Documentation: rewrap to prepare for "git-" vs "git " change
[git/git-svn.git] / builtin-apply.c
blobc0f867daed351d4b32f871941e14773a44508eaf
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 unsigned int recount:1;
158 struct fragment *fragments;
159 char *result;
160 size_t resultsize;
161 char old_sha1_prefix[41];
162 char new_sha1_prefix[41];
163 struct patch *next;
167 * A line in a file, len-bytes long (includes the terminating LF,
168 * except for an incomplete line at the end if the file ends with
169 * one), and its contents hashes to 'hash'.
171 struct line {
172 size_t len;
173 unsigned hash : 24;
174 unsigned flag : 8;
175 #define LINE_COMMON 1
179 * This represents a "file", which is an array of "lines".
181 struct image {
182 char *buf;
183 size_t len;
184 size_t nr;
185 size_t alloc;
186 struct line *line_allocated;
187 struct line *line;
191 * Records filenames that have been touched, in order to handle
192 * the case where more than one patches touch the same file.
195 static struct path_list fn_table;
197 static uint32_t hash_line(const char *cp, size_t len)
199 size_t i;
200 uint32_t h;
201 for (i = 0, h = 0; i < len; i++) {
202 if (!isspace(cp[i])) {
203 h = h * 3 + (cp[i] & 0xff);
206 return h;
209 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
211 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
212 img->line_allocated[img->nr].len = len;
213 img->line_allocated[img->nr].hash = hash_line(bol, len);
214 img->line_allocated[img->nr].flag = flag;
215 img->nr++;
218 static void prepare_image(struct image *image, char *buf, size_t len,
219 int prepare_linetable)
221 const char *cp, *ep;
223 memset(image, 0, sizeof(*image));
224 image->buf = buf;
225 image->len = len;
227 if (!prepare_linetable)
228 return;
230 ep = image->buf + image->len;
231 cp = image->buf;
232 while (cp < ep) {
233 const char *next;
234 for (next = cp; next < ep && *next != '\n'; next++)
236 if (next < ep)
237 next++;
238 add_line_info(image, cp, next - cp, 0);
239 cp = next;
241 image->line = image->line_allocated;
244 static void clear_image(struct image *image)
246 free(image->buf);
247 image->buf = NULL;
248 image->len = 0;
251 static void say_patch_name(FILE *output, const char *pre,
252 struct patch *patch, const char *post)
254 fputs(pre, output);
255 if (patch->old_name && patch->new_name &&
256 strcmp(patch->old_name, patch->new_name)) {
257 quote_c_style(patch->old_name, NULL, output, 0);
258 fputs(" => ", output);
259 quote_c_style(patch->new_name, NULL, output, 0);
260 } else {
261 const char *n = patch->new_name;
262 if (!n)
263 n = patch->old_name;
264 quote_c_style(n, NULL, output, 0);
266 fputs(post, output);
269 #define CHUNKSIZE (8192)
270 #define SLOP (16)
272 static void read_patch_file(struct strbuf *sb, int fd)
274 if (strbuf_read(sb, fd, 0) < 0)
275 die("git-apply: read returned %s", strerror(errno));
278 * Make sure that we have some slop in the buffer
279 * so that we can do speculative "memcmp" etc, and
280 * see to it that it is NUL-filled.
282 strbuf_grow(sb, SLOP);
283 memset(sb->buf + sb->len, 0, SLOP);
286 static unsigned long linelen(const char *buffer, unsigned long size)
288 unsigned long len = 0;
289 while (size--) {
290 len++;
291 if (*buffer++ == '\n')
292 break;
294 return len;
297 static int is_dev_null(const char *str)
299 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
302 #define TERM_SPACE 1
303 #define TERM_TAB 2
305 static int name_terminate(const char *name, int namelen, int c, int terminate)
307 if (c == ' ' && !(terminate & TERM_SPACE))
308 return 0;
309 if (c == '\t' && !(terminate & TERM_TAB))
310 return 0;
312 return 1;
315 static char *find_name(const char *line, char *def, int p_value, int terminate)
317 int len;
318 const char *start = line;
320 if (*line == '"') {
321 struct strbuf name;
324 * Proposed "new-style" GNU patch/diff format; see
325 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
327 strbuf_init(&name, 0);
328 if (!unquote_c_style(&name, line, NULL)) {
329 char *cp;
331 for (cp = name.buf; p_value; p_value--) {
332 cp = strchr(cp, '/');
333 if (!cp)
334 break;
335 cp++;
337 if (cp) {
338 /* name can later be freed, so we need
339 * to memmove, not just return cp
341 strbuf_remove(&name, 0, cp - name.buf);
342 free(def);
343 return strbuf_detach(&name, NULL);
346 strbuf_release(&name);
349 for (;;) {
350 char c = *line;
352 if (isspace(c)) {
353 if (c == '\n')
354 break;
355 if (name_terminate(start, line-start, c, terminate))
356 break;
358 line++;
359 if (c == '/' && !--p_value)
360 start = line;
362 if (!start)
363 return def;
364 len = line - start;
365 if (!len)
366 return def;
369 * Generally we prefer the shorter name, especially
370 * if the other one is just a variation of that with
371 * something else tacked on to the end (ie "file.orig"
372 * or "file~").
