diff --check: detect leftover conflict markers
[git/dscho.git] / builtin-apply.c
blob92f00471bbdeefaf987f77c407eea7cefa52280a
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static const char *fake_ancestor;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|fix|error|error-all>] <patch>...";
50 static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55 } ws_error_action = warn_on_ws_error;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
63 if (!option) {
64 ws_error_action = warn_on_ws_error;
65 return;
67 if (!strcmp(option, "warn")) {
68 ws_error_action = warn_on_ws_error;
69 return;
71 if (!strcmp(option, "nowarn")) {
72 ws_error_action = nowarn_ws_error;
73 return;
75 if (!strcmp(option, "error")) {
76 ws_error_action = die_on_ws_error;
77 return;
79 if (!strcmp(option, "error-all")) {
80 ws_error_action = die_on_ws_error;
81 squelch_whitespace_errors = 0;
82 return;
84 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
85 ws_error_action = correct_ws_error;
86 return;
88 die("unrecognized whitespace option '%s'", option);
91 static void set_default_whitespace_mode(const char *whitespace_option)
93 if (!whitespace_option && !apply_default_whitespace)
94 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
98 * For "diff-stat" like behaviour, we keep track of the biggest change
99 * we've seen, and the longest filename. That allows us to do simple
100 * scaling.
102 static int max_change, max_len;
105 * Various "current state", notably line numbers and what
106 * file (and how) we're patching right now.. The "is_xxxx"
107 * things are flags, where -1 means "don't know yet".
109 static int linenr = 1;
112 * This represents one "hunk" from a patch, starting with
113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
114 * patch text is pointed at by patch, and its byte length
115 * is stored in size. leading and trailing are the number
116 * of context lines.
118 struct fragment {
119 unsigned long leading, trailing;
120 unsigned long oldpos, oldlines;
121 unsigned long newpos, newlines;
122 const char *patch;
123 int size;
124 int rejected;
125 struct fragment *next;
129 * When dealing with a binary patch, we reuse "leading" field
130 * to store the type of the binary hunk, either deflated "delta"
131 * or deflated "literal".
133 #define binary_patch_method leading
134 #define BINARY_DELTA_DEFLATED 1
135 #define BINARY_LITERAL_DEFLATED 2
138 * This represents a "patch" to a file, both metainfo changes
139 * such as creation/deletion, filemode and content changes represented
140 * as a series of fragments.
142 struct patch {
143 char *new_name, *old_name, *def_name;
144 unsigned int old_mode, new_mode;
145 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
146 int rejected;
147 unsigned ws_rule;
148 unsigned long deflate_origlen;
149 int lines_added, lines_deleted;
150 int score;
151 unsigned int is_toplevel_relative:1;
152 unsigned int inaccurate_eof:1;
153 unsigned int is_binary:1;
154 unsigned int is_copy:1;
155 unsigned int is_rename:1;
156 struct fragment *fragments;
157 char *result;
158 size_t resultsize;
159 char old_sha1_prefix[41];
160 char new_sha1_prefix[41];
161 struct patch *next;
165 * A line in a file, len-bytes long (includes the terminating LF,
166 * except for an incomplete line at the end if the file ends with
167 * one), and its contents hashes to 'hash'.
169 struct line {
170 size_t len;
171 unsigned hash : 24;
172 unsigned flag : 8;
173 #define LINE_COMMON 1
177 * This represents a "file", which is an array of "lines".
179 struct image {
180 char *buf;
181 size_t len;
182 size_t nr;
183 size_t alloc;
184 struct line *line_allocated;
185 struct line *line;
188 static uint32_t hash_line(const char *cp, size_t len)
190 size_t i;
191 uint32_t h;
192 for (i = 0, h = 0; i < len; i++) {
193 if (!isspace(cp[i])) {
194 h = h * 3 + (cp[i] & 0xff);
197 return h;
200 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
202 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
203 img->line_allocated[img->nr].len = len;
204 img->line_allocated[img->nr].hash = hash_line(bol, len);
205 img->line_allocated[img->nr].flag = flag;
206 img->nr++;
209 static void prepare_image(struct image *image, char *buf, size_t len,
210 int prepare_linetable)
212 const char *cp, *ep;
214 memset(image, 0, sizeof(*image));
215 image->buf = buf;
216 image->len = len;
218 if (!prepare_linetable)
219 return;
221 ep = image->buf + image->len;
222 cp = image->buf;
223 while (cp < ep) {
224 const char *next;
225 for (next = cp; next < ep && *next != '\n'; next++)
227 if (next < ep)
228 next++;
229 add_line_info(image, cp, next - cp, 0);
230 cp = next;
232 image->line = image->line_allocated;
235 static void clear_image(struct image *image)
237 free(image->buf);
238 image->buf = NULL;
239 image->len = 0;
242 static void say_patch_name(FILE *output, const char *pre,
243 struct patch *patch, const char *post)
245 fputs(pre, output);
246 if (patch->old_name && patch->new_name &&
247 strcmp(patch->old_name, patch->new_name)) {
248 quote_c_style(patch->old_name, NULL, output, 0);
249 fputs(" => ", output);
250 quote_c_style(patch->new_name, NULL, output, 0);
251 } else {
252 const char *n = patch->new_name;
253 if (!n)
254 n = patch->old_name;
255 quote_c_style(n, NULL, output, 0);
257 fputs(post, output);
260 #define CHUNKSIZE (8192)
261 #define SLOP (16)
263 static void read_patch_file(struct strbuf *sb, int fd)
265 if (strbuf_read(sb, fd, 0) < 0)
266 die("git-apply: read returned %s", strerror(errno));
269 * Make sure that we have some slop in the buffer
270 * so that we can do speculative "memcmp" etc, and
271 * see to it that it is NUL-filled.
273 strbuf_grow(sb, SLOP);
274 memset(sb->buf + sb->len, 0, SLOP);
277 static unsigned long linelen(const char *buffer, unsigned long size)
279 unsigned long len = 0;
280 while (size--) {
281 len++;
282 if (*buffer++ == '\n')
283 break;
285 return len;
288 static int is_dev_null(const char *str)
290 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
293 #define TERM_SPACE 1
294 #define TERM_TAB 2
296 static int name_terminate(const char *name, int namelen, int c, int terminate)
298 if (c == ' ' && !(terminate & TERM_SPACE))
299 return 0;
300 if (c == '\t' && !(terminate & TERM_TAB))
301 return 0;
303 return 1;
306 static char *find_name(const char *line, char *def, int p_value, int terminate)
308 int len;
309 const char *start = line;
311 if (*line == '"') {
312 struct strbuf name;
315 * Proposed "new-style" GNU patch/diff format; see
316 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
318 strbuf_init(&name, 0);
319 if (!unquote_c_style(&name, line, NULL)) {
320 char *cp;
322 for (cp = name.buf; p_value; p_value--) {
323 cp = strchr(cp, '/');
324 if (!cp)
325 break;
326 cp++;
328 if (cp) {
329 /* name can later be freed, so we need
330 * to memmove, not just return cp
332 strbuf_remove(&name, 0, cp - name.buf);
333 free(def);
334 return strbuf_detach(&name, NULL);
337 strbuf_release(&name);
340 for (;;) {
341 char c = *line;
343 if (isspace(c)) {
344 if (c == '\n')
345 break;
346 if (name_terminate(start, line-start, c, terminate))
347 break;
349 line++;
350 if (c == '/' && !--p_value)
351 start = line;
353 if (!start)
354 return def;
355 len = line - start;
356 if (!len)
357 return def;
360 * Generally we prefer the shorter name, especially
361 * if the other one is just a variation of that with
362 * something else tacked on to the end (ie "file.orig"
363 * or "file~").
365 if (def) {
366 int deflen = strlen(def);
367 if (deflen < len && !strncmp(start, def, deflen))
368 return def;
369 free(def);
372 return xmemdupz(start, len);
375 static int count_slashes(const char *cp)
377 int cnt = 0;
378 char ch;
380 while ((ch = *cp++))
381 if (ch == '/')
382 cnt++;
383 return cnt;
387 * Given the string after "--- " or "+++ ", guess the appropriate
388 * p_value for the given patch.
390 static int guess_p_value(const char *nameline)
392 char *name, *cp;
393 int val = -1;
395 if (is_dev_null(nameline))
396 return -1;
397 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
398 if (!name)
399 return -1;
400 cp = strchr(name, '/');
401 if (!cp)
402 val = 0;
403 else if (prefix) {
405 * Does it begin with "a/$our-prefix" and such? Then this is
406 * very likely to apply to our directory.
408 if (!strncmp(name, prefix, prefix_length))
409 val = count_slashes(prefix);
410 else {
411 cp++;
412 if (!strncmp(cp, prefix, prefix_length))
413 val = count_slashes(prefix) + 1;
416 free(name);
417 return val;
421 * Get the name etc info from the ---/+++ lines of a traditional patch header
423 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
424 * files, we can happily check the index for a match, but for creating a
425 * new file we should try to match whatever "patch" does. I have no idea.
427 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
429 char *name;
431 first += 4; /* skip "--- " */
432 second += 4; /* skip "+++ " */
433 if (!p_value_known) {
434 int p, q;
435 p = guess_p_value(first);
436 q = guess_p_value(second);
437 if (p < 0) p = q;
438 if (0 <= p && p == q) {
439 p_value = p;
440 p_value_known = 1;
443 if (is_dev_null(first)) {
444 patch->is_new = 1;
445 patch->is_delete = 0;
446 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
447 patch->new_name = name;
448 } else if (is_dev_null(second)) {
449 patch->is_new = 0;
450 patch->is_delete = 1;
451 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
452 patch->old_name = name;
453 } else {
454 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
455 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
456 patch->old_name = patch->new_name = name;
458 if (!name)
459 die("unable to find filename in patch at line %d", linenr);
462 static int gitdiff_hdrend(const char *line, struct patch *patch)
464 return -1;
468 * We're anal about diff header consistency, to make
469 * sure that we don't end up having strange ambiguous
470 * patches floating around.
472 * As a result, gitdiff_{old|new}name() will check
473 * their names against any previous information, just
474 * to make sure..
