apply: reject a hunk that does not do anything
[alt-git.git] / builtin / apply.c
blob606eddd5f6dcd6b3982f9acc6eb6f93eee235649
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "builtin.h"
16 #include "string-list.h"
17 #include "dir.h"
18 #include "diff.h"
19 #include "parse-options.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "rerere.h"
25 * --check turns on checking that the working tree matches the
26 * files that are being modified, but doesn't apply the patch
27 * --stat does just a diffstat, and doesn't actually apply
28 * --numstat does numeric diffstat, and doesn't actually apply
29 * --index-info shows the old and new index info for paths if available.
30 * --index updates the cache as well.
31 * --cached updates only the cache without ever touching the working tree.
33 static const char *prefix;
34 static int prefix_length = -1;
35 static int newfd = -1;
37 static int unidiff_zero;
38 static int p_value = 1;
39 static int p_value_known;
40 static int check_index;
41 static int update_index;
42 static int cached;
43 static int diffstat;
44 static int numstat;
45 static int summary;
46 static int check;
47 static int apply = 1;
48 static int apply_in_reverse;
49 static int apply_with_reject;
50 static int apply_verbosely;
51 static int allow_overlap;
52 static int no_add;
53 static int threeway;
54 static const char *fake_ancestor;
55 static int line_termination = '\n';
56 static unsigned int p_context = UINT_MAX;
57 static const char * const apply_usage[] = {
58 N_("git apply [options] [<patch>...]"),
59 NULL
62 static enum ws_error_action {
63 nowarn_ws_error,
64 warn_on_ws_error,
65 die_on_ws_error,
66 correct_ws_error
67 } ws_error_action = warn_on_ws_error;
68 static int whitespace_error;
69 static int squelch_whitespace_errors = 5;
70 static int applied_after_fixing_ws;
72 static enum ws_ignore {
73 ignore_ws_none,
74 ignore_ws_change
75 } ws_ignore_action = ignore_ws_none;
78 static const char *patch_input_file;
79 static const char *root;
80 static int root_len;
81 static int read_stdin = 1;
82 static int options;
84 static void parse_whitespace_option(const char *option)
86 if (!option) {
87 ws_error_action = warn_on_ws_error;
88 return;
90 if (!strcmp(option, "warn")) {
91 ws_error_action = warn_on_ws_error;
92 return;
94 if (!strcmp(option, "nowarn")) {
95 ws_error_action = nowarn_ws_error;
96 return;
98 if (!strcmp(option, "error")) {
99 ws_error_action = die_on_ws_error;
100 return;
102 if (!strcmp(option, "error-all")) {
103 ws_error_action = die_on_ws_error;
104 squelch_whitespace_errors = 0;
105 return;
107 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
108 ws_error_action = correct_ws_error;
109 return;
111 die(_("unrecognized whitespace option '%s'"), option);
114 static void parse_ignorewhitespace_option(const char *option)
116 if (!option || !strcmp(option, "no") ||
117 !strcmp(option, "false") || !strcmp(option, "never") ||
118 !strcmp(option, "none")) {
119 ws_ignore_action = ignore_ws_none;
120 return;
122 if (!strcmp(option, "change")) {
123 ws_ignore_action = ignore_ws_change;
124 return;
126 die(_("unrecognized whitespace ignore option '%s'"), option);
129 static void set_default_whitespace_mode(const char *whitespace_option)
131 if (!whitespace_option && !apply_default_whitespace)
132 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
136 * For "diff-stat" like behaviour, we keep track of the biggest change
137 * we've seen, and the longest filename. That allows us to do simple
138 * scaling.
140 static int max_change, max_len;
143 * Various "current state", notably line numbers and what
144 * file (and how) we're patching right now.. The "is_xxxx"
145 * things are flags, where -1 means "don't know yet".
147 static int linenr = 1;
150 * This represents one "hunk" from a patch, starting with
151 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
152 * patch text is pointed at by patch, and its byte length
153 * is stored in size. leading and trailing are the number
154 * of context lines.
156 struct fragment {
157 unsigned long leading, trailing;
158 unsigned long oldpos, oldlines;
159 unsigned long newpos, newlines;
161 * 'patch' is usually borrowed from buf in apply_patch(),
162 * but some codepaths store an allocated buffer.
164 const char *patch;
165 unsigned free_patch:1,
166 rejected:1;
167 int size;
168 int linenr;
169 struct fragment *next;
173 * When dealing with a binary patch, we reuse "leading" field
174 * to store the type of the binary hunk, either deflated "delta"
175 * or deflated "literal".
177 #define binary_patch_method leading
178 #define BINARY_DELTA_DEFLATED 1
179 #define BINARY_LITERAL_DEFLATED 2
182 * This represents a "patch" to a file, both metainfo changes
183 * such as creation/deletion, filemode and content changes represented
184 * as a series of fragments.
186 struct patch {
187 char *new_name, *old_name, *def_name;
188 unsigned int old_mode, new_mode;
189 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
190 int rejected;
191 unsigned ws_rule;
192 int lines_added, lines_deleted;
193 int score;
194 unsigned int is_toplevel_relative:1;
195 unsigned int inaccurate_eof:1;
196 unsigned int is_binary:1;
197 unsigned int is_copy:1;
198 unsigned int is_rename:1;
199 unsigned int recount:1;
200 unsigned int conflicted_threeway:1;
201 unsigned int direct_to_threeway:1;
202 struct fragment *fragments;
203 char *result;
204 size_t resultsize;
205 char old_sha1_prefix[41];
206 char new_sha1_prefix[41];
207 struct patch *next;
209 /* three-way fallback result */
210 unsigned char threeway_stage[3][20];
213 static void free_fragment_list(struct fragment *list)
215 while (list) {
216 struct fragment *next = list->next;
217 if (list->free_patch)
218 free((char *)list->patch);
219 free(list);
220 list = next;
224 static void free_patch(struct patch *patch)
226 free_fragment_list(patch->fragments);
227 free(patch->def_name);
228 free(patch->old_name);
229 free(patch->new_name);
230 free(patch->result);
231 free(patch);
234 static void free_patch_list(struct patch *list)
236 while (list) {
237 struct patch *next = list->next;
238 free_patch(list);
239 list = next;
244 * A line in a file, len-bytes long (includes the terminating LF,
245 * except for an incomplete line at the end if the file ends with
246 * one), and its contents hashes to 'hash'.
248 struct line {
249 size_t len;
250 unsigned hash : 24;
251 unsigned flag : 8;
252 #define LINE_COMMON 1
253 #define LINE_PATCHED 2
257 * This represents a "file", which is an array of "lines".
259 struct image {
260 char *buf;
261 size_t len;
262 size_t nr;
263 size_t alloc;
264 struct line *line_allocated;
265 struct line *line;
269 * Records filenames that have been touched, in order to handle
270 * the case where more than one patches touch the same file.
273 static struct string_list fn_table;
275 static uint32_t hash_line(const char *cp, size_t len)
277 size_t i;
278 uint32_t h;
279 for (i = 0, h = 0; i < len; i++) {
280 if (!isspace(cp[i])) {
281 h = h * 3 + (cp[i] & 0xff);
284 return h;
288 * Compare lines s1 of length n1 and s2 of length n2, ignoring
289 * whitespace difference. Returns 1 if they match, 0 otherwise
291 static int fuzzy_matchlines(const char *s1, size_t n1,
292 const char *s2, size_t n2)
294 const char *last1 = s1 + n1 - 1;
295 const char *last2 = s2 + n2 - 1;
296 int result = 0;
298 /* ignore line endings */
299 while ((*last1 == '\r') || (*last1 == '\n'))
300 last1--;
301 while ((*last2 == '\r') || (*last2 == '\n'))
302 last2--;
304 /* skip leading whitespaces, if both begin with whitespace */
305 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
306 while (isspace(*s1) && (s1 <= last1))
307 s1++;
308 while (isspace(*s2) && (s2 <= last2))
309 s2++;
311 /* early return if both lines are empty */
312 if ((s1 > last1) && (s2 > last2))
313 return 1;
314 while (!result) {
315 result = *s1++ - *s2++;
317 * Skip whitespace inside. We check for whitespace on
318 * both buffers because we don't want "a b" to match
319 * "ab"
321 if (isspace(*s1) && isspace(*s2)) {
322 while (isspace(*s1) && s1 <= last1)
323 s1++;
324 while (isspace(*s2) && s2 <= last2)
325 s2++;
328 * If we reached the end on one side only,
329 * lines don't match
331 if (
332 ((s2 > last2) && (s1 <= last1)) ||
333 ((s1 > last1) && (s2 <= last2)))
334 return 0;
335 if ((s1 > last1) && (s2 > last2))
336 break;
339 return !result;
342 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
344 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
345 img->line_allocated[img->nr].len = len;
346 img->line_allocated[img->nr].hash = hash_line(bol, len);
347 img->line_allocated[img->nr].flag = flag;
348 img->nr++;
352 * "buf" has the file contents to be patched (read from various sources).
353 * attach it to "image" and add line-based index to it.
354 * "image" now owns the "buf".
356 static void prepare_image(struct image *image, char *buf, size_t len,
357 int prepare_linetable)
359 const char *cp, *ep;
361 memset(image, 0, sizeof(*image));
362 image->buf = buf;
363 image->len = len;
365 if (!prepare_linetable)
366 return;
368 ep = image->buf + image->len;
369 cp = image->buf;
370 while (cp < ep) {
371 const char *next;
372 for (next = cp; next < ep && *next != '\n'; next++)
374 if (next < ep)
375 next++;
376 add_line_info(image, cp, next - cp, 0);
377 cp = next;
379 image->line = image->line_allocated;
382 static void clear_image(struct image *image)
384 free(image->buf);
385 free(image->line_allocated);
386 memset(image, 0, sizeof(*image));
389 /* fmt must contain _one_ %s and no other substitution */
390 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
392 struct strbuf sb = STRBUF_INIT;
394 if (patch->old_name && patch->new_name &&
395 strcmp(patch->old_name, patch->new_name)) {
396 quote_c_style(patch->old_name, &sb, NULL, 0);
397 strbuf_addstr(&sb, " => ");
398 quote_c_style(patch->new_name, &sb, NULL, 0);
399 } else {
400 const char *n = patch->new_name;
401 if (!n)
402 n = patch->old_name;
403 quote_c_style(n, &sb, NULL, 0);
405 fprintf(output, fmt, sb.buf);
406 fputc('\n', output);
407 strbuf_release(&sb);
410 #define SLOP (16)
412 static void read_patch_file(struct strbuf *sb, int fd)
414 if (strbuf_read(sb, fd, 0) < 0)
415 die_errno("git apply: failed to read");
418 * Make sure that we have some slop in the buffer
419 * so that we can do speculative "memcmp" etc, and
420 * see to it that it is NUL-filled.
422 strbuf_grow(sb, SLOP);
423 memset(sb->buf + sb->len, 0, SLOP);
426 static unsigned long linelen(const char *buffer, unsigned long size)
428 unsigned long len = 0;
429 while (size--) {
430 len++;
431 if (*buffer++ == '\n')
432 break;
434 return len;
437 static int is_dev_null(const char *str)
439 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
442 #define TERM_SPACE 1
443 #define TERM_TAB 2
445 static int name_terminate(const char *name, int namelen, int c, int terminate)
447 if (c == ' ' && !(terminate & TERM_SPACE))
448 return 0;
449 if (c == '\t' && !(terminate & TERM_TAB))
450 return 0;
452 return 1;
455 /* remove double slashes to make --index work with such filenames */
456 static char *squash_slash(char *name)
458 int i = 0, j = 0;
460 if (!name)
461 return NULL;
463 while (name[i]) {
464 if ((name[j++] = name[i++]) == '/')
465 while (name[i] == '/')
466 i++;
468 name[j] = '\0';
469 return name;
472 static char *find_name_gnu(const char *line, const char *def, int p_value)
474 struct strbuf name = STRBUF_INIT;
475 char *cp;
478 * Proposed "new-style" GNU patch/diff format; see
479 * http://marc.info/?l=git&m=112927316408690&w=2
481 if (unquote_c_style(&name, line, NULL)) {
482 strbuf_release(&name);
483 return NULL;
486 for (cp = name.buf; p_value; p_value--) {
487 cp = strchr(cp, '/');
488 if (!cp) {
489 strbuf_release(&name);
490 return NULL;
492 cp++;
495 strbuf_remove(&name, 0, cp - name.buf);
496 if (root)
497 strbuf_insert(&name, 0, root, root_len);
498 return squash_slash(strbuf_detach(&name, NULL));
501 static size_t sane_tz_len(const char *line, size_t len)
503 const char *tz, *p;
505 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
506 return 0;
507 tz = line + len - strlen(" +0500");
509 if (tz[1] != '+' && tz[1] != '-')
510 return 0;
512 for (p = tz + 2; p != line + len; p++)
513 if (!isdigit(*p))
514 return 0;
516 return line + len - tz;
519 static size_t tz_with_colon_len(const char *line, size_t len)
521 const char *tz, *p;
523 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
524 return 0;
525 tz = line + len - strlen(" +08:00");
527 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
528 return 0;
529 p = tz + 2;
530 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
531 !isdigit(*p++) || !isdigit(*p++))
532 return 0;
534 return line + len - tz;
537 static size_t date_len(const char *line, size_t len)
539 const char *date, *p;
541 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
542 return 0;
543 p = date = line + len - strlen("72-02-05");
545 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
546 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
547 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
548 return 0;
550 if (date - line >= strlen("19") &&
551 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
552 date -= strlen("19");
554 return line + len - date;
557 static size_t short_time_len(const char *line, size_t len)
559 const char *time, *p;
561 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
562 return 0;
563 p = time = line + len - strlen(" 07:01:32");
565 /* Permit 1-digit hours? */
566 if (*p++ != ' ' ||
567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
568 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
569 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
570 return 0;
572 return line + len - time;
575 static size_t fractional_time_len(const char *line, size_t len)
577 const char *p;
578 size_t n;
580 /* Expected format: 19:41:17.620000023 */
581 if (!len || !isdigit(line[len - 1]))
582 return 0;
583 p = line + len - 1;
585 /* Fractional seconds. */
586 while (p > line && isdigit(*p))
587 p--;
588 if (*p != '.')
589 return 0;
591 /* Hours, minutes, and whole seconds. */
592 n = short_time_len(line, p - line);
593 if (!n)
594 return 0;
596 return line + len - p + n;
599 static size_t trailing_spaces_len(const char *line, size_t len)
601 const char *p;
603 /* Expected format: ' ' x (1 or more) */
604 if (!len || line[len - 1] != ' ')
605 return 0;
607 p = line + len;
608 while (p != line) {
609 p--;
610 if (*p != ' ')
611 return line + len - (p + 1);
614 /* All spaces! */
615 return len;
618 static size_t diff_timestamp_len(const char *line, size_t len)
620 const char *end = line + len;
621 size_t n;
624 * Posix: 2010-07-05 19:41:17
625 * GNU: 2010-07-05 19:41:17.620000023 -0500
628 if (!isdigit(end[-1]))
629 return 0;
631 n = sane_tz_len(line, end - line);
632 if (!n)
633 n = tz_with_colon_len(line, end - line);
634 end -= n;
636 n = short_time_len(line, end - line);
637 if (!n)
638 n = fractional_time_len(line, end - line);
639 end -= n;
641 n = date_len(line, end - line);
642 if (!n) /* No date. Too bad. */
643 return 0;
644 end -= n;
646 if (end == line) /* No space before date. */
647 return 0;
648 if (end[-1] == '\t') { /* Success! */
649 end--;
650 return line + len - end;
652 if (end[-1] != ' ') /* No space before date. */
653 return 0;
655 /* Whitespace damage. */
656 end -= trailing_spaces_len(line, end - line);
657 return line + len - end;
660 static char *null_strdup(const char *s)
662 return s ? xstrdup(s) : NULL;
665 static char *find_name_common(const char *line, const char *def,
666 int p_value, const char *end, int terminate)
668 int len;
669 const char *start = NULL;
671 if (p_value == 0)
672 start = line;
673 while (line != end) {
674 char c = *line;
676 if (!end && isspace(c)) {
677 if (c == '\n')
678 break;
679 if (name_terminate(start, line-start, c, terminate))
680 break;
682 line++;
683 if (c == '/' && !--p_value)
684 start = line;
686 if (!start)
687 return squash_slash(null_strdup(def));
688 len = line - start;
689 if (!len)
690 return squash_slash(null_strdup(def));
693 * Generally we prefer the shorter name, especially
694 * if the other one is just a variation of that with
695 * something else tacked on to the end (ie "file.orig"
696 * or "file~").
