Make repack less likely to corrupt repository
[alt-git.git] / builtin-apply.c
blob7a1ff041f15cd4e6ce2a5b1d1e3cb668f2f54f5e
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
19 * --check turns on checking that the working tree matches the
20 * files that are being modified, but doesn't apply the patch
21 * --stat does just a diffstat, and doesn't actually apply
22 * --numstat does numeric diffstat, and doesn't actually apply
23 * --index-info shows the old and new index info for paths if available.
24 * --index updates the cache as well.
25 * --cached updates only the cache without ever touching the working tree.
27 static const char *prefix;
28 static int prefix_length = -1;
29 static int newfd = -1;
31 static int unidiff_zero;
32 static int p_value = 1;
33 static int p_value_known;
34 static int check_index;
35 static int update_index;
36 static int cached;
37 static int diffstat;
38 static int numstat;
39 static int summary;
40 static int check;
41 static int apply = 1;
42 static int apply_in_reverse;
43 static int apply_with_reject;
44 static int apply_verbosely;
45 static int no_add;
46 static const char *fake_ancestor;
47 static int line_termination = '\n';
48 static unsigned long p_context = ULONG_MAX;
49 static const char apply_usage[] =
50 "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>...";
52 static enum ws_error_action {
53 nowarn_ws_error,
54 warn_on_ws_error,
55 die_on_ws_error,
56 correct_ws_error,
57 } ws_error_action = warn_on_ws_error;
58 static int whitespace_error;
59 static int squelch_whitespace_errors = 5;
60 static int applied_after_fixing_ws;
61 static const char *patch_input_file;
62 static const char *root;
63 static int root_len;
65 static void parse_whitespace_option(const char *option)
67 if (!option) {
68 ws_error_action = warn_on_ws_error;
69 return;
71 if (!strcmp(option, "warn")) {
72 ws_error_action = warn_on_ws_error;
73 return;
75 if (!strcmp(option, "nowarn")) {
76 ws_error_action = nowarn_ws_error;
77 return;
79 if (!strcmp(option, "error")) {
80 ws_error_action = die_on_ws_error;
81 return;
83 if (!strcmp(option, "error-all")) {
84 ws_error_action = die_on_ws_error;
85 squelch_whitespace_errors = 0;
86 return;
88 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
89 ws_error_action = correct_ws_error;
90 return;
92 die("unrecognized whitespace option '%s'", option);
95 static void set_default_whitespace_mode(const char *whitespace_option)
97 if (!whitespace_option && !apply_default_whitespace)
98 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
102 * For "diff-stat" like behaviour, we keep track of the biggest change
103 * we've seen, and the longest filename. That allows us to do simple
104 * scaling.
106 static int max_change, max_len;
109 * Various "current state", notably line numbers and what
110 * file (and how) we're patching right now.. The "is_xxxx"
111 * things are flags, where -1 means "don't know yet".
113 static int linenr = 1;
116 * This represents one "hunk" from a patch, starting with
117 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
118 * patch text is pointed at by patch, and its byte length
119 * is stored in size. leading and trailing are the number
120 * of context lines.
122 struct fragment {
123 unsigned long leading, trailing;
124 unsigned long oldpos, oldlines;
125 unsigned long newpos, newlines;
126 const char *patch;
127 int size;
128 int rejected;
129 struct fragment *next;
133 * When dealing with a binary patch, we reuse "leading" field
134 * to store the type of the binary hunk, either deflated "delta"
135 * or deflated "literal".
137 #define binary_patch_method leading
138 #define BINARY_DELTA_DEFLATED 1
139 #define BINARY_LITERAL_DEFLATED 2
142 * This represents a "patch" to a file, both metainfo changes
143 * such as creation/deletion, filemode and content changes represented
144 * as a series of fragments.
146 struct patch {
147 char *new_name, *old_name, *def_name;
148 unsigned int old_mode, new_mode;
149 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
150 int rejected;
151 unsigned ws_rule;
152 unsigned long deflate_origlen;
153 int lines_added, lines_deleted;
154 int score;
155 unsigned int is_toplevel_relative:1;
156 unsigned int inaccurate_eof:1;
157 unsigned int is_binary:1;
158 unsigned int is_copy:1;
159 unsigned int is_rename:1;
160 unsigned int recount:1;
161 struct fragment *fragments;
162 char *result;
163 size_t resultsize;
164 char old_sha1_prefix[41];
165 char new_sha1_prefix[41];
166 struct patch *next;
170 * A line in a file, len-bytes long (includes the terminating LF,
171 * except for an incomplete line at the end if the file ends with
172 * one), and its contents hashes to 'hash'.
174 struct line {
175 size_t len;
176 unsigned hash : 24;
177 unsigned flag : 8;
178 #define LINE_COMMON 1
182 * This represents a "file", which is an array of "lines".
184 struct image {
185 char *buf;
186 size_t len;
187 size_t nr;
188 size_t alloc;
189 struct line *line_allocated;
190 struct line *line;
194 * Records filenames that have been touched, in order to handle
195 * the case where more than one patches touch the same file.
198 static struct string_list fn_table;
200 static uint32_t hash_line(const char *cp, size_t len)
202 size_t i;
203 uint32_t h;
204 for (i = 0, h = 0; i < len; i++) {
205 if (!isspace(cp[i])) {
206 h = h * 3 + (cp[i] & 0xff);
209 return h;
212 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
214 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
215 img->line_allocated[img->nr].len = len;
216 img->line_allocated[img->nr].hash = hash_line(bol, len);
217 img->line_allocated[img->nr].flag = flag;
218 img->nr++;
221 static void prepare_image(struct image *image, char *buf, size_t len,
222 int prepare_linetable)
224 const char *cp, *ep;
226 memset(image, 0, sizeof(*image));
227 image->buf = buf;
228 image->len = len;
230 if (!prepare_linetable)
231 return;
233 ep = image->buf + image->len;
234 cp = image->buf;
235 while (cp < ep) {
236 const char *next;
237 for (next = cp; next < ep && *next != '\n'; next++)
239 if (next < ep)
240 next++;
241 add_line_info(image, cp, next - cp, 0);
242 cp = next;
244 image->line = image->line_allocated;
247 static void clear_image(struct image *image)
249 free(image->buf);
250 image->buf = NULL;
251 image->len = 0;
254 static void say_patch_name(FILE *output, const char *pre,
255 struct patch *patch, const char *post)
257 fputs(pre, output);
258 if (patch->old_name && patch->new_name &&
259 strcmp(patch->old_name, patch->new_name)) {
260 quote_c_style(patch->old_name, NULL, output, 0);
261 fputs(" => ", output);
262 quote_c_style(patch->new_name, NULL, output, 0);
263 } else {
264 const char *n = patch->new_name;
265 if (!n)
266 n = patch->old_name;
267 quote_c_style(n, NULL, output, 0);
269 fputs(post, output);
272 #define CHUNKSIZE (8192)
273 #define SLOP (16)
275 static void read_patch_file(struct strbuf *sb, int fd)
277 if (strbuf_read(sb, fd, 0) < 0)
278 die("git apply: read returned %s", strerror(errno));
281 * Make sure that we have some slop in the buffer
282 * so that we can do speculative "memcmp" etc, and
283 * see to it that it is NUL-filled.
285 strbuf_grow(sb, SLOP);
286 memset(sb->buf + sb->len, 0, SLOP);
289 static unsigned long linelen(const char *buffer, unsigned long size)
291 unsigned long len = 0;
292 while (size--) {
293 len++;
294 if (*buffer++ == '\n')
295 break;
297 return len;
300 static int is_dev_null(const char *str)
302 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
305 #define TERM_SPACE 1
306 #define TERM_TAB 2
308 static int name_terminate(const char *name, int namelen, int c, int terminate)
310 if (c == ' ' && !(terminate & TERM_SPACE))
311 return 0;
312 if (c == '\t' && !(terminate & TERM_TAB))
313 return 0;
315 return 1;
318 static char *find_name(const char *line, char *def, int p_value, int terminate)
320 int len;
321 const char *start = line;
323 if (*line == '"') {
324 struct strbuf name;
327 * Proposed "new-style" GNU patch/diff format; see
328 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
330 strbuf_init(&name, 0);
331 if (!unquote_c_style(&name, line, NULL)) {
332 char *cp;
334 for (cp = name.buf; p_value; p_value--) {
335 cp = strchr(cp, '/');
336 if (!cp)
337 break;
338 cp++;
340 if (cp) {
341 /* name can later be freed, so we need
342 * to memmove, not just return cp
344 strbuf_remove(&name, 0, cp - name.buf);
345 free(def);
346 if (root)
347 strbuf_insert(&name, 0, root, root_len);
348 return strbuf_detach(&name, NULL);
351 strbuf_release(&name);
354 for (;;) {
355 char c = *line;
357 if (isspace(c)) {
358 if (c == '\n')
359 break;
360 if (name_terminate(start, line-start, c, terminate))
361 break;
363 line++;
364 if (c == '/' && !--p_value)
365 start = line;
367 if (!start)
368 return def;
369 len = line - start;
370 if (!len)
371 return def;
374 * Generally we prefer the shorter name, especially
375 * if the other one is just a variation of that with
376 * something else tacked on to the end (ie "file.orig"
377 * or "file~").
379 if (def) {
380 int deflen = strlen(def);
381 if (deflen < len && !strncmp(start, def, deflen))
382 return def;
383 free(def);
386 if (root) {
387 char *ret = xmalloc(root_len + len + 1);
388 strcpy(ret, root);
389 memcpy(ret + root_len, start, len);
390 ret[root_len + len] = '\0';
391 return ret;
394 return xmemdupz(start, len);
397 static int count_slashes(const char *cp)
399 int cnt = 0;
400 char ch;
402 while ((ch = *cp++))
403 if (ch == '/')
404 cnt++;
405 return cnt;
409 * Given the string after "--- " or "+++ ", guess the appropriate
410 * p_value for the given patch.
412 static int guess_p_value(const char *nameline)
414 char *name, *cp;
415 int val = -1;
417 if (is_dev_null(nameline))
418 return -1;
419 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
420 if (!name)
421 return -1;
422 cp = strchr(name, '/');
423 if (!cp)
424 val = 0;
425 else if (prefix) {
427 * Does it begin with "a/$our-prefix" and such? Then this is
428 * very likely to apply to our directory.
430 if (!strncmp(name, prefix, prefix_length))
431 val = count_slashes(prefix);
432 else {
433 cp++;
434 if (!strncmp(cp, prefix, prefix_length))
435 val = count_slashes(prefix) + 1;
438 free(name);
439 return val;
443 * Get the name etc info from the ---/+++ lines of a traditional patch header
445 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
446 * files, we can happily check the index for a match, but for creating a
447 * new file we should try to match whatever "patch" does. I have no idea.
