apply --reverse: tie it all together.
[alt-git.git] / builtin-apply.c
blob4737b64c323b6bea1f84932ba80d344ab9997f13
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 <fnmatch.h>
10 #include "cache.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "builtin.h"
18 * --check turns on checking that the working tree matches the
19 * files that are being modified, but doesn't apply the patch
20 * --stat does just a diffstat, and doesn't actually apply
21 * --numstat does numeric diffstat, and doesn't actually apply
22 * --index-info shows the old and new index info for paths if available.
23 * --index updates the cache as well.
24 * --cached updates only the cache without ever touching the working tree.
26 static const char *prefix;
27 static int prefix_length = -1;
28 static int newfd = -1;
30 static int p_value = 1;
31 static int allow_binary_replacement;
32 static int check_index;
33 static int write_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 no_add;
42 static int show_index_info;
43 static int line_termination = '\n';
44 static unsigned long p_context = -1;
45 static const char apply_usage[] =
46 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
48 static enum whitespace_eol {
49 nowarn_whitespace,
50 warn_on_whitespace,
51 error_on_whitespace,
52 strip_whitespace,
53 } new_whitespace = warn_on_whitespace;
54 static int whitespace_error;
55 static int squelch_whitespace_errors = 5;
56 static int applied_after_stripping;
57 static const char *patch_input_file;
59 static void parse_whitespace_option(const char *option)
61 if (!option) {
62 new_whitespace = warn_on_whitespace;
63 return;
65 if (!strcmp(option, "warn")) {
66 new_whitespace = warn_on_whitespace;
67 return;
69 if (!strcmp(option, "nowarn")) {
70 new_whitespace = nowarn_whitespace;
71 return;
73 if (!strcmp(option, "error")) {
74 new_whitespace = error_on_whitespace;
75 return;
77 if (!strcmp(option, "error-all")) {
78 new_whitespace = error_on_whitespace;
79 squelch_whitespace_errors = 0;
80 return;
82 if (!strcmp(option, "strip")) {
83 new_whitespace = strip_whitespace;
84 return;
86 die("unrecognized whitespace option '%s'", option);
89 static void set_default_whitespace_mode(const char *whitespace_option)
91 if (!whitespace_option && !apply_default_whitespace) {
92 new_whitespace = (apply
93 ? warn_on_whitespace
94 : nowarn_whitespace);
99 * For "diff-stat" like behaviour, we keep track of the biggest change
100 * we've seen, and the longest filename. That allows us to do simple
101 * scaling.
103 static int max_change, max_len;
106 * Various "current state", notably line numbers and what
107 * file (and how) we're patching right now.. The "is_xxxx"
108 * things are flags, where -1 means "don't know yet".
110 static int linenr = 1;
113 * This represents one "hunk" from a patch, starting with
114 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
115 * patch text is pointed at by patch, and its byte length
116 * is stored in size. leading and trailing are the number
117 * of context lines.
119 struct fragment {
120 unsigned long leading, trailing;
121 unsigned long oldpos, oldlines;
122 unsigned long newpos, newlines;
123 const char *patch;
124 int size;
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
137 struct patch {
138 char *new_name, *old_name, *def_name;
139 unsigned int old_mode, new_mode;
140 int is_rename, is_copy, is_new, is_delete, is_binary;
141 unsigned long deflate_origlen;
142 int lines_added, lines_deleted;
143 int score;
144 int inaccurate_eof:1;
145 struct fragment *fragments;
146 char *result;
147 unsigned long resultsize;
148 char old_sha1_prefix[41];
149 char new_sha1_prefix[41];
150 struct patch *next;
153 #define CHUNKSIZE (8192)
154 #define SLOP (16)
156 static void *read_patch_file(int fd, unsigned long *sizep)
158 unsigned long size = 0, alloc = CHUNKSIZE;
159 void *buffer = xmalloc(alloc);
161 for (;;) {
162 int nr = alloc - size;
163 if (nr < 1024) {
164 alloc += CHUNKSIZE;
165 buffer = xrealloc(buffer, alloc);
166 nr = alloc - size;
168 nr = xread(fd, (char *) buffer + size, nr);
169 if (!nr)
170 break;
171 if (nr < 0)
172 die("git-apply: read returned %s", strerror(errno));
173 size += nr;
175 *sizep = size;
178 * Make sure that we have some slop in the buffer
179 * so that we can do speculative "memcmp" etc, and
180 * see to it that it is NUL-filled.
182 if (alloc < size + SLOP)
183 buffer = xrealloc(buffer, size + SLOP);
184 memset((char *) buffer + size, 0, SLOP);
185 return buffer;
188 static unsigned long linelen(const char *buffer, unsigned long size)
190 unsigned long len = 0;
191 while (size--) {
192 len++;
193 if (*buffer++ == '\n')
194 break;
196 return len;
199 static int is_dev_null(const char *str)
201 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
204 #define TERM_SPACE 1
205 #define TERM_TAB 2
207 static int name_terminate(const char *name, int namelen, int c, int terminate)
209 if (c == ' ' && !(terminate & TERM_SPACE))
210 return 0;
211 if (c == '\t' && !(terminate & TERM_TAB))
212 return 0;
214 return 1;
217 static char * find_name(const char *line, char *def, int p_value, int terminate)
219 int len;
220 const char *start = line;
221 char *name;
223 if (*line == '"') {
224 /* Proposed "new-style" GNU patch/diff format; see
225 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
227 name = unquote_c_style(line, NULL);
228 if (name) {
229 char *cp = name;
230 while (p_value) {
231 cp = strchr(name, '/');
232 if (!cp)
233 break;
234 cp++;
235 p_value--;
237 if (cp) {
238 /* name can later be freed, so we need
239 * to memmove, not just return cp
241 memmove(name, cp, strlen(cp) + 1);
242 free(def);
243 return name;
245 else {
246 free(name);
247 name = NULL;
252 for (;;) {
253 char c = *line;
255 if (isspace(c)) {
256 if (c == '\n')
257 break;
258 if (name_terminate(start, line-start, c, terminate))
259 break;
261 line++;
262 if (c == '/' && !--p_value)
263 start = line;
265 if (!start)
266 return def;
267 len = line - start;
268 if (!len)
269 return def;
272 * Generally we prefer the shorter name, especially
273 * if the other one is just a variation of that with
274 * something else tacked on to the end (ie "file.orig"
275 * or "file~").
277 if (def) {
278 int deflen = strlen(def);
279 if (deflen < len && !strncmp(start, def, deflen))
280 return def;
283 name = xmalloc(len + 1);
284 memcpy(name, start, len);
285 name[len] = 0;
286 free(def);
287 return name;
291 * Get the name etc info from the --/+++ lines of a traditional patch header
293 * NOTE! This hardcodes "-p1" behaviour in filename detection.
295 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
296 * files, we can happily check the index for a match, but for creating a
297 * new file we should try to match whatever "patch" does. I have no idea.
299 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
301 char *name;
303 first += 4; /* skip "--- " */
304 second += 4; /* skip "+++ " */
305 if (is_dev_null(first)) {
306 patch->is_new = 1;
307 patch->is_delete = 0;
308 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
309 patch->new_name = name;
310 } else if (is_dev_null(second)) {
311 patch->is_new = 0;
312 patch->is_delete = 1;
313 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
314 patch->old_name = name;
315 } else {
316 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
317 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
318 patch->old_name = patch->new_name = name;
320 if (!name)
321 die("unable to find filename in patch at line %d", linenr);
324 static int gitdiff_hdrend(const char *line, struct patch *patch)
326 return -1;
330 * We're anal about diff header consistency, to make
331 * sure that we don't end up having strange ambiguous
332 * patches floating around.
334 * As a result, gitdiff_{old|new}name() will check
335 * their names against any previous information, just
336 * to make sure..