476 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
478 if (!orig_name && !isnull)
479 return find_name(line, NULL, p_value, TERM_TAB);
481 if (orig_name) {
482 int len;
483 const char *name;
484 char *another;
485 name = orig_name;
486 len = strlen(name);
487 if (isnull)
488 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
489 another = find_name(line, NULL, p_value, TERM_TAB);
490 if (!another || memcmp(another, name, len))
491 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
492 free(another);
493 return orig_name;
495 else {
496 /* expect "/dev/null" */
497 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
498 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
499 return NULL;
503 static int gitdiff_oldname(const char *line, struct patch *patch)
505 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
506 return 0;
509 static int gitdiff_newname(const char *line, struct patch *patch)
511 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
512 return 0;
515 static int gitdiff_oldmode(const char *line, struct patch *patch)
517 patch->old_mode = strtoul(line, NULL, 8);
518 return 0;
521 static int gitdiff_newmode(const char *line, struct patch *patch)
523 patch->new_mode = strtoul(line, NULL, 8);
524 return 0;
527 static int gitdiff_delete(const char *line, struct patch *patch)
529 patch->is_delete = 1;
530 patch->old_name = patch->def_name;
531 return gitdiff_oldmode(line, patch);
534 static int gitdiff_newfile(const char *line, struct patch *patch)
536 patch->is_new = 1;
537 patch->new_name = patch->def_name;
538 return gitdiff_newmode(line, patch);
541 static int gitdiff_copysrc(const char *line, struct patch *patch)
543 patch->is_copy = 1;
544 patch->old_name = find_name(line, NULL, 0, 0);
545 return 0;
548 static int gitdiff_copydst(const char *line, struct patch *patch)
550 patch->is_copy = 1;
551 patch->new_name = find_name(line, NULL, 0, 0);
552 return 0;
555 static int gitdiff_renamesrc(const char *line, struct patch *patch)
557 patch->is_rename = 1;
558 patch->old_name = find_name(line, NULL, 0, 0);
559 return 0;
562 static int gitdiff_renamedst(const char *line, struct patch *patch)
564 patch->is_rename = 1;
565 patch->new_name = find_name(line, NULL, 0, 0);
566 return 0;
569 static int gitdiff_similarity(const char *line, struct patch *patch)
571 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
572 patch->score = 0;
573 return 0;
576 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
578 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
579 patch->score = 0;
580 return 0;
583 static int gitdiff_index(const char *line, struct patch *patch)
586 * index line is N hexadecimal, "..", N hexadecimal,
587 * and optional space with octal mode.
589 const char *ptr, *eol;
590 int len;
592 ptr = strchr(line, '.');
593 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
594 return 0;
595 len = ptr - line;
596 memcpy(patch->old_sha1_prefix, line, len);
597 patch->old_sha1_prefix[len] = 0;
599 line = ptr + 2;
600 ptr = strchr(line, ' ');
601 eol = strchr(line, '\n');
603 if (!ptr || eol < ptr)
604 ptr = eol;
605 len = ptr - line;
607 if (40 < len)
608 return 0;
609 memcpy(patch->new_sha1_prefix, line, len);
610 patch->new_sha1_prefix[len] = 0;
611 if (*ptr == ' ')
612 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
613 return 0;
617 * This is normal for a diff that doesn't change anything: we'll fall through
618 * into the next diff. Tell the parser to break out.
620 static int gitdiff_unrecognized(const char *line, struct patch *patch)
622 return -1;
625 static const char *stop_at_slash(const char *line, int llen)
627 int i;
629 for (i = 0; i < llen; i++) {
630 int ch = line[i];
631 if (ch == '/')
632 return line + i;
634 return NULL;
638 * This is to extract the same name that appears on "diff --git"
639 * line. We do not find and return anything if it is a rename
640 * patch, and it is OK because we will find the name elsewhere.
641 * We need to reliably find name only when it is mode-change only,
642 * creation or deletion of an empty file. In any of these cases,
643 * both sides are the same name under a/ and b/ respectively.
645 static char *git_header_name(char *line, int llen)
647 const char *name;
648 const char *second = NULL;
649 size_t len;
651 line += strlen("diff --git ");
652 llen -= strlen("diff --git ");
654 if (*line == '"') {
655 const char *cp;
656 struct strbuf first;
657 struct strbuf sp;
659 strbuf_init(&first, 0);
660 strbuf_init(&sp, 0);
662 if (unquote_c_style(&first, line, &second))
663 goto free_and_fail1;
665 /* advance to the first slash */
666 cp = stop_at_slash(first.buf, first.len);
667 /* we do not accept absolute paths */
668 if (!cp || cp == first.buf)
669 goto free_and_fail1;
670 strbuf_remove(&first, 0, cp + 1 - first.buf);
673 * second points at one past closing dq of name.
674 * find the second name.
676 while ((second < line + llen) && isspace(*second))
677 second++;
679 if (line + llen <= second)
680 goto free_and_fail1;
681 if (*second == '"') {
682 if (unquote_c_style(&sp, second, NULL))
683 goto free_and_fail1;
684 cp = stop_at_slash(sp.buf, sp.len);
685 if (!cp || cp == sp.buf)
686 goto free_and_fail1;
687 /* They must match, otherwise ignore */
688 if (strcmp(cp + 1, first.buf))
689 goto free_and_fail1;
690 strbuf_release(&sp);
691 return strbuf_detach(&first, NULL);
694 /* unquoted second */
695 cp = stop_at_slash(second, line + llen - second);
696 if (!cp || cp == second)
697 goto free_and_fail1;
698 cp++;
699 if (line + llen - cp != first.len + 1 ||
700 memcmp(first.buf, cp, first.len))
701 goto free_and_fail1;
702 return strbuf_detach(&first, NULL);
704 free_and_fail1:
705 strbuf_release(&first);
706 strbuf_release(&sp);
707 return NULL;
710 /* unquoted first name */
711 name = stop_at_slash(line, llen);
712 if (!name || name == line)
713 return NULL;
714 name++;
717 * since the first name is unquoted, a dq if exists must be
718 * the beginning of the second name.
720 for (second = name; second < line + llen; second++) {
721 if (*second == '"') {
722 struct strbuf sp;
723 const char *np;
725 strbuf_init(&sp, 0);
726 if (unquote_c_style(&sp, second, NULL))
727 goto free_and_fail2;
729 np = stop_at_slash(sp.buf, sp.len);
730 if (!np || np == sp.buf)
731 goto free_and_fail2;
732 np++;
734 len = sp.buf + sp.len - np;
735 if (len < second - name &&
736 !strncmp(np, name, len) &&
737 isspace(name[len])) {
738 /* Good */
739 strbuf_remove(&sp, 0, np - sp.buf);
740 return strbuf_detach(&sp, NULL);
743 free_and_fail2:
744 strbuf_release(&sp);
745 return NULL;
750 * Accept a name only if it shows up twice, exactly the same
751 * form.
753 for (len = 0 ; ; len++) {
754 switch (name[len]) {
755 default:
756 continue;
757 case '\n':
758 return NULL;
759 case '\t': case ' ':
760 second = name+len;
761 for (;;) {
762 char c = *second++;
763 if (c == '\n')
764 return NULL;
765 if (c == '/')
766 break;
768 if (second[len] == '\n' && !memcmp(name, second, len)) {
769 return xmemdupz(name, len);
775 /* Verify that we recognize the lines following a git header */
776 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
778 unsigned long offset;
780 /* A git diff has explicit new/delete information, so we don't guess */
781 patch->is_new = 0;
782 patch->is_delete = 0;
785 * Some things may not have the old name in the
786 * rest of the headers anywhere (pure mode changes,
787 * or removing or adding empty files), so we get
788 * the default name from the header.
790 patch->def_name = git_header_name(line, len);
792 line += len;
793 size -= len;
794 linenr++;
795 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
796 static const struct opentry {
797 const char *str;
798 int (*fn)(const char *, struct patch *);
799 } optable[] = {
800 { "@@ -", gitdiff_hdrend },
801 { "--- ", gitdiff_oldname },
802 { "+++ ", gitdiff_newname },
803 { "old mode ", gitdiff_oldmode },
804 { "new mode ", gitdiff_newmode },
805 { "deleted file mode ", gitdiff_delete },
806 { "new file mode ", gitdiff_newfile },
807 { "copy from ", gitdiff_copysrc },
808 { "copy to ", gitdiff_copydst },
809 { "rename old ", gitdiff_renamesrc },
810 { "rename new ", gitdiff_renamedst },
811 { "rename from ", gitdiff_renamesrc },
812 { "rename to ", gitdiff_renamedst },
813 { "similarity index ", gitdiff_similarity },
814 { "dissimilarity index ", gitdiff_dissimilarity },
815 { "index ", gitdiff_index },
816 { "", gitdiff_unrecognized },
818 int i;
820 len = linelen(line, size);
821 if (!len || line[len-1] != '\n')
822 break;
823 for (i = 0; i < ARRAY_SIZE(optable); i++) {
824 const struct opentry *p = optable + i;
825 int oplen = strlen(p->str);
826 if (len < oplen || memcmp(p->str, line, oplen))
827 continue;
828 if (p->fn(line + oplen, patch) < 0)
829 return offset;
830 break;
834 return offset;
837 static int parse_num(const char *line, unsigned long *p)
839 char *ptr;
841 if (!isdigit(*line))
842 return 0;
843 *p = strtoul(line, &ptr, 10);
844 return ptr - line;
847 static int parse_range(const char *line, int len, int offset, const char *expect,
848 unsigned long *p1, unsigned long *p2)
850 int digits, ex;
852 if (offset < 0 || offset >= len)
853 return -1;
854 line += offset;
855 len -= offset;
857 digits = parse_num(line, p1);
858 if (!digits)
859 return -1;
861 offset += digits;
862 line += digits;
863 len -= digits;
865 *p2 = 1;
866 if (*line == ',') {
867 digits = parse_num(line+1, p2);
868 if (!digits)
869 return -1;
871 offset += digits+1;
872 line += digits+1;
873 len -= digits+1;
876 ex = strlen(expect);
877 if (ex > len)
878 return -1;
879 if (memcmp(line, expect, ex))
880 return -1;
882 return offset + ex;
886 * Parse a unified diff fragment header of the
887 * form "@@ -a,b +c,d @@"
889 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
891 int offset;
893 if (!len || line[len-1] != '\n')
894 return -1;
896 /* Figure out the number of lines in a fragment */
897 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
898 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
900 return offset;
903 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
905 unsigned long offset, len;
907 patch->is_toplevel_relative = 0;
908 patch->is_rename = patch->is_copy = 0;
909 patch->is_new = patch->is_delete = -1;
910 patch->old_mode = patch->new_mode = 0;
911 patch->old_name = patch->new_name = NULL;
912 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
913 unsigned long nextlen;
915 len = linelen(line, size);
916 if (!len)
917 break;
919 /* Testing this early allows us to take a few shortcuts.. */
920 if (len < 6)
921 continue;
924 * Make sure we don't find any unconnected patch fragments.
