Avoid some unnecessary lstat() calls
[git/git-p4.git] / builtin-apply.c
blobcaa3f2aa0ca147091d28013c3d2c11346e350675
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static const char *fake_ancestor;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|fix|error|error-all>] <patch>...";
50 static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55 } ws_error_action = warn_on_ws_error;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
63 if (!option) {
64 ws_error_action = warn_on_ws_error;
65 return;
67 if (!strcmp(option, "warn")) {
68 ws_error_action = warn_on_ws_error;
69 return;
71 if (!strcmp(option, "nowarn")) {
72 ws_error_action = nowarn_ws_error;
73 return;
75 if (!strcmp(option, "error")) {
76 ws_error_action = die_on_ws_error;
77 return;
79 if (!strcmp(option, "error-all")) {
80 ws_error_action = die_on_ws_error;
81 squelch_whitespace_errors = 0;
82 return;
84 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
85 ws_error_action = correct_ws_error;
86 return;
88 die("unrecognized whitespace option '%s'", option);
91 static void set_default_whitespace_mode(const char *whitespace_option)
93 if (!whitespace_option && !apply_default_whitespace)
94 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
98 * For "diff-stat" like behaviour, we keep track of the biggest change
99 * we've seen, and the longest filename. That allows us to do simple
100 * scaling.
102 static int max_change, max_len;
105 * Various "current state", notably line numbers and what
106 * file (and how) we're patching right now.. The "is_xxxx"
107 * things are flags, where -1 means "don't know yet".
109 static int linenr = 1;
112 * This represents one "hunk" from a patch, starting with
113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
114 * patch text is pointed at by patch, and its byte length
115 * is stored in size. leading and trailing are the number
116 * of context lines.
118 struct fragment {
119 unsigned long leading, trailing;
120 unsigned long oldpos, oldlines;
121 unsigned long newpos, newlines;
122 const char *patch;
123 int size;
124 int rejected;
125 struct fragment *next;
129 * When dealing with a binary patch, we reuse "leading" field
130 * to store the type of the binary hunk, either deflated "delta"
131 * or deflated "literal".
133 #define binary_patch_method leading
134 #define BINARY_DELTA_DEFLATED 1
135 #define BINARY_LITERAL_DEFLATED 2
138 * This represents a "patch" to a file, both metainfo changes
139 * such as creation/deletion, filemode and content changes represented
140 * as a series of fragments.
142 struct patch {
143 char *new_name, *old_name, *def_name;
144 unsigned int old_mode, new_mode;
145 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
146 int rejected;
147 unsigned ws_rule;
148 unsigned long deflate_origlen;
149 int lines_added, lines_deleted;
150 int score;
151 unsigned int is_toplevel_relative:1;
152 unsigned int inaccurate_eof:1;
153 unsigned int is_binary:1;
154 unsigned int is_copy:1;
155 unsigned int is_rename:1;
156 struct fragment *fragments;
157 char *result;
158 size_t resultsize;
159 char old_sha1_prefix[41];
160 char new_sha1_prefix[41];
161 struct patch *next;
165 * A line in a file, len-bytes long (includes the terminating LF,
166 * except for an incomplete line at the end if the file ends with
167 * one), and its contents hashes to 'hash'.
169 struct line {
170 size_t len;
171 unsigned hash : 24;
172 unsigned flag : 8;
173 #define LINE_COMMON 1
177 * This represents a "file", which is an array of "lines".
179 struct image {
180 char *buf;
181 size_t len;
182 size_t nr;
183 size_t alloc;
184 struct line *line_allocated;
185 struct line *line;
188 static uint32_t hash_line(const char *cp, size_t len)
190 size_t i;
191 uint32_t h;
192 for (i = 0, h = 0; i < len; i++) {
193 if (!isspace(cp[i])) {
194 h = h * 3 + (cp[i] & 0xff);
197 return h;
200 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
202 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
203 img->line_allocated[img->nr].len = len;
204 img->line_allocated[img->nr].hash = hash_line(bol, len);
205 img->line_allocated[img->nr].flag = flag;
206 img->nr++;
209 static void prepare_image(struct image *image, char *buf, size_t len,
210 int prepare_linetable)
212 const char *cp, *ep;
214 memset(image, 0, sizeof(*image));
215 image->buf = buf;
216 image->len = len;
218 if (!prepare_linetable)
219 return;
221 ep = image->buf + image->len;
222 cp = image->buf;
223 while (cp < ep) {
224 const char *next;
225 for (next = cp; next < ep && *next != '\n'; next++)
227 if (next < ep)
228 next++;
229 add_line_info(image, cp, next - cp, 0);
230 cp = next;
232 image->line = image->line_allocated;
235 static void clear_image(struct image *image)
237 free(image->buf);
238 image->buf = NULL;
239 image->len = 0;
242 static void say_patch_name(FILE *output, const char *pre,
243 struct patch *patch, const char *post)
245 fputs(pre, output);
246 if (patch->old_name && patch->new_name &&
247 strcmp(patch->old_name, patch->new_name)) {
248 quote_c_style(patch->old_name, NULL, output, 0);
249 fputs(" => ", output);
250 quote_c_style(patch->new_name, NULL, output, 0);
251 } else {
252 const char *n = patch->new_name;
253 if (!n)
254 n = patch->old_name;
255 quote_c_style(n, NULL, output, 0);
257 fputs(post, output);
260 #define CHUNKSIZE (8192)
261 #define SLOP (16)
263 static void read_patch_file(struct strbuf *sb, int fd)
265 if (strbuf_read(sb, fd, 0) < 0)
266 die("git-apply: read returned %s", strerror(errno));
269 * Make sure that we have some slop in the buffer
270 * so that we can do speculative "memcmp" etc, and
271 * see to it that it is NUL-filled.
273 strbuf_grow(sb, SLOP);
274 memset(sb->buf + sb->len, 0, SLOP);
277 static unsigned long linelen(const char *buffer, unsigned long size)
279 unsigned long len = 0;
280 while (size--) {
281 len++;
282 if (*buffer++ == '\n')
283 break;
285 return len;
288 static int is_dev_null(const char *str)
290 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
293 #define TERM_SPACE 1
294 #define TERM_TAB 2
296 static int name_terminate(const char *name, int namelen, int c, int terminate)
298 if (c == ' ' && !(terminate & TERM_SPACE))
299 return 0;
300 if (c == '\t' && !(terminate & TERM_TAB))
301 return 0;
303 return 1;
306 static char *find_name(const char *line, char *def, int p_value, int terminate)
308 int len;
309 const char *start = line;
311 if (*line == '"') {
312 struct strbuf name;
315 * Proposed "new-style" GNU patch/diff format; see
316 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
318 strbuf_init(&name, 0);
319 if (!unquote_c_style(&name, line, NULL)) {
320 char *cp;
322 for (cp = name.buf; p_value; p_value--) {
323 cp = strchr(cp, '/');
324 if (!cp)
325 break;
326 cp++;
328 if (cp) {
329 /* name can later be freed, so we need
330 * to memmove, not just return cp
332 strbuf_remove(&name, 0, cp - name.buf);
333 free(def);
334 return strbuf_detach(&name, NULL);
337 strbuf_release(&name);
340 for (;;) {
341 char c = *line;
343 if (isspace(c)) {
344 if (c == '\n')
345 break;
346 if (name_terminate(start, line-start, c, terminate))
347 break;
349 line++;
350 if (c == '/' && !--p_value)
351 start = line;
353 if (!start)
354 return def;
355 len = line - start;
356 if (!len)
357 return def;
360 * Generally we prefer the shorter name, especially
361 * if the other one is just a variation of that with
362 * something else tacked on to the end (ie "file.orig"
363 * or "file~").
365 if (def) {
366 int deflen = strlen(def);
367 if (deflen < len && !strncmp(start, def, deflen))
368 return def;
369 free(def);
372 return xmemdupz(start, len);
375 static int count_slashes(const char *cp)
377 int cnt = 0;
378 char ch;
380 while ((ch = *cp++))
381 if (ch == '/')
382 cnt++;
383 return cnt;
387 * Given the string after "--- " or "+++ ", guess the appropriate
388 * p_value for the given patch.
390 static int guess_p_value(const char *nameline)
392 char *name, *cp;
393 int val = -1;
395 if (is_dev_null(nameline))
396 return -1;
397 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
398 if (!name)
399 return -1;
400 cp = strchr(name, '/');
401 if (!cp)
402 val = 0;
403 else if (prefix) {
405 * Does it begin with "a/$our-prefix" and such? Then this is
406 * very likely to apply to our directory.
408 if (!strncmp(name, prefix, prefix_length))
409 val = count_slashes(prefix);
410 else {
411 cp++;
412 if (!strncmp(cp, prefix, prefix_length))
413 val = count_slashes(prefix) + 1;
416 free(name);
417 return val;
421 * Get the name etc info from the --/+++ lines of a traditional patch header
423 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
424 * files, we can happily check the index for a match, but for creating a
425 * new file we should try to match whatever "patch" does. I have no idea.
427 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
429 char *name;
431 first += 4; /* skip "--- " */
432 second += 4; /* skip "+++ " */
433 if (!p_value_known) {
434 int p, q;
435 p = guess_p_value(first);
436 q = guess_p_value(second);
437 if (p < 0) p = q;
438 if (0 <= p && p == q) {
439 p_value = p;
440 p_value_known = 1;
443 if (is_dev_null(first)) {
444 patch->is_new = 1;
445 patch->is_delete = 0;
446 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
447 patch->new_name = name;
448 } else if (is_dev_null(second)) {
449 patch->is_new = 0;
450 patch->is_delete = 1;
451 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
452 patch->old_name = name;
453 } else {
454 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
455 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
456 patch->old_name = patch->new_name = name;
458 if (!name)
459 die("unable to find filename in patch at line %d", linenr);
462 static int gitdiff_hdrend(const char *line, struct patch *patch)
464 return -1;
468 * We're anal about diff header consistency, to make
469 * sure that we don't end up having strange ambiguous
470 * patches floating around.
472 * As a result, gitdiff_{old|new}name() will check
473 * their names against any previous information, just
474 * to make sure..
