Merge branch 'dz/apply-again'
[git/dscho.git] / builtin-apply.c
blob318504a03700d3b50ce2848a0ab374ac5f519950
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 "path-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;
62 static void parse_whitespace_option(const char *option)
64 if (!option) {
65 ws_error_action = warn_on_ws_error;
66 return;
68 if (!strcmp(option, "warn")) {
69 ws_error_action = warn_on_ws_error;
70 return;
72 if (!strcmp(option, "nowarn")) {
73 ws_error_action = nowarn_ws_error;
74 return;
76 if (!strcmp(option, "error")) {
77 ws_error_action = die_on_ws_error;
78 return;
80 if (!strcmp(option, "error-all")) {
81 ws_error_action = die_on_ws_error;
82 squelch_whitespace_errors = 0;
83 return;
85 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
86 ws_error_action = correct_ws_error;
87 return;
89 die("unrecognized whitespace option '%s'", option);
92 static void set_default_whitespace_mode(const char *whitespace_option)
94 if (!whitespace_option && !apply_default_whitespace)
95 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
99 * For "diff-stat" like behaviour, we keep track of the biggest change
100 * we've seen, and the longest filename. That allows us to do simple
101 * scaling.
103 static int max_change, max_len;
106 * Various "current state", notably line numbers and what
107 * file (and how) we're patching right now.. The "is_xxxx"
108 * things are flags, where -1 means "don't know yet".
110 static int linenr = 1;
113 * This represents one "hunk" from a patch, starting with
114 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
115 * patch text is pointed at by patch, and its byte length
116 * is stored in size. leading and trailing are the number
117 * of context lines.
119 struct fragment {
120 unsigned long leading, trailing;
121 unsigned long oldpos, oldlines;
122 unsigned long newpos, newlines;
123 const char *patch;
124 int size;
125 int rejected;
126 struct fragment *next;
130 * When dealing with a binary patch, we reuse "leading" field
131 * to store the type of the binary hunk, either deflated "delta"
132 * or deflated "literal".
134 #define binary_patch_method leading
135 #define BINARY_DELTA_DEFLATED 1
136 #define BINARY_LITERAL_DEFLATED 2
139 * This represents a "patch" to a file, both metainfo changes
140 * such as creation/deletion, filemode and content changes represented
141 * as a series of fragments.
143 struct patch {
144 char *new_name, *old_name, *def_name;
145 unsigned int old_mode, new_mode;
146 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
147 int rejected;
148 unsigned ws_rule;
149 unsigned long deflate_origlen;
150 int lines_added, lines_deleted;
151 int score;
152 unsigned int is_toplevel_relative:1;
153 unsigned int inaccurate_eof:1;
154 unsigned int is_binary:1;
155 unsigned int is_copy:1;
156 unsigned int is_rename:1;
157 struct fragment *fragments;
158 char *result;
159 size_t resultsize;
160 char old_sha1_prefix[41];
161 char new_sha1_prefix[41];
162 struct patch *next;
166 * A line in a file, len-bytes long (includes the terminating LF,
167 * except for an incomplete line at the end if the file ends with
168 * one), and its contents hashes to 'hash'.
170 struct line {
171 size_t len;
172 unsigned hash : 24;
173 unsigned flag : 8;
174 #define LINE_COMMON 1
178 * This represents a "file", which is an array of "lines".
180 struct image {
181 char *buf;
182 size_t len;
183 size_t nr;
184 size_t alloc;
185 struct line *line_allocated;
186 struct line *line;
190 * Records filenames that have been touched, in order to handle
191 * the case where more than one patches touch the same file.
194 static struct path_list fn_table;
196 static uint32_t hash_line(const char *cp, size_t len)
198 size_t i;
199 uint32_t h;
200 for (i = 0, h = 0; i < len; i++) {
201 if (!isspace(cp[i])) {
202 h = h * 3 + (cp[i] & 0xff);
205 return h;
208 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
210 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
211 img->line_allocated[img->nr].len = len;
212 img->line_allocated[img->nr].hash = hash_line(bol, len);
213 img->line_allocated[img->nr].flag = flag;
214 img->nr++;
217 static void prepare_image(struct image *image, char *buf, size_t len,
218 int prepare_linetable)
220 const char *cp, *ep;
222 memset(image, 0, sizeof(*image));
223 image->buf = buf;
224 image->len = len;
226 if (!prepare_linetable)
227 return;
229 ep = image->buf + image->len;
230 cp = image->buf;
231 while (cp < ep) {
232 const char *next;
233 for (next = cp; next < ep && *next != '\n'; next++)
235 if (next < ep)
236 next++;
237 add_line_info(image, cp, next - cp, 0);
238 cp = next;
240 image->line = image->line_allocated;
243 static void clear_image(struct image *image)
245 free(image->buf);
246 image->buf = NULL;
247 image->len = 0;
250 static void say_patch_name(FILE *output, const char *pre,
251 struct patch *patch, const char *post)
253 fputs(pre, output);
254 if (patch->old_name && patch->new_name &&
255 strcmp(patch->old_name, patch->new_name)) {
256 quote_c_style(patch->old_name, NULL, output, 0);
257 fputs(" => ", output);
258 quote_c_style(patch->new_name, NULL, output, 0);
259 } else {
260 const char *n = patch->new_name;
261 if (!n)
262 n = patch->old_name;
263 quote_c_style(n, NULL, output, 0);
265 fputs(post, output);
268 #define CHUNKSIZE (8192)
269 #define SLOP (16)
271 static void read_patch_file(struct strbuf *sb, int fd)
273 if (strbuf_read(sb, fd, 0) < 0)
274 die("git-apply: read returned %s", strerror(errno));
277 * Make sure that we have some slop in the buffer
278 * so that we can do speculative "memcmp" etc, and
279 * see to it that it is NUL-filled.
281 strbuf_grow(sb, SLOP);
282 memset(sb->buf + sb->len, 0, SLOP);
285 static unsigned long linelen(const char *buffer, unsigned long size)
287 unsigned long len = 0;
288 while (size--) {
289 len++;
290 if (*buffer++ == '\n')
291 break;
293 return len;
296 static int is_dev_null(const char *str)
298 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
301 #define TERM_SPACE 1
302 #define TERM_TAB 2
304 static int name_terminate(const char *name, int namelen, int c, int terminate)
306 if (c == ' ' && !(terminate & TERM_SPACE))
307 return 0;
308 if (c == '\t' && !(terminate & TERM_TAB))
309 return 0;
311 return 1;
314 static char *find_name(const char *line, char *def, int p_value, int terminate)
316 int len;
317 const char *start = line;
319 if (*line == '"') {
320 struct strbuf name;
323 * Proposed "new-style" GNU patch/diff format; see
324 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
326 strbuf_init(&name, 0);
327 if (!unquote_c_style(&name, line, NULL)) {
328 char *cp;
330 for (cp = name.buf; p_value; p_value--) {
331 cp = strchr(cp, '/');
332 if (!cp)
333 break;
334 cp++;
336 if (cp) {
337 /* name can later be freed, so we need
338 * to memmove, not just return cp
340 strbuf_remove(&name, 0, cp - name.buf);
341 free(def);
342 return strbuf_detach(&name, NULL);
345 strbuf_release(&name);
348 for (;;) {
349 char c = *line;
351 if (isspace(c)) {
352 if (c == '\n')
353 break;
354 if (name_terminate(start, line-start, c, terminate))
355 break;
357 line++;
358 if (c == '/' && !--p_value)
359 start = line;
361 if (!start)
362 return def;
363 len = line - start;
364 if (!len)
365 return def;
368 * Generally we prefer the shorter name, especially
369 * if the other one is just a variation of that with
370 * something else tacked on to the end (ie "file.orig"
371 * or "file~").
373 if (def) {
374 int deflen = strlen(def);
375 if (deflen < len && !strncmp(start, def, deflen))
376 return def;
377 free(def);
380 return xmemdupz(start, len);
383 static int count_slashes(const char *cp)
385 int cnt = 0;
386 char ch;
388 while ((ch = *cp++))
389 if (ch == '/')
390 cnt++;
391 return cnt;
395 * Given the string after "--- " or "+++ ", guess the appropriate
396 * p_value for the given patch.
398 static int guess_p_value(const char *nameline)
400 char *name, *cp;
401 int val = -1;
403 if (is_dev_null(nameline))
404 return -1;
405 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
406 if (!name)
407 return -1;
408 cp = strchr(name, '/');
409 if (!cp)
410 val = 0;
411 else if (prefix) {
413 * Does it begin with "a/$our-prefix" and such? Then this is
414 * very likely to apply to our directory.
416 if (!strncmp(name, prefix, prefix_length))
417 val = count_slashes(prefix);
418 else {
419 cp++;
420 if (!strncmp(cp, prefix, prefix_length))
421 val = count_slashes(prefix) + 1;
424 free(name);
425 return val;
429 * Get the name etc info from the ---/+++ lines of a traditional patch header
431 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
432 * files, we can happily check the index for a match, but for creating a
433 * new file we should try to match whatever "patch" does. I have no idea.
435 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
437 char *name;
439 first += 4; /* skip "--- " */
440 second += 4; /* skip "+++ " */
441 if (!p_value_known) {
442 int p, q;
443 p = guess_p_value(first);
444 q = guess_p_value(second);
445 if (p < 0) p = q;
446 if (0 <= p && p == q) {
447 p_value = p;
448 p_value_known = 1;
451 if (is_dev_null(first)) {
452 patch->is_new = 1;
453 patch->is_delete = 0;
454 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
455 patch->new_name = name;
456 } else if (is_dev_null(second)) {
457 patch->is_new = 0;
458 patch->is_delete = 1;
459 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
460 patch->old_name = name;
461 } else {
462 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
463 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
464 patch->old_name = patch->new_name = name;
466 if (!name)
467 die("unable to find filename in patch at line %d", linenr);
470 static int gitdiff_hdrend(const char *line, struct patch *patch)
472 return -1;
476 * We're anal about diff header consistency, to make
477 * sure that we don't end up having strange ambiguous
478 * patches floating around.
480 * As a result, gitdiff_{old|new}name() will check
481 * their names against any previous information, just
482 * to make sure..