449 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
451 char *name;
453 first += 4; /* skip "--- " */
454 second += 4; /* skip "+++ " */
455 if (!p_value_known) {
456 int p, q;
457 p = guess_p_value(first);
458 q = guess_p_value(second);
459 if (p < 0) p = q;
460 if (0 <= p && p == q) {
461 p_value = p;
462 p_value_known = 1;
465 if (is_dev_null(first)) {
466 patch->is_new = 1;
467 patch->is_delete = 0;
468 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
469 patch->new_name = name;
470 } else if (is_dev_null(second)) {
471 patch->is_new = 0;
472 patch->is_delete = 1;
473 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
474 patch->old_name = name;
475 } else {
476 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
477 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
478 patch->old_name = patch->new_name = name;
480 if (!name)
481 die("unable to find filename in patch at line %d", linenr);
484 static int gitdiff_hdrend(const char *line, struct patch *patch)
486 return -1;
490 * We're anal about diff header consistency, to make
491 * sure that we don't end up having strange ambiguous
492 * patches floating around.
494 * As a result, gitdiff_{old|new}name() will check
495 * their names against any previous information, just
496 * to make sure..
498 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
500 if (!orig_name && !isnull)
501 return find_name(line, NULL, p_value, TERM_TAB);
503 if (orig_name) {
504 int len;
505 const char *name;
506 char *another;
507 name = orig_name;
508 len = strlen(name);
509 if (isnull)
510 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
511 another = find_name(line, NULL, p_value, TERM_TAB);
512 if (!another || memcmp(another, name, len))
513 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
514 free(another);
515 return orig_name;
517 else {
518 /* expect "/dev/null" */
519 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
520 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
521 return NULL;
525 static int gitdiff_oldname(const char *line, struct patch *patch)
527 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
528 return 0;
531 static int gitdiff_newname(const char *line, struct patch *patch)
533 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
534 return 0;
537 static int gitdiff_oldmode(const char *line, struct patch *patch)
539 patch->old_mode = strtoul(line, NULL, 8);
540 return 0;
543 static int gitdiff_newmode(const char *line, struct patch *patch)
545 patch->new_mode = strtoul(line, NULL, 8);
546 return 0;
549 static int gitdiff_delete(const char *line, struct patch *patch)
551 patch->is_delete = 1;
552 patch->old_name = patch->def_name;
553 return gitdiff_oldmode(line, patch);
556 static int gitdiff_newfile(const char *line, struct patch *patch)
558 patch->is_new = 1;
559 patch->new_name = patch->def_name;
560 return gitdiff_newmode(line, patch);
563 static int gitdiff_copysrc(const char *line, struct patch *patch)
565 patch->is_copy = 1;
566 patch->old_name = find_name(line, NULL, 0, 0);
567 return 0;
570 static int gitdiff_copydst(const char *line, struct patch *patch)
572 patch->is_copy = 1;
573 patch->new_name = find_name(line, NULL, 0, 0);
574 return 0;
577 static int gitdiff_renamesrc(const char *line, struct patch *patch)
579 patch->is_rename = 1;
580 patch->old_name = find_name(line, NULL, 0, 0);
581 return 0;
584 static int gitdiff_renamedst(const char *line, struct patch *patch)
586 patch->is_rename = 1;
587 patch->new_name = find_name(line, NULL, 0, 0);
588 return 0;
591 static int gitdiff_similarity(const char *line, struct patch *patch)
593 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
594 patch->score = 0;
595 return 0;
598 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
600 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
601 patch->score = 0;
602 return 0;
605 static int gitdiff_index(const char *line, struct patch *patch)
608 * index line is N hexadecimal, "..", N hexadecimal,
609 * and optional space with octal mode.
611 const char *ptr, *eol;
612 int len;
614 ptr = strchr(line, '.');
615 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
616 return 0;
617 len = ptr - line;
618 memcpy(patch->old_sha1_prefix, line, len);
619 patch->old_sha1_prefix[len] = 0;
621 line = ptr + 2;
622 ptr = strchr(line, ' ');
623 eol = strchr(line, '\n');
625 if (!ptr || eol < ptr)
626 ptr = eol;
627 len = ptr - line;
629 if (40 < len)
630 return 0;
631 memcpy(patch->new_sha1_prefix, line, len);
632 patch->new_sha1_prefix[len] = 0;
633 if (*ptr == ' ')
634 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
635 return 0;
639 * This is normal for a diff that doesn't change anything: we'll fall through
640 * into the next diff. Tell the parser to break out.
642 static int gitdiff_unrecognized(const char *line, struct patch *patch)
644 return -1;
647 static const char *stop_at_slash(const char *line, int llen)
649 int i;
651 for (i = 0; i < llen; i++) {
652 int ch = line[i];
653 if (ch == '/')
654 return line + i;
656 return NULL;
660 * This is to extract the same name that appears on "diff --git"
661 * line. We do not find and return anything if it is a rename
662 * patch, and it is OK because we will find the name elsewhere.
663 * We need to reliably find name only when it is mode-change only,
664 * creation or deletion of an empty file. In any of these cases,
665 * both sides are the same name under a/ and b/ respectively.
667 static char *git_header_name(char *line, int llen)
669 const char *name;
670 const char *second = NULL;
671 size_t len;
673 line += strlen("diff --git ");
674 llen -= strlen("diff --git ");
676 if (*line == '"') {
677 const char *cp;
678 struct strbuf first;
679 struct strbuf sp;
681 strbuf_init(&first, 0);
682 strbuf_init(&sp, 0);
684 if (unquote_c_style(&first, line, &second))
685 goto free_and_fail1;
687 /* advance to the first slash */
688 cp = stop_at_slash(first.buf, first.len);
689 /* we do not accept absolute paths */
690 if (!cp || cp == first.buf)
691 goto free_and_fail1;
692 strbuf_remove(&first, 0, cp + 1 - first.buf);
695 * second points at one past closing dq of name.
696 * find the second name.
698 while ((second < line + llen) && isspace(*second))
699 second++;
701 if (line + llen <= second)
702 goto free_and_fail1;
703 if (*second == '"') {
704 if (unquote_c_style(&sp, second, NULL))
705 goto free_and_fail1;
706 cp = stop_at_slash(sp.buf, sp.len);
707 if (!cp || cp == sp.buf)
708 goto free_and_fail1;
709 /* They must match, otherwise ignore */
710 if (strcmp(cp + 1, first.buf))
711 goto free_and_fail1;
712 strbuf_release(&sp);
713 return strbuf_detach(&first, NULL);
716 /* unquoted second */
717 cp = stop_at_slash(second, line + llen - second);
718 if (!cp || cp == second)
719 goto free_and_fail1;
720 cp++;
721 if (line + llen - cp != first.len + 1 ||
722 memcmp(first.buf, cp, first.len))
723 goto free_and_fail1;
724 return strbuf_detach(&first, NULL);
726 free_and_fail1:
727 strbuf_release(&first);
728 strbuf_release(&sp);
729 return NULL;
732 /* unquoted first name */
733 name = stop_at_slash(line, llen);
734 if (!name || name == line)
735 return NULL;
736 name++;
739 * since the first name is unquoted, a dq if exists must be
740 * the beginning of the second name.
742 for (second = name; second < line + llen; second++) {
743 if (*second == '"') {
744 struct strbuf sp;
745 const char *np;
747 strbuf_init(&sp, 0);
748 if (unquote_c_style(&sp, second, NULL))
749 goto free_and_fail2;
751 np = stop_at_slash(sp.buf, sp.len);
752 if (!np || np == sp.buf)
753 goto free_and_fail2;
754 np++;
756 len = sp.buf + sp.len - np;
757 if (len < second - name &&
758 !strncmp(np, name, len) &&
759 isspace(name[len])) {
760 /* Good */
761 strbuf_remove(&sp, 0, np - sp.buf);
762 return strbuf_detach(&sp, NULL);
765 free_and_fail2:
766 strbuf_release(&sp);
767 return NULL;
772 * Accept a name only if it shows up twice, exactly the same
773 * form.
775 for (len = 0 ; ; len++) {
776 switch (name[len]) {
777 default:
778 continue;
779 case '\n':
780 return NULL;
781 case '\t': case ' ':
782 second = name+len;
783 for (;;) {
784 char c = *second++;
785 if (c == '\n')
786 return NULL;
787 if (c == '/')
788 break;
790 if (second[len] == '\n' && !memcmp(name, second, len)) {
791 return xmemdupz(name, len);
797 /* Verify that we recognize the lines following a git header */
798 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
800 unsigned long offset;
802 /* A git diff has explicit new/delete information, so we don't guess */
803 patch->is_new = 0;
804 patch->is_delete = 0;
807 * Some things may not have the old name in the
808 * rest of the headers anywhere (pure mode changes,
809 * or removing or adding empty files), so we get
810 * the default name from the header.
812 patch->def_name = git_header_name(line, len);
813 if (patch->def_name && root) {
814 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
815 strcpy(s, root);
816 strcpy(s + root_len, patch->def_name);
817 free(patch->def_name);
818 patch->def_name = s;
821 line += len;
822 size -= len;
823 linenr++;
824 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
825 static const struct opentry {
826 const char *str;
827 int (*fn)(const char *, struct patch *);
828 } optable[] = {
829 { "@@ -", gitdiff_hdrend },
830 { "--- ", gitdiff_oldname },
831 { "+++ ", gitdiff_newname },
832 { "old mode ", gitdiff_oldmode },
833 { "new mode ", gitdiff_newmode },
834 { "deleted file mode ", gitdiff_delete },
835 { "new file mode ", gitdiff_newfile },
836 { "copy from ", gitdiff_copysrc },
837 { "copy to ", gitdiff_copydst },
838 { "rename old ", gitdiff_renamesrc },
839 { "rename new ", gitdiff_renamedst },
840 { "rename from ", gitdiff_renamesrc },
841 { "rename to ", gitdiff_renamedst },
842 { "similarity index ", gitdiff_similarity },
843 { "dissimilarity index ", gitdiff_dissimilarity },
844 { "index ", gitdiff_index },
845 { "", gitdiff_unrecognized },
847 int i;
849 len = linelen(line, size);
850 if (!len || line[len-1] != '\n')
851 break;
852 for (i = 0; i < ARRAY_SIZE(optable); i++) {
853 const struct opentry *p = optable + i;
854 int oplen = strlen(p->str);
855 if (len < oplen || memcmp(p->str, line, oplen))
856 continue;
857 if (p->fn(line + oplen, patch) < 0)
858 return offset;
859 break;
863 return offset;
866 static int parse_num(const char *line, unsigned long *p)
868 char *ptr;
870 if (!isdigit(*line))
871 return 0;
872 *p = strtoul(line, &ptr, 10);
873 return ptr - line;
876 static int parse_range(const char *line, int len, int offset, const char *expect,
877 unsigned long *p1, unsigned long *p2)
879 int digits, ex;
881 if (offset < 0 || offset >= len)
882 return -1;
883 line += offset;
884 len -= offset;
886 digits = parse_num(line, p1);
887 if (!digits)
888 return -1;
890 offset += digits;
891 line += digits;
892 len -= digits;
894 *p2 = 1;
895 if (*line == ',') {
896 digits = parse_num(line+1, p2);
897 if (!digits)
898 return -1;
900 offset += digits+1;
901 line += digits+1;
902 len -= digits+1;
905 ex = strlen(expect);
906 if (ex > len)
907 return -1;
908 if (memcmp(line, expect, ex))
909 return -1;
911 return offset + ex;
914 static void recount_diff(char *line, int size, struct fragment *fragment)
916 int oldlines = 0, newlines = 0, ret = 0;
918 if (size < 1) {
919 warning("recount: ignore empty hunk");
920 return;
923 for (;;) {
924 int len = linelen(line, size);
925 size -= len;
926 line += len;
928 if (size < 1)
929 break;
931 switch (*line) {
932 case ' ': case '\n':
933 newlines++;
934 /* fall through */
935 case '-':
936 oldlines++;
937 continue;
938 case '+':
939 newlines++;
940 continue;
941 case '\\':
942 continue;
943 case '@':
944 ret = size < 3 || prefixcmp(line, "@@ ");
945 break;
946 case 'd':
947 ret = size < 5 || prefixcmp(line, "diff ");
948 break;
949 default:
950 ret = -1;
951 break;
953 if (ret) {
954 warning("recount: unexpected line: %.*s",
955 (int)linelen(line, size), line);
956 return;
958 break;
960 fragment->oldlines = oldlines;
961 fragment->newlines = newlines;
965 * Parse a unified diff fragment header of the
966 * form "@@ -a,b +c,d @@"
968 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
970 int offset;
972 if (!len || line[len-1] != '\n')
973 return -1;
975 /* Figure out the number of lines in a fragment */
976 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
977 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
979 return offset;
982 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
984 unsigned long offset, len;
986 patch->is_toplevel_relative = 0;
987 patch->is_rename = patch->is_copy = 0;
988 patch->is_new = patch->is_delete = -1;
989 patch->old_mode = patch->new_mode = 0;
990 patch->old_name = patch->new_name = NULL;
991 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
992 unsigned long nextlen;
994 len = linelen(line, size);
995 if (!len)
996 break;
998 /* Testing this early allows us to take a few shortcuts.. */
999 if (len < 6)
1000 continue;
1003 * Make sure we don't find any unconnected patch fragments.
