git-apply --directory: make --root more similar to GNU diff
[git/gitweb.git] / builtin-apply.c
blobc242bbd83bd0f154ccf5d4a8be7f85386216cebc
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;
60 static const char *root;
61 static int root_len;
63 static void parse_whitespace_option(const char *option)
65 if (!option) {
66 ws_error_action = warn_on_ws_error;
67 return;
69 if (!strcmp(option, "warn")) {
70 ws_error_action = warn_on_ws_error;
71 return;
73 if (!strcmp(option, "nowarn")) {
74 ws_error_action = nowarn_ws_error;
75 return;
77 if (!strcmp(option, "error")) {
78 ws_error_action = die_on_ws_error;
79 return;
81 if (!strcmp(option, "error-all")) {
82 ws_error_action = die_on_ws_error;
83 squelch_whitespace_errors = 0;
84 return;
86 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
87 ws_error_action = correct_ws_error;
88 return;
90 die("unrecognized whitespace option '%s'", option);
93 static void set_default_whitespace_mode(const char *whitespace_option)
95 if (!whitespace_option && !apply_default_whitespace)
96 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
100 * For "diff-stat" like behaviour, we keep track of the biggest change
101 * we've seen, and the longest filename. That allows us to do simple
102 * scaling.
104 static int max_change, max_len;
107 * Various "current state", notably line numbers and what
108 * file (and how) we're patching right now.. The "is_xxxx"
109 * things are flags, where -1 means "don't know yet".
111 static int linenr = 1;
114 * This represents one "hunk" from a patch, starting with
115 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
116 * patch text is pointed at by patch, and its byte length
117 * is stored in size. leading and trailing are the number
118 * of context lines.
120 struct fragment {
121 unsigned long leading, trailing;
122 unsigned long oldpos, oldlines;
123 unsigned long newpos, newlines;
124 const char *patch;
125 int size;
126 int rejected;
127 struct fragment *next;
131 * When dealing with a binary patch, we reuse "leading" field
132 * to store the type of the binary hunk, either deflated "delta"
133 * or deflated "literal".
135 #define binary_patch_method leading
136 #define BINARY_DELTA_DEFLATED 1
137 #define BINARY_LITERAL_DEFLATED 2
140 * This represents a "patch" to a file, both metainfo changes
141 * such as creation/deletion, filemode and content changes represented
142 * as a series of fragments.
144 struct patch {
145 char *new_name, *old_name, *def_name;
146 unsigned int old_mode, new_mode;
147 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
148 int rejected;
149 unsigned ws_rule;
150 unsigned long deflate_origlen;
151 int lines_added, lines_deleted;
152 int score;
153 unsigned int is_toplevel_relative:1;
154 unsigned int inaccurate_eof:1;
155 unsigned int is_binary:1;
156 unsigned int is_copy:1;
157 unsigned int is_rename:1;
158 struct fragment *fragments;
159 char *result;
160 size_t resultsize;
161 char old_sha1_prefix[41];
162 char new_sha1_prefix[41];
163 struct patch *next;
167 * A line in a file, len-bytes long (includes the terminating LF,
168 * except for an incomplete line at the end if the file ends with
169 * one), and its contents hashes to 'hash'.
171 struct line {
172 size_t len;
173 unsigned hash : 24;
174 unsigned flag : 8;
175 #define LINE_COMMON 1
179 * This represents a "file", which is an array of "lines".
181 struct image {
182 char *buf;
183 size_t len;
184 size_t nr;
185 size_t alloc;
186 struct line *line_allocated;
187 struct line *line;
190 static uint32_t hash_line(const char *cp, size_t len)
192 size_t i;
193 uint32_t h;
194 for (i = 0, h = 0; i < len; i++) {
195 if (!isspace(cp[i])) {
196 h = h * 3 + (cp[i] & 0xff);
199 return h;
202 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
204 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
205 img->line_allocated[img->nr].len = len;
206 img->line_allocated[img->nr].hash = hash_line(bol, len);
207 img->line_allocated[img->nr].flag = flag;
208 img->nr++;
211 static void prepare_image(struct image *image, char *buf, size_t len,
212 int prepare_linetable)
214 const char *cp, *ep;
216 memset(image, 0, sizeof(*image));
217 image->buf = buf;
218 image->len = len;
220 if (!prepare_linetable)
221 return;
223 ep = image->buf + image->len;
224 cp = image->buf;
225 while (cp < ep) {
226 const char *next;
227 for (next = cp; next < ep && *next != '\n'; next++)
229 if (next < ep)
230 next++;
231 add_line_info(image, cp, next - cp, 0);
232 cp = next;
234 image->line = image->line_allocated;
237 static void clear_image(struct image *image)
239 free(image->buf);
240 image->buf = NULL;
241 image->len = 0;
244 static void say_patch_name(FILE *output, const char *pre,
245 struct patch *patch, const char *post)
247 fputs(pre, output);
248 if (patch->old_name && patch->new_name &&
249 strcmp(patch->old_name, patch->new_name)) {
250 quote_c_style(patch->old_name, NULL, output, 0);
251 fputs(" => ", output);
252 quote_c_style(patch->new_name, NULL, output, 0);
253 } else {
254 const char *n = patch->new_name;
255 if (!n)
256 n = patch->old_name;
257 quote_c_style(n, NULL, output, 0);
259 fputs(post, output);
262 #define CHUNKSIZE (8192)
263 #define SLOP (16)
265 static void read_patch_file(struct strbuf *sb, int fd)
267 if (strbuf_read(sb, fd, 0) < 0)
268 die("git-apply: read returned %s", strerror(errno));
271 * Make sure that we have some slop in the buffer
272 * so that we can do speculative "memcmp" etc, and
273 * see to it that it is NUL-filled.
275 strbuf_grow(sb, SLOP);
276 memset(sb->buf + sb->len, 0, SLOP);
279 static unsigned long linelen(const char *buffer, unsigned long size)
281 unsigned long len = 0;
282 while (size--) {
283 len++;
284 if (*buffer++ == '\n')
285 break;
287 return len;
290 static int is_dev_null(const char *str)
292 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
295 #define TERM_SPACE 1
296 #define TERM_TAB 2
298 static int name_terminate(const char *name, int namelen, int c, int terminate)
300 if (c == ' ' && !(terminate & TERM_SPACE))
301 return 0;
302 if (c == '\t' && !(terminate & TERM_TAB))
303 return 0;
305 return 1;
308 static char *find_name(const char *line, char *def, int p_value, int terminate)
310 int len;
311 const char *start = line;
313 if (*line == '"') {
314 struct strbuf name;
317 * Proposed "new-style" GNU patch/diff format; see
318 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
320 strbuf_init(&name, 0);
321 if (!unquote_c_style(&name, line, NULL)) {
322 char *cp;
324 for (cp = name.buf; p_value; p_value--) {
325 cp = strchr(cp, '/');
326 if (!cp)
327 break;
328 cp++;
330 if (cp) {
331 /* name can later be freed, so we need
332 * to memmove, not just return cp
334 strbuf_remove(&name, 0, cp - name.buf);
335 free(def);
336 if (root)
337 strbuf_insert(&name, 0, root, root_len);
338 return strbuf_detach(&name, NULL);
341 strbuf_release(&name);
344 for (;;) {
345 char c = *line;
347 if (isspace(c)) {
348 if (c == '\n')
349 break;
350 if (name_terminate(start, line-start, c, terminate))
351 break;
353 line++;
354 if (c == '/' && !--p_value)
355 start = line;
357 if (!start)
358 return def;
359 len = line - start;
360 if (!len)
361 return def;
364 * Generally we prefer the shorter name, especially
365 * if the other one is just a variation of that with
366 * something else tacked on to the end (ie "file.orig"
367 * or "file~").
369 if (def) {
370 int deflen = strlen(def);
371 if (deflen < len && !strncmp(start, def, deflen))
372 return def;
373 free(def);
376 if (root) {
377 char *ret = xmalloc(root_len + len + 1);
378 strcpy(ret, root);
379 memcpy(ret + root_len, start, len);
380 ret[root_len + len] = '\0';
381 return ret;
384 return xmemdupz(start, len);
387 static int count_slashes(const char *cp)
389 int cnt = 0;
390 char ch;
392 while ((ch = *cp++))
393 if (ch == '/')
394 cnt++;
395 return cnt;
399 * Given the string after "--- " or "+++ ", guess the appropriate
400 * p_value for the given patch.
402 static int guess_p_value(const char *nameline)
404 char *name, *cp;
405 int val = -1;
407 if (is_dev_null(nameline))
408 return -1;
409 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
410 if (!name)
411 return -1;
412 cp = strchr(name, '/');
413 if (!cp)
414 val = 0;
415 else if (prefix) {
417 * Does it begin with "a/$our-prefix" and such? Then this is
418 * very likely to apply to our directory.
420 if (!strncmp(name, prefix, prefix_length))
421 val = count_slashes(prefix);
422 else {
423 cp++;
424 if (!strncmp(cp, prefix, prefix_length))
425 val = count_slashes(prefix) + 1;
428 free(name);
429 return val;
433 * Get the name etc info from the ---/+++ lines of a traditional patch header
435 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
436 * files, we can happily check the index for a match, but for creating a
437 * new file we should try to match whatever "patch" does. I have no idea.
439 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
441 char *name;
443 first += 4; /* skip "--- " */
444 second += 4; /* skip "+++ " */
445 if (!p_value_known) {
446 int p, q;
447 p = guess_p_value(first);
448 q = guess_p_value(second);
449 if (p < 0) p = q;
450 if (0 <= p && p == q) {
451 p_value = p;
452 p_value_known = 1;
455 if (is_dev_null(first)) {
456 patch->is_new = 1;
457 patch->is_delete = 0;
458 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
459 patch->new_name = name;
460 } else if (is_dev_null(second)) {
461 patch->is_new = 0;
462 patch->is_delete = 1;
463 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
464 patch->old_name = name;
465 } else {
466 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
467 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
468 patch->old_name = patch->new_name = name;
470 if (!name)
471 die("unable to find filename in patch at line %d", linenr);
474 static int gitdiff_hdrend(const char *line, struct patch *patch)
476 return -1;
480 * We're anal about diff header consistency, to make
481 * sure that we don't end up having strange ambiguous
482 * patches floating around.
484 * As a result, gitdiff_{old|new}name() will check
485 * their names against any previous information, just
486 * to make sure..