338 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
340 if (!orig_name && !isnull)
341 return find_name(line, NULL, 1, 0);
343 if (orig_name) {
344 int len;
345 const char *name;
346 char *another;
347 name = orig_name;
348 len = strlen(name);
349 if (isnull)
350 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
351 another = find_name(line, NULL, 1, 0);
352 if (!another || memcmp(another, name, len))
353 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
354 free(another);
355 return orig_name;
357 else {
358 /* expect "/dev/null" */
359 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
360 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
361 return NULL;
365 static int gitdiff_oldname(const char *line, struct patch *patch)
367 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
368 return 0;
371 static int gitdiff_newname(const char *line, struct patch *patch)
373 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
374 return 0;
377 static int gitdiff_oldmode(const char *line, struct patch *patch)
379 patch->old_mode = strtoul(line, NULL, 8);
380 return 0;
383 static int gitdiff_newmode(const char *line, struct patch *patch)
385 patch->new_mode = strtoul(line, NULL, 8);
386 return 0;
389 static int gitdiff_delete(const char *line, struct patch *patch)
391 patch->is_delete = 1;
392 patch->old_name = patch->def_name;
393 return gitdiff_oldmode(line, patch);
396 static int gitdiff_newfile(const char *line, struct patch *patch)
398 patch->is_new = 1;
399 patch->new_name = patch->def_name;
400 return gitdiff_newmode(line, patch);
403 static int gitdiff_copysrc(const char *line, struct patch *patch)
405 patch->is_copy = 1;
406 patch->old_name = find_name(line, NULL, 0, 0);
407 return 0;
410 static int gitdiff_copydst(const char *line, struct patch *patch)
412 patch->is_copy = 1;
413 patch->new_name = find_name(line, NULL, 0, 0);
414 return 0;
417 static int gitdiff_renamesrc(const char *line, struct patch *patch)
419 patch->is_rename = 1;
420 patch->old_name = find_name(line, NULL, 0, 0);
421 return 0;
424 static int gitdiff_renamedst(const char *line, struct patch *patch)
426 patch->is_rename = 1;
427 patch->new_name = find_name(line, NULL, 0, 0);
428 return 0;
431 static int gitdiff_similarity(const char *line, struct patch *patch)
433 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
434 patch->score = 0;
435 return 0;
438 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
440 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
441 patch->score = 0;
442 return 0;
445 static int gitdiff_index(const char *line, struct patch *patch)
447 /* index line is N hexadecimal, "..", N hexadecimal,
448 * and optional space with octal mode.
450 const char *ptr, *eol;
451 int len;
453 ptr = strchr(line, '.');
454 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
455 return 0;
456 len = ptr - line;
457 memcpy(patch->old_sha1_prefix, line, len);
458 patch->old_sha1_prefix[len] = 0;
460 line = ptr + 2;
461 ptr = strchr(line, ' ');
462 eol = strchr(line, '\n');
464 if (!ptr || eol < ptr)
465 ptr = eol;
466 len = ptr - line;
468 if (40 < len)
469 return 0;
470 memcpy(patch->new_sha1_prefix, line, len);
471 patch->new_sha1_prefix[len] = 0;
472 if (*ptr == ' ')
473 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
474 return 0;
478 * This is normal for a diff that doesn't change anything: we'll fall through
479 * into the next diff. Tell the parser to break out.
481 static int gitdiff_unrecognized(const char *line, struct patch *patch)
483 return -1;
486 static const char *stop_at_slash(const char *line, int llen)
488 int i;
490 for (i = 0; i < llen; i++) {
491 int ch = line[i];
492 if (ch == '/')
493 return line + i;
495 return NULL;
498 /* This is to extract the same name that appears on "diff --git"
499 * line. We do not find and return anything if it is a rename
500 * patch, and it is OK because we will find the name elsewhere.
501 * We need to reliably find name only when it is mode-change only,
502 * creation or deletion of an empty file. In any of these cases,
503 * both sides are the same name under a/ and b/ respectively.
505 static char *git_header_name(char *line, int llen)
507 int len;
508 const char *name;
509 const char *second = NULL;
511 line += strlen("diff --git ");
512 llen -= strlen("diff --git ");
514 if (*line == '"') {
515 const char *cp;
516 char *first = unquote_c_style(line, &second);
517 if (!first)
518 return NULL;
520 /* advance to the first slash */
521 cp = stop_at_slash(first, strlen(first));
522 if (!cp || cp == first) {
523 /* we do not accept absolute paths */
524 free_first_and_fail:
525 free(first);
526 return NULL;
528 len = strlen(cp+1);
529 memmove(first, cp+1, len+1); /* including NUL */
531 /* second points at one past closing dq of name.
532 * find the second name.
534 while ((second < line + llen) && isspace(*second))
535 second++;
537 if (line + llen <= second)
538 goto free_first_and_fail;
539 if (*second == '"') {
540 char *sp = unquote_c_style(second, NULL);
541 if (!sp)
542 goto free_first_and_fail;
543 cp = stop_at_slash(sp, strlen(sp));
544 if (!cp || cp == sp) {
545 free_both_and_fail:
546 free(sp);
547 goto free_first_and_fail;
549 /* They must match, otherwise ignore */
550 if (strcmp(cp+1, first))
551 goto free_both_and_fail;
552 free(sp);
553 return first;
556 /* unquoted second */
557 cp = stop_at_slash(second, line + llen - second);
558 if (!cp || cp == second)
559 goto free_first_and_fail;
560 cp++;
561 if (line + llen - cp != len + 1 ||
562 memcmp(first, cp, len))
563 goto free_first_and_fail;
564 return first;
567 /* unquoted first name */
568 name = stop_at_slash(line, llen);
569 if (!name || name == line)
570 return NULL;
572 name++;
574 /* since the first name is unquoted, a dq if exists must be
575 * the beginning of the second name.
577 for (second = name; second < line + llen; second++) {
578 if (*second == '"') {
579 const char *cp = second;
580 const char *np;
581 char *sp = unquote_c_style(second, NULL);
583 if (!sp)
584 return NULL;
585 np = stop_at_slash(sp, strlen(sp));
586 if (!np || np == sp) {
587 free_second_and_fail:
588 free(sp);
589 return NULL;
591 np++;
592 len = strlen(np);
593 if (len < cp - name &&
594 !strncmp(np, name, len) &&
595 isspace(name[len])) {
596 /* Good */
597 memmove(sp, np, len + 1);
598 return sp;
600 goto free_second_and_fail;
605 * Accept a name only if it shows up twice, exactly the same
606 * form.
608 for (len = 0 ; ; len++) {
609 char c = name[len];
611 switch (c) {
612 default:
613 continue;
614 case '\n':
615 return NULL;
616 case '\t': case ' ':
617 second = name+len;
618 for (;;) {
619 char c = *second++;
620 if (c == '\n')
621 return NULL;
622 if (c == '/')
623 break;
625 if (second[len] == '\n' && !memcmp(name, second, len)) {
626 char *ret = xmalloc(len + 1);
627 memcpy(ret, name, len);
628 ret[len] = 0;
629 return ret;
633 return NULL;
636 /* Verify that we recognize the lines following a git header */
637 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
639 unsigned long offset;
641 /* A git diff has explicit new/delete information, so we don't guess */
642 patch->is_new = 0;
643 patch->is_delete = 0;
646 * Some things may not have the old name in the
647 * rest of the headers anywhere (pure mode changes,
648 * or removing or adding empty files), so we get
649 * the default name from the header.
