'git foo' program identifies itself without dash in die() messages
[git/dscho.git] / builtin-apply.c
blob4eb263ec5807a771b3a61c147b74a6f2f68dddee
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
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 unidiff_zero;
31 static int p_value = 1;
32 static int p_value_known;
33 static int check_index;
34 static int update_index;
35 static int cached;
36 static int diffstat;
37 static int numstat;
38 static int summary;
39 static int check;
40 static int apply = 1;
41 static int apply_in_reverse;
42 static int apply_with_reject;
43 static int apply_verbosely;
44 static int no_add;
45 static const char *fake_ancestor;
46 static int line_termination = '\n';
47 static unsigned long p_context = ULONG_MAX;
48 static const char apply_usage[] =
49 "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>...";
51 static enum ws_error_action {
52 nowarn_ws_error,
53 warn_on_ws_error,
54 die_on_ws_error,
55 correct_ws_error,
56 } ws_error_action = warn_on_ws_error;
57 static int whitespace_error;
58 static int squelch_whitespace_errors = 5;
59 static int applied_after_fixing_ws;
60 static const char *patch_input_file;
61 static const char *root;
62 static int root_len;
64 static void parse_whitespace_option(const char *option)
66 if (!option) {
67 ws_error_action = warn_on_ws_error;
68 return;
70 if (!strcmp(option, "warn")) {
71 ws_error_action = warn_on_ws_error;
72 return;
74 if (!strcmp(option, "nowarn")) {
75 ws_error_action = nowarn_ws_error;
76 return;
78 if (!strcmp(option, "error")) {
79 ws_error_action = die_on_ws_error;
80 return;
82 if (!strcmp(option, "error-all")) {
83 ws_error_action = die_on_ws_error;
84 squelch_whitespace_errors = 0;
85 return;
87 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
88 ws_error_action = correct_ws_error;
89 return;
91 die("unrecognized whitespace option '%s'", option);
94 static void set_default_whitespace_mode(const char *whitespace_option)
96 if (!whitespace_option && !apply_default_whitespace)
97 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
101 * For "diff-stat" like behaviour, we keep track of the biggest change
102 * we've seen, and the longest filename. That allows us to do simple
103 * scaling.
105 static int max_change, max_len;
108 * Various "current state", notably line numbers and what
109 * file (and how) we're patching right now.. The "is_xxxx"
110 * things are flags, where -1 means "don't know yet".
112 static int linenr = 1;
115 * This represents one "hunk" from a patch, starting with
116 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
117 * patch text is pointed at by patch, and its byte length
118 * is stored in size. leading and trailing are the number
119 * of context lines.
121 struct fragment {
122 unsigned long leading, trailing;
123 unsigned long oldpos, oldlines;
124 unsigned long newpos, newlines;
125 const char *patch;
126 int size;
127 int rejected;
128 struct fragment *next;
132 * When dealing with a binary patch, we reuse "leading" field
133 * to store the type of the binary hunk, either deflated "delta"
134 * or deflated "literal".
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED 1
138 #define BINARY_LITERAL_DEFLATED 2
141 * This represents a "patch" to a file, both metainfo changes
142 * such as creation/deletion, filemode and content changes represented
143 * as a series of fragments.
145 struct patch {
146 char *new_name, *old_name, *def_name;
147 unsigned int old_mode, new_mode;
148 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
149 int rejected;
150 unsigned ws_rule;
151 unsigned long deflate_origlen;
152 int lines_added, lines_deleted;
153 int score;
154 unsigned int is_toplevel_relative:1;
155 unsigned int inaccurate_eof:1;
156 unsigned int is_binary:1;
157 unsigned int is_copy:1;
158 unsigned int is_rename:1;
159 unsigned int recount:1;
160 struct fragment *fragments;
161 char *result;
162 size_t resultsize;
163 char old_sha1_prefix[41];
164 char new_sha1_prefix[41];
165 struct patch *next;
169 * A line in a file, len-bytes long (includes the terminating LF,
170 * except for an incomplete line at the end if the file ends with
171 * one), and its contents hashes to 'hash'.
173 struct line {
174 size_t len;
175 unsigned hash : 24;
176 unsigned flag : 8;
177 #define LINE_COMMON 1
181 * This represents a "file", which is an array of "lines".
183 struct image {
184 char *buf;
185 size_t len;
186 size_t nr;
187 size_t alloc;
188 struct line *line_allocated;
189 struct line *line;
193 * Records filenames that have been touched, in order to handle
194 * the case where more than one patches touch the same file.
197 static struct string_list fn_table;
199 static uint32_t hash_line(const char *cp, size_t len)
201 size_t i;
202 uint32_t h;
203 for (i = 0, h = 0; i < len; i++) {
204 if (!isspace(cp[i])) {
205 h = h * 3 + (cp[i] & 0xff);
208 return h;
211 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
213 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
214 img->line_allocated[img->nr].len = len;
215 img->line_allocated[img->nr].hash = hash_line(bol, len);
216 img->line_allocated[img->nr].flag = flag;
217 img->nr++;
220 static void prepare_image(struct image *image, char *buf, size_t len,
221 int prepare_linetable)
223 const char *cp, *ep;
225 memset(image, 0, sizeof(*image));
226 image->buf = buf;
227 image->len = len;
229 if (!prepare_linetable)
230 return;
232 ep = image->buf + image->len;
233 cp = image->buf;
234 while (cp < ep) {
235 const char *next;
236 for (next = cp; next < ep && *next != '\n'; next++)
238 if (next < ep)
239 next++;
240 add_line_info(image, cp, next - cp, 0);
241 cp = next;
243 image->line = image->line_allocated;
246 static void clear_image(struct image *image)
248 free(image->buf);
249 image->buf = NULL;
250 image->len = 0;
253 static void say_patch_name(FILE *output, const char *pre,
254 struct patch *patch, const char *post)
256 fputs(pre, output);
257 if (patch->old_name && patch->new_name &&
258 strcmp(patch->old_name, patch->new_name)) {
259 quote_c_style(patch->old_name, NULL, output, 0);
260 fputs(" => ", output);
261 quote_c_style(patch->new_name, NULL, output, 0);
262 } else {
263 const char *n = patch->new_name;
264 if (!n)
265 n = patch->old_name;
266 quote_c_style(n, NULL, output, 0);
268 fputs(post, output);
271 #define CHUNKSIZE (8192)
272 #define SLOP (16)
274 static void read_patch_file(struct strbuf *sb, int fd)
276 if (strbuf_read(sb, fd, 0) < 0)
277 die("git apply: read returned %s", strerror(errno));
280 * Make sure that we have some slop in the buffer
281 * so that we can do speculative "memcmp" etc, and
282 * see to it that it is NUL-filled.
284 strbuf_grow(sb, SLOP);
285 memset(sb->buf + sb->len, 0, SLOP);
288 static unsigned long linelen(const char *buffer, unsigned long size)
290 unsigned long len = 0;
291 while (size--) {
292 len++;
293 if (*buffer++ == '\n')
294 break;
296 return len;
299 static int is_dev_null(const char *str)
301 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
304 #define TERM_SPACE 1
305 #define TERM_TAB 2
307 static int name_terminate(const char *name, int namelen, int c, int terminate)
309 if (c == ' ' && !(terminate & TERM_SPACE))
310 return 0;
311 if (c == '\t' && !(terminate & TERM_TAB))
312 return 0;
314 return 1;
317 static char *find_name(const char *line, char *def, int p_value, int terminate)
319 int len;
320 const char *start = line;
322 if (*line == '"') {
323 struct strbuf name;
326 * Proposed "new-style" GNU patch/diff format; see
327 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
329 strbuf_init(&name, 0);
330 if (!unquote_c_style(&name, line, NULL)) {
331 char *cp;
333 for (cp = name.buf; p_value; p_value--) {
334 cp = strchr(cp, '/');
335 if (!cp)
336 break;
337 cp++;
339 if (cp) {
340 /* name can later be freed, so we need
341 * to memmove, not just return cp
343 strbuf_remove(&name, 0, cp - name.buf);
344 free(def);
345 if (root)
346 strbuf_insert(&name, 0, root, root_len);
347 return strbuf_detach(&name, NULL);
350 strbuf_release(&name);
353 for (;;) {
354 char c = *line;
356 if (isspace(c)) {
357 if (c == '\n')
358 break;
359 if (name_terminate(start, line-start, c, terminate))
360 break;
362 line++;
363 if (c == '/' && !--p_value)
364 start = line;
366 if (!start)
367 return def;
368 len = line - start;
369 if (!len)
370 return def;
373 * Generally we prefer the shorter name, especially
374 * if the other one is just a variation of that with
375 * something else tacked on to the end (ie "file.orig"
376 * or "file~").
378 if (def) {
379 int deflen = strlen(def);
380 if (deflen < len && !strncmp(start, def, deflen))
381 return def;
382 free(def);
385 if (root) {
386 char *ret = xmalloc(root_len + len + 1);
387 strcpy(ret, root);
388 memcpy(ret + root_len, start, len);
389 ret[root_len + len] = '\0';
390 return ret;
393 return xmemdupz(start, len);
396 static int count_slashes(const char *cp)
398 int cnt = 0;
399 char ch;
401 while ((ch = *cp++))
402 if (ch == '/')
403 cnt++;
404 return cnt;
408 * Given the string after "--- " or "+++ ", guess the appropriate
409 * p_value for the given patch.
411 static int guess_p_value(const char *nameline)
413 char *name, *cp;
414 int val = -1;
416 if (is_dev_null(nameline))
417 return -1;
418 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
419 if (!name)
420 return -1;
421 cp = strchr(name, '/');
422 if (!cp)
423 val = 0;
424 else if (prefix) {
426 * Does it begin with "a/$our-prefix" and such? Then this is
427 * very likely to apply to our directory.
429 if (!strncmp(name, prefix, prefix_length))
430 val = count_slashes(prefix);
431 else {
432 cp++;
433 if (!strncmp(cp, prefix, prefix_length))
434 val = count_slashes(prefix) + 1;
437 free(name);
438 return val;
442 * Get the name etc info from the ---/+++ lines of a traditional patch header
444 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
445 * files, we can happily check the index for a match, but for creating a
446 * new file we should try to match whatever "patch" does. I have no idea.
448 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
450 char *name;
452 first += 4; /* skip "--- " */
453 second += 4; /* skip "+++ " */
454 if (!p_value_known) {
455 int p, q;
456 p = guess_p_value(first);
457 q = guess_p_value(second);
458 if (p < 0) p = q;
459 if (0 <= p && p == q) {
460 p_value = p;
461 p_value_known = 1;
464 if (is_dev_null(first)) {
465 patch->is_new = 1;
466 patch->is_delete = 0;
467 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
468 patch->new_name = name;
469 } else if (is_dev_null(second)) {
470 patch->is_new = 0;
471 patch->is_delete = 1;
472 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
473 patch->old_name = name;
474 } else {
475 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
476 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
477 patch->old_name = patch->new_name = name;
479 if (!name)
480 die("unable to find filename in patch at line %d", linenr);
483 static int gitdiff_hdrend(const char *line, struct patch *patch)
485 return -1;
489 * We're anal about diff header consistency, to make
490 * sure that we don't end up having strange ambiguous
491 * patches floating around.