488 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
490 if (!orig_name && !isnull)
491 return find_name(line, NULL, p_value, TERM_TAB);
493 if (orig_name) {
494 int len;
495 const char *name;
496 char *another;
497 name = orig_name;
498 len = strlen(name);
499 if (isnull)
500 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
501 another = find_name(line, NULL, p_value, TERM_TAB);
502 if (!another || memcmp(another, name, len))
503 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
504 free(another);
505 return orig_name;
507 else {
508 /* expect "/dev/null" */
509 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
510 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
511 return NULL;
515 static int gitdiff_oldname(const char *line, struct patch *patch)
517 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
518 return 0;
521 static int gitdiff_newname(const char *line, struct patch *patch)
523 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
524 return 0;
527 static int gitdiff_oldmode(const char *line, struct patch *patch)
529 patch->old_mode = strtoul(line, NULL, 8);
530 return 0;
533 static int gitdiff_newmode(const char *line, struct patch *patch)
535 patch->new_mode = strtoul(line, NULL, 8);
536 return 0;
539 static int gitdiff_delete(const char *line, struct patch *patch)
541 patch->is_delete = 1;
542 patch->old_name = patch->def_name;
543 return gitdiff_oldmode(line, patch);
546 static int gitdiff_newfile(const char *line, struct patch *patch)
548 patch->is_new = 1;
549 patch->new_name = patch->def_name;
550 return gitdiff_newmode(line, patch);
553 static int gitdiff_copysrc(const char *line, struct patch *patch)
555 patch->is_copy = 1;
556 patch->old_name = find_name(line, NULL, 0, 0);
557 return 0;
560 static int gitdiff_copydst(const char *line, struct patch *patch)
562 patch->is_copy = 1;
563 patch->new_name = find_name(line, NULL, 0, 0);
564 return 0;
567 static int gitdiff_renamesrc(const char *line, struct patch *patch)
569 patch->is_rename = 1;
570 patch->old_name = find_name(line, NULL, 0, 0);
571 return 0;
574 static int gitdiff_renamedst(const char *line, struct patch *patch)
576 patch->is_rename = 1;
577 patch->new_name = find_name(line, NULL, 0, 0);
578 return 0;
581 static int gitdiff_similarity(const char *line, struct patch *patch)
583 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
584 patch->score = 0;
585 return 0;
588 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
590 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
591 patch->score = 0;
592 return 0;
595 static int gitdiff_index(const char *line, struct patch *patch)
598 * index line is N hexadecimal, "..", N hexadecimal,
599 * and optional space with octal mode.
601 const char *ptr, *eol;
602 int len;
604 ptr = strchr(line, '.');
605 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
606 return 0;
607 len = ptr - line;
608 memcpy(patch->old_sha1_prefix, line, len);
609 patch->old_sha1_prefix[len] = 0;
611 line = ptr + 2;
612 ptr = strchr(line, ' ');
613 eol = strchr(line, '\n');
615 if (!ptr || eol < ptr)
616 ptr = eol;
617 len = ptr - line;
619 if (40 < len)
620 return 0;
621 memcpy(patch->new_sha1_prefix, line, len);
622 patch->new_sha1_prefix[len] = 0;
623 if (*ptr == ' ')
624 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
625 return 0;
629 * This is normal for a diff that doesn't change anything: we'll fall through
630 * into the next diff. Tell the parser to break out.
632 static int gitdiff_unrecognized(const char *line, struct patch *patch)
634 return -1;
637 static const char *stop_at_slash(const char *line, int llen)
639 int i;
641 for (i = 0; i < llen; i++) {
642 int ch = line[i];
643 if (ch == '/')
644 return line + i;
646 return NULL;
650 * This is to extract the same name that appears on "diff --git"
651 * line. We do not find and return anything if it is a rename
652 * patch, and it is OK because we will find the name elsewhere.
653 * We need to reliably find name only when it is mode-change only,
654 * creation or deletion of an empty file. In any of these cases,
655 * both sides are the same name under a/ and b/ respectively.
657 static char *git_header_name(char *line, int llen)
659 const char *name;
660 const char *second = NULL;
661 size_t len;
663 line += strlen("diff --git ");
664 llen -= strlen("diff --git ");
666 if (*line == '"') {
667 const char *cp;
668 struct strbuf first;
669 struct strbuf sp;
671 strbuf_init(&first, 0);
672 strbuf_init(&sp, 0);
674 if (unquote_c_style(&first, line, &second))
675 goto free_and_fail1;
677 /* advance to the first slash */
678 cp = stop_at_slash(first.buf, first.len);
679 /* we do not accept absolute paths */
680 if (!cp || cp == first.buf)
681 goto free_and_fail1;
682 strbuf_remove(&first, 0, cp + 1 - first.buf);
685 * second points at one past closing dq of name.
686 * find the second name.
688 while ((second < line + llen) && isspace(*second))
689 second++;
691 if (line + llen <= second)
692 goto free_and_fail1;
693 if (*second == '"') {
694 if (unquote_c_style(&sp, second, NULL))
695 goto free_and_fail1;
696 cp = stop_at_slash(sp.buf, sp.len);
697 if (!cp || cp == sp.buf)
698 goto free_and_fail1;
699 /* They must match, otherwise ignore */
700 if (strcmp(cp + 1, first.buf))
701 goto free_and_fail1;
702 strbuf_release(&sp);
703 return strbuf_detach(&first, NULL);
706 /* unquoted second */
707 cp = stop_at_slash(second, line + llen - second);
708 if (!cp || cp == second)
709 goto free_and_fail1;
710 cp++;
711 if (line + llen - cp != first.len + 1 ||
712 memcmp(first.buf, cp, first.len))
713 goto free_and_fail1;
714 return strbuf_detach(&first, NULL);
716 free_and_fail1:
717 strbuf_release(&first);
718 strbuf_release(&sp);
719 return NULL;
722 /* unquoted first name */
723 name = stop_at_slash(line, llen);
724 if (!name || name == line)
725 return NULL;
726 name++;
729 * since the first name is unquoted, a dq if exists must be
730 * the beginning of the second name.
732 for (second = name; second < line + llen; second++) {
733 if (*second == '"') {
734 struct strbuf sp;
735 const char *np;
737 strbuf_init(&sp, 0);
738 if (unquote_c_style(&sp, second, NULL))
739 goto free_and_fail2;
741 np = stop_at_slash(sp.buf, sp.len);
742 if (!np || np == sp.buf)
743 goto free_and_fail2;
744 np++;
746 len = sp.buf + sp.len - np;
747 if (len < second - name &&
748 !strncmp(np, name, len) &&
749 isspace(name[len])) {
750 /* Good */
751 strbuf_remove(&sp, 0, np - sp.buf);
752 return strbuf_detach(&sp, NULL);
755 free_and_fail2:
756 strbuf_release(&sp);
757 return NULL;
762 * Accept a name only if it shows up twice, exactly the same
763 * form.
765 for (len = 0 ; ; len++) {
766 switch (name[len]) {
767 default:
768 continue;
769 case '\n':
770 return NULL;
771 case '\t': case ' ':
772 second = name+len;
773 for (;;) {
774 char c = *second++;
775 if (c == '\n')
776 return NULL;
777 if (c == '/')
778 break;
780 if (second[len] == '\n' && !memcmp(name, second, len)) {
781 return xmemdupz(name, len);
787 /* Verify that we recognize the lines following a git header */
788 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
790 unsigned long offset;
792 /* A git diff has explicit new/delete information, so we don't guess */
793 patch->is_new = 0;
794 patch->is_delete = 0;
797 * Some things may not have the old name in the
798 * rest of the headers anywhere (pure mode changes,
799 * or removing or adding empty files), so we get
800 * the default name from the header.
802 patch->def_name = git_header_name(line, len);
804 line += len;
805 size -= len;
806 linenr++;
807 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
808 static const struct opentry {
809 const char *str;
810 int (*fn)(const char *, struct patch *);
811 } optable[] = {
812 { "@@ -", gitdiff_hdrend },
813 { "--- ", gitdiff_oldname },
814 { "+++ ", gitdiff_newname },
815 { "old mode ", gitdiff_oldmode },
816 { "new mode ", gitdiff_newmode },
817 { "deleted file mode ", gitdiff_delete },
818 { "new file mode ", gitdiff_newfile },
819 { "copy from ", gitdiff_copysrc },
820 { "copy to ", gitdiff_copydst },
821 { "rename old ", gitdiff_renamesrc },
822 { "rename new ", gitdiff_renamedst },
823 { "rename from ", gitdiff_renamesrc },
824 { "rename to ", gitdiff_renamedst },
825 { "similarity index ", gitdiff_similarity },
826 { "dissimilarity index ", gitdiff_dissimilarity },
827 { "index ", gitdiff_index },
828 { "", gitdiff_unrecognized },
830 int i;
832 len = linelen(line, size);
833 if (!len || line[len-1] != '\n')
834 break;
835 for (i = 0; i < ARRAY_SIZE(optable); i++) {
836 const struct opentry *p = optable + i;
837 int oplen = strlen(p->str);
838 if (len < oplen || memcmp(p->str, line, oplen))
839 continue;
840 if (p->fn(line + oplen, patch) < 0)
841 return offset;
842 break;
846 return offset;
849 static int parse_num(const char *line, unsigned long *p)
851 char *ptr;
853 if (!isdigit(*line))
854 return 0;
855 *p = strtoul(line, &ptr, 10);
856 return ptr - line;
859 static int parse_range(const char *line, int len, int offset, const char *expect,
860 unsigned long *p1, unsigned long *p2)
862 int digits, ex;
864 if (offset < 0 || offset >= len)
865 return -1;
866 line += offset;
867 len -= offset;
869 digits = parse_num(line, p1);
870 if (!digits)
871 return -1;
873 offset += digits;
874 line += digits;
875 len -= digits;
877 *p2 = 1;
878 if (*line == ',') {
879 digits = parse_num(line+1, p2);
880 if (!digits)
881 return -1;
883 offset += digits+1;
884 line += digits+1;
885 len -= digits+1;
888 ex = strlen(expect);
889 if (ex > len)
890 return -1;
891 if (memcmp(line, expect, ex))
892 return -1;
894 return offset + ex;
898 * Parse a unified diff fragment header of the
899 * form "@@ -a,b +c,d @@"
901 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
903 int offset;
905 if (!len || line[len-1] != '\n')
906 return -1;
908 /* Figure out the number of lines in a fragment */
909 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
910 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
912 return offset;
915 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
917 unsigned long offset, len;
919 patch->is_toplevel_relative = 0;
920 patch->is_rename = patch->is_copy = 0;
921 patch->is_new = patch->is_delete = -1;
922 patch->old_mode = patch->new_mode = 0;
923 patch->old_name = patch->new_name = NULL;
924 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
925 unsigned long nextlen;
927 len = linelen(line, size);
928 if (!len)
929 break;
931 /* Testing this early allows us to take a few shortcuts.. */
932 if (len < 6)
933 continue;
936 * Make sure we don't find any unconnected patch fragments.
