pack-objects: more threaded load balancing fix with often changed paths
[git/dscho.git] / builtin-apply.c
blobf2e9a332ca0e148366f49b2825ed8826576f1bad
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static const char *fake_ancestor;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|fix|error|error-all>] <patch>...";
50 static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55 } ws_error_action = warn_on_ws_error;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
63 if (!option) {
64 ws_error_action = warn_on_ws_error;
65 return;
67 if (!strcmp(option, "warn")) {
68 ws_error_action = warn_on_ws_error;
69 return;
71 if (!strcmp(option, "nowarn")) {
72 ws_error_action = nowarn_ws_error;
73 return;
75 if (!strcmp(option, "error")) {
76 ws_error_action = die_on_ws_error;
77 return;
79 if (!strcmp(option, "error-all")) {
80 ws_error_action = die_on_ws_error;
81 squelch_whitespace_errors = 0;
82 return;
84 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
85 ws_error_action = correct_ws_error;
86 return;
88 die("unrecognized whitespace option '%s'", option);
91 static void set_default_whitespace_mode(const char *whitespace_option)
93 if (!whitespace_option && !apply_default_whitespace)
94 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
98 * For "diff-stat" like behaviour, we keep track of the biggest change
99 * we've seen, and the longest filename. That allows us to do simple
100 * scaling.
102 static int max_change, max_len;
105 * Various "current state", notably line numbers and what
106 * file (and how) we're patching right now.. The "is_xxxx"
107 * things are flags, where -1 means "don't know yet".
109 static int linenr = 1;
112 * This represents one "hunk" from a patch, starting with
113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
114 * patch text is pointed at by patch, and its byte length
115 * is stored in size. leading and trailing are the number
116 * of context lines.
118 struct fragment {
119 unsigned long leading, trailing;
120 unsigned long oldpos, oldlines;
121 unsigned long newpos, newlines;
122 const char *patch;
123 int size;
124 int rejected;
125 struct fragment *next;
129 * When dealing with a binary patch, we reuse "leading" field
130 * to store the type of the binary hunk, either deflated "delta"
131 * or deflated "literal".
133 #define binary_patch_method leading
134 #define BINARY_DELTA_DEFLATED 1
135 #define BINARY_LITERAL_DEFLATED 2
138 * This represents a "patch" to a file, both metainfo changes
139 * such as creation/deletion, filemode and content changes represented
140 * as a series of fragments.
142 struct patch {
143 char *new_name, *old_name, *def_name;
144 unsigned int old_mode, new_mode;
145 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
146 int rejected;
147 unsigned ws_rule;
148 unsigned long deflate_origlen;
149 int lines_added, lines_deleted;
150 int score;
151 unsigned int is_toplevel_relative:1;
152 unsigned int inaccurate_eof:1;
153 unsigned int is_binary:1;
154 unsigned int is_copy:1;
155 unsigned int is_rename:1;
156 struct fragment *fragments;
157 char *result;
158 size_t resultsize;
159 char old_sha1_prefix[41];
160 char new_sha1_prefix[41];
161 struct patch *next;
164 static void say_patch_name(FILE *output, const char *pre,
165 struct patch *patch, const char *post)
167 fputs(pre, output);
168 if (patch->old_name && patch->new_name &&
169 strcmp(patch->old_name, patch->new_name)) {
170 quote_c_style(patch->old_name, NULL, output, 0);
171 fputs(" => ", output);
172 quote_c_style(patch->new_name, NULL, output, 0);
173 } else {
174 const char *n = patch->new_name;
175 if (!n)
176 n = patch->old_name;
177 quote_c_style(n, NULL, output, 0);
179 fputs(post, output);
182 #define CHUNKSIZE (8192)
183 #define SLOP (16)
185 static void read_patch_file(struct strbuf *sb, int fd)
187 if (strbuf_read(sb, fd, 0) < 0)
188 die("git-apply: read returned %s", strerror(errno));
191 * Make sure that we have some slop in the buffer
192 * so that we can do speculative "memcmp" etc, and
193 * see to it that it is NUL-filled.
195 strbuf_grow(sb, SLOP);
196 memset(sb->buf + sb->len, 0, SLOP);
199 static unsigned long linelen(const char *buffer, unsigned long size)
201 unsigned long len = 0;
202 while (size--) {
203 len++;
204 if (*buffer++ == '\n')
205 break;
207 return len;
210 static int is_dev_null(const char *str)
212 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
215 #define TERM_SPACE 1
216 #define TERM_TAB 2
218 static int name_terminate(const char *name, int namelen, int c, int terminate)
220 if (c == ' ' && !(terminate & TERM_SPACE))
221 return 0;
222 if (c == '\t' && !(terminate & TERM_TAB))
223 return 0;
225 return 1;
228 static char *find_name(const char *line, char *def, int p_value, int terminate)
230 int len;
231 const char *start = line;
233 if (*line == '"') {
234 struct strbuf name;
237 * Proposed "new-style" GNU patch/diff format; see
238 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
240 strbuf_init(&name, 0);
241 if (!unquote_c_style(&name, line, NULL)) {
242 char *cp;
244 for (cp = name.buf; p_value; p_value--) {
245 cp = strchr(cp, '/');
246 if (!cp)
247 break;
248 cp++;
250 if (cp) {
251 /* name can later be freed, so we need
252 * to memmove, not just return cp
254 strbuf_remove(&name, 0, cp - name.buf);
255 free(def);
256 return strbuf_detach(&name, NULL);
259 strbuf_release(&name);
262 for (;;) {
263 char c = *line;
265 if (isspace(c)) {
266 if (c == '\n')
267 break;
268 if (name_terminate(start, line-start, c, terminate))
269 break;
271 line++;
272 if (c == '/' && !--p_value)
273 start = line;
275 if (!start)
276 return def;
277 len = line - start;
278 if (!len)
279 return def;
282 * Generally we prefer the shorter name, especially
283 * if the other one is just a variation of that with
284 * something else tacked on to the end (ie "file.orig"
285 * or "file~").
287 if (def) {
288 int deflen = strlen(def);
289 if (deflen < len && !strncmp(start, def, deflen))
290 return def;
291 free(def);
294 return xmemdupz(start, len);
297 static int count_slashes(const char *cp)
299 int cnt = 0;
300 char ch;
302 while ((ch = *cp++))
303 if (ch == '/')
304 cnt++;
305 return cnt;
309 * Given the string after "--- " or "+++ ", guess the appropriate
310 * p_value for the given patch.
312 static int guess_p_value(const char *nameline)
314 char *name, *cp;
315 int val = -1;
317 if (is_dev_null(nameline))
318 return -1;
319 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
320 if (!name)
321 return -1;
322 cp = strchr(name, '/');
323 if (!cp)
324 val = 0;
325 else if (prefix) {
327 * Does it begin with "a/$our-prefix" and such? Then this is
328 * very likely to apply to our directory.
330 if (!strncmp(name, prefix, prefix_length))
331 val = count_slashes(prefix);
332 else {
333 cp++;
334 if (!strncmp(cp, prefix, prefix_length))
335 val = count_slashes(prefix) + 1;
338 free(name);
339 return val;
343 * Get the name etc info from the --/+++ lines of a traditional patch header
345 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
346 * files, we can happily check the index for a match, but for creating a
347 * new file we should try to match whatever "patch" does. I have no idea.
349 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
351 char *name;
353 first += 4; /* skip "--- " */
354 second += 4; /* skip "+++ " */
355 if (!p_value_known) {
356 int p, q;
357 p = guess_p_value(first);
358 q = guess_p_value(second);
359 if (p < 0) p = q;
360 if (0 <= p && p == q) {
361 p_value = p;
362 p_value_known = 1;
365 if (is_dev_null(first)) {
366 patch->is_new = 1;
367 patch->is_delete = 0;
368 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
369 patch->new_name = name;
370 } else if (is_dev_null(second)) {
371 patch->is_new = 0;
372 patch->is_delete = 1;
373 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
374 patch->old_name = name;
375 } else {
376 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
377 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
378 patch->old_name = patch->new_name = name;
380 if (!name)
381 die("unable to find filename in patch at line %d", linenr);
384 static int gitdiff_hdrend(const char *line, struct patch *patch)
386 return -1;
390 * We're anal about diff header consistency, to make
391 * sure that we don't end up having strange ambiguous
392 * patches floating around.
394 * As a result, gitdiff_{old|new}name() will check
395 * their names against any previous information, just
396 * to make sure..
398 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
400 if (!orig_name && !isnull)
401 return find_name(line, NULL, p_value, TERM_TAB);
403 if (orig_name) {
404 int len;
405 const char *name;
406 char *another;
407 name = orig_name;
408 len = strlen(name);
409 if (isnull)
410 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
411 another = find_name(line, NULL, p_value, TERM_TAB);
412 if (!another || memcmp(another, name, len))
413 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
414 free(another);
415 return orig_name;
417 else {
418 /* expect "/dev/null" */
419 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
420 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
421 return NULL;
425 static int gitdiff_oldname(const char *line, struct patch *patch)
427 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
428 return 0;
431 static int gitdiff_newname(const char *line, struct patch *patch)
433 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
434 return 0;
437 static int gitdiff_oldmode(const char *line, struct patch *patch)
439 patch->old_mode = strtoul(line, NULL, 8);
440 return 0;
443 static int gitdiff_newmode(const char *line, struct patch *patch)
445 patch->new_mode = strtoul(line, NULL, 8);
446 return 0;
449 static int gitdiff_delete(const char *line, struct patch *patch)
451 patch->is_delete = 1;
452 patch->old_name = patch->def_name;
453 return gitdiff_oldmode(line, patch);
456 static int gitdiff_newfile(const char *line, struct patch *patch)
458 patch->is_new = 1;
459 patch->new_name = patch->def_name;
460 return gitdiff_newmode(line, patch);
463 static int gitdiff_copysrc(const char *line, struct patch *patch)
465 patch->is_copy = 1;
466 patch->old_name = find_name(line, NULL, 0, 0);
467 return 0;
470 static int gitdiff_copydst(const char *line, struct patch *patch)
472 patch->is_copy = 1;
473 patch->new_name = find_name(line, NULL, 0, 0);
474 return 0;
477 static int gitdiff_renamesrc(const char *line, struct patch *patch)
479 patch->is_rename = 1;
480 patch->old_name = find_name(line, NULL, 0, 0);
481 return 0;
484 static int gitdiff_renamedst(const char *line, struct patch *patch)
486 patch->is_rename = 1;
487 patch->new_name = find_name(line, NULL, 0, 0);
488 return 0;
491 static int gitdiff_similarity(const char *line, struct patch *patch)
493 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
494 patch->score = 0;
495 return 0;
498 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
500 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
501 patch->score = 0;
502 return 0;
505 static int gitdiff_index(const char *line, struct patch *patch)
508 * index line is N hexadecimal, "..", N hexadecimal,
509 * and optional space with octal mode.