1004 * That's a sign that we didn't find a header, and that a
1005 * patch has become corrupted/broken up.
1007 if (!memcmp("@@ -", line, 4)) {
1008 struct fragment dummy;
1009 if (parse_fragment_header(line, len, &dummy) < 0)
1010 continue;
1011 die("patch fragment without header at line %d: %.*s",
1012 linenr, (int)len-1, line);
1015 if (size < len + 6)
1016 break;
1019 * Git patch? It might not have a real patch, just a rename
1020 * or mode change, so we handle that specially
1022 if (!memcmp("diff --git ", line, 11)) {
1023 int git_hdr_len = parse_git_header(line, len, size, patch);
1024 if (git_hdr_len <= len)
1025 continue;
1026 if (!patch->old_name && !patch->new_name) {
1027 if (!patch->def_name)
1028 die("git diff header lacks filename information (line %d)", linenr);
1029 patch->old_name = patch->new_name = patch->def_name;
1031 patch->is_toplevel_relative = 1;
1032 *hdrsize = git_hdr_len;
1033 return offset;
1036 /* --- followed by +++ ? */
1037 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1038 continue;
1041 * We only accept unified patches, so we want it to
1042 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1043 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1045 nextlen = linelen(line + len, size - len);
1046 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1047 continue;
1049 /* Ok, we'll consider it a patch */
1050 parse_traditional_patch(line, line+len, patch);
1051 *hdrsize = len + nextlen;
1052 linenr += 2;
1053 return offset;
1055 return -1;
1058 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1060 char *err;
1061 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1062 if (!result)
1063 return;
1065 whitespace_error++;
1066 if (squelch_whitespace_errors &&
1067 squelch_whitespace_errors < whitespace_error)
1069 else {
1070 err = whitespace_error_string(result);
1071 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1072 patch_input_file, linenr, err, len - 2, line + 1);
1073 free(err);
1078 * Parse a unified diff. Note that this really needs to parse each
1079 * fragment separately, since the only way to know the difference
1080 * between a "---" that is part of a patch, and a "---" that starts
1081 * the next patch is to look at the line counts..
1083 static int parse_fragment(char *line, unsigned long size,
1084 struct patch *patch, struct fragment *fragment)
1086 int added, deleted;
1087 int len = linelen(line, size), offset;
1088 unsigned long oldlines, newlines;
1089 unsigned long leading, trailing;
1091 offset = parse_fragment_header(line, len, fragment);
1092 if (offset < 0)
1093 return -1;
1094 if (offset > 0 && patch->recount)
1095 recount_diff(line + offset, size - offset, fragment);
1096 oldlines = fragment->oldlines;
1097 newlines = fragment->newlines;
1098 leading = 0;
1099 trailing = 0;
1101 /* Parse the thing.. */
1102 line += len;
1103 size -= len;
1104 linenr++;
1105 added = deleted = 0;
1106 for (offset = len;
1107 0 < size;
1108 offset += len, size -= len, line += len, linenr++) {
1109 if (!oldlines && !newlines)
1110 break;
1111 len = linelen(line, size);
1112 if (!len || line[len-1] != '\n')
1113 return -1;
1114 switch (*line) {
1115 default:
1116 return -1;
1117 case '\n': /* newer GNU diff, an empty context line */
1118 case ' ':
1119 oldlines--;
1120 newlines--;
1121 if (!deleted && !added)
1122 leading++;
1123 trailing++;
1124 break;
1125 case '-':
1126 if (apply_in_reverse &&
1127 ws_error_action != nowarn_ws_error)
1128 check_whitespace(line, len, patch->ws_rule);
1129 deleted++;
1130 oldlines--;
1131 trailing = 0;
1132 break;
1133 case '+':
1134 if (!apply_in_reverse &&
1135 ws_error_action != nowarn_ws_error)
1136 check_whitespace(line, len, patch->ws_rule);
1137 added++;
1138 newlines--;
1139 trailing = 0;
1140 break;
1143 * We allow "\ No newline at end of file". Depending
1144 * on locale settings when the patch was produced we
1145 * don't know what this line looks like. The only
1146 * thing we do know is that it begins with "\ ".
1147 * Checking for 12 is just for sanity check -- any
1148 * l10n of "\ No newline..." is at least that long.
1150 case '\\':
1151 if (len < 12 || memcmp(line, "\\ ", 2))
1152 return -1;
1153 break;
1156 if (oldlines || newlines)
1157 return -1;
1158 fragment->leading = leading;
1159 fragment->trailing = trailing;
1162 * If a fragment ends with an incomplete line, we failed to include
1163 * it in the above loop because we hit oldlines == newlines == 0
1164 * before seeing it.
1166 if (12 < size && !memcmp(line, "\\ ", 2))
1167 offset += linelen(line, size);
1169 patch->lines_added += added;
1170 patch->lines_deleted += deleted;
1172 if (0 < patch->is_new && oldlines)
1173 return error("new file depends on old contents");
1174 if (0 < patch->is_delete && newlines)
1175 return error("deleted file still has contents");
1176 return offset;
1179 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1181 unsigned long offset = 0;
1182 unsigned long oldlines = 0, newlines = 0, context = 0;
1183 struct fragment **fragp = &patch->fragments;
1185 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1186 struct fragment *fragment;
1187 int len;
1189 fragment = xcalloc(1, sizeof(*fragment));
1190 len = parse_fragment(line, size, patch, fragment);
1191 if (len <= 0)
1192 die("corrupt patch at line %d", linenr);
1193 fragment->patch = line;
1194 fragment->size = len;
1195 oldlines += fragment->oldlines;
1196 newlines += fragment->newlines;
1197 context += fragment->leading + fragment->trailing;
1199 *fragp = fragment;
1200 fragp = &fragment->next;
1202 offset += len;
1203 line += len;
1204 size -= len;
1208 * If something was removed (i.e. we have old-lines) it cannot
1209 * be creation, and if something was added it cannot be
1210 * deletion. However, the reverse is not true; --unified=0
1211 * patches that only add are not necessarily creation even
1212 * though they do not have any old lines, and ones that only
1213 * delete are not necessarily deletion.
1215 * Unfortunately, a real creation/deletion patch do _not_ have
1216 * any context line by definition, so we cannot safely tell it
1217 * apart with --unified=0 insanity. At least if the patch has
1218 * more than one hunk it is not creation or deletion.
1220 if (patch->is_new < 0 &&
1221 (oldlines || (patch->fragments && patch->fragments->next)))
1222 patch->is_new = 0;
1223 if (patch->is_delete < 0 &&
1224 (newlines || (patch->fragments && patch->fragments->next)))
1225 patch->is_delete = 0;
1227 if (0 < patch->is_new && oldlines)
1228 die("new file %s depends on old contents", patch->new_name);
1229 if (0 < patch->is_delete && newlines)
1230 die("deleted file %s still has contents", patch->old_name);
1231 if (!patch->is_delete && !newlines && context)
1232 fprintf(stderr, "** warning: file %s becomes empty but "
1233 "is not deleted\n", patch->new_name);
1235 return offset;
1238 static inline int metadata_changes(struct patch *patch)
1240 return patch->is_rename > 0 ||
1241 patch->is_copy > 0 ||
1242 patch->is_new > 0 ||
1243 patch->is_delete ||
1244 (patch->old_mode && patch->new_mode &&
1245 patch->old_mode != patch->new_mode);
1248 static char *inflate_it(const void *data, unsigned long size,
1249 unsigned long inflated_size)
1251 z_stream stream;
1252 void *out;
1253 int st;
1255 memset(&stream, 0, sizeof(stream));
1257 stream.next_in = (unsigned char *)data;
1258 stream.avail_in = size;
1259 stream.next_out = out = xmalloc(inflated_size);
1260 stream.avail_out = inflated_size;
1261 inflateInit(&stream);
1262 st = inflate(&stream, Z_FINISH);
1263 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1264 free(out);
1265 return NULL;
1267 return out;
1270 static struct fragment *parse_binary_hunk(char **buf_p,
1271 unsigned long *sz_p,
1272 int *status_p,
1273 int *used_p)
1276 * Expect a line that begins with binary patch method ("literal"
1277 * or "delta"), followed by the length of data before deflating.
1278 * a sequence of 'length-byte' followed by base-85 encoded data
1279 * should follow, terminated by a newline.
1281 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1282 * and we would limit the patch line to 66 characters,
1283 * so one line can fit up to 13 groups that would decode
1284 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1285 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1287 int llen, used;
1288 unsigned long size = *sz_p;
1289 char *buffer = *buf_p;
1290 int patch_method;
1291 unsigned long origlen;
1292 char *data = NULL;
1293 int hunk_size = 0;
1294 struct fragment *frag;
1296 llen = linelen(buffer, size);
1297 used = llen;
1299 *status_p = 0;
1301 if (!prefixcmp(buffer, "delta ")) {
1302 patch_method = BINARY_DELTA_DEFLATED;
1303 origlen = strtoul(buffer + 6, NULL, 10);
1305 else if (!prefixcmp(buffer, "literal ")) {
1306 patch_method = BINARY_LITERAL_DEFLATED;
1307 origlen = strtoul(buffer + 8, NULL, 10);
1309 else
1310 return NULL;
1312 linenr++;
1313 buffer += llen;
1314 while (1) {
1315 int byte_length, max_byte_length, newsize;
1316 llen = linelen(buffer, size);
1317 used += llen;
1318 linenr++;
1319 if (llen == 1) {
1320 /* consume the blank line */
1321 buffer++;
1322 size--;
1323 break;
1326 * Minimum line is "A00000\n" which is 7-byte long,
1327 * and the line length must be multiple of 5 plus 2.