374 if (def) {
375 int deflen = strlen(def);
376 if (deflen < len && !strncmp(start, def, deflen))
377 return def;
378 free(def);
381 return xmemdupz(start, len);
384 static int count_slashes(const char *cp)
386 int cnt = 0;
387 char ch;
389 while ((ch = *cp++))
390 if (ch == '/')
391 cnt++;
392 return cnt;
396 * Given the string after "--- " or "+++ ", guess the appropriate
397 * p_value for the given patch.
399 static int guess_p_value(const char *nameline)
401 char *name, *cp;
402 int val = -1;
404 if (is_dev_null(nameline))
405 return -1;
406 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
407 if (!name)
408 return -1;
409 cp = strchr(name, '/');
410 if (!cp)
411 val = 0;
412 else if (prefix) {
414 * Does it begin with "a/$our-prefix" and such? Then this is
415 * very likely to apply to our directory.
417 if (!strncmp(name, prefix, prefix_length))
418 val = count_slashes(prefix);
419 else {
420 cp++;
421 if (!strncmp(cp, prefix, prefix_length))
422 val = count_slashes(prefix) + 1;
425 free(name);
426 return val;
430 * Get the name etc info from the ---/+++ lines of a traditional patch header
432 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
433 * files, we can happily check the index for a match, but for creating a
434 * new file we should try to match whatever "patch" does. I have no idea.
436 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
438 char *name;
440 first += 4; /* skip "--- " */
441 second += 4; /* skip "+++ " */
442 if (!p_value_known) {
443 int p, q;
444 p = guess_p_value(first);
445 q = guess_p_value(second);
446 if (p < 0) p = q;
447 if (0 <= p && p == q) {
448 p_value = p;
449 p_value_known = 1;
452 if (is_dev_null(first)) {
453 patch->is_new = 1;
454 patch->is_delete = 0;
455 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
456 patch->new_name = name;
457 } else if (is_dev_null(second)) {
458 patch->is_new = 0;
459 patch->is_delete = 1;
460 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
461 patch->old_name = name;
462 } else {
463 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
464 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
465 patch->old_name = patch->new_name = name;
467 if (!name)
468 die("unable to find filename in patch at line %d", linenr);
471 static int gitdiff_hdrend(const char *line, struct patch *patch)
473 return -1;
477 * We're anal about diff header consistency, to make
478 * sure that we don't end up having strange ambiguous
479 * patches floating around.
481 * As a result, gitdiff_{old|new}name() will check
482 * their names against any previous information, just
483 * to make sure..
