 0)
3454 die_errno("closing file '%s'", path);
3455 return 0;
3459 * We optimistically assume that the directories exist,
3460 * which is true 99% of the time anyway. If they don't,
3461 * we create them and try again.
3463 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3465 if (cached)
3466 return;
3467 if (!try_create_file(path, mode, buf, size))
3468 return;
3470 if (errno == ENOENT) {
3471 if (safe_create_leading_directories(path))
3472 return;
3473 if (!try_create_file(path, mode, buf, size))
3474 return;
3477 if (errno == EEXIST || errno == EACCES) {
3478 /* We may be trying to create a file where a directory
3479 * used to be.
3481 struct stat st;
3482 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3483 errno = EEXIST;
3486 if (errno == EEXIST) {
3487 unsigned int nr = getpid();
3489 for (;;) {
3490 char newpath[PATH_MAX];
3491 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3492 if (!try_create_file(newpath, mode, buf, size)) {
3493 if (!rename(newpath, path))
3494 return;
3495 unlink_or_warn(newpath);
3496 break;
3498 if (errno != EEXIST)
3499 break;
3500 ++nr;
3503 die_errno("unable to write file '%s' mode %o", path, mode);
3506 static void create_file(struct patch *patch)
3508 char *path = patch->new_name;
3509 unsigned mode = patch->new_mode;
3510 unsigned long size = patch->resultsize;
3511 char *buf = patch->result;
3513 if (!mode)
3514 mode = S_IFREG | 0644;
3515 create_one_file(path, mode, buf, size);
3516 add_index_file(path, mode, buf, size);
3519 /* phase zero is to remove, phase one is to create */
3520 static void write_out_one_result(struct patch *patch, int phase)
3522 if (patch->is_delete > 0) {
3523 if (phase == 0)
3524 remove_file(patch, 1);
3525 return;
3527 if (patch->is_new > 0 || patch->is_copy) {
3528 if (phase == 1)
3529 create_file(patch);
3530 return;
3533 * Rename or modification boils down to the same
3534 * thing: remove the old, write the new
3536 if (phase == 0)
3537 remove_file(patch, patch->is_rename);
3538 if (phase == 1)
3539 create_file(patch);
3542 static int write_out_one_reject(struct patch *patch)
3544 FILE *rej;
3545 char namebuf[PATH_MAX];
3546 struct fragment *frag;
3547 int cnt = 0;
3549 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3550 if (!frag->rejected)
3551 continue;
3552 cnt++;
3555 if (!cnt) {
3556 if (apply_verbosely)
3557 say_patch_name(stderr,
3558 "Applied patch ", patch, " cleanly.\n");
3559 return 0;
3562 /* This should not happen, because a removal patch that leaves
3563 * contents are marked "rejected" at the patch level.
3565 if (!patch->new_name)
3566 die("internal error");
3568 /* Say this even without --verbose */
3569 say_patch_name(stderr, "Applying patch ", patch, " with");
3570 fprintf(stderr, " %d rejects...\n", cnt);
3572 cnt = strlen(patch->new_name);
3573 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3574 cnt = ARRAY_SIZE(namebuf) - 5;
3575 warning("truncating .rej filename to %.*s.rej",
3576 cnt - 1, patch->new_name);
3578 memcpy(namebuf, patch->new_name, cnt);
3579 memcpy(namebuf + cnt, ".rej", 5);
3581 rej = fopen(namebuf, "w");
3582 if (!rej)
3583 return error("cannot open %s: %s", namebuf, strerror(errno));
3585 /* Normal git tools never deal with .rej, so do not pretend
3586 * this is a git patch by saying --git nor give extended
3587 * headers. While at it, maybe please "kompare" that wants
3588 * the trailing TAB and some garbage at the end of line ;-).