651 patch->def_name = git_header_name(line, len);
653 line += len;
654 size -= len;
655 linenr++;
656 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
657 static const struct opentry {
658 const char *str;
659 int (*fn)(const char *, struct patch *);
660 } optable[] = {
661 { "@@ -", gitdiff_hdrend },
662 { "--- ", gitdiff_oldname },
663 { "+++ ", gitdiff_newname },
664 { "old mode ", gitdiff_oldmode },
665 { "new mode ", gitdiff_newmode },
666 { "deleted file mode ", gitdiff_delete },
667 { "new file mode ", gitdiff_newfile },
668 { "copy from ", gitdiff_copysrc },
669 { "copy to ", gitdiff_copydst },
670 { "rename old ", gitdiff_renamesrc },
671 { "rename new ", gitdiff_renamedst },
672 { "rename from ", gitdiff_renamesrc },
673 { "rename to ", gitdiff_renamedst },
674 { "similarity index ", gitdiff_similarity },
675 { "dissimilarity index ", gitdiff_dissimilarity },
676 { "index ", gitdiff_index },
677 { "", gitdiff_unrecognized },
679 int i;
681 len = linelen(line, size);
682 if (!len || line[len-1] != '\n')
683 break;
684 for (i = 0; i < ARRAY_SIZE(optable); i++) {
685 const struct opentry *p = optable + i;
686 int oplen = strlen(p->str);
687 if (len < oplen || memcmp(p->str, line, oplen))
688 continue;
689 if (p->fn(line + oplen, patch) < 0)
690 return offset;
691 break;
695 return offset;
698 static int parse_num(const char *line, unsigned long *p)
700 char *ptr;
702 if (!isdigit(*line))
703 return 0;
704 *p = strtoul(line, &ptr, 10);
705 return ptr - line;
708 static int parse_range(const char *line, int len, int offset, const char *expect,
709 unsigned long *p1, unsigned long *p2)
711 int digits, ex;
713 if (offset < 0 || offset >= len)
714 return -1;
715 line += offset;
716 len -= offset;
718 digits = parse_num(line, p1);
719 if (!digits)
720 return -1;
722 offset += digits;
723 line += digits;
724 len -= digits;
726 *p2 = 1;
727 if (*line == ',') {
728 digits = parse_num(line+1, p2);
729 if (!digits)
730 return -1;
732 offset += digits+1;
733 line += digits+1;
734 len -= digits+1;
737 ex = strlen(expect);
738 if (ex > len)
739 return -1;
740 if (memcmp(line, expect, ex))
741 return -1;
743 return offset + ex;
747 * Parse a unified diff fragment header of the
748 * form "@@ -a,b +c,d @@"
750 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
752 int offset;
754 if (!len || line[len-1] != '\n')
755 return -1;
757 /* Figure out the number of lines in a fragment */
758 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
759 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
761 return offset;
764 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
766 unsigned long offset, len;
768 patch->is_rename = patch->is_copy = 0;
769 patch->is_new = patch->is_delete = -1;
770 patch->old_mode = patch->new_mode = 0;
771 patch->old_name = patch->new_name = NULL;
772 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
773 unsigned long nextlen;
775 len = linelen(line, size);
776 if (!len)
777 break;
779 /* Testing this early allows us to take a few shortcuts.. */
780 if (len < 6)
781 continue;
784 * Make sure we don't find any unconnected patch fragments.
785 * That's a sign that we didn't find a header, and that a
786 * patch has become corrupted/broken up.
788 if (!memcmp("@@ -", line, 4)) {
789 struct fragment dummy;
790 if (parse_fragment_header(line, len, &dummy) < 0)
791 continue;
792 error("patch fragment without header at line %d: %.*s", linenr, (int)len-1, line);
795 if (size < len + 6)
796 break;
799 * Git patch? It might not have a real patch, just a rename
800 * or mode change, so we handle that specially
802 if (!memcmp("diff --git ", line, 11)) {
803 int git_hdr_len = parse_git_header(line, len, size, patch);
804 if (git_hdr_len <= len)
805 continue;
806 if (!patch->old_name && !patch->new_name) {
807 if (!patch->def_name)
808 die("git diff header lacks filename information (line %d)", linenr);
809 patch->old_name = patch->new_name = patch->def_name;
811 *hdrsize = git_hdr_len;
812 return offset;
815 /** --- followed by +++ ? */
816 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
817 continue;
820 * We only accept unified patches, so we want it to
821 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
822 * minimum
824 nextlen = linelen(line + len, size - len);
825 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
826 continue;
828 /* Ok, we'll consider it a patch */
829 parse_traditional_patch(line, line+len, patch);
830 *hdrsize = len + nextlen;
831 linenr += 2;
832 return offset;
834 return -1;
838 * Parse a unified diff. Note that this really needs
839 * to parse each fragment separately, since the only
840 * way to know the difference between a "---" that is
841 * part of a patch, and a "---" that starts the next
842 * patch is to look at the line counts..
844 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
846 int added, deleted;
847 int len = linelen(line, size), offset;
848 unsigned long oldlines, newlines;
849 unsigned long leading, trailing;
851 offset = parse_fragment_header(line, len, fragment);
852 if (offset < 0)
853 return -1;
854 oldlines = fragment->oldlines;
855 newlines = fragment->newlines;
856 leading = 0;
857 trailing = 0;
859 if (patch->is_new < 0) {
860 patch->is_new = !oldlines;
861 if (!oldlines)
862 patch->old_name = NULL;
864 if (patch->is_delete < 0) {
865 patch->is_delete = !newlines;
866 if (!newlines)
867 patch->new_name = NULL;
870 if (patch->is_new && oldlines)
871 return error("new file depends on old contents");
872 if (patch->is_delete != !newlines) {
873 if (newlines)
874 return error("deleted file still has contents");
875 fprintf(stderr, "** warning: file %s becomes empty but is not deleted\n", patch->new_name);
878 /* Parse the thing.. */
879 line += len;
880 size -= len;
881 linenr++;
882 added = deleted = 0;
883 for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
884 if (!oldlines && !newlines)
885 break;
886 len = linelen(line, size);
887 if (!len || line[len-1] != '\n')
888 return -1;
889 switch (*line) {
890 default:
891 return -1;
892 case ' ':
893 oldlines--;
894 newlines--;
895 if (!deleted && !added)
896 leading++;
897 trailing++;
898 break;
899 case '-':
900 deleted++;
901 oldlines--;
902 trailing = 0;
903 break;
904 case '+':
906 * We know len is at least two, since we have a '+' and
907 * we checked that the last character was a '\n' above.
908 * That is, an addition of an empty line would check
909 * the '+' here. Sneaky...
911 if ((new_whitespace != nowarn_whitespace) &&
912 isspace(line[len-2])) {
913 whitespace_error++;
914 if (squelch_whitespace_errors &&
915 squelch_whitespace_errors <
916 whitespace_error)
918 else {
919 fprintf(stderr, "Adds trailing whitespace.\n%s:%d:%.*s\n",
920 patch_input_file,
921 linenr, len-2, line+1);
924 added++;
925 newlines--;
926 trailing = 0;
927 break;
929 /* We allow "\ No newline at end of file". Depending
930 * on locale settings when the patch was produced we
931 * don't know what this line looks like. The only
932 * thing we do know is that it begins with "\ ".
933 * Checking for 12 is just for sanity check -- any
934 * l10n of "\ No newline..." is at least that long.
936 case '\\':
937 if (len < 12 || memcmp(line, "\\ ", 2))
938 return -1;
939 break;
942 if (oldlines || newlines)
943 return -1;
944 fragment->leading = leading;
945 fragment->trailing = trailing;
947 /* If a fragment ends with an incomplete line, we failed to include
948 * it in the above loop because we hit oldlines == newlines == 0
949 * before seeing it.
951 if (12 < size && !memcmp(line, "\\ ", 2))
952 offset += linelen(line, size);
954 patch->lines_added += added;
955 patch->lines_deleted += deleted;
956 return offset;
959 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
961 unsigned long offset = 0;
962 struct fragment **fragp = &patch->fragments;
964 while (size > 4 && !memcmp(line, "@@ -", 4)) {
965 struct fragment *fragment;
966 int len;
968 fragment = xcalloc(1, sizeof(*fragment));
969 len = parse_fragment(line, size, patch, fragment);
970 if (len <= 0)
971 die("corrupt patch at line %d", linenr);
973 fragment->patch = line;
974 fragment->size = len;
976 *fragp = fragment;
977 fragp = &fragment->next;
979 offset += len;
980 line += len;
981 size -= len;
983 return offset;
986 static inline int metadata_changes(struct patch *patch)
988 return patch->is_rename > 0 ||
989 patch->is_copy > 0 ||
990 patch->is_new > 0 ||
991 patch->is_delete ||
992 (patch->old_mode && patch->new_mode &&
993 patch->old_mode != patch->new_mode);
996 static char *inflate_it(const void *data, unsigned long size,
997 unsigned long inflated_size)
999 z_stream stream;
1000 void *out;
1001 int st;
1003 memset(&stream, 0, sizeof(stream));
1005 stream.next_in = (unsigned char *)data;
1006 stream.avail_in = size;
1007 stream.next_out = out = xmalloc(inflated_size);
1008 stream.avail_out = inflated_size;
1009 inflateInit(&stream);
1010 st = inflate(&stream, Z_FINISH);
1011 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1012 free(out);
1013 return NULL;
1015 return out;
1018 static struct fragment *parse_binary_hunk(char **buf_p,
1019 unsigned long *sz_p,
1020 int *status_p,
1021 int *used_p)
1023 /* Expect a line that begins with binary patch method ("literal"
1024 * or "delta"), followed by the length of data before deflating.