484 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
486 if (!orig_name && !isnull)
487 return find_name(line, NULL, p_value, TERM_TAB);
489 if (orig_name) {
490 int len;
491 const char *name;
492 char *another;
493 name = orig_name;
494 len = strlen(name);
495 if (isnull)
496 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
497 another = find_name(line, NULL, p_value, TERM_TAB);
498 if (!another || memcmp(another, name, len))
499 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
500 free(another);
501 return orig_name;
503 else {
504 /* expect "/dev/null" */
505 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
506 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
507 return NULL;
511 static int gitdiff_oldname(const char *line, struct patch *patch)
513 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
514 return 0;
517 static int gitdiff_newname(const char *line, struct patch *patch)
519 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
520 return 0;
523 static int gitdiff_oldmode(const char *line, struct patch *patch)
525 patch->old_mode = strtoul(line, NULL, 8);
526 return 0;
529 static int gitdiff_newmode(const char *line, struct patch *patch)
531 patch->new_mode = strtoul(line, NULL, 8);
532 return 0;
535 static int gitdiff_delete(const char *line, struct patch *patch)
537 patch->is_delete = 1;
538 patch->old_name = patch->def_name;
539 return gitdiff_oldmode(line, patch);
542 static int gitdiff_newfile(const char *line, struct patch *patch)
544 patch->is_new = 1;
545 patch->new_name = patch->def_name;
546 return gitdiff_newmode(line, patch);
549 static int gitdiff_copysrc(const char *line, struct patch *patch)
551 patch->is_copy = 1;
552 patch->old_name = find_name(line, NULL, 0, 0);
553 return 0;
556 static int gitdiff_copydst(const char *line, struct patch *patch)
558 patch->is_copy = 1;
559 patch->new_name = find_name(line, NULL, 0, 0);
560 return 0;
563 static int gitdiff_renamesrc(const char *line, struct patch *patch)
565 patch->is_rename = 1;
566 patch->old_name = find_name(line, NULL, 0, 0);
567 return 0;
570 static int gitdiff_renamedst(const char *line, struct patch *patch)
572 patch->is_rename = 1;
573 patch->new_name = find_name(line, NULL, 0, 0);
574 return 0;
577 static int gitdiff_similarity(const char *line, struct patch *patch)
579 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
580 patch->score = 0;
581 return 0;
584 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
586 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
587 patch->score = 0;
588 return 0;
591 static int gitdiff_index(const char *line, struct patch *patch)
594 * index line is N hexadecimal, "..", N hexadecimal,
595 * and optional space with octal mode.
597 const char *ptr, *eol;
598 int len;
600 ptr = strchr(line, '.');
601 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
602 return 0;
603 len = ptr - line;
604 memcpy(patch->old_sha1_prefix, line, len);
605 patch->old_sha1_prefix[len] = 0;
607 line = ptr + 2;
608 ptr = strchr(line, ' ');
609 eol = strchr(line, '\n');
611 if (!ptr || eol < ptr)
612 ptr = eol;
613 len = ptr - line;
615 if (40 < len)
616 return 0;
617 memcpy(patch->new_sha1_prefix, line, len);
618 patch->new_sha1_prefix[len] = 0;
619 if (*ptr == ' ')
620 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
621 return 0;
625 * This is normal for a diff that doesn't change anything: we'll fall through
626 * into the next diff. Tell the parser to break out.
628 static int gitdiff_unrecognized(const char *line, struct patch *patch)
630 return -1;
633 static const char *stop_at_slash(const char *line, int llen)
635 int i;
637 for (i = 0; i < llen; i++) {
638 int ch = line[i];
639 if (ch == '/')
640 return line + i;
642 return NULL;
646 * This is to extract the same name that appears on "diff --git"
647 * line. We do not find and return anything if it is a rename
648 * patch, and it is OK because we will find the name elsewhere.
649 * We need to reliably find name only when it is mode-change only,
650 * creation or deletion of an empty file. In any of these cases,
651 * both sides are the same name under a/ and b/ respectively.
653 static char *git_header_name(char *line, int llen)
655 const char *name;
656 const char *second = NULL;
657 size_t len;
659 line += strlen("diff --git ");
660 llen -= strlen("diff --git ");
662 if (*line == '"') {
663 const char *cp;
664 struct strbuf first;
665 struct strbuf sp;
667 strbuf_init(&first, 0);
668 strbuf_init(&sp, 0);
670 if (unquote_c_style(&first, line, &second))
671 goto free_and_fail1;
673 /* advance to the first slash */
674 cp = stop_at_slash(first.buf, first.len);
675 /* we do not accept absolute paths */
676 if (!cp || cp == first.buf)
677 goto free_and_fail1;
678 strbuf_remove(&first, 0, cp + 1 - first.buf);
681 * second points at one past closing dq of name.
682 * find the second name.
684 while ((second < line + llen) && isspace(*second))
685 second++;
687 if (line + llen <= second)
688 goto free_and_fail1;
689 if (*second == '"') {
690 if (unquote_c_style(&sp, second, NULL))
691 goto free_and_fail1;
692 cp = stop_at_slash(sp.buf, sp.len);
693 if (!cp || cp == sp.buf)
694 goto free_and_fail1;
695 /* They must match, otherwise ignore */
696 if (strcmp(cp + 1, first.buf))
697 goto free_and_fail1;
698 strbuf_release(&sp);
699 return strbuf_detach(&first, NULL);
702 /* unquoted second */
703 cp = stop_at_slash(second, line + llen - second);
704 if (!cp || cp == second)
705 goto free_and_fail1;
706 cp++;
707 if (line + llen - cp != first.len + 1 ||
708 memcmp(first.buf, cp, first.len))
709 goto free_and_fail1;
710 return strbuf_detach(&first, NULL);
712 free_and_fail1:
713 strbuf_release(&first);
714 strbuf_release(&sp);
715 return NULL;
718 /* unquoted first name */
719 name = stop_at_slash(line, llen);
720 if (!name || name == line)
721 return NULL;
722 name++;
725 * since the first name is unquoted, a dq if exists must be
726 * the beginning of the second name.
728 for (second = name; second < line + llen; second++) {
729 if (*second == '"') {
730 struct strbuf sp;
731 const char *np;
733 strbuf_init(&sp, 0);
734 if (unquote_c_style(&sp, second, NULL))
735 goto free_and_fail2;
737 np = stop_at_slash(sp.buf, sp.len);
738 if (!np || np == sp.buf)
739 goto free_and_fail2;
740 np++;
742 len = sp.buf + sp.len - np;
743 if (len < second - name &&
744 !strncmp(np, name, len) &&
745 isspace(name[len])) {
746 /* Good */
747 strbuf_remove(&sp, 0, np - sp.buf);
748 return strbuf_detach(&sp, NULL);
751 free_and_fail2:
752 strbuf_release(&sp);
753 return NULL;
758 * Accept a name only if it shows up twice, exactly the same
759 * form.
761 for (len = 0 ; ; len++) {
762 switch (name[len]) {
763 default:
764 continue;
765 case '\n':
766 return NULL;
767 case '\t': case ' ':
768 second = name+len;
769 for (;;) {
770 char c = *second++;
771 if (c == '\n')
772 return NULL;
773 if (c == '/')
774 break;
776 if (second[len] == '\n' && !memcmp(name, second, len)) {
777 return xmemdupz(name, len);
783 /* Verify that we recognize the lines following a git header */
784 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
786 unsigned long offset;
788 /* A git diff has explicit new/delete information, so we don't guess */
789 patch->is_new = 0;
790 patch->is_delete = 0;
793 * Some things may not have the old name in the
794 * rest of the headers anywhere (pure mode changes,
795 * or removing or adding empty files), so we get
796 * the default name from the header.
798 patch->def_name = git_header_name(line, len);
800 line += len;
801 size -= len;
802 linenr++;
803 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
804 static const struct opentry {
805 const char *str;
806 int (*fn)(const char *, struct patch *);
807 } optable[] = {
808 { "@@ -", gitdiff_hdrend },
809 { "--- ", gitdiff_oldname },
810 { "+++ ", gitdiff_newname },
811 { "old mode ", gitdiff_oldmode },
812 { "new mode ", gitdiff_newmode },
813 { "deleted file mode ", gitdiff_delete },
814 { "new file mode ", gitdiff_newfile },
815 { "copy from ", gitdiff_copysrc },
816 { "copy to ", gitdiff_copydst },
817 { "rename old ", gitdiff_renamesrc },
818 { "rename new ", gitdiff_renamedst },
819 { "rename from ", gitdiff_renamesrc },
820 { "rename to ", gitdiff_renamedst },
821 { "similarity index ", gitdiff_similarity },
822 { "dissimilarity index ", gitdiff_dissimilarity },
823 { "index ", gitdiff_index },
824 { "", gitdiff_unrecognized },
826 int i;
828 len = linelen(line, size);
829 if (!len || line[len-1] != '\n')
830 break;
831 for (i = 0; i < ARRAY_SIZE(optable); i++) {
832 const struct opentry *p = optable + i;
833 int oplen = strlen(p->str);
834 if (len < oplen || memcmp(p->str, line, oplen))
835 continue;
836 if (p->fn(line + oplen, patch) < 0)
837 return offset;
838 break;
842 return offset;
845 static int parse_num(const char *line, unsigned long *p)
847 char *ptr;
849 if (!isdigit(*line))
850 return 0;
851 *p = strtoul(line, &ptr, 10);
852 return ptr - line;
855 static int parse_range(const char *line, int len, int offset, const char *expect,
856 unsigned long *p1, unsigned long *p2)
858 int digits, ex;
860 if (offset < 0 || offset >= len)
861 return -1;
862 line += offset;
863 len -= offset;
865 digits = parse_num(line, p1);
866 if (!digits)
867 return -1;
869 offset += digits;
870 line += digits;
871 len -= digits;
873 *p2 = 1;
874 if (*line == ',') {
875 digits = parse_num(line+1, p2);
876 if (!digits)
877 return -1;
879 offset += digits+1;
880 line += digits+1;
881 len -= digits+1;
884 ex = strlen(expect);
885 if (ex > len)
886 return -1;
887 if (memcmp(line, expect, ex))
888 return -1;
890 return offset + ex;
894 * Parse a unified diff fragment header of the
895 * form "@@ -a,b +c,d @@"
897 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
899 int offset;
901 if (!len || line[len-1] != '\n')
902 return -1;
904 /* Figure out the number of lines in a fragment */
905 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
906 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
908 return offset;
911 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
913 unsigned long offset, len;
915 patch->is_toplevel_relative = 0;
916 patch->is_rename = patch->is_copy = 0;
917 patch->is_new = patch->is_delete = -1;
918 patch->old_mode = patch->new_mode = 0;
919 patch->old_name = patch->new_name = NULL;
920 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
921 unsigned long nextlen;
923 len = linelen(line, size);
924 if (!len)
925 break;
927 /* Testing this early allows us to take a few shortcuts.. */
928 if (len < 6)
929 continue;
932 * Make sure we don't find any unconnected patch fragments.
933 * That's a sign that we didn't find a header, and that a
934 * patch has become corrupted/broken up.