511 const char *ptr, *eol;
512 int len;
514 ptr = strchr(line, '.');
515 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
516 return 0;
517 len = ptr - line;
518 memcpy(patch->old_sha1_prefix, line, len);
519 patch->old_sha1_prefix[len] = 0;
521 line = ptr + 2;
522 ptr = strchr(line, ' ');
523 eol = strchr(line, '\n');
525 if (!ptr || eol < ptr)
526 ptr = eol;
527 len = ptr - line;
529 if (40 < len)
530 return 0;
531 memcpy(patch->new_sha1_prefix, line, len);
532 patch->new_sha1_prefix[len] = 0;
533 if (*ptr == ' ')
534 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
535 return 0;
539 * This is normal for a diff that doesn't change anything: we'll fall through
540 * into the next diff. Tell the parser to break out.
542 static int gitdiff_unrecognized(const char *line, struct patch *patch)
544 return -1;
547 static const char *stop_at_slash(const char *line, int llen)
549 int i;
551 for (i = 0; i < llen; i++) {
552 int ch = line[i];
553 if (ch == '/')
554 return line + i;
556 return NULL;
560 * This is to extract the same name that appears on "diff --git"
561 * line. We do not find and return anything if it is a rename
562 * patch, and it is OK because we will find the name elsewhere.
563 * We need to reliably find name only when it is mode-change only,
564 * creation or deletion of an empty file. In any of these cases,
565 * both sides are the same name under a/ and b/ respectively.
567 static char *git_header_name(char *line, int llen)
569 const char *name;
570 const char *second = NULL;
571 size_t len;
573 line += strlen("diff --git ");
574 llen -= strlen("diff --git ");
576 if (*line == '"') {
577 const char *cp;
578 struct strbuf first;
579 struct strbuf sp;
581 strbuf_init(&first, 0);
582 strbuf_init(&sp, 0);
584 if (unquote_c_style(&first, line, &second))
585 goto free_and_fail1;
587 /* advance to the first slash */
588 cp = stop_at_slash(first.buf, first.len);
589 /* we do not accept absolute paths */
590 if (!cp || cp == first.buf)
591 goto free_and_fail1;
592 strbuf_remove(&first, 0, cp + 1 - first.buf);
595 * second points at one past closing dq of name.
596 * find the second name.
598 while ((second < line + llen) && isspace(*second))
599 second++;
601 if (line + llen <= second)
602 goto free_and_fail1;
603 if (*second == '"') {
604 if (unquote_c_style(&sp, second, NULL))
605 goto free_and_fail1;
606 cp = stop_at_slash(sp.buf, sp.len);
607 if (!cp || cp == sp.buf)
608 goto free_and_fail1;
609 /* They must match, otherwise ignore */
610 if (strcmp(cp + 1, first.buf))
611 goto free_and_fail1;
612 strbuf_release(&sp);
613 return strbuf_detach(&first, NULL);
616 /* unquoted second */
617 cp = stop_at_slash(second, line + llen - second);
618 if (!cp || cp == second)
619 goto free_and_fail1;
620 cp++;
621 if (line + llen - cp != first.len + 1 ||
622 memcmp(first.buf, cp, first.len))
623 goto free_and_fail1;
624 return strbuf_detach(&first, NULL);
626 free_and_fail1:
627 strbuf_release(&first);
628 strbuf_release(&sp);
629 return NULL;
632 /* unquoted first name */
633 name = stop_at_slash(line, llen);
634 if (!name || name == line)
635 return NULL;
636 name++;
639 * since the first name is unquoted, a dq if exists must be
640 * the beginning of the second name.
642 for (second = name; second < line + llen; second++) {
643 if (*second == '"') {
644 struct strbuf sp;
645 const char *np;
647 strbuf_init(&sp, 0);
648 if (unquote_c_style(&sp, second, NULL))
649 goto free_and_fail2;
651 np = stop_at_slash(sp.buf, sp.len);
652 if (!np || np == sp.buf)
653 goto free_and_fail2;
654 np++;
656 len = sp.buf + sp.len - np;
657 if (len < second - name &&
658 !strncmp(np, name, len) &&
659 isspace(name[len])) {
660 /* Good */
661 strbuf_remove(&sp, 0, np - sp.buf);
662 return strbuf_detach(&sp, NULL);
665 free_and_fail2:
666 strbuf_release(&sp);
667 return NULL;
672 * Accept a name only if it shows up twice, exactly the same
673 * form.
675 for (len = 0 ; ; len++) {
676 switch (name[len]) {
677 default:
678 continue;
679 case '\n':
680 return NULL;
681 case '\t': case ' ':
682 second = name+len;
683 for (;;) {
684 char c = *second++;
685 if (c == '\n')
686 return NULL;
687 if (c == '/')
688 break;
690 if (second[len] == '\n' && !memcmp(name, second, len)) {
691 return xmemdupz(name, len);
697 /* Verify that we recognize the lines following a git header */
698 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
700 unsigned long offset;
702 /* A git diff has explicit new/delete information, so we don't guess */
703 patch->is_new = 0;
704 patch->is_delete = 0;
707 * Some things may not have the old name in the
708 * rest of the headers anywhere (pure mode changes,
709 * or removing or adding empty files), so we get
710 * the default name from the header.
712 patch->def_name = git_header_name(line, len);
714 line += len;
715 size -= len;
716 linenr++;
717 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
718 static const struct opentry {
719 const char *str;
720 int (*fn)(const char *, struct patch *);
721 } optable[] = {
722 { "@@ -", gitdiff_hdrend },
723 { "--- ", gitdiff_oldname },
724 { "+++ ", gitdiff_newname },
725 { "old mode ", gitdiff_oldmode },
726 { "new mode ", gitdiff_newmode },
727 { "deleted file mode ", gitdiff_delete },
728 { "new file mode ", gitdiff_newfile },
729 { "copy from ", gitdiff_copysrc },
730 { "copy to ", gitdiff_copydst },
731 { "rename old ", gitdiff_renamesrc },
732 { "rename new ", gitdiff_renamedst },
733 { "rename from ", gitdiff_renamesrc },
734 { "rename to ", gitdiff_renamedst },
735 { "similarity index ", gitdiff_similarity },
736 { "dissimilarity index ", gitdiff_dissimilarity },
737 { "index ", gitdiff_index },
738 { "", gitdiff_unrecognized },
740 int i;
742 len = linelen(line, size);
743 if (!len || line[len-1] != '\n')
744 break;
745 for (i = 0; i < ARRAY_SIZE(optable); i++) {
746 const struct opentry *p = optable + i;
747 int oplen = strlen(p->str);
748 if (len < oplen || memcmp(p->str, line, oplen))
749 continue;
750 if (p->fn(line + oplen, patch) < 0)
751 return offset;
752 break;
756 return offset;
759 static int parse_num(const char *line, unsigned long *p)
761 char *ptr;
763 if (!isdigit(*line))
764 return 0;
765 *p = strtoul(line, &ptr, 10);
766 return ptr - line;
769 static int parse_range(const char *line, int len, int offset, const char *expect,
770 unsigned long *p1, unsigned long *p2)
772 int digits, ex;
774 if (offset < 0 || offset >= len)
775 return -1;
776 line += offset;
777 len -= offset;
779 digits = parse_num(line, p1);
780 if (!digits)
781 return -1;
783 offset += digits;
784 line += digits;
785 len -= digits;
787 *p2 = 1;
788 if (*line == ',') {
789 digits = parse_num(line+1, p2);
790 if (!digits)
791 return -1;
793 offset += digits+1;
794 line += digits+1;
795 len -= digits+1;
798 ex = strlen(expect);
799 if (ex > len)
800 return -1;
801 if (memcmp(line, expect, ex))
802 return -1;
804 return offset + ex;
808 * Parse a unified diff fragment header of the
809 * form "@@ -a,b +c,d @@"
811 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
813 int offset;
815 if (!len || line[len-1] != '\n')
816 return -1;
818 /* Figure out the number of lines in a fragment */
819 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
820 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
822 return offset;
825 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
827 unsigned long offset, len;
829 patch->is_toplevel_relative = 0;
830 patch->is_rename = patch->is_copy = 0;
831 patch->is_new = patch->is_delete = -1;
832 patch->old_mode = patch->new_mode = 0;
833 patch->old_name = patch->new_name = NULL;
834 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
835 unsigned long nextlen;
837 len = linelen(line, size);
838 if (!len)
839 break;
841 /* Testing this early allows us to take a few shortcuts.. */
842 if (len < 6)
843 continue;
846 * Make sure we don't find any unconnected patch fragments.
847 * That's a sign that we didn't find a header, and that a
848 * patch has become corrupted/broken up.