485 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
487 if (!orig_name && !isnull)
488 return find_name(line, NULL, p_value, TERM_TAB);
490 if (orig_name) {
491 int len;
492 const char *name;
493 char *another;
494 name = orig_name;
495 len = strlen(name);
496 if (isnull)
497 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
498 another = find_name(line, NULL, p_value, TERM_TAB);
499 if (!another || memcmp(another, name, len))
500 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
501 free(another);
502 return orig_name;
504 else {
505 /* expect "/dev/null" */
506 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
507 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
508 return NULL;
512 static int gitdiff_oldname(const char *line, struct patch *patch)
514 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
515 return 0;
518 static int gitdiff_newname(const char *line, struct patch *patch)
520 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
521 return 0;
524 static int gitdiff_oldmode(const char *line, struct patch *patch)
526 patch->old_mode = strtoul(line, NULL, 8);
527 return 0;
530 static int gitdiff_newmode(const char *line, struct patch *patch)
532 patch->new_mode = strtoul(line, NULL, 8);
533 return 0;
536 static int gitdiff_delete(const char *line, struct patch *patch)
538 patch->is_delete = 1;
539 patch->old_name = patch->def_name;
540 return gitdiff_oldmode(line, patch);
543 static int gitdiff_newfile(const char *line, struct patch *patch)
545 patch->is_new = 1;
546 patch->new_name = patch->def_name;
547 return gitdiff_newmode(line, patch);
550 static int gitdiff_copysrc(const char *line, struct patch *patch)
552 patch->is_copy = 1;
553 patch->old_name = find_name(line, NULL, 0, 0);
554 return 0;
557 static int gitdiff_copydst(const char *line, struct patch *patch)
559 patch->is_copy = 1;
560 patch->new_name = find_name(line, NULL, 0, 0);
561 return 0;
564 static int gitdiff_renamesrc(const char *line, struct patch *patch)
566 patch->is_rename = 1;
567 patch->old_name = find_name(line, NULL, 0, 0);
568 return 0;
571 static int gitdiff_renamedst(const char *line, struct patch *patch)
573 patch->is_rename = 1;
574 patch->new_name = find_name(line, NULL, 0, 0);
575 return 0;
578 static int gitdiff_similarity(const char *line, struct patch *patch)
580 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
581 patch->score = 0;
582 return 0;
585 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
587 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
588 patch->score = 0;
589 return 0;
592 static int gitdiff_index(const char *line, struct patch *patch)
595 * index line is N hexadecimal, "..", N hexadecimal,
596 * and optional space with octal mode.
598 const char *ptr, *eol;
599 int len;
601 ptr = strchr(line, '.');
602 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
603 return 0;
604 len = ptr - line;
605 memcpy(patch->old_sha1_prefix, line, len);
606 patch->old_sha1_prefix[len] = 0;
608 line = ptr + 2;
609 ptr = strchr(line, ' ');
610 eol = strchr(line, '\n');
612 if (!ptr || eol < ptr)
613 ptr = eol;
614 len = ptr - line;
616 if (40 < len)
617 return 0;
618 memcpy(patch->new_sha1_prefix, line, len);
619 patch->new_sha1_prefix[len] = 0;
620 if (*ptr == ' ')
621 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
622 return 0;
626 * This is normal for a diff that doesn't change anything: we'll fall through
627 * into the next diff. Tell the parser to break out.
629 static int gitdiff_unrecognized(const char *line, struct patch *patch)
631 return -1;
634 static const char *stop_at_slash(const char *line, int llen)
636 int i;
638 for (i = 0; i < llen; i++) {
639 int ch = line[i];
640 if (ch == '/')
641 return line + i;
643 return NULL;
647 * This is to extract the same name that appears on "diff --git"
648 * line. We do not find and return anything if it is a rename
649 * patch, and it is OK because we will find the name elsewhere.
650 * We need to reliably find name only when it is mode-change only,
651 * creation or deletion of an empty file. In any of these cases,
652 * both sides are the same name under a/ and b/ respectively.
654 static char *git_header_name(char *line, int llen)
656 const char *name;
657 const char *second = NULL;
658 size_t len;
660 line += strlen("diff --git ");
661 llen -= strlen("diff --git ");
663 if (*line == '"') {
664 const char *cp;
665 struct strbuf first;
666 struct strbuf sp;
668 strbuf_init(&first, 0);
669 strbuf_init(&sp, 0);
671 if (unquote_c_style(&first, line, &second))
672 goto free_and_fail1;
674 /* advance to the first slash */
675 cp = stop_at_slash(first.buf, first.len);
676 /* we do not accept absolute paths */
677 if (!cp || cp == first.buf)
678 goto free_and_fail1;
679 strbuf_remove(&first, 0, cp + 1 - first.buf);
682 * second points at one past closing dq of name.
683 * find the second name.