936 if (!memcmp("@@ -", line, 4)) {
937 struct fragment dummy;
938 if (parse_fragment_header(line, len, &dummy) < 0)
939 continue;
940 die("patch fragment without header at line %d: %.*s",
941 linenr, (int)len-1, line);
944 if (size < len + 6)
945 break;
948 * Git patch? It might not have a real patch, just a rename
949 * or mode change, so we handle that specially
951 if (!memcmp("diff --git ", line, 11)) {
952 int git_hdr_len = parse_git_header(line, len, size, patch);
953 if (git_hdr_len <= len)
954 continue;
955 if (!patch->old_name && !patch->new_name) {
956 if (!patch->def_name)
957 die("git diff header lacks filename information (line %d)", linenr);
958 patch->old_name = patch->new_name = patch->def_name;
960 patch->is_toplevel_relative = 1;
961 *hdrsize = git_hdr_len;
962 return offset;
965 /* --- followed by +++ ? */
966 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
967 continue;
970 * We only accept unified patches, so we want it to
971 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
972 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
974 nextlen = linelen(line + len, size - len);
975 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
976 continue;
978 /* Ok, we'll consider it a patch */
979 parse_traditional_patch(line, line+len, patch);
980 *hdrsize = len + nextlen;
981 linenr += 2;
982 return offset;
984 return -1;
987 static void check_whitespace(const char *line, int len, unsigned ws_rule)
989 char *err;
990 unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
991 NULL, NULL, NULL, NULL);
992 if (!result)
993 return;
995 whitespace_error++;
996 if (squelch_whitespace_errors &&
997 squelch_whitespace_errors < whitespace_error)
999 else {
1000 err = whitespace_error_string(result);
1001 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1002 patch_input_file, linenr, err, len - 2, line + 1);
1003 free(err);
1008 * Parse a unified diff. Note that this really needs to parse each
1009 * fragment separately, since the only way to know the difference
1010 * between a "---" that is part of a patch, and a "---" that starts
1011 * the next patch is to look at the line counts..
1013 static int parse_fragment(char *line, unsigned long size,
1014 struct patch *patch, struct fragment *fragment)
1016 int added, deleted;
1017 int len = linelen(line, size), offset;
1018 unsigned long oldlines, newlines;
1019 unsigned long leading, trailing;
1021 offset = parse_fragment_header(line, len, fragment);
1022 if (offset < 0)
1023 return -1;
1024 oldlines = fragment->oldlines;
1025 newlines = fragment->newlines;
1026 leading = 0;
1027 trailing = 0;
1029 /* Parse the thing.. */
1030 line += len;
1031 size -= len;
1032 linenr++;
1033 added = deleted = 0;
1034 for (offset = len;
1035 0 < size;
1036 offset += len, size -= len, line += len, linenr++) {
1037 if (!oldlines && !newlines)
1038 break;
1039 len = linelen(line, size);
1040 if (!len || line[len-1] != '\n')
1041 return -1;
1042 switch (*line) {
1043 default:
1044 return -1;
1045 case '\n': /* newer GNU diff, an empty context line */
1046 case ' ':
1047 oldlines--;
1048 newlines--;
1049 if (!deleted && !added)
1050 leading++;
1051 trailing++;
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 deleted++;
1058 oldlines--;
1059 trailing = 0;
1060 break;
1061 case '+':
1062 if (!apply_in_reverse &&
1063 ws_error_action != nowarn_ws_error)
1064 check_whitespace(line, len, patch->ws_rule);
1065 added++;
1066 newlines--;
1067 trailing = 0;
1068 break;
1071 * We allow "\ No newline at end of file". Depending
1072 * on locale settings when the patch was produced we
1073 * don't know what this line looks like. The only
1074 * thing we do know is that it begins with "\ ".
1075 * Checking for 12 is just for sanity check -- any
1076 * l10n of "\ No newline..." is at least that long.
1078 case '\\':
1079 if (len < 12 || memcmp(line, "\\ ", 2))
1080 return -1;
1081 break;
1084 if (oldlines || newlines)
1085 return -1;
1086 fragment->leading = leading;
1087 fragment->trailing = trailing;
1090 * If a fragment ends with an incomplete line, we failed to include
1091 * it in the above loop because we hit oldlines == newlines == 0
1092 * before seeing it.
1094 if (12 < size && !memcmp(line, "\\ ", 2))
1095 offset += linelen(line, size);
1097 patch->lines_added += added;
1098 patch->lines_deleted += deleted;
1100 if (0 < patch->is_new && oldlines)
1101 return error("new file depends on old contents");
1102 if (0 < patch->is_delete && newlines)
1103 return error("deleted file still has contents");
1104 return offset;
1107 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1109 unsigned long offset = 0;
1110 unsigned long oldlines = 0, newlines = 0, context = 0;
1111 struct fragment **fragp = &patch->fragments;
1113 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1114 struct fragment *fragment;
1115 int len;
1117 fragment = xcalloc(1, sizeof(*fragment));
1118 len = parse_fragment(line, size, patch, fragment);
1119 if (len <= 0)
1120 die("corrupt patch at line %d", linenr);
1121 fragment->patch = line;
1122 fragment->size = len;
1123 oldlines += fragment->oldlines;
1124 newlines += fragment->newlines;
1125 context += fragment->leading + fragment->trailing;
1127 *fragp = fragment;
1128 fragp = &fragment->next;
1130 offset += len;
1131 line += len;
1132 size -= len;
1136 * If something was removed (i.e. we have old-lines) it cannot
1137 * be creation, and if something was added it cannot be
1138 * deletion. However, the reverse is not true; --unified=0
1139 * patches that only add are not necessarily creation even
1140 * though they do not have any old lines, and ones that only
1141 * delete are not necessarily deletion.
1143 * Unfortunately, a real creation/deletion patch do _not_ have
1144 * any context line by definition, so we cannot safely tell it
1145 * apart with --unified=0 insanity. At least if the patch has
1146 * more than one hunk it is not creation or deletion.
1148 if (patch->is_new < 0 &&
1149 (oldlines || (patch->fragments && patch->fragments->next)))
1150 patch->is_new = 0;
1151 if (patch->is_delete < 0 &&
1152 (newlines || (patch->fragments && patch->fragments->next)))
1153 patch->is_delete = 0;
1155 if (0 < patch->is_new && oldlines)
1156 die("new file %s depends on old contents", patch->new_name);
1157 if (0 < patch->is_delete && newlines)
1158 die("deleted file %s still has contents", patch->old_name);
1159 if (!patch->is_delete && !newlines && context)
1160 fprintf(stderr, "** warning: file %s becomes empty but "
1161 "is not deleted\n", patch->new_name);
1163 return offset;
1166 static inline int metadata_changes(struct patch *patch)
1168 return patch->is_rename > 0 ||
1169 patch->is_copy > 0 ||
1170 patch->is_new > 0 ||
1171 patch->is_delete ||
1172 (patch->old_mode && patch->new_mode &&
1173 patch->old_mode != patch->new_mode);
1176 static char *inflate_it(const void *data, unsigned long size,
1177 unsigned long inflated_size)
1179 z_stream stream;
1180 void *out;
1181 int st;
1183 memset(&stream, 0, sizeof(stream));
1185 stream.next_in = (unsigned char *)data;
1186 stream.avail_in = size;
1187 stream.next_out = out = xmalloc(inflated_size);
1188 stream.avail_out = inflated_size;
1189 inflateInit(&stream);
1190 st = inflate(&stream, Z_FINISH);
1191 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1192 free(out);
1193 return NULL;
1195 return out;
1198 static struct fragment *parse_binary_hunk(char **buf_p,
1199 unsigned long *sz_p,
1200 int *status_p,
1201 int *used_p)
1204 * Expect a line that begins with binary patch method ("literal"
1205 * or "delta"), followed by the length of data before deflating.
1206 * a sequence of 'length-byte' followed by base-85 encoded data
1207 * should follow, terminated by a newline.
1209 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1210 * and we would limit the patch line to 66 characters,
1211 * so one line can fit up to 13 groups that would decode
1212 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1213 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1215 int llen, used;
1216 unsigned long size = *sz_p;
1217 char *buffer = *buf_p;
1218 int patch_method;
1219 unsigned long origlen;
1220 char *data = NULL;
1221 int hunk_size = 0;
1222 struct fragment *frag;
1224 llen = linelen(buffer, size);
1225 used = llen;
1227 *status_p = 0;
1229 if (!prefixcmp(buffer, "delta ")) {
1230 patch_method = BINARY_DELTA_DEFLATED;
1231 origlen = strtoul(buffer + 6, NULL, 10);
1233 else if (!prefixcmp(buffer, "literal ")) {
1234 patch_method = BINARY_LITERAL_DEFLATED;
1235 origlen = strtoul(buffer + 8, NULL, 10);
1237 else
1238 return NULL;
1240 linenr++;
1241 buffer += llen;
1242 while (1) {
1243 int byte_length, max_byte_length, newsize;
1244 llen = linelen(buffer, size);
1245 used += llen;
1246 linenr++;
1247 if (llen == 1) {
1248 /* consume the blank line */
1249 buffer++;
1250 size--;
1251 break;
1254 * Minimum line is "A00000\n" which is 7-byte long,
1255 * and the line length must be multiple of 5 plus 2.
1257 if ((llen < 7) || (llen-2) % 5)
1258 goto corrupt;
1259 max_byte_length = (llen - 2) / 5 * 4;
1260 byte_length = *buffer;
1261 if ('A' <= byte_length && byte_length <= 'Z')
1262 byte_length = byte_length - 'A' + 1;
1263 else if ('a' <= byte_length && byte_length <= 'z')
1264 byte_length = byte_length - 'a' + 27;
1265 else
1266 goto corrupt;
1267 /* if the input length was not multiple of 4, we would
1268 * have filler at the end but the filler should never
1269 * exceed 3 bytes
1271 if (max_byte_length < byte_length ||
1272 byte_length <= max_byte_length - 4)
1273 goto corrupt;
1274 newsize = hunk_size + byte_length;
1275 data = xrealloc(data, newsize);
1276 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1277 goto corrupt;
1278 hunk_size = newsize;
1279 buffer += llen;
1280 size -= llen;
1283 frag = xcalloc(1, sizeof(*frag));
1284 frag->patch = inflate_it(data, hunk_size, origlen);
1285 if (!frag->patch)
1286 goto corrupt;
1287 free(data);
1288 frag->size = origlen;
1289 *buf_p = buffer;
1290 *sz_p = size;
1291 *used_p = used;
1292 frag->binary_patch_method = patch_method;
1293 return frag;
1295 corrupt:
1296 free(data);
1297 *status_p = -1;
1298 error("corrupt binary patch at line %d: %.*s",
1299 linenr-1, llen-1, buffer);
1300 return NULL;
1303 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1306 * We have read "GIT binary patch\n"; what follows is a line
1307 * that says the patch method (currently, either "literal" or
1308 * "delta") and the length of data before deflating; a
1309 * sequence of 'length-byte' followed by base-85 encoded data
1310 * follows.