476 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
478 if (!orig_name && !isnull)
479 return find_name(line, NULL, p_value, TERM_TAB);
481 if (orig_name) {
482 int len;
483 const char *name;
484 char *another;
485 name = orig_name;
486 len = strlen(name);
487 if (isnull)
488 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
489 another = find_name(line, NULL, p_value, TERM_TAB);
490 if (!another || memcmp(another, name, len))
491 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
492 free(another);
493 return orig_name;
495 else {
496 /* expect "/dev/null" */
497 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
498 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
499 return NULL;
503 static int gitdiff_oldname(const char *line, struct patch *patch)
505 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
506 return 0;
509 static int gitdiff_newname(const char *line, struct patch *patch)
511 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
512 return 0;
515 static int gitdiff_oldmode(const char *line, struct patch *patch)
517 patch->old_mode = strtoul(line, NULL, 8);
518 return 0;
521 static int gitdiff_newmode(const char *line, struct patch *patch)
523 patch->new_mode = strtoul(line, NULL, 8);
524 return 0;
527 static int gitdiff_delete(const char *line, struct patch *patch)
529 patch->is_delete = 1;
530 patch->old_name = patch->def_name;
531 return gitdiff_oldmode(line, patch);
534 static int gitdiff_newfile(const char *line, struct patch *patch)
536 patch->is_new = 1;
537 patch->new_name = patch->def_name;
538 return gitdiff_newmode(line, patch);
541 static int gitdiff_copysrc(const char *line, struct patch *patch)
543 patch->is_copy = 1;
544 patch->old_name = find_name(line, NULL, 0, 0);
545 return 0;
548 static int gitdiff_copydst(const char *line, struct patch *patch)
550 patch->is_copy = 1;
551 patch->new_name = find_name(line, NULL, 0, 0);
552 return 0;
555 static int gitdiff_renamesrc(const char *line, struct patch *patch)
557 patch->is_rename = 1;
558 patch->old_name = find_name(line, NULL, 0, 0);
559 return 0;
562 static int gitdiff_renamedst(const char *line, struct patch *patch)
564 patch->is_rename = 1;
565 patch->new_name = find_name(line, NULL, 0, 0);
566 return 0;
569 static int gitdiff_similarity(const char *line, struct patch *patch)
571 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
572 patch->score = 0;
573 return 0;
576 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
578 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
579 patch->score = 0;
580 return 0;
583 static int gitdiff_index(const char *line, struct patch *patch)
586 * index line is N hexadecimal, "..", N hexadecimal,
587 * and optional space with octal mode.
589 const char *ptr, *eol;
590 int len;
592 ptr = strchr(line, '.');
593 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
594 return 0;
595 len = ptr - line;
596 memcpy(patch->old_sha1_prefix, line, len);
597 patch->old_sha1_prefix[len] = 0;
599 line = ptr + 2;
600 ptr = strchr(line, ' ');
601 eol = strchr(line, '\n');
603 if (!ptr || eol < ptr)
604 ptr = eol;
605 len = ptr - line;
607 if (40 < len)
608 return 0;
609 memcpy(patch->new_sha1_prefix, line, len);
610 patch->new_sha1_prefix[len] = 0;
611 if (*ptr == ' ')
612 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
613 return 0;
617 * This is normal for a diff that doesn't change anything: we'll fall through
618 * into the next diff. Tell the parser to break out.
620 static int gitdiff_unrecognized(const char *line, struct patch *patch)
622 return -1;
625 static const char *stop_at_slash(const char *line, int llen)
627 int i;
629 for (i = 0; i < llen; i++) {
630 int ch = line[i];
631 if (ch == '/')
632 return line + i;
634 return NULL;
638 * This is to extract the same name that appears on "diff --git"
639 * line. We do not find and return anything if it is a rename
640 * patch, and it is OK because we will find the name elsewhere.
641 * We need to reliably find name only when it is mode-change only,
642 * creation or deletion of an empty file. In any of these cases,
643 * both sides are the same name under a/ and b/ respectively.
645 static char *git_header_name(char *line, int llen)
647 const char *name;
648 const char *second = NULL;
649 size_t len;
651 line += strlen("diff --git ");
652 llen -= strlen("diff --git ");
654 if (*line == '"') {
655 const char *cp;
656 struct strbuf first;
657 struct strbuf sp;
659 strbuf_init(&first, 0);
660 strbuf_init(&sp, 0);
662 if (unquote_c_style(&first, line, &second))
663 goto free_and_fail1;
665 /* advance to the first slash */
666 cp = stop_at_slash(first.buf, first.len);
667 /* we do not accept absolute paths */
668 if (!cp || cp == first.buf)
669 goto free_and_fail1;
670 strbuf_remove(&first, 0, cp + 1 - first.buf);
673 * second points at one past closing dq of name.
674 * find the second name.
676 while ((second < line + llen) && isspace(*second))
677 second++;
679 if (line + llen <= second)
680 goto free_and_fail1;
681 if (*second == '"') {
682 if (unquote_c_style(&sp, second, NULL))
683 goto free_and_fail1;
684 cp = stop_at_slash(sp.buf, sp.len);
685 if (!cp || cp == sp.buf)
686 goto free_and_fail1;
687 /* They must match, otherwise ignore */
688 if (strcmp(cp + 1, first.buf))
689 goto free_and_fail1;
690 strbuf_release(&sp);
691 return strbuf_detach(&first, NULL);
694 /* unquoted second */
695 cp = stop_at_slash(second, line + llen - second);
696 if (!cp || cp == second)
697 goto free_and_fail1;
698 cp++;
699 if (line + llen - cp != first.len + 1 ||
700 memcmp(first.buf, cp, first.len))
701 goto free_and_fail1;
702 return strbuf_detach(&first, NULL);
704 free_and_fail1:
705 strbuf_release(&first);
706 strbuf_release(&sp);
707 return NULL;
710 /* unquoted first name */
711 name = stop_at_slash(line, llen);
712 if (!name || name == line)
713 return NULL;
714 name++;
717 * since the first name is unquoted, a dq if exists must be
718 * the beginning of the second name.
720 for (second = name; second < line + llen; second++) {
721 if (*second == '"') {
722 struct strbuf sp;
723 const char *np;
725 strbuf_init(&sp, 0);
726 if (unquote_c_style(&sp, second, NULL))
727 goto free_and_fail2;
729 np = stop_at_slash(sp.buf, sp.len);
730 if (!np || np == sp.buf)
731 goto free_and_fail2;
732 np++;
734 len = sp.buf + sp.len - np;
735 if (len < second - name &&
736 !strncmp(np, name, len) &&
737 isspace(name[len])) {
738 /* Good */
739 strbuf_remove(&sp, 0, np - sp.buf);
740 return strbuf_detach(&sp, NULL);
743 free_and_fail2:
744 strbuf_release(&sp);
745 return NULL;
750 * Accept a name only if it shows up twice, exactly the same
751 * form.
753 for (len = 0 ; ; len++) {
754 switch (name[len]) {
755 default:
756 continue;
757 case '\n':
758 return NULL;
759 case '\t': case ' ':
760 second = name+len;
761 for (;;) {
762 char c = *second++;
763 if (c == '\n')
764 return NULL;
765 if (c == '/')
766 break;
768 if (second[len] == '\n' && !memcmp(name, second, len)) {
769 return xmemdupz(name, len);
775 /* Verify that we recognize the lines following a git header */
776 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
778 unsigned long offset;
780 /* A git diff has explicit new/delete information, so we don't guess */
781 patch->is_new = 0;
782 patch->is_delete = 0;
785 * Some things may not have the old name in the
786 * rest of the headers anywhere (pure mode changes,
787 * or removing or adding empty files), so we get
788 * the default name from the header.
790 patch->def_name = git_header_name(line, len);
792 line += len;
793 size -= len;
794 linenr++;
795 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
796 static const struct opentry {
797 const char *str;
798 int (*fn)(const char *, struct patch *);
799 } optable[] = {
800 { "@@ -", gitdiff_hdrend },
801 { "--- ", gitdiff_oldname },
802 { "+++ ", gitdiff_newname },
803 { "old mode ", gitdiff_oldmode },
804 { "new mode ", gitdiff_newmode },
805 { "deleted file mode ", gitdiff_delete },
806 { "new file mode ", gitdiff_newfile },
807 { "copy from ", gitdiff_copysrc },
808 { "copy to ", gitdiff_copydst },
809 { "rename old ", gitdiff_renamesrc },
810 { "rename new ", gitdiff_renamedst },
811 { "rename from ", gitdiff_renamesrc },
812 { "rename to ", gitdiff_renamedst },
813 { "similarity index ", gitdiff_similarity },
814 { "dissimilarity index ", gitdiff_dissimilarity },
815 { "index ", gitdiff_index },
816 { "", gitdiff_unrecognized },
818 int i;
820 len = linelen(line, size);
821 if (!len || line[len-1] != '\n')
822 break;
823 for (i = 0; i < ARRAY_SIZE(optable); i++) {
824 const struct opentry *p = optable + i;
825 int oplen = strlen(p->str);
826 if (len < oplen || memcmp(p->str, line, oplen))
827 continue;
828 if (p->fn(line + oplen, patch) < 0)
829 return offset;
830 break;
834 return offset;
837 static int parse_num(const char *line, unsigned long *p)
839 char *ptr;
841 if (!isdigit(*line))
842 return 0;
843 *p = strtoul(line, &ptr, 10);
844 return ptr - line;
847 static int parse_range(const char *line, int len, int offset, const char *expect,
848 unsigned long *p1, unsigned long *p2)
850 int digits, ex;
852 if (offset < 0 || offset >= len)
853 return -1;
854 line += offset;
855 len -= offset;
857 digits = parse_num(line, p1);
858 if (!digits)
859 return -1;
861 offset += digits;
862 line += digits;
863 len -= digits;
865 *p2 = 1;
866 if (*line == ',') {
867 digits = parse_num(line+1, p2);
868 if (!digits)
869 return -1;
871 offset += digits+1;
872 line += digits+1;
873 len -= digits+1;
876 ex = strlen(expect);
877 if (ex > len)
878 return -1;
879 if (memcmp(line, expect, ex))
880 return -1;
882 return offset + ex;
886 * Parse a unified diff fragment header of the
887 * form "@@ -a,b +c,d @@"
889 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
891 int offset;
893 if (!len || line[len-1] != '\n')
894 return -1;
896 /* Figure out the number of lines in a fragment */
897 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
898 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
900 return offset;
903 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
905 unsigned long offset, len;
907 patch->is_toplevel_relative = 0;
908 patch->is_rename = patch->is_copy = 0;
909 patch->is_new = patch->is_delete = -1;
910 patch->old_mode = patch->new_mode = 0;
911 patch->old_name = patch->new_name = NULL;
912 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
913 unsigned long nextlen;
915 len = linelen(line, size);
916 if (!len)
917 break;
919 /* Testing this early allows us to take a few shortcuts.. */
920 if (len < 6)
921 continue;
924 * Make sure we don't find any unconnected patch fragments.