493 * As a result, gitdiff_{old|new}name() will check
494 * their names against any previous information, just
495 * to make sure..
497 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
499 if (!orig_name && !isnull)
500 return find_name(line, NULL, p_value, TERM_TAB);
502 if (orig_name) {
503 int len;
504 const char *name;
505 char *another;
506 name = orig_name;
507 len = strlen(name);
508 if (isnull)
509 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
510 another = find_name(line, NULL, p_value, TERM_TAB);
511 if (!another || memcmp(another, name, len))
512 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
513 free(another);
514 return orig_name;
516 else {
517 /* expect "/dev/null" */
518 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
519 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
520 return NULL;
524 static int gitdiff_oldname(const char *line, struct patch *patch)
526 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
527 return 0;
530 static int gitdiff_newname(const char *line, struct patch *patch)
532 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
533 return 0;
536 static int gitdiff_oldmode(const char *line, struct patch *patch)
538 patch->old_mode = strtoul(line, NULL, 8);
539 return 0;
542 static int gitdiff_newmode(const char *line, struct patch *patch)
544 patch->new_mode = strtoul(line, NULL, 8);
545 return 0;
548 static int gitdiff_delete(const char *line, struct patch *patch)
550 patch->is_delete = 1;
551 patch->old_name = patch->def_name;
552 return gitdiff_oldmode(line, patch);
555 static int gitdiff_newfile(const char *line, struct patch *patch)
557 patch->is_new = 1;
558 patch->new_name = patch->def_name;
559 return gitdiff_newmode(line, patch);
562 static int gitdiff_copysrc(const char *line, struct patch *patch)
564 patch->is_copy = 1;
565 patch->old_name = find_name(line, NULL, 0, 0);
566 return 0;
569 static int gitdiff_copydst(const char *line, struct patch *patch)
571 patch->is_copy = 1;
572 patch->new_name = find_name(line, NULL, 0, 0);
573 return 0;
576 static int gitdiff_renamesrc(const char *line, struct patch *patch)
578 patch->is_rename = 1;
579 patch->old_name = find_name(line, NULL, 0, 0);
580 return 0;
583 static int gitdiff_renamedst(const char *line, struct patch *patch)
585 patch->is_rename = 1;
586 patch->new_name = find_name(line, NULL, 0, 0);
587 return 0;
590 static int gitdiff_similarity(const char *line, struct patch *patch)
592 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
593 patch->score = 0;
594 return 0;
597 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
599 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
600 patch->score = 0;
601 return 0;
604 static int gitdiff_index(const char *line, struct patch *patch)
607 * index line is N hexadecimal, "..", N hexadecimal,
608 * and optional space with octal mode.
610 const char *ptr, *eol;
611 int len;
613 ptr = strchr(line, '.');
614 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
615 return 0;
616 len = ptr - line;
617 memcpy(patch->old_sha1_prefix, line, len);
618 patch->old_sha1_prefix[len] = 0;
620 line = ptr + 2;
621 ptr = strchr(line, ' ');
622 eol = strchr(line, '\n');
624 if (!ptr || eol < ptr)
625 ptr = eol;
626 len = ptr - line;
628 if (40 < len)
629 return 0;
630 memcpy(patch->new_sha1_prefix, line, len);
631 patch->new_sha1_prefix[len] = 0;
632 if (*ptr == ' ')
633 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
634 return 0;
638 * This is normal for a diff that doesn't change anything: we'll fall through
639 * into the next diff. Tell the parser to break out.
641 static int gitdiff_unrecognized(const char *line, struct patch *patch)
643 return -1;
646 static const char *stop_at_slash(const char *line, int llen)
648 int i;
650 for (i = 0; i < llen; i++) {
651 int ch = line[i];
652 if (ch == '/')
653 return line + i;
655 return NULL;
659 * This is to extract the same name that appears on "diff --git"
660 * line. We do not find and return anything if it is a rename
661 * patch, and it is OK because we will find the name elsewhere.
662 * We need to reliably find name only when it is mode-change only,
663 * creation or deletion of an empty file. In any of these cases,
664 * both sides are the same name under a/ and b/ respectively.
666 static char *git_header_name(char *line, int llen)
668 const char *name;
669 const char *second = NULL;
670 size_t len;
672 line += strlen("diff --git ");
673 llen -= strlen("diff --git ");
675 if (*line == '"') {
676 const char *cp;
677 struct strbuf first;
678 struct strbuf sp;
680 strbuf_init(&first, 0);
681 strbuf_init(&sp, 0);
683 if (unquote_c_style(&first, line, &second))
684 goto free_and_fail1;
686 /* advance to the first slash */
687 cp = stop_at_slash(first.buf, first.len);
688 /* we do not accept absolute paths */
689 if (!cp || cp == first.buf)
690 goto free_and_fail1;
691 strbuf_remove(&first, 0, cp + 1 - first.buf);
694 * second points at one past closing dq of name.
695 * find the second name.
697 while ((second < line + llen) && isspace(*second))
698 second++;
700 if (line + llen <= second)
701 goto free_and_fail1;
702 if (*second == '"') {
703 if (unquote_c_style(&sp, second, NULL))
704 goto free_and_fail1;
705 cp = stop_at_slash(sp.buf, sp.len);
706 if (!cp || cp == sp.buf)
707 goto free_and_fail1;
708 /* They must match, otherwise ignore */
709 if (strcmp(cp + 1, first.buf))
710 goto free_and_fail1;
711 strbuf_release(&sp);
712 return strbuf_detach(&first, NULL);
715 /* unquoted second */
716 cp = stop_at_slash(second, line + llen - second);
717 if (!cp || cp == second)
718 goto free_and_fail1;
719 cp++;
720 if (line + llen - cp != first.len + 1 ||
721 memcmp(first.buf, cp, first.len))
722 goto free_and_fail1;
723 return strbuf_detach(&first, NULL);
725 free_and_fail1:
726 strbuf_release(&first);
727 strbuf_release(&sp);
728 return NULL;
731 /* unquoted first name */
732 name = stop_at_slash(line, llen);
733 if (!name || name == line)
734 return NULL;
735 name++;
738 * since the first name is unquoted, a dq if exists must be
739 * the beginning of the second name.
741 for (second = name; second < line + llen; second++) {
742 if (*second == '"') {
743 struct strbuf sp;
744 const char *np;
746 strbuf_init(&sp, 0);
747 if (unquote_c_style(&sp, second, NULL))
748 goto free_and_fail2;
750 np = stop_at_slash(sp.buf, sp.len);
751 if (!np || np == sp.buf)
752 goto free_and_fail2;
753 np++;
755 len = sp.buf + sp.len - np;
756 if (len < second - name &&
757 !strncmp(np, name, len) &&
758 isspace(name[len])) {
759 /* Good */
760 strbuf_remove(&sp, 0, np - sp.buf);
761 return strbuf_detach(&sp, NULL);
764 free_and_fail2:
765 strbuf_release(&sp);
766 return NULL;
771 * Accept a name only if it shows up twice, exactly the same
772 * form.
774 for (len = 0 ; ; len++) {
775 switch (name[len]) {
776 default:
777 continue;
778 case '\n':
779 return NULL;
780 case '\t': case ' ':
781 second = name+len;
782 for (;;) {
783 char c = *second++;
784 if (c == '\n')
785 return NULL;
786 if (c == '/')
787 break;
789 if (second[len] == '\n' && !memcmp(name, second, len)) {
790 return xmemdupz(name, len);
796 /* Verify that we recognize the lines following a git header */
797 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
799 unsigned long offset;
801 /* A git diff has explicit new/delete information, so we don't guess */
802 patch->is_new = 0;
803 patch->is_delete = 0;
806 * Some things may not have the old name in the
807 * rest of the headers anywhere (pure mode changes,
808 * or removing or adding empty files), so we get
809 * the default name from the header.
811 patch->def_name = git_header_name(line, len);
813 line += len;
814 size -= len;
815 linenr++;
816 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
817 static const struct opentry {
818 const char *str;
819 int (*fn)(const char *, struct patch *);
820 } optable[] = {
821 { "@@ -", gitdiff_hdrend },
822 { "--- ", gitdiff_oldname },
823 { "+++ ", gitdiff_newname },
824 { "old mode ", gitdiff_oldmode },
825 { "new mode ", gitdiff_newmode },
826 { "deleted file mode ", gitdiff_delete },
827 { "new file mode ", gitdiff_newfile },
828 { "copy from ", gitdiff_copysrc },
829 { "copy to ", gitdiff_copydst },
830 { "rename old ", gitdiff_renamesrc },
831 { "rename new ", gitdiff_renamedst },
832 { "rename from ", gitdiff_renamesrc },
833 { "rename to ", gitdiff_renamedst },
834 { "similarity index ", gitdiff_similarity },
835 { "dissimilarity index ", gitdiff_dissimilarity },
836 { "index ", gitdiff_index },
837 { "", gitdiff_unrecognized },
839 int i;
841 len = linelen(line, size);
842 if (!len || line[len-1] != '\n')
843 break;
844 for (i = 0; i < ARRAY_SIZE(optable); i++) {
845 const struct opentry *p = optable + i;
846 int oplen = strlen(p->str);
847 if (len < oplen || memcmp(p->str, line, oplen))
848 continue;
849 if (p->fn(line + oplen, patch) < 0)
850 return offset;
851 break;
855 return offset;
858 static int parse_num(const char *line, unsigned long *p)
860 char *ptr;
862 if (!isdigit(*line))
863 return 0;
864 *p = strtoul(line, &ptr, 10);
865 return ptr - line;
868 static int parse_range(const char *line, int len, int offset, const char *expect,
869 unsigned long *p1, unsigned long *p2)
871 int digits, ex;
873 if (offset < 0 || offset >= len)
874 return -1;
875 line += offset;
876 len -= offset;
878 digits = parse_num(line, p1);
879 if (!digits)
880 return -1;
882 offset += digits;
883 line += digits;
884 len -= digits;
886 *p2 = 1;
887 if (*line == ',') {
888 digits = parse_num(line+1, p2);
889 if (!digits)
890 return -1;
892 offset += digits+1;
893 line += digits+1;
894 len -= digits+1;
897 ex = strlen(expect);
898 if (ex > len)
899 return -1;
900 if (memcmp(line, expect, ex))
901 return -1;
903 return offset + ex;
906 static void recount_diff(char *line, int size, struct fragment *fragment)
908 int oldlines = 0, newlines = 0, ret = 0;
910 if (size < 1) {
911 warning("recount: ignore empty hunk");
912 return;
915 for (;;) {
916 int len = linelen(line, size);
917 size -= len;
918 line += len;
920 if (size < 1)
921 break;
923 switch (*line) {
924 case ' ': case '\n':
925 newlines++;
926 /* fall through */
927 case '-':
928 oldlines++;
929 continue;
930 case '+':
931 newlines++;
932 continue;
933 case '\\':
934 continue;
935 case '@':
936 ret = size < 3 || prefixcmp(line, "@@ ");
937 break;
938 case 'd':
939 ret = size < 5 || prefixcmp(line, "diff ");
940 break;
941 default:
942 ret = -1;
943 break;
945 if (ret) {
946 warning("recount: unexpected line: %.*s",
947 (int)linelen(line, size), line);
948 return;
950 break;
952 fragment->oldlines = oldlines;
953 fragment->newlines = newlines;
957 * Parse a unified diff fragment header of the
958 * form "@@ -a,b +c,d @@"
960 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
962 int offset;
964 if (!len || line[len-1] != '\n')
965 return -1;
967 /* Figure out the number of lines in a fragment */
968 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
969 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
971 return offset;
974 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
976 unsigned long offset, len;
978 patch->is_toplevel_relative = 0;
979 patch->is_rename = patch->is_copy = 0;
980 patch->is_new = patch->is_delete = -1;
981 patch->old_mode = patch->new_mode = 0;
982 patch->old_name = patch->new_name = NULL;
983 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
984 unsigned long nextlen;
986 len = linelen(line, size);
987 if (!len)
988 break;
990 /* Testing this early allows us to take a few shortcuts.. */
991 if (len < 6)
992 continue;
995 * Make sure we don't find any unconnected patch fragments.