698 if (def) {
699 int deflen = strlen(def);
700 if (deflen < len && !strncmp(start, def, deflen))
701 return squash_slash(xstrdup(def));
704 if (root) {
705 char *ret = xmalloc(root_len + len + 1);
706 strcpy(ret, root);
707 memcpy(ret + root_len, start, len);
708 ret[root_len + len] = '\0';
709 return squash_slash(ret);
712 return squash_slash(xmemdupz(start, len));
715 static char *find_name(const char *line, char *def, int p_value, int terminate)
717 if (*line == '"') {
718 char *name = find_name_gnu(line, def, p_value);
719 if (name)
720 return name;
723 return find_name_common(line, def, p_value, NULL, terminate);
726 static char *find_name_traditional(const char *line, char *def, int p_value)
728 size_t len;
729 size_t date_len;
731 if (*line == '"') {
732 char *name = find_name_gnu(line, def, p_value);
733 if (name)
734 return name;
737 len = strchrnul(line, '\n') - line;
738 date_len = diff_timestamp_len(line, len);
739 if (!date_len)
740 return find_name_common(line, def, p_value, NULL, TERM_TAB);
741 len -= date_len;
743 return find_name_common(line, def, p_value, line + len, 0);
746 static int count_slashes(const char *cp)
748 int cnt = 0;
749 char ch;
751 while ((ch = *cp++))
752 if (ch == '/')
753 cnt++;
754 return cnt;
758 * Given the string after "--- " or "+++ ", guess the appropriate
759 * p_value for the given patch.
761 static int guess_p_value(const char *nameline)
763 char *name, *cp;
764 int val = -1;
766 if (is_dev_null(nameline))
767 return -1;
768 name = find_name_traditional(nameline, NULL, 0);
769 if (!name)
770 return -1;
771 cp = strchr(name, '/');
772 if (!cp)
773 val = 0;
774 else if (prefix) {
776 * Does it begin with "a/$our-prefix" and such? Then this is
777 * very likely to apply to our directory.
779 if (!strncmp(name, prefix, prefix_length))
780 val = count_slashes(prefix);
781 else {
782 cp++;
783 if (!strncmp(cp, prefix, prefix_length))
784 val = count_slashes(prefix) + 1;
787 free(name);
788 return val;
792 * Does the ---/+++ line has the POSIX timestamp after the last HT?
793 * GNU diff puts epoch there to signal a creation/deletion event. Is
794 * this such a timestamp?
796 static int has_epoch_timestamp(const char *nameline)
799 * We are only interested in epoch timestamp; any non-zero
800 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
801 * For the same reason, the date must be either 1969-12-31 or
802 * 1970-01-01, and the seconds part must be "00".
804 const char stamp_regexp[] =
805 "^(1969-12-31|1970-01-01)"
807 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
809 "([-+][0-2][0-9]:?[0-5][0-9])\n";
810 const char *timestamp = NULL, *cp, *colon;
811 static regex_t *stamp;
812 regmatch_t m[10];
813 int zoneoffset;
814 int hourminute;
815 int status;
817 for (cp = nameline; *cp != '\n'; cp++) {
818 if (*cp == '\t')
819 timestamp = cp + 1;
821 if (!timestamp)
822 return 0;
823 if (!stamp) {
824 stamp = xmalloc(sizeof(*stamp));
825 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
826 warning(_("Cannot prepare timestamp regexp %s"),
827 stamp_regexp);
828 return 0;
832 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
833 if (status) {
834 if (status != REG_NOMATCH)
835 warning(_("regexec returned %d for input: %s"),
836 status, timestamp);
837 return 0;
840 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
841 if (*colon == ':')
842 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
843 else
844 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
845 if (timestamp[m[3].rm_so] == '-')
846 zoneoffset = -zoneoffset;
849 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
850 * (west of GMT) or 1970-01-01 (east of GMT)
852 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
853 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
854 return 0;
856 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
857 strtol(timestamp + 14, NULL, 10) -
858 zoneoffset);
860 return ((zoneoffset < 0 && hourminute == 1440) ||
861 (0 <= zoneoffset && !hourminute));
865 * Get the name etc info from the ---/+++ lines of a traditional patch header
867 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
868 * files, we can happily check the index for a match, but for creating a
869 * new file we should try to match whatever "patch" does. I have no idea.
871 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
873 char *name;
875 first += 4; /* skip "--- " */
876 second += 4; /* skip "+++ " */
877 if (!p_value_known) {
878 int p, q;
879 p = guess_p_value(first);
880 q = guess_p_value(second);
881 if (p < 0) p = q;
882 if (0 <= p && p == q) {
883 p_value = p;
884 p_value_known = 1;
887 if (is_dev_null(first)) {
888 patch->is_new = 1;
889 patch->is_delete = 0;
890 name = find_name_traditional(second, NULL, p_value);
891 patch->new_name = name;
892 } else if (is_dev_null(second)) {
893 patch->is_new = 0;
894 patch->is_delete = 1;
895 name = find_name_traditional(first, NULL, p_value);
896 patch->old_name = name;
897 } else {
898 char *first_name;
899 first_name = find_name_traditional(first, NULL, p_value);
900 name = find_name_traditional(second, first_name, p_value);
901 free(first_name);
902 if (has_epoch_timestamp(first)) {
903 patch->is_new = 1;
904 patch->is_delete = 0;
905 patch->new_name = name;
906 } else if (has_epoch_timestamp(second)) {
907 patch->is_new = 0;
908 patch->is_delete = 1;
909 patch->old_name = name;
910 } else {
911 patch->old_name = name;
912 patch->new_name = null_strdup(name);
915 if (!name)
916 die(_("unable to find filename in patch at line %d"), linenr);
919 static int gitdiff_hdrend(const char *line, struct patch *patch)
921 return -1;
925 * We're anal about diff header consistency, to make
926 * sure that we don't end up having strange ambiguous
927 * patches floating around.
929 * As a result, gitdiff_{old|new}name() will check
930 * their names against any previous information, just
931 * to make sure..
933 #define DIFF_OLD_NAME 0
934 #define DIFF_NEW_NAME 1
936 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side)
938 if (!orig_name && !isnull)
939 return find_name(line, NULL, p_value, TERM_TAB);
941 if (orig_name) {
942 int len;
943 const char *name;
944 char *another;
945 name = orig_name;
946 len = strlen(name);
947 if (isnull)
948 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr);
949 another = find_name(line, NULL, p_value, TERM_TAB);
950 if (!another || memcmp(another, name, len + 1))
951 die((side == DIFF_NEW_NAME) ?
952 _("git apply: bad git-diff - inconsistent new filename on line %d") :
953 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr);
954 free(another);
955 return orig_name;
957 else {
958 /* expect "/dev/null" */
959 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
960 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr);
961 return NULL;
965 static int gitdiff_oldname(const char *line, struct patch *patch)
967 char *orig = patch->old_name;
968 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name,
969 DIFF_OLD_NAME);
970 if (orig != patch->old_name)
971 free(orig);
972 return 0;
975 static int gitdiff_newname(const char *line, struct patch *patch)
977 char *orig = patch->new_name;
978 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name,
979 DIFF_NEW_NAME);
980 if (orig != patch->new_name)
981 free(orig);
982 return 0;
985 static int gitdiff_oldmode(const char *line, struct patch *patch)
987 patch->old_mode = strtoul(line, NULL, 8);
988 return 0;
991 static int gitdiff_newmode(const char *line, struct patch *patch)
993 patch->new_mode = strtoul(line, NULL, 8);
994 return 0;
997 static int gitdiff_delete(const char *line, struct patch *patch)
999 patch->is_delete = 1;
1000 free(patch->old_name);
1001 patch->old_name = null_strdup(patch->def_name);
1002 return gitdiff_oldmode(line, patch);
1005 static int gitdiff_newfile(const char *line, struct patch *patch)
1007 patch->is_new = 1;
1008 free(patch->new_name);
1009 patch->new_name = null_strdup(patch->def_name);
1010 return gitdiff_newmode(line, patch);
1013 static int gitdiff_copysrc(const char *line, struct patch *patch)
1015 patch->is_copy = 1;
1016 free(patch->old_name);
1017 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1018 return 0;
1021 static int gitdiff_copydst(const char *line, struct patch *patch)
1023 patch->is_copy = 1;
1024 free(patch->new_name);
1025 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1026 return 0;
1029 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1031 patch->is_rename = 1;
1032 free(patch->old_name);
1033 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1034 return 0;
1037 static int gitdiff_renamedst(const char *line, struct patch *patch)
1039 patch->is_rename = 1;
1040 free(patch->new_name);
1041 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);
1042 return 0;
1045 static int gitdiff_similarity(const char *line, struct patch *patch)
1047 unsigned long val = strtoul(line, NULL, 10);
1048 if (val <= 100)
1049 patch->score = val;
1050 return 0;
1053 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1055 unsigned long val = strtoul(line, NULL, 10);
1056 if (val <= 100)
1057 patch->score = val;
1058 return 0;
1061 static int gitdiff_index(const char *line, struct patch *patch)
1064 * index line is N hexadecimal, "..", N hexadecimal,
1065 * and optional space with octal mode.
1067 const char *ptr, *eol;
1068 int len;
1070 ptr = strchr(line, '.');
1071 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1072 return 0;
1073 len = ptr - line;
1074 memcpy(patch->old_sha1_prefix, line, len);
1075 patch->old_sha1_prefix[len] = 0;
1077 line = ptr + 2;
1078 ptr = strchr(line, ' ');
1079 eol = strchrnul(line, '\n');
1081 if (!ptr || eol < ptr)
1082 ptr = eol;
1083 len = ptr - line;
1085 if (40 < len)
1086 return 0;
1087 memcpy(patch->new_sha1_prefix, line, len);
1088 patch->new_sha1_prefix[len] = 0;
1089 if (*ptr == ' ')
1090 patch->old_mode = strtoul(ptr+1, NULL, 8);
1091 return 0;
1095 * This is normal for a diff that doesn't change anything: we'll fall through
1096 * into the next diff. Tell the parser to break out.
1098 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1100 return -1;
1104 * Skip p_value leading components from "line"; as we do not accept
1105 * absolute paths, return NULL in that case.
1107 static const char *skip_tree_prefix(const char *line, int llen)
1109 int nslash;
1110 int i;
1112 if (!p_value)
1113 return (llen && line[0] == '/') ? NULL : line;
1115 nslash = p_value;
1116 for (i = 0; i < llen; i++) {
1117 int ch = line[i];
1118 if (ch == '/' && --nslash <= 0)
1119 return (i == 0) ? NULL : &line[i + 1];
1121 return NULL;
1125 * This is to extract the same name that appears on "diff --git"
1126 * line. We do not find and return anything if it is a rename
1127 * patch, and it is OK because we will find the name elsewhere.
1128 * We need to reliably find name only when it is mode-change only,
1129 * creation or deletion of an empty file. In any of these cases,
1130 * both sides are the same name under a/ and b/ respectively.
1132 static char *git_header_name(const char *line, int llen)
1134 const char *name;
1135 const char *second = NULL;
1136 size_t len, line_len;
1138 line += strlen("diff --git ");
1139 llen -= strlen("diff --git ");
1141 if (*line == '"') {
1142 const char *cp;
1143 struct strbuf first = STRBUF_INIT;
1144 struct strbuf sp = STRBUF_INIT;
1146 if (unquote_c_style(&first, line, &second))
1147 goto free_and_fail1;
1149 /* strip the a/b prefix including trailing slash */
1150 cp = skip_tree_prefix(first.buf, first.len);
1151 if (!cp)
1152 goto free_and_fail1;
1153 strbuf_remove(&first, 0, cp - first.buf);
1156 * second points at one past closing dq of name.
1157 * find the second name.
1159 while ((second < line + llen) && isspace(*second))
1160 second++;
1162 if (line + llen <= second)
1163 goto free_and_fail1;
1164 if (*second == '"') {
1165 if (unquote_c_style(&sp, second, NULL))
1166 goto free_and_fail1;
1167 cp = skip_tree_prefix(sp.buf, sp.len);
1168 if (!cp)
1169 goto free_and_fail1;
1170 /* They must match, otherwise ignore */
1171 if (strcmp(cp, first.buf))
1172 goto free_and_fail1;
1173 strbuf_release(&sp);
1174 return strbuf_detach(&first, NULL);
1177 /* unquoted second */
1178 cp = skip_tree_prefix(second, line + llen - second);
1179 if (!cp)
1180 goto free_and_fail1;
1181 if (line + llen - cp != first.len ||
1182 memcmp(first.buf, cp, first.len))
1183 goto free_and_fail1;
1184 return strbuf_detach(&first, NULL);
1186 free_and_fail1:
1187 strbuf_release(&first);
1188 strbuf_release(&sp);
1189 return NULL;
1192 /* unquoted first name */
1193 name = skip_tree_prefix(line, llen);
1194 if (!name)
1195 return NULL;
1198 * since the first name is unquoted, a dq if exists must be
1199 * the beginning of the second name.
1201 for (second = name; second < line + llen; second++) {
1202 if (*second == '"') {
1203 struct strbuf sp = STRBUF_INIT;
1204 const char *np;
1206 if (unquote_c_style(&sp, second, NULL))
1207 goto free_and_fail2;
1209 np = skip_tree_prefix(sp.buf, sp.len);
1210 if (!np)
1211 goto free_and_fail2;
1213 len = sp.buf + sp.len - np;
1214 if (len < second - name &&
1215 !strncmp(np, name, len) &&
1216 isspace(name[len])) {
1217 /* Good */
1218 strbuf_remove(&sp, 0, np - sp.buf);
1219 return strbuf_detach(&sp, NULL);
1222 free_and_fail2:
1223 strbuf_release(&sp);
1224 return NULL;
1229 * Accept a name only if it shows up twice, exactly the same
1230 * form.
1232 second = strchr(name, '\n');
1233 if (!second)
1234 return NULL;
1235 line_len = second - name;
1236 for (len = 0 ; ; len++) {
1237 switch (name[len]) {
1238 default:
1239 continue;
1240 case '\n':
1241 return NULL;
1242 case '\t': case ' ':
1244 * Is this the separator between the preimage
1245 * and the postimage pathname? Again, we are
1246 * only interested in the case where there is
1247 * no rename, as this is only to set def_name
1248 * and a rename patch has the names elsewhere
1249 * in an unambiguous form.
1251 if (!name[len + 1])
1252 return NULL; /* no postimage name */
1253 second = skip_tree_prefix(name + len + 1,
1254 line_len - (len + 1));
1255 if (!second)
1256 return NULL;
1258 * Does len bytes starting at "name" and "second"
1259 * (that are separated by one HT or SP we just
1260 * found) exactly match?