937 * That's a sign that we didn't find a header, and that a
938 * patch has become corrupted/broken up.
940 if (!memcmp("@@ -", line, 4)) {
941 struct fragment dummy;
942 if (parse_fragment_header(line, len, &dummy) < 0)
943 continue;
944 die("patch fragment without header at line %d: %.*s",
945 linenr, (int)len-1, line);
948 if (size < len + 6)
949 break;
952 * Git patch? It might not have a real patch, just a rename
953 * or mode change, so we handle that specially
955 if (!memcmp("diff --git ", line, 11)) {
956 int git_hdr_len = parse_git_header(line, len, size, patch);
957 if (git_hdr_len <= len)
958 continue;
959 if (!patch->old_name && !patch->new_name) {
960 if (!patch->def_name)
961 die("git diff header lacks filename information (line %d)", linenr);
962 patch->old_name = patch->new_name = patch->def_name;
964 patch->is_toplevel_relative = 1;
965 *hdrsize = git_hdr_len;
966 return offset;
969 /* --- followed by +++ ? */
970 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
971 continue;
974 * We only accept unified patches, so we want it to
975 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
976 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
978 nextlen = linelen(line + len, size - len);
979 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
980 continue;
982 /* Ok, we'll consider it a patch */
983 parse_traditional_patch(line, line+len, patch);
984 *hdrsize = len + nextlen;
985 linenr += 2;
986 return offset;
988 return -1;
991 static void check_whitespace(const char *line, int len, unsigned ws_rule)
993 char *err;
994 unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
995 NULL, NULL, NULL, NULL);
996 if (!result)
997 return;
999 whitespace_error++;
1000 if (squelch_whitespace_errors &&
1001 squelch_whitespace_errors < whitespace_error)
1003 else {
1004 err = whitespace_error_string(result);
1005 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1006 patch_input_file, linenr, err, len - 2, line + 1);
1007 free(err);
1012 * Parse a unified diff. Note that this really needs to parse each
1013 * fragment separately, since the only way to know the difference
1014 * between a "---" that is part of a patch, and a "---" that starts
1015 * the next patch is to look at the line counts..
1017 static int parse_fragment(char *line, unsigned long size,
1018 struct patch *patch, struct fragment *fragment)
1020 int added, deleted;
1021 int len = linelen(line, size), offset;
1022 unsigned long oldlines, newlines;
1023 unsigned long leading, trailing;
1025 offset = parse_fragment_header(line, len, fragment);
1026 if (offset < 0)
1027 return -1;
1028 oldlines = fragment->oldlines;
1029 newlines = fragment->newlines;
1030 leading = 0;
1031 trailing = 0;
1033 /* Parse the thing.. */
1034 line += len;
1035 size -= len;
1036 linenr++;
1037 added = deleted = 0;
1038 for (offset = len;
1039 0 < size;
1040 offset += len, size -= len, line += len, linenr++) {
1041 if (!oldlines && !newlines)
1042 break;
1043 len = linelen(line, size);
1044 if (!len || line[len-1] != '\n')
1045 return -1;
1046 switch (*line) {
1047 default:
1048 return -1;
1049 case '\n': /* newer GNU diff, an empty context line */
1050 case ' ':
1051 oldlines--;
1052 newlines--;
1053 if (!deleted && !added)
1054 leading++;
1055 trailing++;
1056 break;
1057 case '-':
1058 if (apply_in_reverse &&
1059 ws_error_action != nowarn_ws_error)
1060 check_whitespace(line, len, patch->ws_rule);
1061 deleted++;
1062 oldlines--;
1063 trailing = 0;
1064 break;
1065 case '+':
1066 if (!apply_in_reverse &&
1067 ws_error_action != nowarn_ws_error)
1068 check_whitespace(line, len, patch->ws_rule);
1069 added++;
1070 newlines--;
1071 trailing = 0;
1072 break;
1075 * We allow "\ No newline at end of file". Depending
1076 * on locale settings when the patch was produced we
1077 * don't know what this line looks like. The only
1078 * thing we do know is that it begins with "\ ".
1079 * Checking for 12 is just for sanity check -- any
1080 * l10n of "\ No newline..." is at least that long.
1082 case '\\':
1083 if (len < 12 || memcmp(line, "\\ ", 2))
1084 return -1;
1085 break;
1088 if (oldlines || newlines)
1089 return -1;
1090 fragment->leading = leading;
1091 fragment->trailing = trailing;
1094 * If a fragment ends with an incomplete line, we failed to include
1095 * it in the above loop because we hit oldlines == newlines == 0
1096 * before seeing it.
1098 if (12 < size && !memcmp(line, "\\ ", 2))
1099 offset += linelen(line, size);
1101 patch->lines_added += added;
1102 patch->lines_deleted += deleted;
1104 if (0 < patch->is_new && oldlines)
1105 return error("new file depends on old contents");
1106 if (0 < patch->is_delete && newlines)
1107 return error("deleted file still has contents");
1108 return offset;
1111 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1113 unsigned long offset = 0;
1114 unsigned long oldlines = 0, newlines = 0, context = 0;
1115 struct fragment **fragp = &patch->fragments;
1117 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1118 struct fragment *fragment;
1119 int len;
1121 fragment = xcalloc(1, sizeof(*fragment));
1122 len = parse_fragment(line, size, patch, fragment);
1123 if (len <= 0)
1124 die("corrupt patch at line %d", linenr);
1125 fragment->patch = line;
1126 fragment->size = len;
1127 oldlines += fragment->oldlines;
1128 newlines += fragment->newlines;
1129 context += fragment->leading + fragment->trailing;
1131 *fragp = fragment;
1132 fragp = &fragment->next;
1134 offset += len;
1135 line += len;
1136 size -= len;
1140 * If something was removed (i.e. we have old-lines) it cannot
1141 * be creation, and if something was added it cannot be
1142 * deletion. However, the reverse is not true; --unified=0
1143 * patches that only add are not necessarily creation even
1144 * though they do not have any old lines, and ones that only
1145 * delete are not necessarily deletion.
1147 * Unfortunately, a real creation/deletion patch do _not_ have
1148 * any context line by definition, so we cannot safely tell it
1149 * apart with --unified=0 insanity. At least if the patch has
1150 * more than one hunk it is not creation or deletion.
1152 if (patch->is_new < 0 &&
1153 (oldlines || (patch->fragments && patch->fragments->next)))
1154 patch->is_new = 0;
1155 if (patch->is_delete < 0 &&
1156 (newlines || (patch->fragments && patch->fragments->next)))
1157 patch->is_delete = 0;
1159 if (0 < patch->is_new && oldlines)
1160 die("new file %s depends on old contents", patch->new_name);
1161 if (0 < patch->is_delete && newlines)
1162 die("deleted file %s still has contents", patch->old_name);
1163 if (!patch->is_delete && !newlines && context)
1164 fprintf(stderr, "** warning: file %s becomes empty but "
1165 "is not deleted\n", patch->new_name);
1167 return offset;
1170 static inline int metadata_changes(struct patch *patch)
1172 return patch->is_rename > 0 ||
1173 patch->is_copy > 0 ||
1174 patch->is_new > 0 ||
1175 patch->is_delete ||
1176 (patch->old_mode && patch->new_mode &&
1177 patch->old_mode != patch->new_mode);
1180 static char *inflate_it(const void *data, unsigned long size,
1181 unsigned long inflated_size)
1183 z_stream stream;
1184 void *out;
1185 int st;
1187 memset(&stream, 0, sizeof(stream));
1189 stream.next_in = (unsigned char *)data;
1190 stream.avail_in = size;
1191 stream.next_out = out = xmalloc(inflated_size);
1192 stream.avail_out = inflated_size;
1193 inflateInit(&stream);
1194 st = inflate(&stream, Z_FINISH);
1195 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1196 free(out);
1197 return NULL;
1199 return out;
1202 static struct fragment *parse_binary_hunk(char **buf_p,
1203 unsigned long *sz_p,
1204 int *status_p,
1205 int *used_p)
1208 * Expect a line that begins with binary patch method ("literal"
1209 * or "delta"), followed by the length of data before deflating.
1210 * a sequence of 'length-byte' followed by base-85 encoded data
1211 * should follow, terminated by a newline.
1213 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1214 * and we would limit the patch line to 66 characters,
1215 * so one line can fit up to 13 groups that would decode
1216 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1217 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1219 int llen, used;
1220 unsigned long size = *sz_p;
1221 char *buffer = *buf_p;
1222 int patch_method;
1223 unsigned long origlen;
1224 char *data = NULL;
1225 int hunk_size = 0;
1226 struct fragment *frag;
1228 llen = linelen(buffer, size);
1229 used = llen;
1231 *status_p = 0;
1233 if (!prefixcmp(buffer, "delta ")) {
1234 patch_method = BINARY_DELTA_DEFLATED;
1235 origlen = strtoul(buffer + 6, NULL, 10);
1237 else if (!prefixcmp(buffer, "literal ")) {
1238 patch_method = BINARY_LITERAL_DEFLATED;
1239 origlen = strtoul(buffer + 8, NULL, 10);
1241 else
1242 return NULL;
1244 linenr++;
1245 buffer += llen;
1246 while (1) {
1247 int byte_length, max_byte_length, newsize;
1248 llen = linelen(buffer, size);
1249 used += llen;
1250 linenr++;
1251 if (llen == 1) {
1252 /* consume the blank line */
1253 buffer++;
1254 size--;
1255 break;
1258 * Minimum line is "A00000\n" which is 7-byte long,
1259 * and the line length must be multiple of 5 plus 2.
1261 if ((llen < 7) || (llen-2) % 5)
1262 goto corrupt;
1263 max_byte_length = (llen - 2) / 5 * 4;
1264 byte_length = *buffer;
1265 if ('A' <= byte_length && byte_length <= 'Z')
1266 byte_length = byte_length - 'A' + 1;
1267 else if ('a' <= byte_length && byte_length <= 'z')
1268 byte_length = byte_length - 'a' + 27;
1269 else
1270 goto corrupt;
1271 /* if the input length was not multiple of 4, we would
1272 * have filler at the end but the filler should never
1273 * exceed 3 bytes
1275 if (max_byte_length < byte_length ||
1276 byte_length <= max_byte_length - 4)
1277 goto corrupt;
1278 newsize = hunk_size + byte_length;
1279 data = xrealloc(data, newsize);
1280 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1281 goto corrupt;
1282 hunk_size = newsize;
1283 buffer += llen;
1284 size -= llen;
1287 frag = xcalloc(1, sizeof(*frag));
1288 frag->patch = inflate_it(data, hunk_size, origlen);
1289 if (!frag->patch)
1290 goto corrupt;
1291 free(data);
1292 frag->size = origlen;
1293 *buf_p = buffer;
1294 *sz_p = size;
1295 *used_p = used;
1296 frag->binary_patch_method = patch_method;
1297 return frag;
1299 corrupt:
1300 free(data);
1301 *status_p = -1;
1302 error("corrupt binary patch at line %d: %.*s",
1303 linenr-1, llen-1, buffer);
1304 return NULL;
1307 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1310 * We have read "GIT binary patch\n"; what follows is a line
1311 * that says the patch method (currently, either "literal" or
1312 * "delta") and the length of data before deflating; a
1313 * sequence of 'length-byte' followed by base-85 encoded data
1314 * follows.