996 * That's a sign that we didn't find a header, and that a
997 * patch has become corrupted/broken up.
999 if (!memcmp("@@ -", line, 4)) {
1000 struct fragment dummy;
1001 if (parse_fragment_header(line, len, &dummy) < 0)
1002 continue;
1003 die("patch fragment without header at line %d: %.*s",
1004 linenr, (int)len-1, line);
1007 if (size < len + 6)
1008 break;
1011 * Git patch? It might not have a real patch, just a rename
1012 * or mode change, so we handle that specially
1014 if (!memcmp("diff --git ", line, 11)) {
1015 int git_hdr_len = parse_git_header(line, len, size, patch);
1016 if (git_hdr_len <= len)
1017 continue;
1018 if (!patch->old_name && !patch->new_name) {
1019 if (!patch->def_name)
1020 die("git diff header lacks filename information (line %d)", linenr);
1021 patch->old_name = patch->new_name = patch->def_name;
1023 patch->is_toplevel_relative = 1;
1024 *hdrsize = git_hdr_len;
1025 return offset;
1028 /* --- followed by +++ ? */
1029 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1030 continue;
1033 * We only accept unified patches, so we want it to
1034 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1035 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1037 nextlen = linelen(line + len, size - len);
1038 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1039 continue;
1041 /* Ok, we'll consider it a patch */
1042 parse_traditional_patch(line, line+len, patch);
1043 *hdrsize = len + nextlen;
1044 linenr += 2;
1045 return offset;
1047 return -1;
1050 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1052 char *err;
1053 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1054 if (!result)
1055 return;
1057 whitespace_error++;
1058 if (squelch_whitespace_errors &&
1059 squelch_whitespace_errors < whitespace_error)
1061 else {
1062 err = whitespace_error_string(result);
1063 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1064 patch_input_file, linenr, err, len - 2, line + 1);
1065 free(err);
1070 * Parse a unified diff. Note that this really needs to parse each
1071 * fragment separately, since the only way to know the difference
1072 * between a "---" that is part of a patch, and a "---" that starts
1073 * the next patch is to look at the line counts..
1075 static int parse_fragment(char *line, unsigned long size,
1076 struct patch *patch, struct fragment *fragment)
1078 int added, deleted;
1079 int len = linelen(line, size), offset;
1080 unsigned long oldlines, newlines;
1081 unsigned long leading, trailing;
1083 offset = parse_fragment_header(line, len, fragment);
1084 if (offset < 0)
1085 return -1;
1086 if (offset > 0 && patch->recount)
1087 recount_diff(line + offset, size - offset, fragment);
1088 oldlines = fragment->oldlines;
1089 newlines = fragment->newlines;
1090 leading = 0;
1091 trailing = 0;
1093 /* Parse the thing.. */
1094 line += len;
1095 size -= len;
1096 linenr++;
1097 added = deleted = 0;
1098 for (offset = len;
1099 0 < size;
1100 offset += len, size -= len, line += len, linenr++) {
1101 if (!oldlines && !newlines)
1102 break;
1103 len = linelen(line, size);
1104 if (!len || line[len-1] != '\n')
1105 return -1;
1106 switch (*line) {
1107 default:
1108 return -1;
1109 case '\n': /* newer GNU diff, an empty context line */
1110 case ' ':
1111 oldlines--;
1112 newlines--;
1113 if (!deleted && !added)
1114 leading++;
1115 trailing++;
1116 break;
1117 case '-':
1118 if (apply_in_reverse &&
1119 ws_error_action != nowarn_ws_error)
1120 check_whitespace(line, len, patch->ws_rule);
1121 deleted++;
1122 oldlines--;
1123 trailing = 0;
1124 break;
1125 case '+':
1126 if (!apply_in_reverse &&
1127 ws_error_action != nowarn_ws_error)
1128 check_whitespace(line, len, patch->ws_rule);
1129 added++;
1130 newlines--;
1131 trailing = 0;
1132 break;
1135 * We allow "\ No newline at end of file". Depending
1136 * on locale settings when the patch was produced we
1137 * don't know what this line looks like. The only
1138 * thing we do know is that it begins with "\ ".
1139 * Checking for 12 is just for sanity check -- any
1140 * l10n of "\ No newline..." is at least that long.
1142 case '\\':
1143 if (len < 12 || memcmp(line, "\\ ", 2))
1144 return -1;
1145 break;
1148 if (oldlines || newlines)
1149 return -1;
1150 fragment->leading = leading;
1151 fragment->trailing = trailing;
1154 * If a fragment ends with an incomplete line, we failed to include
1155 * it in the above loop because we hit oldlines == newlines == 0
1156 * before seeing it.
1158 if (12 < size && !memcmp(line, "\\ ", 2))
1159 offset += linelen(line, size);
1161 patch->lines_added += added;
1162 patch->lines_deleted += deleted;
1164 if (0 < patch->is_new && oldlines)
1165 return error("new file depends on old contents");
1166 if (0 < patch->is_delete && newlines)
1167 return error("deleted file still has contents");
1168 return offset;
1171 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1173 unsigned long offset = 0;
1174 unsigned long oldlines = 0, newlines = 0, context = 0;
1175 struct fragment **fragp = &patch->fragments;
1177 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1178 struct fragment *fragment;
1179 int len;
1181 fragment = xcalloc(1, sizeof(*fragment));
1182 len = parse_fragment(line, size, patch, fragment);
1183 if (len <= 0)
1184 die("corrupt patch at line %d", linenr);
1185 fragment->patch = line;
1186 fragment->size = len;
1187 oldlines += fragment->oldlines;
1188 newlines += fragment->newlines;
1189 context += fragment->leading + fragment->trailing;
1191 *fragp = fragment;
1192 fragp = &fragment->next;
1194 offset += len;
1195 line += len;
1196 size -= len;
1200 * If something was removed (i.e. we have old-lines) it cannot
1201 * be creation, and if something was added it cannot be
1202 * deletion. However, the reverse is not true; --unified=0
1203 * patches that only add are not necessarily creation even
1204 * though they do not have any old lines, and ones that only
1205 * delete are not necessarily deletion.
1207 * Unfortunately, a real creation/deletion patch do _not_ have
1208 * any context line by definition, so we cannot safely tell it
1209 * apart with --unified=0 insanity. At least if the patch has
1210 * more than one hunk it is not creation or deletion.
1212 if (patch->is_new < 0 &&
1213 (oldlines || (patch->fragments && patch->fragments->next)))
1214 patch->is_new = 0;
1215 if (patch->is_delete < 0 &&
1216 (newlines || (patch->fragments && patch->fragments->next)))
1217 patch->is_delete = 0;
1219 if (0 < patch->is_new && oldlines)
1220 die("new file %s depends on old contents", patch->new_name);
1221 if (0 < patch->is_delete && newlines)
1222 die("deleted file %s still has contents", patch->old_name);
1223 if (!patch->is_delete && !newlines && context)
1224 fprintf(stderr, "** warning: file %s becomes empty but "
1225 "is not deleted\n", patch->new_name);
1227 return offset;
1230 static inline int metadata_changes(struct patch *patch)
1232 return patch->is_rename > 0 ||
1233 patch->is_copy > 0 ||
1234 patch->is_new > 0 ||
1235 patch->is_delete ||
1236 (patch->old_mode && patch->new_mode &&
1237 patch->old_mode != patch->new_mode);
1240 static char *inflate_it(const void *data, unsigned long size,
1241 unsigned long inflated_size)
1243 z_stream stream;
1244 void *out;
1245 int st;
1247 memset(&stream, 0, sizeof(stream));
1249 stream.next_in = (unsigned char *)data;
1250 stream.avail_in = size;
1251 stream.next_out = out = xmalloc(inflated_size);
1252 stream.avail_out = inflated_size;
1253 inflateInit(&stream);
1254 st = inflate(&stream, Z_FINISH);
1255 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1256 free(out);
1257 return NULL;
1259 return out;
1262 static struct fragment *parse_binary_hunk(char **buf_p,
1263 unsigned long *sz_p,
1264 int *status_p,
1265 int *used_p)
1268 * Expect a line that begins with binary patch method ("literal"
1269 * or "delta"), followed by the length of data before deflating.
1270 * a sequence of 'length-byte' followed by base-85 encoded data
1271 * should follow, terminated by a newline.
1273 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1274 * and we would limit the patch line to 66 characters,
1275 * so one line can fit up to 13 groups that would decode
1276 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1277 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1279 int llen, used;
1280 unsigned long size = *sz_p;
1281 char *buffer = *buf_p;
1282 int patch_method;
1283 unsigned long origlen;
1284 char *data = NULL;
1285 int hunk_size = 0;
1286 struct fragment *frag;
1288 llen = linelen(buffer, size);
1289 used = llen;
1291 *status_p = 0;
1293 if (!prefixcmp(buffer, "delta ")) {
1294 patch_method = BINARY_DELTA_DEFLATED;
1295 origlen = strtoul(buffer + 6, NULL, 10);
1297 else if (!prefixcmp(buffer, "literal ")) {
1298 patch_method = BINARY_LITERAL_DEFLATED;
1299 origlen = strtoul(buffer + 8, NULL, 10);
1301 else
1302 return NULL;
1304 linenr++;
1305 buffer += llen;
1306 while (1) {
1307 int byte_length, max_byte_length, newsize;
1308 llen = linelen(buffer, size);
1309 used += llen;
1310 linenr++;
1311 if (llen == 1) {
1312 /* consume the blank line */
1313 buffer++;
1314 size--;
1315 break;
1318 * Minimum line is "A00000\n" which is 7-byte long,
1319 * and the line length must be multiple of 5 plus 2.