1025 * a sequence of 'length-byte' followed by base-85 encoded data
1026 * should follow, terminated by a newline.
1028 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1029 * and we would limit the patch line to 66 characters,
1030 * so one line can fit up to 13 groups that would decode
1031 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1032 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1034 int llen, used;
1035 unsigned long size = *sz_p;
1036 char *buffer = *buf_p;
1037 int patch_method;
1038 unsigned long origlen;
1039 char *data = NULL;
1040 int hunk_size = 0;
1041 struct fragment *frag;
1043 llen = linelen(buffer, size);
1044 used = llen;
1046 *status_p = 0;
1048 if (!strncmp(buffer, "delta ", 6)) {
1049 patch_method = BINARY_DELTA_DEFLATED;
1050 origlen = strtoul(buffer + 6, NULL, 10);
1052 else if (!strncmp(buffer, "literal ", 8)) {
1053 patch_method = BINARY_LITERAL_DEFLATED;
1054 origlen = strtoul(buffer + 8, NULL, 10);
1056 else
1057 return NULL;
1059 linenr++;
1060 buffer += llen;
1061 while (1) {
1062 int byte_length, max_byte_length, newsize;
1063 llen = linelen(buffer, size);
1064 used += llen;
1065 linenr++;
1066 if (llen == 1) {
1067 /* consume the blank line */
1068 buffer++;
1069 size--;
1070 break;
1072 /* Minimum line is "A00000\n" which is 7-byte long,
1073 * and the line length must be multiple of 5 plus 2.
1075 if ((llen < 7) || (llen-2) % 5)
1076 goto corrupt;
1077 max_byte_length = (llen - 2) / 5 * 4;
1078 byte_length = *buffer;
1079 if ('A' <= byte_length && byte_length <= 'Z')
1080 byte_length = byte_length - 'A' + 1;
1081 else if ('a' <= byte_length && byte_length <= 'z')
1082 byte_length = byte_length - 'a' + 27;
1083 else
1084 goto corrupt;
1085 /* if the input length was not multiple of 4, we would
1086 * have filler at the end but the filler should never
1087 * exceed 3 bytes
1089 if (max_byte_length < byte_length ||
1090 byte_length <= max_byte_length - 4)
1091 goto corrupt;
1092 newsize = hunk_size + byte_length;
1093 data = xrealloc(data, newsize);
1094 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1095 goto corrupt;
1096 hunk_size = newsize;
1097 buffer += llen;
1098 size -= llen;
1101 frag = xcalloc(1, sizeof(*frag));
1102 frag->patch = inflate_it(data, hunk_size, origlen);
1103 if (!frag->patch)
1104 goto corrupt;
1105 free(data);
1106 frag->size = origlen;
1107 *buf_p = buffer;
1108 *sz_p = size;
1109 *used_p = used;
1110 frag->binary_patch_method = patch_method;
1111 return frag;
1113 corrupt:
1114 if (data)
1115 free(data);
1116 *status_p = -1;
1117 error("corrupt binary patch at line %d: %.*s",
1118 linenr-1, llen-1, buffer);
1119 return NULL;
1122 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1124 /* We have read "GIT binary patch\n"; what follows is a line
1125 * that says the patch method (currently, either "literal" or
1126 * "delta") and the length of data before deflating; a
1127 * sequence of 'length-byte' followed by base-85 encoded data
1128 * follows.
1130 * When a binary patch is reversible, there is another binary
1131 * hunk in the same format, starting with patch method (either
1132 * "literal" or "delta") with the length of data, and a sequence
1133 * of length-byte + base-85 encoded data, terminated with another
1134 * empty line. This data, when applied to the postimage, produces
1135 * the preimage.
1137 struct fragment *forward;
1138 struct fragment *reverse;
1139 int status;
1140 int used, used_1;
1142 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1143 if (!forward && !status)
1144 /* there has to be one hunk (forward hunk) */
1145 return error("unrecognized binary patch at line %d", linenr-1);
1146 if (status)
1147 /* otherwise we already gave an error message */
1148 return status;
1150 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1151 if (reverse)
1152 used += used_1;
1153 else if (status) {
1154 /* not having reverse hunk is not an error, but having
1155 * a corrupt reverse hunk is.
1157 free((void*) forward->patch);
1158 free(forward);
1159 return status;
1161 forward->next = reverse;
1162 patch->fragments = forward;
1163 patch->is_binary = 1;
1164 return used;
1167 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1169 int hdrsize, patchsize;
1170 int offset = find_header(buffer, size, &hdrsize, patch);
1172 if (offset < 0)
1173 return offset;
1175 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1177 if (!patchsize) {
1178 static const char *binhdr[] = {
1179 "Binary files ",
1180 "Files ",
1181 NULL,
1183 static const char git_binary[] = "GIT binary patch\n";
1184 int i;
1185 int hd = hdrsize + offset;
1186 unsigned long llen = linelen(buffer + hd, size - hd);
1188 if (llen == sizeof(git_binary) - 1 &&
1189 !memcmp(git_binary, buffer + hd, llen)) {
1190 int used;
1191 linenr++;
1192 used = parse_binary(buffer + hd + llen,
1193 size - hd - llen, patch);
1194 if (used)
1195 patchsize = used + llen;
1196 else
1197 patchsize = 0;
1199 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1200 for (i = 0; binhdr[i]; i++) {
1201 int len = strlen(binhdr[i]);
1202 if (len < size - hd &&
1203 !memcmp(binhdr[i], buffer + hd, len)) {
1204 linenr++;
1205 patch->is_binary = 1;
1206 patchsize = llen;
1207 break;
1212 /* Empty patch cannot be applied if:
1213 * - it is a binary patch and we do not do binary_replace, or
1214 * - text patch without metadata change
1216 if ((apply || check) &&
1217 (patch->is_binary
1218 ? !allow_binary_replacement
1219 : !metadata_changes(patch)))
1220 die("patch with only garbage at line %d", linenr);
1223 return offset + hdrsize + patchsize;
1226 #define swap(a,b) myswap((a),(b),sizeof(a))
1228 #define myswap(a, b, size) do { \
1229 unsigned char mytmp[size]; \
1230 memcpy(mytmp, &a, size); \
1231 memcpy(&a, &b, size); \
1232 memcpy(&b, mytmp, size); \
1233 } while (0)
1235 static void reverse_patches(struct patch *p)
1237 for (; p; p = p->next) {
1238 struct fragment *frag = p->fragments;
1240 swap(p->new_name, p->old_name);
1241 swap(p->new_mode, p->old_mode);
1242 swap(p->is_new, p->is_delete);
1243 swap(p->lines_added, p->lines_deleted);
1244 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1246 for (; frag; frag = frag->next) {
1247 swap(frag->newpos, frag->oldpos);
1248 swap(frag->newlines, frag->oldlines);
1253 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1254 static const char minuses[]= "----------------------------------------------------------------------";
1256 static void show_stats(struct patch *patch)
1258 const char *prefix = "";
1259 char *name = patch->new_name;
1260 char *qname = NULL;
1261 int len, max, add, del, total;
1263 if (!name)
1264 name = patch->old_name;
1266 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1267 qname = xmalloc(len + 1);
1268 quote_c_style(name, qname, NULL, 0);
1269 name = qname;
1273 * "scale" the filename
1275 len = strlen(name);
1276 max = max_len;
1277 if (max > 50)
1278 max = 50;
1279 if (len > max) {
1280 char *slash;
1281 prefix = "...";
1282 max -= 3;
1283 name += len - max;
1284 slash = strchr(name, '/');
1285 if (slash)
1286 name = slash;
1288 len = max;
1291 * scale the add/delete
1293 max = max_change;
1294 if (max + len > 70)
1295 max = 70 - len;
1297 add = patch->lines_added;
1298 del = patch->lines_deleted;
1299 total = add + del;
1301 if (max_change > 0) {
1302 total = (total * max + max_change / 2) / max_change;
1303 add = (add * max + max_change / 2) / max_change;
1304 del = total - add;
1306 if (patch->is_binary)
1307 printf(" %s%-*s | Bin\n", prefix, len, name);
1308 else
1309 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1310 len, name, patch->lines_added + patch->lines_deleted,
1311 add, pluses, del, minuses);
1312 if (qname)
1313 free(qname);
1316 static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
1318 int fd;
1319 unsigned long got;
1321 switch (st->st_mode & S_IFMT) {
1322 case S_IFLNK:
1323 return readlink(path, buf, size);
1324 case S_IFREG:
1325 fd = open(path, O_RDONLY);
1326 if (fd < 0)
1327 return error("unable to open %s", path);
1328 got = 0;
1329 for (;;) {
1330 int ret = xread(fd, (char *) buf + got, size - got);
1331 if (ret <= 0)
1332 break;
1333 got += ret;
1335 close(fd);
1336 return got;
1338 default:
1339 return -1;
1343 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1345 int i;
1346 unsigned long start, backwards, forwards;
1348 if (fragsize > size)
1349 return -1;
1351 start = 0;
1352 if (line > 1) {
1353 unsigned long offset = 0;
1354 i = line-1;
1355 while (offset + fragsize <= size) {
1356 if (buf[offset++] == '\n') {
1357 start = offset;
1358 if (!--i)
1359 break;
1364 /* Exact line number? */
1365 if (!memcmp(buf + start, fragment, fragsize))
1366 return start;
1369 * There's probably some smart way to do this, but I'll leave
1370 * that to the smart and beautiful people. I'm simple and stupid.