685 while ((second < line + llen) && isspace(*second))
686 second++;
688 if (line + llen <= second)
689 goto free_and_fail1;
690 if (*second == '"') {
691 if (unquote_c_style(&sp, second, NULL))
692 goto free_and_fail1;
693 cp = stop_at_slash(sp.buf, sp.len);
694 if (!cp || cp == sp.buf)
695 goto free_and_fail1;
696 /* They must match, otherwise ignore */
697 if (strcmp(cp + 1, first.buf))
698 goto free_and_fail1;
699 strbuf_release(&sp);
700 return strbuf_detach(&first, NULL);
703 /* unquoted second */
704 cp = stop_at_slash(second, line + llen - second);
705 if (!cp || cp == second)
706 goto free_and_fail1;
707 cp++;
708 if (line + llen - cp != first.len + 1 ||
709 memcmp(first.buf, cp, first.len))
710 goto free_and_fail1;
711 return strbuf_detach(&first, NULL);
713 free_and_fail1:
714 strbuf_release(&first);
715 strbuf_release(&sp);
716 return NULL;
719 /* unquoted first name */
720 name = stop_at_slash(line, llen);
721 if (!name || name == line)
722 return NULL;
723 name++;
726 * since the first name is unquoted, a dq if exists must be
727 * the beginning of the second name.
729 for (second = name; second < line + llen; second++) {
730 if (*second == '"') {
731 struct strbuf sp;
732 const char *np;
734 strbuf_init(&sp, 0);
735 if (unquote_c_style(&sp, second, NULL))
736 goto free_and_fail2;
738 np = stop_at_slash(sp.buf, sp.len);
739 if (!np || np == sp.buf)
740 goto free_and_fail2;
741 np++;
743 len = sp.buf + sp.len - np;
744 if (len < second - name &&
745 !strncmp(np, name, len) &&
746 isspace(name[len])) {
747 /* Good */
748 strbuf_remove(&sp, 0, np - sp.buf);
749 return strbuf_detach(&sp, NULL);
752 free_and_fail2:
753 strbuf_release(&sp);
754 return NULL;
759 * Accept a name only if it shows up twice, exactly the same
760 * form.
762 for (len = 0 ; ; len++) {
763 switch (name[len]) {
764 default:
765 continue;
766 case '\n':
767 return NULL;
768 case '\t': case ' ':
769 second = name+len;
770 for (;;) {
771 char c = *second++;
772 if (c == '\n')
773 return NULL;
774 if (c == '/')
775 break;
777 if (second[len] == '\n' && !memcmp(name, second, len)) {
778 return xmemdupz(name, len);
784 /* Verify that we recognize the lines following a git header */
785 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
787 unsigned long offset;
789 /* A git diff has explicit new/delete information, so we don't guess */
790 patch->is_new = 0;
791 patch->is_delete = 0;
794 * Some things may not have the old name in the
795 * rest of the headers anywhere (pure mode changes,
796 * or removing or adding empty files), so we get
797 * the default name from the header.
799 patch->def_name = git_header_name(line, len);
801 line += len;
802 size -= len;
803 linenr++;
804 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
805 static const struct opentry {
806 const char *str;
807 int (*fn)(const char *, struct patch *);
808 } optable[] = {
809 { "@@ -", gitdiff_hdrend },
810 { "--- ", gitdiff_oldname },
811 { "+++ ", gitdiff_newname },
812 { "old mode ", gitdiff_oldmode },
813 { "new mode ", gitdiff_newmode },
814 { "deleted file mode ", gitdiff_delete },
815 { "new file mode ", gitdiff_newfile },
816 { "copy from ", gitdiff_copysrc },
817 { "copy to ", gitdiff_copydst },
818 { "rename old ", gitdiff_renamesrc },
819 { "rename new ", gitdiff_renamedst },
820 { "rename from ", gitdiff_renamesrc },
821 { "rename to ", gitdiff_renamedst },
822 { "similarity index ", gitdiff_similarity },
823 { "dissimilarity index ", gitdiff_dissimilarity },
824 { "index ", gitdiff_index },
825 { "", gitdiff_unrecognized },
827 int i;
829 len = linelen(line, size);
830 if (!len || line[len-1] != '\n')
831 break;
832 for (i = 0; i < ARRAY_SIZE(optable); i++) {
833 const struct opentry *p = optable + i;
834 int oplen = strlen(p->str);
835 if (len < oplen || memcmp(p->str, line, oplen))
836 continue;
837 if (p->fn(line + oplen, patch) < 0)
838 return offset;
839 break;
843 return offset;
846 static int parse_num(const char *line, unsigned long *p)
848 char *ptr;
850 if (!isdigit(*line))
851 return 0;
852 *p = strtoul(line, &ptr, 10);
853 return ptr - line;
856 static int parse_range(const char *line, int len, int offset, const char *expect,
857 unsigned long *p1, unsigned long *p2)
859 int digits, ex;
861 if (offset < 0 || offset >= len)
862 return -1;
863 line += offset;
864 len -= offset;
866 digits = parse_num(line, p1);
867 if (!digits)
868 return -1;
870 offset += digits;
871 line += digits;