3590 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3591 patch->new_name, patch->new_name);
3592 for (cnt = 1, frag = patch->fragments;
3593 frag;
3594 cnt++, frag = frag->next) {
3595 if (!frag->rejected) {
3596 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3597 continue;
3599 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3600 fprintf(rej, "%.*s", frag->size, frag->patch);
3601 if (frag->patch[frag->size-1] != '\n')
3602 fputc('\n', rej);
3604 fclose(rej);
3605 return -1;
3608 static int write_out_results(struct patch *list)
3610 int phase;
3611 int errs = 0;
3612 struct patch *l;
3614 for (phase = 0; phase < 2; phase++) {
3615 l = list;
3616 while (l) {
3617 if (l->rejected)
3618 errs = 1;
3619 else {
3620 write_out_one_result(l, phase);
3621 if (phase == 1 && write_out_one_reject(l))
3622 errs = 1;
3624 l = l->next;
3627 return errs;
3630 static struct lock_file lock_file;
3632 static struct string_list limit_by_name;
3633 static int has_include;
3634 static void add_name_limit(const char *name, int exclude)
3636 struct string_list_item *it;
3638 it = string_list_append(&limit_by_name, name);
3639 it->util = exclude ? NULL : (void *) 1;
3642 static int use_patch(struct patch *p)
3644 const char *pathname = p->new_name ? p->new_name : p->old_name;
3645 int i;
3647 /* Paths outside are not touched regardless of "--include" */
3648 if (0 < prefix_length) {
3649 int pathlen = strlen(pathname);
3650 if (pathlen <= prefix_length ||
3651 memcmp(prefix, pathname, prefix_length))
3652 return 0;
3655 /* See if it matches any of exclude/include rule */
3656 for (i = 0; i < limit_by_name.nr; i++) {
3657 struct string_list_item *it = &limit_by_name.items[i];
3658 if (!fnmatch(it->string, pathname, 0))
3659 return (it->util != NULL);
3663 * If we had any include, a path that does not match any rule is
3664 * not used. Otherwise, we saw bunch of exclude rules (or none)
3665 * and such a path is used.
3667 return !has_include;
3671 static void prefix_one(char **name)
3673 char *old_name = *name;
3674 if (!old_name)
3675 return;
3676 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3677 free(old_name);
3680 static void prefix_patches(struct patch *p)
3682 if (!prefix || p->is_toplevel_relative)
3683 return;
3684 for ( ; p; p = p->next) {
3685 if (p->new_name == p->old_name) {
3686 char *prefixed = p->new_name;
3687 prefix_one(&prefixed);
3688 p->new_name = p->old_name = prefixed;
3690 else {
3691 prefix_one(&p->new_name);
3692 prefix_one(&p->old_name);
3697 #define INACCURATE_EOF (1<<0)
3698 #define RECOUNT (1<<1)
3700 static int apply_patch(int fd, const char *filename, int options)
3702 size_t offset;
3703 struct strbuf buf = STRBUF_INIT;
3704 struct patch *list = NULL, **listp = &list;
3705 int skipped_patch = 0;
3707 /* FIXME - memory leak when using multiple patch files as inputs */
3708 memset(&fn_table, 0, sizeof(struct string_list));
3709 patch_input_file = filename;
3710 read_patch_file(&buf, fd);
3711 offset = 0;
3712 while (offset < buf.len) {
3713 struct patch *patch;
3714 int nr;
3716 patch = xcalloc(1, sizeof(*patch));
3717 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3718 patch->recount = !!(options & RECOUNT);
3719 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3720 if (nr < 0)
3721 break;
3722 if (apply_in_reverse)
3723 reverse_patches(patch);
3724 if (prefix)
3725 prefix_patches(patch);
3726 if (use_patch(patch)) {
3727 patch_stats(patch);
3728 *listp = patch;
3729 listp = &patch->next;
3731 else {
3732 /* perhaps free it a bit better? */
3733 free(patch);
3734 skipped_patch++;
3736 offset += nr;
3739 if (!list && !skipped_patch)
3740 die("unrecognized input");
3742 if (whitespace_error && (ws_error_action == die_on_ws_error))
3743 apply = 0;
3745 update_index = check_index && apply;
3746 if (update_index && newfd < 0)
3747 newfd = hold_locked_index(&lock_file, 1);
3749 if (check_index) {
3750 if (read_cache() < 0)
3751 die("unable to read index file");
3754 if ((check || apply) &&