850 if (!memcmp("@@ -", line, 4)) {
851 struct fragment dummy;
852 if (parse_fragment_header(line, len, &dummy) < 0)
853 continue;
854 die("patch fragment without header at line %d: %.*s",
855 linenr, (int)len-1, line);
858 if (size < len + 6)
859 break;
862 * Git patch? It might not have a real patch, just a rename
863 * or mode change, so we handle that specially
865 if (!memcmp("diff --git ", line, 11)) {
866 int git_hdr_len = parse_git_header(line, len, size, patch);
867 if (git_hdr_len <= len)
868 continue;
869 if (!patch->old_name && !patch->new_name) {
870 if (!patch->def_name)
871 die("git diff header lacks filename information (line %d)", linenr);
872 patch->old_name = patch->new_name = patch->def_name;
874 patch->is_toplevel_relative = 1;
875 *hdrsize = git_hdr_len;
876 return offset;
879 /* --- followed by +++ ? */
880 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
881 continue;
884 * We only accept unified patches, so we want it to
885 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
886 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
888 nextlen = linelen(line + len, size - len);
889 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
890 continue;
892 /* Ok, we'll consider it a patch */
893 parse_traditional_patch(line, line+len, patch);
894 *hdrsize = len + nextlen;
895 linenr += 2;
896 return offset;
898 return -1;
901 static void check_whitespace(const char *line, int len, unsigned ws_rule)
903 const char *err = "Adds trailing whitespace";
904 int seen_space = 0;
905 int i;
908 * We know len is at least two, since we have a '+' and we
909 * checked that the last character was a '\n' before calling
910 * this function. That is, an addition of an empty line would
911 * check the '+' here. Sneaky...
913 if ((ws_rule & WS_TRAILING_SPACE) && isspace(line[len-2]))
914 goto error;
917 * Make sure that there is no space followed by a tab in
918 * indentation.
920 if (ws_rule & WS_SPACE_BEFORE_TAB) {
921 err = "Space in indent is followed by a tab";
922 for (i = 1; i < len; i++) {
923 if (line[i] == '\t') {
924 if (seen_space)
925 goto error;
927 else if (line[i] == ' ')
928 seen_space = 1;
929 else
930 break;
935 * Make sure that the indentation does not contain more than
936 * 8 spaces.
938 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
939 (8 < len) && !strncmp("+ ", line, 9)) {
940 err = "Indent more than 8 places with spaces";
941 goto error;
943 return;
945 error:
946 whitespace_error++;
947 if (squelch_whitespace_errors &&
948 squelch_whitespace_errors < whitespace_error)
950 else
951 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
952 err, patch_input_file, linenr, len-2, line+1);
956 * Parse a unified diff. Note that this really needs to parse each
957 * fragment separately, since the only way to know the difference
958 * between a "---" that is part of a patch, and a "---" that starts
959 * the next patch is to look at the line counts..
961 static int parse_fragment(char *line, unsigned long size,
962 struct patch *patch, struct fragment *fragment)
964 int added, deleted;
965 int len = linelen(line, size), offset;
966 unsigned long oldlines, newlines;
967 unsigned long leading, trailing;
969 offset = parse_fragment_header(line, len, fragment);
970 if (offset < 0)
971 return -1;
972 oldlines = fragment->oldlines;
973 newlines = fragment->newlines;
974 leading = 0;
975 trailing = 0;
977 /* Parse the thing.. */
978 line += len;
979 size -= len;
980 linenr++;
981 added = deleted = 0;
982 for (offset = len;
983 0 < size;
984 offset += len, size -= len, line += len, linenr++) {
985 if (!oldlines && !newlines)
986 break;
987 len = linelen(line, size);
988 if (!len || line[len-1] != '\n')
989 return -1;
990 switch (*line) {
991 default:
992 return -1;
993 case '\n': /* newer GNU diff, an empty context line */
994 case ' ':
995 oldlines--;
996 newlines--;
997 if (!deleted && !added)
998 leading++;
999 trailing++;
1000 break;
1001 case '-':
1002 if (apply_in_reverse &&
1003 ws_error_action != nowarn_ws_error)
1004 check_whitespace(line, len, patch->ws_rule);
1005 deleted++;
1006 oldlines--;
1007 trailing = 0;
1008 break;
1009 case '+':
1010 if (!apply_in_reverse &&
1011 ws_error_action != nowarn_ws_error)
1012 check_whitespace(line, len, patch->ws_rule);
1013 added++;
1014 newlines--;
1015 trailing = 0;
1016 break;
1019 * We allow "\ No newline at end of file". Depending
1020 * on locale settings when the patch was produced we
1021 * don't know what this line looks like. The only
1022 * thing we do know is that it begins with "\ ".
1023 * Checking for 12 is just for sanity check -- any
1024 * l10n of "\ No newline..." is at least that long.
1026 case '\\':
1027 if (len < 12 || memcmp(line, "\\ ", 2))
1028 return -1;
1029 break;
1032 if (oldlines || newlines)
1033 return -1;
1034 fragment->leading = leading;
1035 fragment->trailing = trailing;
1038 * If a fragment ends with an incomplete line, we failed to include
1039 * it in the above loop because we hit oldlines == newlines == 0
1040 * before seeing it.
1042 if (12 < size && !memcmp(line, "\\ ", 2))
1043 offset += linelen(line, size);
1045 patch->lines_added += added;
1046 patch->lines_deleted += deleted;
1048 if (0 < patch->is_new && oldlines)
1049 return error("new file depends on old contents");
1050 if (0 < patch->is_delete && newlines)
1051 return error("deleted file still has contents");
1052 return offset;
1055 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1057 unsigned long offset = 0;
1058 unsigned long oldlines = 0, newlines = 0, context = 0;
1059 struct fragment **fragp = &patch->fragments;
1061 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1062 struct fragment *fragment;
1063 int len;
1065 fragment = xcalloc(1, sizeof(*fragment));
1066 len = parse_fragment(line, size, patch, fragment);
1067 if (len <= 0)
1068 die("corrupt patch at line %d", linenr);
1069 fragment->patch = line;
1070 fragment->size = len;
1071 oldlines += fragment->oldlines;
1072 newlines += fragment->newlines;
1073 context += fragment->leading + fragment->trailing;
1075 *fragp = fragment;
1076 fragp = &fragment->next;
1078 offset += len;
1079 line += len;
1080 size -= len;
1084 * If something was removed (i.e. we have old-lines) it cannot
1085 * be creation, and if something was added it cannot be
1086 * deletion. However, the reverse is not true; --unified=0
1087 * patches that only add are not necessarily creation even
1088 * though they do not have any old lines, and ones that only
1089 * delete are not necessarily deletion.
1091 * Unfortunately, a real creation/deletion patch do _not_ have
1092 * any context line by definition, so we cannot safely tell it
1093 * apart with --unified=0 insanity. At least if the patch has
1094 * more than one hunk it is not creation or deletion.
1096 if (patch->is_new < 0 &&
1097 (oldlines || (patch->fragments && patch->fragments->next)))
1098 patch->is_new = 0;
1099 if (patch->is_delete < 0 &&
1100 (newlines || (patch->fragments && patch->fragments->next)))
1101 patch->is_delete = 0;
1102 if (!unidiff_zero || context) {
1103 /* If the user says the patch is not generated with
1104 * --unified=0, or if we have seen context lines,
1105 * then not having oldlines means the patch is creation,
1106 * and not having newlines means the patch is deletion.
1108 if (patch->is_new < 0 && !oldlines) {
1109 patch->is_new = 1;
1110 patch->old_name = NULL;
1112 if (patch->is_delete < 0 && !newlines) {
1113 patch->is_delete = 1;
1114 patch->new_name = NULL;
1118 if (0 < patch->is_new && oldlines)
1119 die("new file %s depends on old contents", patch->new_name);
1120 if (0 < patch->is_delete && newlines)
1121 die("deleted file %s still has contents", patch->old_name);
1122 if (!patch->is_delete && !newlines && context)
1123 fprintf(stderr, "** warning: file %s becomes empty but "
1124 "is not deleted\n", patch->new_name);
1126 return offset;
1129 static inline int metadata_changes(struct patch *patch)
1131 return patch->is_rename > 0 ||
1132 patch->is_copy > 0 ||
1133 patch->is_new > 0 ||
1134 patch->is_delete ||
1135 (patch->old_mode && patch->new_mode &&
1136 patch->old_mode != patch->new_mode);
1139 static char *inflate_it(const void *data, unsigned long size,
1140 unsigned long inflated_size)
1142 z_stream stream;
1143 void *out;
1144 int st;
1146 memset(&stream, 0, sizeof(stream));
1148 stream.next_in = (unsigned char *)data;
1149 stream.avail_in = size;
1150 stream.next_out = out = xmalloc(inflated_size);
1151 stream.avail_out = inflated_size;
1152 inflateInit(&stream);
1153 st = inflate(&stream, Z_FINISH);
1154 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1155 free(out);
1156 return NULL;
1158 return out;
1161 static struct fragment *parse_binary_hunk(char **buf_p,
1162 unsigned long *sz_p,
1163 int *status_p,
1164 int *used_p)
1167 * Expect a line that begins with binary patch method ("literal"
1168 * or "delta"), followed by the length of data before deflating.
1169 * a sequence of 'length-byte' followed by base-85 encoded data
1170 * should follow, terminated by a newline.