1321 if ((llen < 7) || (llen-2) % 5)
1322 goto corrupt;
1323 max_byte_length = (llen - 2) / 5 * 4;
1324 byte_length = *buffer;
1325 if ('A' <= byte_length && byte_length <= 'Z')
1326 byte_length = byte_length - 'A' + 1;
1327 else if ('a' <= byte_length && byte_length <= 'z')
1328 byte_length = byte_length - 'a' + 27;
1329 else
1330 goto corrupt;
1331 /* if the input length was not multiple of 4, we would
1332 * have filler at the end but the filler should never
1333 * exceed 3 bytes
1335 if (max_byte_length < byte_length ||
1336 byte_length <= max_byte_length - 4)
1337 goto corrupt;
1338 newsize = hunk_size + byte_length;
1339 data = xrealloc(data, newsize);
1340 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1341 goto corrupt;
1342 hunk_size = newsize;
1343 buffer += llen;
1344 size -= llen;
1347 frag = xcalloc(1, sizeof(*frag));
1348 frag->patch = inflate_it(data, hunk_size, origlen);
1349 if (!frag->patch)
1350 goto corrupt;
1351 free(data);
1352 frag->size = origlen;
1353 *buf_p = buffer;
1354 *sz_p = size;
1355 *used_p = used;
1356 frag->binary_patch_method = patch_method;
1357 return frag;
1359 corrupt:
1360 free(data);
1361 *status_p = -1;
1362 error("corrupt binary patch at line %d: %.*s",
1363 linenr-1, llen-1, buffer);
1364 return NULL;
1367 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1370 * We have read "GIT binary patch\n"; what follows is a line
1371 * that says the patch method (currently, either "literal" or
1372 * "delta") and the length of data before deflating; a
1373 * sequence of 'length-byte' followed by base-85 encoded data
1374 * follows.
1376 * When a binary patch is reversible, there is another binary
1377 * hunk in the same format, starting with patch method (either
1378 * "literal" or "delta") with the length of data, and a sequence
1379 * of length-byte + base-85 encoded data, terminated with another
1380 * empty line. This data, when applied to the postimage, produces
1381 * the preimage.
1383 struct fragment *forward;
1384 struct fragment *reverse;
1385 int status;
1386 int used, used_1;
1388 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1389 if (!forward && !status)
1390 /* there has to be one hunk (forward hunk) */
1391 return error("unrecognized binary patch at line %d", linenr-1);
1392 if (status)
1393 /* otherwise we already gave an error message */
1394 return status;
1396 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1397 if (reverse)
1398 used += used_1;
1399 else if (status) {
1401 * Not having reverse hunk is not an error, but having
1402 * a corrupt reverse hunk is.
1404 free((void*) forward->patch);
1405 free(forward);
1406 return status;
1408 forward->next = reverse;
1409 patch->fragments = forward;
1410 patch->is_binary = 1;
1411 return used;
1414 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1416 int hdrsize, patchsize;
1417 int offset = find_header(buffer, size, &hdrsize, patch);
1419 if (offset < 0)
1420 return offset;
1422 patch->ws_rule = whitespace_rule(patch->new_name
1423 ? patch->new_name
1424 : patch->old_name);
1426 patchsize = parse_single_patch(buffer + offset + hdrsize,
1427 size - offset - hdrsize, patch);
1429 if (!patchsize) {
1430 static const char *binhdr[] = {
1431 "Binary files ",
1432 "Files ",
1433 NULL,
1435 static const char git_binary[] = "GIT binary patch\n";
1436 int i;
1437 int hd = hdrsize + offset;
1438 unsigned long llen = linelen(buffer + hd, size - hd);
1440 if (llen == sizeof(git_binary) - 1 &&
1441 !memcmp(git_binary, buffer + hd, llen)) {
1442 int used;
1443 linenr++;
1444 used = parse_binary(buffer + hd + llen,
1445 size - hd - llen, patch);
1446 if (used)
1447 patchsize = used + llen;
1448 else
1449 patchsize = 0;
1451 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1452 for (i = 0; binhdr[i]; i++) {
1453 int len = strlen(binhdr[i]);
1454 if (len < size - hd &&
1455 !memcmp(binhdr[i], buffer + hd, len)) {
1456 linenr++;
1457 patch->is_binary = 1;
1458 patchsize = llen;
1459 break;
1464 /* Empty patch cannot be applied if it is a text patch
1465 * without metadata change. A binary patch appears
1466 * empty to us here.
1468 if ((apply || check) &&
1469 (!patch->is_binary && !metadata_changes(patch)))
1470 die("patch with only garbage at line %d", linenr);
1473 return offset + hdrsize + patchsize;
1476 #define swap(a,b) myswap((a),(b),sizeof(a))
1478 #define myswap(a, b, size) do { \
1479 unsigned char mytmp[size]; \
1480 memcpy(mytmp, &a, size); \
1481 memcpy(&a, &b, size); \
1482 memcpy(&b, mytmp, size); \
1483 } while (0)
1485 static void reverse_patches(struct patch *p)
1487 for (; p; p = p->next) {
1488 struct fragment *frag = p->fragments;
1490 swap(p->new_name, p->old_name);
1491 swap(p->new_mode, p->old_mode);
1492 swap(p->is_new, p->is_delete);
1493 swap(p->lines_added, p->lines_deleted);
1494 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1496 for (; frag; frag = frag->next) {
1497 swap(frag->newpos, frag->oldpos);
1498 swap(frag->newlines, frag->oldlines);
1503 static const char pluses[] =
1504 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1505 static const char minuses[]=
1506 "----------------------------------------------------------------------";
1508 static void show_stats(struct patch *patch)
1510 struct strbuf qname;
1511 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1512 int max, add, del;
1514 strbuf_init(&qname, 0);
1515 quote_c_style(cp, &qname, NULL, 0);
1518 * "scale" the filename
1520 max = max_len;
1521 if (max > 50)
1522 max = 50;
1524 if (qname.len > max) {
1525 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1526 if (!cp)
1527 cp = qname.buf + qname.len + 3 - max;
1528 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1531 if (patch->is_binary) {
1532 printf(" %-*s | Bin\n", max, qname.buf);
1533 strbuf_release(&qname);
1534 return;
1537 printf(" %-*s |", max, qname.buf);
1538 strbuf_release(&qname);
1541 * scale the add/delete
1543 max = max + max_change > 70 ? 70 - max : max_change;
1544 add = patch->lines_added;
1545 del = patch->lines_deleted;
1547 if (max_change > 0) {
1548 int total = ((add + del) * max + max_change / 2) / max_change;
1549 add = (add * max + max_change / 2) / max_change;
1550 del = total - add;
1552 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1553 add, pluses, del, minuses);
1556 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1558 switch (st->st_mode & S_IFMT) {
1559 case S_IFLNK:
1560 strbuf_grow(buf, st->st_size);
1561 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1562 return -1;
1563 strbuf_setlen(buf, st->st_size);
1564 return 0;
1565 case S_IFREG:
1566 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1567 return error("unable to open or read %s", path);
1568 convert_to_git(path, buf->buf, buf->len, buf, 0);
1569 return 0;
1570 default:
1571 return -1;
1575 static void update_pre_post_images(struct image *preimage,
1576 struct image *postimage,
1577 char *buf,
1578 size_t len)
1580 int i, ctx;
1581 char *new, *old, *fixed;
1582 struct image fixed_preimage;
1585 * Update the preimage with whitespace fixes. Note that we
1586 * are not losing preimage->buf -- apply_one_fragment() will
1587 * free "oldlines".
1589 prepare_image(&fixed_preimage, buf, len, 1);
1590 assert(fixed_preimage.nr == preimage->nr);
1591 for (i = 0; i < preimage->nr; i++)
1592 fixed_preimage.line[i].flag = preimage->line[i].flag;
1593 free(preimage->line_allocated);
1594 *preimage = fixed_preimage;
1597 * Adjust the common context lines in postimage, in place.
1598 * This is possible because whitespace fixing does not make
1599 * the string grow.
1601 new = old = postimage->buf;
1602 fixed = preimage->buf;
1603 for (i = ctx = 0; i < postimage->nr; i++) {
1604 size_t len = postimage->line[i].len;
1605 if (!(postimage->line[i].flag & LINE_COMMON)) {
1606 /* an added line -- no counterparts in preimage */
1607 memmove(new, old, len);
1608 old += len;
1609 new += len;
1610 continue;
1613 /* a common context -- skip it in the original postimage */
1614 old += len;
1616 /* and find the corresponding one in the fixed preimage */
1617 while (ctx < preimage->nr &&
1618 !(preimage->line[ctx].flag & LINE_COMMON)) {
1619 fixed += preimage->line[ctx].len;
1620 ctx++;
1622 if (preimage->nr <= ctx)
1623 die("oops");
1625 /* and copy it in, while fixing the line length */
1626 len = preimage->line[ctx].len;
1627 memcpy(new, fixed, len);
1628 new += len;
1629 fixed += len;
1630 postimage->line[i].len = len;
1631 ctx++;
1634 /* Fix the length of the whole thing */
1635 postimage->len = new - postimage->buf;
1638 static int match_fragment(struct image *img,
1639 struct image *preimage,
1640 struct image *postimage,
1641 unsigned long try,
1642 int try_lno,
1643 unsigned ws_rule,
1644 int match_beginning, int match_end)
1646 int i;
1647 char *fixed_buf, *buf, *orig, *target;
1649 if (preimage->nr + try_lno > img->nr)
1650 return 0;
1652 if (match_beginning && try_lno)
1653 return 0;
1655 if (match_end && preimage->nr + try_lno != img->nr)
1656 return 0;
1658 /* Quick hash check */
1659 for (i = 0; i < preimage->nr; i++)
1660 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1661 return 0;
1664 * Do we have an exact match? If we were told to match
1665 * at the end, size must be exactly at try+fragsize,
1666 * otherwise try+fragsize must be still within the preimage,
1667 * and either case, the old piece should match the preimage
1668 * exactly.
1670 if ((match_end
1671 ? (try + preimage->len == img->len)
1672 : (try + preimage->len <= img->len)) &&
1673 !memcmp(img->buf + try, preimage->buf, preimage->len))
1674 return 1;
1676 if (ws_error_action != correct_ws_error)
1677 return 0;
1680 * The hunk does not apply byte-by-byte, but the hash says
1681 * it might with whitespace fuzz.
1683 fixed_buf = xmalloc(preimage->len + 1);
1684 buf = fixed_buf;
1685 orig = preimage->buf;
1686 target = img->buf + try;
1687 for (i = 0; i < preimage->nr; i++) {
1688 size_t fixlen; /* length after fixing the preimage */
1689 size_t oldlen = preimage->line[i].len;
1690 size_t tgtlen = img->line[try_lno + i].len;
1691 size_t tgtfixlen; /* length after fixing the target line */
1692 char tgtfixbuf[1024], *tgtfix;
1693 int match;
1695 /* Try fixing the line in the preimage */
1696 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1698 /* Try fixing the line in the target */
1699 if (sizeof(tgtfixbuf) < tgtlen)
1700 tgtfix = tgtfixbuf;
1701 else
1702 tgtfix = xmalloc(tgtlen);
1703 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1706 * If they match, either the preimage was based on
1707 * a version before our tree fixed whitespace breakage,
1708 * or we are lacking a whitespace-fix patch the tree
1709 * the preimage was based on already had (i.e. target
1710 * has whitespace breakage, the preimage doesn't).
1711 * In either case, we are fixing the whitespace breakages
1712 * so we might as well take the fix together with their
1713 * real change.