1329 if ((llen < 7) || (llen-2) % 5)
1330 goto corrupt;
1331 max_byte_length = (llen - 2) / 5 * 4;
1332 byte_length = *buffer;
1333 if ('A' <= byte_length && byte_length <= 'Z')
1334 byte_length = byte_length - 'A' + 1;
1335 else if ('a' <= byte_length && byte_length <= 'z')
1336 byte_length = byte_length - 'a' + 27;
1337 else
1338 goto corrupt;
1339 /* if the input length was not multiple of 4, we would
1340 * have filler at the end but the filler should never
1341 * exceed 3 bytes
1343 if (max_byte_length < byte_length ||
1344 byte_length <= max_byte_length - 4)
1345 goto corrupt;
1346 newsize = hunk_size + byte_length;
1347 data = xrealloc(data, newsize);
1348 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1349 goto corrupt;
1350 hunk_size = newsize;
1351 buffer += llen;
1352 size -= llen;
1355 frag = xcalloc(1, sizeof(*frag));
1356 frag->patch = inflate_it(data, hunk_size, origlen);
1357 if (!frag->patch)
1358 goto corrupt;
1359 free(data);
1360 frag->size = origlen;
1361 *buf_p = buffer;
1362 *sz_p = size;
1363 *used_p = used;
1364 frag->binary_patch_method = patch_method;
1365 return frag;
1367 corrupt:
1368 free(data);
1369 *status_p = -1;
1370 error("corrupt binary patch at line %d: %.*s",
1371 linenr-1, llen-1, buffer);
1372 return NULL;
1375 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1378 * We have read "GIT binary patch\n"; what follows is a line
1379 * that says the patch method (currently, either "literal" or
1380 * "delta") and the length of data before deflating; a
1381 * sequence of 'length-byte' followed by base-85 encoded data
1382 * follows.
1384 * When a binary patch is reversible, there is another binary
1385 * hunk in the same format, starting with patch method (either
1386 * "literal" or "delta") with the length of data, and a sequence
1387 * of length-byte + base-85 encoded data, terminated with another
1388 * empty line. This data, when applied to the postimage, produces
1389 * the preimage.
1391 struct fragment *forward;
1392 struct fragment *reverse;
1393 int status;
1394 int used, used_1;
1396 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1397 if (!forward && !status)
1398 /* there has to be one hunk (forward hunk) */
1399 return error("unrecognized binary patch at line %d", linenr-1);
1400 if (status)
1401 /* otherwise we already gave an error message */
1402 return status;
1404 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1405 if (reverse)
1406 used += used_1;
1407 else if (status) {
1409 * Not having reverse hunk is not an error, but having
1410 * a corrupt reverse hunk is.
1412 free((void*) forward->patch);
1413 free(forward);
1414 return status;
1416 forward->next = reverse;
1417 patch->fragments = forward;
1418 patch->is_binary = 1;
1419 return used;
1422 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1424 int hdrsize, patchsize;
1425 int offset = find_header(buffer, size, &hdrsize, patch);
1427 if (offset < 0)
1428 return offset;
1430 patch->ws_rule = whitespace_rule(patch->new_name
1431 ? patch->new_name
1432 : patch->old_name);
1434 patchsize = parse_single_patch(buffer + offset + hdrsize,
1435 size - offset - hdrsize, patch);
1437 if (!patchsize) {
1438 static const char *binhdr[] = {
1439 "Binary files ",
1440 "Files ",
1441 NULL,
1443 static const char git_binary[] = "GIT binary patch\n";
1444 int i;
1445 int hd = hdrsize + offset;
1446 unsigned long llen = linelen(buffer + hd, size - hd);
1448 if (llen == sizeof(git_binary) - 1 &&
1449 !memcmp(git_binary, buffer + hd, llen)) {
1450 int used;
1451 linenr++;
1452 used = parse_binary(buffer + hd + llen,
1453 size - hd - llen, patch);
1454 if (used)
1455 patchsize = used + llen;
1456 else
1457 patchsize = 0;
1459 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1460 for (i = 0; binhdr[i]; i++) {
1461 int len = strlen(binhdr[i]);
1462 if (len < size - hd &&
1463 !memcmp(binhdr[i], buffer + hd, len)) {
1464 linenr++;
1465 patch->is_binary = 1;
1466 patchsize = llen;
1467 break;
1472 /* Empty patch cannot be applied if it is a text patch
1473 * without metadata change. A binary patch appears
1474 * empty to us here.
1476 if ((apply || check) &&
1477 (!patch->is_binary && !metadata_changes(patch)))
1478 die("patch with only garbage at line %d", linenr);
1481 return offset + hdrsize + patchsize;
1484 #define swap(a,b) myswap((a),(b),sizeof(a))
1486 #define myswap(a, b, size) do { \
1487 unsigned char mytmp[size]; \
1488 memcpy(mytmp, &a, size); \
1489 memcpy(&a, &b, size); \
1490 memcpy(&b, mytmp, size); \
1491 } while (0)
1493 static void reverse_patches(struct patch *p)
1495 for (; p; p = p->next) {
1496 struct fragment *frag = p->fragments;
1498 swap(p->new_name, p->old_name);
1499 swap(p->new_mode, p->old_mode);
1500 swap(p->is_new, p->is_delete);
1501 swap(p->lines_added, p->lines_deleted);
1502 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1504 for (; frag; frag = frag->next) {
1505 swap(frag->newpos, frag->oldpos);
1506 swap(frag->newlines, frag->oldlines);
1511 static const char pluses[] =
1512 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1513 static const char minuses[]=
1514 "----------------------------------------------------------------------";
1516 static void show_stats(struct patch *patch)
1518 struct strbuf qname;
1519 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1520 int max, add, del;
1522 strbuf_init(&qname, 0);
1523 quote_c_style(cp, &qname, NULL, 0);
1526 * "scale" the filename
1528 max = max_len;
1529 if (max > 50)
1530 max = 50;
1532 if (qname.len > max) {
1533 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1534 if (!cp)
1535 cp = qname.buf + qname.len + 3 - max;
1536 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1539 if (patch->is_binary) {
1540 printf(" %-*s | Bin\n", max, qname.buf);
1541 strbuf_release(&qname);
1542 return;
1545 printf(" %-*s |", max, qname.buf);
1546 strbuf_release(&qname);
1549 * scale the add/delete
1551 max = max + max_change > 70 ? 70 - max : max_change;
1552 add = patch->lines_added;
1553 del = patch->lines_deleted;
1555 if (max_change > 0) {
1556 int total = ((add + del) * max + max_change / 2) / max_change;
1557 add = (add * max + max_change / 2) / max_change;
1558 del = total - add;
1560 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1561 add, pluses, del, minuses);
1564 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1566 switch (st->st_mode & S_IFMT) {
1567 case S_IFLNK:
1568 strbuf_grow(buf, st->st_size);
1569 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1570 return -1;
1571 strbuf_setlen(buf, st->st_size);
1572 return 0;
1573 case S_IFREG:
1574 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1575 return error("unable to open or read %s", path);
1576 convert_to_git(path, buf->buf, buf->len, buf, 0);
1577 return 0;
1578 default:
1579 return -1;
1583 static void update_pre_post_images(struct image *preimage,
1584 struct image *postimage,
1585 char *buf,
1586 size_t len)
1588 int i, ctx;
1589 char *new, *old, *fixed;
1590 struct image fixed_preimage;
1593 * Update the preimage with whitespace fixes. Note that we
1594 * are not losing preimage->buf -- apply_one_fragment() will
1595 * free "oldlines".
1597 prepare_image(&fixed_preimage, buf, len, 1);
1598 assert(fixed_preimage.nr == preimage->nr);
1599 for (i = 0; i < preimage->nr; i++)
1600 fixed_preimage.line[i].flag = preimage->line[i].flag;
1601 free(preimage->line_allocated);
1602 *preimage = fixed_preimage;
1605 * Adjust the common context lines in postimage, in place.
1606 * This is possible because whitespace fixing does not make
1607 * the string grow.
1609 new = old = postimage->buf;
1610 fixed = preimage->buf;
1611 for (i = ctx = 0; i < postimage->nr; i++) {
1612 size_t len = postimage->line[i].len;
1613 if (!(postimage->line[i].flag & LINE_COMMON)) {
1614 /* an added line -- no counterparts in preimage */
1615 memmove(new, old, len);
1616 old += len;
1617 new += len;
1618 continue;
1621 /* a common context -- skip it in the original postimage */
1622 old += len;
1624 /* and find the corresponding one in the fixed preimage */
1625 while (ctx < preimage->nr &&
1626 !(preimage->line[ctx].flag & LINE_COMMON)) {
1627 fixed += preimage->line[ctx].len;
1628 ctx++;
1630 if (preimage->nr <= ctx)
1631 die("oops");
1633 /* and copy it in, while fixing the line length */
1634 len = preimage->line[ctx].len;
1635 memcpy(new, fixed, len);
1636 new += len;
1637 fixed += len;
1638 postimage->line[i].len = len;
1639 ctx++;
1642 /* Fix the length of the whole thing */
1643 postimage->len = new - postimage->buf;
1646 static int match_fragment(struct image *img,
1647 struct image *preimage,
1648 struct image *postimage,
1649 unsigned long try,
1650 int try_lno,
1651 unsigned ws_rule,
1652 int match_beginning, int match_end)
1654 int i;
1655 char *fixed_buf, *buf, *orig, *target;
1657 if (preimage->nr + try_lno > img->nr)
1658 return 0;
1660 if (match_beginning && try_lno)
1661 return 0;
1663 if (match_end && preimage->nr + try_lno != img->nr)
1664 return 0;
1666 /* Quick hash check */
1667 for (i = 0; i < preimage->nr; i++)
1668 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1669 return 0;
1672 * Do we have an exact match? If we were told to match
1673 * at the end, size must be exactly at try+fragsize,
1674 * otherwise try+fragsize must be still within the preimage,
1675 * and either case, the old piece should match the preimage
1676 * exactly.
1678 if ((match_end
1679 ? (try + preimage->len == img->len)
1680 : (try + preimage->len <= img->len)) &&
1681 !memcmp(img->buf + try, preimage->buf, preimage->len))
1682 return 1;
1684 if (ws_error_action != correct_ws_error)
1685 return 0;
1688 * The hunk does not apply byte-by-byte, but the hash says
1689 * it might with whitespace fuzz.
1691 fixed_buf = xmalloc(preimage->len + 1);
1692 buf = fixed_buf;
1693 orig = preimage->buf;
1694 target = img->buf + try;
1695 for (i = 0; i < preimage->nr; i++) {
1696 size_t fixlen; /* length after fixing the preimage */
1697 size_t oldlen = preimage->line[i].len;
1698 size_t tgtlen = img->line[try_lno + i].len;
1699 size_t tgtfixlen; /* length after fixing the target line */
1700 char tgtfixbuf[1024], *tgtfix;
1701 int match;
1703 /* Try fixing the line in the preimage */
1704 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1706 /* Try fixing the line in the target */
1707 if (sizeof(tgtfixbuf) > tgtlen)
1708 tgtfix = tgtfixbuf;
1709 else
1710 tgtfix = xmalloc(tgtlen);
1711 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1714 * If they match, either the preimage was based on
1715 * a version before our tree fixed whitespace breakage,
1716 * or we are lacking a whitespace-fix patch the tree
1717 * the preimage was based on already had (i.e. target
1718 * has whitespace breakage, the preimage doesn't).