1312 * When a binary patch is reversible, there is another binary
1313 * hunk in the same format, starting with patch method (either
1314 * "literal" or "delta") with the length of data, and a sequence
1315 * of length-byte + base-85 encoded data, terminated with another
1316 * empty line. This data, when applied to the postimage, produces
1317 * the preimage.
1319 struct fragment *forward;
1320 struct fragment *reverse;
1321 int status;
1322 int used, used_1;
1324 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1325 if (!forward && !status)
1326 /* there has to be one hunk (forward hunk) */
1327 return error("unrecognized binary patch at line %d", linenr-1);
1328 if (status)
1329 /* otherwise we already gave an error message */
1330 return status;
1332 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1333 if (reverse)
1334 used += used_1;
1335 else if (status) {
1337 * Not having reverse hunk is not an error, but having
1338 * a corrupt reverse hunk is.
1340 free((void*) forward->patch);
1341 free(forward);
1342 return status;
1344 forward->next = reverse;
1345 patch->fragments = forward;
1346 patch->is_binary = 1;
1347 return used;
1350 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1352 int hdrsize, patchsize;
1353 int offset = find_header(buffer, size, &hdrsize, patch);
1355 if (offset < 0)
1356 return offset;
1358 patch->ws_rule = whitespace_rule(patch->new_name
1359 ? patch->new_name
1360 : patch->old_name);
1362 patchsize = parse_single_patch(buffer + offset + hdrsize,
1363 size - offset - hdrsize, patch);
1365 if (!patchsize) {
1366 static const char *binhdr[] = {
1367 "Binary files ",
1368 "Files ",
1369 NULL,
1371 static const char git_binary[] = "GIT binary patch\n";
1372 int i;
1373 int hd = hdrsize + offset;
1374 unsigned long llen = linelen(buffer + hd, size - hd);
1376 if (llen == sizeof(git_binary) - 1 &&
1377 !memcmp(git_binary, buffer + hd, llen)) {
1378 int used;
1379 linenr++;
1380 used = parse_binary(buffer + hd + llen,
1381 size - hd - llen, patch);
1382 if (used)
1383 patchsize = used + llen;
1384 else
1385 patchsize = 0;
1387 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1388 for (i = 0; binhdr[i]; i++) {
1389 int len = strlen(binhdr[i]);
1390 if (len < size - hd &&
1391 !memcmp(binhdr[i], buffer + hd, len)) {
1392 linenr++;
1393 patch->is_binary = 1;
1394 patchsize = llen;
1395 break;
1400 /* Empty patch cannot be applied if it is a text patch
1401 * without metadata change. A binary patch appears
1402 * empty to us here.
1404 if ((apply || check) &&
1405 (!patch->is_binary && !metadata_changes(patch)))
1406 die("patch with only garbage at line %d", linenr);
1409 return offset + hdrsize + patchsize;
1412 #define swap(a,b) myswap((a),(b),sizeof(a))
1414 #define myswap(a, b, size) do { \
1415 unsigned char mytmp[size]; \
1416 memcpy(mytmp, &a, size); \
1417 memcpy(&a, &b, size); \
1418 memcpy(&b, mytmp, size); \
1419 } while (0)
1421 static void reverse_patches(struct patch *p)
1423 for (; p; p = p->next) {
1424 struct fragment *frag = p->fragments;
1426 swap(p->new_name, p->old_name);
1427 swap(p->new_mode, p->old_mode);
1428 swap(p->is_new, p->is_delete);
1429 swap(p->lines_added, p->lines_deleted);
1430 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1432 for (; frag; frag = frag->next) {
1433 swap(frag->newpos, frag->oldpos);
1434 swap(frag->newlines, frag->oldlines);
1439 static const char pluses[] =
1440 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1441 static const char minuses[]=
1442 "----------------------------------------------------------------------";
1444 static void show_stats(struct patch *patch)
1446 struct strbuf qname;
1447 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1448 int max, add, del;
1450 strbuf_init(&qname, 0);
1451 quote_c_style(cp, &qname, NULL, 0);
1454 * "scale" the filename
1456 max = max_len;
1457 if (max > 50)
1458 max = 50;
1460 if (qname.len > max) {
1461 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1462 if (!cp)
1463 cp = qname.buf + qname.len + 3 - max;
1464 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1467 if (patch->is_binary) {
1468 printf(" %-*s | Bin\n", max, qname.buf);
1469 strbuf_release(&qname);
1470 return;
1473 printf(" %-*s |", max, qname.buf);
1474 strbuf_release(&qname);
1477 * scale the add/delete
1479 max = max + max_change > 70 ? 70 - max : max_change;
1480 add = patch->lines_added;
1481 del = patch->lines_deleted;
1483 if (max_change > 0) {
1484 int total = ((add + del) * max + max_change / 2) / max_change;
1485 add = (add * max + max_change / 2) / max_change;
1486 del = total - add;
1488 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1489 add, pluses, del, minuses);
1492 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1494 switch (st->st_mode & S_IFMT) {
1495 case S_IFLNK:
1496 strbuf_grow(buf, st->st_size);
1497 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1498 return -1;
1499 strbuf_setlen(buf, st->st_size);
1500 return 0;
1501 case S_IFREG:
1502 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1503 return error("unable to open or read %s", path);
1504 convert_to_git(path, buf->buf, buf->len, buf, 0);
1505 return 0;
1506 default:
1507 return -1;
1511 static void update_pre_post_images(struct image *preimage,
1512 struct image *postimage,
1513 char *buf,
1514 size_t len)
1516 int i, ctx;
1517 char *new, *old, *fixed;
1518 struct image fixed_preimage;
1521 * Update the preimage with whitespace fixes. Note that we
1522 * are not losing preimage->buf -- apply_one_fragment() will
1523 * free "oldlines".
1525 prepare_image(&fixed_preimage, buf, len, 1);
1526 assert(fixed_preimage.nr == preimage->nr);
1527 for (i = 0; i < preimage->nr; i++)
1528 fixed_preimage.line[i].flag = preimage->line[i].flag;
1529 free(preimage->line_allocated);
1530 *preimage = fixed_preimage;
1533 * Adjust the common context lines in postimage, in place.
1534 * This is possible because whitespace fixing does not make
1535 * the string grow.
1537 new = old = postimage->buf;
1538 fixed = preimage->buf;
1539 for (i = ctx = 0; i < postimage->nr; i++) {
1540 size_t len = postimage->line[i].len;
1541 if (!(postimage->line[i].flag & LINE_COMMON)) {
1542 /* an added line -- no counterparts in preimage */
1543 memmove(new, old, len);
1544 old += len;
1545 new += len;
1546 continue;
1549 /* a common context -- skip it in the original postimage */
1550 old += len;
1552 /* and find the corresponding one in the fixed preimage */
1553 while (ctx < preimage->nr &&
1554 !(preimage->line[ctx].flag & LINE_COMMON)) {
1555 fixed += preimage->line[ctx].len;
1556 ctx++;
1558 if (preimage->nr <= ctx)
1559 die("oops");
1561 /* and copy it in, while fixing the line length */
1562 len = preimage->line[ctx].len;
1563 memcpy(new, fixed, len);
1564 new += len;
1565 fixed += len;
1566 postimage->line[i].len = len;
1567 ctx++;
1570 /* Fix the length of the whole thing */
1571 postimage->len = new - postimage->buf;
1574 static int match_fragment(struct image *img,
1575 struct image *preimage,
1576 struct image *postimage,
1577 unsigned long try,
1578 int try_lno,
1579 unsigned ws_rule,
1580 int match_beginning, int match_end)
1582 int i;
1583 char *fixed_buf, *buf, *orig, *target;
1585 if (preimage->nr + try_lno > img->nr)
1586 return 0;
1588 if (match_beginning && try_lno)
1589 return 0;
1591 if (match_end && preimage->nr + try_lno != img->nr)
1592 return 0;
1594 /* Quick hash check */
1595 for (i = 0; i < preimage->nr; i++)
1596 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1597 return 0;
1600 * Do we have an exact match? If we were told to match
1601 * at the end, size must be exactly at try+fragsize,
1602 * otherwise try+fragsize must be still within the preimage,
1603 * and either case, the old piece should match the preimage
1604 * exactly.
1606 if ((match_end
1607 ? (try + preimage->len == img->len)
1608 : (try + preimage->len <= img->len)) &&
1609 !memcmp(img->buf + try, preimage->buf, preimage->len))
1610 return 1;
1612 if (ws_error_action != correct_ws_error)
1613 return 0;
1616 * The hunk does not apply byte-by-byte, but the hash says
1617 * it might with whitespace fuzz.
1619 fixed_buf = xmalloc(preimage->len + 1);
1620 buf = fixed_buf;
1621 orig = preimage->buf;
1622 target = img->buf + try;
1623 for (i = 0; i < preimage->nr; i++) {
1624 size_t fixlen; /* length after fixing the preimage */
1625 size_t oldlen = preimage->line[i].len;
1626 size_t tgtlen = img->line[try_lno + i].len;
1627 size_t tgtfixlen; /* length after fixing the target line */
1628 char tgtfixbuf[1024], *tgtfix;
1629 int match;
1631 /* Try fixing the line in the preimage */
1632 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1634 /* Try fixing the line in the target */
1635 if (sizeof(tgtfixbuf) < tgtlen)
1636 tgtfix = tgtfixbuf;
1637 else
1638 tgtfix = xmalloc(tgtlen);
1639 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1642 * If they match, either the preimage was based on
1643 * a version before our tree fixed whitespace breakage,
1644 * or we are lacking a whitespace-fix patch the tree
1645 * the preimage was based on already had (i.e. target
1646 * has whitespace breakage, the preimage doesn't).
1647 * In either case, we are fixing the whitespace breakages
1648 * so we might as well take the fix together with their
1649 * real change.
1651 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1653 if (tgtfix != tgtfixbuf)
1654 free(tgtfix);
1655 if (!match)
1656 goto unmatch_exit;
1658 orig += oldlen;
1659 buf += fixlen;
1660 target += tgtlen;
1664 * Yes, the preimage is based on an older version that still
1665 * has whitespace breakages unfixed, and fixing them makes the
1666 * hunk match. Update the context lines in the postimage.
1668 update_pre_post_images(preimage, postimage,
1669 fixed_buf, buf - fixed_buf);
1670 return 1;
1672 unmatch_exit:
1673 free(fixed_buf);
1674 return 0;
1677 static int find_pos(struct image *img,
1678 struct image *preimage,
1679 struct image *postimage,
1680 int line,
1681 unsigned ws_rule,
1682 int match_beginning, int match_end)
1684 int i;
1685 unsigned long backwards, forwards, try;
1686 int backwards_lno, forwards_lno, try_lno;
1688 if (preimage->nr > img->nr)
1689 return -1;
1692 * If match_begining or match_end is specified, there is no
1693 * point starting from a wrong line that will never match and
1694 * wander around and wait for a match at the specified end.