1372 backwards = start;
1373 forwards = start;
1374 for (i = 0; ; i++) {
1375 unsigned long try;
1376 int n;
1378 /* "backward" */
1379 if (i & 1) {
1380 if (!backwards) {
1381 if (forwards + fragsize > size)
1382 break;
1383 continue;
1385 do {
1386 --backwards;
1387 } while (backwards && buf[backwards-1] != '\n');
1388 try = backwards;
1389 } else {
1390 while (forwards + fragsize <= size) {
1391 if (buf[forwards++] == '\n')
1392 break;
1394 try = forwards;
1397 if (try + fragsize > size)
1398 continue;
1399 if (memcmp(buf + try, fragment, fragsize))
1400 continue;
1401 n = (i >> 1)+1;
1402 if (i & 1)
1403 n = -n;
1404 *lines = n;
1405 return try;
1409 * We should start searching forward and backward.
1411 return -1;
1414 static void remove_first_line(const char **rbuf, int *rsize)
1416 const char *buf = *rbuf;
1417 int size = *rsize;
1418 unsigned long offset;
1419 offset = 0;
1420 while (offset <= size) {
1421 if (buf[offset++] == '\n')
1422 break;
1424 *rsize = size - offset;
1425 *rbuf = buf + offset;
1428 static void remove_last_line(const char **rbuf, int *rsize)
1430 const char *buf = *rbuf;
1431 int size = *rsize;
1432 unsigned long offset;
1433 offset = size - 1;
1434 while (offset > 0) {
1435 if (buf[--offset] == '\n')
1436 break;
1438 *rsize = offset + 1;
1441 struct buffer_desc {
1442 char *buffer;
1443 unsigned long size;
1444 unsigned long alloc;
1447 static int apply_line(char *output, const char *patch, int plen)
1449 /* plen is number of bytes to be copied from patch,
1450 * starting at patch+1 (patch[0] is '+'). Typically
1451 * patch[plen] is '\n'.
1453 int add_nl_to_tail = 0;
1454 if ((new_whitespace == strip_whitespace) &&
1455 1 < plen && isspace(patch[plen-1])) {
1456 if (patch[plen] == '\n')
1457 add_nl_to_tail = 1;
1458 plen--;
1459 while (0 < plen && isspace(patch[plen]))
1460 plen--;
1461 applied_after_stripping++;
1463 memcpy(output, patch + 1, plen);
1464 if (add_nl_to_tail)
1465 output[plen++] = '\n';
1466 return plen;
1469 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag, int inaccurate_eof)
1471 int match_beginning, match_end;
1472 char *buf = desc->buffer;
1473 const char *patch = frag->patch;
1474 int offset, size = frag->size;
1475 char *old = xmalloc(size);
1476 char *new = xmalloc(size);
1477 const char *oldlines, *newlines;
1478 int oldsize = 0, newsize = 0;
1479 unsigned long leading, trailing;
1480 int pos, lines;
1482 while (size > 0) {
1483 char first;
1484 int len = linelen(patch, size);
1485 int plen;
1487 if (!len)
1488 break;
1491 * "plen" is how much of the line we should use for
1492 * the actual patch data. Normally we just remove the
1493 * first character on the line, but if the line is
1494 * followed by "\ No newline", then we also remove the
1495 * last one (which is the newline, of course).
1497 plen = len-1;
1498 if (len < size && patch[len] == '\\')
1499 plen--;
1500 first = *patch;
1501 if (apply_in_reverse) {
1502 if (first == '-')
1503 first = '+';
1504 else if (first == '+')
1505 first = '-';
1507 switch (first) {
1508 case ' ':
1509 case '-':
1510 memcpy(old + oldsize, patch + 1, plen);
1511 oldsize += plen;
1512 if (first == '-')
1513 break;
1514 /* Fall-through for ' ' */
1515 case '+':
1516 if (first != '+' || !no_add)
1517 newsize += apply_line(new + newsize, patch,
1518 plen);
1519 break;
1520 case '@': case '\\':
1521 /* Ignore it, we already handled it */
1522 break;
1523 default:
1524 return -1;
1526 patch += len;
1527 size -= len;
1530 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1531 newsize > 0 && new[newsize - 1] == '\n') {
1532 oldsize--;
1533 newsize--;
1536 oldlines = old;
1537 newlines = new;
1538 leading = frag->leading;
1539 trailing = frag->trailing;
1542 * If we don't have any leading/trailing data in the patch,
1543 * we want it to match at the beginning/end of the file.
1545 match_beginning = !leading && (frag->oldpos == 1);
1546 match_end = !trailing;
1548 lines = 0;
1549 pos = frag->newpos;
1550 for (;;) {
1551 offset = find_offset(buf, desc->size, oldlines, oldsize, pos, &lines);
1552 if (match_end && offset + oldsize != desc->size)
1553 offset = -1;
1554 if (match_beginning && offset)
1555 offset = -1;
1556 if (offset >= 0) {
1557 int diff = newsize - oldsize;
1558 unsigned long size = desc->size + diff;
1559 unsigned long alloc = desc->alloc;
1561 /* Warn if it was necessary to reduce the number
1562 * of context lines.
1564 if ((leading != frag->leading) || (trailing != frag->trailing))
1565 fprintf(stderr, "Context reduced to (%ld/%ld) to apply fragment at %d\n",
1566 leading, trailing, pos + lines);
1568 if (size > alloc) {
1569 alloc = size + 8192;
1570 desc->alloc = alloc;
1571 buf = xrealloc(buf, alloc);
1572 desc->buffer = buf;
1574 desc->size = size;
1575 memmove(buf + offset + newsize, buf + offset + oldsize, size - offset - newsize);
1576 memcpy(buf + offset, newlines, newsize);
1577 offset = 0;
1579 break;
1582 /* Am I at my context limits? */
1583 if ((leading <= p_context) && (trailing <= p_context))
1584 break;
1585 if (match_beginning || match_end) {
1586 match_beginning = match_end = 0;
1587 continue;
1589 /* Reduce the number of context lines
1590 * Reduce both leading and trailing if they are equal
1591 * otherwise just reduce the larger context.
1593 if (leading >= trailing) {
1594 remove_first_line(&oldlines, &oldsize);
1595 remove_first_line(&newlines, &newsize);
1596 pos--;
1597 leading--;
1599 if (trailing > leading) {
1600 remove_last_line(&oldlines, &oldsize);
1601 remove_last_line(&newlines, &newsize);
1602 trailing--;
1606 free(old);
1607 free(new);
1608 return offset;
1611 static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1613 unsigned long dst_size;
1614 struct fragment *fragment = patch->fragments;
1615 void *data;
1616 void *result;
1618 /* Binary patch is irreversible without the optional second hunk */
1619 if (apply_in_reverse) {
1620 if (!fragment->next)
1621 return error("cannot reverse-apply a binary patch "
1622 "without the reverse hunk to '%s'",
1623 patch->new_name
1624 ? patch->new_name : patch->old_name);
1625 fragment = fragment->next;
1627 data = (void*) fragment->patch;
1628 switch (fragment->binary_patch_method) {
1629 case BINARY_DELTA_DEFLATED:
1630 result = patch_delta(desc->buffer, desc->size,
1631 data,
1632 fragment->size,
1633 &dst_size);
1634 free(desc->buffer);
1635 desc->buffer = result;
1636 break;
1637 case BINARY_LITERAL_DEFLATED:
1638 free(desc->buffer);
1639 desc->buffer = data;
1640 dst_size = fragment->size;
1641 break;
1643 if (!desc->buffer)
1644 return -1;
1645 desc->size = desc->alloc = dst_size;
1646 return 0;
1649 static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1651 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1652 unsigned char sha1[20];
1653 unsigned char hdr[50];
1654 int hdrlen;
1656 if (!allow_binary_replacement)
1657 return error("cannot apply binary patch to '%s' "
1658 "without --allow-binary-replacement",
1659 name);
1661 /* For safety, we require patch index line to contain
1662 * full 40-byte textual SHA1 for old and new, at least for now.