1316 * When a binary patch is reversible, there is another binary
1317 * hunk in the same format, starting with patch method (either
1318 * "literal" or "delta") with the length of data, and a sequence
1319 * of length-byte + base-85 encoded data, terminated with another
1320 * empty line. This data, when applied to the postimage, produces
1321 * the preimage.
1323 struct fragment *forward;
1324 struct fragment *reverse;
1325 int status;
1326 int used, used_1;
1328 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1329 if (!forward && !status)
1330 /* there has to be one hunk (forward hunk) */
1331 return error("unrecognized binary patch at line %d", linenr-1);
1332 if (status)
1333 /* otherwise we already gave an error message */
1334 return status;
1336 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1337 if (reverse)
1338 used += used_1;
1339 else if (status) {
1341 * Not having reverse hunk is not an error, but having
1342 * a corrupt reverse hunk is.
1344 free((void*) forward->patch);
1345 free(forward);
1346 return status;
1348 forward->next = reverse;
1349 patch->fragments = forward;
1350 patch->is_binary = 1;
1351 return used;
1354 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1356 int hdrsize, patchsize;
1357 int offset = find_header(buffer, size, &hdrsize, patch);
1359 if (offset < 0)
1360 return offset;
1362 patch->ws_rule = whitespace_rule(patch->new_name
1363 ? patch->new_name
1364 : patch->old_name);
1366 patchsize = parse_single_patch(buffer + offset + hdrsize,
1367 size - offset - hdrsize, patch);
1369 if (!patchsize) {
1370 static const char *binhdr[] = {
1371 "Binary files ",
1372 "Files ",
1373 NULL,
1375 static const char git_binary[] = "GIT binary patch\n";
1376 int i;
1377 int hd = hdrsize + offset;
1378 unsigned long llen = linelen(buffer + hd, size - hd);
1380 if (llen == sizeof(git_binary) - 1 &&
1381 !memcmp(git_binary, buffer + hd, llen)) {
1382 int used;
1383 linenr++;
1384 used = parse_binary(buffer + hd + llen,
1385 size - hd - llen, patch);
1386 if (used)
1387 patchsize = used + llen;
1388 else
1389 patchsize = 0;
1391 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1392 for (i = 0; binhdr[i]; i++) {
1393 int len = strlen(binhdr[i]);
1394 if (len < size - hd &&
1395 !memcmp(binhdr[i], buffer + hd, len)) {
1396 linenr++;
1397 patch->is_binary = 1;
1398 patchsize = llen;
1399 break;
1404 /* Empty patch cannot be applied if it is a text patch
1405 * without metadata change. A binary patch appears
1406 * empty to us here.
1408 if ((apply || check) &&
1409 (!patch->is_binary && !metadata_changes(patch)))
1410 die("patch with only garbage at line %d", linenr);
1413 return offset + hdrsize + patchsize;
1416 #define swap(a,b) myswap((a),(b),sizeof(a))
1418 #define myswap(a, b, size) do { \
1419 unsigned char mytmp[size]; \
1420 memcpy(mytmp, &a, size); \
1421 memcpy(&a, &b, size); \
1422 memcpy(&b, mytmp, size); \
1423 } while (0)
1425 static void reverse_patches(struct patch *p)
1427 for (; p; p = p->next) {
1428 struct fragment *frag = p->fragments;
1430 swap(p->new_name, p->old_name);
1431 swap(p->new_mode, p->old_mode);
1432 swap(p->is_new, p->is_delete);
1433 swap(p->lines_added, p->lines_deleted);
1434 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1436 for (; frag; frag = frag->next) {
1437 swap(frag->newpos, frag->oldpos);
1438 swap(frag->newlines, frag->oldlines);
1443 static const char pluses[] =
1444 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1445 static const char minuses[]=
1446 "----------------------------------------------------------------------";
1448 static void show_stats(struct patch *patch)
1450 struct strbuf qname;
1451 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1452 int max, add, del;
1454 strbuf_init(&qname, 0);
1455 quote_c_style(cp, &qname, NULL, 0);
1458 * "scale" the filename
1460 max = max_len;
1461 if (max > 50)
1462 max = 50;
1464 if (qname.len > max) {
1465 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1466 if (!cp)
1467 cp = qname.buf + qname.len + 3 - max;
1468 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1471 if (patch->is_binary) {
1472 printf(" %-*s | Bin\n", max, qname.buf);
1473 strbuf_release(&qname);
1474 return;
1477 printf(" %-*s |", max, qname.buf);
1478 strbuf_release(&qname);
1481 * scale the add/delete
1483 max = max + max_change > 70 ? 70 - max : max_change;
1484 add = patch->lines_added;
1485 del = patch->lines_deleted;
1487 if (max_change > 0) {
1488 int total = ((add + del) * max + max_change / 2) / max_change;
1489 add = (add * max + max_change / 2) / max_change;
1490 del = total - add;
1492 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1493 add, pluses, del, minuses);
1496 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1498 switch (st->st_mode & S_IFMT) {
1499 case S_IFLNK:
1500 strbuf_grow(buf, st->st_size);
1501 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1502 return -1;
1503 strbuf_setlen(buf, st->st_size);
1504 return 0;
1505 case S_IFREG:
1506 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1507 return error("unable to open or read %s", path);
1508 convert_to_git(path, buf->buf, buf->len, buf, 0);
1509 return 0;
1510 default:
1511 return -1;
1515 static void update_pre_post_images(struct image *preimage,
1516 struct image *postimage,
1517 char *buf,
1518 size_t len)
1520 int i, ctx;
1521 char *new, *old, *fixed;
1522 struct image fixed_preimage;
1525 * Update the preimage with whitespace fixes. Note that we
1526 * are not losing preimage->buf -- apply_one_fragment() will
1527 * free "oldlines".
1529 prepare_image(&fixed_preimage, buf, len, 1);
1530 assert(fixed_preimage.nr == preimage->nr);
1531 for (i = 0; i < preimage->nr; i++)
1532 fixed_preimage.line[i].flag = preimage->line[i].flag;
1533 free(preimage->line_allocated);
1534 *preimage = fixed_preimage;
1537 * Adjust the common context lines in postimage, in place.
1538 * This is possible because whitespace fixing does not make
1539 * the string grow.
1541 new = old = postimage->buf;
1542 fixed = preimage->buf;
1543 for (i = ctx = 0; i < postimage->nr; i++) {
1544 size_t len = postimage->line[i].len;
1545 if (!(postimage->line[i].flag & LINE_COMMON)) {
1546 /* an added line -- no counterparts in preimage */
1547 memmove(new, old, len);
1548 old += len;
1549 new += len;
1550 continue;
1553 /* a common context -- skip it in the original postimage */
1554 old += len;
1556 /* and find the corresponding one in the fixed preimage */
1557 while (ctx < preimage->nr &&
1558 !(preimage->line[ctx].flag & LINE_COMMON)) {
1559 fixed += preimage->line[ctx].len;
1560 ctx++;
1562 if (preimage->nr <= ctx)
1563 die("oops");
1565 /* and copy it in, while fixing the line length */
1566 len = preimage->line[ctx].len;
1567 memcpy(new, fixed, len);
1568 new += len;
1569 fixed += len;
1570 postimage->line[i].len = len;
1571 ctx++;
1574 /* Fix the length of the whole thing */
1575 postimage->len = new - postimage->buf;
1578 static int match_fragment(struct image *img,
1579 struct image *preimage,
1580 struct image *postimage,
1581 unsigned long try,
1582 int try_lno,
1583 unsigned ws_rule,
1584 int match_beginning, int match_end)
1586 int i;
1587 char *fixed_buf, *buf, *orig, *target;
1589 if (preimage->nr + try_lno > img->nr)
1590 return 0;
1592 if (match_beginning && try_lno)
1593 return 0;
1595 if (match_end && preimage->nr + try_lno != img->nr)
1596 return 0;
1598 /* Quick hash check */
1599 for (i = 0; i < preimage->nr; i++)
1600 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1601 return 0;
1604 * Do we have an exact match? If we were told to match
1605 * at the end, size must be exactly at try+fragsize,
1606 * otherwise try+fragsize must be still within the preimage,
1607 * and either case, the old piece should match the preimage
1608 * exactly.
1610 if ((match_end
1611 ? (try + preimage->len == img->len)
1612 : (try + preimage->len <= img->len)) &&
1613 !memcmp(img->buf + try, preimage->buf, preimage->len))
1614 return 1;
1616 if (ws_error_action != correct_ws_error)
1617 return 0;
1620 * The hunk does not apply byte-by-byte, but the hash says
1621 * it might with whitespace fuzz.
1623 fixed_buf = xmalloc(preimage->len + 1);
1624 buf = fixed_buf;
1625 orig = preimage->buf;
1626 target = img->buf + try;
1627 for (i = 0; i < preimage->nr; i++) {
1628 size_t fixlen; /* length after fixing the preimage */
1629 size_t oldlen = preimage->line[i].len;
1630 size_t tgtlen = img->line[try_lno + i].len;
1631 size_t tgtfixlen; /* length after fixing the target line */
1632 char tgtfixbuf[1024], *tgtfix;
1633 int match;
1635 /* Try fixing the line in the preimage */
1636 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1638 /* Try fixing the line in the target */
1639 if (sizeof(tgtfixbuf) < tgtlen)
1640 tgtfix = tgtfixbuf;
1641 else
1642 tgtfix = xmalloc(tgtlen);
1643 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1646 * If they match, either the preimage was based on
1647 * a version before our tree fixed whitespace breakage,
1648 * or we are lacking a whitespace-fix patch the tree
1649 * the preimage was based on already had (i.e. target
1650 * has whitespace breakage, the preimage doesn't).
1651 * In either case, we are fixing the whitespace breakages
1652 * so we might as well take the fix together with their
1653 * real change.