1262 if (second[len] == '\n' && !strncmp(name, second, len))
1263 return xmemdupz(name, len);
1268 /* Verify that we recognize the lines following a git header */
1269 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1271 unsigned long offset;
1273 /* A git diff has explicit new/delete information, so we don't guess */
1274 patch->is_new = 0;
1275 patch->is_delete = 0;
1278 * Some things may not have the old name in the
1279 * rest of the headers anywhere (pure mode changes,
1280 * or removing or adding empty files), so we get
1281 * the default name from the header.
1283 patch->def_name = git_header_name(line, len);
1284 if (patch->def_name && root) {
1285 char *s = xstrfmt("%s%s", root, patch->def_name);
1286 free(patch->def_name);
1287 patch->def_name = s;
1290 line += len;
1291 size -= len;
1292 linenr++;
1293 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
1294 static const struct opentry {
1295 const char *str;
1296 int (*fn)(const char *, struct patch *);
1297 } optable[] = {
1298 { "@@ -", gitdiff_hdrend },
1299 { "--- ", gitdiff_oldname },
1300 { "+++ ", gitdiff_newname },
1301 { "old mode ", gitdiff_oldmode },
1302 { "new mode ", gitdiff_newmode },
1303 { "deleted file mode ", gitdiff_delete },
1304 { "new file mode ", gitdiff_newfile },
1305 { "copy from ", gitdiff_copysrc },
1306 { "copy to ", gitdiff_copydst },
1307 { "rename old ", gitdiff_renamesrc },
1308 { "rename new ", gitdiff_renamedst },
1309 { "rename from ", gitdiff_renamesrc },
1310 { "rename to ", gitdiff_renamedst },
1311 { "similarity index ", gitdiff_similarity },
1312 { "dissimilarity index ", gitdiff_dissimilarity },
1313 { "index ", gitdiff_index },
1314 { "", gitdiff_unrecognized },
1316 int i;
1318 len = linelen(line, size);
1319 if (!len || line[len-1] != '\n')
1320 break;
1321 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1322 const struct opentry *p = optable + i;
1323 int oplen = strlen(p->str);
1324 if (len < oplen || memcmp(p->str, line, oplen))
1325 continue;
1326 if (p->fn(line + oplen, patch) < 0)
1327 return offset;
1328 break;
1332 return offset;
1335 static int parse_num(const char *line, unsigned long *p)
1337 char *ptr;
1339 if (!isdigit(*line))
1340 return 0;
1341 *p = strtoul(line, &ptr, 10);
1342 return ptr - line;
1345 static int parse_range(const char *line, int len, int offset, const char *expect,
1346 unsigned long *p1, unsigned long *p2)
1348 int digits, ex;
1350 if (offset < 0 || offset >= len)
1351 return -1;
1352 line += offset;
1353 len -= offset;
1355 digits = parse_num(line, p1);
1356 if (!digits)
1357 return -1;
1359 offset += digits;
1360 line += digits;
1361 len -= digits;
1363 *p2 = 1;
1364 if (*line == ',') {
1365 digits = parse_num(line+1, p2);
1366 if (!digits)
1367 return -1;
1369 offset += digits+1;
1370 line += digits+1;
1371 len -= digits+1;
1374 ex = strlen(expect);
1375 if (ex > len)
1376 return -1;
1377 if (memcmp(line, expect, ex))
1378 return -1;
1380 return offset + ex;
1383 static void recount_diff(const char *line, int size, struct fragment *fragment)
1385 int oldlines = 0, newlines = 0, ret = 0;
1387 if (size < 1) {
1388 warning("recount: ignore empty hunk");
1389 return;
1392 for (;;) {
1393 int len = linelen(line, size);
1394 size -= len;
1395 line += len;
1397 if (size < 1)
1398 break;
1400 switch (*line) {
1401 case ' ': case '\n':
1402 newlines++;
1403 /* fall through */
1404 case '-':
1405 oldlines++;
1406 continue;
1407 case '+':
1408 newlines++;
1409 continue;
1410 case '\\':
1411 continue;
1412 case '@':
1413 ret = size < 3 || !starts_with(line, "@@ ");
1414 break;
1415 case 'd':
1416 ret = size < 5 || !starts_with(line, "diff ");
1417 break;
1418 default:
1419 ret = -1;
1420 break;
1422 if (ret) {
1423 warning(_("recount: unexpected line: %.*s"),
1424 (int)linelen(line, size), line);
1425 return;
1427 break;
1429 fragment->oldlines = oldlines;
1430 fragment->newlines = newlines;
1434 * Parse a unified diff fragment header of the
1435 * form "@@ -a,b +c,d @@"
1437 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1439 int offset;
1441 if (!len || line[len-1] != '\n')
1442 return -1;
1444 /* Figure out the number of lines in a fragment */
1445 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1446 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1448 return offset;
1451 static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)
1453 unsigned long offset, len;
1455 patch->is_toplevel_relative = 0;
1456 patch->is_rename = patch->is_copy = 0;
1457 patch->is_new = patch->is_delete = -1;
1458 patch->old_mode = patch->new_mode = 0;
1459 patch->old_name = patch->new_name = NULL;
1460 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1461 unsigned long nextlen;
1463 len = linelen(line, size);
1464 if (!len)
1465 break;
1467 /* Testing this early allows us to take a few shortcuts.. */
1468 if (len < 6)
1469 continue;
1472 * Make sure we don't find any unconnected patch fragments.
1473 * That's a sign that we didn't find a header, and that a
1474 * patch has become corrupted/broken up.
1476 if (!memcmp("@@ -", line, 4)) {
1477 struct fragment dummy;
1478 if (parse_fragment_header(line, len, &dummy) < 0)
1479 continue;
1480 die(_("patch fragment without header at line %d: %.*s"),
1481 linenr, (int)len-1, line);
1484 if (size < len + 6)
1485 break;
1488 * Git patch? It might not have a real patch, just a rename
1489 * or mode change, so we handle that specially
1491 if (!memcmp("diff --git ", line, 11)) {
1492 int git_hdr_len = parse_git_header(line, len, size, patch);
1493 if (git_hdr_len <= len)
1494 continue;
1495 if (!patch->old_name && !patch->new_name) {
1496 if (!patch->def_name)
1497 die(Q_("git diff header lacks filename information when removing "
1498 "%d leading pathname component (line %d)",
1499 "git diff header lacks filename information when removing "
1500 "%d leading pathname components (line %d)",
1501 p_value),
1502 p_value, linenr);
1503 patch->old_name = xstrdup(patch->def_name);
1504 patch->new_name = xstrdup(patch->def_name);
1506 if (!patch->is_delete && !patch->new_name)
1507 die("git diff header lacks filename information "
1508 "(line %d)", linenr);
1509 patch->is_toplevel_relative = 1;
1510 *hdrsize = git_hdr_len;
1511 return offset;
1514 /* --- followed by +++ ? */
1515 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1516 continue;
1519 * We only accept unified patches, so we want it to
1520 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1521 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1523 nextlen = linelen(line + len, size - len);
1524 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1525 continue;
1527 /* Ok, we'll consider it a patch */
1528 parse_traditional_patch(line, line+len, patch);
1529 *hdrsize = len + nextlen;
1530 linenr += 2;
1531 return offset;
1533 return -1;
1536 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1538 char *err;
1540 if (!result)
1541 return;
1543 whitespace_error++;
1544 if (squelch_whitespace_errors &&
1545 squelch_whitespace_errors < whitespace_error)
1546 return;
1548 err = whitespace_error_string(result);
1549 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1550 patch_input_file, linenr, err, len, line);
1551 free(err);
1554 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1556 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1558 record_ws_error(result, line + 1, len - 2, linenr);
1562 * Parse a unified diff. Note that this really needs to parse each
1563 * fragment separately, since the only way to know the difference
1564 * between a "---" that is part of a patch, and a "---" that starts
1565 * the next patch is to look at the line counts..
1567 static int parse_fragment(const char *line, unsigned long size,
1568 struct patch *patch, struct fragment *fragment)
1570 int added, deleted;
1571 int len = linelen(line, size), offset;
1572 unsigned long oldlines, newlines;
1573 unsigned long leading, trailing;
1575 offset = parse_fragment_header(line, len, fragment);
1576 if (offset < 0)
1577 return -1;
1578 if (offset > 0 && patch->recount)
1579 recount_diff(line + offset, size - offset, fragment);
1580 oldlines = fragment->oldlines;
1581 newlines = fragment->newlines;
1582 leading = 0;
1583 trailing = 0;
1585 /* Parse the thing.. */
1586 line += len;
1587 size -= len;
1588 linenr++;
1589 added = deleted = 0;
1590 for (offset = len;
1591 0 < size;
1592 offset += len, size -= len, line += len, linenr++) {
1593 if (!oldlines && !newlines)
1594 break;
1595 len = linelen(line, size);
1596 if (!len || line[len-1] != '\n')
1597 return -1;
1598 switch (*line) {
1599 default:
1600 return -1;
1601 case '\n': /* newer GNU diff, an empty context line */
1602 case ' ':
1603 oldlines--;
1604 newlines--;
1605 if (!deleted && !added)
1606 leading++;
1607 trailing++;
1608 break;
1609 case '-':
1610 if (apply_in_reverse &&
1611 ws_error_action != nowarn_ws_error)
1612 check_whitespace(line, len, patch->ws_rule);
1613 deleted++;
1614 oldlines--;
1615 trailing = 0;
1616 break;
1617 case '+':
1618 if (!apply_in_reverse &&
1619 ws_error_action != nowarn_ws_error)
1620 check_whitespace(line, len, patch->ws_rule);
1621 added++;
1622 newlines--;
1623 trailing = 0;
1624 break;
1627 * We allow "\ No newline at end of file". Depending
1628 * on locale settings when the patch was produced we
1629 * don't know what this line looks like. The only
1630 * thing we do know is that it begins with "\ ".
1631 * Checking for 12 is just for sanity check -- any
1632 * l10n of "\ No newline..." is at least that long.
1634 case '\\':
1635 if (len < 12 || memcmp(line, "\\ ", 2))
1636 return -1;
1637 break;
1640 if (oldlines || newlines)
1641 return -1;
1642 if (!deleted && !added)
1643 return -1;
1645 fragment->leading = leading;
1646 fragment->trailing = trailing;
1649 * If a fragment ends with an incomplete line, we failed to include
1650 * it in the above loop because we hit oldlines == newlines == 0
1651 * before seeing it.
1653 if (12 < size && !memcmp(line, "\\ ", 2))
1654 offset += linelen(line, size);
1656 patch->lines_added += added;
1657 patch->lines_deleted += deleted;
1659 if (0 < patch->is_new && oldlines)
1660 return error(_("new file depends on old contents"));
1661 if (0 < patch->is_delete && newlines)
1662 return error(_("deleted file still has contents"));
1663 return offset;
1667 * We have seen "diff --git a/... b/..." header (or a traditional patch
1668 * header). Read hunks that belong to this patch into fragments and hang
1669 * them to the given patch structure.
1671 * The (fragment->patch, fragment->size) pair points into the memory given
1672 * by the caller, not a copy, when we return.
1674 static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)
1676 unsigned long offset = 0;
1677 unsigned long oldlines = 0, newlines = 0, context = 0;
1678 struct fragment **fragp = &patch->fragments;
1680 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1681 struct fragment *fragment;
1682 int len;
1684 fragment = xcalloc(1, sizeof(*fragment));
1685 fragment->linenr = linenr;
1686 len = parse_fragment(line, size, patch, fragment);
1687 if (len <= 0)
1688 die(_("corrupt patch at line %d"), linenr);
1689 fragment->patch = line;
1690 fragment->size = len;
1691 oldlines += fragment->oldlines;
1692 newlines += fragment->newlines;
1693 context += fragment->leading + fragment->trailing;
1695 *fragp = fragment;
1696 fragp = &fragment->next;
1698 offset += len;
1699 line += len;
1700 size -= len;
1704 * If something was removed (i.e. we have old-lines) it cannot
1705 * be creation, and if something was added it cannot be
1706 * deletion. However, the reverse is not true; --unified=0
1707 * patches that only add are not necessarily creation even
1708 * though they do not have any old lines, and ones that only
1709 * delete are not necessarily deletion.
1711 * Unfortunately, a real creation/deletion patch do _not_ have
1712 * any context line by definition, so we cannot safely tell it
1713 * apart with --unified=0 insanity. At least if the patch has
1714 * more than one hunk it is not creation or deletion.
1716 if (patch->is_new < 0 &&
1717 (oldlines || (patch->fragments && patch->fragments->next)))
1718 patch->is_new = 0;
1719 if (patch->is_delete < 0 &&
1720 (newlines || (patch->fragments && patch->fragments->next)))
1721 patch->is_delete = 0;
1723 if (0 < patch->is_new && oldlines)
1724 die(_("new file %s depends on old contents"), patch->new_name);
1725 if (0 < patch->is_delete && newlines)
1726 die(_("deleted file %s still has contents"), patch->old_name);
1727 if (!patch->is_delete && !newlines && context)
1728 fprintf_ln(stderr,
1729 _("** warning: "
1730 "file %s becomes empty but is not deleted"),
1731 patch->new_name);
1733 return offset;
1736 static inline int metadata_changes(struct patch *patch)
1738 return patch->is_rename > 0 ||
1739 patch->is_copy > 0 ||
1740 patch->is_new > 0 ||
1741 patch->is_delete ||
1742 (patch->old_mode && patch->new_mode &&
1743 patch->old_mode != patch->new_mode);
1746 static char *inflate_it(const void *data, unsigned long size,
1747 unsigned long inflated_size)
1749 git_zstream stream;
1750 void *out;
1751 int st;
1753 memset(&stream, 0, sizeof(stream));
1755 stream.next_in = (unsigned char *)data;
1756 stream.avail_in = size;
1757 stream.next_out = out = xmalloc(inflated_size);
1758 stream.avail_out = inflated_size;
1759 git_inflate_init(&stream);
1760 st = git_inflate(&stream, Z_FINISH);
1761 git_inflate_end(&stream);
1762 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1763 free(out);
1764 return NULL;
1766 return out;
1770 * Read a binary hunk and return a new fragment; fragment->patch
1771 * points at an allocated memory that the caller must free, so
1772 * it is marked as "->free_patch = 1".
1774 static struct fragment *parse_binary_hunk(char **buf_p,
1775 unsigned long *sz_p,
1776 int *status_p,
1777 int *used_p)
1780 * Expect a line that begins with binary patch method ("literal"
1781 * or "delta"), followed by the length of data before deflating.
1782 * a sequence of 'length-byte' followed by base-85 encoded data
1783 * should follow, terminated by a newline.
1785 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1786 * and we would limit the patch line to 66 characters,
1787 * so one line can fit up to 13 groups that would decode
1788 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1789 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1791 int llen, used;
1792 unsigned long size = *sz_p;
1793 char *buffer = *buf_p;
1794 int patch_method;
1795 unsigned long origlen;
1796 char *data = NULL;
1797 int hunk_size = 0;
1798 struct fragment *frag;
1800 llen = linelen(buffer, size);
1801 used = llen;
1803 *status_p = 0;
1805 if (starts_with(buffer, "delta ")) {
1806 patch_method = BINARY_DELTA_DEFLATED;
1807 origlen = strtoul(buffer + 6, NULL, 10);
1809 else if (starts_with(buffer, "literal ")) {
1810 patch_method = BINARY_LITERAL_DEFLATED;
1811 origlen = strtoul(buffer + 8, NULL, 10);
1813 else
1814 return NULL;
1816 linenr++;
1817 buffer += llen;
1818 while (1) {
1819 int byte_length, max_byte_length, newsize;
1820 llen = linelen(buffer, size);
1821 used += llen;
1822 linenr++;
1823 if (llen == 1) {
1824 /* consume the blank line */
1825 buffer++;
1826 size--;
1827 break;
1830 * Minimum line is "A00000\n" which is 7-byte long,
1831 * and the line length must be multiple of 5 plus 2.