1172 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1173 * and we would limit the patch line to 66 characters,
1174 * so one line can fit up to 13 groups that would decode
1175 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1176 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1178 int llen, used;
1179 unsigned long size = *sz_p;
1180 char *buffer = *buf_p;
1181 int patch_method;
1182 unsigned long origlen;
1183 char *data = NULL;
1184 int hunk_size = 0;
1185 struct fragment *frag;
1187 llen = linelen(buffer, size);
1188 used = llen;
1190 *status_p = 0;
1192 if (!prefixcmp(buffer, "delta ")) {
1193 patch_method = BINARY_DELTA_DEFLATED;
1194 origlen = strtoul(buffer + 6, NULL, 10);
1196 else if (!prefixcmp(buffer, "literal ")) {
1197 patch_method = BINARY_LITERAL_DEFLATED;
1198 origlen = strtoul(buffer + 8, NULL, 10);
1200 else
1201 return NULL;
1203 linenr++;
1204 buffer += llen;
1205 while (1) {
1206 int byte_length, max_byte_length, newsize;
1207 llen = linelen(buffer, size);
1208 used += llen;
1209 linenr++;
1210 if (llen == 1) {
1211 /* consume the blank line */
1212 buffer++;
1213 size--;
1214 break;
1217 * Minimum line is "A00000\n" which is 7-byte long,
1218 * and the line length must be multiple of 5 plus 2.
1220 if ((llen < 7) || (llen-2) % 5)
1221 goto corrupt;
1222 max_byte_length = (llen - 2) / 5 * 4;
1223 byte_length = *buffer;
1224 if ('A' <= byte_length && byte_length <= 'Z')
1225 byte_length = byte_length - 'A' + 1;
1226 else if ('a' <= byte_length && byte_length <= 'z')
1227 byte_length = byte_length - 'a' + 27;
1228 else
1229 goto corrupt;
1230 /* if the input length was not multiple of 4, we would
1231 * have filler at the end but the filler should never
1232 * exceed 3 bytes
1234 if (max_byte_length < byte_length ||
1235 byte_length <= max_byte_length - 4)
1236 goto corrupt;
1237 newsize = hunk_size + byte_length;
1238 data = xrealloc(data, newsize);
1239 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1240 goto corrupt;
1241 hunk_size = newsize;
1242 buffer += llen;
1243 size -= llen;
1246 frag = xcalloc(1, sizeof(*frag));
1247 frag->patch = inflate_it(data, hunk_size, origlen);
1248 if (!frag->patch)
1249 goto corrupt;
1250 free(data);
1251 frag->size = origlen;
1252 *buf_p = buffer;
1253 *sz_p = size;
1254 *used_p = used;
1255 frag->binary_patch_method = patch_method;
1256 return frag;
1258 corrupt:
1259 free(data);
1260 *status_p = -1;
1261 error("corrupt binary patch at line %d: %.*s",
1262 linenr-1, llen-1, buffer);
1263 return NULL;
1266 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1269 * We have read "GIT binary patch\n"; what follows is a line
1270 * that says the patch method (currently, either "literal" or
1271 * "delta") and the length of data before deflating; a
1272 * sequence of 'length-byte' followed by base-85 encoded data
1273 * follows.
1275 * When a binary patch is reversible, there is another binary
1276 * hunk in the same format, starting with patch method (either
1277 * "literal" or "delta") with the length of data, and a sequence
1278 * of length-byte + base-85 encoded data, terminated with another
1279 * empty line. This data, when applied to the postimage, produces
1280 * the preimage.
1282 struct fragment *forward;
1283 struct fragment *reverse;
1284 int status;
1285 int used, used_1;
1287 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1288 if (!forward && !status)
1289 /* there has to be one hunk (forward hunk) */
1290 return error("unrecognized binary patch at line %d", linenr-1);
1291 if (status)
1292 /* otherwise we already gave an error message */
1293 return status;
1295 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1296 if (reverse)
1297 used += used_1;
1298 else if (status) {
1300 * Not having reverse hunk is not an error, but having
1301 * a corrupt reverse hunk is.
1303 free((void*) forward->patch);
1304 free(forward);
1305 return status;
1307 forward->next = reverse;
1308 patch->fragments = forward;
1309 patch->is_binary = 1;
1310 return used;
1313 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1315 int hdrsize, patchsize;
1316 int offset = find_header(buffer, size, &hdrsize, patch);
1318 if (offset < 0)
1319 return offset;
1321 patch->ws_rule = whitespace_rule(patch->new_name
1322 ? patch->new_name
1323 : patch->old_name);
1325 patchsize = parse_single_patch(buffer + offset + hdrsize,
1326 size - offset - hdrsize, patch);
1328 if (!patchsize) {
1329 static const char *binhdr[] = {
1330 "Binary files ",
1331 "Files ",
1332 NULL,
1334 static const char git_binary[] = "GIT binary patch\n";
1335 int i;
1336 int hd = hdrsize + offset;
1337 unsigned long llen = linelen(buffer + hd, size - hd);
1339 if (llen == sizeof(git_binary) - 1 &&
1340 !memcmp(git_binary, buffer + hd, llen)) {
1341 int used;
1342 linenr++;
1343 used = parse_binary(buffer + hd + llen,
1344 size - hd - llen, patch);
1345 if (used)
1346 patchsize = used + llen;
1347 else
1348 patchsize = 0;
1350 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1351 for (i = 0; binhdr[i]; i++) {
1352 int len = strlen(binhdr[i]);
1353 if (len < size - hd &&
1354 !memcmp(binhdr[i], buffer + hd, len)) {
1355 linenr++;
1356 patch->is_binary = 1;
1357 patchsize = llen;
1358 break;
1363 /* Empty patch cannot be applied if it is a text patch
1364 * without metadata change. A binary patch appears
1365 * empty to us here.
1367 if ((apply || check) &&
1368 (!patch->is_binary && !metadata_changes(patch)))
1369 die("patch with only garbage at line %d", linenr);
1372 return offset + hdrsize + patchsize;
1375 #define swap(a,b) myswap((a),(b),sizeof(a))
1377 #define myswap(a, b, size) do { \
1378 unsigned char mytmp[size]; \
1379 memcpy(mytmp, &a, size); \
1380 memcpy(&a, &b, size); \
1381 memcpy(&b, mytmp, size); \
1382 } while (0)
1384 static void reverse_patches(struct patch *p)
1386 for (; p; p = p->next) {
1387 struct fragment *frag = p->fragments;
1389 swap(p->new_name, p->old_name);
1390 swap(p->new_mode, p->old_mode);
1391 swap(p->is_new, p->is_delete);
1392 swap(p->lines_added, p->lines_deleted);
1393 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1395 for (; frag; frag = frag->next) {
1396 swap(frag->newpos, frag->oldpos);
1397 swap(frag->newlines, frag->oldlines);
1402 static const char pluses[] =
1403 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1404 static const char minuses[]=
1405 "----------------------------------------------------------------------";
1407 static void show_stats(struct patch *patch)
1409 struct strbuf qname;
1410 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1411 int max, add, del;
1413 strbuf_init(&qname, 0);
1414 quote_c_style(cp, &qname, NULL, 0);
1417 * "scale" the filename
1419 max = max_len;
1420 if (max > 50)
1421 max = 50;
1423 if (qname.len > max) {
1424 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1425 if (!cp)
1426 cp = qname.buf + qname.len + 3 - max;
1427 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1430 if (patch->is_binary) {
1431 printf(" %-*s | Bin\n", max, qname.buf);
1432 strbuf_release(&qname);
1433 return;
1436 printf(" %-*s |", max, qname.buf);
1437 strbuf_release(&qname);
1440 * scale the add/delete
1442 max = max + max_change > 70 ? 70 - max : max_change;
1443 add = patch->lines_added;
1444 del = patch->lines_deleted;
1446 if (max_change > 0) {
1447 int total = ((add + del) * max + max_change / 2) / max_change;
1448 add = (add * max + max_change / 2) / max_change;
1449 del = total - add;
1451 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1452 add, pluses, del, minuses);
1455 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1457 switch (st->st_mode & S_IFMT) {
1458 case S_IFLNK:
1459 strbuf_grow(buf, st->st_size);
1460 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1461 return -1;
1462 strbuf_setlen(buf, st->st_size);
1463 return 0;
1464 case S_IFREG:
1465 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1466 return error("unable to open or read %s", path);
1467 convert_to_git(path, buf->buf, buf->len, buf);
1468 return 0;
1469 default:
1470 return -1;
1474 static int find_offset(const char *buf, unsigned long size,
1475 const char *fragment, unsigned long fragsize,
1476 int line, int *lines)
1478 int i;
1479 unsigned long start, backwards, forwards;
1481 if (fragsize > size)
1482 return -1;
1484 start = 0;
1485 if (line > 1) {
1486 unsigned long offset = 0;
1487 i = line-1;
1488 while (offset + fragsize <= size) {
1489 if (buf[offset++] == '\n') {
1490 start = offset;
1491 if (!--i)
1492 break;
1497 /* Exact line number? */
1498 if ((start + fragsize <= size) &&
1499 !memcmp(buf + start, fragment, fragsize))
1500 return start;
1503 * There's probably some smart way to do this, but I'll leave
1504 * that to the smart and beautiful people. I'm simple and stupid.
1506 backwards = start;
1507 forwards = start;
1508 for (i = 0; ; i++) {
1509 unsigned long try;
1510 int n;
1512 /* "backward" */
1513 if (i & 1) {
1514 if (!backwards) {
1515 if (forwards + fragsize > size)
1516 break;
1517 continue;
1519 do {
1520 --backwards;
1521 } while (backwards && buf[backwards-1] != '\n');
1522 try = backwards;
1523 } else {
1524 while (forwards + fragsize <= size) {
1525 if (buf[forwards++] == '\n')
1526 break;
1528 try = forwards;
1531 if (try + fragsize > size)
1532 continue;
1533 if (memcmp(buf + try, fragment, fragsize))
1534 continue;
1535 n = (i >> 1)+1;
1536 if (i & 1)
1537 n = -n;
1538 *lines = n;
1539 return try;
1543 * We should start searching forward and backward.