1719 * In either case, we are fixing the whitespace breakages
1720 * so we might as well take the fix together with their
1721 * real change.
1723 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1725 if (tgtfix != tgtfixbuf)
1726 free(tgtfix);
1727 if (!match)
1728 goto unmatch_exit;
1730 orig += oldlen;
1731 buf += fixlen;
1732 target += tgtlen;
1736 * Yes, the preimage is based on an older version that still
1737 * has whitespace breakages unfixed, and fixing them makes the
1738 * hunk match. Update the context lines in the postimage.
1740 update_pre_post_images(preimage, postimage,
1741 fixed_buf, buf - fixed_buf);
1742 return 1;
1744 unmatch_exit:
1745 free(fixed_buf);
1746 return 0;
1749 static int find_pos(struct image *img,
1750 struct image *preimage,
1751 struct image *postimage,
1752 int line,
1753 unsigned ws_rule,
1754 int match_beginning, int match_end)
1756 int i;
1757 unsigned long backwards, forwards, try;
1758 int backwards_lno, forwards_lno, try_lno;
1760 if (preimage->nr > img->nr)
1761 return -1;
1764 * If match_begining or match_end is specified, there is no
1765 * point starting from a wrong line that will never match and
1766 * wander around and wait for a match at the specified end.
1768 if (match_beginning)
1769 line = 0;
1770 else if (match_end)
1771 line = img->nr - preimage->nr;
1773 if (line > img->nr)
1774 line = img->nr;
1776 try = 0;
1777 for (i = 0; i < line; i++)
1778 try += img->line[i].len;
1781 * There's probably some smart way to do this, but I'll leave
1782 * that to the smart and beautiful people. I'm simple and stupid.
1784 backwards = try;
1785 backwards_lno = line;
1786 forwards = try;
1787 forwards_lno = line;
1788 try_lno = line;
1790 for (i = 0; ; i++) {
1791 if (match_fragment(img, preimage, postimage,
1792 try, try_lno, ws_rule,
1793 match_beginning, match_end))
1794 return try_lno;
1796 again:
1797 if (backwards_lno == 0 && forwards_lno == img->nr)
1798 break;
1800 if (i & 1) {
1801 if (backwards_lno == 0) {
1802 i++;
1803 goto again;
1805 backwards_lno--;
1806 backwards -= img->line[backwards_lno].len;
1807 try = backwards;
1808 try_lno = backwards_lno;
1809 } else {
1810 if (forwards_lno == img->nr) {
1811 i++;
1812 goto again;
1814 forwards += img->line[forwards_lno].len;
1815 forwards_lno++;
1816 try = forwards;
1817 try_lno = forwards_lno;
1821 return -1;
1824 static void remove_first_line(struct image *img)
1826 img->buf += img->line[0].len;
1827 img->len -= img->line[0].len;
1828 img->line++;
1829 img->nr--;
1832 static void remove_last_line(struct image *img)
1834 img->len -= img->line[--img->nr].len;
1837 static void update_image(struct image *img,
1838 int applied_pos,
1839 struct image *preimage,
1840 struct image *postimage)
1843 * remove the copy of preimage at offset in img
1844 * and replace it with postimage
1846 int i, nr;
1847 size_t remove_count, insert_count, applied_at = 0;
1848 char *result;
1850 for (i = 0; i < applied_pos; i++)
1851 applied_at += img->line[i].len;
1853 remove_count = 0;
1854 for (i = 0; i < preimage->nr; i++)
1855 remove_count += img->line[applied_pos + i].len;
1856 insert_count = postimage->len;
1858 /* Adjust the contents */
1859 result = xmalloc(img->len + insert_count - remove_count + 1);
1860 memcpy(result, img->buf, applied_at);
1861 memcpy(result + applied_at, postimage->buf, postimage->len);
1862 memcpy(result + applied_at + postimage->len,
1863 img->buf + (applied_at + remove_count),
1864 img->len - (applied_at + remove_count));
1865 free(img->buf);
1866 img->buf = result;
1867 img->len += insert_count - remove_count;
1868 result[img->len] = '\0';
1870 /* Adjust the line table */
1871 nr = img->nr + postimage->nr - preimage->nr;
1872 if (preimage->nr < postimage->nr) {
1874 * NOTE: this knows that we never call remove_first_line()
1875 * on anything other than pre/post image.
1877 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1878 img->line_allocated = img->line;
1880 if (preimage->nr != postimage->nr)
1881 memmove(img->line + applied_pos + postimage->nr,
1882 img->line + applied_pos + preimage->nr,
1883 (img->nr - (applied_pos + preimage->nr)) *
1884 sizeof(*img->line));
1885 memcpy(img->line + applied_pos,
1886 postimage->line,
1887 postimage->nr * sizeof(*img->line));
1888 img->nr = nr;
1891 static int apply_one_fragment(struct image *img, struct fragment *frag,
1892 int inaccurate_eof, unsigned ws_rule)
1894 int match_beginning, match_end;
1895 const char *patch = frag->patch;
1896 int size = frag->size;
1897 char *old, *new, *oldlines, *newlines;
1898 int new_blank_lines_at_end = 0;
1899 unsigned long leading, trailing;
1900 int pos, applied_pos;
1901 struct image preimage;
1902 struct image postimage;
1904 memset(&preimage, 0, sizeof(preimage));
1905 memset(&postimage, 0, sizeof(postimage));
1906 oldlines = xmalloc(size);
1907 newlines = xmalloc(size);
1909 old = oldlines;
1910 new = newlines;
1911 while (size > 0) {
1912 char first;
1913 int len = linelen(patch, size);
1914 int plen, added;
1915 int added_blank_line = 0;
1917 if (!len)
1918 break;
1921 * "plen" is how much of the line we should use for
1922 * the actual patch data. Normally we just remove the
1923 * first character on the line, but if the line is
1924 * followed by "\ No newline", then we also remove the
1925 * last one (which is the newline, of course).
1927 plen = len - 1;
1928 if (len < size && patch[len] == '\\')
1929 plen--;
1930 first = *patch;
1931 if (apply_in_reverse) {
1932 if (first == '-')
1933 first = '+';
1934 else if (first == '+')
1935 first = '-';
1938 switch (first) {
1939 case '\n':
1940 /* Newer GNU diff, empty context line */
1941 if (plen < 0)
1942 /* ... followed by '\No newline'; nothing */
1943 break;
1944 *old++ = '\n';
1945 *new++ = '\n';
1946 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1947 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1948 break;
1949 case ' ':
1950 case '-':
1951 memcpy(old, patch + 1, plen);
1952 add_line_info(&preimage, old, plen,
1953 (first == ' ' ? LINE_COMMON : 0));
1954 old += plen;
1955 if (first == '-')
1956 break;
1957 /* Fall-through for ' ' */
1958 case '+':
1959 /* --no-add does not add new lines */
1960 if (first == '+' && no_add)
1961 break;
1963 if (first != '+' ||
1964 !whitespace_error ||
1965 ws_error_action != correct_ws_error) {
1966 memcpy(new, patch + 1, plen);
1967 added = plen;
1969 else {
1970 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1972 add_line_info(&postimage, new, added,
1973 (first == '+' ? 0 : LINE_COMMON));
1974 new += added;
1975 if (first == '+' &&
1976 added == 1 && new[-1] == '\n')
1977 added_blank_line = 1;
1978 break;
1979 case '@': case '\\':
1980 /* Ignore it, we already handled it */
1981 break;
1982 default:
1983 if (apply_verbosely)
1984 error("invalid start of line: '%c'", first);
1985 return -1;
1987 if (added_blank_line)
1988 new_blank_lines_at_end++;
1989 else
1990 new_blank_lines_at_end = 0;
1991 patch += len;
1992 size -= len;
1994 if (inaccurate_eof &&
1995 old > oldlines && old[-1] == '\n' &&
1996 new > newlines && new[-1] == '\n') {
1997 old--;
1998 new--;
2001 leading = frag->leading;
2002 trailing = frag->trailing;
2005 * A hunk to change lines at the beginning would begin with
2006 * @@ -1,L +N,M @@
2007 * but we need to be careful. -U0 that inserts before the second
2008 * line also has this pattern.
2010 * And a hunk to add to an empty file would begin with
2011 * @@ -0,0 +N,M @@
2013 * In other words, a hunk that is (frag->oldpos <= 1) with or
2014 * without leading context must match at the beginning.
2016 match_beginning = (!frag->oldpos ||
2017 (frag->oldpos == 1 && !unidiff_zero));
2020 * A hunk without trailing lines must match at the end.
2021 * However, we simply cannot tell if a hunk must match end
2022 * from the lack of trailing lines if the patch was generated
2023 * with unidiff without any context.
2025 match_end = !unidiff_zero && !trailing;
2027 pos = frag->newpos ? (frag->newpos - 1) : 0;
2028 preimage.buf = oldlines;
2029 preimage.len = old - oldlines;
2030 postimage.buf = newlines;
2031 postimage.len = new - newlines;
2032 preimage.line = preimage.line_allocated;
2033 postimage.line = postimage.line_allocated;
2035 for (;;) {
2037 applied_pos = find_pos(img, &preimage, &postimage, pos,
2038 ws_rule, match_beginning, match_end);
2040 if (applied_pos >= 0)
2041 break;
2043 /* Am I at my context limits? */
2044 if ((leading <= p_context) && (trailing <= p_context))
2045 break;
2046 if (match_beginning || match_end) {
2047 match_beginning = match_end = 0;
2048 continue;
2052 * Reduce the number of context lines; reduce both
2053 * leading and trailing if they are equal otherwise
2054 * just reduce the larger context.
2056 if (leading >= trailing) {
2057 remove_first_line(&preimage);
2058 remove_first_line(&postimage);
2059 pos--;
2060 leading--;
2062 if (trailing > leading) {
2063 remove_last_line(&preimage);
2064 remove_last_line(&postimage);
2065 trailing--;
2069 if (applied_pos >= 0) {
2070 if (ws_error_action == correct_ws_error &&
2071 new_blank_lines_at_end &&
2072 postimage.nr + applied_pos == img->nr) {
2074 * If the patch application adds blank lines
2075 * at the end, and if the patch applies at the
2076 * end of the image, remove those added blank
2077 * lines.
2079 while (new_blank_lines_at_end--)
2080 remove_last_line(&postimage);
2084 * Warn if it was necessary to reduce the number
2085 * of context lines.