1696 if (match_beginning)
1697 line = 0;
1698 else if (match_end)
1699 line = img->nr - preimage->nr;
1701 if (line > img->nr)
1702 line = img->nr;
1704 try = 0;
1705 for (i = 0; i < line; i++)
1706 try += img->line[i].len;
1709 * There's probably some smart way to do this, but I'll leave
1710 * that to the smart and beautiful people. I'm simple and stupid.
1712 backwards = try;
1713 backwards_lno = line;
1714 forwards = try;
1715 forwards_lno = line;
1716 try_lno = line;
1718 for (i = 0; ; i++) {
1719 if (match_fragment(img, preimage, postimage,
1720 try, try_lno, ws_rule,
1721 match_beginning, match_end))
1722 return try_lno;
1724 again:
1725 if (backwards_lno == 0 && forwards_lno == img->nr)
1726 break;
1728 if (i & 1) {
1729 if (backwards_lno == 0) {
1730 i++;
1731 goto again;
1733 backwards_lno--;
1734 backwards -= img->line[backwards_lno].len;
1735 try = backwards;
1736 try_lno = backwards_lno;
1737 } else {
1738 if (forwards_lno == img->nr) {
1739 i++;
1740 goto again;
1742 forwards += img->line[forwards_lno].len;
1743 forwards_lno++;
1744 try = forwards;
1745 try_lno = forwards_lno;
1749 return -1;
1752 static void remove_first_line(struct image *img)
1754 img->buf += img->line[0].len;
1755 img->len -= img->line[0].len;
1756 img->line++;
1757 img->nr--;
1760 static void remove_last_line(struct image *img)
1762 img->len -= img->line[--img->nr].len;
1765 static void update_image(struct image *img,
1766 int applied_pos,
1767 struct image *preimage,
1768 struct image *postimage)
1771 * remove the copy of preimage at offset in img
1772 * and replace it with postimage
1774 int i, nr;
1775 size_t remove_count, insert_count, applied_at = 0;
1776 char *result;
1778 for (i = 0; i < applied_pos; i++)
1779 applied_at += img->line[i].len;
1781 remove_count = 0;
1782 for (i = 0; i < preimage->nr; i++)
1783 remove_count += img->line[applied_pos + i].len;
1784 insert_count = postimage->len;
1786 /* Adjust the contents */
1787 result = xmalloc(img->len + insert_count - remove_count + 1);
1788 memcpy(result, img->buf, applied_at);
1789 memcpy(result + applied_at, postimage->buf, postimage->len);
1790 memcpy(result + applied_at + postimage->len,
1791 img->buf + (applied_at + remove_count),
1792 img->len - (applied_at + remove_count));
1793 free(img->buf);
1794 img->buf = result;
1795 img->len += insert_count - remove_count;
1796 result[img->len] = '\0';
1798 /* Adjust the line table */
1799 nr = img->nr + postimage->nr - preimage->nr;
1800 if (preimage->nr < postimage->nr) {
1802 * NOTE: this knows that we never call remove_first_line()
1803 * on anything other than pre/post image.
1805 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1806 img->line_allocated = img->line;
1808 if (preimage->nr != postimage->nr)
1809 memmove(img->line + applied_pos + postimage->nr,
1810 img->line + applied_pos + preimage->nr,
1811 (img->nr - (applied_pos + preimage->nr)) *
1812 sizeof(*img->line));
1813 memcpy(img->line + applied_pos,
1814 postimage->line,
1815 postimage->nr * sizeof(*img->line));
1816 img->nr = nr;
1819 static int apply_one_fragment(struct image *img, struct fragment *frag,
1820 int inaccurate_eof, unsigned ws_rule)
1822 int match_beginning, match_end;
1823 const char *patch = frag->patch;
1824 int size = frag->size;
1825 char *old, *new, *oldlines, *newlines;
1826 int new_blank_lines_at_end = 0;
1827 unsigned long leading, trailing;
1828 int pos, applied_pos;
1829 struct image preimage;
1830 struct image postimage;
1832 memset(&preimage, 0, sizeof(preimage));
1833 memset(&postimage, 0, sizeof(postimage));
1834 oldlines = xmalloc(size);
1835 newlines = xmalloc(size);
1837 old = oldlines;
1838 new = newlines;
1839 while (size > 0) {
1840 char first;
1841 int len = linelen(patch, size);
1842 int plen, added;
1843 int added_blank_line = 0;
1845 if (!len)
1846 break;
1849 * "plen" is how much of the line we should use for
1850 * the actual patch data. Normally we just remove the
1851 * first character on the line, but if the line is
1852 * followed by "\ No newline", then we also remove the
1853 * last one (which is the newline, of course).
1855 plen = len - 1;
1856 if (len < size && patch[len] == '\\')
1857 plen--;
1858 first = *patch;
1859 if (apply_in_reverse) {
1860 if (first == '-')
1861 first = '+';
1862 else if (first == '+')
1863 first = '-';
1866 switch (first) {
1867 case '\n':
1868 /* Newer GNU diff, empty context line */
1869 if (plen < 0)
1870 /* ... followed by '\No newline'; nothing */
1871 break;
1872 *old++ = '\n';
1873 *new++ = '\n';
1874 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1875 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1876 break;
1877 case ' ':
1878 case '-':
1879 memcpy(old, patch + 1, plen);
1880 add_line_info(&preimage, old, plen,
1881 (first == ' ' ? LINE_COMMON : 0));
1882 old += plen;
1883 if (first == '-')
1884 break;
1885 /* Fall-through for ' ' */
1886 case '+':
1887 /* --no-add does not add new lines */
1888 if (first == '+' && no_add)
1889 break;
1891 if (first != '+' ||
1892 !whitespace_error ||
1893 ws_error_action != correct_ws_error) {
1894 memcpy(new, patch + 1, plen);
1895 added = plen;
1897 else {
1898 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1900 add_line_info(&postimage, new, added,
1901 (first == '+' ? 0 : LINE_COMMON));
1902 new += added;
1903 if (first == '+' &&
1904 added == 1 && new[-1] == '\n')
1905 added_blank_line = 1;
1906 break;
1907 case '@': case '\\':
1908 /* Ignore it, we already handled it */
1909 break;
1910 default:
1911 if (apply_verbosely)
1912 error("invalid start of line: '%c'", first);
1913 return -1;
1915 if (added_blank_line)
1916 new_blank_lines_at_end++;
1917 else
1918 new_blank_lines_at_end = 0;
1919 patch += len;
1920 size -= len;
1922 if (inaccurate_eof &&
1923 old > oldlines && old[-1] == '\n' &&
1924 new > newlines && new[-1] == '\n') {
1925 old--;
1926 new--;
1929 leading = frag->leading;
1930 trailing = frag->trailing;
1933 * A hunk to change lines at the beginning would begin with
1934 * @@ -1,L +N,M @@
1936 * And a hunk to add to an empty file would begin with
1937 * @@ -0,0 +N,M @@
1939 * In other words, a hunk that is (frag->oldpos <= 1) with or
1940 * without leading context must match at the beginning.
1942 match_beginning = frag->oldpos <= 1;
1945 * A hunk without trailing lines must match at the end.
1946 * However, we simply cannot tell if a hunk must match end
1947 * from the lack of trailing lines if the patch was generated
1948 * with unidiff without any context.
1950 match_end = !unidiff_zero && !trailing;
1952 pos = frag->newpos ? (frag->newpos - 1) : 0;
1953 preimage.buf = oldlines;
1954 preimage.len = old - oldlines;
1955 postimage.buf = newlines;
1956 postimage.len = new - newlines;
1957 preimage.line = preimage.line_allocated;
1958 postimage.line = postimage.line_allocated;
1960 for (;;) {
1962 applied_pos = find_pos(img, &preimage, &postimage, pos,
1963 ws_rule, match_beginning, match_end);
1965 if (applied_pos >= 0)
1966 break;
1968 /* Am I at my context limits? */
1969 if ((leading <= p_context) && (trailing <= p_context))
1970 break;
1971 if (match_beginning || match_end) {
1972 match_beginning = match_end = 0;
1973 continue;
1977 * Reduce the number of context lines; reduce both
1978 * leading and trailing if they are equal otherwise
1979 * just reduce the larger context.
1981 if (leading >= trailing) {
1982 remove_first_line(&preimage);
1983 remove_first_line(&postimage);
1984 pos--;
1985 leading--;
1987 if (trailing > leading) {
1988 remove_last_line(&preimage);
1989 remove_last_line(&postimage);
1990 trailing--;
1994 if (applied_pos >= 0) {
1995 if (ws_error_action == correct_ws_error &&
1996 new_blank_lines_at_end &&
1997 postimage.nr + applied_pos == img->nr) {
1999 * If the patch application adds blank lines
2000 * at the end, and if the patch applies at the
2001 * end of the image, remove those added blank
2002 * lines.
2004 while (new_blank_lines_at_end--)
2005 remove_last_line(&postimage);
2009 * Warn if it was necessary to reduce the number
2010 * of context lines.
2012 if ((leading != frag->leading) ||
2013 (trailing != frag->trailing))
2014 fprintf(stderr, "Context reduced to (%ld/%ld)"
2015 " to apply fragment at %d\n",
2016 leading, trailing, applied_pos+1);
2017 update_image(img, applied_pos, &preimage, &postimage);
2018 } else {
2019 if (apply_verbosely)
2020 error("while searching for:\n%.*s",
2021 (int)(old - oldlines), oldlines);
2024 free(oldlines);
2025 free(newlines);
2026 free(preimage.line_allocated);
2027 free(postimage.line_allocated);
2029 return (applied_pos < 0);
2032 static int apply_binary_fragment(struct image *img, struct patch *patch)
2034 struct fragment *fragment = patch->fragments;
2035 unsigned long len;
2036 void *dst;
2038 /* Binary patch is irreversible without the optional second hunk */
2039 if (apply_in_reverse) {
2040 if (!fragment->next)
2041 return error("cannot reverse-apply a binary patch "
2042 "without the reverse hunk to '%s'",
2043 patch->new_name
2044 ? patch->new_name : patch->old_name);
2045 fragment = fragment->next;
2047 switch (fragment->binary_patch_method) {
2048 case BINARY_DELTA_DEFLATED:
2049 dst = patch_delta(img->buf, img->len, fragment->patch,
2050 fragment->size, &len);
2051 if (!dst)
2052 return -1;
2053 clear_image(img);
2054 img->buf = dst;
2055 img->len = len;
2056 return 0;
2057 case BINARY_LITERAL_DEFLATED:
2058 clear_image(img);
2059 img->len = fragment->size;
2060 img->buf = xmalloc(img->len+1);
2061 memcpy(img->buf, fragment->patch, img->len);
2062 img->buf[img->len] = '\0';
2063 return 0;
2065 return -1;
2068 static int apply_binary(struct image *img, struct patch *patch)
2070 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2071 unsigned char sha1[20];
2074 * For safety, we require patch index line to contain
2075 * full 40-byte textual SHA1 for old and new, at least for now.