1655 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1657 if (tgtfix != tgtfixbuf)
1658 free(tgtfix);
1659 if (!match)
1660 goto unmatch_exit;
1662 orig += oldlen;
1663 buf += fixlen;
1664 target += tgtlen;
1668 * Yes, the preimage is based on an older version that still
1669 * has whitespace breakages unfixed, and fixing them makes the
1670 * hunk match. Update the context lines in the postimage.
1672 update_pre_post_images(preimage, postimage,
1673 fixed_buf, buf - fixed_buf);
1674 return 1;
1676 unmatch_exit:
1677 free(fixed_buf);
1678 return 0;
1681 static int find_pos(struct image *img,
1682 struct image *preimage,
1683 struct image *postimage,
1684 int line,
1685 unsigned ws_rule,
1686 int match_beginning, int match_end)
1688 int i;
1689 unsigned long backwards, forwards, try;
1690 int backwards_lno, forwards_lno, try_lno;
1692 if (preimage->nr > img->nr)
1693 return -1;
1696 * If match_begining or match_end is specified, there is no
1697 * point starting from a wrong line that will never match and
1698 * wander around and wait for a match at the specified end.
1700 if (match_beginning)
1701 line = 0;
1702 else if (match_end)
1703 line = img->nr - preimage->nr;
1705 if (line > img->nr)
1706 line = img->nr;
1708 try = 0;
1709 for (i = 0; i < line; i++)
1710 try += img->line[i].len;
1713 * There's probably some smart way to do this, but I'll leave
1714 * that to the smart and beautiful people. I'm simple and stupid.
1716 backwards = try;
1717 backwards_lno = line;
1718 forwards = try;
1719 forwards_lno = line;
1720 try_lno = line;
1722 for (i = 0; ; i++) {
1723 if (match_fragment(img, preimage, postimage,
1724 try, try_lno, ws_rule,
1725 match_beginning, match_end))
1726 return try_lno;
1728 again:
1729 if (backwards_lno == 0 && forwards_lno == img->nr)
1730 break;
1732 if (i & 1) {
1733 if (backwards_lno == 0) {
1734 i++;
1735 goto again;
1737 backwards_lno--;
1738 backwards -= img->line[backwards_lno].len;
1739 try = backwards;
1740 try_lno = backwards_lno;
1741 } else {
1742 if (forwards_lno == img->nr) {
1743 i++;
1744 goto again;
1746 forwards += img->line[forwards_lno].len;
1747 forwards_lno++;
1748 try = forwards;
1749 try_lno = forwards_lno;
1753 return -1;
1756 static void remove_first_line(struct image *img)
1758 img->buf += img->line[0].len;
1759 img->len -= img->line[0].len;
1760 img->line++;
1761 img->nr--;
1764 static void remove_last_line(struct image *img)
1766 img->len -= img->line[--img->nr].len;
1769 static void update_image(struct image *img,
1770 int applied_pos,
1771 struct image *preimage,
1772 struct image *postimage)
1775 * remove the copy of preimage at offset in img
1776 * and replace it with postimage
1778 int i, nr;
1779 size_t remove_count, insert_count, applied_at = 0;
1780 char *result;
1782 for (i = 0; i < applied_pos; i++)
1783 applied_at += img->line[i].len;
1785 remove_count = 0;
1786 for (i = 0; i < preimage->nr; i++)
1787 remove_count += img->line[applied_pos + i].len;
1788 insert_count = postimage->len;
1790 /* Adjust the contents */
1791 result = xmalloc(img->len + insert_count - remove_count + 1);
1792 memcpy(result, img->buf, applied_at);
1793 memcpy(result + applied_at, postimage->buf, postimage->len);
1794 memcpy(result + applied_at + postimage->len,
1795 img->buf + (applied_at + remove_count),
1796 img->len - (applied_at + remove_count));
1797 free(img->buf);
1798 img->buf = result;
1799 img->len += insert_count - remove_count;
1800 result[img->len] = '\0';
1802 /* Adjust the line table */
1803 nr = img->nr + postimage->nr - preimage->nr;
1804 if (preimage->nr < postimage->nr) {
1806 * NOTE: this knows that we never call remove_first_line()
1807 * on anything other than pre/post image.
1809 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1810 img->line_allocated = img->line;
1812 if (preimage->nr != postimage->nr)
1813 memmove(img->line + applied_pos + postimage->nr,
1814 img->line + applied_pos + preimage->nr,
1815 (img->nr - (applied_pos + preimage->nr)) *
1816 sizeof(*img->line));
1817 memcpy(img->line + applied_pos,
1818 postimage->line,
1819 postimage->nr * sizeof(*img->line));
1820 img->nr = nr;
1823 static int apply_one_fragment(struct image *img, struct fragment *frag,
1824 int inaccurate_eof, unsigned ws_rule)
1826 int match_beginning, match_end;
1827 const char *patch = frag->patch;
1828 int size = frag->size;
1829 char *old, *new, *oldlines, *newlines;
1830 int new_blank_lines_at_end = 0;
1831 unsigned long leading, trailing;
1832 int pos, applied_pos;
1833 struct image preimage;
1834 struct image postimage;
1836 memset(&preimage, 0, sizeof(preimage));
1837 memset(&postimage, 0, sizeof(postimage));
1838 oldlines = xmalloc(size);
1839 newlines = xmalloc(size);
1841 old = oldlines;
1842 new = newlines;
1843 while (size > 0) {
1844 char first;
1845 int len = linelen(patch, size);
1846 int plen, added;
1847 int added_blank_line = 0;
1849 if (!len)
1850 break;
1853 * "plen" is how much of the line we should use for
1854 * the actual patch data. Normally we just remove the
1855 * first character on the line, but if the line is
1856 * followed by "\ No newline", then we also remove the
1857 * last one (which is the newline, of course).
1859 plen = len - 1;
1860 if (len < size && patch[len] == '\\')
1861 plen--;
1862 first = *patch;
1863 if (apply_in_reverse) {
1864 if (first == '-')
1865 first = '+';
1866 else if (first == '+')
1867 first = '-';
1870 switch (first) {
1871 case '\n':
1872 /* Newer GNU diff, empty context line */
1873 if (plen < 0)
1874 /* ... followed by '\No newline'; nothing */
1875 break;
1876 *old++ = '\n';
1877 *new++ = '\n';
1878 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1879 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1880 break;
1881 case ' ':
1882 case '-':
1883 memcpy(old, patch + 1, plen);
1884 add_line_info(&preimage, old, plen,
1885 (first == ' ' ? LINE_COMMON : 0));
1886 old += plen;
1887 if (first == '-')
1888 break;
1889 /* Fall-through for ' ' */
1890 case '+':
1891 /* --no-add does not add new lines */
1892 if (first == '+' && no_add)
1893 break;
1895 if (first != '+' ||
1896 !whitespace_error ||
1897 ws_error_action != correct_ws_error) {
1898 memcpy(new, patch + 1, plen);
1899 added = plen;
1901 else {
1902 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1904 add_line_info(&postimage, new, added,
1905 (first == '+' ? 0 : LINE_COMMON));
1906 new += added;
1907 if (first == '+' &&
1908 added == 1 && new[-1] == '\n')
1909 added_blank_line = 1;
1910 break;
1911 case '@': case '\\':
1912 /* Ignore it, we already handled it */
1913 break;
1914 default:
1915 if (apply_verbosely)
1916 error("invalid start of line: '%c'", first);
1917 return -1;
1919 if (added_blank_line)
1920 new_blank_lines_at_end++;
1921 else
1922 new_blank_lines_at_end = 0;
1923 patch += len;
1924 size -= len;
1926 if (inaccurate_eof &&
1927 old > oldlines && old[-1] == '\n' &&
1928 new > newlines && new[-1] == '\n') {
1929 old--;
1930 new--;
1933 leading = frag->leading;
1934 trailing = frag->trailing;
1937 * A hunk to change lines at the beginning would begin with
1938 * @@ -1,L +N,M @@
1940 * And a hunk to add to an empty file would begin with
1941 * @@ -0,0 +N,M @@
1943 * In other words, a hunk that is (frag->oldpos <= 1) with or
1944 * without leading context must match at the beginning.
1946 match_beginning = frag->oldpos <= 1;
1949 * A hunk without trailing lines must match at the end.
1950 * However, we simply cannot tell if a hunk must match end
1951 * from the lack of trailing lines if the patch was generated
1952 * with unidiff without any context.
1954 match_end = !unidiff_zero && !trailing;
1956 pos = frag->newpos ? (frag->newpos - 1) : 0;
1957 preimage.buf = oldlines;
1958 preimage.len = old - oldlines;
1959 postimage.buf = newlines;
1960 postimage.len = new - newlines;
1961 preimage.line = preimage.line_allocated;
1962 postimage.line = postimage.line_allocated;
1964 for (;;) {
1966 applied_pos = find_pos(img, &preimage, &postimage, pos,
1967 ws_rule, match_beginning, match_end);
1969 if (applied_pos >= 0)
1970 break;
1972 /* Am I at my context limits? */
1973 if ((leading <= p_context) && (trailing <= p_context))
1974 break;
1975 if (match_beginning || match_end) {
1976 match_beginning = match_end = 0;
1977 continue;
1981 * Reduce the number of context lines; reduce both
1982 * leading and trailing if they are equal otherwise
1983 * just reduce the larger context.
1985 if (leading >= trailing) {
1986 remove_first_line(&preimage);
1987 remove_first_line(&postimage);
1988 pos--;
1989 leading--;
1991 if (trailing > leading) {
1992 remove_last_line(&preimage);
1993 remove_last_line(&postimage);
1994 trailing--;
1998 if (applied_pos >= 0) {
1999 if (ws_error_action == correct_ws_error &&
2000 new_blank_lines_at_end &&
2001 postimage.nr + applied_pos == img->nr) {
2003 * If the patch application adds blank lines
2004 * at the end, and if the patch applies at the
2005 * end of the image, remove those added blank
2006 * lines.
2008 while (new_blank_lines_at_end--)
2009 remove_last_line(&postimage);
2013 * Warn if it was necessary to reduce the number
2014 * of context lines.