1664 if (strlen(patch->old_sha1_prefix) != 40 ||
1665 strlen(patch->new_sha1_prefix) != 40 ||
1666 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1667 get_sha1_hex(patch->new_sha1_prefix, sha1))
1668 return error("cannot apply binary patch to '%s' "
1669 "without full index line", name);
1671 if (patch->old_name) {
1672 /* See if the old one matches what the patch
1673 * applies to.
1675 write_sha1_file_prepare(desc->buffer, desc->size,
1676 blob_type, sha1, hdr, &hdrlen);
1677 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1678 return error("the patch applies to '%s' (%s), "
1679 "which does not match the "
1680 "current contents.",
1681 name, sha1_to_hex(sha1));
1683 else {
1684 /* Otherwise, the old one must be empty. */
1685 if (desc->size)
1686 return error("the patch applies to an empty "
1687 "'%s' but it is not empty", name);
1690 get_sha1_hex(patch->new_sha1_prefix, sha1);
1691 if (is_null_sha1(sha1)) {
1692 free(desc->buffer);
1693 desc->alloc = desc->size = 0;
1694 desc->buffer = NULL;
1695 return 0; /* deletion patch */
1698 if (has_sha1_file(sha1)) {
1699 /* We already have the postimage */
1700 char type[10];
1701 unsigned long size;
1703 free(desc->buffer);
1704 desc->buffer = read_sha1_file(sha1, type, &size);
1705 if (!desc->buffer)
1706 return error("the necessary postimage %s for "
1707 "'%s' cannot be read",
1708 patch->new_sha1_prefix, name);
1709 desc->alloc = desc->size = size;
1711 else {
1712 /* We have verified desc matches the preimage;
1713 * apply the patch data to it, which is stored
1714 * in the patch->fragments->{patch,size}.
1716 if (apply_binary_fragment(desc, patch))
1717 return error("binary patch does not apply to '%s'",
1718 name);
1720 /* verify that the result matches */
1721 write_sha1_file_prepare(desc->buffer, desc->size, blob_type,
1722 sha1, hdr, &hdrlen);
1723 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1724 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)", name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1727 return 0;
1730 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1732 struct fragment *frag = patch->fragments;
1733 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1735 if (patch->is_binary)
1736 return apply_binary(desc, patch);
1738 while (frag) {
1739 if (apply_one_fragment(desc, frag, patch->inaccurate_eof) < 0)
1740 return error("patch failed: %s:%ld",
1741 name, frag->oldpos);
1742 frag = frag->next;
1744 return 0;
1747 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1749 char *buf;
1750 unsigned long size, alloc;
1751 struct buffer_desc desc;
1753 size = 0;
1754 alloc = 0;
1755 buf = NULL;
1756 if (cached) {
1757 if (ce) {
1758 char type[20];
1759 buf = read_sha1_file(ce->sha1, type, &size);
1760 if (!buf)
1761 return error("read of %s failed",
1762 patch->old_name);
1763 alloc = size;
1766 else if (patch->old_name) {
1767 size = st->st_size;
1768 alloc = size + 8192;
1769 buf = xmalloc(alloc);
1770 if (read_old_data(st, patch->old_name, buf, alloc) != size)
1771 return error("read of %s failed", patch->old_name);
1774 desc.size = size;
1775 desc.alloc = alloc;
1776 desc.buffer = buf;
1777 if (apply_fragments(&desc, patch) < 0)
1778 return -1;
1780 /* NUL terminate the result */
1781 if (desc.alloc <= desc.size)
1782 desc.buffer = xrealloc(desc.buffer, desc.size + 1);
1783 desc.buffer[desc.size] = 0;
1785 patch->result = desc.buffer;
1786 patch->resultsize = desc.size;
1788 if (patch->is_delete && patch->resultsize)
1789 return error("removal patch leaves file contents");
1791 return 0;
1794 static int check_patch(struct patch *patch, struct patch *prev_patch)
1796 struct stat st;
1797 const char *old_name = patch->old_name;
1798 const char *new_name = patch->new_name;
1799 const char *name = old_name ? old_name : new_name;
1800 struct cache_entry *ce = NULL;
1801 int ok_if_exists;
1803 if (old_name) {
1804 int changed = 0;
1805 int stat_ret = 0;
1806 unsigned st_mode = 0;
1808 if (!cached)
1809 stat_ret = lstat(old_name, &st);
1810 if (check_index) {
1811 int pos = cache_name_pos(old_name, strlen(old_name));
1812 if (pos < 0)
1813 return error("%s: does not exist in index",
1814 old_name);
1815 ce = active_cache[pos];
1816 if (stat_ret < 0) {
1817 struct checkout costate;
1818 if (errno != ENOENT)
1819 return error("%s: %s", old_name,
1820 strerror(errno));
1821 /* checkout */
1822 costate.base_dir = "";
1823 costate.base_dir_len = 0;
1824 costate.force = 0;
1825 costate.quiet = 0;
1826 costate.not_new = 0;
1827 costate.refresh_cache = 1;
1828 if (checkout_entry(ce,
1829 &costate,
1830 NULL) ||
1831 lstat(old_name, &st))
1832 return -1;
1834 if (!cached)
1835 changed = ce_match_stat(ce, &st, 1);
1836 if (changed)
1837 return error("%s: does not match index",
1838 old_name);
1839 if (cached)
1840 st_mode = ntohl(ce->ce_mode);
1842 else if (stat_ret < 0)
1843 return error("%s: %s", old_name, strerror(errno));
1845 if (!cached)
1846 st_mode = ntohl(create_ce_mode(st.st_mode));
1848 if (patch->is_new < 0)
1849 patch->is_new = 0;
1850 if (!patch->old_mode)
1851 patch->old_mode = st_mode;
1852 if ((st_mode ^ patch->old_mode) & S_IFMT)
1853 return error("%s: wrong type", old_name);
1854 if (st_mode != patch->old_mode)
1855 fprintf(stderr, "warning: %s has type %o, expected %o\n",
1856 old_name, st_mode, patch->old_mode);
1859 if (new_name && prev_patch && prev_patch->is_delete &&
1860 !strcmp(prev_patch->old_name, new_name))
1861 /* A type-change diff is always split into a patch to
1862 * delete old, immediately followed by a patch to
1863 * create new (see diff.c::run_diff()); in such a case
1864 * it is Ok that the entry to be deleted by the
1865 * previous patch is still in the working tree and in
1866 * the index.
1868 ok_if_exists = 1;
1869 else
1870 ok_if_exists = 0;
1872 if (new_name && (patch->is_new | patch->is_rename | patch->is_copy)) {
1873 if (check_index &&
1874 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
1875 !ok_if_exists)
1876 return error("%s: already exists in index", new_name);
1877 if (!cached) {
1878 struct stat nst;
1879 if (!lstat(new_name, &nst)) {
1880 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1881 ; /* ok */
1882 else
1883 return error("%s: already exists in working directory", new_name);
1885 else if ((errno != ENOENT) && (errno != ENOTDIR))
1886 return error("%s: %s", new_name, strerror(errno));
1888 if (!patch->new_mode) {
1889 if (patch->is_new)
1890 patch->new_mode = S_IFREG | 0644;
1891 else
1892 patch->new_mode = patch->old_mode;
1896 if (new_name && old_name) {
1897 int same = !strcmp(old_name, new_name);
1898 if (!patch->new_mode)
1899 patch->new_mode = patch->old_mode;
1900 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
1901 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
1902 patch->new_mode, new_name, patch->old_mode,
1903 same ? "" : " of ", same ? "" : old_name);
1906 if (apply_data(patch, &st, ce) < 0)
1907 return error("%s: patch does not apply", name);
1908 return 0;
1911 static int check_patch_list(struct patch *patch)
1913 struct patch *prev_patch = NULL;
1914 int error = 0;
1916 for (prev_patch = NULL; patch ; patch = patch->next) {
1917 error |= check_patch(patch, prev_patch);
1918 prev_patch = patch;
1920 return error;
1923 static void show_index_list(struct patch *list)
1925 struct patch *patch;
1927 /* Once we start supporting the reverse patch, it may be
1928 * worth showing the new sha1 prefix, but until then...