925 * That's a sign that we didn't find a header, and that a
926 * patch has become corrupted/broken up.
928 if (!memcmp("@@ -", line, 4)) {
929 struct fragment dummy;
930 if (parse_fragment_header(line, len, &dummy) < 0)
931 continue;
932 die("patch fragment without header at line %d: %.*s",
933 linenr, (int)len-1, line);
936 if (size < len + 6)
937 break;
940 * Git patch? It might not have a real patch, just a rename
941 * or mode change, so we handle that specially
943 if (!memcmp("diff --git ", line, 11)) {
944 int git_hdr_len = parse_git_header(line, len, size, patch);
945 if (git_hdr_len <= len)
946 continue;
947 if (!patch->old_name && !patch->new_name) {
948 if (!patch->def_name)
949 die("git diff header lacks filename information (line %d)", linenr);
950 patch->old_name = patch->new_name = patch->def_name;
952 patch->is_toplevel_relative = 1;
953 *hdrsize = git_hdr_len;
954 return offset;
957 /* --- followed by +++ ? */
958 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
959 continue;
962 * We only accept unified patches, so we want it to
963 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
964 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
966 nextlen = linelen(line + len, size - len);
967 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
968 continue;
970 /* Ok, we'll consider it a patch */
971 parse_traditional_patch(line, line+len, patch);
972 *hdrsize = len + nextlen;
973 linenr += 2;
974 return offset;
976 return -1;
979 static void check_whitespace(const char *line, int len, unsigned ws_rule)
981 char *err;
982 unsigned result = ws_check(line + 1, len - 1, ws_rule);
983 if (!result)
984 return;
986 whitespace_error++;
987 if (squelch_whitespace_errors &&
988 squelch_whitespace_errors < whitespace_error)
990 else {
991 err = whitespace_error_string(result);
992 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
993 patch_input_file, linenr, err, len - 2, line + 1);
994 free(err);
999 * Parse a unified diff. Note that this really needs to parse each
1000 * fragment separately, since the only way to know the difference
1001 * between a "---" that is part of a patch, and a "---" that starts
1002 * the next patch is to look at the line counts..
1004 static int parse_fragment(char *line, unsigned long size,
1005 struct patch *patch, struct fragment *fragment)
1007 int added, deleted;
1008 int len = linelen(line, size), offset;
1009 unsigned long oldlines, newlines;
1010 unsigned long leading, trailing;
1012 offset = parse_fragment_header(line, len, fragment);
1013 if (offset < 0)
1014 return -1;
1015 oldlines = fragment->oldlines;
1016 newlines = fragment->newlines;
1017 leading = 0;
1018 trailing = 0;
1020 /* Parse the thing.. */
1021 line += len;
1022 size -= len;
1023 linenr++;
1024 added = deleted = 0;
1025 for (offset = len;
1026 0 < size;
1027 offset += len, size -= len, line += len, linenr++) {
1028 if (!oldlines && !newlines)
1029 break;
1030 len = linelen(line, size);
1031 if (!len || line[len-1] != '\n')
1032 return -1;
1033 switch (*line) {
1034 default:
1035 return -1;
1036 case '\n': /* newer GNU diff, an empty context line */
1037 case ' ':
1038 oldlines--;
1039 newlines--;
1040 if (!deleted && !added)
1041 leading++;
1042 trailing++;
1043 break;
1044 case '-':
1045 if (apply_in_reverse &&
1046 ws_error_action != nowarn_ws_error)
1047 check_whitespace(line, len, patch->ws_rule);
1048 deleted++;
1049 oldlines--;
1050 trailing = 0;
1051 break;
1052 case '+':
1053 if (!apply_in_reverse &&
1054 ws_error_action != nowarn_ws_error)
1055 check_whitespace(line, len, patch->ws_rule);
1056 added++;
1057 newlines--;
1058 trailing = 0;
1059 break;
1062 * We allow "\ No newline at end of file". Depending
1063 * on locale settings when the patch was produced we
1064 * don't know what this line looks like. The only
1065 * thing we do know is that it begins with "\ ".
1066 * Checking for 12 is just for sanity check -- any
1067 * l10n of "\ No newline..." is at least that long.
1069 case '\\':
1070 if (len < 12 || memcmp(line, "\\ ", 2))
1071 return -1;
1072 break;
1075 if (oldlines || newlines)
1076 return -1;
1077 fragment->leading = leading;
1078 fragment->trailing = trailing;
1081 * If a fragment ends with an incomplete line, we failed to include
1082 * it in the above loop because we hit oldlines == newlines == 0
1083 * before seeing it.
1085 if (12 < size && !memcmp(line, "\\ ", 2))
1086 offset += linelen(line, size);
1088 patch->lines_added += added;
1089 patch->lines_deleted += deleted;
1091 if (0 < patch->is_new && oldlines)
1092 return error("new file depends on old contents");
1093 if (0 < patch->is_delete && newlines)
1094 return error("deleted file still has contents");
1095 return offset;
1098 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1100 unsigned long offset = 0;
1101 unsigned long oldlines = 0, newlines = 0, context = 0;
1102 struct fragment **fragp = &patch->fragments;
1104 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1105 struct fragment *fragment;
1106 int len;
1108 fragment = xcalloc(1, sizeof(*fragment));
1109 len = parse_fragment(line, size, patch, fragment);
1110 if (len <= 0)
1111 die("corrupt patch at line %d", linenr);
1112 fragment->patch = line;
1113 fragment->size = len;
1114 oldlines += fragment->oldlines;
1115 newlines += fragment->newlines;
1116 context += fragment->leading + fragment->trailing;
1118 *fragp = fragment;
1119 fragp = &fragment->next;
1121 offset += len;
1122 line += len;
1123 size -= len;
1127 * If something was removed (i.e. we have old-lines) it cannot
1128 * be creation, and if something was added it cannot be
1129 * deletion. However, the reverse is not true; --unified=0
1130 * patches that only add are not necessarily creation even
1131 * though they do not have any old lines, and ones that only
1132 * delete are not necessarily deletion.
1134 * Unfortunately, a real creation/deletion patch do _not_ have
1135 * any context line by definition, so we cannot safely tell it
1136 * apart with --unified=0 insanity. At least if the patch has
1137 * more than one hunk it is not creation or deletion.
1139 if (patch->is_new < 0 &&
1140 (oldlines || (patch->fragments && patch->fragments->next)))
1141 patch->is_new = 0;
1142 if (patch->is_delete < 0 &&
1143 (newlines || (patch->fragments && patch->fragments->next)))
1144 patch->is_delete = 0;
1146 if (0 < patch->is_new && oldlines)
1147 die("new file %s depends on old contents", patch->new_name);
1148 if (0 < patch->is_delete && newlines)
1149 die("deleted file %s still has contents", patch->old_name);
1150 if (!patch->is_delete && !newlines && context)
1151 fprintf(stderr, "** warning: file %s becomes empty but "
1152 "is not deleted\n", patch->new_name);
1154 return offset;
1157 static inline int metadata_changes(struct patch *patch)
1159 return patch->is_rename > 0 ||
1160 patch->is_copy > 0 ||
1161 patch->is_new > 0 ||
1162 patch->is_delete ||
1163 (patch->old_mode && patch->new_mode &&
1164 patch->old_mode != patch->new_mode);
1167 static char *inflate_it(const void *data, unsigned long size,
1168 unsigned long inflated_size)
1170 z_stream stream;
1171 void *out;
1172 int st;
1174 memset(&stream, 0, sizeof(stream));
1176 stream.next_in = (unsigned char *)data;
1177 stream.avail_in = size;
1178 stream.next_out = out = xmalloc(inflated_size);
1179 stream.avail_out = inflated_size;
1180 inflateInit(&stream);
1181 st = inflate(&stream, Z_FINISH);
1182 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1183 free(out);
1184 return NULL;
1186 return out;
1189 static struct fragment *parse_binary_hunk(char **buf_p,
1190 unsigned long *sz_p,
1191 int *status_p,
1192 int *used_p)
1195 * Expect a line that begins with binary patch method ("literal"
1196 * or "delta"), followed by the length of data before deflating.
1197 * a sequence of 'length-byte' followed by base-85 encoded data
1198 * should follow, terminated by a newline.
1200 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1201 * and we would limit the patch line to 66 characters,
1202 * so one line can fit up to 13 groups that would decode
1203 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1204 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1206 int llen, used;
1207 unsigned long size = *sz_p;
1208 char *buffer = *buf_p;
1209 int patch_method;
1210 unsigned long origlen;
1211 char *data = NULL;
1212 int hunk_size = 0;
1213 struct fragment *frag;
1215 llen = linelen(buffer, size);
1216 used = llen;
1218 *status_p = 0;
1220 if (!prefixcmp(buffer, "delta ")) {
1221 patch_method = BINARY_DELTA_DEFLATED;
1222 origlen = strtoul(buffer + 6, NULL, 10);
1224 else if (!prefixcmp(buffer, "literal ")) {
1225 patch_method = BINARY_LITERAL_DEFLATED;
1226 origlen = strtoul(buffer + 8, NULL, 10);
1228 else
1229 return NULL;
1231 linenr++;
1232 buffer += llen;
1233 while (1) {
1234 int byte_length, max_byte_length, newsize;
1235 llen = linelen(buffer, size);
1236 used += llen;
1237 linenr++;
1238 if (llen == 1) {
1239 /* consume the blank line */
1240 buffer++;
1241 size--;
1242 break;
1245 * Minimum line is "A00000\n" which is 7-byte long,
1246 * and the line length must be multiple of 5 plus 2.
1248 if ((llen < 7) || (llen-2) % 5)
1249 goto corrupt;
1250 max_byte_length = (llen - 2) / 5 * 4;
1251 byte_length = *buffer;
1252 if ('A' <= byte_length && byte_length <= 'Z')
1253 byte_length = byte_length - 'A' + 1;
1254 else if ('a' <= byte_length && byte_length <= 'z')
1255 byte_length = byte_length - 'a' + 27;
1256 else
1257 goto corrupt;
1258 /* if the input length was not multiple of 4, we would
1259 * have filler at the end but the filler should never
1260 * exceed 3 bytes
1262 if (max_byte_length < byte_length ||
1263 byte_length <= max_byte_length - 4)
1264 goto corrupt;
1265 newsize = hunk_size + byte_length;
1266 data = xrealloc(data, newsize);
1267 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1268 goto corrupt;
1269 hunk_size = newsize;
1270 buffer += llen;
1271 size -= llen;
1274 frag = xcalloc(1, sizeof(*frag));
1275 frag->patch = inflate_it(data, hunk_size, origlen);
1276 if (!frag->patch)
1277 goto corrupt;
1278 free(data);
1279 frag->size = origlen;
1280 *buf_p = buffer;
1281 *sz_p = size;
1282 *used_p = used;
1283 frag->binary_patch_method = patch_method;
1284 return frag;
1286 corrupt:
1287 free(data);
1288 *status_p = -1;
1289 error("corrupt binary patch at line %d: %.*s",
1290 linenr-1, llen-1, buffer);
1291 return NULL;
1294 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1297 * We have read "GIT binary patch\n"; what follows is a line
1298 * that says the patch method (currently, either "literal" or
1299 * "delta") and the length of data before deflating; a
1300 * sequence of 'length-byte' followed by base-85 encoded data
1301 * follows.