1833 if ((llen < 7) || (llen-2) % 5)
1834 goto corrupt;
1835 max_byte_length = (llen - 2) / 5 * 4;
1836 byte_length = *buffer;
1837 if ('A' <= byte_length && byte_length <= 'Z')
1838 byte_length = byte_length - 'A' + 1;
1839 else if ('a' <= byte_length && byte_length <= 'z')
1840 byte_length = byte_length - 'a' + 27;
1841 else
1842 goto corrupt;
1843 /* if the input length was not multiple of 4, we would
1844 * have filler at the end but the filler should never
1845 * exceed 3 bytes
1847 if (max_byte_length < byte_length ||
1848 byte_length <= max_byte_length - 4)
1849 goto corrupt;
1850 newsize = hunk_size + byte_length;
1851 data = xrealloc(data, newsize);
1852 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1853 goto corrupt;
1854 hunk_size = newsize;
1855 buffer += llen;
1856 size -= llen;
1859 frag = xcalloc(1, sizeof(*frag));
1860 frag->patch = inflate_it(data, hunk_size, origlen);
1861 frag->free_patch = 1;
1862 if (!frag->patch)
1863 goto corrupt;
1864 free(data);
1865 frag->size = origlen;
1866 *buf_p = buffer;
1867 *sz_p = size;
1868 *used_p = used;
1869 frag->binary_patch_method = patch_method;
1870 return frag;
1872 corrupt:
1873 free(data);
1874 *status_p = -1;
1875 error(_("corrupt binary patch at line %d: %.*s"),
1876 linenr-1, llen-1, buffer);
1877 return NULL;
1880 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1883 * We have read "GIT binary patch\n"; what follows is a line
1884 * that says the patch method (currently, either "literal" or
1885 * "delta") and the length of data before deflating; a
1886 * sequence of 'length-byte' followed by base-85 encoded data
1887 * follows.
1889 * When a binary patch is reversible, there is another binary
1890 * hunk in the same format, starting with patch method (either
1891 * "literal" or "delta") with the length of data, and a sequence
1892 * of length-byte + base-85 encoded data, terminated with another
1893 * empty line. This data, when applied to the postimage, produces
1894 * the preimage.
1896 struct fragment *forward;
1897 struct fragment *reverse;
1898 int status;
1899 int used, used_1;
1901 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1902 if (!forward && !status)
1903 /* there has to be one hunk (forward hunk) */
1904 return error(_("unrecognized binary patch at line %d"), linenr-1);
1905 if (status)
1906 /* otherwise we already gave an error message */
1907 return status;
1909 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1910 if (reverse)
1911 used += used_1;
1912 else if (status) {
1914 * Not having reverse hunk is not an error, but having
1915 * a corrupt reverse hunk is.
1917 free((void*) forward->patch);
1918 free(forward);
1919 return status;
1921 forward->next = reverse;
1922 patch->fragments = forward;
1923 patch->is_binary = 1;
1924 return used;
1927 static void prefix_one(char **name)
1929 char *old_name = *name;
1930 if (!old_name)
1931 return;
1932 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
1933 free(old_name);
1936 static void prefix_patch(struct patch *p)
1938 if (!prefix || p->is_toplevel_relative)
1939 return;
1940 prefix_one(&p->new_name);
1941 prefix_one(&p->old_name);
1945 * include/exclude
1948 static struct string_list limit_by_name;
1949 static int has_include;
1950 static void add_name_limit(const char *name, int exclude)
1952 struct string_list_item *it;
1954 it = string_list_append(&limit_by_name, name);
1955 it->util = exclude ? NULL : (void *) 1;
1958 static int use_patch(struct patch *p)
1960 const char *pathname = p->new_name ? p->new_name : p->old_name;
1961 int i;
1963 /* Paths outside are not touched regardless of "--include" */
1964 if (0 < prefix_length) {
1965 int pathlen = strlen(pathname);
1966 if (pathlen <= prefix_length ||
1967 memcmp(prefix, pathname, prefix_length))
1968 return 0;
1971 /* See if it matches any of exclude/include rule */
1972 for (i = 0; i < limit_by_name.nr; i++) {
1973 struct string_list_item *it = &limit_by_name.items[i];
1974 if (!wildmatch(it->string, pathname, 0, NULL))
1975 return (it->util != NULL);
1979 * If we had any include, a path that does not match any rule is
1980 * not used. Otherwise, we saw bunch of exclude rules (or none)
1981 * and such a path is used.
1983 return !has_include;
1988 * Read the patch text in "buffer" that extends for "size" bytes; stop
1989 * reading after seeing a single patch (i.e. changes to a single file).
1990 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1991 * Return the number of bytes consumed, so that the caller can call us
1992 * again for the next patch.
1994 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1996 int hdrsize, patchsize;
1997 int offset = find_header(buffer, size, &hdrsize, patch);
1999 if (offset < 0)
2000 return offset;
2002 prefix_patch(patch);
2004 if (!use_patch(patch))
2005 patch->ws_rule = 0;
2006 else
2007 patch->ws_rule = whitespace_rule(patch->new_name
2008 ? patch->new_name
2009 : patch->old_name);
2011 patchsize = parse_single_patch(buffer + offset + hdrsize,
2012 size - offset - hdrsize, patch);
2014 if (!patchsize) {
2015 static const char git_binary[] = "GIT binary patch\n";
2016 int hd = hdrsize + offset;
2017 unsigned long llen = linelen(buffer + hd, size - hd);
2019 if (llen == sizeof(git_binary) - 1 &&
2020 !memcmp(git_binary, buffer + hd, llen)) {
2021 int used;
2022 linenr++;
2023 used = parse_binary(buffer + hd + llen,
2024 size - hd - llen, patch);
2025 if (used)
2026 patchsize = used + llen;
2027 else
2028 patchsize = 0;
2030 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2031 static const char *binhdr[] = {
2032 "Binary files ",
2033 "Files ",
2034 NULL,
2036 int i;
2037 for (i = 0; binhdr[i]; i++) {
2038 int len = strlen(binhdr[i]);
2039 if (len < size - hd &&
2040 !memcmp(binhdr[i], buffer + hd, len)) {
2041 linenr++;
2042 patch->is_binary = 1;
2043 patchsize = llen;
2044 break;
2049 /* Empty patch cannot be applied if it is a text patch
2050 * without metadata change. A binary patch appears
2051 * empty to us here.
2053 if ((apply || check) &&
2054 (!patch->is_binary && !metadata_changes(patch)))
2055 die(_("patch with only garbage at line %d"), linenr);
2058 return offset + hdrsize + patchsize;
2061 #define swap(a,b) myswap((a),(b),sizeof(a))
2063 #define myswap(a, b, size) do { \
2064 unsigned char mytmp[size]; \
2065 memcpy(mytmp, &a, size); \
2066 memcpy(&a, &b, size); \
2067 memcpy(&b, mytmp, size); \
2068 } while (0)
2070 static void reverse_patches(struct patch *p)
2072 for (; p; p = p->next) {
2073 struct fragment *frag = p->fragments;
2075 swap(p->new_name, p->old_name);
2076 swap(p->new_mode, p->old_mode);
2077 swap(p->is_new, p->is_delete);
2078 swap(p->lines_added, p->lines_deleted);
2079 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2081 for (; frag; frag = frag->next) {
2082 swap(frag->newpos, frag->oldpos);
2083 swap(frag->newlines, frag->oldlines);
2088 static const char pluses[] =
2089 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2090 static const char minuses[]=
2091 "----------------------------------------------------------------------";
2093 static void show_stats(struct patch *patch)
2095 struct strbuf qname = STRBUF_INIT;
2096 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2097 int max, add, del;
2099 quote_c_style(cp, &qname, NULL, 0);
2102 * "scale" the filename
2104 max = max_len;
2105 if (max > 50)
2106 max = 50;
2108 if (qname.len > max) {
2109 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2110 if (!cp)
2111 cp = qname.buf + qname.len + 3 - max;
2112 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2115 if (patch->is_binary) {
2116 printf(" %-*s | Bin\n", max, qname.buf);
2117 strbuf_release(&qname);
2118 return;
2121 printf(" %-*s |", max, qname.buf);
2122 strbuf_release(&qname);
2125 * scale the add/delete
2127 max = max + max_change > 70 ? 70 - max : max_change;
2128 add = patch->lines_added;
2129 del = patch->lines_deleted;
2131 if (max_change > 0) {
2132 int total = ((add + del) * max + max_change / 2) / max_change;
2133 add = (add * max + max_change / 2) / max_change;
2134 del = total - add;
2136 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2137 add, pluses, del, minuses);
2140 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2142 switch (st->st_mode & S_IFMT) {
2143 case S_IFLNK:
2144 if (strbuf_readlink(buf, path, st->st_size) < 0)
2145 return error(_("unable to read symlink %s"), path);
2146 return 0;
2147 case S_IFREG:
2148 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2149 return error(_("unable to open or read %s"), path);
2150 convert_to_git(path, buf->buf, buf->len, buf, 0);
2151 return 0;
2152 default:
2153 return -1;
2158 * Update the preimage, and the common lines in postimage,
2159 * from buffer buf of length len. If postlen is 0 the postimage
2160 * is updated in place, otherwise it's updated on a new buffer
2161 * of length postlen
2164 static void update_pre_post_images(struct image *preimage,
2165 struct image *postimage,
2166 char *buf,
2167 size_t len, size_t postlen)
2169 int i, ctx, reduced;
2170 char *new, *old, *fixed;
2171 struct image fixed_preimage;
2174 * Update the preimage with whitespace fixes. Note that we
2175 * are not losing preimage->buf -- apply_one_fragment() will
2176 * free "oldlines".
2178 prepare_image(&fixed_preimage, buf, len, 1);
2179 assert(postlen
2180 ? fixed_preimage.nr == preimage->nr
2181 : fixed_preimage.nr <= preimage->nr);
2182 for (i = 0; i < fixed_preimage.nr; i++)
2183 fixed_preimage.line[i].flag = preimage->line[i].flag;
2184 free(preimage->line_allocated);
2185 *preimage = fixed_preimage;
2188 * Adjust the common context lines in postimage. This can be
2189 * done in-place when we are shrinking it with whitespace
2190 * fixing, but needs a new buffer when ignoring whitespace or
2191 * expanding leading tabs to spaces.
2193 * We trust the caller to tell us if the update can be done
2194 * in place (postlen==0) or not.
2196 old = postimage->buf;
2197 if (postlen)
2198 new = postimage->buf = xmalloc(postlen);
2199 else
2200 new = old;
2201 fixed = preimage->buf;
2203 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2204 size_t len = postimage->line[i].len;
2205 if (!(postimage->line[i].flag & LINE_COMMON)) {
2206 /* an added line -- no counterparts in preimage */
2207 memmove(new, old, len);
2208 old += len;
2209 new += len;
2210 continue;
2213 /* a common context -- skip it in the original postimage */
2214 old += len;
2216 /* and find the corresponding one in the fixed preimage */
2217 while (ctx < preimage->nr &&
2218 !(preimage->line[ctx].flag & LINE_COMMON)) {
2219 fixed += preimage->line[ctx].len;
2220 ctx++;
2224 * preimage is expected to run out, if the caller
2225 * fixed addition of trailing blank lines.
2227 if (preimage->nr <= ctx) {
2228 reduced++;
2229 continue;
2232 /* and copy it in, while fixing the line length */
2233 len = preimage->line[ctx].len;
2234 memcpy(new, fixed, len);
2235 new += len;
2236 fixed += len;
2237 postimage->line[i].len = len;
2238 ctx++;
2241 /* Fix the length of the whole thing */
2242 postimage->len = new - postimage->buf;
2243 postimage->nr -= reduced;
2246 static int match_fragment(struct image *img,
2247 struct image *preimage,
2248 struct image *postimage,
2249 unsigned long try,
2250 int try_lno,
2251 unsigned ws_rule,
2252 int match_beginning, int match_end)
2254 int i;
2255 char *fixed_buf, *buf, *orig, *target;
2256 struct strbuf fixed;
2257 size_t fixed_len, postlen;
2258 int preimage_limit;
2260 if (preimage->nr + try_lno <= img->nr) {
2262 * The hunk falls within the boundaries of img.
2264 preimage_limit = preimage->nr;
2265 if (match_end && (preimage->nr + try_lno != img->nr))
2266 return 0;
2267 } else if (ws_error_action == correct_ws_error &&
2268 (ws_rule & WS_BLANK_AT_EOF)) {
2270 * This hunk extends beyond the end of img, and we are
2271 * removing blank lines at the end of the file. This
2272 * many lines from the beginning of the preimage must
2273 * match with img, and the remainder of the preimage
2274 * must be blank.
2276 preimage_limit = img->nr - try_lno;
2277 } else {
2279 * The hunk extends beyond the end of the img and
2280 * we are not removing blanks at the end, so we
2281 * should reject the hunk at this position.
2283 return 0;
2286 if (match_beginning && try_lno)
2287 return 0;
2289 /* Quick hash check */
2290 for (i = 0; i < preimage_limit; i++)
2291 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2292 (preimage->line[i].hash != img->line[try_lno + i].hash))
2293 return 0;
2295 if (preimage_limit == preimage->nr) {
2297 * Do we have an exact match? If we were told to match
2298 * at the end, size must be exactly at try+fragsize,
2299 * otherwise try+fragsize must be still within the preimage,
2300 * and either case, the old piece should match the preimage
2301 * exactly.
2303 if ((match_end
2304 ? (try + preimage->len == img->len)
2305 : (try + preimage->len <= img->len)) &&
2306 !memcmp(img->buf + try, preimage->buf, preimage->len))
2307 return 1;
2308 } else {
2310 * The preimage extends beyond the end of img, so
2311 * there cannot be an exact match.
2313 * There must be one non-blank context line that match
2314 * a line before the end of img.
2316 char *buf_end;
2318 buf = preimage->buf;
2319 buf_end = buf;
2320 for (i = 0; i < preimage_limit; i++)
2321 buf_end += preimage->line[i].len;
2323 for ( ; buf < buf_end; buf++)
2324 if (!isspace(*buf))
2325 break;
2326 if (buf == buf_end)
2327 return 0;
2331 * No exact match. If we are ignoring whitespace, run a line-by-line
2332 * fuzzy matching. We collect all the line length information because
2333 * we need it to adjust whitespace if we match.
2335 if (ws_ignore_action == ignore_ws_change) {
2336 size_t imgoff = 0;
2337 size_t preoff = 0;
2338 size_t postlen = postimage->len;
2339 size_t extra_chars;
2340 char *preimage_eof;
2341 char *preimage_end;
2342 for (i = 0; i < preimage_limit; i++) {
2343 size_t prelen = preimage->line[i].len;
2344 size_t imglen = img->line[try_lno+i].len;
2346 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2347 preimage->buf + preoff, prelen))
2348 return 0;
2349 if (preimage->line[i].flag & LINE_COMMON)
2350 postlen += imglen - prelen;
2351 imgoff += imglen;
2352 preoff += prelen;
2356 * Ok, the preimage matches with whitespace fuzz.
2358 * imgoff now holds the true length of the target that
2359 * matches the preimage before the end of the file.