1715 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1717 if (tgtfix != tgtfixbuf)
1718 free(tgtfix);
1719 if (!match)
1720 goto unmatch_exit;
1722 orig += oldlen;
1723 buf += fixlen;
1724 target += tgtlen;
1728 * Yes, the preimage is based on an older version that still
1729 * has whitespace breakages unfixed, and fixing them makes the
1730 * hunk match. Update the context lines in the postimage.
1732 update_pre_post_images(preimage, postimage,
1733 fixed_buf, buf - fixed_buf);
1734 return 1;
1736 unmatch_exit:
1737 free(fixed_buf);
1738 return 0;
1741 static int find_pos(struct image *img,
1742 struct image *preimage,
1743 struct image *postimage,
1744 int line,
1745 unsigned ws_rule,
1746 int match_beginning, int match_end)
1748 int i;
1749 unsigned long backwards, forwards, try;
1750 int backwards_lno, forwards_lno, try_lno;
1752 if (preimage->nr > img->nr)
1753 return -1;
1756 * If match_begining or match_end is specified, there is no
1757 * point starting from a wrong line that will never match and
1758 * wander around and wait for a match at the specified end.
1760 if (match_beginning)
1761 line = 0;
1762 else if (match_end)
1763 line = img->nr - preimage->nr;
1765 if (line > img->nr)
1766 line = img->nr;
1768 try = 0;
1769 for (i = 0; i < line; i++)
1770 try += img->line[i].len;
1773 * There's probably some smart way to do this, but I'll leave
1774 * that to the smart and beautiful people. I'm simple and stupid.
1776 backwards = try;
1777 backwards_lno = line;
1778 forwards = try;
1779 forwards_lno = line;
1780 try_lno = line;
1782 for (i = 0; ; i++) {
1783 if (match_fragment(img, preimage, postimage,
1784 try, try_lno, ws_rule,
1785 match_beginning, match_end))
1786 return try_lno;
1788 again:
1789 if (backwards_lno == 0 && forwards_lno == img->nr)
1790 break;
1792 if (i & 1) {
1793 if (backwards_lno == 0) {
1794 i++;
1795 goto again;
1797 backwards_lno--;
1798 backwards -= img->line[backwards_lno].len;
1799 try = backwards;
1800 try_lno = backwards_lno;
1801 } else {
1802 if (forwards_lno == img->nr) {
1803 i++;
1804 goto again;
1806 forwards += img->line[forwards_lno].len;
1807 forwards_lno++;
1808 try = forwards;
1809 try_lno = forwards_lno;
1813 return -1;
1816 static void remove_first_line(struct image *img)
1818 img->buf += img->line[0].len;
1819 img->len -= img->line[0].len;
1820 img->line++;
1821 img->nr--;
1824 static void remove_last_line(struct image *img)
1826 img->len -= img->line[--img->nr].len;
1829 static void update_image(struct image *img,
1830 int applied_pos,
1831 struct image *preimage,
1832 struct image *postimage)
1835 * remove the copy of preimage at offset in img
1836 * and replace it with postimage
1838 int i, nr;
1839 size_t remove_count, insert_count, applied_at = 0;
1840 char *result;
1842 for (i = 0; i < applied_pos; i++)
1843 applied_at += img->line[i].len;
1845 remove_count = 0;
1846 for (i = 0; i < preimage->nr; i++)
1847 remove_count += img->line[applied_pos + i].len;
1848 insert_count = postimage->len;
1850 /* Adjust the contents */
1851 result = xmalloc(img->len + insert_count - remove_count + 1);
1852 memcpy(result, img->buf, applied_at);
1853 memcpy(result + applied_at, postimage->buf, postimage->len);
1854 memcpy(result + applied_at + postimage->len,
1855 img->buf + (applied_at + remove_count),
1856 img->len - (applied_at + remove_count));
1857 free(img->buf);
1858 img->buf = result;
1859 img->len += insert_count - remove_count;
1860 result[img->len] = '\0';
1862 /* Adjust the line table */
1863 nr = img->nr + postimage->nr - preimage->nr;
1864 if (preimage->nr < postimage->nr) {
1866 * NOTE: this knows that we never call remove_first_line()
1867 * on anything other than pre/post image.
1869 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1870 img->line_allocated = img->line;
1872 if (preimage->nr != postimage->nr)
1873 memmove(img->line + applied_pos + postimage->nr,
1874 img->line + applied_pos + preimage->nr,
1875 (img->nr - (applied_pos + preimage->nr)) *
1876 sizeof(*img->line));
1877 memcpy(img->line + applied_pos,
1878 postimage->line,
1879 postimage->nr * sizeof(*img->line));
1880 img->nr = nr;
1883 static int apply_one_fragment(struct image *img, struct fragment *frag,
1884 int inaccurate_eof, unsigned ws_rule)
1886 int match_beginning, match_end;
1887 const char *patch = frag->patch;
1888 int size = frag->size;
1889 char *old, *new, *oldlines, *newlines;
1890 int new_blank_lines_at_end = 0;
1891 unsigned long leading, trailing;
1892 int pos, applied_pos;
1893 struct image preimage;
1894 struct image postimage;
1896 memset(&preimage, 0, sizeof(preimage));
1897 memset(&postimage, 0, sizeof(postimage));
1898 oldlines = xmalloc(size);
1899 newlines = xmalloc(size);
1901 old = oldlines;
1902 new = newlines;
1903 while (size > 0) {
1904 char first;
1905 int len = linelen(patch, size);
1906 int plen, added;
1907 int added_blank_line = 0;
1909 if (!len)
1910 break;
1913 * "plen" is how much of the line we should use for
1914 * the actual patch data. Normally we just remove the
1915 * first character on the line, but if the line is
1916 * followed by "\ No newline", then we also remove the
1917 * last one (which is the newline, of course).
1919 plen = len - 1;
1920 if (len < size && patch[len] == '\\')
1921 plen--;
1922 first = *patch;
1923 if (apply_in_reverse) {
1924 if (first == '-')
1925 first = '+';
1926 else if (first == '+')
1927 first = '-';
1930 switch (first) {
1931 case '\n':
1932 /* Newer GNU diff, empty context line */
1933 if (plen < 0)
1934 /* ... followed by '\No newline'; nothing */
1935 break;
1936 *old++ = '\n';
1937 *new++ = '\n';
1938 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1939 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1940 break;
1941 case ' ':
1942 case '-':
1943 memcpy(old, patch + 1, plen);
1944 add_line_info(&preimage, old, plen,
1945 (first == ' ' ? LINE_COMMON : 0));
1946 old += plen;
1947 if (first == '-')
1948 break;
1949 /* Fall-through for ' ' */
1950 case '+':
1951 /* --no-add does not add new lines */
1952 if (first == '+' && no_add)
1953 break;
1955 if (first != '+' ||
1956 !whitespace_error ||
1957 ws_error_action != correct_ws_error) {
1958 memcpy(new, patch + 1, plen);
1959 added = plen;
1961 else {
1962 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1964 add_line_info(&postimage, new, added,
1965 (first == '+' ? 0 : LINE_COMMON));
1966 new += added;
1967 if (first == '+' &&
1968 added == 1 && new[-1] == '\n')
1969 added_blank_line = 1;
1970 break;
1971 case '@': case '\\':
1972 /* Ignore it, we already handled it */
1973 break;
1974 default:
1975 if (apply_verbosely)
1976 error("invalid start of line: '%c'", first);
1977 return -1;
1979 if (added_blank_line)
1980 new_blank_lines_at_end++;
1981 else
1982 new_blank_lines_at_end = 0;
1983 patch += len;
1984 size -= len;
1986 if (inaccurate_eof &&
1987 old > oldlines && old[-1] == '\n' &&
1988 new > newlines && new[-1] == '\n') {
1989 old--;
1990 new--;
1993 leading = frag->leading;
1994 trailing = frag->trailing;
1997 * A hunk to change lines at the beginning would begin with
1998 * @@ -1,L +N,M @@
2000 * And a hunk to add to an empty file would begin with
2001 * @@ -0,0 +N,M @@
2003 * In other words, a hunk that is (frag->oldpos <= 1) with or
2004 * without leading context must match at the beginning.
2006 match_beginning = frag->oldpos <= 1;
2009 * A hunk without trailing lines must match at the end.
2010 * However, we simply cannot tell if a hunk must match end
2011 * from the lack of trailing lines if the patch was generated
2012 * with unidiff without any context.
2014 match_end = !unidiff_zero && !trailing;
2016 pos = frag->newpos ? (frag->newpos - 1) : 0;
2017 preimage.buf = oldlines;
2018 preimage.len = old - oldlines;
2019 postimage.buf = newlines;
2020 postimage.len = new - newlines;
2021 preimage.line = preimage.line_allocated;
2022 postimage.line = postimage.line_allocated;
2024 for (;;) {
2026 applied_pos = find_pos(img, &preimage, &postimage, pos,
2027 ws_rule, match_beginning, match_end);
2029 if (applied_pos >= 0)
2030 break;
2032 /* Am I at my context limits? */
2033 if ((leading <= p_context) && (trailing <= p_context))
2034 break;
2035 if (match_beginning || match_end) {
2036 match_beginning = match_end = 0;
2037 continue;
2041 * Reduce the number of context lines; reduce both
2042 * leading and trailing if they are equal otherwise
2043 * just reduce the larger context.
2045 if (leading >= trailing) {
2046 remove_first_line(&preimage);
2047 remove_first_line(&postimage);
2048 pos--;
2049 leading--;
2051 if (trailing > leading) {
2052 remove_last_line(&preimage);
2053 remove_last_line(&postimage);
2054 trailing--;
2058 if (applied_pos >= 0) {
2059 if (ws_error_action == correct_ws_error &&
2060 new_blank_lines_at_end &&
2061 postimage.nr + applied_pos == img->nr) {
2063 * If the patch application adds blank lines
2064 * at the end, and if the patch applies at the
2065 * end of the image, remove those added blank
2066 * lines.
2068 while (new_blank_lines_at_end--)
2069 remove_last_line(&postimage);
2073 * Warn if it was necessary to reduce the number
2074 * of context lines.
2076 if ((leading != frag->leading) ||
2077 (trailing != frag->trailing))
2078 fprintf(stderr, "Context reduced to (%ld/%ld)"
2079 " to apply fragment at %d\n",
2080 leading, trailing, applied_pos+1);
2081 update_image(img, applied_pos, &preimage, &postimage);
2082 } else {
2083 if (apply_verbosely)
2084 error("while searching for:\n%.*s",
2085 (int)(old - oldlines), oldlines);
2088 free(oldlines);
2089 free(newlines);
2090 free(preimage.line_allocated);
2091 free(postimage.line_allocated);
2093 return (applied_pos < 0);
2096 static int apply_binary_fragment(struct image *img, struct patch *patch)
2098 struct fragment *fragment = patch->fragments;
2099 unsigned long len;
2100 void *dst;
2102 /* Binary patch is irreversible without the optional second hunk */
2103 if (apply_in_reverse) {
2104 if (!fragment->next)
2105 return error("cannot reverse-apply a binary patch "
2106 "without the reverse hunk to '%s'",
2107 patch->new_name
2108 ? patch->new_name : patch->old_name);
2109 fragment = fragment->next;
2111 switch (fragment->binary_patch_method) {
2112 case BINARY_DELTA_DEFLATED:
2113 dst = patch_delta(img->buf, img->len, fragment->patch,
2114 fragment->size, &len);
2115 if (!dst)
2116 return -1;
2117 clear_image(img);
2118 img->buf = dst;
2119 img->len = len;
2120 return 0;
2121 case BINARY_LITERAL_DEFLATED:
2122 clear_image(img);
2123 img->len = fragment->size;
2124 img->buf = xmalloc(img->len+1);
2125 memcpy(img->buf, fragment->patch, img->len);
2126 img->buf[img->len] = '\0';
2127 return 0;
2129 return -1;
2132 static int apply_binary(struct image *img, struct patch *patch)
2134 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2135 unsigned char sha1[20];
2138 * For safety, we require patch index line to contain
2139 * full 40-byte textual SHA1 for old and new, at least for now.