2087 if ((leading != frag->leading) ||
2088 (trailing != frag->trailing))
2089 fprintf(stderr, "Context reduced to (%ld/%ld)"
2090 " to apply fragment at %d\n",
2091 leading, trailing, applied_pos+1);
2092 update_image(img, applied_pos, &preimage, &postimage);
2093 } else {
2094 if (apply_verbosely)
2095 error("while searching for:\n%.*s",
2096 (int)(old - oldlines), oldlines);
2099 free(oldlines);
2100 free(newlines);
2101 free(preimage.line_allocated);
2102 free(postimage.line_allocated);
2104 return (applied_pos < 0);
2107 static int apply_binary_fragment(struct image *img, struct patch *patch)
2109 struct fragment *fragment = patch->fragments;
2110 unsigned long len;
2111 void *dst;
2113 /* Binary patch is irreversible without the optional second hunk */
2114 if (apply_in_reverse) {
2115 if (!fragment->next)
2116 return error("cannot reverse-apply a binary patch "
2117 "without the reverse hunk to '%s'",
2118 patch->new_name
2119 ? patch->new_name : patch->old_name);
2120 fragment = fragment->next;
2122 switch (fragment->binary_patch_method) {
2123 case BINARY_DELTA_DEFLATED:
2124 dst = patch_delta(img->buf, img->len, fragment->patch,
2125 fragment->size, &len);
2126 if (!dst)
2127 return -1;
2128 clear_image(img);
2129 img->buf = dst;
2130 img->len = len;
2131 return 0;
2132 case BINARY_LITERAL_DEFLATED:
2133 clear_image(img);
2134 img->len = fragment->size;
2135 img->buf = xmalloc(img->len+1);
2136 memcpy(img->buf, fragment->patch, img->len);
2137 img->buf[img->len] = '\0';
2138 return 0;
2140 return -1;
2143 static int apply_binary(struct image *img, struct patch *patch)
2145 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2146 unsigned char sha1[20];
2149 * For safety, we require patch index line to contain
2150 * full 40-byte textual SHA1 for old and new, at least for now.
2152 if (strlen(patch->old_sha1_prefix) != 40 ||
2153 strlen(patch->new_sha1_prefix) != 40 ||
2154 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2155 get_sha1_hex(patch->new_sha1_prefix, sha1))
2156 return error("cannot apply binary patch to '%s' "
2157 "without full index line", name);
2159 if (patch->old_name) {
2161 * See if the old one matches what the patch
2162 * applies to.
2164 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2165 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2166 return error("the patch applies to '%s' (%s), "
2167 "which does not match the "
2168 "current contents.",
2169 name, sha1_to_hex(sha1));
2171 else {
2172 /* Otherwise, the old one must be empty. */
2173 if (img->len)
2174 return error("the patch applies to an empty "
2175 "'%s' but it is not empty", name);
2178 get_sha1_hex(patch->new_sha1_prefix, sha1);
2179 if (is_null_sha1(sha1)) {
2180 clear_image(img);
2181 return 0; /* deletion patch */
2184 if (has_sha1_file(sha1)) {
2185 /* We already have the postimage */
2186 enum object_type type;
2187 unsigned long size;
2188 char *result;
2190 result = read_sha1_file(sha1, &type, &size);
2191 if (!result)
2192 return error("the necessary postimage %s for "
2193 "'%s' cannot be read",
2194 patch->new_sha1_prefix, name);
2195 clear_image(img);
2196 img->buf = result;
2197 img->len = size;
2198 } else {
2200 * We have verified buf matches the preimage;
2201 * apply the patch data to it, which is stored
2202 * in the patch->fragments->{patch,size}.
2204 if (apply_binary_fragment(img, patch))
2205 return error("binary patch does not apply to '%s'",
2206 name);
2208 /* verify that the result matches */
2209 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2210 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2211 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2212 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2215 return 0;
2218 static int apply_fragments(struct image *img, struct patch *patch)
2220 struct fragment *frag = patch->fragments;
2221 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2222 unsigned ws_rule = patch->ws_rule;
2223 unsigned inaccurate_eof = patch->inaccurate_eof;
2225 if (patch->is_binary)
2226 return apply_binary(img, patch);
2228 while (frag) {
2229 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2230 error("patch failed: %s:%ld", name, frag->oldpos);
2231 if (!apply_with_reject)
2232 return -1;
2233 frag->rejected = 1;
2235 frag = frag->next;
2237 return 0;
2240 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2242 if (!ce)
2243 return 0;
2245 if (S_ISGITLINK(ce->ce_mode)) {
2246 strbuf_grow(buf, 100);
2247 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2248 } else {
2249 enum object_type type;
2250 unsigned long sz;
2251 char *result;
2253 result = read_sha1_file(ce->sha1, &type, &sz);
2254 if (!result)
2255 return -1;
2256 /* XXX read_sha1_file NUL-terminates */
2257 strbuf_attach(buf, result, sz, sz + 1);
2259 return 0;
2262 static struct patch *in_fn_table(const char *name)
2264 struct string_list_item *item;
2266 if (name == NULL)
2267 return NULL;
2269 item = string_list_lookup(name, &fn_table);
2270 if (item != NULL)
2271 return (struct patch *)item->util;
2273 return NULL;
2276 static void add_to_fn_table(struct patch *patch)
2278 struct string_list_item *item;
2281 * Always add new_name unless patch is a deletion
2282 * This should cover the cases for normal diffs,
2283 * file creations and copies
2285 if (patch->new_name != NULL) {
2286 item = string_list_insert(patch->new_name, &fn_table);
2287 item->util = patch;
2291 * store a failure on rename/deletion cases because
2292 * later chunks shouldn't patch old names
2294 if ((patch->new_name == NULL) || (patch->is_rename)) {
2295 item = string_list_insert(patch->old_name, &fn_table);
2296 item->util = (struct patch *) -1;
2300 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2302 struct strbuf buf;
2303 struct image image;
2304 size_t len;
2305 char *img;
2306 struct patch *tpatch;
2308 strbuf_init(&buf, 0);
2310 if (!(patch->is_copy || patch->is_rename) &&
2311 ((tpatch = in_fn_table(patch->old_name)) != NULL)) {
2312 if (tpatch == (struct patch *) -1) {
2313 return error("patch %s has been renamed/deleted",
2314 patch->old_name);
2316 /* We have a patched copy in memory use that */
2317 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2318 } else if (cached) {
2319 if (read_file_or_gitlink(ce, &buf))
2320 return error("read of %s failed", patch->old_name);
2321 } else if (patch->old_name) {
2322 if (S_ISGITLINK(patch->old_mode)) {
2323 if (ce) {
2324 read_file_or_gitlink(ce, &buf);
2325 } else {
2327 * There is no way to apply subproject
2328 * patch without looking at the index.
2330 patch->fragments = NULL;
2332 } else {
2333 if (read_old_data(st, patch->old_name, &buf))
2334 return error("read of %s failed", patch->old_name);
2338 img = strbuf_detach(&buf, &len);
2339 prepare_image(&image, img, len, !patch->is_binary);
2341 if (apply_fragments(&image, patch) < 0)
2342 return -1; /* note with --reject this succeeds. */
2343 patch->result = image.buf;
2344 patch->resultsize = image.len;
2345 add_to_fn_table(patch);
2346 free(image.line_allocated);
2348 if (0 < patch->is_delete && patch->resultsize)
2349 return error("removal patch leaves file contents");
2351 return 0;
2354 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2356 struct stat nst;
2357 if (!lstat(new_name, &nst)) {
2358 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2359 return 0;
2361 * A leading component of new_name might be a symlink
2362 * that is going to be removed with this patch, but
2363 * still pointing at somewhere that has the path.
2364 * In such a case, path "new_name" does not exist as
2365 * far as git is concerned.
2367 if (has_symlink_leading_path(strlen(new_name), new_name))
2368 return 0;
2370 return error("%s: already exists in working directory", new_name);
2372 else if ((errno != ENOENT) && (errno != ENOTDIR))
2373 return error("%s: %s", new_name, strerror(errno));
2374 return 0;
2377 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2379 if (S_ISGITLINK(ce->ce_mode)) {
2380 if (!S_ISDIR(st->st_mode))
2381 return -1;
2382 return 0;
2384 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2387 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2389 const char *old_name = patch->old_name;
2390 struct patch *tpatch = NULL;
2391 int stat_ret = 0;
2392 unsigned st_mode = 0;
2395 * Make sure that we do not have local modifications from the
2396 * index when we are looking at the index. Also make sure
2397 * we have the preimage file to be patched in the work tree,
2398 * unless --cached, which tells git to apply only in the index.
2400 if (!old_name)
2401 return 0;
2403 assert(patch->is_new <= 0);
2405 if (!(patch->is_copy || patch->is_rename) &&
2406 (tpatch = in_fn_table(old_name)) != NULL) {
2407 if (tpatch == (struct patch *) -1) {
2408 return error("%s: has been deleted/renamed", old_name);
2410 st_mode = tpatch->new_mode;
2411 } else if (!cached) {
2412 stat_ret = lstat(old_name, st);
2413 if (stat_ret && errno != ENOENT)
2414 return error("%s: %s", old_name, strerror(errno));
2417 if (check_index && !tpatch) {
2418 int pos = cache_name_pos(old_name, strlen(old_name));
2419 if (pos < 0) {
2420 if (patch->is_new < 0)
2421 goto is_new;
2422 return error("%s: does not exist in index", old_name);
2424 *ce = active_cache[pos];
2425 if (stat_ret < 0) {
2426 struct checkout costate;
2427 /* checkout */
2428 costate.base_dir = "";
2429 costate.base_dir_len = 0;
2430 costate.force = 0;
2431 costate.quiet = 0;
2432 costate.not_new = 0;
2433 costate.refresh_cache = 1;
2434 if (checkout_entry(*ce, &costate, NULL) ||
2435 lstat(old_name, st))
2436 return -1;
2438 if (!cached && verify_index_match(*ce, st))
2439 return error("%s: does not match index", old_name);
2440 if (cached)
2441 st_mode = (*ce)->ce_mode;
2442 } else if (stat_ret < 0) {
2443 if (patch->is_new < 0)
2444 goto is_new;
2445 return error("%s: %s", old_name, strerror(errno));
2448 if (!cached && !tpatch)
2449 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2451 if (patch->is_new < 0)
2452 patch->is_new = 0;
2453 if (!patch->old_mode)
2454 patch->old_mode = st_mode;
2455 if ((st_mode ^ patch->old_mode) & S_IFMT)
2456 return error("%s: wrong type", old_name);
2457 if (st_mode != patch->old_mode)
2458 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2459 old_name, st_mode, patch->old_mode);
2460 return 0;
2462 is_new:
2463 patch->is_new = 1;
2464 patch->is_delete = 0;
2465 patch->old_name = NULL;
2466 return 0;
2469 static int check_patch(struct patch *patch)
2471 struct stat st;
2472 const char *old_name = patch->old_name;
2473 const char *new_name = patch->new_name;
2474 const char *name = old_name ? old_name : new_name;
2475 struct cache_entry *ce = NULL;
2476 int ok_if_exists;
2477 int status;
2479 patch->rejected = 1; /* we will drop this after we succeed */
2481 status = check_preimage(patch, &ce, &st);
2482 if (status)
2483 return status;
2484 old_name = patch->old_name;
2486 if (in_fn_table(new_name) == (struct patch *) -1)
2488 * A type-change diff is always split into a patch to
2489 * delete old, immediately followed by a patch to
2490 * create new (see diff.c::run_diff()); in such a case
2491 * it is Ok that the entry to be deleted by the
2492 * previous patch is still in the working tree and in
2493 * the index.