872 len -= digits;
874 *p2 = 1;
875 if (*line == ',') {
876 digits = parse_num(line+1, p2);
877 if (!digits)
878 return -1;
880 offset += digits+1;
881 line += digits+1;
882 len -= digits+1;
885 ex = strlen(expect);
886 if (ex > len)
887 return -1;
888 if (memcmp(line, expect, ex))
889 return -1;
891 return offset + ex;
894 static void recount_diff(char *line, int size, struct fragment *fragment)
896 int oldlines = 0, newlines = 0, ret = 0;
898 if (size < 1) {
899 warning("recount: ignore empty hunk");
900 return;
903 for (;;) {
904 int len = linelen(line, size);
905 size -= len;
906 line += len;
908 if (size < 1)
909 break;
911 switch (*line) {
912 case ' ': case '\n':
913 newlines++;
914 /* fall through */
915 case '-':
916 oldlines++;
917 continue;
918 case '+':
919 newlines++;
920 continue;
921 case '\\':
922 continue;
923 case '@':
924 ret = size < 3 || prefixcmp(line, "@@ ");
925 break;
926 case 'd':
927 ret = size < 5 || prefixcmp(line, "diff ");
928 break;
929 default:
930 ret = -1;
931 break;
933 if (ret) {
934 warning("recount: unexpected line: %.*s",
935 (int)linelen(line, size), line);
936 return;
938 break;
940 fragment->oldlines = oldlines;
941 fragment->newlines = newlines;
945 * Parse a unified diff fragment header of the
946 * form "@@ -a,b +c,d @@"
948 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
950 int offset;
952 if (!len || line[len-1] != '\n')
953 return -1;
955 /* Figure out the number of lines in a fragment */
956 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
957 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
959 return offset;
962 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
964 unsigned long offset, len;
966 patch->is_toplevel_relative = 0;
967 patch->is_rename = patch->is_copy = 0;
968 patch->is_new = patch->is_delete = -1;
969 patch->old_mode = patch->new_mode = 0;
970 patch->old_name = patch->new_name = NULL;
971 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
972 unsigned long nextlen;
974 len = linelen(line, size);
975 if (!len)
976 break;
978 /* Testing this early allows us to take a few shortcuts.. */
979 if (len < 6)
980 continue;
983 * Make sure we don't find any unconnected patch fragments.
984 * That's a sign that we didn't find a header, and that a
985 * patch has become corrupted/broken up.
987 if (!memcmp("@@ -", line, 4)) {
988 struct fragment dummy;
989 if (parse_fragment_header(line, len, &dummy) < 0)
990 continue;
991 die("patch fragment without header at line %d: %.*s",
992 linenr, (int)len-1, line);
995 if (size < len + 6)
996 break;
999 * Git patch? It might not have a real patch, just a rename
1000 * or mode change, so we handle that specially
1002 if (!memcmp("diff --git ", line, 11)) {
1003 int git_hdr_len = parse_git_header(line, len, size, patch);
1004 if (git_hdr_len <= len)
1005 continue;
1006 if (!patch->old_name && !patch->new_name) {
1007 if (!patch->def_name)
1008 die("git diff header lacks filename information (line %d)", linenr);
1009 patch->old_name = patch->new_name = patch->def_name;
1011 patch->is_toplevel_relative = 1;
1012 *hdrsize = git_hdr_len;
1013 return offset;
1016 /* --- followed by +++ ? */
1017 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1018 continue;
1021 * We only accept unified patches, so we want it to
1022 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1023 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1025 nextlen = linelen(line + len, size - len);
1026 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1027 continue;
1029 /* Ok, we'll consider it a patch */
1030 parse_traditional_patch(line, line+len, patch);
1031 *hdrsize = len + nextlen;
1032 linenr += 2;
1033 return offset;
1035 return -1;
1038 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1040 char *err;
1041 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1042 if (!result)
1043 return;
1045 whitespace_error++;
1046 if (squelch_whitespace_errors &&
1047 squelch_whitespace_errors < whitespace_error)
1049 else {
1050 err = whitespace_error_string(result);
1051 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1052 patch_input_file, linenr, err, len - 2, line + 1);
1053 free(err);
1058 * Parse a unified diff. Note that this really needs to parse each
1059 * fragment separately, since the only way to know the difference
1060 * between a "---" that is part of a patch, and a "---" that starts
1061 * the next patch is to look at the line counts..