925 * That's a sign that we didn't find a header, and that a
926 * patch has become corrupted/broken up.
928 if (!memcmp("@@ -", line, 4)) {
929 struct fragment dummy;
930 if (parse_fragment_header(line, len, &dummy) < 0)
931 continue;
932 die("patch fragment without header at line %d: %.*s",
933 linenr, (int)len-1, line);
936 if (size < len + 6)
937 break;
940 * Git patch? It might not have a real patch, just a rename
941 * or mode change, so we handle that specially
943 if (!memcmp("diff --git ", line, 11)) {
944 int git_hdr_len = parse_git_header(line, len, size, patch);
945 if (git_hdr_len <= len)
946 continue;
947 if (!patch->old_name && !patch->new_name) {
948 if (!patch->def_name)
949 die("git diff header lacks filename information (line %d)", linenr);
950 patch->old_name = patch->new_name = patch->def_name;
952 patch->is_toplevel_relative = 1;
953 *hdrsize = git_hdr_len;
954 return offset;
957 /* --- followed by +++ ? */
958 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
959 continue;
962 * We only accept unified patches, so we want it to
963 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
964 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
966 nextlen = linelen(line + len, size - len);
967 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
968 continue;
970 /* Ok, we'll consider it a patch */
971 parse_traditional_patch(line, line+len, patch);
972 *hdrsize = len + nextlen;
973 linenr += 2;
974 return offset;
976 return -1;
979 static void check_whitespace(const char *line, int len, unsigned ws_rule)
981 char *err;
982 unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
983 NULL, NULL, NULL, NULL);
984 if (!result)
985 return;
987 whitespace_error++;
988 if (squelch_whitespace_errors &&
989 squelch_whitespace_errors < whitespace_error)
991 else {
992 err = whitespace_error_string(result);
993 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
994 patch_input_file, linenr, err, len - 2, line + 1);
995 free(err);
1000 * Parse a unified diff. Note that this really needs to parse each
1001 * fragment separately, since the only way to know the difference
1002 * between a "---" that is part of a patch, and a "---" that starts
1003 * the next patch is to look at the line counts..
1005 static int parse_fragment(char *line, unsigned long size,
1006 struct patch *patch, struct fragment *fragment)
1008 int added, deleted;
1009 int len = linelen(line, size), offset;
1010 unsigned long oldlines, newlines;
1011 unsigned long leading, trailing;
1013 offset = parse_fragment_header(line, len, fragment);
1014 if (offset < 0)
1015 return -1;
1016 oldlines = fragment->oldlines;
1017 newlines = fragment->newlines;
1018 leading = 0;
1019 trailing = 0;
1021 /* Parse the thing.. */
1022 line += len;
1023 size -= len;
1024 linenr++;
1025 added = deleted = 0;
1026 for (offset = len;
1027 0 < size;
1028 offset += len, size -= len, line += len, linenr++) {
1029 if (!oldlines && !newlines)
1030 break;
1031 len = linelen(line, size);
1032 if (!len || line[len-1] != '\n')
1033 return -1;
1034 switch (*line) {
1035 default:
1036 return -1;
1037 case '\n': /* newer GNU diff, an empty context line */
1038 case ' ':
1039 oldlines--;
1040 newlines--;
1041 if (!deleted && !added)
1042 leading++;
1043 trailing++;
1044 break;
1045 case '-':
1046 if (apply_in_reverse &&
1047 ws_error_action != nowarn_ws_error)
1048 check_whitespace(line, len, patch->ws_rule);
1049 deleted++;
1050 oldlines--;
1051 trailing = 0;
1052 break;
1053 case '+':
1054 if (!apply_in_reverse &&
1055 ws_error_action != nowarn_ws_error)
1056 check_whitespace(line, len, patch->ws_rule);
1057 added++;
1058 newlines--;
1059 trailing = 0;
1060 break;
1063 * We allow "\ No newline at end of file". Depending
1064 * on locale settings when the patch was produced we
1065 * don't know what this line looks like. The only
1066 * thing we do know is that it begins with "\ ".
1067 * Checking for 12 is just for sanity check -- any
1068 * l10n of "\ No newline..." is at least that long.
1070 case '\\':
1071 if (len < 12 || memcmp(line, "\\ ", 2))
1072 return -1;
1073 break;
1076 if (oldlines || newlines)
1077 return -1;
1078 fragment->leading = leading;
1079 fragment->trailing = trailing;
1082 * If a fragment ends with an incomplete line, we failed to include
1083 * it in the above loop because we hit oldlines == newlines == 0
1084 * before seeing it.
1086 if (12 < size && !memcmp(line, "\\ ", 2))
1087 offset += linelen(line, size);
1089 patch->lines_added += added;
1090 patch->lines_deleted += deleted;
1092 if (0 < patch->is_new && oldlines)
1093 return error("new file depends on old contents");
1094 if (0 < patch->is_delete && newlines)
1095 return error("deleted file still has contents");
1096 return offset;
1099 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1101 unsigned long offset = 0;
1102 unsigned long oldlines = 0, newlines = 0, context = 0;
1103 struct fragment **fragp = &patch->fragments;
1105 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1106 struct fragment *fragment;
1107 int len;
1109 fragment = xcalloc(1, sizeof(*fragment));
1110 len = parse_fragment(line, size, patch, fragment);
1111 if (len <= 0)
1112 die("corrupt patch at line %d", linenr);
1113 fragment->patch = line;
1114 fragment->size = len;
1115 oldlines += fragment->oldlines;
1116 newlines += fragment->newlines;
1117 context += fragment->leading + fragment->trailing;
1119 *fragp = fragment;
1120 fragp = &fragment->next;
1122 offset += len;
1123 line += len;
1124 size -= len;
1128 * If something was removed (i.e. we have old-lines) it cannot
1129 * be creation, and if something was added it cannot be
1130 * deletion. However, the reverse is not true; --unified=0
1131 * patches that only add are not necessarily creation even
1132 * though they do not have any old lines, and ones that only
1133 * delete are not necessarily deletion.
1135 * Unfortunately, a real creation/deletion patch do _not_ have
1136 * any context line by definition, so we cannot safely tell it
1137 * apart with --unified=0 insanity. At least if the patch has
1138 * more than one hunk it is not creation or deletion.
1140 if (patch->is_new < 0 &&
1141 (oldlines || (patch->fragments && patch->fragments->next)))
1142 patch->is_new = 0;
1143 if (patch->is_delete < 0 &&
1144 (newlines || (patch->fragments && patch->fragments->next)))
1145 patch->is_delete = 0;
1146 if (!unidiff_zero || context) {
1147 /* If the user says the patch is not generated with
1148 * --unified=0, or if we have seen context lines,
1149 * then not having oldlines means the patch is creation,
1150 * and not having newlines means the patch is deletion.
1152 if (patch->is_new < 0 && !oldlines) {
1153 patch->is_new = 1;
1154 patch->old_name = NULL;
1156 if (patch->is_delete < 0 && !newlines) {
1157 patch->is_delete = 1;
1158 patch->new_name = NULL;
1162 if (0 < patch->is_new && oldlines)
1163 die("new file %s depends on old contents", patch->new_name);
1164 if (0 < patch->is_delete && newlines)
1165 die("deleted file %s still has contents", patch->old_name);
1166 if (!patch->is_delete && !newlines && context)
1167 fprintf(stderr, "** warning: file %s becomes empty but "
1168 "is not deleted\n", patch->new_name);
1170 return offset;
1173 static inline int metadata_changes(struct patch *patch)
1175 return patch->is_rename > 0 ||
1176 patch->is_copy > 0 ||
1177 patch->is_new > 0 ||
1178 patch->is_delete ||
1179 (patch->old_mode && patch->new_mode &&
1180 patch->old_mode != patch->new_mode);
1183 static char *inflate_it(const void *data, unsigned long size,
1184 unsigned long inflated_size)
1186 z_stream stream;
1187 void *out;
1188 int st;
1190 memset(&stream, 0, sizeof(stream));
1192 stream.next_in = (unsigned char *)data;
1193 stream.avail_in = size;
1194 stream.next_out = out = xmalloc(inflated_size);
1195 stream.avail_out = inflated_size;
1196 inflateInit(&stream);
1197 st = inflate(&stream, Z_FINISH);
1198 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1199 free(out);
1200 return NULL;
1202 return out;
1205 static struct fragment *parse_binary_hunk(char **buf_p,
1206 unsigned long *sz_p,
1207 int *status_p,
1208 int *used_p)
1211 * Expect a line that begins with binary patch method ("literal"
1212 * or "delta"), followed by the length of data before deflating.
1213 * a sequence of 'length-byte' followed by base-85 encoded data
1214 * should follow, terminated by a newline.
1216 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1217 * and we would limit the patch line to 66 characters,
1218 * so one line can fit up to 13 groups that would decode
1219 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1220 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1222 int llen, used;
1223 unsigned long size = *sz_p;
1224 char *buffer = *buf_p;
1225 int patch_method;
1226 unsigned long origlen;
1227 char *data = NULL;
1228 int hunk_size = 0;
1229 struct fragment *frag;
1231 llen = linelen(buffer, size);
1232 used = llen;
1234 *status_p = 0;
1236 if (!prefixcmp(buffer, "delta ")) {
1237 patch_method = BINARY_DELTA_DEFLATED;
1238 origlen = strtoul(buffer + 6, NULL, 10);
1240 else if (!prefixcmp(buffer, "literal ")) {
1241 patch_method = BINARY_LITERAL_DEFLATED;
1242 origlen = strtoul(buffer + 8, NULL, 10);
1244 else
1245 return NULL;
1247 linenr++;
1248 buffer += llen;
1249 while (1) {
1250 int byte_length, max_byte_length, newsize;
1251 llen = linelen(buffer, size);
1252 used += llen;
1253 linenr++;
1254 if (llen == 1) {
1255 /* consume the blank line */
1256 buffer++;
1257 size--;
1258 break;
1261 * Minimum line is "A00000\n" which is 7-byte long,
1262 * and the line length must be multiple of 5 plus 2.