1930 for (patch = list; patch; patch = patch->next) {
1931 const unsigned char *sha1_ptr;
1932 unsigned char sha1[20];
1933 const char *name;
1935 name = patch->old_name ? patch->old_name : patch->new_name;
1936 if (patch->is_new)
1937 sha1_ptr = null_sha1;
1938 else if (get_sha1(patch->old_sha1_prefix, sha1))
1939 die("sha1 information is lacking or useless (%s).",
1940 name);
1941 else
1942 sha1_ptr = sha1;
1944 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
1945 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1946 quote_c_style(name, NULL, stdout, 0);
1947 else
1948 fputs(name, stdout);
1949 putchar(line_termination);
1953 static void stat_patch_list(struct patch *patch)
1955 int files, adds, dels;
1957 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
1958 files++;
1959 adds += patch->lines_added;
1960 dels += patch->lines_deleted;
1961 show_stats(patch);
1964 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
1967 static void numstat_patch_list(struct patch *patch)
1969 for ( ; patch; patch = patch->next) {
1970 const char *name;
1971 name = patch->new_name ? patch->new_name : patch->old_name;
1972 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
1973 if (line_termination && quote_c_style(name, NULL, NULL, 0))
1974 quote_c_style(name, NULL, stdout, 0);
1975 else
1976 fputs(name, stdout);
1977 putchar('\n');
1981 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
1983 if (mode)
1984 printf(" %s mode %06o %s\n", newdelete, mode, name);
1985 else
1986 printf(" %s %s\n", newdelete, name);
1989 static void show_mode_change(struct patch *p, int show_name)
1991 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
1992 if (show_name)
1993 printf(" mode change %06o => %06o %s\n",
1994 p->old_mode, p->new_mode, p->new_name);
1995 else
1996 printf(" mode change %06o => %06o\n",
1997 p->old_mode, p->new_mode);
2001 static void show_rename_copy(struct patch *p)
2003 const char *renamecopy = p->is_rename ? "rename" : "copy";
2004 const char *old, *new;
2006 /* Find common prefix */
2007 old = p->old_name;
2008 new = p->new_name;
2009 while (1) {
2010 const char *slash_old, *slash_new;
2011 slash_old = strchr(old, '/');
2012 slash_new = strchr(new, '/');
2013 if (!slash_old ||
2014 !slash_new ||
2015 slash_old - old != slash_new - new ||
2016 memcmp(old, new, slash_new - new))
2017 break;
2018 old = slash_old + 1;
2019 new = slash_new + 1;
2021 /* p->old_name thru old is the common prefix, and old and new
2022 * through the end of names are renames
2024 if (old != p->old_name)
2025 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2026 (int)(old - p->old_name), p->old_name,
2027 old, new, p->score);
2028 else
2029 printf(" %s %s => %s (%d%%)\n", renamecopy,
2030 p->old_name, p->new_name, p->score);
2031 show_mode_change(p, 0);
2034 static void summary_patch_list(struct patch *patch)
2036 struct patch *p;
2038 for (p = patch; p; p = p->next) {
2039 if (p->is_new)
2040 show_file_mode_name("create", p->new_mode, p->new_name);
2041 else if (p->is_delete)
2042 show_file_mode_name("delete", p->old_mode, p->old_name);
2043 else {
2044 if (p->is_rename || p->is_copy)
2045 show_rename_copy(p);
2046 else {
2047 if (p->score) {
2048 printf(" rewrite %s (%d%%)\n",
2049 p->new_name, p->score);
2050 show_mode_change(p, 0);
2052 else
2053 show_mode_change(p, 1);
2059 static void patch_stats(struct patch *patch)
2061 int lines = patch->lines_added + patch->lines_deleted;
2063 if (lines > max_change)
2064 max_change = lines;
2065 if (patch->old_name) {
2066 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2067 if (!len)
2068 len = strlen(patch->old_name);
2069 if (len > max_len)
2070 max_len = len;
2072 if (patch->new_name) {
2073 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2074 if (!len)
2075 len = strlen(patch->new_name);
2076 if (len > max_len)
2077 max_len = len;
2081 static void remove_file(struct patch *patch)
2083 if (write_index) {
2084 if (remove_file_from_cache(patch->old_name) < 0)
2085 die("unable to remove %s from index", patch->old_name);
2086 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2088 if (!cached)
2089 unlink(patch->old_name);
2092 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2094 struct stat st;
2095 struct cache_entry *ce;
2096 int namelen = strlen(path);
2097 unsigned ce_size = cache_entry_size(namelen);
2099 if (!write_index)
2100 return;
2102 ce = xcalloc(1, ce_size);
2103 memcpy(ce->name, path, namelen);
2104 ce->ce_mode = create_ce_mode(mode);
2105 ce->ce_flags = htons(namelen);
2106 if (!cached) {
2107 if (lstat(path, &st) < 0)
2108 die("unable to stat newly created file %s", path);
2109 fill_stat_cache_info(ce, &st);
2111 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2112 die("unable to create backing store for newly created file %s", path);
2113 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2114 die("unable to add cache entry for %s", path);
2117 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2119 int fd;
2121 if (S_ISLNK(mode))
2122 /* Although buf:size is counted string, it also is NUL
2123 * terminated.
2125 return symlink(buf, path);
2126 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2127 if (fd < 0)
2128 return -1;
2129 while (size) {
2130 int written = xwrite(fd, buf, size);
2131 if (written < 0)
2132 die("writing file %s: %s", path, strerror(errno));
2133 if (!written)
2134 die("out of space writing file %s", path);
2135 buf += written;
2136 size -= written;
2138 if (close(fd) < 0)
2139 die("closing file %s: %s", path, strerror(errno));
2140 return 0;
2144 * We optimistically assume that the directories exist,
2145 * which is true 99% of the time anyway. If they don't,
2146 * we create them and try again.
2148 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2150 if (cached)
2151 return;
2152 if (!try_create_file(path, mode, buf, size))
2153 return;
2155 if (errno == ENOENT) {
2156 if (safe_create_leading_directories(path))
2157 return;
2158 if (!try_create_file(path, mode, buf, size))
2159 return;
2162 if (errno == EEXIST || errno == EACCES) {
2163 /* We may be trying to create a file where a directory
2164 * used to be.