2361 * Count the number of characters in the preimage that fall
2362 * beyond the end of the file and make sure that all of them
2363 * are whitespace characters. (This can only happen if
2364 * we are removing blank lines at the end of the file.)
2366 buf = preimage_eof = preimage->buf + preoff;
2367 for ( ; i < preimage->nr; i++)
2368 preoff += preimage->line[i].len;
2369 preimage_end = preimage->buf + preoff;
2370 for ( ; buf < preimage_end; buf++)
2371 if (!isspace(*buf))
2372 return 0;
2375 * Update the preimage and the common postimage context
2376 * lines to use the same whitespace as the target.
2377 * If whitespace is missing in the target (i.e.
2378 * if the preimage extends beyond the end of the file),
2379 * use the whitespace from the preimage.
2381 extra_chars = preimage_end - preimage_eof;
2382 strbuf_init(&fixed, imgoff + extra_chars);
2383 strbuf_add(&fixed, img->buf + try, imgoff);
2384 strbuf_add(&fixed, preimage_eof, extra_chars);
2385 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2386 update_pre_post_images(preimage, postimage,
2387 fixed_buf, fixed_len, postlen);
2388 return 1;
2391 if (ws_error_action != correct_ws_error)
2392 return 0;
2395 * The hunk does not apply byte-by-byte, but the hash says
2396 * it might with whitespace fuzz. We haven't been asked to
2397 * ignore whitespace, we were asked to correct whitespace
2398 * errors, so let's try matching after whitespace correction.
2400 * The preimage may extend beyond the end of the file,
2401 * but in this loop we will only handle the part of the
2402 * preimage that falls within the file.
2404 strbuf_init(&fixed, preimage->len + 1);
2405 orig = preimage->buf;
2406 target = img->buf + try;
2407 postlen = 0;
2408 for (i = 0; i < preimage_limit; i++) {
2409 size_t oldlen = preimage->line[i].len;
2410 size_t tgtlen = img->line[try_lno + i].len;
2411 size_t fixstart = fixed.len;
2412 struct strbuf tgtfix;
2413 int match;
2415 /* Try fixing the line in the preimage */
2416 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2418 /* Try fixing the line in the target */
2419 strbuf_init(&tgtfix, tgtlen);
2420 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2423 * If they match, either the preimage was based on
2424 * a version before our tree fixed whitespace breakage,
2425 * or we are lacking a whitespace-fix patch the tree
2426 * the preimage was based on already had (i.e. target
2427 * has whitespace breakage, the preimage doesn't).
2428 * In either case, we are fixing the whitespace breakages
2429 * so we might as well take the fix together with their
2430 * real change.
2432 match = (tgtfix.len == fixed.len - fixstart &&
2433 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2434 fixed.len - fixstart));
2435 postlen += tgtfix.len;
2437 strbuf_release(&tgtfix);
2438 if (!match)
2439 goto unmatch_exit;
2441 orig += oldlen;
2442 target += tgtlen;
2447 * Now handle the lines in the preimage that falls beyond the
2448 * end of the file (if any). They will only match if they are
2449 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2450 * false).
2452 for ( ; i < preimage->nr; i++) {
2453 size_t fixstart = fixed.len; /* start of the fixed preimage */
2454 size_t oldlen = preimage->line[i].len;
2455 int j;
2457 /* Try fixing the line in the preimage */
2458 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2460 for (j = fixstart; j < fixed.len; j++)
2461 if (!isspace(fixed.buf[j]))
2462 goto unmatch_exit;
2464 orig += oldlen;
2468 * Yes, the preimage is based on an older version that still
2469 * has whitespace breakages unfixed, and fixing them makes the
2470 * hunk match. Update the context lines in the postimage.
2472 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2473 if (postlen < postimage->len)
2474 postlen = 0;
2475 update_pre_post_images(preimage, postimage,
2476 fixed_buf, fixed_len, postlen);
2477 return 1;
2479 unmatch_exit:
2480 strbuf_release(&fixed);
2481 return 0;
2484 static int find_pos(struct image *img,
2485 struct image *preimage,
2486 struct image *postimage,
2487 int line,
2488 unsigned ws_rule,
2489 int match_beginning, int match_end)
2491 int i;
2492 unsigned long backwards, forwards, try;
2493 int backwards_lno, forwards_lno, try_lno;
2496 * If match_beginning or match_end is specified, there is no
2497 * point starting from a wrong line that will never match and
2498 * wander around and wait for a match at the specified end.
2500 if (match_beginning)
2501 line = 0;
2502 else if (match_end)
2503 line = img->nr - preimage->nr;
2506 * Because the comparison is unsigned, the following test
2507 * will also take care of a negative line number that can
2508 * result when match_end and preimage is larger than the target.
2510 if ((size_t) line > img->nr)
2511 line = img->nr;
2513 try = 0;
2514 for (i = 0; i < line; i++)
2515 try += img->line[i].len;
2518 * There's probably some smart way to do this, but I'll leave
2519 * that to the smart and beautiful people. I'm simple and stupid.
2521 backwards = try;
2522 backwards_lno = line;
2523 forwards = try;
2524 forwards_lno = line;
2525 try_lno = line;
2527 for (i = 0; ; i++) {
2528 if (match_fragment(img, preimage, postimage,
2529 try, try_lno, ws_rule,
2530 match_beginning, match_end))
2531 return try_lno;
2533 again:
2534 if (backwards_lno == 0 && forwards_lno == img->nr)
2535 break;
2537 if (i & 1) {
2538 if (backwards_lno == 0) {
2539 i++;
2540 goto again;
2542 backwards_lno--;
2543 backwards -= img->line[backwards_lno].len;
2544 try = backwards;
2545 try_lno = backwards_lno;
2546 } else {
2547 if (forwards_lno == img->nr) {
2548 i++;
2549 goto again;
2551 forwards += img->line[forwards_lno].len;
2552 forwards_lno++;
2553 try = forwards;
2554 try_lno = forwards_lno;
2558 return -1;
2561 static void remove_first_line(struct image *img)
2563 img->buf += img->line[0].len;
2564 img->len -= img->line[0].len;
2565 img->line++;
2566 img->nr--;
2569 static void remove_last_line(struct image *img)
2571 img->len -= img->line[--img->nr].len;
2575 * The change from "preimage" and "postimage" has been found to
2576 * apply at applied_pos (counts in line numbers) in "img".
2577 * Update "img" to remove "preimage" and replace it with "postimage".
2579 static void update_image(struct image *img,
2580 int applied_pos,
2581 struct image *preimage,
2582 struct image *postimage)
2585 * remove the copy of preimage at offset in img
2586 * and replace it with postimage
2588 int i, nr;
2589 size_t remove_count, insert_count, applied_at = 0;
2590 char *result;
2591 int preimage_limit;
2594 * If we are removing blank lines at the end of img,
2595 * the preimage may extend beyond the end.
2596 * If that is the case, we must be careful only to
2597 * remove the part of the preimage that falls within
2598 * the boundaries of img. Initialize preimage_limit
2599 * to the number of lines in the preimage that falls
2600 * within the boundaries.
2602 preimage_limit = preimage->nr;
2603 if (preimage_limit > img->nr - applied_pos)
2604 preimage_limit = img->nr - applied_pos;
2606 for (i = 0; i < applied_pos; i++)
2607 applied_at += img->line[i].len;
2609 remove_count = 0;
2610 for (i = 0; i < preimage_limit; i++)
2611 remove_count += img->line[applied_pos + i].len;
2612 insert_count = postimage->len;
2614 /* Adjust the contents */
2615 result = xmalloc(img->len + insert_count - remove_count + 1);
2616 memcpy(result, img->buf, applied_at);
2617 memcpy(result + applied_at, postimage->buf, postimage->len);
2618 memcpy(result + applied_at + postimage->len,
2619 img->buf + (applied_at + remove_count),
2620 img->len - (applied_at + remove_count));
2621 free(img->buf);
2622 img->buf = result;
2623 img->len += insert_count - remove_count;
2624 result[img->len] = '\0';
2626 /* Adjust the line table */
2627 nr = img->nr + postimage->nr - preimage_limit;
2628 if (preimage_limit < postimage->nr) {
2630 * NOTE: this knows that we never call remove_first_line()
2631 * on anything other than pre/post image.
2633 REALLOC_ARRAY(img->line, nr);
2634 img->line_allocated = img->line;
2636 if (preimage_limit != postimage->nr)
2637 memmove(img->line + applied_pos + postimage->nr,
2638 img->line + applied_pos + preimage_limit,
2639 (img->nr - (applied_pos + preimage_limit)) *
2640 sizeof(*img->line));
2641 memcpy(img->line + applied_pos,
2642 postimage->line,
2643 postimage->nr * sizeof(*img->line));
2644 if (!allow_overlap)
2645 for (i = 0; i < postimage->nr; i++)
2646 img->line[applied_pos + i].flag |= LINE_PATCHED;
2647 img->nr = nr;
2651 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2652 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2653 * replace the part of "img" with "postimage" text.
2655 static int apply_one_fragment(struct image *img, struct fragment *frag,
2656 int inaccurate_eof, unsigned ws_rule,
2657 int nth_fragment)
2659 int match_beginning, match_end;
2660 const char *patch = frag->patch;
2661 int size = frag->size;
2662 char *old, *oldlines;
2663 struct strbuf newlines;
2664 int new_blank_lines_at_end = 0;
2665 int found_new_blank_lines_at_end = 0;
2666 int hunk_linenr = frag->linenr;
2667 unsigned long leading, trailing;
2668 int pos, applied_pos;
2669 struct image preimage;
2670 struct image postimage;
2672 memset(&preimage, 0, sizeof(preimage));
2673 memset(&postimage, 0, sizeof(postimage));
2674 oldlines = xmalloc(size);
2675 strbuf_init(&newlines, size);
2677 old = oldlines;
2678 while (size > 0) {
2679 char first;
2680 int len = linelen(patch, size);
2681 int plen;
2682 int added_blank_line = 0;
2683 int is_blank_context = 0;
2684 size_t start;
2686 if (!len)
2687 break;
2690 * "plen" is how much of the line we should use for
2691 * the actual patch data. Normally we just remove the
2692 * first character on the line, but if the line is
2693 * followed by "\ No newline", then we also remove the
2694 * last one (which is the newline, of course).
2696 plen = len - 1;
2697 if (len < size && patch[len] == '\\')
2698 plen--;
2699 first = *patch;
2700 if (apply_in_reverse) {
2701 if (first == '-')
2702 first = '+';
2703 else if (first == '+')
2704 first = '-';
2707 switch (first) {
2708 case '\n':
2709 /* Newer GNU diff, empty context line */
2710 if (plen < 0)
2711 /* ... followed by '\No newline'; nothing */
2712 break;
2713 *old++ = '\n';
2714 strbuf_addch(&newlines, '\n');
2715 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2716 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2717 is_blank_context = 1;
2718 break;
2719 case ' ':
2720 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2721 ws_blank_line(patch + 1, plen, ws_rule))
2722 is_blank_context = 1;
2723 case '-':
2724 memcpy(old, patch + 1, plen);
2725 add_line_info(&preimage, old, plen,
2726 (first == ' ' ? LINE_COMMON : 0));
2727 old += plen;
2728 if (first == '-')
2729 break;
2730 /* Fall-through for ' ' */
2731 case '+':
2732 /* --no-add does not add new lines */
2733 if (first == '+' && no_add)
2734 break;
2736 start = newlines.len;
2737 if (first != '+' ||
2738 !whitespace_error ||
2739 ws_error_action != correct_ws_error) {
2740 strbuf_add(&newlines, patch + 1, plen);
2742 else {
2743 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2745 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2746 (first == '+' ? 0 : LINE_COMMON));
2747 if (first == '+' &&
2748 (ws_rule & WS_BLANK_AT_EOF) &&
2749 ws_blank_line(patch + 1, plen, ws_rule))
2750 added_blank_line = 1;
2751 break;
2752 case '@': case '\\':
2753 /* Ignore it, we already handled it */
2754 break;
2755 default:
2756 if (apply_verbosely)
2757 error(_("invalid start of line: '%c'"), first);
2758 return -1;
2760 if (added_blank_line) {
2761 if (!new_blank_lines_at_end)
2762 found_new_blank_lines_at_end = hunk_linenr;
2763 new_blank_lines_at_end++;
2765 else if (is_blank_context)
2767 else
2768 new_blank_lines_at_end = 0;
2769 patch += len;
2770 size -= len;
2771 hunk_linenr++;
2773 if (inaccurate_eof &&
2774 old > oldlines && old[-1] == '\n' &&
2775 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2776 old--;
2777 strbuf_setlen(&newlines, newlines.len - 1);
2780 leading = frag->leading;
2781 trailing = frag->trailing;
2784 * A hunk to change lines at the beginning would begin with
2785 * @@ -1,L +N,M @@
2786 * but we need to be careful. -U0 that inserts before the second
2787 * line also has this pattern.
2789 * And a hunk to add to an empty file would begin with
2790 * @@ -0,0 +N,M @@
2792 * In other words, a hunk that is (frag->oldpos <= 1) with or
2793 * without leading context must match at the beginning.
2795 match_beginning = (!frag->oldpos ||
2796 (frag->oldpos == 1 && !unidiff_zero));
2799 * A hunk without trailing lines must match at the end.
2800 * However, we simply cannot tell if a hunk must match end
2801 * from the lack of trailing lines if the patch was generated
2802 * with unidiff without any context.
2804 match_end = !unidiff_zero && !trailing;
2806 pos = frag->newpos ? (frag->newpos - 1) : 0;
2807 preimage.buf = oldlines;
2808 preimage.len = old - oldlines;
2809 postimage.buf = newlines.buf;
2810 postimage.len = newlines.len;
2811 preimage.line = preimage.line_allocated;
2812 postimage.line = postimage.line_allocated;
2814 for (;;) {
2816 applied_pos = find_pos(img, &preimage, &postimage, pos,
2817 ws_rule, match_beginning, match_end);
2819 if (applied_pos >= 0)
2820 break;
2822 /* Am I at my context limits? */
2823 if ((leading <= p_context) && (trailing <= p_context))
2824 break;
2825 if (match_beginning || match_end) {
2826 match_beginning = match_end = 0;
2827 continue;
2831 * Reduce the number of context lines; reduce both
2832 * leading and trailing if they are equal otherwise
2833 * just reduce the larger context.
2835 if (leading >= trailing) {
2836 remove_first_line(&preimage);
2837 remove_first_line(&postimage);
2838 pos--;
2839 leading--;
2841 if (trailing > leading) {
2842 remove_last_line(&preimage);
2843 remove_last_line(&postimage);
2844 trailing--;
2848 if (applied_pos >= 0) {
2849 if (new_blank_lines_at_end &&
2850 preimage.nr + applied_pos >= img->nr &&
2851 (ws_rule & WS_BLANK_AT_EOF) &&
2852 ws_error_action != nowarn_ws_error) {
2853 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2854 found_new_blank_lines_at_end);
2855 if (ws_error_action == correct_ws_error) {
2856 while (new_blank_lines_at_end--)
2857 remove_last_line(&postimage);
2860 * We would want to prevent write_out_results()
2861 * from taking place in apply_patch() that follows
2862 * the callchain led us here, which is:
2863 * apply_patch->check_patch_list->check_patch->
2864 * apply_data->apply_fragments->apply_one_fragment
2866 if (ws_error_action == die_on_ws_error)
2867 apply = 0;
2870 if (apply_verbosely && applied_pos != pos) {
2871 int offset = applied_pos - pos;
2872 if (apply_in_reverse)
2873 offset = 0 - offset;
2874 fprintf_ln(stderr,
2875 Q_("Hunk #%d succeeded at %d (offset %d line).",
2876 "Hunk #%d succeeded at %d (offset %d lines).",
2877 offset),
2878 nth_fragment, applied_pos + 1, offset);
2882 * Warn if it was necessary to reduce the number
2883 * of context lines.