1063 static int parse_fragment(char *line, unsigned long size,
1064 struct patch *patch, struct fragment *fragment)
1066 int added, deleted;
1067 int len = linelen(line, size), offset;
1068 unsigned long oldlines, newlines;
1069 unsigned long leading, trailing;
1071 offset = parse_fragment_header(line, len, fragment);
1072 if (offset < 0)
1073 return -1;
1074 if (offset > 0 && patch->recount)
1075 recount_diff(line + offset, size - offset, fragment);
1076 oldlines = fragment->oldlines;
1077 newlines = fragment->newlines;
1078 leading = 0;
1079 trailing = 0;
1081 /* Parse the thing.. */
1082 line += len;
1083 size -= len;
1084 linenr++;
1085 added = deleted = 0;
1086 for (offset = len;
1087 0 < size;
1088 offset += len, size -= len, line += len, linenr++) {
1089 if (!oldlines && !newlines)
1090 break;
1091 len = linelen(line, size);
1092 if (!len || line[len-1] != '\n')
1093 return -1;
1094 switch (*line) {
1095 default:
1096 return -1;
1097 case '\n': /* newer GNU diff, an empty context line */
1098 case ' ':
1099 oldlines--;
1100 newlines--;
1101 if (!deleted && !added)
1102 leading++;
1103 trailing++;
1104 break;
1105 case '-':
1106 if (apply_in_reverse &&
1107 ws_error_action != nowarn_ws_error)
1108 check_whitespace(line, len, patch->ws_rule);
1109 deleted++;
1110 oldlines--;
1111 trailing = 0;
1112 break;
1113 case '+':
1114 if (!apply_in_reverse &&
1115 ws_error_action != nowarn_ws_error)
1116 check_whitespace(line, len, patch->ws_rule);
1117 added++;
1118 newlines--;
1119 trailing = 0;
1120 break;
1123 * We allow "\ No newline at end of file". Depending
1124 * on locale settings when the patch was produced we
1125 * don't know what this line looks like. The only
1126 * thing we do know is that it begins with "\ ".
1127 * Checking for 12 is just for sanity check -- any
1128 * l10n of "\ No newline..." is at least that long.
1130 case '\\':
1131 if (len < 12 || memcmp(line, "\\ ", 2))
1132 return -1;
1133 break;
1136 if (oldlines || newlines)
1137 return -1;
1138 fragment->leading = leading;
1139 fragment->trailing = trailing;
1142 * If a fragment ends with an incomplete line, we failed to include
1143 * it in the above loop because we hit oldlines == newlines == 0
1144 * before seeing it.
1146 if (12 < size && !memcmp(line, "\\ ", 2))
1147 offset += linelen(line, size);
1149 patch->lines_added += added;
1150 patch->lines_deleted += deleted;
1152 if (0 < patch->is_new && oldlines)
1153 return error("new file depends on old contents");
1154 if (0 < patch->is_delete && newlines)
1155 return error("deleted file still has contents");
1156 return offset;
1159 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1161 unsigned long offset = 0;
1162 unsigned long oldlines = 0, newlines = 0, context = 0;
1163 struct fragment **fragp = &patch->fragments;
1165 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1166 struct fragment *fragment;
1167 int len;
1169 fragment = xcalloc(1, sizeof(*fragment));
1170 len = parse_fragment(line, size, patch, fragment);
1171 if (len <= 0)
1172 die("corrupt patch at line %d", linenr);
1173 fragment->patch = line;
1174 fragment->size = len;
1175 oldlines += fragment->oldlines;
1176 newlines += fragment->newlines;
1177 context += fragment->leading + fragment->trailing;
1179 *fragp = fragment;
1180 fragp = &fragment->next;
1182 offset += len;
1183 line += len;
1184 size -= len;
1188 * If something was removed (i.e. we have old-lines) it cannot
1189 * be creation, and if something was added it cannot be
1190 * deletion. However, the reverse is not true; --unified=0
1191 * patches that only add are not necessarily creation even
1192 * though they do not have any old lines, and ones that only
1193 * delete are not necessarily deletion.