2141 if (strlen(patch->old_sha1_prefix) != 40 ||
2142 strlen(patch->new_sha1_prefix) != 40 ||
2143 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2144 get_sha1_hex(patch->new_sha1_prefix, sha1))
2145 return error("cannot apply binary patch to '%s' "
2146 "without full index line", name);
2148 if (patch->old_name) {
2150 * See if the old one matches what the patch
2151 * applies to.
2153 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2154 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2155 return error("the patch applies to '%s' (%s), "
2156 "which does not match the "
2157 "current contents.",
2158 name, sha1_to_hex(sha1));
2160 else {
2161 /* Otherwise, the old one must be empty. */
2162 if (img->len)
2163 return error("the patch applies to an empty "
2164 "'%s' but it is not empty", name);
2167 get_sha1_hex(patch->new_sha1_prefix, sha1);
2168 if (is_null_sha1(sha1)) {
2169 clear_image(img);
2170 return 0; /* deletion patch */
2173 if (has_sha1_file(sha1)) {
2174 /* We already have the postimage */
2175 enum object_type type;
2176 unsigned long size;
2177 char *result;
2179 result = read_sha1_file(sha1, &type, &size);
2180 if (!result)
2181 return error("the necessary postimage %s for "
2182 "'%s' cannot be read",
2183 patch->new_sha1_prefix, name);
2184 clear_image(img);
2185 img->buf = result;
2186 img->len = size;
2187 } else {
2189 * We have verified buf matches the preimage;
2190 * apply the patch data to it, which is stored
2191 * in the patch->fragments->{patch,size}.
2193 if (apply_binary_fragment(img, patch))
2194 return error("binary patch does not apply to '%s'",
2195 name);
2197 /* verify that the result matches */
2198 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2199 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2200 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2201 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2204 return 0;
2207 static int apply_fragments(struct image *img, struct patch *patch)
2209 struct fragment *frag = patch->fragments;
2210 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2211 unsigned ws_rule = patch->ws_rule;
2212 unsigned inaccurate_eof = patch->inaccurate_eof;
2214 if (patch->is_binary)
2215 return apply_binary(img, patch);
2217 while (frag) {
2218 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2219 error("patch failed: %s:%ld", name, frag->oldpos);
2220 if (!apply_with_reject)
2221 return -1;
2222 frag->rejected = 1;
2224 frag = frag->next;
2226 return 0;
2229 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2231 if (!ce)
2232 return 0;
2234 if (S_ISGITLINK(ce->ce_mode)) {
2235 strbuf_grow(buf, 100);
2236 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2237 } else {
2238 enum object_type type;
2239 unsigned long sz;
2240 char *result;
2242 result = read_sha1_file(ce->sha1, &type, &sz);
2243 if (!result)
2244 return -1;
2245 /* XXX read_sha1_file NUL-terminates */
2246 strbuf_attach(buf, result, sz, sz + 1);
2248 return 0;
2251 static struct patch *in_fn_table(const char *name)
2253 struct string_list_item *item;
2255 if (name == NULL)
2256 return NULL;
2258 item = string_list_lookup(name, &fn_table);
2259 if (item != NULL)
2260 return (struct patch *)item->util;
2262 return NULL;
2265 static void add_to_fn_table(struct patch *patch)
2267 struct string_list_item *item;
2270 * Always add new_name unless patch is a deletion
2271 * This should cover the cases for normal diffs,
2272 * file creations and copies
2274 if (patch->new_name != NULL) {
2275 item = string_list_insert(patch->new_name, &fn_table);
2276 item->util = patch;
2280 * store a failure on rename/deletion cases because
2281 * later chunks shouldn't patch old names
2283 if ((patch->new_name == NULL) || (patch->is_rename)) {
2284 item = string_list_insert(patch->old_name, &fn_table);
2285 item->util = (struct patch *) -1;
2289 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2291 struct strbuf buf;
2292 struct image image;
2293 size_t len;
2294 char *img;
2295 struct patch *tpatch;
2297 strbuf_init(&buf, 0);
2299 if (!(patch->is_copy || patch->is_rename) &&
2300 ((tpatch = in_fn_table(patch->old_name)) != NULL)) {
2301 if (tpatch == (struct patch *) -1) {
2302 return error("patch %s has been renamed/deleted",
2303 patch->old_name);
2305 /* We have a patched copy in memory use that */
2306 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2307 } else if (cached) {
2308 if (read_file_or_gitlink(ce, &buf))
2309 return error("read of %s failed", patch->old_name);
2310 } else if (patch->old_name) {
2311 if (S_ISGITLINK(patch->old_mode)) {
2312 if (ce) {
2313 read_file_or_gitlink(ce, &buf);
2314 } else {
2316 * There is no way to apply subproject
2317 * patch without looking at the index.
2319 patch->fragments = NULL;
2321 } else {
2322 if (read_old_data(st, patch->old_name, &buf))
2323 return error("read of %s failed", patch->old_name);
2327 img = strbuf_detach(&buf, &len);
2328 prepare_image(&image, img, len, !patch->is_binary);
2330 if (apply_fragments(&image, patch) < 0)
2331 return -1; /* note with --reject this succeeds. */
2332 patch->result = image.buf;
2333 patch->resultsize = image.len;
2334 add_to_fn_table(patch);
2335 free(image.line_allocated);
2337 if (0 < patch->is_delete && patch->resultsize)
2338 return error("removal patch leaves file contents");
2340 return 0;
2343 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2345 struct stat nst;
2346 if (!lstat(new_name, &nst)) {
2347 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2348 return 0;
2350 * A leading component of new_name might be a symlink
2351 * that is going to be removed with this patch, but
2352 * still pointing at somewhere that has the path.
2353 * In such a case, path "new_name" does not exist as
2354 * far as git is concerned.
2356 if (has_symlink_leading_path(strlen(new_name), new_name))
2357 return 0;
2359 return error("%s: already exists in working directory", new_name);
2361 else if ((errno != ENOENT) && (errno != ENOTDIR))
2362 return error("%s: %s", new_name, strerror(errno));
2363 return 0;
2366 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2368 if (S_ISGITLINK(ce->ce_mode)) {
2369 if (!S_ISDIR(st->st_mode))
2370 return -1;
2371 return 0;
2373 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2376 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2378 const char *old_name = patch->old_name;
2379 struct patch *tpatch = NULL;
2380 int stat_ret = 0;
2381 unsigned st_mode = 0;
2384 * Make sure that we do not have local modifications from the
2385 * index when we are looking at the index. Also make sure
2386 * we have the preimage file to be patched in the work tree,
2387 * unless --cached, which tells git to apply only in the index.
2389 if (!old_name)
2390 return 0;
2392 assert(patch->is_new <= 0);
2394 if (!(patch->is_copy || patch->is_rename) &&
2395 (tpatch = in_fn_table(old_name)) != NULL) {
2396 if (tpatch == (struct patch *) -1) {
2397 return error("%s: has been deleted/renamed", old_name);
2399 st_mode = tpatch->new_mode;
2400 } else if (!cached) {
2401 stat_ret = lstat(old_name, st);
2402 if (stat_ret && errno != ENOENT)
2403 return error("%s: %s", old_name, strerror(errno));
2406 if (check_index && !tpatch) {
2407 int pos = cache_name_pos(old_name, strlen(old_name));
2408 if (pos < 0) {
2409 if (patch->is_new < 0)
2410 goto is_new;
2411 return error("%s: does not exist in index", old_name);
2413 *ce = active_cache[pos];
2414 if (stat_ret < 0) {
2415 struct checkout costate;
2416 /* checkout */
2417 costate.base_dir = "";
2418 costate.base_dir_len = 0;
2419 costate.force = 0;
2420 costate.quiet = 0;
2421 costate.not_new = 0;
2422 costate.refresh_cache = 1;
2423 if (checkout_entry(*ce, &costate, NULL) ||
2424 lstat(old_name, st))
2425 return -1;
2427 if (!cached && verify_index_match(*ce, st))
2428 return error("%s: does not match index", old_name);
2429 if (cached)
2430 st_mode = (*ce)->ce_mode;
2431 } else if (stat_ret < 0) {
2432 if (patch->is_new < 0)
2433 goto is_new;
2434 return error("%s: %s", old_name, strerror(errno));
2437 if (!cached)
2438 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2440 if (patch->is_new < 0)
2441 patch->is_new = 0;
2442 if (!patch->old_mode)
2443 patch->old_mode = st_mode;
2444 if ((st_mode ^ patch->old_mode) & S_IFMT)
2445 return error("%s: wrong type", old_name);
2446 if (st_mode != patch->old_mode)
2447 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2448 old_name, st_mode, patch->old_mode);
2449 return 0;
2451 is_new:
2452 patch->is_new = 1;
2453 patch->is_delete = 0;
2454 patch->old_name = NULL;
2455 return 0;
2458 static int check_patch(struct patch *patch)
2460 struct stat st;
2461 const char *old_name = patch->old_name;
2462 const char *new_name = patch->new_name;
2463 const char *name = old_name ? old_name : new_name;
2464 struct cache_entry *ce = NULL;
2465 int ok_if_exists;
2466 int status;
2468 patch->rejected = 1; /* we will drop this after we succeed */
2470 status = check_preimage(patch, &ce, &st);
2471 if (status)
2472 return status;
2473 old_name = patch->old_name;
2475 if (in_fn_table(new_name) == (struct patch *) -1)
2477 * A type-change diff is always split into a patch to
2478 * delete old, immediately followed by a patch to
2479 * create new (see diff.c::run_diff()); in such a case
2480 * it is Ok that the entry to be deleted by the
2481 * previous patch is still in the working tree and in
2482 * the index.