1264 if ((llen < 7) || (llen-2) % 5)
1265 goto corrupt;
1266 max_byte_length = (llen - 2) / 5 * 4;
1267 byte_length = *buffer;
1268 if ('A' <= byte_length && byte_length <= 'Z')
1269 byte_length = byte_length - 'A' + 1;
1270 else if ('a' <= byte_length && byte_length <= 'z')
1271 byte_length = byte_length - 'a' + 27;
1272 else
1273 goto corrupt;
1274 /* if the input length was not multiple of 4, we would
1275 * have filler at the end but the filler should never
1276 * exceed 3 bytes
1278 if (max_byte_length < byte_length ||
1279 byte_length <= max_byte_length - 4)
1280 goto corrupt;
1281 newsize = hunk_size + byte_length;
1282 data = xrealloc(data, newsize);
1283 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1284 goto corrupt;
1285 hunk_size = newsize;
1286 buffer += llen;
1287 size -= llen;
1290 frag = xcalloc(1, sizeof(*frag));
1291 frag->patch = inflate_it(data, hunk_size, origlen);
1292 if (!frag->patch)
1293 goto corrupt;
1294 free(data);
1295 frag->size = origlen;
1296 *buf_p = buffer;
1297 *sz_p = size;
1298 *used_p = used;
1299 frag->binary_patch_method = patch_method;
1300 return frag;
1302 corrupt:
1303 free(data);
1304 *status_p = -1;
1305 error("corrupt binary patch at line %d: %.*s",
1306 linenr-1, llen-1, buffer);
1307 return NULL;
1310 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1313 * We have read "GIT binary patch\n"; what follows is a line
1314 * that says the patch method (currently, either "literal" or
1315 * "delta") and the length of data before deflating; a
1316 * sequence of 'length-byte' followed by base-85 encoded data
1317 * follows.
1319 * When a binary patch is reversible, there is another binary
1320 * hunk in the same format, starting with patch method (either
1321 * "literal" or "delta") with the length of data, and a sequence
1322 * of length-byte + base-85 encoded data, terminated with another
1323 * empty line. This data, when applied to the postimage, produces
1324 * the preimage.
1326 struct fragment *forward;
1327 struct fragment *reverse;
1328 int status;
1329 int used, used_1;
1331 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1332 if (!forward && !status)
1333 /* there has to be one hunk (forward hunk) */
1334 return error("unrecognized binary patch at line %d", linenr-1);
1335 if (status)
1336 /* otherwise we already gave an error message */
1337 return status;
1339 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1340 if (reverse)
1341 used += used_1;
1342 else if (status) {
1344 * Not having reverse hunk is not an error, but having
1345 * a corrupt reverse hunk is.
1347 free((void*) forward->patch);
1348 free(forward);
1349 return status;
1351 forward->next = reverse;
1352 patch->fragments = forward;
1353 patch->is_binary = 1;
1354 return used;
1357 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1359 int hdrsize, patchsize;
1360 int offset = find_header(buffer, size, &hdrsize, patch);
1362 if (offset < 0)
1363 return offset;
1365 patch->ws_rule = whitespace_rule(patch->new_name
1366 ? patch->new_name
1367 : patch->old_name);
1369 patchsize = parse_single_patch(buffer + offset + hdrsize,
1370 size - offset - hdrsize, patch);
1372 if (!patchsize) {
1373 static const char *binhdr[] = {
1374 "Binary files ",
1375 "Files ",
1376 NULL,
1378 static const char git_binary[] = "GIT binary patch\n";
1379 int i;
1380 int hd = hdrsize + offset;
1381 unsigned long llen = linelen(buffer + hd, size - hd);
1383 if (llen == sizeof(git_binary) - 1 &&
1384 !memcmp(git_binary, buffer + hd, llen)) {
1385 int used;
1386 linenr++;
1387 used = parse_binary(buffer + hd + llen,
1388 size - hd - llen, patch);
1389 if (used)
1390 patchsize = used + llen;
1391 else
1392 patchsize = 0;
1394 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1395 for (i = 0; binhdr[i]; i++) {
1396 int len = strlen(binhdr[i]);
1397 if (len < size - hd &&
1398 !memcmp(binhdr[i], buffer + hd, len)) {
1399 linenr++;
1400 patch->is_binary = 1;
1401 patchsize = llen;
1402 break;
1407 /* Empty patch cannot be applied if it is a text patch
1408 * without metadata change. A binary patch appears
1409 * empty to us here.
1411 if ((apply || check) &&
1412 (!patch->is_binary && !metadata_changes(patch)))
1413 die("patch with only garbage at line %d", linenr);
1416 return offset + hdrsize + patchsize;
1419 #define swap(a,b) myswap((a),(b),sizeof(a))
1421 #define myswap(a, b, size) do { \
1422 unsigned char mytmp[size]; \
1423 memcpy(mytmp, &a, size); \
1424 memcpy(&a, &b, size); \
1425 memcpy(&b, mytmp, size); \
1426 } while (0)
1428 static void reverse_patches(struct patch *p)
1430 for (; p; p = p->next) {
1431 struct fragment *frag = p->fragments;
1433 swap(p->new_name, p->old_name);
1434 swap(p->new_mode, p->old_mode);
1435 swap(p->is_new, p->is_delete);
1436 swap(p->lines_added, p->lines_deleted);
1437 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1439 for (; frag; frag = frag->next) {
1440 swap(frag->newpos, frag->oldpos);
1441 swap(frag->newlines, frag->oldlines);
1446 static const char pluses[] =
1447 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1448 static const char minuses[]=
1449 "----------------------------------------------------------------------";
1451 static void show_stats(struct patch *patch)
1453 struct strbuf qname;
1454 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1455 int max, add, del;
1457 strbuf_init(&qname, 0);
1458 quote_c_style(cp, &qname, NULL, 0);
1461 * "scale" the filename
1463 max = max_len;
1464 if (max > 50)
1465 max = 50;
1467 if (qname.len > max) {
1468 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1469 if (!cp)
1470 cp = qname.buf + qname.len + 3 - max;
1471 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1474 if (patch->is_binary) {
1475 printf(" %-*s | Bin\n", max, qname.buf);
1476 strbuf_release(&qname);
1477 return;
1480 printf(" %-*s |", max, qname.buf);
1481 strbuf_release(&qname);
1484 * scale the add/delete
1486 max = max + max_change > 70 ? 70 - max : max_change;
1487 add = patch->lines_added;
1488 del = patch->lines_deleted;
1490 if (max_change > 0) {
1491 int total = ((add + del) * max + max_change / 2) / max_change;
1492 add = (add * max + max_change / 2) / max_change;
1493 del = total - add;
1495 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1496 add, pluses, del, minuses);
1499 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1501 switch (st->st_mode & S_IFMT) {
1502 case S_IFLNK:
1503 strbuf_grow(buf, st->st_size);
1504 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1505 return -1;
1506 strbuf_setlen(buf, st->st_size);
1507 return 0;
1508 case S_IFREG:
1509 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1510 return error("unable to open or read %s", path);
1511 convert_to_git(path, buf->buf, buf->len, buf, 0);
1512 return 0;
1513 default:
1514 return -1;
1518 static void update_pre_post_images(struct image *preimage,
1519 struct image *postimage,
1520 char *buf,
1521 size_t len)
1523 int i, ctx;
1524 char *new, *old, *fixed;
1525 struct image fixed_preimage;
1528 * Update the preimage with whitespace fixes. Note that we
1529 * are not losing preimage->buf -- apply_one_fragment() will
1530 * free "oldlines".
1532 prepare_image(&fixed_preimage, buf, len, 1);
1533 assert(fixed_preimage.nr == preimage->nr);
1534 for (i = 0; i < preimage->nr; i++)
1535 fixed_preimage.line[i].flag = preimage->line[i].flag;
1536 free(preimage->line_allocated);
1537 *preimage = fixed_preimage;
1540 * Adjust the common context lines in postimage, in place.
1541 * This is possible because whitespace fixing does not make
1542 * the string grow.
1544 new = old = postimage->buf;
1545 fixed = preimage->buf;
1546 for (i = ctx = 0; i < postimage->nr; i++) {
1547 size_t len = postimage->line[i].len;
1548 if (!(postimage->line[i].flag & LINE_COMMON)) {
1549 /* an added line -- no counterparts in preimage */
1550 memmove(new, old, len);
1551 old += len;
1552 new += len;
1553 continue;
1556 /* a common context -- skip it in the original postimage */
1557 old += len;
1559 /* and find the corresponding one in the fixed preimage */
1560 while (ctx < preimage->nr &&
1561 !(preimage->line[ctx].flag & LINE_COMMON)) {
1562 fixed += preimage->line[ctx].len;
1563 ctx++;
1565 if (preimage->nr <= ctx)
1566 die("oops");
1568 /* and copy it in, while fixing the line length */
1569 len = preimage->line[ctx].len;
1570 memcpy(new, fixed, len);
1571 new += len;
1572 fixed += len;
1573 postimage->line[i].len = len;
1574 ctx++;
1577 /* Fix the length of the whole thing */
1578 postimage->len = new - postimage->buf;
1581 static int match_fragment(struct image *img,
1582 struct image *preimage,
1583 struct image *postimage,
1584 unsigned long try,
1585 int try_lno,
1586 unsigned ws_rule,
1587 int match_beginning, int match_end)
1589 int i;
1590 char *fixed_buf, *buf, *orig, *target;
1592 if (preimage->nr + try_lno > img->nr)
1593 return 0;
1595 if (match_beginning && try_lno)
1596 return 0;
1598 if (match_end && preimage->nr + try_lno != img->nr)
1599 return 0;
1601 /* Quick hash check */
1602 for (i = 0; i < preimage->nr; i++)
1603 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1604 return 0;
1607 * Do we have an exact match? If we were told to match
1608 * at the end, size must be exactly at try+fragsize,
1609 * otherwise try+fragsize must be still within the preimage,
1610 * and either case, the old piece should match the preimage
1611 * exactly.
1613 if ((match_end
1614 ? (try + preimage->len == img->len)
1615 : (try + preimage->len <= img->len)) &&
1616 !memcmp(img->buf + try, preimage->buf, preimage->len))
1617 return 1;
1619 if (ws_error_action != correct_ws_error)
1620 return 0;
1623 * The hunk does not apply byte-by-byte, but the hash says
1624 * it might with whitespace fuzz.
1626 fixed_buf = xmalloc(preimage->len + 1);
1627 buf = fixed_buf;
1628 orig = preimage->buf;
1629 target = img->buf + try;
1630 for (i = 0; i < preimage->nr; i++) {
1631 size_t fixlen; /* length after fixing the preimage */
1632 size_t oldlen = preimage->line[i].len;
1633 size_t tgtlen = img->line[try_lno + i].len;
1634 size_t tgtfixlen; /* length after fixing the target line */
1635 char tgtfixbuf[1024], *tgtfix;
1636 int match;
1638 /* Try fixing the line in the preimage */
1639 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1641 /* Try fixing the line in the target */
1642 if (sizeof(tgtfixbuf) < tgtlen)
1643 tgtfix = tgtfixbuf;
1644 else
1645 tgtfix = xmalloc(tgtlen);
1646 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1649 * If they match, either the preimage was based on
1650 * a version before our tree fixed whitespace breakage,
1651 * or we are lacking a whitespace-fix patch the tree
1652 * the preimage was based on already had (i.e. target
1653 * has whitespace breakage, the preimage doesn't).