2885 if ((leading != frag->leading) ||
2886 (trailing != frag->trailing))
2887 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2888 " to apply fragment at %d"),
2889 leading, trailing, applied_pos+1);
2890 update_image(img, applied_pos, &preimage, &postimage);
2891 } else {
2892 if (apply_verbosely)
2893 error(_("while searching for:\n%.*s"),
2894 (int)(old - oldlines), oldlines);
2897 free(oldlines);
2898 strbuf_release(&newlines);
2899 free(preimage.line_allocated);
2900 free(postimage.line_allocated);
2902 return (applied_pos < 0);
2905 static int apply_binary_fragment(struct image *img, struct patch *patch)
2907 struct fragment *fragment = patch->fragments;
2908 unsigned long len;
2909 void *dst;
2911 if (!fragment)
2912 return error(_("missing binary patch data for '%s'"),
2913 patch->new_name ?
2914 patch->new_name :
2915 patch->old_name);
2917 /* Binary patch is irreversible without the optional second hunk */
2918 if (apply_in_reverse) {
2919 if (!fragment->next)
2920 return error("cannot reverse-apply a binary patch "
2921 "without the reverse hunk to '%s'",
2922 patch->new_name
2923 ? patch->new_name : patch->old_name);
2924 fragment = fragment->next;
2926 switch (fragment->binary_patch_method) {
2927 case BINARY_DELTA_DEFLATED:
2928 dst = patch_delta(img->buf, img->len, fragment->patch,
2929 fragment->size, &len);
2930 if (!dst)
2931 return -1;
2932 clear_image(img);
2933 img->buf = dst;
2934 img->len = len;
2935 return 0;
2936 case BINARY_LITERAL_DEFLATED:
2937 clear_image(img);
2938 img->len = fragment->size;
2939 img->buf = xmemdupz(fragment->patch, img->len);
2940 return 0;
2942 return -1;
2946 * Replace "img" with the result of applying the binary patch.
2947 * The binary patch data itself in patch->fragment is still kept
2948 * but the preimage prepared by the caller in "img" is freed here
2949 * or in the helper function apply_binary_fragment() this calls.
2951 static int apply_binary(struct image *img, struct patch *patch)
2953 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2954 unsigned char sha1[20];
2957 * For safety, we require patch index line to contain
2958 * full 40-byte textual SHA1 for old and new, at least for now.
2960 if (strlen(patch->old_sha1_prefix) != 40 ||
2961 strlen(patch->new_sha1_prefix) != 40 ||
2962 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2963 get_sha1_hex(patch->new_sha1_prefix, sha1))
2964 return error("cannot apply binary patch to '%s' "
2965 "without full index line", name);
2967 if (patch->old_name) {
2969 * See if the old one matches what the patch
2970 * applies to.
2972 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2973 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2974 return error("the patch applies to '%s' (%s), "
2975 "which does not match the "
2976 "current contents.",
2977 name, sha1_to_hex(sha1));
2979 else {
2980 /* Otherwise, the old one must be empty. */
2981 if (img->len)
2982 return error("the patch applies to an empty "
2983 "'%s' but it is not empty", name);
2986 get_sha1_hex(patch->new_sha1_prefix, sha1);
2987 if (is_null_sha1(sha1)) {
2988 clear_image(img);
2989 return 0; /* deletion patch */
2992 if (has_sha1_file(sha1)) {
2993 /* We already have the postimage */
2994 enum object_type type;
2995 unsigned long size;
2996 char *result;
2998 result = read_sha1_file(sha1, &type, &size);
2999 if (!result)
3000 return error("the necessary postimage %s for "
3001 "'%s' cannot be read",
3002 patch->new_sha1_prefix, name);
3003 clear_image(img);
3004 img->buf = result;
3005 img->len = size;
3006 } else {
3008 * We have verified buf matches the preimage;
3009 * apply the patch data to it, which is stored
3010 * in the patch->fragments->{patch,size}.
3012 if (apply_binary_fragment(img, patch))
3013 return error(_("binary patch does not apply to '%s'"),
3014 name);
3016 /* verify that the result matches */
3017 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3018 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3019 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3020 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3023 return 0;
3026 static int apply_fragments(struct image *img, struct patch *patch)
3028 struct fragment *frag = patch->fragments;
3029 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3030 unsigned ws_rule = patch->ws_rule;
3031 unsigned inaccurate_eof = patch->inaccurate_eof;
3032 int nth = 0;
3034 if (patch->is_binary)
3035 return apply_binary(img, patch);
3037 while (frag) {
3038 nth++;
3039 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {
3040 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3041 if (!apply_with_reject)
3042 return -1;
3043 frag->rejected = 1;
3045 frag = frag->next;
3047 return 0;
3050 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3052 if (S_ISGITLINK(mode)) {
3053 strbuf_grow(buf, 100);
3054 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3055 } else {
3056 enum object_type type;
3057 unsigned long sz;
3058 char *result;
3060 result = read_sha1_file(sha1, &type, &sz);
3061 if (!result)
3062 return -1;
3063 /* XXX read_sha1_file NUL-terminates */
3064 strbuf_attach(buf, result, sz, sz + 1);
3066 return 0;
3069 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3071 if (!ce)
3072 return 0;
3073 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3076 static struct patch *in_fn_table(const char *name)
3078 struct string_list_item *item;
3080 if (name == NULL)
3081 return NULL;
3083 item = string_list_lookup(&fn_table, name);
3084 if (item != NULL)
3085 return (struct patch *)item->util;
3087 return NULL;
3091 * item->util in the filename table records the status of the path.
3092 * Usually it points at a patch (whose result records the contents
3093 * of it after applying it), but it could be PATH_WAS_DELETED for a
3094 * path that a previously applied patch has already removed, or
3095 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3097 * The latter is needed to deal with a case where two paths A and B
3098 * are swapped by first renaming A to B and then renaming B to A;
3099 * moving A to B should not be prevented due to presence of B as we
3100 * will remove it in a later patch.
3102 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3103 #define PATH_WAS_DELETED ((struct patch *) -1)
3105 static int to_be_deleted(struct patch *patch)
3107 return patch == PATH_TO_BE_DELETED;
3110 static int was_deleted(struct patch *patch)
3112 return patch == PATH_WAS_DELETED;
3115 static void add_to_fn_table(struct patch *patch)
3117 struct string_list_item *item;
3120 * Always add new_name unless patch is a deletion
3121 * This should cover the cases for normal diffs,
3122 * file creations and copies
3124 if (patch->new_name != NULL) {
3125 item = string_list_insert(&fn_table, patch->new_name);
3126 item->util = patch;
3130 * store a failure on rename/deletion cases because
3131 * later chunks shouldn't patch old names
3133 if ((patch->new_name == NULL) || (patch->is_rename)) {
3134 item = string_list_insert(&fn_table, patch->old_name);
3135 item->util = PATH_WAS_DELETED;
3139 static void prepare_fn_table(struct patch *patch)
3142 * store information about incoming file deletion
3144 while (patch) {
3145 if ((patch->new_name == NULL) || (patch->is_rename)) {
3146 struct string_list_item *item;
3147 item = string_list_insert(&fn_table, patch->old_name);
3148 item->util = PATH_TO_BE_DELETED;
3150 patch = patch->next;
3154 static int checkout_target(struct index_state *istate,
3155 struct cache_entry *ce, struct stat *st)
3157 struct checkout costate;
3159 memset(&costate, 0, sizeof(costate));
3160 costate.base_dir = "";
3161 costate.refresh_cache = 1;
3162 costate.istate = istate;
3163 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3164 return error(_("cannot checkout %s"), ce->name);
3165 return 0;
3168 static struct patch *previous_patch(struct patch *patch, int *gone)
3170 struct patch *previous;
3172 *gone = 0;
3173 if (patch->is_copy || patch->is_rename)
3174 return NULL; /* "git" patches do not depend on the order */
3176 previous = in_fn_table(patch->old_name);
3177 if (!previous)
3178 return NULL;
3180 if (to_be_deleted(previous))
3181 return NULL; /* the deletion hasn't happened yet */
3183 if (was_deleted(previous))
3184 *gone = 1;
3186 return previous;
3189 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3191 if (S_ISGITLINK(ce->ce_mode)) {
3192 if (!S_ISDIR(st->st_mode))
3193 return -1;
3194 return 0;
3196 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3199 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3201 static int load_patch_target(struct strbuf *buf,
3202 const struct cache_entry *ce,
3203 struct stat *st,
3204 const char *name,
3205 unsigned expected_mode)
3207 if (cached) {
3208 if (read_file_or_gitlink(ce, buf))
3209 return error(_("read of %s failed"), name);
3210 } else if (name) {
3211 if (S_ISGITLINK(expected_mode)) {
3212 if (ce)
3213 return read_file_or_gitlink(ce, buf);
3214 else
3215 return SUBMODULE_PATCH_WITHOUT_INDEX;
3216 } else {
3217 if (read_old_data(st, name, buf))
3218 return error(_("read of %s failed"), name);
3221 return 0;
3225 * We are about to apply "patch"; populate the "image" with the
3226 * current version we have, from the working tree or from the index,
3227 * depending on the situation e.g. --cached/--index. If we are
3228 * applying a non-git patch that incrementally updates the tree,
3229 * we read from the result of a previous diff.
3231 static int load_preimage(struct image *image,
3232 struct patch *patch, struct stat *st,
3233 const struct cache_entry *ce)
3235 struct strbuf buf = STRBUF_INIT;
3236 size_t len;
3237 char *img;
3238 struct patch *previous;
3239 int status;
3241 previous = previous_patch(patch, &status);
3242 if (status)
3243 return error(_("path %s has been renamed/deleted"),
3244 patch->old_name);
3245 if (previous) {
3246 /* We have a patched copy in memory; use that. */
3247 strbuf_add(&buf, previous->result, previous->resultsize);
3248 } else {
3249 status = load_patch_target(&buf, ce, st,
3250 patch->old_name, patch->old_mode);
3251 if (status < 0)
3252 return status;
3253 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3255 * There is no way to apply subproject
3256 * patch without looking at the index.
3257 * NEEDSWORK: shouldn't this be flagged
3258 * as an error???
3260 free_fragment_list(patch->fragments);
3261 patch->fragments = NULL;
3262 } else if (status) {
3263 return error(_("read of %s failed"), patch->old_name);
3267 img = strbuf_detach(&buf, &len);
3268 prepare_image(image, img, len, !patch->is_binary);
3269 return 0;
3272 static int three_way_merge(struct image *image,
3273 char *path,
3274 const unsigned char *base,
3275 const unsigned char *ours,
3276 const unsigned char *theirs)
3278 mmfile_t base_file, our_file, their_file;
3279 mmbuffer_t result = { NULL };
3280 int status;
3282 read_mmblob(&base_file, base);
3283 read_mmblob(&our_file, ours);
3284 read_mmblob(&their_file, theirs);
3285 status = ll_merge(&result, path,
3286 &base_file, "base",
3287 &our_file, "ours",
3288 &their_file, "theirs", NULL);
3289 free(base_file.ptr);
3290 free(our_file.ptr);
3291 free(their_file.ptr);
3292 if (status < 0 || !result.ptr) {
3293 free(result.ptr);
3294 return -1;
3296 clear_image(image);
3297 image->buf = result.ptr;
3298 image->len = result.size;
3300 return status;
3304 * When directly falling back to add/add three-way merge, we read from
3305 * the current contents of the new_name. In no cases other than that
3306 * this function will be called.
3308 static int load_current(struct image *image, struct patch *patch)
3310 struct strbuf buf = STRBUF_INIT;
3311 int status, pos;
3312 size_t len;
3313 char *img;
3314 struct stat st;
3315 struct cache_entry *ce;
3316 char *name = patch->new_name;
3317 unsigned mode = patch->new_mode;
3319 if (!patch->is_new)
3320 die("BUG: patch to %s is not a creation", patch->old_name);
3322 pos = cache_name_pos(name, strlen(name));
3323 if (pos < 0)
3324 return error(_("%s: does not exist in index"), name);
3325 ce = active_cache[pos];
3326 if (lstat(name, &st)) {
3327 if (errno != ENOENT)
3328 return error(_("%s: %s"), name, strerror(errno));
3329 if (checkout_target(&the_index, ce, &st))
3330 return -1;
3332 if (verify_index_match(ce, &st))
3333 return error(_("%s: does not match index"), name);
3335 status = load_patch_target(&buf, ce, &st, name, mode);
3336 if (status < 0)
3337 return status;
3338 else if (status)
3339 return -1;
3340 img = strbuf_detach(&buf, &len);
3341 prepare_image(image, img, len, !patch->is_binary);
3342 return 0;
3345 static int try_threeway(struct image *image, struct patch *patch,
3346 struct stat *st, const struct cache_entry *ce)
3348 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3349 struct strbuf buf = STRBUF_INIT;
3350 size_t len;
3351 int status;
3352 char *img;
3353 struct image tmp_image;
3355 /* No point falling back to 3-way merge in these cases */
3356 if (patch->is_delete ||
3357 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3358 return -1;
3360 /* Preimage the patch was prepared for */
3361 if (patch->is_new)
3362 write_sha1_file("", 0, blob_type, pre_sha1);
3363 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3364 read_blob_object(&buf, pre_sha1, patch->old_mode))
3365 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3367 fprintf(stderr, "Falling back to three-way merge...\n");
3369 img = strbuf_detach(&buf, &len);
3370 prepare_image(&tmp_image, img, len, 1);
3371 /* Apply the patch to get the post image */
3372 if (apply_fragments(&tmp_image, patch) < 0) {
3373 clear_image(&tmp_image);
3374 return -1;
3376 /* post_sha1[] is theirs */
3377 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3378 clear_image(&tmp_image);
3380 /* our_sha1[] is ours */
3381 if (patch->is_new) {
3382 if (load_current(&tmp_image, patch))
3383 return error("cannot read the current contents of '%s'",
3384 patch->new_name);
3385 } else {
3386 if (load_preimage(&tmp_image, patch, st, ce))
3387 return error("cannot read the current contents of '%s'",
3388 patch->old_name);
3390 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3391 clear_image(&tmp_image);
3393 /* in-core three-way merge between post and our using pre as base */
3394 status = three_way_merge(image, patch->new_name,
3395 pre_sha1, our_sha1, post_sha1);
3396 if (status < 0) {
3397 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3398 return status;
3401 if (status) {
3402 patch->conflicted_threeway = 1;
3403 if (patch->is_new)
3404 hashclr(patch->threeway_stage[0]);
3405 else
3406 hashcpy(patch->threeway_stage[0], pre_sha1);
3407 hashcpy(patch->threeway_stage[1], our_sha1);
3408 hashcpy(patch->threeway_stage[2], post_sha1);
3409 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3410 } else {
3411 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3413 return 0;
3416 static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)
3418 struct image image;
3420 if (load_preimage(&image, patch, st, ce) < 0)
3421 return -1;
3423 if (patch->direct_to_threeway ||
3424 apply_fragments(&image, patch) < 0) {
3425 /* Note: with --reject, apply_fragments() returns 0 */
3426 if (!threeway || try_threeway(&image, patch, st, ce) < 0)
3427 return -1;
3429 patch->result = image.buf;
3430 patch->resultsize = image.len;
3431 add_to_fn_table(patch);
3432 free(image.line_allocated);
3434 if (0 < patch->is_delete && patch->resultsize)
3435 return error(_("removal patch leaves file contents"));
3437 return 0;
3441 * If "patch" that we are looking at modifies or deletes what we have,
3442 * we would want it not to lose any local modification we have, either
3443 * in the working tree or in the index.