2484 ok_if_exists = 1;
2485 else
2486 ok_if_exists = 0;
2488 if (new_name &&
2489 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2490 if (check_index &&
2491 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2492 !ok_if_exists)
2493 return error("%s: already exists in index", new_name);
2494 if (!cached) {
2495 int err = check_to_create_blob(new_name, ok_if_exists);
2496 if (err)
2497 return err;
2499 if (!patch->new_mode) {
2500 if (0 < patch->is_new)
2501 patch->new_mode = S_IFREG | 0644;
2502 else
2503 patch->new_mode = patch->old_mode;
2507 if (new_name && old_name) {
2508 int same = !strcmp(old_name, new_name);
2509 if (!patch->new_mode)
2510 patch->new_mode = patch->old_mode;
2511 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2512 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2513 patch->new_mode, new_name, patch->old_mode,
2514 same ? "" : " of ", same ? "" : old_name);
2517 if (apply_data(patch, &st, ce) < 0)
2518 return error("%s: patch does not apply", name);
2519 patch->rejected = 0;
2520 return 0;
2523 static int check_patch_list(struct patch *patch)
2525 int err = 0;
2527 while (patch) {
2528 if (apply_verbosely)
2529 say_patch_name(stderr,
2530 "Checking patch ", patch, "...\n");
2531 err |= check_patch(patch);
2532 patch = patch->next;
2534 return err;
2537 /* This function tries to read the sha1 from the current index */
2538 static int get_current_sha1(const char *path, unsigned char *sha1)
2540 int pos;
2542 if (read_cache() < 0)
2543 return -1;
2544 pos = cache_name_pos(path, strlen(path));
2545 if (pos < 0)
2546 return -1;
2547 hashcpy(sha1, active_cache[pos]->sha1);
2548 return 0;
2551 /* Build an index that contains the just the files needed for a 3way merge */
2552 static void build_fake_ancestor(struct patch *list, const char *filename)
2554 struct patch *patch;
2555 struct index_state result = { 0 };
2556 int fd;
2558 /* Once we start supporting the reverse patch, it may be
2559 * worth showing the new sha1 prefix, but until then...
2561 for (patch = list; patch; patch = patch->next) {
2562 const unsigned char *sha1_ptr;
2563 unsigned char sha1[20];
2564 struct cache_entry *ce;
2565 const char *name;
2567 name = patch->old_name ? patch->old_name : patch->new_name;
2568 if (0 < patch->is_new)
2569 continue;
2570 else if (get_sha1(patch->old_sha1_prefix, sha1))
2571 /* git diff has no index line for mode/type changes */
2572 if (!patch->lines_added && !patch->lines_deleted) {
2573 if (get_current_sha1(patch->new_name, sha1) ||
2574 get_current_sha1(patch->old_name, sha1))
2575 die("mode change for %s, which is not "
2576 "in current HEAD", name);
2577 sha1_ptr = sha1;
2578 } else
2579 die("sha1 information is lacking or useless "
2580 "(%s).", name);
2581 else
2582 sha1_ptr = sha1;
2584 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2585 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2586 die ("Could not add %s to temporary index", name);
2589 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2590 if (fd < 0 || write_index(&result, fd) || close(fd))
2591 die ("Could not write temporary index to %s", filename);
2593 discard_index(&result);
2596 static void stat_patch_list(struct patch *patch)
2598 int files, adds, dels;
2600 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2601 files++;
2602 adds += patch->lines_added;
2603 dels += patch->lines_deleted;
2604 show_stats(patch);
2607 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2610 static void numstat_patch_list(struct patch *patch)
2612 for ( ; patch; patch = patch->next) {
2613 const char *name;
2614 name = patch->new_name ? patch->new_name : patch->old_name;
2615 if (patch->is_binary)
2616 printf("-\t-\t");
2617 else
2618 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2619 write_name_quoted(name, stdout, line_termination);
2623 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2625 if (mode)
2626 printf(" %s mode %06o %s\n", newdelete, mode, name);
2627 else
2628 printf(" %s %s\n", newdelete, name);
2631 static void show_mode_change(struct patch *p, int show_name)
2633 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2634 if (show_name)
2635 printf(" mode change %06o => %06o %s\n",
2636 p->old_mode, p->new_mode, p->new_name);
2637 else
2638 printf(" mode change %06o => %06o\n",
2639 p->old_mode, p->new_mode);
2643 static void show_rename_copy(struct patch *p)
2645 const char *renamecopy = p->is_rename ? "rename" : "copy";
2646 const char *old, *new;
2648 /* Find common prefix */
2649 old = p->old_name;
2650 new = p->new_name;
2651 while (1) {
2652 const char *slash_old, *slash_new;
2653 slash_old = strchr(old, '/');
2654 slash_new = strchr(new, '/');
2655 if (!slash_old ||
2656 !slash_new ||
2657 slash_old - old != slash_new - new ||
2658 memcmp(old, new, slash_new - new))
2659 break;
2660 old = slash_old + 1;
2661 new = slash_new + 1;
2663 /* p->old_name thru old is the common prefix, and old and new
2664 * through the end of names are renames
2666 if (old != p->old_name)
2667 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2668 (int)(old - p->old_name), p->old_name,
2669 old, new, p->score);
2670 else
2671 printf(" %s %s => %s (%d%%)\n", renamecopy,
2672 p->old_name, p->new_name, p->score);
2673 show_mode_change(p, 0);
2676 static void summary_patch_list(struct patch *patch)
2678 struct patch *p;
2680 for (p = patch; p; p = p->next) {
2681 if (p->is_new)
2682 show_file_mode_name("create", p->new_mode, p->new_name);
2683 else if (p->is_delete)
2684 show_file_mode_name("delete", p->old_mode, p->old_name);
2685 else {
2686 if (p->is_rename || p->is_copy)
2687 show_rename_copy(p);
2688 else {
2689 if (p->score) {
2690 printf(" rewrite %s (%d%%)\n",
2691 p->new_name, p->score);
2692 show_mode_change(p, 0);
2694 else
2695 show_mode_change(p, 1);
2701 static void patch_stats(struct patch *patch)
2703 int lines = patch->lines_added + patch->lines_deleted;
2705 if (lines > max_change)
2706 max_change = lines;
2707 if (patch->old_name) {
2708 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2709 if (!len)
2710 len = strlen(patch->old_name);
2711 if (len > max_len)
2712 max_len = len;
2714 if (patch->new_name) {
2715 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2716 if (!len)
2717 len = strlen(patch->new_name);
2718 if (len > max_len)
2719 max_len = len;
2723 static void remove_file(struct patch *patch, int rmdir_empty)
2725 if (update_index) {
2726 if (remove_file_from_cache(patch->old_name) < 0)
2727 die("unable to remove %s from index", patch->old_name);
2729 if (!cached) {
2730 if (S_ISGITLINK(patch->old_mode)) {
2731 if (rmdir(patch->old_name))
2732 warning("unable to remove submodule %s",
2733 patch->old_name);
2734 } else if (!unlink(patch->old_name) && rmdir_empty) {
2735 char *name = xstrdup(patch->old_name);
2736 char *end = strrchr(name, '/');
2737 while (end) {
2738 *end = 0;
2739 if (rmdir(name))
2740 break;
2741 end = strrchr(name, '/');
2743 free(name);
2748 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2750 struct stat st;
2751 struct cache_entry *ce;
2752 int namelen = strlen(path);
2753 unsigned ce_size = cache_entry_size(namelen);
2755 if (!update_index)
2756 return;
2758 ce = xcalloc(1, ce_size);
2759 memcpy(ce->name, path, namelen);
2760 ce->ce_mode = create_ce_mode(mode);
2761 ce->ce_flags = namelen;
2762 if (S_ISGITLINK(mode)) {
2763 const char *s = buf;
2765 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2766 die("corrupt patch for subproject %s", path);
2767 } else {
2768 if (!cached) {
2769 if (lstat(path, &st) < 0)
2770 die("unable to stat newly created file %s",
2771 path);
2772 fill_stat_cache_info(ce, &st);
2774 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2775 die("unable to create backing store for newly created file %s", path);
2777 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2778 die("unable to add cache entry for %s", path);
2781 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2783 int fd;
2784 struct strbuf nbuf;
2786 if (S_ISGITLINK(mode)) {
2787 struct stat st;
2788 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2789 return 0;
2790 return mkdir(path, 0777);
2793 if (has_symlinks && S_ISLNK(mode))
2794 /* Although buf:size is counted string, it also is NUL
2795 * terminated.
2797 return symlink(buf, path);
2799 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2800 if (fd < 0)
2801 return -1;
2803 strbuf_init(&nbuf, 0);
2804 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2805 size = nbuf.len;
2806 buf = nbuf.buf;
2808 write_or_die(fd, buf, size);
2809 strbuf_release(&nbuf);
2811 if (close(fd) < 0)
2812 die("closing file %s: %s", path, strerror(errno));
2813 return 0;
2817 * We optimistically assume that the directories exist,
2818 * which is true 99% of the time anyway. If they don't,
2819 * we create them and try again.
2821 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2823 if (cached)
2824 return;
2825 if (!try_create_file(path, mode, buf, size))
2826 return;
2828 if (errno == ENOENT) {
2829 if (safe_create_leading_directories(path))
2830 return;
2831 if (!try_create_file(path, mode, buf, size))
2832 return;
2835 if (errno == EEXIST || errno == EACCES) {
2836 /* We may be trying to create a file where a directory
2837 * used to be.
2839 struct stat st;
2840 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2841 errno = EEXIST;
2844 if (errno == EEXIST) {
2845 unsigned int nr = getpid();
2847 for (;;) {
2848 const char *newpath;
2849 newpath = mkpath("%s~%u", path, nr);
2850 if (!try_create_file(newpath, mode, buf, size)) {
2851 if (!rename(newpath, path))
2852 return;
2853 unlink(newpath);
2854 break;
2856 if (errno != EEXIST)
2857 break;
2858 ++nr;
2861 die("unable to write file %s mode %o", path, mode);
2864 static void create_file(struct patch *patch)
2866 char *path = patch->new_name;
2867 unsigned mode = patch->new_mode;
2868 unsigned long size = patch->resultsize;
2869 char *buf = patch->result;
2871 if (!mode)
2872 mode = S_IFREG | 0644;
2873 create_one_file(path, mode, buf, size);
2874 add_index_file(path, mode, buf, size);
2877 /* phase zero is to remove, phase one is to create */
2878 static void write_out_one_result(struct patch *patch, int phase)
2880 if (patch->is_delete > 0) {
2881 if (phase == 0)
2882 remove_file(patch, 1);
2883 return;
2885 if (patch->is_new > 0 || patch->is_copy) {
2886 if (phase == 1)
2887 create_file(patch);
2888 return;
2891 * Rename or modification boils down to the same
2892 * thing: remove the old, write the new
2894 if (phase == 0)
2895 remove_file(patch, patch->is_rename);
2896 if (phase == 1)
2897 create_file(patch);
2900 static int write_out_one_reject(struct patch *patch)
2902 FILE *rej;
2903 char namebuf[PATH_MAX];
2904 struct fragment *frag;
2905 int cnt = 0;
2907 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2908 if (!frag->rejected)
2909 continue;
2910 cnt++;
2913 if (!cnt) {
2914 if (apply_verbosely)
2915 say_patch_name(stderr,
2916 "Applied patch ", patch, " cleanly.\n");
2917 return 0;
2920 /* This should not happen, because a removal patch that leaves
2921 * contents are marked "rejected" at the patch level.