2016 if ((leading != frag->leading) ||
2017 (trailing != frag->trailing))
2018 fprintf(stderr, "Context reduced to (%ld/%ld)"
2019 " to apply fragment at %d\n",
2020 leading, trailing, applied_pos+1);
2021 update_image(img, applied_pos, &preimage, &postimage);
2022 } else {
2023 if (apply_verbosely)
2024 error("while searching for:\n%.*s",
2025 (int)(old - oldlines), oldlines);
2028 free(oldlines);
2029 free(newlines);
2030 free(preimage.line_allocated);
2031 free(postimage.line_allocated);
2033 return (applied_pos < 0);
2036 static int apply_binary_fragment(struct image *img, struct patch *patch)
2038 struct fragment *fragment = patch->fragments;
2039 unsigned long len;
2040 void *dst;
2042 /* Binary patch is irreversible without the optional second hunk */
2043 if (apply_in_reverse) {
2044 if (!fragment->next)
2045 return error("cannot reverse-apply a binary patch "
2046 "without the reverse hunk to '%s'",
2047 patch->new_name
2048 ? patch->new_name : patch->old_name);
2049 fragment = fragment->next;
2051 switch (fragment->binary_patch_method) {
2052 case BINARY_DELTA_DEFLATED:
2053 dst = patch_delta(img->buf, img->len, fragment->patch,
2054 fragment->size, &len);
2055 if (!dst)
2056 return -1;
2057 clear_image(img);
2058 img->buf = dst;
2059 img->len = len;
2060 return 0;
2061 case BINARY_LITERAL_DEFLATED:
2062 clear_image(img);
2063 img->len = fragment->size;
2064 img->buf = xmalloc(img->len+1);
2065 memcpy(img->buf, fragment->patch, img->len);
2066 img->buf[img->len] = '\0';
2067 return 0;
2069 return -1;
2072 static int apply_binary(struct image *img, struct patch *patch)
2074 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2075 unsigned char sha1[20];
2078 * For safety, we require patch index line to contain
2079 * full 40-byte textual SHA1 for old and new, at least for now.
2081 if (strlen(patch->old_sha1_prefix) != 40 ||
2082 strlen(patch->new_sha1_prefix) != 40 ||
2083 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2084 get_sha1_hex(patch->new_sha1_prefix, sha1))
2085 return error("cannot apply binary patch to '%s' "
2086 "without full index line", name);
2088 if (patch->old_name) {
2090 * See if the old one matches what the patch
2091 * applies to.
2093 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2094 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2095 return error("the patch applies to '%s' (%s), "
2096 "which does not match the "
2097 "current contents.",
2098 name, sha1_to_hex(sha1));
2100 else {
2101 /* Otherwise, the old one must be empty. */
2102 if (img->len)
2103 return error("the patch applies to an empty "
2104 "'%s' but it is not empty", name);
2107 get_sha1_hex(patch->new_sha1_prefix, sha1);
2108 if (is_null_sha1(sha1)) {
2109 clear_image(img);
2110 return 0; /* deletion patch */
2113 if (has_sha1_file(sha1)) {
2114 /* We already have the postimage */
2115 enum object_type type;
2116 unsigned long size;
2117 char *result;
2119 result = read_sha1_file(sha1, &type, &size);
2120 if (!result)
2121 return error("the necessary postimage %s for "
2122 "'%s' cannot be read",
2123 patch->new_sha1_prefix, name);
2124 clear_image(img);
2125 img->buf = result;
2126 img->len = size;
2127 } else {
2129 * We have verified buf matches the preimage;
2130 * apply the patch data to it, which is stored
2131 * in the patch->fragments->{patch,size}.
2133 if (apply_binary_fragment(img, patch))
2134 return error("binary patch does not apply to '%s'",
2135 name);
2137 /* verify that the result matches */
2138 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2139 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2140 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2141 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2144 return 0;
2147 static int apply_fragments(struct image *img, struct patch *patch)
2149 struct fragment *frag = patch->fragments;
2150 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2151 unsigned ws_rule = patch->ws_rule;
2152 unsigned inaccurate_eof = patch->inaccurate_eof;
2154 if (patch->is_binary)
2155 return apply_binary(img, patch);
2157 while (frag) {
2158 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2159 error("patch failed: %s:%ld", name, frag->oldpos);
2160 if (!apply_with_reject)
2161 return -1;
2162 frag->rejected = 1;
2164 frag = frag->next;
2166 return 0;
2169 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2171 if (!ce)
2172 return 0;
2174 if (S_ISGITLINK(ce->ce_mode)) {
2175 strbuf_grow(buf, 100);
2176 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2177 } else {
2178 enum object_type type;
2179 unsigned long sz;
2180 char *result;
2182 result = read_sha1_file(ce->sha1, &type, &sz);
2183 if (!result)
2184 return -1;
2185 /* XXX read_sha1_file NUL-terminates */
2186 strbuf_attach(buf, result, sz, sz + 1);
2188 return 0;
2191 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2193 struct strbuf buf;
2194 struct image image;
2195 size_t len;
2196 char *img;
2198 strbuf_init(&buf, 0);
2199 if (cached) {
2200 if (read_file_or_gitlink(ce, &buf))
2201 return error("read of %s failed", patch->old_name);
2202 } else if (patch->old_name) {
2203 if (S_ISGITLINK(patch->old_mode)) {
2204 if (ce) {
2205 read_file_or_gitlink(ce, &buf);
2206 } else {
2208 * There is no way to apply subproject
2209 * patch without looking at the index.
2211 patch->fragments = NULL;
2213 } else {
2214 if (read_old_data(st, patch->old_name, &buf))
2215 return error("read of %s failed", patch->old_name);
2219 img = strbuf_detach(&buf, &len);
2220 prepare_image(&image, img, len, !patch->is_binary);
2222 if (apply_fragments(&image, patch) < 0)
2223 return -1; /* note with --reject this succeeds. */
2224 patch->result = image.buf;
2225 patch->resultsize = image.len;
2226 free(image.line_allocated);
2228 if (0 < patch->is_delete && patch->resultsize)
2229 return error("removal patch leaves file contents");
2231 return 0;
2234 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2236 struct stat nst;
2237 if (!lstat(new_name, &nst)) {
2238 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2239 return 0;
2241 * A leading component of new_name might be a symlink
2242 * that is going to be removed with this patch, but
2243 * still pointing at somewhere that has the path.
2244 * In such a case, path "new_name" does not exist as
2245 * far as git is concerned.
2247 if (has_symlink_leading_path(strlen(new_name), new_name))
2248 return 0;
2250 return error("%s: already exists in working directory", new_name);
2252 else if ((errno != ENOENT) && (errno != ENOTDIR))
2253 return error("%s: %s", new_name, strerror(errno));
2254 return 0;
2257 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2259 if (S_ISGITLINK(ce->ce_mode)) {
2260 if (!S_ISDIR(st->st_mode))
2261 return -1;
2262 return 0;
2264 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2267 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2269 const char *old_name = patch->old_name;
2270 int stat_ret = 0;
2271 unsigned st_mode = 0;
2274 * Make sure that we do not have local modifications from the
2275 * index when we are looking at the index. Also make sure
2276 * we have the preimage file to be patched in the work tree,
2277 * unless --cached, which tells git to apply only in the index.
2279 if (!old_name)
2280 return 0;
2282 assert(patch->is_new <= 0);
2283 if (!cached) {
2284 stat_ret = lstat(old_name, st);
2285 if (stat_ret && errno != ENOENT)
2286 return error("%s: %s", old_name, strerror(errno));
2288 if (check_index) {
2289 int pos = cache_name_pos(old_name, strlen(old_name));
2290 if (pos < 0) {
2291 if (patch->is_new < 0)
2292 goto is_new;
2293 return error("%s: does not exist in index", old_name);
2295 *ce = active_cache[pos];
2296 if (stat_ret < 0) {
2297 struct checkout costate;
2298 /* checkout */
2299 costate.base_dir = "";
2300 costate.base_dir_len = 0;
2301 costate.force = 0;
2302 costate.quiet = 0;
2303 costate.not_new = 0;
2304 costate.refresh_cache = 1;
2305 if (checkout_entry(*ce, &costate, NULL) ||
2306 lstat(old_name, st))
2307 return -1;
2309 if (!cached && verify_index_match(*ce, st))
2310 return error("%s: does not match index", old_name);
2311 if (cached)
2312 st_mode = (*ce)->ce_mode;
2313 } else if (stat_ret < 0) {
2314 if (patch->is_new < 0)
2315 goto is_new;
2316 return error("%s: %s", old_name, strerror(errno));
2319 if (!cached)
2320 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2322 if (patch->is_new < 0)
2323 patch->is_new = 0;
2324 if (!patch->old_mode)
2325 patch->old_mode = st_mode;
2326 if ((st_mode ^ patch->old_mode) & S_IFMT)
2327 return error("%s: wrong type", old_name);
2328 if (st_mode != patch->old_mode)
2329 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2330 old_name, st_mode, patch->old_mode);
2331 return 0;
2333 is_new:
2334 patch->is_new = 1;
2335 patch->is_delete = 0;
2336 patch->old_name = NULL;
2337 return 0;
2340 static int check_patch(struct patch *patch, struct patch *prev_patch)
2342 struct stat st;
2343 const char *old_name = patch->old_name;
2344 const char *new_name = patch->new_name;
2345 const char *name = old_name ? old_name : new_name;
2346 struct cache_entry *ce = NULL;
2347 int ok_if_exists;
2348 int status;
2350 patch->rejected = 1; /* we will drop this after we succeed */
2352 status = check_preimage(patch, &ce, &st);
2353 if (status)
2354 return status;
2355 old_name = patch->old_name;
2357 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2358 !strcmp(prev_patch->old_name, new_name))
2360 * A type-change diff is always split into a patch to
2361 * delete old, immediately followed by a patch to
2362 * create new (see diff.c::run_diff()); in such a case
2363 * it is Ok that the entry to be deleted by the
2364 * previous patch is still in the working tree and in
2365 * the index.
2367 ok_if_exists = 1;
2368 else
2369 ok_if_exists = 0;
2371 if (new_name &&
2372 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2373 if (check_index &&
2374 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2375 !ok_if_exists)
2376 return error("%s: already exists in index", new_name);
2377 if (!cached) {
2378 int err = check_to_create_blob(new_name, ok_if_exists);
2379 if (err)
2380 return err;
2382 if (!patch->new_mode) {
2383 if (0 < patch->is_new)
2384 patch->new_mode = S_IFREG | 0644;
2385 else
2386 patch->new_mode = patch->old_mode;
2390 if (new_name && old_name) {
2391 int same = !strcmp(old_name, new_name);
2392 if (!patch->new_mode)
2393 patch->new_mode = patch->old_mode;
2394 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2395 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2396 patch->new_mode, new_name, patch->old_mode,
2397 same ? "" : " of ", same ? "" : old_name);
2400 if (apply_data(patch, &st, ce) < 0)
2401 return error("%s: patch does not apply", name);
2402 patch->rejected = 0;
2403 return 0;
2406 static int check_patch_list(struct patch *patch)
2408 struct patch *prev_patch = NULL;
2409 int err = 0;
2411 for (prev_patch = NULL; patch ; patch = patch->next) {
2412 if (apply_verbosely)
2413 say_patch_name(stderr,
2414 "Checking patch ", patch, "...\n");
2415 err |= check_patch(patch, prev_patch);
2416 prev_patch = patch;
2418 return err;
2421 /* This function tries to read the sha1 from the current index */
2422 static int get_current_sha1(const char *path, unsigned char *sha1)
2424 int pos;
2426 if (read_cache() < 0)
2427 return -1;
2428 pos = cache_name_pos(path, strlen(path));
2429 if (pos < 0)
2430 return -1;
2431 hashcpy(sha1, active_cache[pos]->sha1);
2432 return 0;
2435 /* Build an index that contains the just the files needed for a 3way merge */
2436 static void build_fake_ancestor(struct patch *list, const char *filename)
2438 struct patch *patch;
2439 struct index_state result = { 0 };
2440 int fd;
2442 /* Once we start supporting the reverse patch, it may be
2443 * worth showing the new sha1 prefix, but until then...