3445 * This also decides if a non-git patch is a creation patch or a
3446 * modification to an existing empty file. We do not check the state
3447 * of the current tree for a creation patch in this function; the caller
3448 * check_patch() separately makes sure (and errors out otherwise) that
3449 * the path the patch creates does not exist in the current tree.
3451 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
3453 const char *old_name = patch->old_name;
3454 struct patch *previous = NULL;
3455 int stat_ret = 0, status;
3456 unsigned st_mode = 0;
3458 if (!old_name)
3459 return 0;
3461 assert(patch->is_new <= 0);
3462 previous = previous_patch(patch, &status);
3464 if (status)
3465 return error(_("path %s has been renamed/deleted"), old_name);
3466 if (previous) {
3467 st_mode = previous->new_mode;
3468 } else if (!cached) {
3469 stat_ret = lstat(old_name, st);
3470 if (stat_ret && errno != ENOENT)
3471 return error(_("%s: %s"), old_name, strerror(errno));
3474 if (check_index && !previous) {
3475 int pos = cache_name_pos(old_name, strlen(old_name));
3476 if (pos < 0) {
3477 if (patch->is_new < 0)
3478 goto is_new;
3479 return error(_("%s: does not exist in index"), old_name);
3481 *ce = active_cache[pos];
3482 if (stat_ret < 0) {
3483 if (checkout_target(&the_index, *ce, st))
3484 return -1;
3486 if (!cached && verify_index_match(*ce, st))
3487 return error(_("%s: does not match index"), old_name);
3488 if (cached)
3489 st_mode = (*ce)->ce_mode;
3490 } else if (stat_ret < 0) {
3491 if (patch->is_new < 0)
3492 goto is_new;
3493 return error(_("%s: %s"), old_name, strerror(errno));
3496 if (!cached && !previous)
3497 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3499 if (patch->is_new < 0)
3500 patch->is_new = 0;
3501 if (!patch->old_mode)
3502 patch->old_mode = st_mode;
3503 if ((st_mode ^ patch->old_mode) & S_IFMT)
3504 return error(_("%s: wrong type"), old_name);
3505 if (st_mode != patch->old_mode)
3506 warning(_("%s has type %o, expected %o"),
3507 old_name, st_mode, patch->old_mode);
3508 if (!patch->new_mode && !patch->is_delete)
3509 patch->new_mode = st_mode;
3510 return 0;
3512 is_new:
3513 patch->is_new = 1;
3514 patch->is_delete = 0;
3515 free(patch->old_name);
3516 patch->old_name = NULL;
3517 return 0;
3521 #define EXISTS_IN_INDEX 1
3522 #define EXISTS_IN_WORKTREE 2
3524 static int check_to_create(const char *new_name, int ok_if_exists)
3526 struct stat nst;
3528 if (check_index &&
3529 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3530 !ok_if_exists)
3531 return EXISTS_IN_INDEX;
3532 if (cached)
3533 return 0;
3535 if (!lstat(new_name, &nst)) {
3536 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3537 return 0;
3539 * A leading component of new_name might be a symlink
3540 * that is going to be removed with this patch, but
3541 * still pointing at somewhere that has the path.
3542 * In such a case, path "new_name" does not exist as
3543 * far as git is concerned.
3545 if (has_symlink_leading_path(new_name, strlen(new_name)))
3546 return 0;
3548 return EXISTS_IN_WORKTREE;
3549 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3550 return error("%s: %s", new_name, strerror(errno));
3552 return 0;
3556 * Check and apply the patch in-core; leave the result in patch->result
3557 * for the caller to write it out to the final destination.
3559 static int check_patch(struct patch *patch)
3561 struct stat st;
3562 const char *old_name = patch->old_name;
3563 const char *new_name = patch->new_name;
3564 const char *name = old_name ? old_name : new_name;
3565 struct cache_entry *ce = NULL;
3566 struct patch *tpatch;
3567 int ok_if_exists;
3568 int status;
3570 patch->rejected = 1; /* we will drop this after we succeed */
3572 status = check_preimage(patch, &ce, &st);
3573 if (status)
3574 return status;
3575 old_name = patch->old_name;
3578 * A type-change diff is always split into a patch to delete
3579 * old, immediately followed by a patch to create new (see
3580 * diff.c::run_diff()); in such a case it is Ok that the entry
3581 * to be deleted by the previous patch is still in the working
3582 * tree and in the index.
3584 * A patch to swap-rename between A and B would first rename A
3585 * to B and then rename B to A. While applying the first one,
3586 * the presence of B should not stop A from getting renamed to
3587 * B; ask to_be_deleted() about the later rename. Removal of
3588 * B and rename from A to B is handled the same way by asking
3589 * was_deleted().
3591 if ((tpatch = in_fn_table(new_name)) &&
3592 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3593 ok_if_exists = 1;
3594 else
3595 ok_if_exists = 0;
3597 if (new_name &&
3598 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3599 int err = check_to_create(new_name, ok_if_exists);
3601 if (err && threeway) {
3602 patch->direct_to_threeway = 1;
3603 } else switch (err) {
3604 case 0:
3605 break; /* happy */
3606 case EXISTS_IN_INDEX:
3607 return error(_("%s: already exists in index"), new_name);
3608 break;
3609 case EXISTS_IN_WORKTREE:
3610 return error(_("%s: already exists in working directory"),
3611 new_name);
3612 default:
3613 return err;
3616 if (!patch->new_mode) {
3617 if (0 < patch->is_new)
3618 patch->new_mode = S_IFREG | 0644;
3619 else
3620 patch->new_mode = patch->old_mode;
3624 if (new_name && old_name) {
3625 int same = !strcmp(old_name, new_name);
3626 if (!patch->new_mode)
3627 patch->new_mode = patch->old_mode;
3628 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3629 if (same)
3630 return error(_("new mode (%o) of %s does not "
3631 "match old mode (%o)"),
3632 patch->new_mode, new_name,
3633 patch->old_mode);
3634 else
3635 return error(_("new mode (%o) of %s does not "
3636 "match old mode (%o) of %s"),
3637 patch->new_mode, new_name,
3638 patch->old_mode, old_name);
3642 if (apply_data(patch, &st, ce) < 0)
3643 return error(_("%s: patch does not apply"), name);
3644 patch->rejected = 0;
3645 return 0;
3648 static int check_patch_list(struct patch *patch)
3650 int err = 0;
3652 prepare_fn_table(patch);
3653 while (patch) {
3654 if (apply_verbosely)
3655 say_patch_name(stderr,
3656 _("Checking patch %s..."), patch);
3657 err |= check_patch(patch);
3658 patch = patch->next;
3660 return err;
3663 /* This function tries to read the sha1 from the current index */
3664 static int get_current_sha1(const char *path, unsigned char *sha1)
3666 int pos;
3668 if (read_cache() < 0)
3669 return -1;
3670 pos = cache_name_pos(path, strlen(path));
3671 if (pos < 0)
3672 return -1;
3673 hashcpy(sha1, active_cache[pos]->sha1);
3674 return 0;
3677 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3680 * A usable gitlink patch has only one fragment (hunk) that looks like:
3681 * @@ -1 +1 @@
3682 * -Subproject commit <old sha1>
3683 * +Subproject commit <new sha1>
3684 * or
3685 * @@ -1 +0,0 @@
3686 * -Subproject commit <old sha1>
3687 * for a removal patch.
3689 struct fragment *hunk = p->fragments;
3690 static const char heading[] = "-Subproject commit ";
3691 char *preimage;
3693 if (/* does the patch have only one hunk? */
3694 hunk && !hunk->next &&
3695 /* is its preimage one line? */
3696 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3697 /* does preimage begin with the heading? */
3698 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3699 starts_with(++preimage, heading) &&
3700 /* does it record full SHA-1? */
3701 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3702 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3703 /* does the abbreviated name on the index line agree with it? */
3704 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3705 return 0; /* it all looks fine */
3707 /* we may have full object name on the index line */
3708 return get_sha1_hex(p->old_sha1_prefix, sha1);
3711 /* Build an index that contains the just the files needed for a 3way merge */
3712 static void build_fake_ancestor(struct patch *list, const char *filename)
3714 struct patch *patch;
3715 struct index_state result = { NULL };
3716 static struct lock_file lock;
3718 /* Once we start supporting the reverse patch, it may be
3719 * worth showing the new sha1 prefix, but until then...
3721 for (patch = list; patch; patch = patch->next) {
3722 unsigned char sha1[20];
3723 struct cache_entry *ce;
3724 const char *name;
3726 name = patch->old_name ? patch->old_name : patch->new_name;
3727 if (0 < patch->is_new)
3728 continue;
3730 if (S_ISGITLINK(patch->old_mode)) {
3731 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3732 ; /* ok, the textual part looks sane */
3733 else
3734 die("sha1 information is lacking or useless for submoule %s",
3735 name);
3736 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3737 ; /* ok */
3738 } else if (!patch->lines_added && !patch->lines_deleted) {
3739 /* mode-only change: update the current */
3740 if (get_current_sha1(patch->old_name, sha1))
3741 die("mode change for %s, which is not "
3742 "in current HEAD", name);
3743 } else
3744 die("sha1 information is lacking or useless "
3745 "(%s).", name);
3747 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3748 if (!ce)
3749 die(_("make_cache_entry failed for path '%s'"), name);
3750 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3751 die ("Could not add %s to temporary index", name);
3754 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3755 if (write_locked_index(&result, &lock, COMMIT_LOCK))
3756 die ("Could not write temporary index to %s", filename);
3758 discard_index(&result);
3761 static void stat_patch_list(struct patch *patch)
3763 int files, adds, dels;
3765 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3766 files++;
3767 adds += patch->lines_added;
3768 dels += patch->lines_deleted;
3769 show_stats(patch);
3772 print_stat_summary(stdout, files, adds, dels);
3775 static void numstat_patch_list(struct patch *patch)
3777 for ( ; patch; patch = patch->next) {
3778 const char *name;
3779 name = patch->new_name ? patch->new_name : patch->old_name;
3780 if (patch->is_binary)
3781 printf("-\t-\t");
3782 else
3783 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3784 write_name_quoted(name, stdout, line_termination);
3788 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3790 if (mode)
3791 printf(" %s mode %06o %s\n", newdelete, mode, name);
3792 else
3793 printf(" %s %s\n", newdelete, name);
3796 static void show_mode_change(struct patch *p, int show_name)
3798 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
3799 if (show_name)
3800 printf(" mode change %06o => %06o %s\n",
3801 p->old_mode, p->new_mode, p->new_name);
3802 else
3803 printf(" mode change %06o => %06o\n",
3804 p->old_mode, p->new_mode);
3808 static void show_rename_copy(struct patch *p)
3810 const char *renamecopy = p->is_rename ? "rename" : "copy";
3811 const char *old, *new;
3813 /* Find common prefix */
3814 old = p->old_name;
3815 new = p->new_name;
3816 while (1) {
3817 const char *slash_old, *slash_new;
3818 slash_old = strchr(old, '/');
3819 slash_new = strchr(new, '/');
3820 if (!slash_old ||
3821 !slash_new ||
3822 slash_old - old != slash_new - new ||
3823 memcmp(old, new, slash_new - new))
3824 break;
3825 old = slash_old + 1;
3826 new = slash_new + 1;
3828 /* p->old_name thru old is the common prefix, and old and new
3829 * through the end of names are renames
3831 if (old != p->old_name)
3832 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
3833 (int)(old - p->old_name), p->old_name,
3834 old, new, p->score);
3835 else
3836 printf(" %s %s => %s (%d%%)\n", renamecopy,
3837 p->old_name, p->new_name, p->score);
3838 show_mode_change(p, 0);
3841 static void summary_patch_list(struct patch *patch)
3843 struct patch *p;
3845 for (p = patch; p; p = p->next) {
3846 if (p->is_new)
3847 show_file_mode_name("create", p->new_mode, p->new_name);
3848 else if (p->is_delete)
3849 show_file_mode_name("delete", p->old_mode, p->old_name);
3850 else {
3851 if (p->is_rename || p->is_copy)
3852 show_rename_copy(p);
3853 else {
3854 if (p->score) {
3855 printf(" rewrite %s (%d%%)\n",
3856 p->new_name, p->score);
3857 show_mode_change(p, 0);
3859 else
3860 show_mode_change(p, 1);
3866 static void patch_stats(struct patch *patch)
3868 int lines = patch->lines_added + patch->lines_deleted;
3870 if (lines > max_change)
3871 max_change = lines;
3872 if (patch->old_name) {
3873 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3874 if (!len)
3875 len = strlen(patch->old_name);
3876 if (len > max_len)
3877 max_len = len;
3879 if (patch->new_name) {
3880 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3881 if (!len)
3882 len = strlen(patch->new_name);
3883 if (len > max_len)
3884 max_len = len;
3888 static void remove_file(struct patch *patch, int rmdir_empty)
3890 if (update_index) {
3891 if (remove_file_from_cache(patch->old_name) < 0)
3892 die(_("unable to remove %s from index"), patch->old_name);
3894 if (!cached) {
3895 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
3896 remove_path(patch->old_name);
3901 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3903 struct stat st;
3904 struct cache_entry *ce;
3905 int namelen = strlen(path);
3906 unsigned ce_size = cache_entry_size(namelen);
3908 if (!update_index)
3909 return;
3911 ce = xcalloc(1, ce_size);
3912 memcpy(ce->name, path, namelen);
3913 ce->ce_mode = create_ce_mode(mode);
3914 ce->ce_flags = create_ce_flags(0);
3915 ce->ce_namelen = namelen;
3916 if (S_ISGITLINK(mode)) {
3917 const char *s;
3919 if (!skip_prefix(buf, "Subproject commit ", &s) ||
3920 get_sha1_hex(s, ce->sha1))
3921 die(_("corrupt patch for submodule %s"), path);
3922 } else {
3923 if (!cached) {
3924 if (lstat(path, &st) < 0)
3925 die_errno(_("unable to stat newly created file '%s'"),
3926 path);
3927 fill_stat_cache_info(ce, &st);
3929 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3930 die(_("unable to create backing store for newly created file %s"), path);
3932 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3933 die(_("unable to add cache entry for %s"), path);
3936 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3938 int fd;
3939 struct strbuf nbuf = STRBUF_INIT;
3941 if (S_ISGITLINK(mode)) {
3942 struct stat st;
3943 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3944 return 0;
3945 return mkdir(path, 0777);
3948 if (has_symlinks && S_ISLNK(mode))
3949 /* Although buf:size is counted string, it also is NUL
3950 * terminated.
3952 return symlink(buf, path);
3954 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3955 if (fd < 0)
3956 return -1;
3958 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3959 size = nbuf.len;
3960 buf = nbuf.buf;
3962 write_or_die(fd, buf, size);
3963 strbuf_release(&nbuf);
3965 if (close(fd) < 0)
3966 die_errno(_("closing file '%s'"), path);
3967 return 0;
3971 * We optimistically assume that the directories exist,
3972 * which is true 99% of the time anyway. If they don't,
3973 * we create them and try again.
3975 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3977 if (cached)
3978 return;
3979 if (!try_create_file(path, mode, buf, size))
3980 return;
3982 if (errno == ENOENT) {
3983 if (safe_create_leading_directories(path))
3984 return;
3985 if (!try_create_file(path, mode, buf, size))
3986 return;
3989 if (errno == EEXIST || errno == EACCES) {
3990 /* We may be trying to create a file where a directory
3991 * used to be.