1303 * When a binary patch is reversible, there is another binary
1304 * hunk in the same format, starting with patch method (either
1305 * "literal" or "delta") with the length of data, and a sequence
1306 * of length-byte + base-85 encoded data, terminated with another
1307 * empty line. This data, when applied to the postimage, produces
1308 * the preimage.
1310 struct fragment *forward;
1311 struct fragment *reverse;
1312 int status;
1313 int used, used_1;
1315 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1316 if (!forward && !status)
1317 /* there has to be one hunk (forward hunk) */
1318 return error("unrecognized binary patch at line %d", linenr-1);
1319 if (status)
1320 /* otherwise we already gave an error message */
1321 return status;
1323 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1324 if (reverse)
1325 used += used_1;
1326 else if (status) {
1328 * Not having reverse hunk is not an error, but having
1329 * a corrupt reverse hunk is.
1331 free((void*) forward->patch);
1332 free(forward);
1333 return status;
1335 forward->next = reverse;
1336 patch->fragments = forward;
1337 patch->is_binary = 1;
1338 return used;
1341 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1343 int hdrsize, patchsize;
1344 int offset = find_header(buffer, size, &hdrsize, patch);
1346 if (offset < 0)
1347 return offset;
1349 patch->ws_rule = whitespace_rule(patch->new_name
1350 ? patch->new_name
1351 : patch->old_name);
1353 patchsize = parse_single_patch(buffer + offset + hdrsize,
1354 size - offset - hdrsize, patch);
1356 if (!patchsize) {
1357 static const char *binhdr[] = {
1358 "Binary files ",
1359 "Files ",
1360 NULL,
1362 static const char git_binary[] = "GIT binary patch\n";
1363 int i;
1364 int hd = hdrsize + offset;
1365 unsigned long llen = linelen(buffer + hd, size - hd);
1367 if (llen == sizeof(git_binary) - 1 &&
1368 !memcmp(git_binary, buffer + hd, llen)) {
1369 int used;
1370 linenr++;
1371 used = parse_binary(buffer + hd + llen,
1372 size - hd - llen, patch);
1373 if (used)
1374 patchsize = used + llen;
1375 else
1376 patchsize = 0;
1378 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1379 for (i = 0; binhdr[i]; i++) {
1380 int len = strlen(binhdr[i]);
1381 if (len < size - hd &&
1382 !memcmp(binhdr[i], buffer + hd, len)) {
1383 linenr++;
1384 patch->is_binary = 1;
1385 patchsize = llen;
1386 break;
1391 /* Empty patch cannot be applied if it is a text patch
1392 * without metadata change. A binary patch appears
1393 * empty to us here.
1395 if ((apply || check) &&
1396 (!patch->is_binary && !metadata_changes(patch)))
1397 die("patch with only garbage at line %d", linenr);
1400 return offset + hdrsize + patchsize;
1403 #define swap(a,b) myswap((a),(b),sizeof(a))
1405 #define myswap(a, b, size) do { \
1406 unsigned char mytmp[size]; \
1407 memcpy(mytmp, &a, size); \
1408 memcpy(&a, &b, size); \
1409 memcpy(&b, mytmp, size); \
1410 } while (0)
1412 static void reverse_patches(struct patch *p)
1414 for (; p; p = p->next) {
1415 struct fragment *frag = p->fragments;
1417 swap(p->new_name, p->old_name);
1418 swap(p->new_mode, p->old_mode);
1419 swap(p->is_new, p->is_delete);
1420 swap(p->lines_added, p->lines_deleted);
1421 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1423 for (; frag; frag = frag->next) {
1424 swap(frag->newpos, frag->oldpos);
1425 swap(frag->newlines, frag->oldlines);
1430 static const char pluses[] =
1431 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1432 static const char minuses[]=
1433 "----------------------------------------------------------------------";
1435 static void show_stats(struct patch *patch)
1437 struct strbuf qname;
1438 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1439 int max, add, del;
1441 strbuf_init(&qname, 0);
1442 quote_c_style(cp, &qname, NULL, 0);
1445 * "scale" the filename
1447 max = max_len;
1448 if (max > 50)
1449 max = 50;
1451 if (qname.len > max) {
1452 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1453 if (!cp)
1454 cp = qname.buf + qname.len + 3 - max;
1455 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1458 if (patch->is_binary) {
1459 printf(" %-*s | Bin\n", max, qname.buf);
1460 strbuf_release(&qname);
1461 return;
1464 printf(" %-*s |", max, qname.buf);
1465 strbuf_release(&qname);
1468 * scale the add/delete
1470 max = max + max_change > 70 ? 70 - max : max_change;
1471 add = patch->lines_added;
1472 del = patch->lines_deleted;
1474 if (max_change > 0) {
1475 int total = ((add + del) * max + max_change / 2) / max_change;
1476 add = (add * max + max_change / 2) / max_change;
1477 del = total - add;
1479 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1480 add, pluses, del, minuses);
1483 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1485 switch (st->st_mode & S_IFMT) {
1486 case S_IFLNK:
1487 strbuf_grow(buf, st->st_size);
1488 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1489 return -1;
1490 strbuf_setlen(buf, st->st_size);
1491 return 0;
1492 case S_IFREG:
1493 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1494 return error("unable to open or read %s", path);
1495 convert_to_git(path, buf->buf, buf->len, buf, 0);
1496 return 0;
1497 default:
1498 return -1;
1502 static void update_pre_post_images(struct image *preimage,
1503 struct image *postimage,
1504 char *buf,
1505 size_t len)
1507 int i, ctx;
1508 char *new, *old, *fixed;
1509 struct image fixed_preimage;
1512 * Update the preimage with whitespace fixes. Note that we
1513 * are not losing preimage->buf -- apply_one_fragment() will
1514 * free "oldlines".
1516 prepare_image(&fixed_preimage, buf, len, 1);
1517 assert(fixed_preimage.nr == preimage->nr);
1518 for (i = 0; i < preimage->nr; i++)
1519 fixed_preimage.line[i].flag = preimage->line[i].flag;
1520 free(preimage->line_allocated);
1521 *preimage = fixed_preimage;
1524 * Adjust the common context lines in postimage, in place.
1525 * This is possible because whitespace fixing does not make
1526 * the string grow.
1528 new = old = postimage->buf;
1529 fixed = preimage->buf;
1530 for (i = ctx = 0; i < postimage->nr; i++) {
1531 size_t len = postimage->line[i].len;
1532 if (!(postimage->line[i].flag & LINE_COMMON)) {
1533 /* an added line -- no counterparts in preimage */
1534 memmove(new, old, len);
1535 old += len;
1536 new += len;
1537 continue;
1540 /* a common context -- skip it in the original postimage */
1541 old += len;
1543 /* and find the corresponding one in the fixed preimage */
1544 while (ctx < preimage->nr &&
1545 !(preimage->line[ctx].flag & LINE_COMMON)) {
1546 fixed += preimage->line[ctx].len;
1547 ctx++;
1549 if (preimage->nr <= ctx)
1550 die("oops");
1552 /* and copy it in, while fixing the line length */
1553 len = preimage->line[ctx].len;
1554 memcpy(new, fixed, len);
1555 new += len;
1556 fixed += len;
1557 postimage->line[i].len = len;
1558 ctx++;
1561 /* Fix the length of the whole thing */
1562 postimage->len = new - postimage->buf;
1565 static int match_fragment(struct image *img,
1566 struct image *preimage,
1567 struct image *postimage,
1568 unsigned long try,
1569 int try_lno,
1570 unsigned ws_rule,
1571 int match_beginning, int match_end)
1573 int i;
1574 char *fixed_buf, *buf, *orig, *target;
1576 if (preimage->nr + try_lno > img->nr)
1577 return 0;
1579 if (match_beginning && try_lno)
1580 return 0;
1582 if (match_end && preimage->nr + try_lno != img->nr)
1583 return 0;
1585 /* Quick hash check */
1586 for (i = 0; i < preimage->nr; i++)
1587 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1588 return 0;
1591 * Do we have an exact match? If we were told to match
1592 * at the end, size must be exactly at try+fragsize,
1593 * otherwise try+fragsize must be still within the preimage,
1594 * and either case, the old piece should match the preimage
1595 * exactly.
1597 if ((match_end
1598 ? (try + preimage->len == img->len)
1599 : (try + preimage->len <= img->len)) &&
1600 !memcmp(img->buf + try, preimage->buf, preimage->len))
1601 return 1;
1603 if (ws_error_action != correct_ws_error)
1604 return 0;
1607 * The hunk does not apply byte-by-byte, but the hash says
1608 * it might with whitespace fuzz.
1610 fixed_buf = xmalloc(preimage->len + 1);
1611 buf = fixed_buf;
1612 orig = preimage->buf;
1613 target = img->buf + try;
1614 for (i = 0; i < preimage->nr; i++) {
1615 size_t fixlen; /* length after fixing the preimage */
1616 size_t oldlen = preimage->line[i].len;
1617 size_t tgtlen = img->line[try_lno + i].len;
1618 size_t tgtfixlen; /* length after fixing the target line */
1619 char tgtfixbuf[1024], *tgtfix;
1620 int match;
1622 /* Try fixing the line in the preimage */
1623 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1625 /* Try fixing the line in the target */
1626 if (sizeof(tgtfixbuf) < tgtlen)
1627 tgtfix = tgtfixbuf;
1628 else
1629 tgtfix = xmalloc(tgtlen);
1630 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1633 * If they match, either the preimage was based on
1634 * a version before our tree fixed whitespace breakage,
1635 * or we are lacking a whitespace-fix patch the tree
1636 * the preimage was based on already had (i.e. target
1637 * has whitespace breakage, the preimage doesn't).
1638 * In either case, we are fixing the whitespace breakages
1639 * so we might as well take the fix together with their
1640 * real change.