1654 * In either case, we are fixing the whitespace breakages
1655 * so we might as well take the fix together with their
1656 * real change.
1658 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1660 if (tgtfix != tgtfixbuf)
1661 free(tgtfix);
1662 if (!match)
1663 goto unmatch_exit;
1665 orig += oldlen;
1666 buf += fixlen;
1667 target += tgtlen;
1671 * Yes, the preimage is based on an older version that still
1672 * has whitespace breakages unfixed, and fixing them makes the
1673 * hunk match. Update the context lines in the postimage.
1675 update_pre_post_images(preimage, postimage,
1676 fixed_buf, buf - fixed_buf);
1677 return 1;
1679 unmatch_exit:
1680 free(fixed_buf);
1681 return 0;
1684 static int find_pos(struct image *img,
1685 struct image *preimage,
1686 struct image *postimage,
1687 int line,
1688 unsigned ws_rule,
1689 int match_beginning, int match_end)
1691 int i;
1692 unsigned long backwards, forwards, try;
1693 int backwards_lno, forwards_lno, try_lno;
1695 if (preimage->nr > img->nr)
1696 return -1;
1699 * If match_begining or match_end is specified, there is no
1700 * point starting from a wrong line that will never match and
1701 * wander around and wait for a match at the specified end.
1703 if (match_beginning)
1704 line = 0;
1705 else if (match_end)
1706 line = img->nr - preimage->nr;
1708 if (line > img->nr)
1709 line = img->nr;
1711 try = 0;
1712 for (i = 0; i < line; i++)
1713 try += img->line[i].len;
1716 * There's probably some smart way to do this, but I'll leave
1717 * that to the smart and beautiful people. I'm simple and stupid.
1719 backwards = try;
1720 backwards_lno = line;
1721 forwards = try;
1722 forwards_lno = line;
1723 try_lno = line;
1725 for (i = 0; ; i++) {
1726 if (match_fragment(img, preimage, postimage,
1727 try, try_lno, ws_rule,
1728 match_beginning, match_end))
1729 return try_lno;
1731 again:
1732 if (backwards_lno == 0 && forwards_lno == img->nr)
1733 break;
1735 if (i & 1) {
1736 if (backwards_lno == 0) {
1737 i++;
1738 goto again;
1740 backwards_lno--;
1741 backwards -= img->line[backwards_lno].len;
1742 try = backwards;
1743 try_lno = backwards_lno;
1744 } else {
1745 if (forwards_lno == img->nr) {
1746 i++;
1747 goto again;
1749 forwards += img->line[forwards_lno].len;
1750 forwards_lno++;
1751 try = forwards;
1752 try_lno = forwards_lno;
1756 return -1;
1759 static void remove_first_line(struct image *img)
1761 img->buf += img->line[0].len;
1762 img->len -= img->line[0].len;
1763 img->line++;
1764 img->nr--;
1767 static void remove_last_line(struct image *img)
1769 img->len -= img->line[--img->nr].len;
1772 static void update_image(struct image *img,
1773 int applied_pos,
1774 struct image *preimage,
1775 struct image *postimage)
1778 * remove the copy of preimage at offset in img
1779 * and replace it with postimage
1781 int i, nr;
1782 size_t remove_count, insert_count, applied_at = 0;
1783 char *result;
1785 for (i = 0; i < applied_pos; i++)
1786 applied_at += img->line[i].len;
1788 remove_count = 0;
1789 for (i = 0; i < preimage->nr; i++)
1790 remove_count += img->line[applied_pos + i].len;
1791 insert_count = postimage->len;
1793 /* Adjust the contents */
1794 result = xmalloc(img->len + insert_count - remove_count + 1);
1795 memcpy(result, img->buf, applied_at);
1796 memcpy(result + applied_at, postimage->buf, postimage->len);
1797 memcpy(result + applied_at + postimage->len,
1798 img->buf + (applied_at + remove_count),
1799 img->len - (applied_at + remove_count));
1800 free(img->buf);
1801 img->buf = result;
1802 img->len += insert_count - remove_count;
1803 result[img->len] = '\0';
1805 /* Adjust the line table */
1806 nr = img->nr + postimage->nr - preimage->nr;
1807 if (preimage->nr < postimage->nr) {
1809 * NOTE: this knows that we never call remove_first_line()
1810 * on anything other than pre/post image.
1812 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1813 img->line_allocated = img->line;
1815 if (preimage->nr != postimage->nr)
1816 memmove(img->line + applied_pos + postimage->nr,
1817 img->line + applied_pos + preimage->nr,
1818 (img->nr - (applied_pos + preimage->nr)) *
1819 sizeof(*img->line));
1820 memcpy(img->line + applied_pos,
1821 postimage->line,
1822 postimage->nr * sizeof(*img->line));
1823 img->nr = nr;
1826 static int apply_one_fragment(struct image *img, struct fragment *frag,
1827 int inaccurate_eof, unsigned ws_rule)
1829 int match_beginning, match_end;
1830 const char *patch = frag->patch;
1831 int size = frag->size;
1832 char *old, *new, *oldlines, *newlines;
1833 int new_blank_lines_at_end = 0;
1834 unsigned long leading, trailing;
1835 int pos, applied_pos;
1836 struct image preimage;
1837 struct image postimage;
1839 memset(&preimage, 0, sizeof(preimage));
1840 memset(&postimage, 0, sizeof(postimage));
1841 oldlines = xmalloc(size);
1842 newlines = xmalloc(size);
1844 old = oldlines;
1845 new = newlines;
1846 while (size > 0) {
1847 char first;
1848 int len = linelen(patch, size);
1849 int plen, added;
1850 int added_blank_line = 0;
1852 if (!len)
1853 break;
1856 * "plen" is how much of the line we should use for
1857 * the actual patch data. Normally we just remove the
1858 * first character on the line, but if the line is
1859 * followed by "\ No newline", then we also remove the
1860 * last one (which is the newline, of course).
1862 plen = len - 1;
1863 if (len < size && patch[len] == '\\')
1864 plen--;
1865 first = *patch;
1866 if (apply_in_reverse) {
1867 if (first == '-')
1868 first = '+';
1869 else if (first == '+')
1870 first = '-';
1873 switch (first) {
1874 case '\n':
1875 /* Newer GNU diff, empty context line */
1876 if (plen < 0)
1877 /* ... followed by '\No newline'; nothing */
1878 break;
1879 *old++ = '\n';
1880 *new++ = '\n';
1881 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1882 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1883 break;
1884 case ' ':
1885 case '-':
1886 memcpy(old, patch + 1, plen);
1887 add_line_info(&preimage, old, plen,
1888 (first == ' ' ? LINE_COMMON : 0));
1889 old += plen;
1890 if (first == '-')
1891 break;
1892 /* Fall-through for ' ' */
1893 case '+':
1894 /* --no-add does not add new lines */
1895 if (first == '+' && no_add)
1896 break;
1898 if (first != '+' ||
1899 !whitespace_error ||
1900 ws_error_action != correct_ws_error) {
1901 memcpy(new, patch + 1, plen);
1902 added = plen;
1904 else {
1905 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1907 add_line_info(&postimage, new, added,
1908 (first == '+' ? 0 : LINE_COMMON));
1909 new += added;
1910 if (first == '+' &&
1911 added == 1 && new[-1] == '\n')
1912 added_blank_line = 1;
1913 break;
1914 case '@': case '\\':
1915 /* Ignore it, we already handled it */
1916 break;
1917 default:
1918 if (apply_verbosely)
1919 error("invalid start of line: '%c'", first);
1920 return -1;
1922 if (added_blank_line)
1923 new_blank_lines_at_end++;
1924 else
1925 new_blank_lines_at_end = 0;
1926 patch += len;
1927 size -= len;
1929 if (inaccurate_eof &&
1930 old > oldlines && old[-1] == '\n' &&
1931 new > newlines && new[-1] == '\n') {
1932 old--;
1933 new--;
1936 leading = frag->leading;
1937 trailing = frag->trailing;
1940 * A hunk to change lines at the beginning would begin with
1941 * @@ -1,L +N,M @@
1943 * And a hunk to add to an empty file would begin with
1944 * @@ -0,0 +N,M @@
1946 * In other words, a hunk that is (frag->oldpos <= 1) with or
1947 * without leading context must match at the beginning.
1949 match_beginning = frag->oldpos <= 1;
1952 * A hunk without trailing lines must match at the end.
1953 * However, we simply cannot tell if a hunk must match end
1954 * from the lack of trailing lines if the patch was generated
1955 * with unidiff without any context.
1957 match_end = !unidiff_zero && !trailing;
1959 pos = frag->newpos ? (frag->newpos - 1) : 0;
1960 preimage.buf = oldlines;
1961 preimage.len = old - oldlines;
1962 postimage.buf = newlines;
1963 postimage.len = new - newlines;
1964 preimage.line = preimage.line_allocated;
1965 postimage.line = postimage.line_allocated;
1967 for (;;) {
1969 applied_pos = find_pos(img, &preimage, &postimage, pos,
1970 ws_rule, match_beginning, match_end);
1972 if (applied_pos >= 0)
1973 break;
1975 /* Am I at my context limits? */
1976 if ((leading <= p_context) && (trailing <= p_context))
1977 break;
1978 if (match_beginning || match_end) {
1979 match_beginning = match_end = 0;
1980 continue;
1984 * Reduce the number of context lines; reduce both
1985 * leading and trailing if they are equal otherwise
1986 * just reduce the larger context.
1988 if (leading >= trailing) {
1989 remove_first_line(&preimage);
1990 remove_first_line(&postimage);
1991 pos--;
1992 leading--;
1994 if (trailing > leading) {
1995 remove_last_line(&preimage);
1996 remove_last_line(&postimage);
1997 trailing--;
2001 if (applied_pos >= 0) {
2002 if (ws_error_action == correct_ws_error &&
2003 new_blank_lines_at_end &&
2004 postimage.nr + applied_pos == img->nr) {
2006 * If the patch application adds blank lines
2007 * at the end, and if the patch applies at the
2008 * end of the image, remove those added blank
2009 * lines.
2011 while (new_blank_lines_at_end--)
2012 remove_last_line(&postimage);
2016 * Warn if it was necessary to reduce the number
2017 * of context lines.