1545 return -1;
1548 static void remove_first_line(const char **rbuf, int *rsize)
1550 const char *buf = *rbuf;
1551 int size = *rsize;
1552 unsigned long offset;
1553 offset = 0;
1554 while (offset <= size) {
1555 if (buf[offset++] == '\n')
1556 break;
1558 *rsize = size - offset;
1559 *rbuf = buf + offset;
1562 static void remove_last_line(const char **rbuf, int *rsize)
1564 const char *buf = *rbuf;
1565 int size = *rsize;
1566 unsigned long offset;
1567 offset = size - 1;
1568 while (offset > 0) {
1569 if (buf[--offset] == '\n')
1570 break;
1572 *rsize = offset + 1;
1575 static int apply_line(char *output, const char *patch, int plen,
1576 unsigned ws_rule)
1579 * plen is number of bytes to be copied from patch,
1580 * starting at patch+1 (patch[0] is '+'). Typically
1581 * patch[plen] is '\n', unless this is the incomplete
1582 * last line.
1584 int i;
1585 int add_nl_to_tail = 0;
1586 int fixed = 0;
1587 int last_tab_in_indent = -1;
1588 int last_space_in_indent = -1;
1589 int need_fix_leading_space = 0;
1590 char *buf;
1592 if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1593 *patch != '+') {
1594 memcpy(output, patch + 1, plen);
1595 return plen;
1599 * Strip trailing whitespace
1601 if ((ws_rule & WS_TRAILING_SPACE) &&
1602 (1 < plen && isspace(patch[plen-1]))) {
1603 if (patch[plen] == '\n')
1604 add_nl_to_tail = 1;
1605 plen--;
1606 while (0 < plen && isspace(patch[plen]))
1607 plen--;
1608 fixed = 1;
1612 * Check leading whitespaces (indent)
1614 for (i = 1; i < plen; i++) {
1615 char ch = patch[i];
1616 if (ch == '\t') {
1617 last_tab_in_indent = i;
1618 if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1619 0 <= last_space_in_indent)
1620 need_fix_leading_space = 1;
1621 } else if (ch == ' ') {
1622 last_space_in_indent = i;
1623 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1624 last_tab_in_indent < 0 &&
1625 8 <= i)
1626 need_fix_leading_space = 1;
1628 else
1629 break;
1632 buf = output;
1633 if (need_fix_leading_space) {
1634 int consecutive_spaces = 0;
1635 int last = last_tab_in_indent + 1;
1637 if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1638 /* have "last" point at one past the indent */
1639 if (last_tab_in_indent < last_space_in_indent)
1640 last = last_space_in_indent + 1;
1641 else
1642 last = last_tab_in_indent + 1;
1646 * between patch[1..last], strip the funny spaces,
1647 * updating them to tab as needed.
1649 for (i = 1; i < last; i++, plen--) {
1650 char ch = patch[i];
1651 if (ch != ' ') {
1652 consecutive_spaces = 0;
1653 *output++ = ch;
1654 } else {
1655 consecutive_spaces++;
1656 if (consecutive_spaces == 8) {
1657 *output++ = '\t';
1658 consecutive_spaces = 0;
1662 while (0 < consecutive_spaces--)
1663 *output++ = ' ';
1664 fixed = 1;
1665 i = last;
1667 else
1668 i = 1;
1670 memcpy(output, patch + i, plen);
1671 if (add_nl_to_tail)
1672 output[plen++] = '\n';
1673 if (fixed)
1674 applied_after_fixing_ws++;
1675 return output + plen - buf;
1678 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag,
1679 int inaccurate_eof, unsigned ws_rule)
1681 int match_beginning, match_end;
1682 const char *patch = frag->patch;
1683 int offset, size = frag->size;
1684 char *old = xmalloc(size);
1685 char *new = xmalloc(size);
1686 const char *oldlines, *newlines;
1687 int oldsize = 0, newsize = 0;
1688 int new_blank_lines_at_end = 0;
1689 unsigned long leading, trailing;
1690 int pos, lines;
1692 while (size > 0) {
1693 char first;
1694 int len = linelen(patch, size);
1695 int plen;
1696 int added_blank_line = 0;
1698 if (!len)
1699 break;
1702 * "plen" is how much of the line we should use for
1703 * the actual patch data. Normally we just remove the
1704 * first character on the line, but if the line is
1705 * followed by "\ No newline", then we also remove the
1706 * last one (which is the newline, of course).
1708 plen = len-1;
1709 if (len < size && patch[len] == '\\')
1710 plen--;
1711 first = *patch;
1712 if (apply_in_reverse) {
1713 if (first == '-')
1714 first = '+';
1715 else if (first == '+')
1716 first = '-';
1719 switch (first) {
1720 case '\n':
1721 /* Newer GNU diff, empty context line */
1722 if (plen < 0)
1723 /* ... followed by '\No newline'; nothing */
1724 break;
1725 old[oldsize++] = '\n';
1726 new[newsize++] = '\n';
1727 break;
1728 case ' ':
1729 case '-':
1730 memcpy(old + oldsize, patch + 1, plen);
1731 oldsize += plen;
1732 if (first == '-')
1733 break;
1734 /* Fall-through for ' ' */
1735 case '+':
1736 if (first != '+' || !no_add) {
1737 int added = apply_line(new + newsize, patch,
1738 plen, ws_rule);
1739 newsize += added;
1740 if (first == '+' &&
1741 added == 1 && new[newsize-1] == '\n')
1742 added_blank_line = 1;
1744 break;
1745 case '@': case '\\':
1746 /* Ignore it, we already handled it */
1747 break;
1748 default:
1749 if (apply_verbosely)
1750 error("invalid start of line: '%c'", first);
1751 return -1;
1753 if (added_blank_line)
1754 new_blank_lines_at_end++;
1755 else
1756 new_blank_lines_at_end = 0;
1757 patch += len;
1758 size -= len;
1761 if (inaccurate_eof &&
1762 oldsize > 0 && old[oldsize - 1] == '\n' &&
1763 newsize > 0 && new[newsize - 1] == '\n') {
1764 oldsize--;
1765 newsize--;
1768 oldlines = old;
1769 newlines = new;
1770 leading = frag->leading;
1771 trailing = frag->trailing;
1774 * If we don't have any leading/trailing data in the patch,
1775 * we want it to match at the beginning/end of the file.
1777 * But that would break if the patch is generated with
1778 * --unified=0; sane people wouldn't do that to cause us
1779 * trouble, but we try to please not so sane ones as well.
1781 if (unidiff_zero) {
1782 match_beginning = (!leading && !frag->oldpos);
1783 match_end = 0;
1785 else {
1786 match_beginning = !leading && (frag->oldpos == 1);
1787 match_end = !trailing;
1790 lines = 0;
1791 pos = frag->newpos;
1792 for (;;) {
1793 offset = find_offset(buf->buf, buf->len,
1794 oldlines, oldsize, pos, &lines);
1795 if (match_end && offset + oldsize != buf->len)
1796 offset = -1;
1797 if (match_beginning && offset)
1798 offset = -1;
1799 if (offset >= 0) {
1800 if (ws_error_action == correct_ws_error &&
1801 (buf->len - oldsize - offset == 0)) /* end of file? */
1802 newsize -= new_blank_lines_at_end;
1804 /* Warn if it was necessary to reduce the number
1805 * of context lines.
1807 if ((leading != frag->leading) ||
1808 (trailing != frag->trailing))
1809 fprintf(stderr, "Context reduced to (%ld/%ld)"
1810 " to apply fragment at %d\n",
1811 leading, trailing, pos + lines);
1813 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1814 offset = 0;
1815 break;
1818 /* Am I at my context limits? */
1819 if ((leading <= p_context) && (trailing <= p_context))
1820 break;
1821 if (match_beginning || match_end) {
1822 match_beginning = match_end = 0;
1823 continue;
1826 * Reduce the number of context lines; reduce both
1827 * leading and trailing if they are equal otherwise
1828 * just reduce the larger context.
1830 if (leading >= trailing) {
1831 remove_first_line(&oldlines, &oldsize);
1832 remove_first_line(&newlines, &newsize);
1833 pos--;
1834 leading--;
1836 if (trailing > leading) {
1837 remove_last_line(&oldlines, &oldsize);
1838 remove_last_line(&newlines, &newsize);
1839 trailing--;
1843 if (offset && apply_verbosely)
1844 error("while searching for:\n%.*s", oldsize, oldlines);
1846 free(old);
1847 free(new);
1848 return offset;
1851 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1853 struct fragment *fragment = patch->fragments;
1854 unsigned long len;
1855 void *dst;
1857 /* Binary patch is irreversible without the optional second hunk */
1858 if (apply_in_reverse) {
1859 if (!fragment->next)
1860 return error("cannot reverse-apply a binary patch "
1861 "without the reverse hunk to '%s'",
1862 patch->new_name
1863 ? patch->new_name : patch->old_name);
1864 fragment = fragment->next;
1866 switch (fragment->binary_patch_method) {
1867 case BINARY_DELTA_DEFLATED:
1868 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1869 fragment->size, &len);
1870 if (!dst)
1871 return -1;
1872 /* XXX patch_delta NUL-terminates */
1873 strbuf_attach(buf, dst, len, len + 1);
1874 return 0;
1875 case BINARY_LITERAL_DEFLATED:
1876 strbuf_reset(buf);
1877 strbuf_add(buf, fragment->patch, fragment->size);
1878 return 0;
1880 return -1;
1883 static int apply_binary(struct strbuf *buf, struct patch *patch)
1885 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1886 unsigned char sha1[20];
1889 * For safety, we require patch index line to contain
1890 * full 40-byte textual SHA1 for old and new, at least for now.