1642 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1644 if (tgtfix != tgtfixbuf)
1645 free(tgtfix);
1646 if (!match)
1647 goto unmatch_exit;
1649 orig += oldlen;
1650 buf += fixlen;
1651 target += tgtlen;
1655 * Yes, the preimage is based on an older version that still
1656 * has whitespace breakages unfixed, and fixing them makes the
1657 * hunk match. Update the context lines in the postimage.
1659 update_pre_post_images(preimage, postimage,
1660 fixed_buf, buf - fixed_buf);
1661 return 1;
1663 unmatch_exit:
1664 free(fixed_buf);
1665 return 0;
1668 static int find_pos(struct image *img,
1669 struct image *preimage,
1670 struct image *postimage,
1671 int line,
1672 unsigned ws_rule,
1673 int match_beginning, int match_end)
1675 int i;
1676 unsigned long backwards, forwards, try;
1677 int backwards_lno, forwards_lno, try_lno;
1679 if (preimage->nr > img->nr)
1680 return -1;
1683 * If match_begining or match_end is specified, there is no
1684 * point starting from a wrong line that will never match and
1685 * wander around and wait for a match at the specified end.
1687 if (match_beginning)
1688 line = 0;
1689 else if (match_end)
1690 line = img->nr - preimage->nr;
1692 if (line > img->nr)
1693 line = img->nr;
1695 try = 0;
1696 for (i = 0; i < line; i++)
1697 try += img->line[i].len;
1700 * There's probably some smart way to do this, but I'll leave
1701 * that to the smart and beautiful people. I'm simple and stupid.
1703 backwards = try;
1704 backwards_lno = line;
1705 forwards = try;
1706 forwards_lno = line;
1707 try_lno = line;
1709 for (i = 0; ; i++) {
1710 if (match_fragment(img, preimage, postimage,
1711 try, try_lno, ws_rule,
1712 match_beginning, match_end))
1713 return try_lno;
1715 again:
1716 if (backwards_lno == 0 && forwards_lno == img->nr)
1717 break;
1719 if (i & 1) {
1720 if (backwards_lno == 0) {
1721 i++;
1722 goto again;
1724 backwards_lno--;
1725 backwards -= img->line[backwards_lno].len;
1726 try = backwards;
1727 try_lno = backwards_lno;
1728 } else {
1729 if (forwards_lno == img->nr) {
1730 i++;
1731 goto again;
1733 forwards += img->line[forwards_lno].len;
1734 forwards_lno++;
1735 try = forwards;
1736 try_lno = forwards_lno;
1740 return -1;
1743 static void remove_first_line(struct image *img)
1745 img->buf += img->line[0].len;
1746 img->len -= img->line[0].len;
1747 img->line++;
1748 img->nr--;
1751 static void remove_last_line(struct image *img)
1753 img->len -= img->line[--img->nr].len;
1756 static void update_image(struct image *img,
1757 int applied_pos,
1758 struct image *preimage,
1759 struct image *postimage)
1762 * remove the copy of preimage at offset in img
1763 * and replace it with postimage
1765 int i, nr;
1766 size_t remove_count, insert_count, applied_at = 0;
1767 char *result;
1769 for (i = 0; i < applied_pos; i++)
1770 applied_at += img->line[i].len;
1772 remove_count = 0;
1773 for (i = 0; i < preimage->nr; i++)
1774 remove_count += img->line[applied_pos + i].len;
1775 insert_count = postimage->len;
1777 /* Adjust the contents */
1778 result = xmalloc(img->len + insert_count - remove_count + 1);
1779 memcpy(result, img->buf, applied_at);
1780 memcpy(result + applied_at, postimage->buf, postimage->len);
1781 memcpy(result + applied_at + postimage->len,
1782 img->buf + (applied_at + remove_count),
1783 img->len - (applied_at + remove_count));
1784 free(img->buf);
1785 img->buf = result;
1786 img->len += insert_count - remove_count;
1787 result[img->len] = '\0';
1789 /* Adjust the line table */
1790 nr = img->nr + postimage->nr - preimage->nr;
1791 if (preimage->nr < postimage->nr) {
1793 * NOTE: this knows that we never call remove_first_line()
1794 * on anything other than pre/post image.
1796 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1797 img->line_allocated = img->line;
1799 if (preimage->nr != postimage->nr)
1800 memmove(img->line + applied_pos + postimage->nr,
1801 img->line + applied_pos + preimage->nr,
1802 (img->nr - (applied_pos + preimage->nr)) *
1803 sizeof(*img->line));
1804 memcpy(img->line + applied_pos,
1805 postimage->line,
1806 postimage->nr * sizeof(*img->line));
1807 img->nr = nr;
1810 static int apply_one_fragment(struct image *img, struct fragment *frag,
1811 int inaccurate_eof, unsigned ws_rule)
1813 int match_beginning, match_end;
1814 const char *patch = frag->patch;
1815 int size = frag->size;
1816 char *old, *new, *oldlines, *newlines;
1817 int new_blank_lines_at_end = 0;
1818 unsigned long leading, trailing;
1819 int pos, applied_pos;
1820 struct image preimage;
1821 struct image postimage;
1823 memset(&preimage, 0, sizeof(preimage));
1824 memset(&postimage, 0, sizeof(postimage));
1825 oldlines = xmalloc(size);
1826 newlines = xmalloc(size);
1828 old = oldlines;
1829 new = newlines;
1830 while (size > 0) {
1831 char first;
1832 int len = linelen(patch, size);
1833 int plen, added;
1834 int added_blank_line = 0;
1836 if (!len)
1837 break;
1840 * "plen" is how much of the line we should use for
1841 * the actual patch data. Normally we just remove the
1842 * first character on the line, but if the line is
1843 * followed by "\ No newline", then we also remove the
1844 * last one (which is the newline, of course).
1846 plen = len - 1;
1847 if (len < size && patch[len] == '\\')
1848 plen--;
1849 first = *patch;
1850 if (apply_in_reverse) {
1851 if (first == '-')
1852 first = '+';
1853 else if (first == '+')
1854 first = '-';
1857 switch (first) {
1858 case '\n':
1859 /* Newer GNU diff, empty context line */
1860 if (plen < 0)
1861 /* ... followed by '\No newline'; nothing */
1862 break;
1863 *old++ = '\n';
1864 *new++ = '\n';
1865 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1866 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1867 break;
1868 case ' ':
1869 case '-':
1870 memcpy(old, patch + 1, plen);
1871 add_line_info(&preimage, old, plen,
1872 (first == ' ' ? LINE_COMMON : 0));
1873 old += plen;
1874 if (first == '-')
1875 break;
1876 /* Fall-through for ' ' */
1877 case '+':
1878 /* --no-add does not add new lines */
1879 if (first == '+' && no_add)
1880 break;
1882 if (first != '+' ||
1883 !whitespace_error ||
1884 ws_error_action != correct_ws_error) {
1885 memcpy(new, patch + 1, plen);
1886 added = plen;
1888 else {
1889 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1891 add_line_info(&postimage, new, added,
1892 (first == '+' ? 0 : LINE_COMMON));
1893 new += added;
1894 if (first == '+' &&
1895 added == 1 && new[-1] == '\n')
1896 added_blank_line = 1;
1897 break;
1898 case '@': case '\\':
1899 /* Ignore it, we already handled it */
1900 break;
1901 default:
1902 if (apply_verbosely)
1903 error("invalid start of line: '%c'", first);
1904 return -1;
1906 if (added_blank_line)
1907 new_blank_lines_at_end++;
1908 else
1909 new_blank_lines_at_end = 0;
1910 patch += len;
1911 size -= len;
1913 if (inaccurate_eof &&
1914 old > oldlines && old[-1] == '\n' &&
1915 new > newlines && new[-1] == '\n') {
1916 old--;
1917 new--;
1920 leading = frag->leading;
1921 trailing = frag->trailing;
1924 * A hunk to change lines at the beginning would begin with
1925 * @@ -1,L +N,M @@
1927 * And a hunk to add to an empty file would begin with
1928 * @@ -0,0 +N,M @@
1930 * In other words, a hunk that is (frag->oldpos <= 1) with or
1931 * without leading context must match at the beginning.
1933 match_beginning = frag->oldpos <= 1;
1936 * A hunk without trailing lines must match at the end.
1937 * However, we simply cannot tell if a hunk must match end
1938 * from the lack of trailing lines if the patch was generated
1939 * with unidiff without any context.
1941 match_end = !unidiff_zero && !trailing;
1943 pos = frag->newpos ? (frag->newpos - 1) : 0;
1944 preimage.buf = oldlines;
1945 preimage.len = old - oldlines;
1946 postimage.buf = newlines;
1947 postimage.len = new - newlines;
1948 preimage.line = preimage.line_allocated;
1949 postimage.line = postimage.line_allocated;
1951 for (;;) {
1953 applied_pos = find_pos(img, &preimage, &postimage, pos,
1954 ws_rule, match_beginning, match_end);
1956 if (applied_pos >= 0)
1957 break;
1959 /* Am I at my context limits? */
1960 if ((leading <= p_context) && (trailing <= p_context))
1961 break;
1962 if (match_beginning || match_end) {
1963 match_beginning = match_end = 0;
1964 continue;
1968 * Reduce the number of context lines; reduce both
1969 * leading and trailing if they are equal otherwise
1970 * just reduce the larger context.
1972 if (leading >= trailing) {
1973 remove_first_line(&preimage);
1974 remove_first_line(&postimage);
1975 pos--;
1976 leading--;
1978 if (trailing > leading) {
1979 remove_last_line(&preimage);
1980 remove_last_line(&postimage);
1981 trailing--;
1985 if (applied_pos >= 0) {
1986 if (ws_error_action == correct_ws_error &&
1987 new_blank_lines_at_end &&
1988 postimage.nr + applied_pos == img->nr) {
1990 * If the patch application adds blank lines
1991 * at the end, and if the patch applies at the
1992 * end of the image, remove those added blank
1993 * lines.
1995 while (new_blank_lines_at_end--)
1996 remove_last_line(&postimage);
2000 * Warn if it was necessary to reduce the number
2001 * of context lines.