2445 for (patch = list; patch; patch = patch->next) {
2446 const unsigned char *sha1_ptr;
2447 unsigned char sha1[20];
2448 struct cache_entry *ce;
2449 const char *name;
2451 name = patch->old_name ? patch->old_name : patch->new_name;
2452 if (0 < patch->is_new)
2453 continue;
2454 else if (get_sha1(patch->old_sha1_prefix, sha1))
2455 /* git diff has no index line for mode/type changes */
2456 if (!patch->lines_added && !patch->lines_deleted) {
2457 if (get_current_sha1(patch->new_name, sha1) ||
2458 get_current_sha1(patch->old_name, sha1))
2459 die("mode change for %s, which is not "
2460 "in current HEAD", name);
2461 sha1_ptr = sha1;
2462 } else
2463 die("sha1 information is lacking or useless "
2464 "(%s).", name);
2465 else
2466 sha1_ptr = sha1;
2468 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2469 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2470 die ("Could not add %s to temporary index", name);
2473 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2474 if (fd < 0 || write_index(&result, fd) || close(fd))
2475 die ("Could not write temporary index to %s", filename);
2477 discard_index(&result);
2480 static void stat_patch_list(struct patch *patch)
2482 int files, adds, dels;
2484 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2485 files++;
2486 adds += patch->lines_added;
2487 dels += patch->lines_deleted;
2488 show_stats(patch);
2491 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2494 static void numstat_patch_list(struct patch *patch)
2496 for ( ; patch; patch = patch->next) {
2497 const char *name;
2498 name = patch->new_name ? patch->new_name : patch->old_name;
2499 if (patch->is_binary)
2500 printf("-\t-\t");
2501 else
2502 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2503 write_name_quoted(name, stdout, line_termination);
2507 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2509 if (mode)
2510 printf(" %s mode %06o %s\n", newdelete, mode, name);
2511 else
2512 printf(" %s %s\n", newdelete, name);
2515 static void show_mode_change(struct patch *p, int show_name)
2517 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2518 if (show_name)
2519 printf(" mode change %06o => %06o %s\n",
2520 p->old_mode, p->new_mode, p->new_name);
2521 else
2522 printf(" mode change %06o => %06o\n",
2523 p->old_mode, p->new_mode);
2527 static void show_rename_copy(struct patch *p)
2529 const char *renamecopy = p->is_rename ? "rename" : "copy";
2530 const char *old, *new;
2532 /* Find common prefix */
2533 old = p->old_name;
2534 new = p->new_name;
2535 while (1) {
2536 const char *slash_old, *slash_new;
2537 slash_old = strchr(old, '/');
2538 slash_new = strchr(new, '/');
2539 if (!slash_old ||
2540 !slash_new ||
2541 slash_old - old != slash_new - new ||
2542 memcmp(old, new, slash_new - new))
2543 break;
2544 old = slash_old + 1;
2545 new = slash_new + 1;
2547 /* p->old_name thru old is the common prefix, and old and new
2548 * through the end of names are renames
2550 if (old != p->old_name)
2551 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2552 (int)(old - p->old_name), p->old_name,
2553 old, new, p->score);
2554 else
2555 printf(" %s %s => %s (%d%%)\n", renamecopy,
2556 p->old_name, p->new_name, p->score);
2557 show_mode_change(p, 0);
2560 static void summary_patch_list(struct patch *patch)
2562 struct patch *p;
2564 for (p = patch; p; p = p->next) {
2565 if (p->is_new)
2566 show_file_mode_name("create", p->new_mode, p->new_name);
2567 else if (p->is_delete)
2568 show_file_mode_name("delete", p->old_mode, p->old_name);
2569 else {
2570 if (p->is_rename || p->is_copy)
2571 show_rename_copy(p);
2572 else {
2573 if (p->score) {
2574 printf(" rewrite %s (%d%%)\n",
2575 p->new_name, p->score);
2576 show_mode_change(p, 0);
2578 else
2579 show_mode_change(p, 1);
2585 static void patch_stats(struct patch *patch)
2587 int lines = patch->lines_added + patch->lines_deleted;
2589 if (lines > max_change)
2590 max_change = lines;
2591 if (patch->old_name) {
2592 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2593 if (!len)
2594 len = strlen(patch->old_name);
2595 if (len > max_len)
2596 max_len = len;
2598 if (patch->new_name) {
2599 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2600 if (!len)
2601 len = strlen(patch->new_name);
2602 if (len > max_len)
2603 max_len = len;
2607 static void remove_file(struct patch *patch, int rmdir_empty)
2609 if (update_index) {
2610 if (remove_file_from_cache(patch->old_name) < 0)
2611 die("unable to remove %s from index", patch->old_name);
2613 if (!cached) {
2614 if (S_ISGITLINK(patch->old_mode)) {
2615 if (rmdir(patch->old_name))
2616 warning("unable to remove submodule %s",
2617 patch->old_name);
2618 } else if (!unlink(patch->old_name) && rmdir_empty) {
2619 char *name = xstrdup(patch->old_name);
2620 char *end = strrchr(name, '/');
2621 while (end) {
2622 *end = 0;
2623 if (rmdir(name))
2624 break;
2625 end = strrchr(name, '/');
2627 free(name);
2632 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2634 struct stat st;
2635 struct cache_entry *ce;
2636 int namelen = strlen(path);
2637 unsigned ce_size = cache_entry_size(namelen);
2639 if (!update_index)
2640 return;
2642 ce = xcalloc(1, ce_size);
2643 memcpy(ce->name, path, namelen);
2644 ce->ce_mode = create_ce_mode(mode);
2645 ce->ce_flags = namelen;
2646 if (S_ISGITLINK(mode)) {
2647 const char *s = buf;
2649 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2650 die("corrupt patch for subproject %s", path);
2651 } else {
2652 if (!cached) {
2653 if (lstat(path, &st) < 0)
2654 die("unable to stat newly created file %s",
2655 path);
2656 fill_stat_cache_info(ce, &st);
2658 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2659 die("unable to create backing store for newly created file %s", path);
2661 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2662 die("unable to add cache entry for %s", path);
2665 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2667 int fd;
2668 struct strbuf nbuf;
2670 if (S_ISGITLINK(mode)) {
2671 struct stat st;
2672 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2673 return 0;
2674 return mkdir(path, 0777);
2677 if (has_symlinks && S_ISLNK(mode))
2678 /* Although buf:size is counted string, it also is NUL
2679 * terminated.
2681 return symlink(buf, path);
2683 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2684 if (fd < 0)
2685 return -1;
2687 strbuf_init(&nbuf, 0);
2688 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2689 size = nbuf.len;
2690 buf = nbuf.buf;
2692 write_or_die(fd, buf, size);
2693 strbuf_release(&nbuf);
2695 if (close(fd) < 0)
2696 die("closing file %s: %s", path, strerror(errno));
2697 return 0;
2701 * We optimistically assume that the directories exist,
2702 * which is true 99% of the time anyway. If they don't,
2703 * we create them and try again.
2705 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2707 if (cached)
2708 return;
2709 if (!try_create_file(path, mode, buf, size))
2710 return;
2712 if (errno == ENOENT) {
2713 if (safe_create_leading_directories(path))
2714 return;
2715 if (!try_create_file(path, mode, buf, size))
2716 return;
2719 if (errno == EEXIST || errno == EACCES) {
2720 /* We may be trying to create a file where a directory
2721 * used to be.
2723 struct stat st;
2724 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2725 errno = EEXIST;
2728 if (errno == EEXIST) {
2729 unsigned int nr = getpid();
2731 for (;;) {
2732 const char *newpath;
2733 newpath = mkpath("%s~%u", path, nr);
2734 if (!try_create_file(newpath, mode, buf, size)) {
2735 if (!rename(newpath, path))
2736 return;
2737 unlink(newpath);
2738 break;
2740 if (errno != EEXIST)
2741 break;
2742 ++nr;
2745 die("unable to write file %s mode %o", path, mode);
2748 static void create_file(struct patch *patch)
2750 char *path = patch->new_name;
2751 unsigned mode = patch->new_mode;
2752 unsigned long size = patch->resultsize;
2753 char *buf = patch->result;
2755 if (!mode)
2756 mode = S_IFREG | 0644;
2757 create_one_file(path, mode, buf, size);
2758 add_index_file(path, mode, buf, size);
2761 /* phase zero is to remove, phase one is to create */
2762 static void write_out_one_result(struct patch *patch, int phase)
2764 if (patch->is_delete > 0) {
2765 if (phase == 0)
2766 remove_file(patch, 1);
2767 return;
2769 if (patch->is_new > 0 || patch->is_copy) {
2770 if (phase == 1)
2771 create_file(patch);
2772 return;
2775 * Rename or modification boils down to the same
2776 * thing: remove the old, write the new
2778 if (phase == 0)
2779 remove_file(patch, patch->is_rename);
2780 if (phase == 1)
2781 create_file(patch);
2784 static int write_out_one_reject(struct patch *patch)
2786 FILE *rej;
2787 char namebuf[PATH_MAX];
2788 struct fragment *frag;
2789 int cnt = 0;
2791 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2792 if (!frag->rejected)
2793 continue;
2794 cnt++;
2797 if (!cnt) {
2798 if (apply_verbosely)
2799 say_patch_name(stderr,
2800 "Applied patch ", patch, " cleanly.\n");
2801 return 0;
2804 /* This should not happen, because a removal patch that leaves
2805 * contents are marked "rejected" at the patch level.