2019 if ((leading != frag->leading) ||
2020 (trailing != frag->trailing))
2021 fprintf(stderr, "Context reduced to (%ld/%ld)"
2022 " to apply fragment at %d\n",
2023 leading, trailing, applied_pos+1);
2024 update_image(img, applied_pos, &preimage, &postimage);
2025 } else {
2026 if (apply_verbosely)
2027 error("while searching for:\n%.*s",
2028 (int)(old - oldlines), oldlines);
2031 free(oldlines);
2032 free(newlines);
2033 free(preimage.line_allocated);
2034 free(postimage.line_allocated);
2036 return (applied_pos < 0);
2039 static int apply_binary_fragment(struct image *img, struct patch *patch)
2041 struct fragment *fragment = patch->fragments;
2042 unsigned long len;
2043 void *dst;
2045 /* Binary patch is irreversible without the optional second hunk */
2046 if (apply_in_reverse) {
2047 if (!fragment->next)
2048 return error("cannot reverse-apply a binary patch "
2049 "without the reverse hunk to '%s'",
2050 patch->new_name
2051 ? patch->new_name : patch->old_name);
2052 fragment = fragment->next;
2054 switch (fragment->binary_patch_method) {
2055 case BINARY_DELTA_DEFLATED:
2056 dst = patch_delta(img->buf, img->len, fragment->patch,
2057 fragment->size, &len);
2058 if (!dst)
2059 return -1;
2060 clear_image(img);
2061 img->buf = dst;
2062 img->len = len;
2063 return 0;
2064 case BINARY_LITERAL_DEFLATED:
2065 clear_image(img);
2066 img->len = fragment->size;
2067 img->buf = xmalloc(img->len+1);
2068 memcpy(img->buf, fragment->patch, img->len);
2069 img->buf[img->len] = '\0';
2070 return 0;
2072 return -1;
2075 static int apply_binary(struct image *img, struct patch *patch)
2077 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2078 unsigned char sha1[20];
2081 * For safety, we require patch index line to contain
2082 * full 40-byte textual SHA1 for old and new, at least for now.
2084 if (strlen(patch->old_sha1_prefix) != 40 ||
2085 strlen(patch->new_sha1_prefix) != 40 ||
2086 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2087 get_sha1_hex(patch->new_sha1_prefix, sha1))
2088 return error("cannot apply binary patch to '%s' "
2089 "without full index line", name);
2091 if (patch->old_name) {
2093 * See if the old one matches what the patch
2094 * applies to.
2096 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2097 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2098 return error("the patch applies to '%s' (%s), "
2099 "which does not match the "
2100 "current contents.",
2101 name, sha1_to_hex(sha1));
2103 else {
2104 /* Otherwise, the old one must be empty. */
2105 if (img->len)
2106 return error("the patch applies to an empty "
2107 "'%s' but it is not empty", name);
2110 get_sha1_hex(patch->new_sha1_prefix, sha1);
2111 if (is_null_sha1(sha1)) {
2112 clear_image(img);
2113 return 0; /* deletion patch */
2116 if (has_sha1_file(sha1)) {
2117 /* We already have the postimage */
2118 enum object_type type;
2119 unsigned long size;
2120 char *result;
2122 result = read_sha1_file(sha1, &type, &size);
2123 if (!result)
2124 return error("the necessary postimage %s for "
2125 "'%s' cannot be read",
2126 patch->new_sha1_prefix, name);
2127 clear_image(img);
2128 img->buf = result;
2129 img->len = size;
2130 } else {
2132 * We have verified buf matches the preimage;
2133 * apply the patch data to it, which is stored
2134 * in the patch->fragments->{patch,size}.
2136 if (apply_binary_fragment(img, patch))
2137 return error("binary patch does not apply to '%s'",
2138 name);
2140 /* verify that the result matches */
2141 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2142 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2143 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2144 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2147 return 0;
2150 static int apply_fragments(struct image *img, struct patch *patch)
2152 struct fragment *frag = patch->fragments;
2153 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2154 unsigned ws_rule = patch->ws_rule;
2155 unsigned inaccurate_eof = patch->inaccurate_eof;
2157 if (patch->is_binary)
2158 return apply_binary(img, patch);
2160 while (frag) {
2161 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2162 error("patch failed: %s:%ld", name, frag->oldpos);
2163 if (!apply_with_reject)
2164 return -1;
2165 frag->rejected = 1;
2167 frag = frag->next;
2169 return 0;
2172 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2174 if (!ce)
2175 return 0;
2177 if (S_ISGITLINK(ce->ce_mode)) {
2178 strbuf_grow(buf, 100);
2179 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2180 } else {
2181 enum object_type type;
2182 unsigned long sz;
2183 char *result;
2185 result = read_sha1_file(ce->sha1, &type, &sz);
2186 if (!result)
2187 return -1;
2188 /* XXX read_sha1_file NUL-terminates */
2189 strbuf_attach(buf, result, sz, sz + 1);
2191 return 0;
2194 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2196 struct strbuf buf;
2197 struct image image;
2198 size_t len;
2199 char *img;
2201 strbuf_init(&buf, 0);
2202 if (cached) {
2203 if (read_file_or_gitlink(ce, &buf))
2204 return error("read of %s failed", patch->old_name);
2205 } else if (patch->old_name) {
2206 if (S_ISGITLINK(patch->old_mode)) {
2207 if (ce) {
2208 read_file_or_gitlink(ce, &buf);
2209 } else {
2211 * There is no way to apply subproject
2212 * patch without looking at the index.
2214 patch->fragments = NULL;
2216 } else {
2217 if (read_old_data(st, patch->old_name, &buf))
2218 return error("read of %s failed", patch->old_name);
2222 img = strbuf_detach(&buf, &len);
2223 prepare_image(&image, img, len, !patch->is_binary);
2225 if (apply_fragments(&image, patch) < 0)
2226 return -1; /* note with --reject this succeeds. */
2227 patch->result = image.buf;
2228 patch->resultsize = image.len;
2229 free(image.line_allocated);
2231 if (0 < patch->is_delete && patch->resultsize)
2232 return error("removal patch leaves file contents");
2234 return 0;
2237 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2239 struct stat nst;
2240 if (!lstat(new_name, &nst)) {
2241 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2242 return 0;
2244 * A leading component of new_name might be a symlink
2245 * that is going to be removed with this patch, but
2246 * still pointing at somewhere that has the path.
2247 * In such a case, path "new_name" does not exist as
2248 * far as git is concerned.
2250 if (has_symlink_leading_path(new_name, NULL))
2251 return 0;
2253 return error("%s: already exists in working directory", new_name);
2255 else if ((errno != ENOENT) && (errno != ENOTDIR))
2256 return error("%s: %s", new_name, strerror(errno));
2257 return 0;
2260 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2262 if (S_ISGITLINK(ce->ce_mode)) {
2263 if (!S_ISDIR(st->st_mode))
2264 return -1;
2265 return 0;
2267 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2270 static int check_patch(struct patch *patch, struct patch *prev_patch)
2272 struct stat st;
2273 const char *old_name = patch->old_name;
2274 const char *new_name = patch->new_name;
2275 const char *name = old_name ? old_name : new_name;
2276 struct cache_entry *ce = NULL;
2277 int ok_if_exists;
2279 patch->rejected = 1; /* we will drop this after we succeed */
2282 * Make sure that we do not have local modifications from the
2283 * index when we are looking at the index. Also make sure
2284 * we have the preimage file to be patched in the work tree,
2285 * unless --cached, which tells git to apply only in the index.
2287 if (old_name) {
2288 int stat_ret = 0;
2289 unsigned st_mode = 0;
2291 if (!cached)
2292 stat_ret = lstat(old_name, &st);
2293 if (check_index) {
2294 int pos = cache_name_pos(old_name, strlen(old_name));
2295 if (pos < 0)
2296 return error("%s: does not exist in index",
2297 old_name);
2298 ce = active_cache[pos];
2299 if (stat_ret < 0) {
2300 struct checkout costate;
2301 if (errno != ENOENT)
2302 return error("%s: %s", old_name,
2303 strerror(errno));
2304 /* checkout */
2305 costate.base_dir = "";
2306 costate.base_dir_len = 0;
2307 costate.force = 0;
2308 costate.quiet = 0;
2309 costate.not_new = 0;
2310 costate.refresh_cache = 1;
2311 if (checkout_entry(ce,
2312 &costate,
2313 NULL) ||
2314 lstat(old_name, &st))
2315 return -1;
2317 if (!cached && verify_index_match(ce, &st))
2318 return error("%s: does not match index",
2319 old_name);
2320 if (cached)
2321 st_mode = ce->ce_mode;
2322 } else if (stat_ret < 0)
2323 return error("%s: %s", old_name, strerror(errno));
2325 if (!cached)
2326 st_mode = ce_mode_from_stat(ce, st.st_mode);
2328 if (patch->is_new < 0)
2329 patch->is_new = 0;
2330 if (!patch->old_mode)
2331 patch->old_mode = st_mode;
2332 if ((st_mode ^ patch->old_mode) & S_IFMT)
2333 return error("%s: wrong type", old_name);
2334 if (st_mode != patch->old_mode)
2335 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2336 old_name, st_mode, patch->old_mode);
2339 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2340 !strcmp(prev_patch->old_name, new_name))
2342 * A type-change diff is always split into a patch to
2343 * delete old, immediately followed by a patch to
2344 * create new (see diff.c::run_diff()); in such a case
2345 * it is Ok that the entry to be deleted by the
2346 * previous patch is still in the working tree and in
2347 * the index.
2349 ok_if_exists = 1;
2350 else
2351 ok_if_exists = 0;
2353 if (new_name &&
2354 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2355 if (check_index &&
2356 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2357 !ok_if_exists)
2358 return error("%s: already exists in index", new_name);
2359 if (!cached) {
2360 int err = check_to_create_blob(new_name, ok_if_exists);
2361 if (err)
2362 return err;
2364 if (!patch->new_mode) {
2365 if (0 < patch->is_new)
2366 patch->new_mode = S_IFREG | 0644;
2367 else
2368 patch->new_mode = patch->old_mode;
2372 if (new_name && old_name) {
2373 int same = !strcmp(old_name, new_name);
2374 if (!patch->new_mode)
2375 patch->new_mode = patch->old_mode;
2376 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2377 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2378 patch->new_mode, new_name, patch->old_mode,
2379 same ? "" : " of ", same ? "" : old_name);
2382 if (apply_data(patch, &st, ce) < 0)
2383 return error("%s: patch does not apply", name);
2384 patch->rejected = 0;
2385 return 0;
2388 static int check_patch_list(struct patch *patch)
2390 struct patch *prev_patch = NULL;
2391 int err = 0;
2393 for (prev_patch = NULL; patch ; patch = patch->next) {
2394 if (apply_verbosely)
2395 say_patch_name(stderr,
2396 "Checking patch ", patch, "...\n");
2397 err |= check_patch(patch, prev_patch);
2398 prev_patch = patch;
2400 return err;
2403 /* This function tries to read the sha1 from the current index */
2404 static int get_current_sha1(const char *path, unsigned char *sha1)
2406 int pos;
2408 if (read_cache() < 0)
2409 return -1;
2410 pos = cache_name_pos(path, strlen(path));
2411 if (pos < 0)
2412 return -1;
2413 hashcpy(sha1, active_cache[pos]->sha1);
2414 return 0;
2417 /* Build an index that contains the just the files needed for a 3way merge */
2418 static void build_fake_ancestor(struct patch *list, const char *filename)
2420 struct patch *patch;
2421 struct index_state result = { 0 };
2422 int fd;
2424 /* Once we start supporting the reverse patch, it may be
2425 * worth showing the new sha1 prefix, but until then...