2077 if (strlen(patch->old_sha1_prefix) != 40 ||
2078 strlen(patch->new_sha1_prefix) != 40 ||
2079 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2080 get_sha1_hex(patch->new_sha1_prefix, sha1))
2081 return error("cannot apply binary patch to '%s' "
2082 "without full index line", name);
2084 if (patch->old_name) {
2086 * See if the old one matches what the patch
2087 * applies to.
2089 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2090 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2091 return error("the patch applies to '%s' (%s), "
2092 "which does not match the "
2093 "current contents.",
2094 name, sha1_to_hex(sha1));
2096 else {
2097 /* Otherwise, the old one must be empty. */
2098 if (img->len)
2099 return error("the patch applies to an empty "
2100 "'%s' but it is not empty", name);
2103 get_sha1_hex(patch->new_sha1_prefix, sha1);
2104 if (is_null_sha1(sha1)) {
2105 clear_image(img);
2106 return 0; /* deletion patch */
2109 if (has_sha1_file(sha1)) {
2110 /* We already have the postimage */
2111 enum object_type type;
2112 unsigned long size;
2113 char *result;
2115 result = read_sha1_file(sha1, &type, &size);
2116 if (!result)
2117 return error("the necessary postimage %s for "
2118 "'%s' cannot be read",
2119 patch->new_sha1_prefix, name);
2120 clear_image(img);
2121 img->buf = result;
2122 img->len = size;
2123 } else {
2125 * We have verified buf matches the preimage;
2126 * apply the patch data to it, which is stored
2127 * in the patch->fragments->{patch,size}.
2129 if (apply_binary_fragment(img, patch))
2130 return error("binary patch does not apply to '%s'",
2131 name);
2133 /* verify that the result matches */
2134 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2135 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2136 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2137 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2140 return 0;
2143 static int apply_fragments(struct image *img, struct patch *patch)
2145 struct fragment *frag = patch->fragments;
2146 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2147 unsigned ws_rule = patch->ws_rule;
2148 unsigned inaccurate_eof = patch->inaccurate_eof;
2150 if (patch->is_binary)
2151 return apply_binary(img, patch);
2153 while (frag) {
2154 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2155 error("patch failed: %s:%ld", name, frag->oldpos);
2156 if (!apply_with_reject)
2157 return -1;
2158 frag->rejected = 1;
2160 frag = frag->next;
2162 return 0;
2165 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2167 if (!ce)
2168 return 0;
2170 if (S_ISGITLINK(ce->ce_mode)) {
2171 strbuf_grow(buf, 100);
2172 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2173 } else {
2174 enum object_type type;
2175 unsigned long sz;
2176 char *result;
2178 result = read_sha1_file(ce->sha1, &type, &sz);
2179 if (!result)
2180 return -1;
2181 /* XXX read_sha1_file NUL-terminates */
2182 strbuf_attach(buf, result, sz, sz + 1);
2184 return 0;
2187 static struct patch *in_fn_table(const char *name)
2189 struct path_list_item *item;
2191 if (name == NULL)
2192 return NULL;
2194 item = path_list_lookup(name, &fn_table);
2195 if (item != NULL)
2196 return (struct patch *)item->util;
2198 return NULL;
2201 static void add_to_fn_table(struct patch *patch)
2203 struct path_list_item *item;
2206 * Always add new_name unless patch is a deletion
2207 * This should cover the cases for normal diffs,
2208 * file creations and copies
2210 if (patch->new_name != NULL) {
2211 item = path_list_insert(patch->new_name, &fn_table);
2212 item->util = patch;
2216 * store a failure on rename/deletion cases because
2217 * later chunks shouldn't patch old names
2219 if ((patch->new_name == NULL) || (patch->is_rename)) {
2220 item = path_list_insert(patch->old_name, &fn_table);
2221 item->util = (struct patch *) -1;
2225 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2227 struct strbuf buf;
2228 struct image image;
2229 size_t len;
2230 char *img;
2231 struct patch *tpatch;
2233 strbuf_init(&buf, 0);
2235 if ((tpatch = in_fn_table(patch->old_name)) != NULL) {
2236 if (tpatch == (struct patch *) -1) {
2237 return error("patch %s has been renamed/deleted",
2238 patch->old_name);
2240 /* We have a patched copy in memory use that */
2241 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2242 } else if (cached) {
2243 if (read_file_or_gitlink(ce, &buf))
2244 return error("read of %s failed", patch->old_name);
2245 } else if (patch->old_name) {
2246 if (S_ISGITLINK(patch->old_mode)) {
2247 if (ce) {
2248 read_file_or_gitlink(ce, &buf);
2249 } else {
2251 * There is no way to apply subproject
2252 * patch without looking at the index.
2254 patch->fragments = NULL;
2256 } else {
2257 if (read_old_data(st, patch->old_name, &buf))
2258 return error("read of %s failed", patch->old_name);
2262 img = strbuf_detach(&buf, &len);
2263 prepare_image(&image, img, len, !patch->is_binary);
2265 if (apply_fragments(&image, patch) < 0)
2266 return -1; /* note with --reject this succeeds. */
2267 patch->result = image.buf;
2268 patch->resultsize = image.len;
2269 add_to_fn_table(patch);
2270 free(image.line_allocated);
2272 if (0 < patch->is_delete && patch->resultsize)
2273 return error("removal patch leaves file contents");
2275 return 0;
2278 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2280 struct stat nst;
2281 if (!lstat(new_name, &nst)) {
2282 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2283 return 0;
2285 * A leading component of new_name might be a symlink
2286 * that is going to be removed with this patch, but
2287 * still pointing at somewhere that has the path.
2288 * In such a case, path "new_name" does not exist as
2289 * far as git is concerned.
2291 if (has_symlink_leading_path(strlen(new_name), new_name))
2292 return 0;
2294 return error("%s: already exists in working directory", new_name);
2296 else if ((errno != ENOENT) && (errno != ENOTDIR))
2297 return error("%s: %s", new_name, strerror(errno));
2298 return 0;
2301 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2303 if (S_ISGITLINK(ce->ce_mode)) {
2304 if (!S_ISDIR(st->st_mode))
2305 return -1;
2306 return 0;
2308 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2311 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2313 const char *old_name = patch->old_name;
2314 struct patch *tpatch;
2315 int stat_ret = 0;
2316 unsigned st_mode = 0;
2319 * Make sure that we do not have local modifications from the
2320 * index when we are looking at the index. Also make sure
2321 * we have the preimage file to be patched in the work tree,
2322 * unless --cached, which tells git to apply only in the index.
2324 if (!old_name)
2325 return 0;
2327 assert(patch->is_new <= 0);
2328 if ((tpatch = in_fn_table(old_name)) != NULL) {
2329 if (tpatch == (struct patch *) -1) {
2330 return error("%s: has been deleted/renamed", old_name);
2332 st_mode = tpatch->new_mode;
2333 } else if (!cached) {
2334 stat_ret = lstat(old_name, st);
2335 if (stat_ret && errno != ENOENT)
2336 return error("%s: %s", old_name, strerror(errno));
2338 if (check_index && !tpatch) {
2339 int pos = cache_name_pos(old_name, strlen(old_name));
2340 if (pos < 0) {
2341 if (patch->is_new < 0)
2342 goto is_new;
2343 return error("%s: does not exist in index", old_name);
2345 *ce = active_cache[pos];
2346 if (stat_ret < 0) {
2347 struct checkout costate;
2348 /* checkout */
2349 costate.base_dir = "";
2350 costate.base_dir_len = 0;
2351 costate.force = 0;
2352 costate.quiet = 0;
2353 costate.not_new = 0;
2354 costate.refresh_cache = 1;
2355 if (checkout_entry(*ce, &costate, NULL) ||
2356 lstat(old_name, st))
2357 return -1;
2359 if (!cached && verify_index_match(*ce, st))
2360 return error("%s: does not match index", old_name);
2361 if (cached)
2362 st_mode = (*ce)->ce_mode;
2363 } else if (stat_ret < 0) {
2364 if (patch->is_new < 0)
2365 goto is_new;
2366 return error("%s: %s", old_name, strerror(errno));
2369 if (!cached)
2370 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2372 if (patch->is_new < 0)
2373 patch->is_new = 0;
2374 if (!patch->old_mode)
2375 patch->old_mode = st_mode;
2376 if ((st_mode ^ patch->old_mode) & S_IFMT)
2377 return error("%s: wrong type", old_name);
2378 if (st_mode != patch->old_mode)
2379 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2380 old_name, st_mode, patch->old_mode);
2381 return 0;
2383 is_new:
2384 patch->is_new = 1;
2385 patch->is_delete = 0;
2386 patch->old_name = NULL;
2387 return 0;
2390 static int check_patch(struct patch *patch)
2392 struct stat st;
2393 const char *old_name = patch->old_name;
2394 const char *new_name = patch->new_name;
2395 const char *name = old_name ? old_name : new_name;
2396 struct cache_entry *ce = NULL;
2397 int ok_if_exists;
2398 int status;
2400 patch->rejected = 1; /* we will drop this after we succeed */
2402 status = check_preimage(patch, &ce, &st);
2403 if (status)
2404 return status;
2405 old_name = patch->old_name;
2407 if (in_fn_table(new_name) == (struct patch *) -1)
2409 * A type-change diff is always split into a patch to
2410 * delete old, immediately followed by a patch to
2411 * create new (see diff.c::run_diff()); in such a case
2412 * it is Ok that the entry to be deleted by the
2413 * previous patch is still in the working tree and in
2414 * the index.