2923 if (!patch->new_name)
2924 die("internal error");
2926 /* Say this even without --verbose */
2927 say_patch_name(stderr, "Applying patch ", patch, " with");
2928 fprintf(stderr, " %d rejects...\n", cnt);
2930 cnt = strlen(patch->new_name);
2931 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2932 cnt = ARRAY_SIZE(namebuf) - 5;
2933 fprintf(stderr,
2934 "warning: truncating .rej filename to %.*s.rej",
2935 cnt - 1, patch->new_name);
2937 memcpy(namebuf, patch->new_name, cnt);
2938 memcpy(namebuf + cnt, ".rej", 5);
2940 rej = fopen(namebuf, "w");
2941 if (!rej)
2942 return error("cannot open %s: %s", namebuf, strerror(errno));
2944 /* Normal git tools never deal with .rej, so do not pretend
2945 * this is a git patch by saying --git nor give extended
2946 * headers. While at it, maybe please "kompare" that wants
2947 * the trailing TAB and some garbage at the end of line ;-).
2949 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2950 patch->new_name, patch->new_name);
2951 for (cnt = 1, frag = patch->fragments;
2952 frag;
2953 cnt++, frag = frag->next) {
2954 if (!frag->rejected) {
2955 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2956 continue;
2958 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2959 fprintf(rej, "%.*s", frag->size, frag->patch);
2960 if (frag->patch[frag->size-1] != '\n')
2961 fputc('\n', rej);
2963 fclose(rej);
2964 return -1;
2967 static int write_out_results(struct patch *list, int skipped_patch)
2969 int phase;
2970 int errs = 0;
2971 struct patch *l;
2973 if (!list && !skipped_patch)
2974 return error("No changes");
2976 for (phase = 0; phase < 2; phase++) {
2977 l = list;
2978 while (l) {
2979 if (l->rejected)
2980 errs = 1;
2981 else {
2982 write_out_one_result(l, phase);
2983 if (phase == 1 && write_out_one_reject(l))
2984 errs = 1;
2986 l = l->next;
2989 return errs;
2992 static struct lock_file lock_file;
2994 static struct excludes {
2995 struct excludes *next;
2996 const char *path;
2997 } *excludes;
2999 static int use_patch(struct patch *p)
3001 const char *pathname = p->new_name ? p->new_name : p->old_name;
3002 struct excludes *x = excludes;
3003 while (x) {
3004 if (fnmatch(x->path, pathname, 0) == 0)
3005 return 0;
3006 x = x->next;
3008 if (0 < prefix_length) {
3009 int pathlen = strlen(pathname);
3010 if (pathlen <= prefix_length ||
3011 memcmp(prefix, pathname, prefix_length))
3012 return 0;
3014 return 1;
3017 static void prefix_one(char **name)
3019 char *old_name = *name;
3020 if (!old_name)
3021 return;
3022 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3023 free(old_name);
3026 static void prefix_patches(struct patch *p)
3028 if (!prefix || p->is_toplevel_relative)
3029 return;
3030 for ( ; p; p = p->next) {
3031 if (p->new_name == p->old_name) {
3032 char *prefixed = p->new_name;
3033 prefix_one(&prefixed);
3034 p->new_name = p->old_name = prefixed;
3036 else {
3037 prefix_one(&p->new_name);
3038 prefix_one(&p->old_name);
3043 #define INACCURATE_EOF (1<<0)
3044 #define RECOUNT (1<<1)
3046 static int apply_patch(int fd, const char *filename, int options)
3048 size_t offset;
3049 struct strbuf buf;
3050 struct patch *list = NULL, **listp = &list;
3051 int skipped_patch = 0;
3053 /* FIXME - memory leak when using multiple patch files as inputs */
3054 memset(&fn_table, 0, sizeof(struct string_list));
3055 strbuf_init(&buf, 0);
3056 patch_input_file = filename;
3057 read_patch_file(&buf, fd);
3058 offset = 0;
3059 while (offset < buf.len) {
3060 struct patch *patch;
3061 int nr;
3063 patch = xcalloc(1, sizeof(*patch));
3064 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3065 patch->recount = !!(options & RECOUNT);
3066 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3067 if (nr < 0)
3068 break;
3069 if (apply_in_reverse)
3070 reverse_patches(patch);
3071 if (prefix)
3072 prefix_patches(patch);
3073 if (use_patch(patch)) {
3074 patch_stats(patch);
3075 *listp = patch;
3076 listp = &patch->next;
3078 else {
3079 /* perhaps free it a bit better? */
3080 free(patch);
3081 skipped_patch++;
3083 offset += nr;
3086 if (whitespace_error && (ws_error_action == die_on_ws_error))
3087 apply = 0;
3089 update_index = check_index && apply;
3090 if (update_index && newfd < 0)
3091 newfd = hold_locked_index(&lock_file, 1);
3093 if (check_index) {
3094 if (read_cache() < 0)
3095 die("unable to read index file");
3098 if ((check || apply) &&
3099 check_patch_list(list) < 0 &&
3100 !apply_with_reject)
3101 exit(1);
3103 if (apply && write_out_results(list, skipped_patch))
3104 exit(1);
3106 if (fake_ancestor)
3107 build_fake_ancestor(list, fake_ancestor);
3109 if (diffstat)
3110 stat_patch_list(list);
3112 if (numstat)
3113 numstat_patch_list(list);
3115 if (summary)
3116 summary_patch_list(list);
3118 strbuf_release(&buf);
3119 return 0;
3122 static int git_apply_config(const char *var, const char *value, void *cb)
3124 if (!strcmp(var, "apply.whitespace"))
3125 return git_config_string(&apply_default_whitespace, var, value);
3126 return git_default_config(var, value, cb);
3130 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3132 int i;
3133 int read_stdin = 1;
3134 int options = 0;
3135 int errs = 0;
3136 int is_not_gitdir;
3138 const char *whitespace_option = NULL;
3140 prefix = setup_git_directory_gently(&is_not_gitdir);
3141 prefix_length = prefix ? strlen(prefix) : 0;
3142 git_config(git_apply_config, NULL);
3143 if (apply_default_whitespace)
3144 parse_whitespace_option(apply_default_whitespace);
3146 for (i = 1; i < argc; i++) {
3147 const char *arg = argv[i];
3148 char *end;
3149 int fd;
3151 if (!strcmp(arg, "-")) {
3152 errs |= apply_patch(0, "<stdin>", options);
3153 read_stdin = 0;
3154 continue;
3156 if (!prefixcmp(arg, "--exclude=")) {
3157 struct excludes *x = xmalloc(sizeof(*x));
3158 x->path = arg + 10;
3159 x->next = excludes;
3160 excludes = x;
3161 continue;
3163 if (!prefixcmp(arg, "-p")) {
3164 p_value = atoi(arg + 2);
3165 p_value_known = 1;
3166 continue;
3168 if (!strcmp(arg, "--no-add")) {
3169 no_add = 1;
3170 continue;
3172 if (!strcmp(arg, "--stat")) {
3173 apply = 0;
3174 diffstat = 1;
3175 continue;
3177 if (!strcmp(arg, "--allow-binary-replacement") ||
3178 !strcmp(arg, "--binary")) {
3179 continue; /* now no-op */
3181 if (!strcmp(arg, "--numstat")) {
3182 apply = 0;
3183 numstat = 1;
3184 continue;
3186 if (!strcmp(arg, "--summary")) {
3187 apply = 0;
3188 summary = 1;
3189 continue;
3191 if (!strcmp(arg, "--check")) {
3192 apply = 0;
3193 check = 1;
3194 continue;
3196 if (!strcmp(arg, "--index")) {
3197 if (is_not_gitdir)
3198 die("--index outside a repository");
3199 check_index = 1;
3200 continue;
3202 if (!strcmp(arg, "--cached")) {
3203 if (is_not_gitdir)
3204 die("--cached outside a repository");
3205 check_index = 1;
3206 cached = 1;
3207 continue;
3209 if (!strcmp(arg, "--apply")) {
3210 apply = 1;
3211 continue;
3213 if (!strcmp(arg, "--build-fake-ancestor")) {
3214 apply = 0;
3215 if (++i >= argc)
3216 die ("need a filename");
3217 fake_ancestor = argv[i];
3218 continue;
3220 if (!strcmp(arg, "-z")) {
3221 line_termination = 0;
3222 continue;
3224 if (!prefixcmp(arg, "-C")) {
3225 p_context = strtoul(arg + 2, &end, 0);
3226 if (*end != '\0')
3227 die("unrecognized context count '%s'", arg + 2);
3228 continue;
3230 if (!prefixcmp(arg, "--whitespace=")) {
3231 whitespace_option = arg + 13;
3232 parse_whitespace_option(arg + 13);
3233 continue;
3235 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3236 apply_in_reverse = 1;
3237 continue;
3239 if (!strcmp(arg, "--unidiff-zero")) {
3240 unidiff_zero = 1;
3241 continue;
3243 if (!strcmp(arg, "--reject")) {
3244 apply = apply_with_reject = apply_verbosely = 1;
3245 continue;
3247 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3248 apply_verbosely = 1;
3249 continue;
3251 if (!strcmp(arg, "--inaccurate-eof")) {
3252 options |= INACCURATE_EOF;
3253 continue;
3255 if (!strcmp(arg, "--recount")) {
3256 options |= RECOUNT;
3257 continue;
3259 if (!prefixcmp(arg, "--directory=")) {
3260 arg += strlen("--directory=");
3261 root_len = strlen(arg);
3262 if (root_len && arg[root_len - 1] != '/') {
3263 char *new_root;
3264 root = new_root = xmalloc(root_len + 2);
3265 strcpy(new_root, arg);
3266 strcpy(new_root + root_len++, "/");
3267 } else
3268 root = arg;
3269 continue;
3271 if (0 < prefix_length)
3272 arg = prefix_filename(prefix, prefix_length, arg);
3274 fd = open(arg, O_RDONLY);
3275 if (fd < 0)
3276 die("can't open patch '%s': %s", arg, strerror(errno));
3277 read_stdin = 0;
3278 set_default_whitespace_mode(whitespace_option);
3279 errs |= apply_patch(fd, arg, options);
3280 close(fd);
3282 set_default_whitespace_mode(whitespace_option);
3283 if (read_stdin)
3284 errs |= apply_patch(0, "<stdin>", options);
3285 if (whitespace_error) {
3286 if (squelch_whitespace_errors &&
3287 squelch_whitespace_errors < whitespace_error) {
3288 int squelched =
3289 whitespace_error - squelch_whitespace_errors;
3290 fprintf(stderr, "warning: squelched %d "
3291 "whitespace error%s\n",
3292 squelched,
3293 squelched == 1 ? "" : "s");
3295 if (ws_error_action == die_on_ws_error)
3296 die("%d line%s add%s whitespace errors.",
3297 whitespace_error,
3298 whitespace_error == 1 ? "" : "s",
3299 whitespace_error == 1 ? "s" : "");
3300 if (applied_after_fixing_ws && apply)
3301 fprintf(stderr, "warning: %d line%s applied after"
3302 " fixing whitespace errors.\n",
3303 applied_after_fixing_ws,
3304 applied_after_fixing_ws == 1 ? "" : "s");
3305 else if (whitespace_error)
3306 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3307 whitespace_error,
3308 whitespace_error == 1 ? "" : "s",
3309 whitespace_error == 1 ? "s" : "");
3312 if (update_index) {
3313 if (write_cache(newfd, active_cache, active_nr) ||
3314 commit_locked_index(&lock_file))
3315 die("Unable to write new index file");
3318 return !!errs;