1892 if (strlen(patch->old_sha1_prefix) != 40 ||
1893 strlen(patch->new_sha1_prefix) != 40 ||
1894 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1895 get_sha1_hex(patch->new_sha1_prefix, sha1))
1896 return error("cannot apply binary patch to '%s' "
1897 "without full index line", name);
1899 if (patch->old_name) {
1901 * See if the old one matches what the patch
1902 * applies to.
1904 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1905 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1906 return error("the patch applies to '%s' (%s), "
1907 "which does not match the "
1908 "current contents.",
1909 name, sha1_to_hex(sha1));
1911 else {
1912 /* Otherwise, the old one must be empty. */
1913 if (buf->len)
1914 return error("the patch applies to an empty "
1915 "'%s' but it is not empty", name);
1918 get_sha1_hex(patch->new_sha1_prefix, sha1);
1919 if (is_null_sha1(sha1)) {
1920 strbuf_release(buf);
1921 return 0; /* deletion patch */
1924 if (has_sha1_file(sha1)) {
1925 /* We already have the postimage */
1926 enum object_type type;
1927 unsigned long size;
1928 char *result;
1930 result = read_sha1_file(sha1, &type, &size);
1931 if (!result)
1932 return error("the necessary postimage %s for "
1933 "'%s' cannot be read",
1934 patch->new_sha1_prefix, name);
1935 /* XXX read_sha1_file NUL-terminates */
1936 strbuf_attach(buf, result, size, size + 1);
1937 } else {
1939 * We have verified buf matches the preimage;
1940 * apply the patch data to it, which is stored
1941 * in the patch->fragments->{patch,size}.
1943 if (apply_binary_fragment(buf, patch))
1944 return error("binary patch does not apply to '%s'",
1945 name);
1947 /* verify that the result matches */
1948 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1949 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1950 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1951 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1954 return 0;
1957 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1959 struct fragment *frag = patch->fragments;
1960 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1961 unsigned ws_rule = patch->ws_rule;
1962 unsigned inaccurate_eof = patch->inaccurate_eof;
1964 if (patch->is_binary)
1965 return apply_binary(buf, patch);
1967 while (frag) {
1968 if (apply_one_fragment(buf, frag, inaccurate_eof, ws_rule)) {
1969 error("patch failed: %s:%ld", name, frag->oldpos);
1970 if (!apply_with_reject)
1971 return -1;
1972 frag->rejected = 1;
1974 frag = frag->next;
1976 return 0;
1979 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1981 if (!ce)
1982 return 0;
1984 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1985 strbuf_grow(buf, 100);
1986 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1987 } else {
1988 enum object_type type;
1989 unsigned long sz;
1990 char *result;
1992 result = read_sha1_file(ce->sha1, &type, &sz);
1993 if (!result)
1994 return -1;
1995 /* XXX read_sha1_file NUL-terminates */
1996 strbuf_attach(buf, result, sz, sz + 1);
1998 return 0;
2001 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2003 struct strbuf buf;
2005 strbuf_init(&buf, 0);
2006 if (cached) {
2007 if (read_file_or_gitlink(ce, &buf))
2008 return error("read of %s failed", patch->old_name);
2009 } else if (patch->old_name) {
2010 if (S_ISGITLINK(patch->old_mode)) {
2011 if (ce) {
2012 read_file_or_gitlink(ce, &buf);
2013 } else {
2015 * There is no way to apply subproject
2016 * patch without looking at the index.
2018 patch->fragments = NULL;
2020 } else {
2021 if (read_old_data(st, patch->old_name, &buf))
2022 return error("read of %s failed", patch->old_name);
2026 if (apply_fragments(&buf, patch) < 0)
2027 return -1; /* note with --reject this succeeds. */
2028 patch->result = strbuf_detach(&buf, &patch->resultsize);
2030 if (0 < patch->is_delete && patch->resultsize)
2031 return error("removal patch leaves file contents");
2033 return 0;
2036 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2038 struct stat nst;
2039 if (!lstat(new_name, &nst)) {
2040 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2041 return 0;
2043 * A leading component of new_name might be a symlink
2044 * that is going to be removed with this patch, but
2045 * still pointing at somewhere that has the path.
2046 * In such a case, path "new_name" does not exist as
2047 * far as git is concerned.
2049 if (has_symlink_leading_path(new_name, NULL))
2050 return 0;
2052 return error("%s: already exists in working directory", new_name);
2054 else if ((errno != ENOENT) && (errno != ENOTDIR))
2055 return error("%s: %s", new_name, strerror(errno));
2056 return 0;
2059 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2061 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2062 if (!S_ISDIR(st->st_mode))
2063 return -1;
2064 return 0;
2066 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2069 static int check_patch(struct patch *patch, struct patch *prev_patch)
2071 struct stat st;
2072 const char *old_name = patch->old_name;
2073 const char *new_name = patch->new_name;
2074 const char *name = old_name ? old_name : new_name;
2075 struct cache_entry *ce = NULL;
2076 int ok_if_exists;
2078 patch->rejected = 1; /* we will drop this after we succeed */
2081 * Make sure that we do not have local modifications from the
2082 * index when we are looking at the index. Also make sure
2083 * we have the preimage file to be patched in the work tree,
2084 * unless --cached, which tells git to apply only in the index.
2086 if (old_name) {
2087 int stat_ret = 0;
2088 unsigned st_mode = 0;
2090 if (!cached)
2091 stat_ret = lstat(old_name, &st);
2092 if (check_index) {
2093 int pos = cache_name_pos(old_name, strlen(old_name));
2094 if (pos < 0)
2095 return error("%s: does not exist in index",
2096 old_name);
2097 ce = active_cache[pos];
2098 if (stat_ret < 0) {
2099 struct checkout costate;
2100 if (errno != ENOENT)
2101 return error("%s: %s", old_name,
2102 strerror(errno));
2103 /* checkout */
2104 costate.base_dir = "";
2105 costate.base_dir_len = 0;
2106 costate.force = 0;
2107 costate.quiet = 0;
2108 costate.not_new = 0;
2109 costate.refresh_cache = 1;
2110 if (checkout_entry(ce,
2111 &costate,
2112 NULL) ||
2113 lstat(old_name, &st))
2114 return -1;
2116 if (!cached && verify_index_match(ce, &st))
2117 return error("%s: does not match index",
2118 old_name);
2119 if (cached)
2120 st_mode = ntohl(ce->ce_mode);
2121 } else if (stat_ret < 0)
2122 return error("%s: %s", old_name, strerror(errno));
2124 if (!cached)
2125 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2127 if (patch->is_new < 0)
2128 patch->is_new = 0;
2129 if (!patch->old_mode)
2130 patch->old_mode = st_mode;
2131 if ((st_mode ^ patch->old_mode) & S_IFMT)
2132 return error("%s: wrong type", old_name);
2133 if (st_mode != patch->old_mode)
2134 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2135 old_name, st_mode, patch->old_mode);
2138 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2139 !strcmp(prev_patch->old_name, new_name))
2141 * A type-change diff is always split into a patch to
2142 * delete old, immediately followed by a patch to
2143 * create new (see diff.c::run_diff()); in such a case
2144 * it is Ok that the entry to be deleted by the
2145 * previous patch is still in the working tree and in
2146 * the index.
2148 ok_if_exists = 1;
2149 else
2150 ok_if_exists = 0;
2152 if (new_name &&
2153 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2154 if (check_index &&
2155 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2156 !ok_if_exists)
2157 return error("%s: already exists in index", new_name);
2158 if (!cached) {
2159 int err = check_to_create_blob(new_name, ok_if_exists);
2160 if (err)
2161 return err;
2163 if (!patch->new_mode) {
2164 if (0 < patch->is_new)
2165 patch->new_mode = S_IFREG | 0644;
2166 else
2167 patch->new_mode = patch->old_mode;
2171 if (new_name && old_name) {
2172 int same = !strcmp(old_name, new_name);
2173 if (!patch->new_mode)
2174 patch->new_mode = patch->old_mode;
2175 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2176 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2177 patch->new_mode, new_name, patch->old_mode,
2178 same ? "" : " of ", same ? "" : old_name);
2181 if (apply_data(patch, &st, ce) < 0)
2182 return error("%s: patch does not apply", name);
2183 patch->rejected = 0;
2184 return 0;
2187 static int check_patch_list(struct patch *patch)
2189 struct patch *prev_patch = NULL;
2190 int err = 0;
2192 for (prev_patch = NULL; patch ; patch = patch->next) {
2193 if (apply_verbosely)
2194 say_patch_name(stderr,
2195 "Checking patch ", patch, "...\n");
2196 err |= check_patch(patch, prev_patch);
2197 prev_patch = patch;
2199 return err;
2202 /* This function tries to read the sha1 from the current index */
2203 static int get_current_sha1(const char *path, unsigned char *sha1)
2205 int pos;
2207 if (read_cache() < 0)
2208 return -1;
2209 pos = cache_name_pos(path, strlen(path));
2210 if (pos < 0)
2211 return -1;
2212 hashcpy(sha1, active_cache[pos]->sha1);
2213 return 0;
2216 /* Build an index that contains the just the files needed for a 3way merge */
2217 static void build_fake_ancestor(struct patch *list, const char *filename)
2219 struct patch *patch;
2220 struct index_state result = { 0 };
2221 int fd;
2223 /* Once we start supporting the reverse patch, it may be
2224 * worth showing the new sha1 prefix, but until then...