2807 if (!patch->new_name)
2808 die("internal error");
2810 /* Say this even without --verbose */
2811 say_patch_name(stderr, "Applying patch ", patch, " with");
2812 fprintf(stderr, " %d rejects...\n", cnt);
2814 cnt = strlen(patch->new_name);
2815 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2816 cnt = ARRAY_SIZE(namebuf) - 5;
2817 fprintf(stderr,
2818 "warning: truncating .rej filename to %.*s.rej",
2819 cnt - 1, patch->new_name);
2821 memcpy(namebuf, patch->new_name, cnt);
2822 memcpy(namebuf + cnt, ".rej", 5);
2824 rej = fopen(namebuf, "w");
2825 if (!rej)
2826 return error("cannot open %s: %s", namebuf, strerror(errno));
2828 /* Normal git tools never deal with .rej, so do not pretend
2829 * this is a git patch by saying --git nor give extended
2830 * headers. While at it, maybe please "kompare" that wants
2831 * the trailing TAB and some garbage at the end of line ;-).
2833 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2834 patch->new_name, patch->new_name);
2835 for (cnt = 1, frag = patch->fragments;
2836 frag;
2837 cnt++, frag = frag->next) {
2838 if (!frag->rejected) {
2839 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2840 continue;
2842 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2843 fprintf(rej, "%.*s", frag->size, frag->patch);
2844 if (frag->patch[frag->size-1] != '\n')
2845 fputc('\n', rej);
2847 fclose(rej);
2848 return -1;
2851 static int write_out_results(struct patch *list, int skipped_patch)
2853 int phase;
2854 int errs = 0;
2855 struct patch *l;
2857 if (!list && !skipped_patch)
2858 return error("No changes");
2860 for (phase = 0; phase < 2; phase++) {
2861 l = list;
2862 while (l) {
2863 if (l->rejected)
2864 errs = 1;
2865 else {
2866 write_out_one_result(l, phase);
2867 if (phase == 1 && write_out_one_reject(l))
2868 errs = 1;
2870 l = l->next;
2873 return errs;
2876 static struct lock_file lock_file;
2878 static struct excludes {
2879 struct excludes *next;
2880 const char *path;
2881 } *excludes;
2883 static int use_patch(struct patch *p)
2885 const char *pathname = p->new_name ? p->new_name : p->old_name;
2886 struct excludes *x = excludes;
2887 while (x) {
2888 if (fnmatch(x->path, pathname, 0) == 0)
2889 return 0;
2890 x = x->next;
2892 if (0 < prefix_length) {
2893 int pathlen = strlen(pathname);
2894 if (pathlen <= prefix_length ||
2895 memcmp(prefix, pathname, prefix_length))
2896 return 0;
2898 return 1;
2901 static void prefix_one(char **name)
2903 char *old_name = *name;
2904 if (!old_name)
2905 return;
2906 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2907 free(old_name);
2910 static void prefix_patches(struct patch *p)
2912 if (!prefix || p->is_toplevel_relative)
2913 return;
2914 for ( ; p; p = p->next) {
2915 if (p->new_name == p->old_name) {
2916 char *prefixed = p->new_name;
2917 prefix_one(&prefixed);
2918 p->new_name = p->old_name = prefixed;
2920 else {
2921 prefix_one(&p->new_name);
2922 prefix_one(&p->old_name);
2927 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2929 size_t offset;
2930 struct strbuf buf;
2931 struct patch *list = NULL, **listp = &list;
2932 int skipped_patch = 0;
2934 strbuf_init(&buf, 0);
2935 patch_input_file = filename;
2936 read_patch_file(&buf, fd);
2937 offset = 0;
2938 while (offset < buf.len) {
2939 struct patch *patch;
2940 int nr;
2942 patch = xcalloc(1, sizeof(*patch));
2943 patch->inaccurate_eof = inaccurate_eof;
2944 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2945 if (nr < 0)
2946 break;
2947 if (apply_in_reverse)
2948 reverse_patches(patch);
2949 if (prefix)
2950 prefix_patches(patch);
2951 if (use_patch(patch)) {
2952 patch_stats(patch);
2953 *listp = patch;
2954 listp = &patch->next;
2956 else {
2957 /* perhaps free it a bit better? */
2958 free(patch);
2959 skipped_patch++;
2961 offset += nr;
2964 if (whitespace_error && (ws_error_action == die_on_ws_error))
2965 apply = 0;
2967 update_index = check_index && apply;
2968 if (update_index && newfd < 0)
2969 newfd = hold_locked_index(&lock_file, 1);
2971 if (check_index) {
2972 if (read_cache() < 0)
2973 die("unable to read index file");
2976 if ((check || apply) &&
2977 check_patch_list(list) < 0 &&
2978 !apply_with_reject)
2979 exit(1);
2981 if (apply && write_out_results(list, skipped_patch))
2982 exit(1);
2984 if (fake_ancestor)
2985 build_fake_ancestor(list, fake_ancestor);
2987 if (diffstat)
2988 stat_patch_list(list);
2990 if (numstat)
2991 numstat_patch_list(list);
2993 if (summary)
2994 summary_patch_list(list);
2996 strbuf_release(&buf);
2997 return 0;
3000 static int git_apply_config(const char *var, const char *value, void *cb)
3002 if (!strcmp(var, "apply.whitespace"))
3003 return git_config_string(&apply_default_whitespace, var, value);
3004 return git_default_config(var, value, cb);
3008 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3010 int i;
3011 int read_stdin = 1;
3012 int inaccurate_eof = 0;
3013 int errs = 0;
3014 int is_not_gitdir;
3016 const char *whitespace_option = NULL;
3018 prefix = setup_git_directory_gently(&is_not_gitdir);
3019 prefix_length = prefix ? strlen(prefix) : 0;
3020 git_config(git_apply_config, NULL);
3021 if (apply_default_whitespace)
3022 parse_whitespace_option(apply_default_whitespace);
3024 for (i = 1; i < argc; i++) {
3025 const char *arg = argv[i];
3026 char *end;
3027 int fd;
3029 if (!strcmp(arg, "-")) {
3030 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3031 read_stdin = 0;
3032 continue;
3034 if (!prefixcmp(arg, "--exclude=")) {
3035 struct excludes *x = xmalloc(sizeof(*x));
3036 x->path = arg + 10;
3037 x->next = excludes;
3038 excludes = x;
3039 continue;
3041 if (!prefixcmp(arg, "-p")) {
3042 p_value = atoi(arg + 2);
3043 p_value_known = 1;
3044 continue;
3046 if (!strcmp(arg, "--no-add")) {
3047 no_add = 1;
3048 continue;
3050 if (!strcmp(arg, "--stat")) {
3051 apply = 0;
3052 diffstat = 1;
3053 continue;
3055 if (!strcmp(arg, "--allow-binary-replacement") ||
3056 !strcmp(arg, "--binary")) {
3057 continue; /* now no-op */
3059 if (!strcmp(arg, "--numstat")) {
3060 apply = 0;
3061 numstat = 1;
3062 continue;
3064 if (!strcmp(arg, "--summary")) {
3065 apply = 0;
3066 summary = 1;
3067 continue;
3069 if (!strcmp(arg, "--check")) {
3070 apply = 0;
3071 check = 1;
3072 continue;
3074 if (!strcmp(arg, "--index")) {
3075 if (is_not_gitdir)
3076 die("--index outside a repository");
3077 check_index = 1;
3078 continue;
3080 if (!strcmp(arg, "--cached")) {
3081 if (is_not_gitdir)
3082 die("--cached outside a repository");
3083 check_index = 1;
3084 cached = 1;
3085 continue;
3087 if (!strcmp(arg, "--apply")) {
3088 apply = 1;
3089 continue;
3091 if (!strcmp(arg, "--build-fake-ancestor")) {
3092 apply = 0;
3093 if (++i >= argc)
3094 die ("need a filename");
3095 fake_ancestor = argv[i];
3096 continue;
3098 if (!strcmp(arg, "-z")) {
3099 line_termination = 0;
3100 continue;
3102 if (!prefixcmp(arg, "-C")) {
3103 p_context = strtoul(arg + 2, &end, 0);
3104 if (*end != '\0')
3105 die("unrecognized context count '%s'", arg + 2);
3106 continue;
3108 if (!prefixcmp(arg, "--whitespace=")) {
3109 whitespace_option = arg + 13;
3110 parse_whitespace_option(arg + 13);
3111 continue;
3113 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3114 apply_in_reverse = 1;
3115 continue;
3117 if (!strcmp(arg, "--unidiff-zero")) {
3118 unidiff_zero = 1;
3119 continue;
3121 if (!strcmp(arg, "--reject")) {
3122 apply = apply_with_reject = apply_verbosely = 1;
3123 continue;
3125 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3126 apply_verbosely = 1;
3127 continue;
3129 if (!strcmp(arg, "--inaccurate-eof")) {
3130 inaccurate_eof = 1;
3131 continue;
3133 if (!prefixcmp(arg, "--directory=")) {
3134 arg += strlen("--directory=");
3135 root_len = strlen(arg);
3136 if (root_len && arg[root_len - 1] != '/') {
3137 char *new_root;
3138 root = new_root = xmalloc(root_len + 2);
3139 strcpy(new_root, arg);
3140 strcpy(new_root + root_len++, "/");
3141 } else
3142 root = arg;
3143 continue;
3145 if (0 < prefix_length)
3146 arg = prefix_filename(prefix, prefix_length, arg);
3148 fd = open(arg, O_RDONLY);
3149 if (fd < 0)
3150 die("can't open patch '%s': %s", arg, strerror(errno));
3151 read_stdin = 0;
3152 set_default_whitespace_mode(whitespace_option);
3153 errs |= apply_patch(fd, arg, inaccurate_eof);
3154 close(fd);
3156 set_default_whitespace_mode(whitespace_option);
3157 if (read_stdin)
3158 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3159 if (whitespace_error) {
3160 if (squelch_whitespace_errors &&
3161 squelch_whitespace_errors < whitespace_error) {
3162 int squelched =
3163 whitespace_error - squelch_whitespace_errors;
3164 fprintf(stderr, "warning: squelched %d "
3165 "whitespace error%s\n",
3166 squelched,
3167 squelched == 1 ? "" : "s");
3169 if (ws_error_action == die_on_ws_error)
3170 die("%d line%s add%s whitespace errors.",
3171 whitespace_error,
3172 whitespace_error == 1 ? "" : "s",
3173 whitespace_error == 1 ? "s" : "");
3174 if (applied_after_fixing_ws && apply)
3175 fprintf(stderr, "warning: %d line%s applied after"
3176 " fixing whitespace errors.\n",
3177 applied_after_fixing_ws,
3178 applied_after_fixing_ws == 1 ? "" : "s");
3179 else if (whitespace_error)
3180 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3181 whitespace_error,
3182 whitespace_error == 1 ? "" : "s",
3183 whitespace_error == 1 ? "s" : "");
3186 if (update_index) {
3187 if (write_cache(newfd, active_cache, active_nr) ||
3188 commit_locked_index(&lock_file))
3189 die("Unable to write new index file");
3192 return !!errs;