2003 if ((leading != frag->leading) ||
2004 (trailing != frag->trailing))
2005 fprintf(stderr, "Context reduced to (%ld/%ld)"
2006 " to apply fragment at %d\n",
2007 leading, trailing, applied_pos+1);
2008 update_image(img, applied_pos, &preimage, &postimage);
2009 } else {
2010 if (apply_verbosely)
2011 error("while searching for:\n%.*s",
2012 (int)(old - oldlines), oldlines);
2015 free(oldlines);
2016 free(newlines);
2017 free(preimage.line_allocated);
2018 free(postimage.line_allocated);
2020 return (applied_pos < 0);
2023 static int apply_binary_fragment(struct image *img, struct patch *patch)
2025 struct fragment *fragment = patch->fragments;
2026 unsigned long len;
2027 void *dst;
2029 /* Binary patch is irreversible without the optional second hunk */
2030 if (apply_in_reverse) {
2031 if (!fragment->next)
2032 return error("cannot reverse-apply a binary patch "
2033 "without the reverse hunk to '%s'",
2034 patch->new_name
2035 ? patch->new_name : patch->old_name);
2036 fragment = fragment->next;
2038 switch (fragment->binary_patch_method) {
2039 case BINARY_DELTA_DEFLATED:
2040 dst = patch_delta(img->buf, img->len, fragment->patch,
2041 fragment->size, &len);
2042 if (!dst)
2043 return -1;
2044 clear_image(img);
2045 img->buf = dst;
2046 img->len = len;
2047 return 0;
2048 case BINARY_LITERAL_DEFLATED:
2049 clear_image(img);
2050 img->len = fragment->size;
2051 img->buf = xmalloc(img->len+1);
2052 memcpy(img->buf, fragment->patch, img->len);
2053 img->buf[img->len] = '\0';
2054 return 0;
2056 return -1;
2059 static int apply_binary(struct image *img, struct patch *patch)
2061 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2062 unsigned char sha1[20];
2065 * For safety, we require patch index line to contain
2066 * full 40-byte textual SHA1 for old and new, at least for now.
2068 if (strlen(patch->old_sha1_prefix) != 40 ||
2069 strlen(patch->new_sha1_prefix) != 40 ||
2070 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2071 get_sha1_hex(patch->new_sha1_prefix, sha1))
2072 return error("cannot apply binary patch to '%s' "
2073 "without full index line", name);
2075 if (patch->old_name) {
2077 * See if the old one matches what the patch
2078 * applies to.
2080 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2081 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2082 return error("the patch applies to '%s' (%s), "
2083 "which does not match the "
2084 "current contents.",
2085 name, sha1_to_hex(sha1));
2087 else {
2088 /* Otherwise, the old one must be empty. */
2089 if (img->len)
2090 return error("the patch applies to an empty "
2091 "'%s' but it is not empty", name);
2094 get_sha1_hex(patch->new_sha1_prefix, sha1);
2095 if (is_null_sha1(sha1)) {
2096 clear_image(img);
2097 return 0; /* deletion patch */
2100 if (has_sha1_file(sha1)) {
2101 /* We already have the postimage */
2102 enum object_type type;
2103 unsigned long size;
2104 char *result;
2106 result = read_sha1_file(sha1, &type, &size);
2107 if (!result)
2108 return error("the necessary postimage %s for "
2109 "'%s' cannot be read",
2110 patch->new_sha1_prefix, name);
2111 clear_image(img);
2112 img->buf = result;
2113 img->len = size;
2114 } else {
2116 * We have verified buf matches the preimage;
2117 * apply the patch data to it, which is stored
2118 * in the patch->fragments->{patch,size}.
2120 if (apply_binary_fragment(img, patch))
2121 return error("binary patch does not apply to '%s'",
2122 name);
2124 /* verify that the result matches */
2125 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2126 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2127 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2128 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2131 return 0;
2134 static int apply_fragments(struct image *img, struct patch *patch)
2136 struct fragment *frag = patch->fragments;
2137 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2138 unsigned ws_rule = patch->ws_rule;
2139 unsigned inaccurate_eof = patch->inaccurate_eof;
2141 if (patch->is_binary)
2142 return apply_binary(img, patch);
2144 while (frag) {
2145 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2146 error("patch failed: %s:%ld", name, frag->oldpos);
2147 if (!apply_with_reject)
2148 return -1;
2149 frag->rejected = 1;
2151 frag = frag->next;
2153 return 0;
2156 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2158 if (!ce)
2159 return 0;
2161 if (S_ISGITLINK(ce->ce_mode)) {
2162 strbuf_grow(buf, 100);
2163 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2164 } else {
2165 enum object_type type;
2166 unsigned long sz;
2167 char *result;
2169 result = read_sha1_file(ce->sha1, &type, &sz);
2170 if (!result)
2171 return -1;
2172 /* XXX read_sha1_file NUL-terminates */
2173 strbuf_attach(buf, result, sz, sz + 1);
2175 return 0;
2178 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2180 struct strbuf buf;
2181 struct image image;
2182 size_t len;
2183 char *img;
2185 strbuf_init(&buf, 0);
2186 if (cached) {
2187 if (read_file_or_gitlink(ce, &buf))
2188 return error("read of %s failed", patch->old_name);
2189 } else if (patch->old_name) {
2190 if (S_ISGITLINK(patch->old_mode)) {
2191 if (ce) {
2192 read_file_or_gitlink(ce, &buf);
2193 } else {
2195 * There is no way to apply subproject
2196 * patch without looking at the index.
2198 patch->fragments = NULL;
2200 } else {
2201 if (read_old_data(st, patch->old_name, &buf))
2202 return error("read of %s failed", patch->old_name);
2206 img = strbuf_detach(&buf, &len);
2207 prepare_image(&image, img, len, !patch->is_binary);
2209 if (apply_fragments(&image, patch) < 0)
2210 return -1; /* note with --reject this succeeds. */
2211 patch->result = image.buf;
2212 patch->resultsize = image.len;
2213 free(image.line_allocated);
2215 if (0 < patch->is_delete && patch->resultsize)
2216 return error("removal patch leaves file contents");
2218 return 0;
2221 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2223 struct stat nst;
2224 if (!lstat(new_name, &nst)) {
2225 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2226 return 0;
2228 * A leading component of new_name might be a symlink
2229 * that is going to be removed with this patch, but
2230 * still pointing at somewhere that has the path.
2231 * In such a case, path "new_name" does not exist as
2232 * far as git is concerned.
2234 if (has_symlink_leading_path(strlen(new_name), new_name))
2235 return 0;
2237 return error("%s: already exists in working directory", new_name);
2239 else if ((errno != ENOENT) && (errno != ENOTDIR))
2240 return error("%s: %s", new_name, strerror(errno));
2241 return 0;
2244 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2246 if (S_ISGITLINK(ce->ce_mode)) {
2247 if (!S_ISDIR(st->st_mode))
2248 return -1;
2249 return 0;
2251 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2254 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2256 const char *old_name = patch->old_name;
2257 int stat_ret = 0;
2258 unsigned st_mode = 0;
2261 * Make sure that we do not have local modifications from the
2262 * index when we are looking at the index. Also make sure
2263 * we have the preimage file to be patched in the work tree,
2264 * unless --cached, which tells git to apply only in the index.
2266 if (!old_name)
2267 return 0;
2269 assert(patch->is_new <= 0);
2270 if (!cached) {
2271 stat_ret = lstat(old_name, st);
2272 if (stat_ret && errno != ENOENT)
2273 return error("%s: %s", old_name, strerror(errno));
2275 if (check_index) {
2276 int pos = cache_name_pos(old_name, strlen(old_name));
2277 if (pos < 0) {
2278 if (patch->is_new < 0)
2279 goto is_new;
2280 return error("%s: does not exist in index", old_name);
2282 *ce = active_cache[pos];
2283 if (stat_ret < 0) {
2284 struct checkout costate;
2285 /* checkout */
2286 costate.base_dir = "";
2287 costate.base_dir_len = 0;
2288 costate.force = 0;
2289 costate.quiet = 0;
2290 costate.not_new = 0;
2291 costate.refresh_cache = 1;
2292 if (checkout_entry(*ce, &costate, NULL) ||
2293 lstat(old_name, st))
2294 return -1;
2296 if (!cached && verify_index_match(*ce, st))
2297 return error("%s: does not match index", old_name);
2298 if (cached)
2299 st_mode = (*ce)->ce_mode;
2300 } else if (stat_ret < 0) {
2301 if (patch->is_new < 0)
2302 goto is_new;
2303 return error("%s: %s", old_name, strerror(errno));
2306 if (!cached)
2307 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2309 if (patch->is_new < 0)
2310 patch->is_new = 0;
2311 if (!patch->old_mode)
2312 patch->old_mode = st_mode;
2313 if ((st_mode ^ patch->old_mode) & S_IFMT)
2314 return error("%s: wrong type", old_name);
2315 if (st_mode != patch->old_mode)
2316 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2317 old_name, st_mode, patch->old_mode);
2318 return 0;
2320 is_new:
2321 patch->is_new = 1;
2322 patch->is_delete = 0;
2323 patch->old_name = NULL;
2324 return 0;
2327 static int check_patch(struct patch *patch, struct patch *prev_patch)
2329 struct stat st;
2330 const char *old_name = patch->old_name;
2331 const char *new_name = patch->new_name;
2332 const char *name = old_name ? old_name : new_name;
2333 struct cache_entry *ce = NULL;
2334 int ok_if_exists;
2335 int status;
2337 patch->rejected = 1; /* we will drop this after we succeed */
2339 status = check_preimage(patch, &ce, &st);
2340 if (status)
2341 return status;
2342 old_name = patch->old_name;
2344 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2345 !strcmp(prev_patch->old_name, new_name))
2347 * A type-change diff is always split into a patch to
2348 * delete old, immediately followed by a patch to
2349 * create new (see diff.c::run_diff()); in such a case
2350 * it is Ok that the entry to be deleted by the
2351 * previous patch is still in the working tree and in
2352 * the index.
2354 ok_if_exists = 1;
2355 else
2356 ok_if_exists = 0;
2358 if (new_name &&
2359 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2360 if (check_index &&
2361 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2362 !ok_if_exists)
2363 return error("%s: already exists in index", new_name);
2364 if (!cached) {
2365 int err = check_to_create_blob(new_name, ok_if_exists);
2366 if (err)
2367 return err;
2369 if (!patch->new_mode) {
2370 if (0 < patch->is_new)
2371 patch->new_mode = S_IFREG | 0644;
2372 else
2373 patch->new_mode = patch->old_mode;
2377 if (new_name && old_name) {
2378 int same = !strcmp(old_name, new_name);
2379 if (!patch->new_mode)
2380 patch->new_mode = patch->old_mode;
2381 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2382 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2383 patch->new_mode, new_name, patch->old_mode,
2384 same ? "" : " of ", same ? "" : old_name);
2387 if (apply_data(patch, &st, ce) < 0)
2388 return error("%s: patch does not apply", name);
2389 patch->rejected = 0;
2390 return 0;
2393 static int check_patch_list(struct patch *patch)
2395 struct patch *prev_patch = NULL;
2396 int err = 0;
2398 for (prev_patch = NULL; patch ; patch = patch->next) {
2399 if (apply_verbosely)
2400 say_patch_name(stderr,
2401 "Checking patch ", patch, "...\n");
2402 err |= check_patch(patch, prev_patch);
2403 prev_patch = patch;
2405 return err;
2408 /* This function tries to read the sha1 from the current index */
2409 static int get_current_sha1(const char *path, unsigned char *sha1)
2411 int pos;
2413 if (read_cache() < 0)
2414 return -1;
2415 pos = cache_name_pos(path, strlen(path));
2416 if (pos < 0)
2417 return -1;
2418 hashcpy(sha1, active_cache[pos]->sha1);
2419 return 0;
2422 /* Build an index that contains the just the files needed for a 3way merge */
2423 static void build_fake_ancestor(struct patch *list, const char *filename)
2425 struct patch *patch;
2426 struct index_state result = { 0 };
2427 int fd;
2429 /* Once we start supporting the reverse patch, it may be
2430 * worth showing the new sha1 prefix, but until then...