2166 struct stat st;
2167 errno = 0;
2168 if (!lstat(path, &st) && S_ISDIR(st.st_mode) && !rmdir(path))
2169 errno = EEXIST;
2172 if (errno == EEXIST) {
2173 unsigned int nr = getpid();
2175 for (;;) {
2176 const char *newpath;
2177 newpath = mkpath("%s~%u", path, nr);
2178 if (!try_create_file(newpath, mode, buf, size)) {
2179 if (!rename(newpath, path))
2180 return;
2181 unlink(newpath);
2182 break;
2184 if (errno != EEXIST)
2185 break;
2186 ++nr;
2189 die("unable to write file %s mode %o", path, mode);
2192 static void create_file(struct patch *patch)
2194 char *path = patch->new_name;
2195 unsigned mode = patch->new_mode;
2196 unsigned long size = patch->resultsize;
2197 char *buf = patch->result;
2199 if (!mode)
2200 mode = S_IFREG | 0644;
2201 create_one_file(path, mode, buf, size);
2202 add_index_file(path, mode, buf, size);
2203 cache_tree_invalidate_path(active_cache_tree, path);
2206 /* phase zero is to remove, phase one is to create */
2207 static void write_out_one_result(struct patch *patch, int phase)
2209 if (patch->is_delete > 0) {
2210 if (phase == 0)
2211 remove_file(patch);
2212 return;
2214 if (patch->is_new > 0 || patch->is_copy) {
2215 if (phase == 1)
2216 create_file(patch);
2217 return;
2220 * Rename or modification boils down to the same
2221 * thing: remove the old, write the new
2223 if (phase == 0)
2224 remove_file(patch);
2225 if (phase == 1)
2226 create_file(patch);
2229 static void write_out_results(struct patch *list, int skipped_patch)
2231 int phase;
2233 if (!list && !skipped_patch)
2234 die("No changes");
2236 for (phase = 0; phase < 2; phase++) {
2237 struct patch *l = list;
2238 while (l) {
2239 write_out_one_result(l, phase);
2240 l = l->next;
2245 static struct lock_file lock_file;
2247 static struct excludes {
2248 struct excludes *next;
2249 const char *path;
2250 } *excludes;
2252 static int use_patch(struct patch *p)
2254 const char *pathname = p->new_name ? p->new_name : p->old_name;
2255 struct excludes *x = excludes;
2256 while (x) {
2257 if (fnmatch(x->path, pathname, 0) == 0)
2258 return 0;
2259 x = x->next;
2261 if (0 < prefix_length) {
2262 int pathlen = strlen(pathname);
2263 if (pathlen <= prefix_length ||
2264 memcmp(prefix, pathname, prefix_length))
2265 return 0;
2267 return 1;
2270 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2272 unsigned long offset, size;
2273 char *buffer = read_patch_file(fd, &size);
2274 struct patch *list = NULL, **listp = &list;
2275 int skipped_patch = 0;
2277 patch_input_file = filename;
2278 if (!buffer)
2279 return -1;
2280 offset = 0;
2281 while (size > 0) {
2282 struct patch *patch;
2283 int nr;
2285 patch = xcalloc(1, sizeof(*patch));
2286 patch->inaccurate_eof = inaccurate_eof;
2287 nr = parse_chunk(buffer + offset, size, patch);
2288 if (nr < 0)
2289 break;
2290 if (apply_in_reverse)
2291 reverse_patches(patch);
2292 if (use_patch(patch)) {
2293 patch_stats(patch);
2294 *listp = patch;
2295 listp = &patch->next;
2296 } else {
2297 /* perhaps free it a bit better? */
2298 free(patch);
2299 skipped_patch++;
2301 offset += nr;
2302 size -= nr;
2305 if (whitespace_error && (new_whitespace == error_on_whitespace))
2306 apply = 0;
2308 write_index = check_index && apply;
2309 if (write_index && newfd < 0)
2310 newfd = hold_lock_file_for_update(&lock_file,
2311 get_index_file(), 1);
2312 if (check_index) {
2313 if (read_cache() < 0)
2314 die("unable to read index file");
2317 if ((check || apply) && check_patch_list(list) < 0)
2318 exit(1);
2320 if (apply)
2321 write_out_results(list, skipped_patch);
2323 if (show_index_info)
2324 show_index_list(list);
2326 if (diffstat)
2327 stat_patch_list(list);
2329 if (numstat)
2330 numstat_patch_list(list);
2332 if (summary)
2333 summary_patch_list(list);
2335 free(buffer);
2336 return 0;
2339 static int git_apply_config(const char *var, const char *value)
2341 if (!strcmp(var, "apply.whitespace")) {
2342 apply_default_whitespace = strdup(value);
2343 return 0;
2345 return git_default_config(var, value);
2349 int cmd_apply(int argc, const char **argv, const char *prefix)
2351 int i;
2352 int read_stdin = 1;
2353 int inaccurate_eof = 0;
2355 const char *whitespace_option = NULL;
2357 for (i = 1; i < argc; i++) {
2358 const char *arg = argv[i];
2359 char *end;
2360 int fd;
2362 if (!strcmp(arg, "-")) {
2363 apply_patch(0, "<stdin>", inaccurate_eof);
2364 read_stdin = 0;
2365 continue;
2367 if (!strncmp(arg, "--exclude=", 10)) {
2368 struct excludes *x = xmalloc(sizeof(*x));
2369 x->path = arg + 10;
2370 x->next = excludes;
2371 excludes = x;
2372 continue;
2374 if (!strncmp(arg, "-p", 2)) {
2375 p_value = atoi(arg + 2);
2376 continue;
2378 if (!strcmp(arg, "--no-add")) {
2379 no_add = 1;
2380 continue;
2382 if (!strcmp(arg, "--stat")) {
2383 apply = 0;
2384 diffstat = 1;
2385 continue;
2387 if (!strcmp(arg, "--allow-binary-replacement") ||
2388 !strcmp(arg, "--binary")) {
2389 allow_binary_replacement = 1;
2390 continue;
2392 if (!strcmp(arg, "--numstat")) {
2393 apply = 0;
2394 numstat = 1;
2395 continue;
2397 if (!strcmp(arg, "--summary")) {
2398 apply = 0;
2399 summary = 1;
2400 continue;
2402 if (!strcmp(arg, "--check")) {
2403 apply = 0;
2404 check = 1;
2405 continue;
2407 if (!strcmp(arg, "--index")) {
2408 check_index = 1;
2409 continue;
2411 if (!strcmp(arg, "--cached")) {
2412 check_index = 1;
2413 cached = 1;
2414 continue;
2416 if (!strcmp(arg, "--apply")) {
2417 apply = 1;
2418 continue;
2420 if (!strcmp(arg, "--index-info")) {
2421 apply = 0;
2422 show_index_info = 1;
2423 continue;
2425 if (!strcmp(arg, "-z")) {
2426 line_termination = 0;
2427 continue;
2429 if (!strncmp(arg, "-C", 2)) {
2430 p_context = strtoul(arg + 2, &end, 0);
2431 if (*end != '\0')
2432 die("unrecognized context count '%s'", arg + 2);
2433 continue;
2435 if (!strncmp(arg, "--whitespace=", 13)) {
2436 whitespace_option = arg + 13;
2437 parse_whitespace_option(arg + 13);
2438 continue;
2440 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2441 apply_in_reverse = 1;
2442 continue;
2444 if (!strcmp(arg, "--inaccurate-eof")) {
2445 inaccurate_eof = 1;
2446 continue;
2449 if (check_index && prefix_length < 0) {
2450 prefix = setup_git_directory();
2451 prefix_length = prefix ? strlen(prefix) : 0;
2452 git_config(git_apply_config);
2453 if (!whitespace_option && apply_default_whitespace)
2454 parse_whitespace_option(apply_default_whitespace);
2456 if (0 < prefix_length)
2457 arg = prefix_filename(prefix, prefix_length, arg);
2459 fd = open(arg, O_RDONLY);
2460 if (fd < 0)
2461 usage(apply_usage);
2462 read_stdin = 0;
2463 set_default_whitespace_mode(whitespace_option);
2464 apply_patch(fd, arg, inaccurate_eof);
2465 close(fd);
2467 set_default_whitespace_mode(whitespace_option);
2468 if (read_stdin)
2469 apply_patch(0, "<stdin>", inaccurate_eof);
2470 if (whitespace_error) {
2471 if (squelch_whitespace_errors &&
2472 squelch_whitespace_errors < whitespace_error) {
2473 int squelched =
2474 whitespace_error - squelch_whitespace_errors;
2475 fprintf(stderr, "warning: squelched %d whitespace error%s\n",
2476 squelched,
2477 squelched == 1 ? "" : "s");
2479 if (new_whitespace == error_on_whitespace)
2480 die("%d line%s add%s trailing whitespaces.",
2481 whitespace_error,
2482 whitespace_error == 1 ? "" : "s",
2483 whitespace_error == 1 ? "s" : "");
2484 if (applied_after_stripping)
2485 fprintf(stderr, "warning: %d line%s applied after"
2486 " stripping trailing whitespaces.\n",
2487 applied_after_stripping,
2488 applied_after_stripping == 1 ? "" : "s");
2489 else if (whitespace_error)
2490 fprintf(stderr, "warning: %d line%s add%s trailing"
2491 " whitespaces.\n",
2492 whitespace_error,
2493 whitespace_error == 1 ? "" : "s",
2494 whitespace_error == 1 ? "s" : "");
2497 if (write_index) {
2498 if (write_cache(newfd, active_cache, active_nr) ||
2499 close(newfd) || commit_lock_file(&lock_file))
2500 die("Unable to write new index file");
2503 return 0;