2416 ok_if_exists = 1;
2417 else
2418 ok_if_exists = 0;
2420 if (new_name &&
2421 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2422 if (check_index &&
2423 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2424 !ok_if_exists)
2425 return error("%s: already exists in index", new_name);
2426 if (!cached) {
2427 int err = check_to_create_blob(new_name, ok_if_exists);
2428 if (err)
2429 return err;
2431 if (!patch->new_mode) {
2432 if (0 < patch->is_new)
2433 patch->new_mode = S_IFREG | 0644;
2434 else
2435 patch->new_mode = patch->old_mode;
2439 if (new_name && old_name) {
2440 int same = !strcmp(old_name, new_name);
2441 if (!patch->new_mode)
2442 patch->new_mode = patch->old_mode;
2443 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2444 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2445 patch->new_mode, new_name, patch->old_mode,
2446 same ? "" : " of ", same ? "" : old_name);
2449 if (apply_data(patch, &st, ce) < 0)
2450 return error("%s: patch does not apply", name);
2451 patch->rejected = 0;
2452 return 0;
2455 static int check_patch_list(struct patch *patch)
2457 int err = 0;
2459 while (patch) {
2460 if (apply_verbosely)
2461 say_patch_name(stderr,
2462 "Checking patch ", patch, "...\n");
2463 err |= check_patch(patch);
2464 patch = patch->next;
2466 return err;
2469 /* This function tries to read the sha1 from the current index */
2470 static int get_current_sha1(const char *path, unsigned char *sha1)
2472 int pos;
2474 if (read_cache() < 0)
2475 return -1;
2476 pos = cache_name_pos(path, strlen(path));
2477 if (pos < 0)
2478 return -1;
2479 hashcpy(sha1, active_cache[pos]->sha1);
2480 return 0;
2483 /* Build an index that contains the just the files needed for a 3way merge */
2484 static void build_fake_ancestor(struct patch *list, const char *filename)
2486 struct patch *patch;
2487 struct index_state result = { 0 };
2488 int fd;
2490 /* Once we start supporting the reverse patch, it may be
2491 * worth showing the new sha1 prefix, but until then...
2493 for (patch = list; patch; patch = patch->next) {
2494 const unsigned char *sha1_ptr;
2495 unsigned char sha1[20];
2496 struct cache_entry *ce;
2497 const char *name;
2499 name = patch->old_name ? patch->old_name : patch->new_name;
2500 if (0 < patch->is_new)
2501 continue;
2502 else if (get_sha1(patch->old_sha1_prefix, sha1))
2503 /* git diff has no index line for mode/type changes */
2504 if (!patch->lines_added && !patch->lines_deleted) {
2505 if (get_current_sha1(patch->new_name, sha1) ||
2506 get_current_sha1(patch->old_name, sha1))
2507 die("mode change for %s, which is not "
2508 "in current HEAD", name);
2509 sha1_ptr = sha1;
2510 } else
2511 die("sha1 information is lacking or useless "
2512 "(%s).", name);
2513 else
2514 sha1_ptr = sha1;
2516 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2517 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2518 die ("Could not add %s to temporary index", name);
2521 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2522 if (fd < 0 || write_index(&result, fd) || close(fd))
2523 die ("Could not write temporary index to %s", filename);
2525 discard_index(&result);
2528 static void stat_patch_list(struct patch *patch)
2530 int files, adds, dels;
2532 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2533 files++;
2534 adds += patch->lines_added;
2535 dels += patch->lines_deleted;
2536 show_stats(patch);
2539 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2542 static void numstat_patch_list(struct patch *patch)
2544 for ( ; patch; patch = patch->next) {
2545 const char *name;
2546 name = patch->new_name ? patch->new_name : patch->old_name;
2547 if (patch->is_binary)
2548 printf("-\t-\t");
2549 else
2550 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2551 write_name_quoted(name, stdout, line_termination);
2555 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2557 if (mode)
2558 printf(" %s mode %06o %s\n", newdelete, mode, name);
2559 else
2560 printf(" %s %s\n", newdelete, name);
2563 static void show_mode_change(struct patch *p, int show_name)
2565 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2566 if (show_name)
2567 printf(" mode change %06o => %06o %s\n",
2568 p->old_mode, p->new_mode, p->new_name);
2569 else
2570 printf(" mode change %06o => %06o\n",
2571 p->old_mode, p->new_mode);
2575 static void show_rename_copy(struct patch *p)
2577 const char *renamecopy = p->is_rename ? "rename" : "copy";
2578 const char *old, *new;
2580 /* Find common prefix */
2581 old = p->old_name;
2582 new = p->new_name;
2583 while (1) {
2584 const char *slash_old, *slash_new;
2585 slash_old = strchr(old, '/');
2586 slash_new = strchr(new, '/');
2587 if (!slash_old ||
2588 !slash_new ||
2589 slash_old - old != slash_new - new ||
2590 memcmp(old, new, slash_new - new))
2591 break;
2592 old = slash_old + 1;
2593 new = slash_new + 1;
2595 /* p->old_name thru old is the common prefix, and old and new
2596 * through the end of names are renames
2598 if (old != p->old_name)
2599 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2600 (int)(old - p->old_name), p->old_name,
2601 old, new, p->score);
2602 else
2603 printf(" %s %s => %s (%d%%)\n", renamecopy,
2604 p->old_name, p->new_name, p->score);
2605 show_mode_change(p, 0);
2608 static void summary_patch_list(struct patch *patch)
2610 struct patch *p;
2612 for (p = patch; p; p = p->next) {
2613 if (p->is_new)
2614 show_file_mode_name("create", p->new_mode, p->new_name);
2615 else if (p->is_delete)
2616 show_file_mode_name("delete", p->old_mode, p->old_name);
2617 else {
2618 if (p->is_rename || p->is_copy)
2619 show_rename_copy(p);
2620 else {
2621 if (p->score) {
2622 printf(" rewrite %s (%d%%)\n",
2623 p->new_name, p->score);
2624 show_mode_change(p, 0);
2626 else
2627 show_mode_change(p, 1);
2633 static void patch_stats(struct patch *patch)
2635 int lines = patch->lines_added + patch->lines_deleted;
2637 if (lines > max_change)
2638 max_change = lines;
2639 if (patch->old_name) {
2640 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2641 if (!len)
2642 len = strlen(patch->old_name);
2643 if (len > max_len)
2644 max_len = len;
2646 if (patch->new_name) {
2647 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2648 if (!len)
2649 len = strlen(patch->new_name);
2650 if (len > max_len)
2651 max_len = len;
2655 static void remove_file(struct patch *patch, int rmdir_empty)
2657 if (update_index) {
2658 if (remove_file_from_cache(patch->old_name) < 0)
2659 die("unable to remove %s from index", patch->old_name);
2661 if (!cached) {
2662 if (S_ISGITLINK(patch->old_mode)) {
2663 if (rmdir(patch->old_name))
2664 warning("unable to remove submodule %s",
2665 patch->old_name);
2666 } else if (!unlink(patch->old_name) && rmdir_empty) {
2667 char *name = xstrdup(patch->old_name);
2668 char *end = strrchr(name, '/');
2669 while (end) {
2670 *end = 0;
2671 if (rmdir(name))
2672 break;
2673 end = strrchr(name, '/');
2675 free(name);
2680 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2682 struct stat st;
2683 struct cache_entry *ce;
2684 int namelen = strlen(path);
2685 unsigned ce_size = cache_entry_size(namelen);
2687 if (!update_index)
2688 return;
2690 ce = xcalloc(1, ce_size);
2691 memcpy(ce->name, path, namelen);
2692 ce->ce_mode = create_ce_mode(mode);
2693 ce->ce_flags = namelen;
2694 if (S_ISGITLINK(mode)) {
2695 const char *s = buf;
2697 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2698 die("corrupt patch for subproject %s", path);
2699 } else {
2700 if (!cached) {
2701 if (lstat(path, &st) < 0)
2702 die("unable to stat newly created file %s",
2703 path);
2704 fill_stat_cache_info(ce, &st);
2706 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2707 die("unable to create backing store for newly created file %s", path);
2709 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2710 die("unable to add cache entry for %s", path);
2713 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2715 int fd;
2716 struct strbuf nbuf;
2718 if (S_ISGITLINK(mode)) {
2719 struct stat st;
2720 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2721 return 0;
2722 return mkdir(path, 0777);
2725 if (has_symlinks && S_ISLNK(mode))
2726 /* Although buf:size is counted string, it also is NUL
2727 * terminated.
2729 return symlink(buf, path);
2731 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2732 if (fd < 0)
2733 return -1;
2735 strbuf_init(&nbuf, 0);
2736 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2737 size = nbuf.len;
2738 buf = nbuf.buf;
2740 write_or_die(fd, buf, size);
2741 strbuf_release(&nbuf);
2743 if (close(fd) < 0)
2744 die("closing file %s: %s", path, strerror(errno));
2745 return 0;
2749 * We optimistically assume that the directories exist,
2750 * which is true 99% of the time anyway. If they don't,
2751 * we create them and try again.
2753 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2755 if (cached)
2756 return;
2757 if (!try_create_file(path, mode, buf, size))
2758 return;
2760 if (errno == ENOENT) {
2761 if (safe_create_leading_directories(path))
2762 return;
2763 if (!try_create_file(path, mode, buf, size))
2764 return;
2767 if (errno == EEXIST || errno == EACCES) {
2768 /* We may be trying to create a file where a directory
2769 * used to be.
2771 struct stat st;
2772 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2773 errno = EEXIST;
2776 if (errno == EEXIST) {
2777 unsigned int nr = getpid();
2779 for (;;) {
2780 const char *newpath;
2781 newpath = mkpath("%s~%u", path, nr);
2782 if (!try_create_file(newpath, mode, buf, size)) {
2783 if (!rename(newpath, path))
2784 return;
2785 unlink(newpath);
2786 break;
2788 if (errno != EEXIST)
2789 break;
2790 ++nr;
2793 die("unable to write file %s mode %o", path, mode);
2796 static void create_file(struct patch *patch)
2798 char *path = patch->new_name;
2799 unsigned mode = patch->new_mode;
2800 unsigned long size = patch->resultsize;
2801 char *buf = patch->result;
2803 if (!mode)
2804 mode = S_IFREG | 0644;
2805 create_one_file(path, mode, buf, size);
2806 add_index_file(path, mode, buf, size);
2809 /* phase zero is to remove, phase one is to create */
2810 static void write_out_one_result(struct patch *patch, int phase)
2812 if (patch->is_delete > 0) {
2813 if (phase == 0)
2814 remove_file(patch, 1);
2815 return;
2817 if (patch->is_new > 0 || patch->is_copy) {
2818 if (phase == 1)
2819 create_file(patch);
2820 return;
2823 * Rename or modification boils down to the same
2824 * thing: remove the old, write the new
2826 if (phase == 0)
2827 remove_file(patch, patch->is_rename);
2828 if (phase == 1)
2829 create_file(patch);
2832 static int write_out_one_reject(struct patch *patch)
2834 FILE *rej;
2835 char namebuf[PATH_MAX];
2836 struct fragment *frag;
2837 int cnt = 0;
2839 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2840 if (!frag->rejected)
2841 continue;
2842 cnt++;
2845 if (!cnt) {
2846 if (apply_verbosely)
2847 say_patch_name(stderr,
2848 "Applied patch ", patch, " cleanly.\n");
2849 return 0;
2852 /* This should not happen, because a removal patch that leaves
2853 * contents are marked "rejected" at the patch level.