2495 ok_if_exists = 1;
2496 else
2497 ok_if_exists = 0;
2499 if (new_name &&
2500 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2501 if (check_index &&
2502 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2503 !ok_if_exists)
2504 return error("%s: already exists in index", new_name);
2505 if (!cached) {
2506 int err = check_to_create_blob(new_name, ok_if_exists);
2507 if (err)
2508 return err;
2510 if (!patch->new_mode) {
2511 if (0 < patch->is_new)
2512 patch->new_mode = S_IFREG | 0644;
2513 else
2514 patch->new_mode = patch->old_mode;
2518 if (new_name && old_name) {
2519 int same = !strcmp(old_name, new_name);
2520 if (!patch->new_mode)
2521 patch->new_mode = patch->old_mode;
2522 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2523 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2524 patch->new_mode, new_name, patch->old_mode,
2525 same ? "" : " of ", same ? "" : old_name);
2528 if (apply_data(patch, &st, ce) < 0)
2529 return error("%s: patch does not apply", name);
2530 patch->rejected = 0;
2531 return 0;
2534 static int check_patch_list(struct patch *patch)
2536 int err = 0;
2538 while (patch) {
2539 if (apply_verbosely)
2540 say_patch_name(stderr,
2541 "Checking patch ", patch, "...\n");
2542 err |= check_patch(patch);
2543 patch = patch->next;
2545 return err;
2548 /* This function tries to read the sha1 from the current index */
2549 static int get_current_sha1(const char *path, unsigned char *sha1)
2551 int pos;
2553 if (read_cache() < 0)
2554 return -1;
2555 pos = cache_name_pos(path, strlen(path));
2556 if (pos < 0)
2557 return -1;
2558 hashcpy(sha1, active_cache[pos]->sha1);
2559 return 0;
2562 /* Build an index that contains the just the files needed for a 3way merge */
2563 static void build_fake_ancestor(struct patch *list, const char *filename)
2565 struct patch *patch;
2566 struct index_state result = { 0 };
2567 int fd;
2569 /* Once we start supporting the reverse patch, it may be
2570 * worth showing the new sha1 prefix, but until then...
2572 for (patch = list; patch; patch = patch->next) {
2573 const unsigned char *sha1_ptr;
2574 unsigned char sha1[20];
2575 struct cache_entry *ce;
2576 const char *name;
2578 name = patch->old_name ? patch->old_name : patch->new_name;
2579 if (0 < patch->is_new)
2580 continue;
2581 else if (get_sha1(patch->old_sha1_prefix, sha1))
2582 /* git diff has no index line for mode/type changes */
2583 if (!patch->lines_added && !patch->lines_deleted) {
2584 if (get_current_sha1(patch->new_name, sha1) ||
2585 get_current_sha1(patch->old_name, sha1))
2586 die("mode change for %s, which is not "
2587 "in current HEAD", name);
2588 sha1_ptr = sha1;
2589 } else
2590 die("sha1 information is lacking or useless "
2591 "(%s).", name);
2592 else
2593 sha1_ptr = sha1;
2595 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2596 if (!ce)
2597 die("make_cache_entry failed for path '%s'", name);
2598 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2599 die ("Could not add %s to temporary index", name);
2602 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2603 if (fd < 0 || write_index(&result, fd) || close(fd))
2604 die ("Could not write temporary index to %s", filename);
2606 discard_index(&result);
2609 static void stat_patch_list(struct patch *patch)
2611 int files, adds, dels;
2613 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2614 files++;
2615 adds += patch->lines_added;
2616 dels += patch->lines_deleted;
2617 show_stats(patch);
2620 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2623 static void numstat_patch_list(struct patch *patch)
2625 for ( ; patch; patch = patch->next) {
2626 const char *name;
2627 name = patch->new_name ? patch->new_name : patch->old_name;
2628 if (patch->is_binary)
2629 printf("-\t-\t");
2630 else
2631 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2632 write_name_quoted(name, stdout, line_termination);
2636 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2638 if (mode)
2639 printf(" %s mode %06o %s\n", newdelete, mode, name);
2640 else
2641 printf(" %s %s\n", newdelete, name);
2644 static void show_mode_change(struct patch *p, int show_name)
2646 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2647 if (show_name)
2648 printf(" mode change %06o => %06o %s\n",
2649 p->old_mode, p->new_mode, p->new_name);
2650 else
2651 printf(" mode change %06o => %06o\n",
2652 p->old_mode, p->new_mode);
2656 static void show_rename_copy(struct patch *p)
2658 const char *renamecopy = p->is_rename ? "rename" : "copy";
2659 const char *old, *new;
2661 /* Find common prefix */
2662 old = p->old_name;
2663 new = p->new_name;
2664 while (1) {
2665 const char *slash_old, *slash_new;
2666 slash_old = strchr(old, '/');
2667 slash_new = strchr(new, '/');
2668 if (!slash_old ||
2669 !slash_new ||
2670 slash_old - old != slash_new - new ||
2671 memcmp(old, new, slash_new - new))
2672 break;
2673 old = slash_old + 1;
2674 new = slash_new + 1;
2676 /* p->old_name thru old is the common prefix, and old and new
2677 * through the end of names are renames
2679 if (old != p->old_name)
2680 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2681 (int)(old - p->old_name), p->old_name,
2682 old, new, p->score);
2683 else
2684 printf(" %s %s => %s (%d%%)\n", renamecopy,
2685 p->old_name, p->new_name, p->score);
2686 show_mode_change(p, 0);
2689 static void summary_patch_list(struct patch *patch)
2691 struct patch *p;
2693 for (p = patch; p; p = p->next) {
2694 if (p->is_new)
2695 show_file_mode_name("create", p->new_mode, p->new_name);
2696 else if (p->is_delete)
2697 show_file_mode_name("delete", p->old_mode, p->old_name);
2698 else {
2699 if (p->is_rename || p->is_copy)
2700 show_rename_copy(p);
2701 else {
2702 if (p->score) {
2703 printf(" rewrite %s (%d%%)\n",
2704 p->new_name, p->score);
2705 show_mode_change(p, 0);
2707 else
2708 show_mode_change(p, 1);
2714 static void patch_stats(struct patch *patch)
2716 int lines = patch->lines_added + patch->lines_deleted;
2718 if (lines > max_change)
2719 max_change = lines;
2720 if (patch->old_name) {
2721 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2722 if (!len)
2723 len = strlen(patch->old_name);
2724 if (len > max_len)
2725 max_len = len;
2727 if (patch->new_name) {
2728 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2729 if (!len)
2730 len = strlen(patch->new_name);
2731 if (len > max_len)
2732 max_len = len;
2736 static void remove_file(struct patch *patch, int rmdir_empty)
2738 if (update_index) {
2739 if (remove_file_from_cache(patch->old_name) < 0)
2740 die("unable to remove %s from index", patch->old_name);
2742 if (!cached) {
2743 if (S_ISGITLINK(patch->old_mode)) {
2744 if (rmdir(patch->old_name))
2745 warning("unable to remove submodule %s",
2746 patch->old_name);
2747 } else if (!unlink(patch->old_name) && rmdir_empty) {
2748 remove_path(patch->old_name);
2753 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2755 struct stat st;
2756 struct cache_entry *ce;
2757 int namelen = strlen(path);
2758 unsigned ce_size = cache_entry_size(namelen);
2760 if (!update_index)
2761 return;
2763 ce = xcalloc(1, ce_size);
2764 memcpy(ce->name, path, namelen);
2765 ce->ce_mode = create_ce_mode(mode);
2766 ce->ce_flags = namelen;
2767 if (S_ISGITLINK(mode)) {
2768 const char *s = buf;
2770 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2771 die("corrupt patch for subproject %s", path);
2772 } else {
2773 if (!cached) {
2774 if (lstat(path, &st) < 0)
2775 die("unable to stat newly created file %s",
2776 path);
2777 fill_stat_cache_info(ce, &st);
2779 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2780 die("unable to create backing store for newly created file %s", path);
2782 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2783 die("unable to add cache entry for %s", path);
2786 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2788 int fd;
2789 struct strbuf nbuf;
2791 if (S_ISGITLINK(mode)) {
2792 struct stat st;
2793 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2794 return 0;
2795 return mkdir(path, 0777);
2798 if (has_symlinks && S_ISLNK(mode))
2799 /* Although buf:size is counted string, it also is NUL
2800 * terminated.
2802 return symlink(buf, path);
2804 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2805 if (fd < 0)
2806 return -1;
2808 strbuf_init(&nbuf, 0);
2809 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2810 size = nbuf.len;
2811 buf = nbuf.buf;
2813 write_or_die(fd, buf, size);
2814 strbuf_release(&nbuf);
2816 if (close(fd) < 0)
2817 die("closing file %s: %s", path, strerror(errno));
2818 return 0;
2822 * We optimistically assume that the directories exist,
2823 * which is true 99% of the time anyway. If they don't,
2824 * we create them and try again.
2826 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2828 if (cached)
2829 return;
2830 if (!try_create_file(path, mode, buf, size))
2831 return;
2833 if (errno == ENOENT) {
2834 if (safe_create_leading_directories(path))
2835 return;
2836 if (!try_create_file(path, mode, buf, size))
2837 return;
2840 if (errno == EEXIST || errno == EACCES) {
2841 /* We may be trying to create a file where a directory
2842 * used to be.
2844 struct stat st;
2845 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2846 errno = EEXIST;
2849 if (errno == EEXIST) {
2850 unsigned int nr = getpid();
2852 for (;;) {
2853 char newpath[PATH_MAX];
2854 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
2855 if (!try_create_file(newpath, mode, buf, size)) {
2856 if (!rename(newpath, path))
2857 return;
2858 unlink(newpath);
2859 break;
2861 if (errno != EEXIST)
2862 break;
2863 ++nr;
2866 die("unable to write file %s mode %o", path, mode);
2869 static void create_file(struct patch *patch)
2871 char *path = patch->new_name;
2872 unsigned mode = patch->new_mode;
2873 unsigned long size = patch->resultsize;
2874 char *buf = patch->result;
2876 if (!mode)
2877 mode = S_IFREG | 0644;
2878 create_one_file(path, mode, buf, size);
2879 add_index_file(path, mode, buf, size);
2882 /* phase zero is to remove, phase one is to create */
2883 static void write_out_one_result(struct patch *patch, int phase)
2885 if (patch->is_delete > 0) {
2886 if (phase == 0)
2887 remove_file(patch, 1);
2888 return;
2890 if (patch->is_new > 0 || patch->is_copy) {
2891 if (phase == 1)
2892 create_file(patch);
2893 return;
2896 * Rename or modification boils down to the same
2897 * thing: remove the old, write the new
2899 if (phase == 0)
2900 remove_file(patch, patch->is_rename);
2901 if (phase == 1)
2902 create_file(patch);
2905 static int write_out_one_reject(struct patch *patch)
2907 FILE *rej;
2908 char namebuf[PATH_MAX];
2909 struct fragment *frag;
2910 int cnt = 0;
2912 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2913 if (!frag->rejected)
2914 continue;
2915 cnt++;
2918 if (!cnt) {
2919 if (apply_verbosely)
2920 say_patch_name(stderr,
2921 "Applied patch ", patch, " cleanly.\n");
2922 return 0;
2925 /* This should not happen, because a removal patch that leaves
2926 * contents are marked "rejected" at the patch level.