3993 struct stat st;
3994 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3995 errno = EEXIST;
3998 if (errno == EEXIST) {
3999 unsigned int nr = getpid();
4001 for (;;) {
4002 char newpath[PATH_MAX];
4003 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4004 if (!try_create_file(newpath, mode, buf, size)) {
4005 if (!rename(newpath, path))
4006 return;
4007 unlink_or_warn(newpath);
4008 break;
4010 if (errno != EEXIST)
4011 break;
4012 ++nr;
4015 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4018 static void add_conflicted_stages_file(struct patch *patch)
4020 int stage, namelen;
4021 unsigned ce_size, mode;
4022 struct cache_entry *ce;
4024 if (!update_index)
4025 return;
4026 namelen = strlen(patch->new_name);
4027 ce_size = cache_entry_size(namelen);
4028 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4030 remove_file_from_cache(patch->new_name);
4031 for (stage = 1; stage < 4; stage++) {
4032 if (is_null_sha1(patch->threeway_stage[stage - 1]))
4033 continue;
4034 ce = xcalloc(1, ce_size);
4035 memcpy(ce->name, patch->new_name, namelen);
4036 ce->ce_mode = create_ce_mode(mode);
4037 ce->ce_flags = create_ce_flags(stage);
4038 ce->ce_namelen = namelen;
4039 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);
4040 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4041 die(_("unable to add cache entry for %s"), patch->new_name);
4045 static void create_file(struct patch *patch)
4047 char *path = patch->new_name;
4048 unsigned mode = patch->new_mode;
4049 unsigned long size = patch->resultsize;
4050 char *buf = patch->result;
4052 if (!mode)
4053 mode = S_IFREG | 0644;
4054 create_one_file(path, mode, buf, size);
4056 if (patch->conflicted_threeway)
4057 add_conflicted_stages_file(patch);
4058 else
4059 add_index_file(path, mode, buf, size);
4062 /* phase zero is to remove, phase one is to create */
4063 static void write_out_one_result(struct patch *patch, int phase)
4065 if (patch->is_delete > 0) {
4066 if (phase == 0)
4067 remove_file(patch, 1);
4068 return;
4070 if (patch->is_new > 0 || patch->is_copy) {
4071 if (phase == 1)
4072 create_file(patch);
4073 return;
4076 * Rename or modification boils down to the same
4077 * thing: remove the old, write the new
4079 if (phase == 0)
4080 remove_file(patch, patch->is_rename);
4081 if (phase == 1)
4082 create_file(patch);
4085 static int write_out_one_reject(struct patch *patch)
4087 FILE *rej;
4088 char namebuf[PATH_MAX];
4089 struct fragment *frag;
4090 int cnt = 0;
4091 struct strbuf sb = STRBUF_INIT;
4093 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4094 if (!frag->rejected)
4095 continue;
4096 cnt++;
4099 if (!cnt) {
4100 if (apply_verbosely)
4101 say_patch_name(stderr,
4102 _("Applied patch %s cleanly."), patch);
4103 return 0;
4106 /* This should not happen, because a removal patch that leaves
4107 * contents are marked "rejected" at the patch level.
4109 if (!patch->new_name)
4110 die(_("internal error"));
4112 /* Say this even without --verbose */
4113 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4114 "Applying patch %%s with %d rejects...",
4115 cnt),
4116 cnt);
4117 say_patch_name(stderr, sb.buf, patch);
4118 strbuf_release(&sb);
4120 cnt = strlen(patch->new_name);
4121 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4122 cnt = ARRAY_SIZE(namebuf) - 5;
4123 warning(_("truncating .rej filename to %.*s.rej"),
4124 cnt - 1, patch->new_name);
4126 memcpy(namebuf, patch->new_name, cnt);
4127 memcpy(namebuf + cnt, ".rej", 5);
4129 rej = fopen(namebuf, "w");
4130 if (!rej)
4131 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4133 /* Normal git tools never deal with .rej, so do not pretend
4134 * this is a git patch by saying --git or giving extended
4135 * headers. While at it, maybe please "kompare" that wants
4136 * the trailing TAB and some garbage at the end of line ;-).
4138 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4139 patch->new_name, patch->new_name);
4140 for (cnt = 1, frag = patch->fragments;
4141 frag;
4142 cnt++, frag = frag->next) {
4143 if (!frag->rejected) {
4144 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4145 continue;
4147 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4148 fprintf(rej, "%.*s", frag->size, frag->patch);
4149 if (frag->patch[frag->size-1] != '\n')
4150 fputc('\n', rej);
4152 fclose(rej);
4153 return -1;
4156 static int write_out_results(struct patch *list)
4158 int phase;
4159 int errs = 0;
4160 struct patch *l;
4161 struct string_list cpath = STRING_LIST_INIT_DUP;
4163 for (phase = 0; phase < 2; phase++) {
4164 l = list;
4165 while (l) {
4166 if (l->rejected)
4167 errs = 1;
4168 else {
4169 write_out_one_result(l, phase);
4170 if (phase == 1) {
4171 if (write_out_one_reject(l))
4172 errs = 1;
4173 if (l->conflicted_threeway) {
4174 string_list_append(&cpath, l->new_name);
4175 errs = 1;
4179 l = l->next;
4183 if (cpath.nr) {
4184 struct string_list_item *item;
4186 sort_string_list(&cpath);
4187 for_each_string_list_item(item, &cpath)
4188 fprintf(stderr, "U %s\n", item->string);
4189 string_list_clear(&cpath, 0);
4191 rerere(0);
4194 return errs;
4197 static struct lock_file lock_file;
4199 #define INACCURATE_EOF (1<<0)
4200 #define RECOUNT (1<<1)
4202 static int apply_patch(int fd, const char *filename, int options)
4204 size_t offset;
4205 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4206 struct patch *list = NULL, **listp = &list;
4207 int skipped_patch = 0;
4209 patch_input_file = filename;
4210 read_patch_file(&buf, fd);
4211 offset = 0;
4212 while (offset < buf.len) {
4213 struct patch *patch;
4214 int nr;
4216 patch = xcalloc(1, sizeof(*patch));
4217 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4218 patch->recount = !!(options & RECOUNT);
4219 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
4220 if (nr < 0)
4221 break;
4222 if (apply_in_reverse)
4223 reverse_patches(patch);
4224 if (use_patch(patch)) {
4225 patch_stats(patch);
4226 *listp = patch;
4227 listp = &patch->next;
4229 else {
4230 free_patch(patch);
4231 skipped_patch++;
4233 offset += nr;
4236 if (!list && !skipped_patch)
4237 die(_("unrecognized input"));
4239 if (whitespace_error && (ws_error_action == die_on_ws_error))
4240 apply = 0;
4242 update_index = check_index && apply;
4243 if (update_index && newfd < 0)
4244 newfd = hold_locked_index(&lock_file, 1);
4246 if (check_index) {
4247 if (read_cache() < 0)
4248 die(_("unable to read index file"));
4251 if ((check || apply) &&
4252 check_patch_list(list) < 0 &&
4253 !apply_with_reject)
4254 exit(1);
4256 if (apply && write_out_results(list)) {
4257 if (apply_with_reject)
4258 exit(1);
4259 /* with --3way, we still need to write the index out */
4260 return 1;
4263 if (fake_ancestor)
4264 build_fake_ancestor(list, fake_ancestor);
4266 if (diffstat)
4267 stat_patch_list(list);
4269 if (numstat)
4270 numstat_patch_list(list);
4272 if (summary)
4273 summary_patch_list(list);
4275 free_patch_list(list);
4276 strbuf_release(&buf);
4277 string_list_clear(&fn_table, 0);
4278 return 0;
4281 static void git_apply_config(void)
4283 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4284 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4285 git_config(git_default_config, NULL);
4288 static int option_parse_exclude(const struct option *opt,
4289 const char *arg, int unset)
4291 add_name_limit(arg, 1);
4292 return 0;
4295 static int option_parse_include(const struct option *opt,
4296 const char *arg, int unset)
4298 add_name_limit(arg, 0);
4299 has_include = 1;
4300 return 0;
4303 static int option_parse_p(const struct option *opt,
4304 const char *arg, int unset)
4306 p_value = atoi(arg);
4307 p_value_known = 1;
4308 return 0;
4311 static int option_parse_z(const struct option *opt,
4312 const char *arg, int unset)
4314 if (unset)
4315 line_termination = '\n';
4316 else
4317 line_termination = 0;
4318 return 0;
4321 static int option_parse_space_change(const struct option *opt,
4322 const char *arg, int unset)
4324 if (unset)
4325 ws_ignore_action = ignore_ws_none;
4326 else
4327 ws_ignore_action = ignore_ws_change;
4328 return 0;
4331 static int option_parse_whitespace(const struct option *opt,
4332 const char *arg, int unset)
4334 const char **whitespace_option = opt->value;
4336 *whitespace_option = arg;
4337 parse_whitespace_option(arg);
4338 return 0;
4341 static int option_parse_directory(const struct option *opt,
4342 const char *arg, int unset)
4344 root_len = strlen(arg);
4345 if (root_len && arg[root_len - 1] != '/') {
4346 char *new_root;
4347 root = new_root = xmalloc(root_len + 2);
4348 strcpy(new_root, arg);
4349 strcpy(new_root + root_len++, "/");
4350 } else
4351 root = arg;
4352 return 0;
4355 int cmd_apply(int argc, const char **argv, const char *prefix_)
4357 int i;
4358 int errs = 0;
4359 int is_not_gitdir = !startup_info->have_repository;
4360 int force_apply = 0;
4362 const char *whitespace_option = NULL;
4364 struct option builtin_apply_options[] = {
4365 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4366 N_("don't apply changes matching the given path"),
4367 0, option_parse_exclude },
4368 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4369 N_("apply changes matching the given path"),
4370 0, option_parse_include },
4371 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4372 N_("remove <num> leading slashes from traditional diff paths"),
4373 0, option_parse_p },
4374 OPT_BOOL(0, "no-add", &no_add,
4375 N_("ignore additions made by the patch")),
4376 OPT_BOOL(0, "stat", &diffstat,
4377 N_("instead of applying the patch, output diffstat for the input")),
4378 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4379 OPT_NOOP_NOARG(0, "binary"),
4380 OPT_BOOL(0, "numstat", &numstat,
4381 N_("show number of added and deleted lines in decimal notation")),
4382 OPT_BOOL(0, "summary", &summary,
4383 N_("instead of applying the patch, output a summary for the input")),
4384 OPT_BOOL(0, "check", &check,
4385 N_("instead of applying the patch, see if the patch is applicable")),
4386 OPT_BOOL(0, "index", &check_index,
4387 N_("make sure the patch is applicable to the current index")),
4388 OPT_BOOL(0, "cached", &cached,
4389 N_("apply a patch without touching the working tree")),
4390 OPT_BOOL(0, "apply", &force_apply,
4391 N_("also apply the patch (use with --stat/--summary/--check)")),
4392 OPT_BOOL('3', "3way", &threeway,
4393 N_( "attempt three-way merge if a patch does not apply")),
4394 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4395 N_("build a temporary index based on embedded index information")),
4396 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
4397 N_("paths are separated with NUL character"),
4398 PARSE_OPT_NOARG, option_parse_z },
4399 OPT_INTEGER('C', NULL, &p_context,
4400 N_("ensure at least <n> lines of context match")),
4401 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4402 N_("detect new or modified lines that have whitespace errors"),
4403 0, option_parse_whitespace },
4404 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4405 N_("ignore changes in whitespace when finding context"),
4406 PARSE_OPT_NOARG, option_parse_space_change },
4407 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4408 N_("ignore changes in whitespace when finding context"),
4409 PARSE_OPT_NOARG, option_parse_space_change },
4410 OPT_BOOL('R', "reverse", &apply_in_reverse,
4411 N_("apply the patch in reverse")),
4412 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,
4413 N_("don't expect at least one line of context")),
4414 OPT_BOOL(0, "reject", &apply_with_reject,
4415 N_("leave the rejected hunks in corresponding *.rej files")),
4416 OPT_BOOL(0, "allow-overlap", &allow_overlap,
4417 N_("allow overlapping hunks")),
4418 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),
4419 OPT_BIT(0, "inaccurate-eof", &options,
4420 N_("tolerate incorrectly detected missing new-line at the end of file"),
4421 INACCURATE_EOF),
4422 OPT_BIT(0, "recount", &options,
4423 N_("do not trust the line counts in the hunk headers"),
4424 RECOUNT),
4425 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4426 N_("prepend <root> to all filenames"),
4427 0, option_parse_directory },
4428 OPT_END()
4431 prefix = prefix_;
4432 prefix_length = prefix ? strlen(prefix) : 0;
4433 git_apply_config();
4434 if (apply_default_whitespace)
4435 parse_whitespace_option(apply_default_whitespace);
4436 if (apply_default_ignorewhitespace)
4437 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4439 argc = parse_options(argc, argv, prefix, builtin_apply_options,
4440 apply_usage, 0);
4442 if (apply_with_reject && threeway)
4443 die("--reject and --3way cannot be used together.");
4444 if (cached && threeway)
4445 die("--cached and --3way cannot be used together.");
4446 if (threeway) {
4447 if (is_not_gitdir)
4448 die(_("--3way outside a repository"));
4449 check_index = 1;
4451 if (apply_with_reject)
4452 apply = apply_verbosely = 1;
4453 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
4454 apply = 0;
4455 if (check_index && is_not_gitdir)
4456 die(_("--index outside a repository"));
4457 if (cached) {
4458 if (is_not_gitdir)
4459 die(_("--cached outside a repository"));
4460 check_index = 1;
4462 for (i = 0; i < argc; i++) {
4463 const char *arg = argv[i];
4464 int fd;
4466 if (!strcmp(arg, "-")) {
4467 errs |= apply_patch(0, "<stdin>", options);
4468 read_stdin = 0;
4469 continue;
4470 } else if (0 < prefix_length)
4471 arg = prefix_filename(prefix, prefix_length, arg);
4473 fd = open(arg, O_RDONLY);
4474 if (fd < 0)
4475 die_errno(_("can't open patch '%s'"), arg);
4476 read_stdin = 0;
4477 set_default_whitespace_mode(whitespace_option);
4478 errs |= apply_patch(fd, arg, options);
4479 close(fd);
4481 set_default_whitespace_mode(whitespace_option);
4482 if (read_stdin)
4483 errs |= apply_patch(0, "<stdin>", options);
4484 if (whitespace_error) {
4485 if (squelch_whitespace_errors &&
4486 squelch_whitespace_errors < whitespace_error) {
4487 int squelched =
4488 whitespace_error - squelch_whitespace_errors;
4489 warning(Q_("squelched %d whitespace error",
4490 "squelched %d whitespace errors",
4491 squelched),
4492 squelched);
4494 if (ws_error_action == die_on_ws_error)
4495 die(Q_("%d line adds whitespace errors.",
4496 "%d lines add whitespace errors.",
4497 whitespace_error),
4498 whitespace_error);
4499 if (applied_after_fixing_ws && apply)
4500 warning("%d line%s applied after"
4501 " fixing whitespace errors.",
4502 applied_after_fixing_ws,
4503 applied_after_fixing_ws == 1 ? "" : "s");
4504 else if (whitespace_error)
4505 warning(Q_("%d line adds whitespace errors.",
4506 "%d lines add whitespace errors.",
4507 whitespace_error),
4508 whitespace_error);
4511 if (update_index) {
4512 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4513 die(_("Unable to write new index file"));
4516 return !!errs;