2855 if (!patch->new_name)
2856 die("internal error");
2858 /* Say this even without --verbose */
2859 say_patch_name(stderr, "Applying patch ", patch, " with");
2860 fprintf(stderr, " %d rejects...\n", cnt);
2862 cnt = strlen(patch->new_name);
2863 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2864 cnt = ARRAY_SIZE(namebuf) - 5;
2865 fprintf(stderr,
2866 "warning: truncating .rej filename to %.*s.rej",
2867 cnt - 1, patch->new_name);
2869 memcpy(namebuf, patch->new_name, cnt);
2870 memcpy(namebuf + cnt, ".rej", 5);
2872 rej = fopen(namebuf, "w");
2873 if (!rej)
2874 return error("cannot open %s: %s", namebuf, strerror(errno));
2876 /* Normal git tools never deal with .rej, so do not pretend
2877 * this is a git patch by saying --git nor give extended
2878 * headers. While at it, maybe please "kompare" that wants
2879 * the trailing TAB and some garbage at the end of line ;-).
2881 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2882 patch->new_name, patch->new_name);
2883 for (cnt = 1, frag = patch->fragments;
2884 frag;
2885 cnt++, frag = frag->next) {
2886 if (!frag->rejected) {
2887 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2888 continue;
2890 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2891 fprintf(rej, "%.*s", frag->size, frag->patch);
2892 if (frag->patch[frag->size-1] != '\n')
2893 fputc('\n', rej);
2895 fclose(rej);
2896 return -1;
2899 static int write_out_results(struct patch *list, int skipped_patch)
2901 int phase;
2902 int errs = 0;
2903 struct patch *l;
2905 if (!list && !skipped_patch)
2906 return error("No changes");
2908 for (phase = 0; phase < 2; phase++) {
2909 l = list;
2910 while (l) {
2911 if (l->rejected)
2912 errs = 1;
2913 else {
2914 write_out_one_result(l, phase);
2915 if (phase == 1 && write_out_one_reject(l))
2916 errs = 1;
2918 l = l->next;
2921 return errs;
2924 static struct lock_file lock_file;
2926 static struct excludes {
2927 struct excludes *next;
2928 const char *path;
2929 } *excludes;
2931 static int use_patch(struct patch *p)
2933 const char *pathname = p->new_name ? p->new_name : p->old_name;
2934 struct excludes *x = excludes;
2935 while (x) {
2936 if (fnmatch(x->path, pathname, 0) == 0)
2937 return 0;
2938 x = x->next;
2940 if (0 < prefix_length) {
2941 int pathlen = strlen(pathname);
2942 if (pathlen <= prefix_length ||
2943 memcmp(prefix, pathname, prefix_length))
2944 return 0;
2946 return 1;
2949 static void prefix_one(char **name)
2951 char *old_name = *name;
2952 if (!old_name)
2953 return;
2954 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2955 free(old_name);
2958 static void prefix_patches(struct patch *p)
2960 if (!prefix || p->is_toplevel_relative)
2961 return;
2962 for ( ; p; p = p->next) {
2963 if (p->new_name == p->old_name) {
2964 char *prefixed = p->new_name;
2965 prefix_one(&prefixed);
2966 p->new_name = p->old_name = prefixed;
2968 else {
2969 prefix_one(&p->new_name);
2970 prefix_one(&p->old_name);
2975 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2977 size_t offset;
2978 struct strbuf buf;
2979 struct patch *list = NULL, **listp = &list;
2980 int skipped_patch = 0;
2982 /* FIXME - memory leak when using multiple patch files as inputs */
2983 memset(&fn_table, 0, sizeof(struct path_list));
2984 strbuf_init(&buf, 0);
2985 patch_input_file = filename;
2986 read_patch_file(&buf, fd);
2987 offset = 0;
2988 while (offset < buf.len) {
2989 struct patch *patch;
2990 int nr;
2992 patch = xcalloc(1, sizeof(*patch));
2993 patch->inaccurate_eof = inaccurate_eof;
2994 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2995 if (nr < 0)
2996 break;
2997 if (apply_in_reverse)
2998 reverse_patches(patch);
2999 if (prefix)
3000 prefix_patches(patch);
3001 if (use_patch(patch)) {
3002 patch_stats(patch);
3003 *listp = patch;
3004 listp = &patch->next;
3006 else {
3007 /* perhaps free it a bit better? */
3008 free(patch);
3009 skipped_patch++;
3011 offset += nr;
3014 if (whitespace_error && (ws_error_action == die_on_ws_error))
3015 apply = 0;
3017 update_index = check_index && apply;
3018 if (update_index && newfd < 0)
3019 newfd = hold_locked_index(&lock_file, 1);
3021 if (check_index) {
3022 if (read_cache() < 0)
3023 die("unable to read index file");
3026 if ((check || apply) &&
3027 check_patch_list(list) < 0 &&
3028 !apply_with_reject)
3029 exit(1);
3031 if (apply && write_out_results(list, skipped_patch))
3032 exit(1);
3034 if (fake_ancestor)
3035 build_fake_ancestor(list, fake_ancestor);
3037 if (diffstat)
3038 stat_patch_list(list);
3040 if (numstat)
3041 numstat_patch_list(list);
3043 if (summary)
3044 summary_patch_list(list);
3046 strbuf_release(&buf);
3047 return 0;
3050 static int git_apply_config(const char *var, const char *value, void *cb)
3052 if (!strcmp(var, "apply.whitespace"))
3053 return git_config_string(&apply_default_whitespace, var, value);
3054 return git_default_config(var, value, cb);
3058 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3060 int i;
3061 int read_stdin = 1;
3062 int inaccurate_eof = 0;
3063 int errs = 0;
3064 int is_not_gitdir;
3066 const char *whitespace_option = NULL;
3068 prefix = setup_git_directory_gently(&is_not_gitdir);
3069 prefix_length = prefix ? strlen(prefix) : 0;
3070 git_config(git_apply_config, NULL);
3071 if (apply_default_whitespace)
3072 parse_whitespace_option(apply_default_whitespace);
3074 for (i = 1; i < argc; i++) {
3075 const char *arg = argv[i];
3076 char *end;
3077 int fd;
3079 if (!strcmp(arg, "-")) {
3080 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3081 read_stdin = 0;
3082 continue;
3084 if (!prefixcmp(arg, "--exclude=")) {
3085 struct excludes *x = xmalloc(sizeof(*x));
3086 x->path = arg + 10;
3087 x->next = excludes;
3088 excludes = x;
3089 continue;
3091 if (!prefixcmp(arg, "-p")) {
3092 p_value = atoi(arg + 2);
3093 p_value_known = 1;
3094 continue;
3096 if (!strcmp(arg, "--no-add")) {
3097 no_add = 1;
3098 continue;
3100 if (!strcmp(arg, "--stat")) {
3101 apply = 0;
3102 diffstat = 1;
3103 continue;
3105 if (!strcmp(arg, "--allow-binary-replacement") ||
3106 !strcmp(arg, "--binary")) {
3107 continue; /* now no-op */
3109 if (!strcmp(arg, "--numstat")) {
3110 apply = 0;
3111 numstat = 1;
3112 continue;
3114 if (!strcmp(arg, "--summary")) {
3115 apply = 0;
3116 summary = 1;
3117 continue;
3119 if (!strcmp(arg, "--check")) {
3120 apply = 0;
3121 check = 1;
3122 continue;
3124 if (!strcmp(arg, "--index")) {
3125 if (is_not_gitdir)
3126 die("--index outside a repository");
3127 check_index = 1;
3128 continue;
3130 if (!strcmp(arg, "--cached")) {
3131 if (is_not_gitdir)
3132 die("--cached outside a repository");
3133 check_index = 1;
3134 cached = 1;
3135 continue;
3137 if (!strcmp(arg, "--apply")) {
3138 apply = 1;
3139 continue;
3141 if (!strcmp(arg, "--build-fake-ancestor")) {
3142 apply = 0;
3143 if (++i >= argc)
3144 die ("need a filename");
3145 fake_ancestor = argv[i];
3146 continue;
3148 if (!strcmp(arg, "-z")) {
3149 line_termination = 0;
3150 continue;
3152 if (!prefixcmp(arg, "-C")) {
3153 p_context = strtoul(arg + 2, &end, 0);
3154 if (*end != '\0')
3155 die("unrecognized context count '%s'", arg + 2);
3156 continue;
3158 if (!prefixcmp(arg, "--whitespace=")) {
3159 whitespace_option = arg + 13;
3160 parse_whitespace_option(arg + 13);
3161 continue;
3163 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3164 apply_in_reverse = 1;
3165 continue;
3167 if (!strcmp(arg, "--unidiff-zero")) {
3168 unidiff_zero = 1;
3169 continue;
3171 if (!strcmp(arg, "--reject")) {
3172 apply = apply_with_reject = apply_verbosely = 1;
3173 continue;
3175 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3176 apply_verbosely = 1;
3177 continue;
3179 if (!strcmp(arg, "--inaccurate-eof")) {
3180 inaccurate_eof = 1;
3181 continue;
3183 if (0 < prefix_length)
3184 arg = prefix_filename(prefix, prefix_length, arg);
3186 fd = open(arg, O_RDONLY);
3187 if (fd < 0)
3188 die("can't open patch '%s': %s", arg, strerror(errno));
3189 read_stdin = 0;
3190 set_default_whitespace_mode(whitespace_option);
3191 errs |= apply_patch(fd, arg, inaccurate_eof);
3192 close(fd);
3194 set_default_whitespace_mode(whitespace_option);
3195 if (read_stdin)
3196 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3197 if (whitespace_error) {
3198 if (squelch_whitespace_errors &&
3199 squelch_whitespace_errors < whitespace_error) {
3200 int squelched =
3201 whitespace_error - squelch_whitespace_errors;
3202 fprintf(stderr, "warning: squelched %d "
3203 "whitespace error%s\n",
3204 squelched,
3205 squelched == 1 ? "" : "s");
3207 if (ws_error_action == die_on_ws_error)
3208 die("%d line%s add%s whitespace errors.",
3209 whitespace_error,
3210 whitespace_error == 1 ? "" : "s",
3211 whitespace_error == 1 ? "s" : "");
3212 if (applied_after_fixing_ws && apply)
3213 fprintf(stderr, "warning: %d line%s applied after"
3214 " fixing whitespace errors.\n",
3215 applied_after_fixing_ws,
3216 applied_after_fixing_ws == 1 ? "" : "s");
3217 else if (whitespace_error)
3218 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3219 whitespace_error,
3220 whitespace_error == 1 ? "" : "s",
3221 whitespace_error == 1 ? "s" : "");
3224 if (update_index) {
3225 if (write_cache(newfd, active_cache, active_nr) ||
3226 commit_locked_index(&lock_file))
3227 die("Unable to write new index file");
3230 return !!errs;