3755 check_patch_list(list) < 0 &&
3756 !apply_with_reject)
3757 exit(1);
3759 if (apply && write_out_results(list))
3760 exit(1);
3762 if (fake_ancestor)
3763 build_fake_ancestor(list, fake_ancestor);
3765 if (diffstat)
3766 stat_patch_list(list);
3768 if (numstat)
3769 numstat_patch_list(list);
3771 if (summary)
3772 summary_patch_list(list);
3774 strbuf_release(&buf);
3775 return 0;
3778 static int git_apply_config(const char *var, const char *value, void *cb)
3780 if (!strcmp(var, "apply.whitespace"))
3781 return git_config_string(&apply_default_whitespace, var, value);
3782 else if (!strcmp(var, "apply.ignorewhitespace"))
3783 return git_config_string(&apply_default_ignorewhitespace, var, value);
3784 return git_default_config(var, value, cb);
3787 static int option_parse_exclude(const struct option *opt,
3788 const char *arg, int unset)
3790 add_name_limit(arg, 1);
3791 return 0;
3794 static int option_parse_include(const struct option *opt,
3795 const char *arg, int unset)
3797 add_name_limit(arg, 0);
3798 has_include = 1;
3799 return 0;
3802 static int option_parse_p(const struct option *opt,
3803 const char *arg, int unset)
3805 p_value = atoi(arg);
3806 p_value_known = 1;
3807 return 0;
3810 static int option_parse_z(const struct option *opt,
3811 const char *arg, int unset)
3813 if (unset)
3814 line_termination = '\n';
3815 else
3816 line_termination = 0;
3817 return 0;
3820 static int option_parse_space_change(const struct option *opt,
3821 const char *arg, int unset)
3823 if (unset)
3824 ws_ignore_action = ignore_ws_none;
3825 else
3826 ws_ignore_action = ignore_ws_change;
3827 return 0;
3830 static int option_parse_whitespace(const struct option *opt,
3831 const char *arg, int unset)
3833 const char **whitespace_option = opt->value;
3835 *whitespace_option = arg;
3836 parse_whitespace_option(arg);
3837 return 0;
3840 static int option_parse_directory(const struct option *opt,
3841 const char *arg, int unset)
3843 root_len = strlen(arg);
3844 if (root_len && arg[root_len - 1] != '/') {
3845 char *new_root;
3846 root = new_root = xmalloc(root_len + 2);
3847 strcpy(new_root, arg);
3848 strcpy(new_root + root_len++, "/");
3849 } else
3850 root = arg;
3851 return 0;
3854 int cmd_apply(int argc, const char **argv, const char *prefix_)
3856 int i;
3857 int errs = 0;
3858 int is_not_gitdir = !startup_info->have_repository;
3859 int force_apply = 0;
3861 const char *whitespace_option = NULL;
3863 struct option builtin_apply_options[] = {
3864 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3865 "don't apply changes matching the given path",
3866 0, option_parse_exclude },
3867 { OPTION_CALLBACK, 0, "include", NULL, "path",
3868 "apply changes matching the given path",
3869 0, option_parse_include },
3870 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3871 "remove <num> leading slashes from traditional diff paths",
3872 0, option_parse_p },
3873 OPT_BOOLEAN(0, "no-add", &no_add,
3874 "ignore additions made by the patch"),
3875 OPT_BOOLEAN(0, "stat", &diffstat,
3876 "instead of applying the patch, output diffstat for the input"),
3877 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
3878 OPT_NOOP_NOARG(0, "binary"),
3879 OPT_BOOLEAN(0, "numstat", &numstat,
3880 "shows number of added and deleted lines in decimal notation"),
3881 OPT_BOOLEAN(0, "summary", &summary,
3882 "instead of applying the patch, output a summary for the input"),
3883 OPT_BOOLEAN(0, "check", &check,
3884 "instead of applying the patch, see if the patch is applicable"),
3885 OPT_BOOLEAN(0, "index", &check_index,
3886 "make sure the patch is applicable to the current index"),
3887 OPT_BOOLEAN(0, "cached", &cached,
3888 "apply a patch without touching the working tree"),
3889 OPT_BOOLEAN(0, "apply", &force_apply,
3890 "also apply the patch (use with --stat/--summary/--check)"),
3891 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3892 "build a temporary index based on embedded index information"),