2432 for (patch = list; patch; patch = patch->next) {
2433 const unsigned char *sha1_ptr;
2434 unsigned char sha1[20];
2435 struct cache_entry *ce;
2436 const char *name;
2438 name = patch->old_name ? patch->old_name : patch->new_name;
2439 if (0 < patch->is_new)
2440 continue;
2441 else if (get_sha1(patch->old_sha1_prefix, sha1))
2442 /* git diff has no index line for mode/type changes */
2443 if (!patch->lines_added && !patch->lines_deleted) {
2444 if (get_current_sha1(patch->new_name, sha1) ||
2445 get_current_sha1(patch->old_name, sha1))
2446 die("mode change for %s, which is not "
2447 "in current HEAD", name);
2448 sha1_ptr = sha1;
2449 } else
2450 die("sha1 information is lacking or useless "
2451 "(%s).", name);
2452 else
2453 sha1_ptr = sha1;
2455 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2456 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2457 die ("Could not add %s to temporary index", name);
2460 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2461 if (fd < 0 || write_index(&result, fd) || close(fd))
2462 die ("Could not write temporary index to %s", filename);
2464 discard_index(&result);
2467 static void stat_patch_list(struct patch *patch)
2469 int files, adds, dels;
2471 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2472 files++;
2473 adds += patch->lines_added;
2474 dels += patch->lines_deleted;
2475 show_stats(patch);
2478 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2481 static void numstat_patch_list(struct patch *patch)
2483 for ( ; patch; patch = patch->next) {
2484 const char *name;
2485 name = patch->new_name ? patch->new_name : patch->old_name;
2486 if (patch->is_binary)
2487 printf("-\t-\t");
2488 else
2489 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2490 write_name_quoted(name, stdout, line_termination);
2494 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2496 if (mode)
2497 printf(" %s mode %06o %s\n", newdelete, mode, name);
2498 else
2499 printf(" %s %s\n", newdelete, name);
2502 static void show_mode_change(struct patch *p, int show_name)
2504 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2505 if (show_name)
2506 printf(" mode change %06o => %06o %s\n",
2507 p->old_mode, p->new_mode, p->new_name);
2508 else
2509 printf(" mode change %06o => %06o\n",
2510 p->old_mode, p->new_mode);
2514 static void show_rename_copy(struct patch *p)
2516 const char *renamecopy = p->is_rename ? "rename" : "copy";
2517 const char *old, *new;
2519 /* Find common prefix */
2520 old = p->old_name;
2521 new = p->new_name;
2522 while (1) {
2523 const char *slash_old, *slash_new;
2524 slash_old = strchr(old, '/');
2525 slash_new = strchr(new, '/');
2526 if (!slash_old ||
2527 !slash_new ||
2528 slash_old - old != slash_new - new ||
2529 memcmp(old, new, slash_new - new))
2530 break;
2531 old = slash_old + 1;
2532 new = slash_new + 1;
2534 /* p->old_name thru old is the common prefix, and old and new
2535 * through the end of names are renames
2537 if (old != p->old_name)
2538 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2539 (int)(old - p->old_name), p->old_name,
2540 old, new, p->score);
2541 else
2542 printf(" %s %s => %s (%d%%)\n", renamecopy,
2543 p->old_name, p->new_name, p->score);
2544 show_mode_change(p, 0);
2547 static void summary_patch_list(struct patch *patch)
2549 struct patch *p;
2551 for (p = patch; p; p = p->next) {
2552 if (p->is_new)
2553 show_file_mode_name("create", p->new_mode, p->new_name);
2554 else if (p->is_delete)
2555 show_file_mode_name("delete", p->old_mode, p->old_name);
2556 else {
2557 if (p->is_rename || p->is_copy)
2558 show_rename_copy(p);
2559 else {
2560 if (p->score) {
2561 printf(" rewrite %s (%d%%)\n",
2562 p->new_name, p->score);
2563 show_mode_change(p, 0);
2565 else
2566 show_mode_change(p, 1);
2572 static void patch_stats(struct patch *patch)
2574 int lines = patch->lines_added + patch->lines_deleted;
2576 if (lines > max_change)
2577 max_change = lines;
2578 if (patch->old_name) {
2579 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2580 if (!len)
2581 len = strlen(patch->old_name);
2582 if (len > max_len)
2583 max_len = len;
2585 if (patch->new_name) {
2586 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2587 if (!len)
2588 len = strlen(patch->new_name);
2589 if (len > max_len)
2590 max_len = len;
2594 static void remove_file(struct patch *patch, int rmdir_empty)
2596 if (update_index) {
2597 if (remove_file_from_cache(patch->old_name) < 0)
2598 die("unable to remove %s from index", patch->old_name);
2600 if (!cached) {
2601 if (S_ISGITLINK(patch->old_mode)) {
2602 if (rmdir(patch->old_name))
2603 warning("unable to remove submodule %s",
2604 patch->old_name);
2605 } else if (!unlink(patch->old_name) && rmdir_empty) {
2606 char *name = xstrdup(patch->old_name);
2607 char *end = strrchr(name, '/');
2608 while (end) {
2609 *end = 0;
2610 if (rmdir(name))
2611 break;
2612 end = strrchr(name, '/');
2614 free(name);
2619 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2621 struct stat st;
2622 struct cache_entry *ce;
2623 int namelen = strlen(path);
2624 unsigned ce_size = cache_entry_size(namelen);
2626 if (!update_index)
2627 return;
2629 ce = xcalloc(1, ce_size);
2630 memcpy(ce->name, path, namelen);
2631 ce->ce_mode = create_ce_mode(mode);
2632 ce->ce_flags = namelen;
2633 if (S_ISGITLINK(mode)) {
2634 const char *s = buf;
2636 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2637 die("corrupt patch for subproject %s", path);
2638 } else {
2639 if (!cached) {
2640 if (lstat(path, &st) < 0)
2641 die("unable to stat newly created file %s",
2642 path);
2643 fill_stat_cache_info(ce, &st);
2645 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2646 die("unable to create backing store for newly created file %s", path);
2648 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2649 die("unable to add cache entry for %s", path);
2652 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2654 int fd;
2655 struct strbuf nbuf;
2657 if (S_ISGITLINK(mode)) {
2658 struct stat st;
2659 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2660 return 0;
2661 return mkdir(path, 0777);
2664 if (has_symlinks && S_ISLNK(mode))
2665 /* Although buf:size is counted string, it also is NUL
2666 * terminated.
2668 return symlink(buf, path);
2670 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2671 if (fd < 0)
2672 return -1;
2674 strbuf_init(&nbuf, 0);
2675 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2676 size = nbuf.len;
2677 buf = nbuf.buf;
2679 write_or_die(fd, buf, size);
2680 strbuf_release(&nbuf);
2682 if (close(fd) < 0)
2683 die("closing file %s: %s", path, strerror(errno));
2684 return 0;
2688 * We optimistically assume that the directories exist,
2689 * which is true 99% of the time anyway. If they don't,
2690 * we create them and try again.
2692 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2694 if (cached)
2695 return;
2696 if (!try_create_file(path, mode, buf, size))
2697 return;
2699 if (errno == ENOENT) {
2700 if (safe_create_leading_directories(path))
2701 return;
2702 if (!try_create_file(path, mode, buf, size))
2703 return;
2706 if (errno == EEXIST || errno == EACCES) {
2707 /* We may be trying to create a file where a directory
2708 * used to be.
2710 struct stat st;
2711 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2712 errno = EEXIST;
2715 if (errno == EEXIST) {
2716 unsigned int nr = getpid();
2718 for (;;) {
2719 const char *newpath;
2720 newpath = mkpath("%s~%u", path, nr);
2721 if (!try_create_file(newpath, mode, buf, size)) {
2722 if (!rename(newpath, path))
2723 return;
2724 unlink(newpath);
2725 break;
2727 if (errno != EEXIST)
2728 break;
2729 ++nr;
2732 die("unable to write file %s mode %o", path, mode);
2735 static void create_file(struct patch *patch)
2737 char *path = patch->new_name;
2738 unsigned mode = patch->new_mode;
2739 unsigned long size = patch->resultsize;
2740 char *buf = patch->result;
2742 if (!mode)
2743 mode = S_IFREG | 0644;
2744 create_one_file(path, mode, buf, size);
2745 add_index_file(path, mode, buf, size);
2748 /* phase zero is to remove, phase one is to create */
2749 static void write_out_one_result(struct patch *patch, int phase)
2751 if (patch->is_delete > 0) {
2752 if (phase == 0)
2753 remove_file(patch, 1);
2754 return;
2756 if (patch->is_new > 0 || patch->is_copy) {
2757 if (phase == 1)
2758 create_file(patch);
2759 return;
2762 * Rename or modification boils down to the same
2763 * thing: remove the old, write the new
2765 if (phase == 0)
2766 remove_file(patch, patch->is_rename);
2767 if (phase == 1)
2768 create_file(patch);
2771 static int write_out_one_reject(struct patch *patch)
2773 FILE *rej;
2774 char namebuf[PATH_MAX];
2775 struct fragment *frag;
2776 int cnt = 0;
2778 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2779 if (!frag->rejected)
2780 continue;
2781 cnt++;
2784 if (!cnt) {
2785 if (apply_verbosely)
2786 say_patch_name(stderr,
2787 "Applied patch ", patch, " cleanly.\n");
2788 return 0;
2791 /* This should not happen, because a removal patch that leaves
2792 * contents are marked "rejected" at the patch level.