2427 for (patch = list; patch; patch = patch->next) {
2428 const unsigned char *sha1_ptr;
2429 unsigned char sha1[20];
2430 struct cache_entry *ce;
2431 const char *name;
2433 name = patch->old_name ? patch->old_name : patch->new_name;
2434 if (0 < patch->is_new)
2435 continue;
2436 else if (get_sha1(patch->old_sha1_prefix, sha1))
2437 /* git diff has no index line for mode/type changes */
2438 if (!patch->lines_added && !patch->lines_deleted) {
2439 if (get_current_sha1(patch->new_name, sha1) ||
2440 get_current_sha1(patch->old_name, sha1))
2441 die("mode change for %s, which is not "
2442 "in current HEAD", name);
2443 sha1_ptr = sha1;
2444 } else
2445 die("sha1 information is lacking or useless "
2446 "(%s).", name);
2447 else
2448 sha1_ptr = sha1;
2450 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2451 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2452 die ("Could not add %s to temporary index", name);
2455 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2456 if (fd < 0 || write_index(&result, fd) || close(fd))
2457 die ("Could not write temporary index to %s", filename);
2459 discard_index(&result);
2462 static void stat_patch_list(struct patch *patch)
2464 int files, adds, dels;
2466 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2467 files++;
2468 adds += patch->lines_added;
2469 dels += patch->lines_deleted;
2470 show_stats(patch);
2473 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2476 static void numstat_patch_list(struct patch *patch)
2478 for ( ; patch; patch = patch->next) {
2479 const char *name;
2480 name = patch->new_name ? patch->new_name : patch->old_name;
2481 if (patch->is_binary)
2482 printf("-\t-\t");
2483 else
2484 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2485 write_name_quoted(name, stdout, line_termination);
2489 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2491 if (mode)
2492 printf(" %s mode %06o %s\n", newdelete, mode, name);
2493 else
2494 printf(" %s %s\n", newdelete, name);
2497 static void show_mode_change(struct patch *p, int show_name)
2499 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2500 if (show_name)
2501 printf(" mode change %06o => %06o %s\n",
2502 p->old_mode, p->new_mode, p->new_name);
2503 else
2504 printf(" mode change %06o => %06o\n",
2505 p->old_mode, p->new_mode);
2509 static void show_rename_copy(struct patch *p)
2511 const char *renamecopy = p->is_rename ? "rename" : "copy";
2512 const char *old, *new;
2514 /* Find common prefix */
2515 old = p->old_name;
2516 new = p->new_name;
2517 while (1) {
2518 const char *slash_old, *slash_new;
2519 slash_old = strchr(old, '/');
2520 slash_new = strchr(new, '/');
2521 if (!slash_old ||
2522 !slash_new ||
2523 slash_old - old != slash_new - new ||
2524 memcmp(old, new, slash_new - new))
2525 break;
2526 old = slash_old + 1;
2527 new = slash_new + 1;
2529 /* p->old_name thru old is the common prefix, and old and new
2530 * through the end of names are renames
2532 if (old != p->old_name)
2533 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2534 (int)(old - p->old_name), p->old_name,
2535 old, new, p->score);
2536 else
2537 printf(" %s %s => %s (%d%%)\n", renamecopy,
2538 p->old_name, p->new_name, p->score);
2539 show_mode_change(p, 0);
2542 static void summary_patch_list(struct patch *patch)
2544 struct patch *p;
2546 for (p = patch; p; p = p->next) {
2547 if (p->is_new)
2548 show_file_mode_name("create", p->new_mode, p->new_name);
2549 else if (p->is_delete)
2550 show_file_mode_name("delete", p->old_mode, p->old_name);
2551 else {
2552 if (p->is_rename || p->is_copy)
2553 show_rename_copy(p);
2554 else {
2555 if (p->score) {
2556 printf(" rewrite %s (%d%%)\n",
2557 p->new_name, p->score);
2558 show_mode_change(p, 0);
2560 else
2561 show_mode_change(p, 1);
2567 static void patch_stats(struct patch *patch)
2569 int lines = patch->lines_added + patch->lines_deleted;
2571 if (lines > max_change)
2572 max_change = lines;
2573 if (patch->old_name) {
2574 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2575 if (!len)
2576 len = strlen(patch->old_name);
2577 if (len > max_len)
2578 max_len = len;
2580 if (patch->new_name) {
2581 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2582 if (!len)
2583 len = strlen(patch->new_name);
2584 if (len > max_len)
2585 max_len = len;
2589 static void remove_file(struct patch *patch, int rmdir_empty)
2591 if (update_index) {
2592 if (remove_file_from_cache(patch->old_name) < 0)
2593 die("unable to remove %s from index", patch->old_name);
2595 if (!cached) {
2596 if (S_ISGITLINK(patch->old_mode)) {
2597 if (rmdir(patch->old_name))
2598 warning("unable to remove submodule %s",
2599 patch->old_name);
2600 } else if (!unlink(patch->old_name) && rmdir_empty) {
2601 char *name = xstrdup(patch->old_name);
2602 char *end = strrchr(name, '/');
2603 while (end) {
2604 *end = 0;
2605 if (rmdir(name))
2606 break;
2607 end = strrchr(name, '/');
2609 free(name);
2614 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2616 struct stat st;
2617 struct cache_entry *ce;
2618 int namelen = strlen(path);
2619 unsigned ce_size = cache_entry_size(namelen);
2621 if (!update_index)
2622 return;
2624 ce = xcalloc(1, ce_size);
2625 memcpy(ce->name, path, namelen);
2626 ce->ce_mode = create_ce_mode(mode);
2627 ce->ce_flags = namelen;
2628 if (S_ISGITLINK(mode)) {
2629 const char *s = buf;
2631 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2632 die("corrupt patch for subproject %s", path);
2633 } else {
2634 if (!cached) {
2635 if (lstat(path, &st) < 0)
2636 die("unable to stat newly created file %s",
2637 path);
2638 fill_stat_cache_info(ce, &st);
2640 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2641 die("unable to create backing store for newly created file %s", path);
2643 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2644 die("unable to add cache entry for %s", path);
2647 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2649 int fd;
2650 struct strbuf nbuf;
2652 if (S_ISGITLINK(mode)) {
2653 struct stat st;
2654 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2655 return 0;
2656 return mkdir(path, 0777);
2659 if (has_symlinks && S_ISLNK(mode))
2660 /* Although buf:size is counted string, it also is NUL
2661 * terminated.
2663 return symlink(buf, path);
2665 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2666 if (fd < 0)
2667 return -1;
2669 strbuf_init(&nbuf, 0);
2670 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2671 size = nbuf.len;
2672 buf = nbuf.buf;
2674 write_or_die(fd, buf, size);
2675 strbuf_release(&nbuf);
2677 if (close(fd) < 0)
2678 die("closing file %s: %s", path, strerror(errno));
2679 return 0;
2683 * We optimistically assume that the directories exist,
2684 * which is true 99% of the time anyway. If they don't,
2685 * we create them and try again.
2687 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2689 if (cached)
2690 return;
2691 if (!try_create_file(path, mode, buf, size))
2692 return;
2694 if (errno == ENOENT) {
2695 if (safe_create_leading_directories(path))
2696 return;
2697 if (!try_create_file(path, mode, buf, size))
2698 return;
2701 if (errno == EEXIST || errno == EACCES) {
2702 /* We may be trying to create a file where a directory
2703 * used to be.
2705 struct stat st;
2706 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2707 errno = EEXIST;
2710 if (errno == EEXIST) {
2711 unsigned int nr = getpid();
2713 for (;;) {
2714 const char *newpath;
2715 newpath = mkpath("%s~%u", path, nr);
2716 if (!try_create_file(newpath, mode, buf, size)) {
2717 if (!rename(newpath, path))
2718 return;
2719 unlink(newpath);
2720 break;
2722 if (errno != EEXIST)
2723 break;
2724 ++nr;
2727 die("unable to write file %s mode %o", path, mode);
2730 static void create_file(struct patch *patch)
2732 char *path = patch->new_name;
2733 unsigned mode = patch->new_mode;
2734 unsigned long size = patch->resultsize;
2735 char *buf = patch->result;
2737 if (!mode)
2738 mode = S_IFREG | 0644;
2739 create_one_file(path, mode, buf, size);
2740 add_index_file(path, mode, buf, size);
2743 /* phase zero is to remove, phase one is to create */
2744 static void write_out_one_result(struct patch *patch, int phase)
2746 if (patch->is_delete > 0) {
2747 if (phase == 0)
2748 remove_file(patch, 1);
2749 return;
2751 if (patch->is_new > 0 || patch->is_copy) {
2752 if (phase == 1)
2753 create_file(patch);
2754 return;
2757 * Rename or modification boils down to the same
2758 * thing: remove the old, write the new
2760 if (phase == 0)
2761 remove_file(patch, patch->is_rename);
2762 if (phase == 1)
2763 create_file(patch);
2766 static int write_out_one_reject(struct patch *patch)
2768 FILE *rej;
2769 char namebuf[PATH_MAX];
2770 struct fragment *frag;
2771 int cnt = 0;
2773 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2774 if (!frag->rejected)
2775 continue;
2776 cnt++;
2779 if (!cnt) {
2780 if (apply_verbosely)
2781 say_patch_name(stderr,
2782 "Applied patch ", patch, " cleanly.\n");
2783 return 0;
2786 /* This should not happen, because a removal patch that leaves
2787 * contents are marked "rejected" at the patch level.