2226 for (patch = list; patch; patch = patch->next) {
2227 const unsigned char *sha1_ptr;
2228 unsigned char sha1[20];
2229 struct cache_entry *ce;
2230 const char *name;
2232 name = patch->old_name ? patch->old_name : patch->new_name;
2233 if (0 < patch->is_new)
2234 continue;
2235 else if (get_sha1(patch->old_sha1_prefix, sha1))
2236 /* git diff has no index line for mode/type changes */
2237 if (!patch->lines_added && !patch->lines_deleted) {
2238 if (get_current_sha1(patch->new_name, sha1) ||
2239 get_current_sha1(patch->old_name, sha1))
2240 die("mode change for %s, which is not "
2241 "in current HEAD", name);
2242 sha1_ptr = sha1;
2243 } else
2244 die("sha1 information is lacking or useless "
2245 "(%s).", name);
2246 else
2247 sha1_ptr = sha1;
2249 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2250 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2251 die ("Could not add %s to temporary index", name);
2254 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2255 if (fd < 0 || write_index(&result, fd) || close(fd))
2256 die ("Could not write temporary index to %s", filename);
2258 discard_index(&result);
2261 static void stat_patch_list(struct patch *patch)
2263 int files, adds, dels;
2265 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2266 files++;
2267 adds += patch->lines_added;
2268 dels += patch->lines_deleted;
2269 show_stats(patch);
2272 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2275 static void numstat_patch_list(struct patch *patch)
2277 for ( ; patch; patch = patch->next) {
2278 const char *name;
2279 name = patch->new_name ? patch->new_name : patch->old_name;
2280 if (patch->is_binary)
2281 printf("-\t-\t");
2282 else
2283 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2284 write_name_quoted(name, stdout, line_termination);
2288 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2290 if (mode)
2291 printf(" %s mode %06o %s\n", newdelete, mode, name);
2292 else
2293 printf(" %s %s\n", newdelete, name);
2296 static void show_mode_change(struct patch *p, int show_name)
2298 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2299 if (show_name)
2300 printf(" mode change %06o => %06o %s\n",
2301 p->old_mode, p->new_mode, p->new_name);
2302 else
2303 printf(" mode change %06o => %06o\n",
2304 p->old_mode, p->new_mode);
2308 static void show_rename_copy(struct patch *p)
2310 const char *renamecopy = p->is_rename ? "rename" : "copy";
2311 const char *old, *new;
2313 /* Find common prefix */
2314 old = p->old_name;
2315 new = p->new_name;
2316 while (1) {
2317 const char *slash_old, *slash_new;
2318 slash_old = strchr(old, '/');
2319 slash_new = strchr(new, '/');
2320 if (!slash_old ||
2321 !slash_new ||
2322 slash_old - old != slash_new - new ||
2323 memcmp(old, new, slash_new - new))
2324 break;
2325 old = slash_old + 1;
2326 new = slash_new + 1;
2328 /* p->old_name thru old is the common prefix, and old and new
2329 * through the end of names are renames
2331 if (old != p->old_name)
2332 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2333 (int)(old - p->old_name), p->old_name,
2334 old, new, p->score);
2335 else
2336 printf(" %s %s => %s (%d%%)\n", renamecopy,
2337 p->old_name, p->new_name, p->score);
2338 show_mode_change(p, 0);
2341 static void summary_patch_list(struct patch *patch)
2343 struct patch *p;
2345 for (p = patch; p; p = p->next) {
2346 if (p->is_new)
2347 show_file_mode_name("create", p->new_mode, p->new_name);
2348 else if (p->is_delete)
2349 show_file_mode_name("delete", p->old_mode, p->old_name);
2350 else {
2351 if (p->is_rename || p->is_copy)
2352 show_rename_copy(p);
2353 else {
2354 if (p->score) {
2355 printf(" rewrite %s (%d%%)\n",
2356 p->new_name, p->score);
2357 show_mode_change(p, 0);
2359 else
2360 show_mode_change(p, 1);
2366 static void patch_stats(struct patch *patch)
2368 int lines = patch->lines_added + patch->lines_deleted;
2370 if (lines > max_change)
2371 max_change = lines;
2372 if (patch->old_name) {
2373 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2374 if (!len)
2375 len = strlen(patch->old_name);
2376 if (len > max_len)
2377 max_len = len;
2379 if (patch->new_name) {
2380 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2381 if (!len)
2382 len = strlen(patch->new_name);
2383 if (len > max_len)
2384 max_len = len;
2388 static void remove_file(struct patch *patch, int rmdir_empty)
2390 if (update_index) {
2391 if (remove_file_from_cache(patch->old_name) < 0)
2392 die("unable to remove %s from index", patch->old_name);
2394 if (!cached) {
2395 if (S_ISGITLINK(patch->old_mode)) {
2396 if (rmdir(patch->old_name))
2397 warning("unable to remove submodule %s",
2398 patch->old_name);
2399 } else if (!unlink(patch->old_name) && rmdir_empty) {
2400 char *name = xstrdup(patch->old_name);
2401 char *end = strrchr(name, '/');
2402 while (end) {
2403 *end = 0;
2404 if (rmdir(name))
2405 break;
2406 end = strrchr(name, '/');
2408 free(name);
2413 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2415 struct stat st;
2416 struct cache_entry *ce;
2417 int namelen = strlen(path);
2418 unsigned ce_size = cache_entry_size(namelen);
2420 if (!update_index)
2421 return;
2423 ce = xcalloc(1, ce_size);
2424 memcpy(ce->name, path, namelen);
2425 ce->ce_mode = create_ce_mode(mode);
2426 ce->ce_flags = htons(namelen);
2427 if (S_ISGITLINK(mode)) {
2428 const char *s = buf;
2430 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2431 die("corrupt patch for subproject %s", path);
2432 } else {
2433 if (!cached) {
2434 if (lstat(path, &st) < 0)
2435 die("unable to stat newly created file %s",
2436 path);
2437 fill_stat_cache_info(ce, &st);
2439 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2440 die("unable to create backing store for newly created file %s", path);
2442 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2443 die("unable to add cache entry for %s", path);
2446 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2448 int fd;
2449 struct strbuf nbuf;
2451 if (S_ISGITLINK(mode)) {
2452 struct stat st;
2453 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2454 return 0;
2455 return mkdir(path, 0777);
2458 if (has_symlinks && S_ISLNK(mode))
2459 /* Although buf:size is counted string, it also is NUL
2460 * terminated.
2462 return symlink(buf, path);
2464 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2465 if (fd < 0)
2466 return -1;
2468 strbuf_init(&nbuf, 0);
2469 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2470 size = nbuf.len;
2471 buf = nbuf.buf;
2473 write_or_die(fd, buf, size);
2474 strbuf_release(&nbuf);
2476 if (close(fd) < 0)
2477 die("closing file %s: %s", path, strerror(errno));
2478 return 0;
2482 * We optimistically assume that the directories exist,
2483 * which is true 99% of the time anyway. If they don't,
2484 * we create them and try again.
2486 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2488 if (cached)
2489 return;
2490 if (!try_create_file(path, mode, buf, size))
2491 return;
2493 if (errno == ENOENT) {
2494 if (safe_create_leading_directories(path))
2495 return;
2496 if (!try_create_file(path, mode, buf, size))
2497 return;
2500 if (errno == EEXIST || errno == EACCES) {
2501 /* We may be trying to create a file where a directory
2502 * used to be.
2504 struct stat st;
2505 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2506 errno = EEXIST;
2509 if (errno == EEXIST) {
2510 unsigned int nr = getpid();
2512 for (;;) {
2513 const char *newpath;
2514 newpath = mkpath("%s~%u", path, nr);
2515 if (!try_create_file(newpath, mode, buf, size)) {
2516 if (!rename(newpath, path))
2517 return;
2518 unlink(newpath);
2519 break;
2521 if (errno != EEXIST)
2522 break;
2523 ++nr;
2526 die("unable to write file %s mode %o", path, mode);
2529 static void create_file(struct patch *patch)
2531 char *path = patch->new_name;
2532 unsigned mode = patch->new_mode;
2533 unsigned long size = patch->resultsize;
2534 char *buf = patch->result;
2536 if (!mode)
2537 mode = S_IFREG | 0644;
2538 create_one_file(path, mode, buf, size);
2539 add_index_file(path, mode, buf, size);
2542 /* phase zero is to remove, phase one is to create */
2543 static void write_out_one_result(struct patch *patch, int phase)
2545 if (patch->is_delete > 0) {
2546 if (phase == 0)
2547 remove_file(patch, 1);
2548 return;
2550 if (patch->is_new > 0 || patch->is_copy) {
2551 if (phase == 1)
2552 create_file(patch);
2553 return;
2556 * Rename or modification boils down to the same
2557 * thing: remove the old, write the new
2559 if (phase == 0)
2560 remove_file(patch, patch->is_rename);
2561 if (phase == 1)
2562 create_file(patch);
2565 static int write_out_one_reject(struct patch *patch)
2567 FILE *rej;
2568 char namebuf[PATH_MAX];
2569 struct fragment *frag;
2570 int cnt = 0;
2572 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2573 if (!frag->rejected)
2574 continue;
2575 cnt++;
2578 if (!cnt) {
2579 if (apply_verbosely)
2580 say_patch_name(stderr,
2581 "Applied patch ", patch, " cleanly.\n");
2582 return 0;
2585 /* This should not happen, because a removal patch that leaves
2586 * contents are marked "rejected" at the patch level.
2588 if (!patch->new_name)
2589 die("internal error");
2591 /* Say this even without --verbose */
2592 say_patch_name(stderr, "Applying patch ", patch, " with");
2593 fprintf(stderr, " %d rejects...\n", cnt);
2595 cnt = strlen(patch->new_name);
2596 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2597 cnt = ARRAY_SIZE(namebuf) - 5;
2598 fprintf(stderr,
2599 "warning: truncating .rej filename to %.*s.rej",
2600 cnt - 1, patch->new_name);
2602 memcpy(namebuf, patch->new_name, cnt);
2603 memcpy(namebuf + cnt, ".rej", 5);
2605 rej = fopen(namebuf, "w");
2606 if (!rej)
2607 return error("cannot open %s: %s", namebuf, strerror(errno));
2609 /* Normal git tools never deal with .rej, so do not pretend
2610 * this is a git patch by saying --git nor give extended
2611 * headers. While at it, maybe please "kompare" that wants
2612 * the trailing TAB and some garbage at the end of line ;-).