3893 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3894 "paths are separated with NUL character",
3895 PARSE_OPT_NOARG, option_parse_z },
3896 OPT_INTEGER('C', NULL, &p_context,
3897 "ensure at least <n> lines of context match"),
3898 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3899 "detect new or modified lines that have whitespace errors",
3900 0, option_parse_whitespace },
3901 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3902 "ignore changes in whitespace when finding context",
3903 PARSE_OPT_NOARG, option_parse_space_change },
3904 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3905 "ignore changes in whitespace when finding context",
3906 PARSE_OPT_NOARG, option_parse_space_change },
3907 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3908 "apply the patch in reverse"),
3909 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3910 "don't expect at least one line of context"),
3911 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3912 "leave the rejected hunks in corresponding *.rej files"),
3913 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,
3914 "allow overlapping hunks"),
3915 OPT__VERBOSE(&apply_verbosely, "be verbose"),
3916 OPT_BIT(0, "inaccurate-eof", &options,
3917 "tolerate incorrectly detected missing new-line at the end of file",
3918 INACCURATE_EOF),
3919 OPT_BIT(0, "recount", &options,
3920 "do not trust the line counts in the hunk headers",
3921 RECOUNT),
3922 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3923 "prepend <root> to all filenames",
3924 0, option_parse_directory },
3925 OPT_END()
3928 prefix = prefix_;
3929 prefix_length = prefix ? strlen(prefix) : 0;
3930 git_config(git_apply_config, NULL);
3931 if (apply_default_whitespace)
3932 parse_whitespace_option(apply_default_whitespace);
3933 if (apply_default_ignorewhitespace)
3934 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3936 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3937 apply_usage, 0);
3939 if (apply_with_reject)
3940 apply = apply_verbosely = 1;
3941 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3942 apply = 0;
3943 if (check_index && is_not_gitdir)
3944 die("--index outside a repository");
3945 if (cached) {
3946 if (is_not_gitdir)
3947 die("--cached outside a repository");
3948 check_index = 1;
3950 for (i = 0; i < argc; i++) {
3951 const char *arg = argv[i];
3952 int fd;
3954 if (!strcmp(arg, "-")) {
3955 errs |= apply_patch(0, "<stdin>", options);
3956 read_stdin = 0;
3957 continue;
3958 } else if (0 < prefix_length)
3959 arg = prefix_filename(prefix, prefix_length, arg);
3961 fd = open(arg, O_RDONLY);
3962 if (fd < 0)
3963 die_errno("can't open patch '%s'", arg);
3964 read_stdin = 0;
3965 set_default_whitespace_mode(whitespace_option);
3966 errs |= apply_patch(fd, arg, options);
3967 close(fd);
3969 set_default_whitespace_mode(whitespace_option);
3970 if (read_stdin)
3971 errs |= apply_patch(0, "<stdin>", options);
3972 if (whitespace_error) {
3973 if (squelch_whitespace_errors &&
3974 squelch_whitespace_errors < whitespace_error) {
3975 int squelched =
3976 whitespace_error - squelch_whitespace_errors;
3977 warning("squelched %d "
3978 "whitespace error%s",
3979 squelched,
3980 squelched == 1 ? "" : "s");
3982 if (ws_error_action == die_on_ws_error)
3983 die("%d line%s add%s whitespace errors.",
3984 whitespace_error,
3985 whitespace_error == 1 ? "" : "s",
3986 whitespace_error == 1 ? "s" : "");
3987 if (applied_after_fixing_ws && apply)
3988 warning("%d line%s applied after"
3989 " fixing whitespace errors.",
3990 applied_after_fixing_ws,
3991 applied_after_fixing_ws == 1 ? "" : "s");
3992 else if (whitespace_error)
3993 warning("%d line%s add%s whitespace errors.",
3994 whitespace_error,
3995 whitespace_error == 1 ? "" : "s",
3996 whitespace_error == 1 ? "s" : "");
3999 if (update_index) {
4000 if (write_cache(newfd, active_cache, active_nr) ||
4001 commit_locked_index(&lock_file))
4002 die("Unable to write new index file");
4005 return !!errs;