2789 if (!patch->new_name)
2790 die("internal error");
2792 /* Say this even without --verbose */
2793 say_patch_name(stderr, "Applying patch ", patch, " with");
2794 fprintf(stderr, " %d rejects...\n", cnt);
2796 cnt = strlen(patch->new_name);
2797 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2798 cnt = ARRAY_SIZE(namebuf) - 5;
2799 fprintf(stderr,
2800 "warning: truncating .rej filename to %.*s.rej",
2801 cnt - 1, patch->new_name);
2803 memcpy(namebuf, patch->new_name, cnt);
2804 memcpy(namebuf + cnt, ".rej", 5);
2806 rej = fopen(namebuf, "w");
2807 if (!rej)
2808 return error("cannot open %s: %s", namebuf, strerror(errno));
2810 /* Normal git tools never deal with .rej, so do not pretend
2811 * this is a git patch by saying --git nor give extended
2812 * headers. While at it, maybe please "kompare" that wants
2813 * the trailing TAB and some garbage at the end of line ;-).
2815 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2816 patch->new_name, patch->new_name);
2817 for (cnt = 1, frag = patch->fragments;
2818 frag;
2819 cnt++, frag = frag->next) {
2820 if (!frag->rejected) {
2821 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2822 continue;
2824 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2825 fprintf(rej, "%.*s", frag->size, frag->patch);
2826 if (frag->patch[frag->size-1] != '\n')
2827 fputc('\n', rej);
2829 fclose(rej);
2830 return -1;
2833 static int write_out_results(struct patch *list, int skipped_patch)
2835 int phase;
2836 int errs = 0;
2837 struct patch *l;
2839 if (!list && !skipped_patch)
2840 return error("No changes");
2842 for (phase = 0; phase < 2; phase++) {
2843 l = list;
2844 while (l) {
2845 if (l->rejected)
2846 errs = 1;
2847 else {
2848 write_out_one_result(l, phase);
2849 if (phase == 1 && write_out_one_reject(l))
2850 errs = 1;
2852 l = l->next;
2855 return errs;
2858 static struct lock_file lock_file;
2860 static struct excludes {
2861 struct excludes *next;
2862 const char *path;
2863 } *excludes;
2865 static int use_patch(struct patch *p)
2867 const char *pathname = p->new_name ? p->new_name : p->old_name;
2868 struct excludes *x = excludes;
2869 while (x) {
2870 if (fnmatch(x->path, pathname, 0) == 0)
2871 return 0;
2872 x = x->next;
2874 if (0 < prefix_length) {
2875 int pathlen = strlen(pathname);
2876 if (pathlen <= prefix_length ||
2877 memcmp(prefix, pathname, prefix_length))
2878 return 0;
2880 return 1;
2883 static void prefix_one(char **name)
2885 char *old_name = *name;
2886 if (!old_name)
2887 return;
2888 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2889 free(old_name);
2892 static void prefix_patches(struct patch *p)
2894 if (!prefix || p->is_toplevel_relative)
2895 return;
2896 for ( ; p; p = p->next) {
2897 if (p->new_name == p->old_name) {
2898 char *prefixed = p->new_name;
2899 prefix_one(&prefixed);
2900 p->new_name = p->old_name = prefixed;
2902 else {
2903 prefix_one(&p->new_name);
2904 prefix_one(&p->old_name);
2909 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2911 size_t offset;
2912 struct strbuf buf;
2913 struct patch *list = NULL, **listp = &list;
2914 int skipped_patch = 0;
2916 strbuf_init(&buf, 0);
2917 patch_input_file = filename;
2918 read_patch_file(&buf, fd);
2919 offset = 0;
2920 while (offset < buf.len) {
2921 struct patch *patch;
2922 int nr;
2924 patch = xcalloc(1, sizeof(*patch));
2925 patch->inaccurate_eof = inaccurate_eof;
2926 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2927 if (nr < 0)
2928 break;
2929 if (apply_in_reverse)
2930 reverse_patches(patch);
2931 if (prefix)
2932 prefix_patches(patch);
2933 if (use_patch(patch)) {
2934 patch_stats(patch);
2935 *listp = patch;
2936 listp = &patch->next;
2938 else {
2939 /* perhaps free it a bit better? */
2940 free(patch);
2941 skipped_patch++;
2943 offset += nr;
2946 if (whitespace_error && (ws_error_action == die_on_ws_error))
2947 apply = 0;
2949 update_index = check_index && apply;
2950 if (update_index && newfd < 0)
2951 newfd = hold_locked_index(&lock_file, 1);
2953 if (check_index) {
2954 if (read_cache() < 0)
2955 die("unable to read index file");
2958 if ((check || apply) &&
2959 check_patch_list(list) < 0 &&
2960 !apply_with_reject)
2961 exit(1);
2963 if (apply && write_out_results(list, skipped_patch))
2964 exit(1);
2966 if (fake_ancestor)
2967 build_fake_ancestor(list, fake_ancestor);
2969 if (diffstat)
2970 stat_patch_list(list);
2972 if (numstat)
2973 numstat_patch_list(list);
2975 if (summary)
2976 summary_patch_list(list);
2978 strbuf_release(&buf);
2979 return 0;
2982 static int git_apply_config(const char *var, const char *value)
2984 if (!strcmp(var, "apply.whitespace"))
2985 return git_config_string(&apply_default_whitespace, var, value);
2986 return git_default_config(var, value);
2990 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2992 int i;
2993 int read_stdin = 1;
2994 int inaccurate_eof = 0;
2995 int errs = 0;
2996 int is_not_gitdir;
2998 const char *whitespace_option = NULL;
3000 prefix = setup_git_directory_gently(&is_not_gitdir);
3001 prefix_length = prefix ? strlen(prefix) : 0;
3002 git_config(git_apply_config);
3003 if (apply_default_whitespace)
3004 parse_whitespace_option(apply_default_whitespace);
3006 for (i = 1; i < argc; i++) {
3007 const char *arg = argv[i];
3008 char *end;
3009 int fd;
3011 if (!strcmp(arg, "-")) {
3012 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3013 read_stdin = 0;
3014 continue;
3016 if (!prefixcmp(arg, "--exclude=")) {
3017 struct excludes *x = xmalloc(sizeof(*x));
3018 x->path = arg + 10;
3019 x->next = excludes;
3020 excludes = x;
3021 continue;
3023 if (!prefixcmp(arg, "-p")) {
3024 p_value = atoi(arg + 2);
3025 p_value_known = 1;
3026 continue;
3028 if (!strcmp(arg, "--no-add")) {
3029 no_add = 1;
3030 continue;
3032 if (!strcmp(arg, "--stat")) {
3033 apply = 0;
3034 diffstat = 1;
3035 continue;
3037 if (!strcmp(arg, "--allow-binary-replacement") ||
3038 !strcmp(arg, "--binary")) {
3039 continue; /* now no-op */
3041 if (!strcmp(arg, "--numstat")) {
3042 apply = 0;
3043 numstat = 1;
3044 continue;
3046 if (!strcmp(arg, "--summary")) {
3047 apply = 0;
3048 summary = 1;
3049 continue;
3051 if (!strcmp(arg, "--check")) {
3052 apply = 0;
3053 check = 1;
3054 continue;
3056 if (!strcmp(arg, "--index")) {
3057 if (is_not_gitdir)
3058 die("--index outside a repository");
3059 check_index = 1;
3060 continue;
3062 if (!strcmp(arg, "--cached")) {
3063 if (is_not_gitdir)
3064 die("--cached outside a repository");
3065 check_index = 1;
3066 cached = 1;
3067 continue;
3069 if (!strcmp(arg, "--apply")) {
3070 apply = 1;
3071 continue;
3073 if (!strcmp(arg, "--build-fake-ancestor")) {
3074 apply = 0;
3075 if (++i >= argc)
3076 die ("need a filename");
3077 fake_ancestor = argv[i];
3078 continue;
3080 if (!strcmp(arg, "-z")) {
3081 line_termination = 0;
3082 continue;
3084 if (!prefixcmp(arg, "-C")) {
3085 p_context = strtoul(arg + 2, &end, 0);
3086 if (*end != '\0')
3087 die("unrecognized context count '%s'", arg + 2);
3088 continue;
3090 if (!prefixcmp(arg, "--whitespace=")) {
3091 whitespace_option = arg + 13;
3092 parse_whitespace_option(arg + 13);
3093 continue;
3095 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3096 apply_in_reverse = 1;
3097 continue;
3099 if (!strcmp(arg, "--unidiff-zero")) {
3100 unidiff_zero = 1;
3101 continue;
3103 if (!strcmp(arg, "--reject")) {
3104 apply = apply_with_reject = apply_verbosely = 1;
3105 continue;
3107 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3108 apply_verbosely = 1;
3109 continue;
3111 if (!strcmp(arg, "--inaccurate-eof")) {
3112 inaccurate_eof = 1;
3113 continue;
3115 if (0 < prefix_length)
3116 arg = prefix_filename(prefix, prefix_length, arg);
3118 fd = open(arg, O_RDONLY);
3119 if (fd < 0)
3120 die("can't open patch '%s': %s", arg, strerror(errno));
3121 read_stdin = 0;
3122 set_default_whitespace_mode(whitespace_option);
3123 errs |= apply_patch(fd, arg, inaccurate_eof);
3124 close(fd);
3126 set_default_whitespace_mode(whitespace_option);
3127 if (read_stdin)
3128 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3129 if (whitespace_error) {
3130 if (squelch_whitespace_errors &&
3131 squelch_whitespace_errors < whitespace_error) {
3132 int squelched =
3133 whitespace_error - squelch_whitespace_errors;
3134 fprintf(stderr, "warning: squelched %d "
3135 "whitespace error%s\n",
3136 squelched,
3137 squelched == 1 ? "" : "s");
3139 if (ws_error_action == die_on_ws_error)
3140 die("%d line%s add%s whitespace errors.",
3141 whitespace_error,
3142 whitespace_error == 1 ? "" : "s",
3143 whitespace_error == 1 ? "s" : "");
3144 if (applied_after_fixing_ws && apply)
3145 fprintf(stderr, "warning: %d line%s applied after"
3146 " fixing whitespace errors.\n",
3147 applied_after_fixing_ws,
3148 applied_after_fixing_ws == 1 ? "" : "s");
3149 else if (whitespace_error)
3150 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3151 whitespace_error,
3152 whitespace_error == 1 ? "" : "s",
3153 whitespace_error == 1 ? "s" : "");
3156 if (update_index) {
3157 if (write_cache(newfd, active_cache, active_nr) ||
3158 commit_locked_index(&lock_file))
3159 die("Unable to write new index file");
3162 return !!errs;