2614 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2615 patch->new_name, patch->new_name);
2616 for (cnt = 1, frag = patch->fragments;
2617 frag;
2618 cnt++, frag = frag->next) {
2619 if (!frag->rejected) {
2620 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2621 continue;
2623 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2624 fprintf(rej, "%.*s", frag->size, frag->patch);
2625 if (frag->patch[frag->size-1] != '\n')
2626 fputc('\n', rej);
2628 fclose(rej);
2629 return -1;
2632 static int write_out_results(struct patch *list, int skipped_patch)
2634 int phase;
2635 int errs = 0;
2636 struct patch *l;
2638 if (!list && !skipped_patch)
2639 return error("No changes");
2641 for (phase = 0; phase < 2; phase++) {
2642 l = list;
2643 while (l) {
2644 if (l->rejected)
2645 errs = 1;
2646 else {
2647 write_out_one_result(l, phase);
2648 if (phase == 1 && write_out_one_reject(l))
2649 errs = 1;
2651 l = l->next;
2654 return errs;
2657 static struct lock_file lock_file;
2659 static struct excludes {
2660 struct excludes *next;
2661 const char *path;
2662 } *excludes;
2664 static int use_patch(struct patch *p)
2666 const char *pathname = p->new_name ? p->new_name : p->old_name;
2667 struct excludes *x = excludes;
2668 while (x) {
2669 if (fnmatch(x->path, pathname, 0) == 0)
2670 return 0;
2671 x = x->next;
2673 if (0 < prefix_length) {
2674 int pathlen = strlen(pathname);
2675 if (pathlen <= prefix_length ||
2676 memcmp(prefix, pathname, prefix_length))
2677 return 0;
2679 return 1;
2682 static void prefix_one(char **name)
2684 char *old_name = *name;
2685 if (!old_name)
2686 return;
2687 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2688 free(old_name);
2691 static void prefix_patches(struct patch *p)
2693 if (!prefix || p->is_toplevel_relative)
2694 return;
2695 for ( ; p; p = p->next) {
2696 if (p->new_name == p->old_name) {
2697 char *prefixed = p->new_name;
2698 prefix_one(&prefixed);
2699 p->new_name = p->old_name = prefixed;
2701 else {
2702 prefix_one(&p->new_name);
2703 prefix_one(&p->old_name);
2708 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2710 size_t offset;
2711 struct strbuf buf;
2712 struct patch *list = NULL, **listp = &list;
2713 int skipped_patch = 0;
2715 strbuf_init(&buf, 0);
2716 patch_input_file = filename;
2717 read_patch_file(&buf, fd);
2718 offset = 0;
2719 while (offset < buf.len) {
2720 struct patch *patch;
2721 int nr;
2723 patch = xcalloc(1, sizeof(*patch));
2724 patch->inaccurate_eof = inaccurate_eof;
2725 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2726 if (nr < 0)
2727 break;
2728 if (apply_in_reverse)
2729 reverse_patches(patch);
2730 if (prefix)
2731 prefix_patches(patch);
2732 if (use_patch(patch)) {
2733 patch_stats(patch);
2734 *listp = patch;
2735 listp = &patch->next;
2737 else {
2738 /* perhaps free it a bit better? */
2739 free(patch);
2740 skipped_patch++;
2742 offset += nr;
2745 if (whitespace_error && (ws_error_action == die_on_ws_error))
2746 apply = 0;
2748 update_index = check_index && apply;
2749 if (update_index && newfd < 0)
2750 newfd = hold_locked_index(&lock_file, 1);
2752 if (check_index) {
2753 if (read_cache() < 0)
2754 die("unable to read index file");
2757 if ((check || apply) &&
2758 check_patch_list(list) < 0 &&
2759 !apply_with_reject)
2760 exit(1);
2762 if (apply && write_out_results(list, skipped_patch))
2763 exit(1);
2765 if (fake_ancestor)
2766 build_fake_ancestor(list, fake_ancestor);
2768 if (diffstat)
2769 stat_patch_list(list);
2771 if (numstat)
2772 numstat_patch_list(list);
2774 if (summary)
2775 summary_patch_list(list);
2777 strbuf_release(&buf);
2778 return 0;
2781 static int git_apply_config(const char *var, const char *value)
2783 if (!strcmp(var, "apply.whitespace")) {
2784 apply_default_whitespace = xstrdup(value);
2785 return 0;
2787 return git_default_config(var, value);
2791 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2793 int i;
2794 int read_stdin = 1;
2795 int inaccurate_eof = 0;
2796 int errs = 0;
2797 int is_not_gitdir = 0;
2799 const char *whitespace_option = NULL;
2801 prefix = setup_git_directory_gently(&is_not_gitdir);
2802 prefix_length = prefix ? strlen(prefix) : 0;
2803 git_config(git_apply_config);
2804 if (apply_default_whitespace)
2805 parse_whitespace_option(apply_default_whitespace);
2807 for (i = 1; i < argc; i++) {
2808 const char *arg = argv[i];
2809 char *end;
2810 int fd;
2812 if (!strcmp(arg, "-")) {
2813 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2814 read_stdin = 0;
2815 continue;
2817 if (!prefixcmp(arg, "--exclude=")) {
2818 struct excludes *x = xmalloc(sizeof(*x));
2819 x->path = arg + 10;
2820 x->next = excludes;
2821 excludes = x;
2822 continue;
2824 if (!prefixcmp(arg, "-p")) {
2825 p_value = atoi(arg + 2);
2826 p_value_known = 1;
2827 continue;
2829 if (!strcmp(arg, "--no-add")) {
2830 no_add = 1;
2831 continue;
2833 if (!strcmp(arg, "--stat")) {
2834 apply = 0;
2835 diffstat = 1;
2836 continue;
2838 if (!strcmp(arg, "--allow-binary-replacement") ||
2839 !strcmp(arg, "--binary")) {
2840 continue; /* now no-op */
2842 if (!strcmp(arg, "--numstat")) {
2843 apply = 0;
2844 numstat = 1;
2845 continue;
2847 if (!strcmp(arg, "--summary")) {
2848 apply = 0;
2849 summary = 1;
2850 continue;
2852 if (!strcmp(arg, "--check")) {
2853 apply = 0;
2854 check = 1;
2855 continue;
2857 if (!strcmp(arg, "--index")) {
2858 if (is_not_gitdir)
2859 die("--index outside a repository");
2860 check_index = 1;
2861 continue;
2863 if (!strcmp(arg, "--cached")) {
2864 if (is_not_gitdir)
2865 die("--cached outside a repository");
2866 check_index = 1;
2867 cached = 1;
2868 continue;
2870 if (!strcmp(arg, "--apply")) {
2871 apply = 1;
2872 continue;
2874 if (!strcmp(arg, "--build-fake-ancestor")) {
2875 apply = 0;
2876 if (++i >= argc)
2877 die ("need a filename");
2878 fake_ancestor = argv[i];
2879 continue;
2881 if (!strcmp(arg, "-z")) {
2882 line_termination = 0;
2883 continue;
2885 if (!prefixcmp(arg, "-C")) {
2886 p_context = strtoul(arg + 2, &end, 0);
2887 if (*end != '\0')
2888 die("unrecognized context count '%s'", arg + 2);
2889 continue;
2891 if (!prefixcmp(arg, "--whitespace=")) {
2892 whitespace_option = arg + 13;
2893 parse_whitespace_option(arg + 13);
2894 continue;
2896 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2897 apply_in_reverse = 1;
2898 continue;
2900 if (!strcmp(arg, "--unidiff-zero")) {
2901 unidiff_zero = 1;
2902 continue;
2904 if (!strcmp(arg, "--reject")) {
2905 apply = apply_with_reject = apply_verbosely = 1;
2906 continue;
2908 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2909 apply_verbosely = 1;
2910 continue;
2912 if (!strcmp(arg, "--inaccurate-eof")) {
2913 inaccurate_eof = 1;
2914 continue;
2916 if (0 < prefix_length)
2917 arg = prefix_filename(prefix, prefix_length, arg);
2919 fd = open(arg, O_RDONLY);
2920 if (fd < 0)
2921 usage(apply_usage);
2922 read_stdin = 0;
2923 set_default_whitespace_mode(whitespace_option);
2924 errs |= apply_patch(fd, arg, inaccurate_eof);
2925 close(fd);
2927 set_default_whitespace_mode(whitespace_option);
2928 if (read_stdin)
2929 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2930 if (whitespace_error) {
2931 if (squelch_whitespace_errors &&
2932 squelch_whitespace_errors < whitespace_error) {
2933 int squelched =
2934 whitespace_error - squelch_whitespace_errors;
2935 fprintf(stderr, "warning: squelched %d "
2936 "whitespace error%s\n",
2937 squelched,
2938 squelched == 1 ? "" : "s");
2940 if (ws_error_action == die_on_ws_error)
2941 die("%d line%s add%s whitespace errors.",
2942 whitespace_error,
2943 whitespace_error == 1 ? "" : "s",
2944 whitespace_error == 1 ? "s" : "");
2945 if (applied_after_fixing_ws)
2946 fprintf(stderr, "warning: %d line%s applied after"
2947 " fixing whitespace errors.\n",
2948 applied_after_fixing_ws,
2949 applied_after_fixing_ws == 1 ? "" : "s");
2950 else if (whitespace_error)
2951 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2952 whitespace_error,
2953 whitespace_error == 1 ? "" : "s",
2954 whitespace_error == 1 ? "s" : "");
2957 if (update_index) {
2958 if (write_cache(newfd, active_cache, active_nr) ||
2959 close(newfd) || commit_locked_index(&lock_file))
2960 die("Unable to write new index file");
2963 return !!errs;