2928 if (!patch->new_name)
2929 die("internal error");
2931 /* Say this even without --verbose */
2932 say_patch_name(stderr, "Applying patch ", patch, " with");
2933 fprintf(stderr, " %d rejects...\n", cnt);
2935 cnt = strlen(patch->new_name);
2936 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2937 cnt = ARRAY_SIZE(namebuf) - 5;
2938 fprintf(stderr,
2939 "warning: truncating .rej filename to %.*s.rej",
2940 cnt - 1, patch->new_name);
2942 memcpy(namebuf, patch->new_name, cnt);
2943 memcpy(namebuf + cnt, ".rej", 5);
2945 rej = fopen(namebuf, "w");
2946 if (!rej)
2947 return error("cannot open %s: %s", namebuf, strerror(errno));
2949 /* Normal git tools never deal with .rej, so do not pretend
2950 * this is a git patch by saying --git nor give extended
2951 * headers. While at it, maybe please "kompare" that wants
2952 * the trailing TAB and some garbage at the end of line ;-).
2954 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2955 patch->new_name, patch->new_name);
2956 for (cnt = 1, frag = patch->fragments;
2957 frag;
2958 cnt++, frag = frag->next) {
2959 if (!frag->rejected) {
2960 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2961 continue;
2963 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2964 fprintf(rej, "%.*s", frag->size, frag->patch);
2965 if (frag->patch[frag->size-1] != '\n')
2966 fputc('\n', rej);
2968 fclose(rej);
2969 return -1;
2972 static int write_out_results(struct patch *list, int skipped_patch)
2974 int phase;
2975 int errs = 0;
2976 struct patch *l;
2978 if (!list && !skipped_patch)
2979 return error("No changes");
2981 for (phase = 0; phase < 2; phase++) {
2982 l = list;
2983 while (l) {
2984 if (l->rejected)
2985 errs = 1;
2986 else {
2987 write_out_one_result(l, phase);
2988 if (phase == 1 && write_out_one_reject(l))
2989 errs = 1;
2991 l = l->next;
2994 return errs;
2997 static struct lock_file lock_file;
2999 static struct excludes {
3000 struct excludes *next;
3001 const char *path;
3002 } *excludes;
3004 static int use_patch(struct patch *p)
3006 const char *pathname = p->new_name ? p->new_name : p->old_name;
3007 struct excludes *x = excludes;
3008 while (x) {
3009 if (fnmatch(x->path, pathname, 0) == 0)
3010 return 0;
3011 x = x->next;
3013 if (0 < prefix_length) {
3014 int pathlen = strlen(pathname);
3015 if (pathlen <= prefix_length ||
3016 memcmp(prefix, pathname, prefix_length))
3017 return 0;
3019 return 1;
3022 static void prefix_one(char **name)
3024 char *old_name = *name;
3025 if (!old_name)
3026 return;
3027 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3028 free(old_name);
3031 static void prefix_patches(struct patch *p)
3033 if (!prefix || p->is_toplevel_relative)
3034 return;
3035 for ( ; p; p = p->next) {
3036 if (p->new_name == p->old_name) {
3037 char *prefixed = p->new_name;
3038 prefix_one(&prefixed);
3039 p->new_name = p->old_name = prefixed;
3041 else {
3042 prefix_one(&p->new_name);
3043 prefix_one(&p->old_name);
3048 #define INACCURATE_EOF (1<<0)
3049 #define RECOUNT (1<<1)
3051 static int apply_patch(int fd, const char *filename, int options)
3053 size_t offset;
3054 struct strbuf buf;
3055 struct patch *list = NULL, **listp = &list;
3056 int skipped_patch = 0;
3058 /* FIXME - memory leak when using multiple patch files as inputs */
3059 memset(&fn_table, 0, sizeof(struct string_list));
3060 strbuf_init(&buf, 0);
3061 patch_input_file = filename;
3062 read_patch_file(&buf, fd);
3063 offset = 0;
3064 while (offset < buf.len) {
3065 struct patch *patch;
3066 int nr;
3068 patch = xcalloc(1, sizeof(*patch));
3069 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3070 patch->recount = !!(options & RECOUNT);
3071 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3072 if (nr < 0)
3073 break;
3074 if (apply_in_reverse)
3075 reverse_patches(patch);
3076 if (prefix)
3077 prefix_patches(patch);
3078 if (use_patch(patch)) {
3079 patch_stats(patch);
3080 *listp = patch;
3081 listp = &patch->next;
3083 else {
3084 /* perhaps free it a bit better? */
3085 free(patch);
3086 skipped_patch++;
3088 offset += nr;
3091 if (whitespace_error && (ws_error_action == die_on_ws_error))
3092 apply = 0;
3094 update_index = check_index && apply;
3095 if (update_index && newfd < 0)
3096 newfd = hold_locked_index(&lock_file, 1);
3098 if (check_index) {
3099 if (read_cache() < 0)
3100 die("unable to read index file");
3103 if ((check || apply) &&
3104 check_patch_list(list) < 0 &&
3105 !apply_with_reject)
3106 exit(1);
3108 if (apply && write_out_results(list, skipped_patch))
3109 exit(1);
3111 if (fake_ancestor)
3112 build_fake_ancestor(list, fake_ancestor);
3114 if (diffstat)
3115 stat_patch_list(list);
3117 if (numstat)
3118 numstat_patch_list(list);
3120 if (summary)
3121 summary_patch_list(list);
3123 strbuf_release(&buf);
3124 return 0;
3127 static int git_apply_config(const char *var, const char *value, void *cb)
3129 if (!strcmp(var, "apply.whitespace"))
3130 return git_config_string(&apply_default_whitespace, var, value);
3131 return git_default_config(var, value, cb);
3135 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3137 int i;
3138 int read_stdin = 1;
3139 int options = 0;
3140 int errs = 0;
3141 int is_not_gitdir;
3143 const char *whitespace_option = NULL;
3145 prefix = setup_git_directory_gently(&is_not_gitdir);
3146 prefix_length = prefix ? strlen(prefix) : 0;
3147 git_config(git_apply_config, NULL);
3148 if (apply_default_whitespace)
3149 parse_whitespace_option(apply_default_whitespace);
3151 for (i = 1; i < argc; i++) {
3152 const char *arg = argv[i];
3153 char *end;
3154 int fd;
3156 if (!strcmp(arg, "-")) {
3157 errs |= apply_patch(0, "<stdin>", options);
3158 read_stdin = 0;
3159 continue;
3161 if (!prefixcmp(arg, "--exclude=")) {
3162 struct excludes *x = xmalloc(sizeof(*x));
3163 x->path = arg + 10;
3164 x->next = excludes;
3165 excludes = x;
3166 continue;
3168 if (!prefixcmp(arg, "-p")) {
3169 p_value = atoi(arg + 2);
3170 p_value_known = 1;
3171 continue;
3173 if (!strcmp(arg, "--no-add")) {
3174 no_add = 1;
3175 continue;
3177 if (!strcmp(arg, "--stat")) {
3178 apply = 0;
3179 diffstat = 1;
3180 continue;
3182 if (!strcmp(arg, "--allow-binary-replacement") ||
3183 !strcmp(arg, "--binary")) {
3184 continue; /* now no-op */
3186 if (!strcmp(arg, "--numstat")) {
3187 apply = 0;
3188 numstat = 1;
3189 continue;
3191 if (!strcmp(arg, "--summary")) {
3192 apply = 0;
3193 summary = 1;
3194 continue;
3196 if (!strcmp(arg, "--check")) {
3197 apply = 0;
3198 check = 1;
3199 continue;
3201 if (!strcmp(arg, "--index")) {
3202 if (is_not_gitdir)
3203 die("--index outside a repository");
3204 check_index = 1;
3205 continue;
3207 if (!strcmp(arg, "--cached")) {
3208 if (is_not_gitdir)
3209 die("--cached outside a repository");
3210 check_index = 1;
3211 cached = 1;
3212 continue;
3214 if (!strcmp(arg, "--apply")) {
3215 apply = 1;
3216 continue;
3218 if (!strcmp(arg, "--build-fake-ancestor")) {
3219 apply = 0;
3220 if (++i >= argc)
3221 die ("need a filename");
3222 fake_ancestor = argv[i];
3223 continue;
3225 if (!strcmp(arg, "-z")) {
3226 line_termination = 0;
3227 continue;
3229 if (!prefixcmp(arg, "-C")) {
3230 p_context = strtoul(arg + 2, &end, 0);
3231 if (*end != '\0')
3232 die("unrecognized context count '%s'", arg + 2);
3233 continue;
3235 if (!prefixcmp(arg, "--whitespace=")) {
3236 whitespace_option = arg + 13;
3237 parse_whitespace_option(arg + 13);
3238 continue;
3240 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3241 apply_in_reverse = 1;
3242 continue;
3244 if (!strcmp(arg, "--unidiff-zero")) {
3245 unidiff_zero = 1;
3246 continue;
3248 if (!strcmp(arg, "--reject")) {
3249 apply = apply_with_reject = apply_verbosely = 1;
3250 continue;
3252 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3253 apply_verbosely = 1;
3254 continue;
3256 if (!strcmp(arg, "--inaccurate-eof")) {
3257 options |= INACCURATE_EOF;
3258 continue;
3260 if (!strcmp(arg, "--recount")) {
3261 options |= RECOUNT;
3262 continue;
3264 if (!prefixcmp(arg, "--directory=")) {
3265 arg += strlen("--directory=");
3266 root_len = strlen(arg);
3267 if (root_len && arg[root_len - 1] != '/') {
3268 char *new_root;
3269 root = new_root = xmalloc(root_len + 2);
3270 strcpy(new_root, arg);
3271 strcpy(new_root + root_len++, "/");
3272 } else
3273 root = arg;
3274 continue;
3276 if (0 < prefix_length)
3277 arg = prefix_filename(prefix, prefix_length, arg);
3279 fd = open(arg, O_RDONLY);
3280 if (fd < 0)
3281 die("can't open patch '%s': %s", arg, strerror(errno));
3282 read_stdin = 0;
3283 set_default_whitespace_mode(whitespace_option);
3284 errs |= apply_patch(fd, arg, options);
3285 close(fd);
3287 set_default_whitespace_mode(whitespace_option);
3288 if (read_stdin)
3289 errs |= apply_patch(0, "<stdin>", options);
3290 if (whitespace_error) {
3291 if (squelch_whitespace_errors &&
3292 squelch_whitespace_errors < whitespace_error) {
3293 int squelched =
3294 whitespace_error - squelch_whitespace_errors;
3295 fprintf(stderr, "warning: squelched %d "
3296 "whitespace error%s\n",
3297 squelched,
3298 squelched == 1 ? "" : "s");
3300 if (ws_error_action == die_on_ws_error)
3301 die("%d line%s add%s whitespace errors.",
3302 whitespace_error,
3303 whitespace_error == 1 ? "" : "s",
3304 whitespace_error == 1 ? "s" : "");
3305 if (applied_after_fixing_ws && apply)
3306 fprintf(stderr, "warning: %d line%s applied after"
3307 " fixing whitespace errors.\n",
3308 applied_after_fixing_ws,
3309 applied_after_fixing_ws == 1 ? "" : "s");
3310 else if (whitespace_error)
3311 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3312 whitespace_error,
3313 whitespace_error == 1 ? "" : "s",
3314 whitespace_error == 1 ? "s" : "");
3317 if (update_index) {
3318 if (write_cache(newfd, active_cache, active_nr) ||
3319 commit_locked_index(&lock_file))
3320 die("Unable to write new index file");
3323 return !!errs;