2794 if (!patch->new_name)
2795 die("internal error");
2797 /* Say this even without --verbose */
2798 say_patch_name(stderr, "Applying patch ", patch, " with");
2799 fprintf(stderr, " %d rejects...\n", cnt);
2801 cnt = strlen(patch->new_name);
2802 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2803 cnt = ARRAY_SIZE(namebuf) - 5;
2804 fprintf(stderr,
2805 "warning: truncating .rej filename to %.*s.rej",
2806 cnt - 1, patch->new_name);
2808 memcpy(namebuf, patch->new_name, cnt);
2809 memcpy(namebuf + cnt, ".rej", 5);
2811 rej = fopen(namebuf, "w");
2812 if (!rej)
2813 return error("cannot open %s: %s", namebuf, strerror(errno));
2815 /* Normal git tools never deal with .rej, so do not pretend
2816 * this is a git patch by saying --git nor give extended
2817 * headers. While at it, maybe please "kompare" that wants
2818 * the trailing TAB and some garbage at the end of line ;-).
2820 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2821 patch->new_name, patch->new_name);
2822 for (cnt = 1, frag = patch->fragments;
2823 frag;
2824 cnt++, frag = frag->next) {
2825 if (!frag->rejected) {
2826 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2827 continue;
2829 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2830 fprintf(rej, "%.*s", frag->size, frag->patch);
2831 if (frag->patch[frag->size-1] != '\n')
2832 fputc('\n', rej);
2834 fclose(rej);
2835 return -1;
2838 static int write_out_results(struct patch *list, int skipped_patch)
2840 int phase;
2841 int errs = 0;
2842 struct patch *l;
2844 if (!list && !skipped_patch)
2845 return error("No changes");
2847 for (phase = 0; phase < 2; phase++) {
2848 l = list;
2849 while (l) {
2850 if (l->rejected)
2851 errs = 1;
2852 else {
2853 write_out_one_result(l, phase);
2854 if (phase == 1 && write_out_one_reject(l))
2855 errs = 1;
2857 l = l->next;
2860 return errs;
2863 static struct lock_file lock_file;
2865 static struct excludes {
2866 struct excludes *next;
2867 const char *path;
2868 } *excludes;
2870 static int use_patch(struct patch *p)
2872 const char *pathname = p->new_name ? p->new_name : p->old_name;
2873 struct excludes *x = excludes;
2874 while (x) {
2875 if (fnmatch(x->path, pathname, 0) == 0)
2876 return 0;
2877 x = x->next;
2879 if (0 < prefix_length) {
2880 int pathlen = strlen(pathname);
2881 if (pathlen <= prefix_length ||
2882 memcmp(prefix, pathname, prefix_length))
2883 return 0;
2885 return 1;
2888 static void prefix_one(char **name)
2890 char *old_name = *name;
2891 if (!old_name)
2892 return;
2893 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2894 free(old_name);
2897 static void prefix_patches(struct patch *p)
2899 if (!prefix || p->is_toplevel_relative)
2900 return;
2901 for ( ; p; p = p->next) {
2902 if (p->new_name == p->old_name) {
2903 char *prefixed = p->new_name;
2904 prefix_one(&prefixed);
2905 p->new_name = p->old_name = prefixed;
2907 else {
2908 prefix_one(&p->new_name);
2909 prefix_one(&p->old_name);
2914 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2916 size_t offset;
2917 struct strbuf buf;
2918 struct patch *list = NULL, **listp = &list;
2919 int skipped_patch = 0;
2921 strbuf_init(&buf, 0);
2922 patch_input_file = filename;
2923 read_patch_file(&buf, fd);
2924 offset = 0;
2925 while (offset < buf.len) {
2926 struct patch *patch;
2927 int nr;
2929 patch = xcalloc(1, sizeof(*patch));
2930 patch->inaccurate_eof = inaccurate_eof;
2931 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2932 if (nr < 0)
2933 break;
2934 if (apply_in_reverse)
2935 reverse_patches(patch);
2936 if (prefix)
2937 prefix_patches(patch);
2938 if (use_patch(patch)) {
2939 patch_stats(patch);
2940 *listp = patch;
2941 listp = &patch->next;
2943 else {
2944 /* perhaps free it a bit better? */
2945 free(patch);
2946 skipped_patch++;
2948 offset += nr;
2951 if (whitespace_error && (ws_error_action == die_on_ws_error))
2952 apply = 0;
2954 update_index = check_index && apply;
2955 if (update_index && newfd < 0)
2956 newfd = hold_locked_index(&lock_file, 1);
2958 if (check_index) {
2959 if (read_cache() < 0)
2960 die("unable to read index file");
2963 if ((check || apply) &&
2964 check_patch_list(list) < 0 &&
2965 !apply_with_reject)
2966 exit(1);
2968 if (apply && write_out_results(list, skipped_patch))
2969 exit(1);
2971 if (fake_ancestor)
2972 build_fake_ancestor(list, fake_ancestor);
2974 if (diffstat)
2975 stat_patch_list(list);
2977 if (numstat)
2978 numstat_patch_list(list);
2980 if (summary)
2981 summary_patch_list(list);
2983 strbuf_release(&buf);
2984 return 0;
2987 static int git_apply_config(const char *var, const char *value, void *cb)
2989 if (!strcmp(var, "apply.whitespace"))
2990 return git_config_string(&apply_default_whitespace, var, value);
2991 return git_default_config(var, value, cb);
2995 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2997 int i;
2998 int read_stdin = 1;
2999 int inaccurate_eof = 0;
3000 int errs = 0;
3001 int is_not_gitdir;
3003 const char *whitespace_option = NULL;
3005 prefix = setup_git_directory_gently(&is_not_gitdir);
3006 prefix_length = prefix ? strlen(prefix) : 0;
3007 git_config(git_apply_config, NULL);
3008 if (apply_default_whitespace)
3009 parse_whitespace_option(apply_default_whitespace);
3011 for (i = 1; i < argc; i++) {
3012 const char *arg = argv[i];
3013 char *end;
3014 int fd;
3016 if (!strcmp(arg, "-")) {
3017 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3018 read_stdin = 0;
3019 continue;
3021 if (!prefixcmp(arg, "--exclude=")) {
3022 struct excludes *x = xmalloc(sizeof(*x));
3023 x->path = arg + 10;
3024 x->next = excludes;
3025 excludes = x;
3026 continue;
3028 if (!prefixcmp(arg, "-p")) {
3029 p_value = atoi(arg + 2);
3030 p_value_known = 1;
3031 continue;
3033 if (!strcmp(arg, "--no-add")) {
3034 no_add = 1;
3035 continue;
3037 if (!strcmp(arg, "--stat")) {
3038 apply = 0;
3039 diffstat = 1;
3040 continue;
3042 if (!strcmp(arg, "--allow-binary-replacement") ||
3043 !strcmp(arg, "--binary")) {
3044 continue; /* now no-op */
3046 if (!strcmp(arg, "--numstat")) {
3047 apply = 0;
3048 numstat = 1;
3049 continue;
3051 if (!strcmp(arg, "--summary")) {
3052 apply = 0;
3053 summary = 1;
3054 continue;
3056 if (!strcmp(arg, "--check")) {
3057 apply = 0;
3058 check = 1;
3059 continue;
3061 if (!strcmp(arg, "--index")) {
3062 if (is_not_gitdir)
3063 die("--index outside a repository");
3064 check_index = 1;
3065 continue;
3067 if (!strcmp(arg, "--cached")) {
3068 if (is_not_gitdir)
3069 die("--cached outside a repository");
3070 check_index = 1;
3071 cached = 1;
3072 continue;
3074 if (!strcmp(arg, "--apply")) {
3075 apply = 1;
3076 continue;
3078 if (!strcmp(arg, "--build-fake-ancestor")) {
3079 apply = 0;
3080 if (++i >= argc)
3081 die ("need a filename");
3082 fake_ancestor = argv[i];
3083 continue;
3085 if (!strcmp(arg, "-z")) {
3086 line_termination = 0;
3087 continue;
3089 if (!prefixcmp(arg, "-C")) {
3090 p_context = strtoul(arg + 2, &end, 0);
3091 if (*end != '\0')
3092 die("unrecognized context count '%s'", arg + 2);
3093 continue;
3095 if (!prefixcmp(arg, "--whitespace=")) {
3096 whitespace_option = arg + 13;
3097 parse_whitespace_option(arg + 13);
3098 continue;
3100 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3101 apply_in_reverse = 1;
3102 continue;
3104 if (!strcmp(arg, "--unidiff-zero")) {
3105 unidiff_zero = 1;
3106 continue;
3108 if (!strcmp(arg, "--reject")) {
3109 apply = apply_with_reject = apply_verbosely = 1;
3110 continue;
3112 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3113 apply_verbosely = 1;
3114 continue;
3116 if (!strcmp(arg, "--inaccurate-eof")) {
3117 inaccurate_eof = 1;
3118 continue;
3120 if (0 < prefix_length)
3121 arg = prefix_filename(prefix, prefix_length, arg);
3123 fd = open(arg, O_RDONLY);
3124 if (fd < 0)
3125 die("can't open patch '%s': %s", arg, strerror(errno));
3126 read_stdin = 0;
3127 set_default_whitespace_mode(whitespace_option);
3128 errs |= apply_patch(fd, arg, inaccurate_eof);
3129 close(fd);
3131 set_default_whitespace_mode(whitespace_option);
3132 if (read_stdin)
3133 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3134 if (whitespace_error) {
3135 if (squelch_whitespace_errors &&
3136 squelch_whitespace_errors < whitespace_error) {
3137 int squelched =
3138 whitespace_error - squelch_whitespace_errors;
3139 fprintf(stderr, "warning: squelched %d "
3140 "whitespace error%s\n",
3141 squelched,
3142 squelched == 1 ? "" : "s");
3144 if (ws_error_action == die_on_ws_error)
3145 die("%d line%s add%s whitespace errors.",
3146 whitespace_error,
3147 whitespace_error == 1 ? "" : "s",
3148 whitespace_error == 1 ? "s" : "");
3149 if (applied_after_fixing_ws && apply)
3150 fprintf(stderr, "warning: %d line%s applied after"
3151 " fixing whitespace errors.\n",
3152 applied_after_fixing_ws,
3153 applied_after_fixing_ws == 1 ? "" : "s");
3154 else if (whitespace_error)
3155 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3156 whitespace_error,
3157 whitespace_error == 1 ? "" : "s",
3158 whitespace_error == 1 ? "s" : "");
3161 if (update_index) {
3162 if (write_cache(newfd, active_cache, active_nr) ||
3163 commit_locked_index(&lock_file))
3164 die("Unable to write new index file");
3167 return !!errs;