1195 * Unfortunately, a real creation/deletion patch do _not_ have
1196 * any context line by definition, so we cannot safely tell it
1197 * apart with --unified=0 insanity. At least if the patch has
1198 * more than one hunk it is not creation or deletion.
1200 if (patch->is_new < 0 &&
1201 (oldlines || (patch->fragments && patch->fragments->next)))
1202 patch->is_new = 0;
1203 if (patch->is_delete < 0 &&
1204 (newlines || (patch->fragments && patch->fragments->next)))
1205 patch->is_delete = 0;
1207 if (0 < patch->is_new && oldlines)
1208 die("new file %s depends on old contents", patch->new_name);
1209 if (0 < patch->is_delete && newlines)
1210 die("deleted file %s still has contents", patch->old_name);
1211 if (!patch->is_delete && !newlines && context)
1212 fprintf(stderr, "** warning: file %s becomes empty but "
1213 "is not deleted\n", patch->new_name);
1215 return offset;
1218 static inline int metadata_changes(struct patch *patch)
1220 return patch->is_rename > 0 ||
1221 patch->is_copy > 0 ||
1222 patch->is_new > 0 ||
1223 patch->is_delete ||
1224 (patch->old_mode && patch->new_mode &&
1225 patch->old_mode != patch->new_mode);
1228 static char *inflate_it(const void *data, unsigned long size,
1229 unsigned long inflated_size)
1231 z_stream stream;
1232 void *out;
1233 int st;
1235 memset(&stream, 0, sizeof(stream));
1237 stream.next_in = (unsigned char *)data;
1238 stream.avail_in = size;
1239 stream.next_out = out = xmalloc(inflated_size);
1240 stream.avail_out = inflated_size;
1241 inflateInit(&stream);
1242 st = inflate(&stream, Z_FINISH);
1243 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1244 free(out);
1245 return NULL;
1247 return out;
1250 static struct fragment *parse_binary_hunk(char **buf_p,
1251 unsigned long *sz_p,
1252 int *status_p,
1253 int *used_p)
1256 * Expect a line that begins with binary patch method ("literal"
1257 * or "delta"), followed by the length of data before deflating.
1258 * a sequence of 'length-byte' followed by base-85 encoded data
1259 * should follow, terminated by a newline.
1261 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1262 * and we would limit the patch line to 66 characters,
1263 * so one line can fit up to 13 groups that would decode
1264 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1265 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1267 int llen, used;
1268 unsigned long size = *sz_p;
1269 char *buffer = *buf_p;
1270 int patch_method;
1271 unsigned long origlen;
1272 char *data = NULL;
1273 int hunk_size = 0;
1274 struct fragment *frag;
1276 llen = linelen(buffer, size);
1277 used = llen;
1279 *status_p = 0;
1281 if (!prefixcmp(buffer, "delta ")) {
1282 patch_method = BINARY_DELTA_DEFLATED;
1283 origlen = strtoul(buffer + 6, NULL, 10);
1285 else if (!prefixcmp(buffer, "literal ")) {
1286 patch_method = BINARY_LITERAL_DEFLATED;
1287 origlen = strtoul(buffer + 8, NULL, 10);
1289 else
1290 return NULL;
1292 linenr++;
1293 buffer += llen;
1294 while (1) {
1295 int byte_length, max_byte_length, newsize;
1296 llen = linelen(buffer, size);
1297 used += llen;
1298 linenr++;
1299 if (llen == 1) {
1300 /* consume the blank line */
1301 buffer++;
1302 size--;
1303 break;
1306 * Minimum line is "A00000\n" which is 7-byte long,
1307 * and the line length must be multiple of 5 plus 2.
1309 if ((llen < 7) || (llen-2) % 5)
1310 goto corrupt;
1311 max_byte_length = (llen - 2) / 5 * 4;
1312 byte_length = *buffer;
1313 if ('A' <= byte_length && byte_length <= 'Z')
1314 byte_length = byte_length - 'A' + 1;
1315 else if ('a' <= byte_length && byte_length <= 'z')
1316 byte_length = byte_length - 'a' + 27;
1317 else
1318 goto corrupt;
1319 /* if the input length was not multiple of 4, we would
1320 * have filler at the end but the filler should never
1321 * exceed 3 bytes
1323 if (max_byte_length < byte_length ||
1324 byte_length <= max_byte_length - 4)
1325 goto corrupt;
1326 newsize = hunk_size + byte_length;
1327 data = xrealloc(data, newsize);
1328 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1329 goto corrupt;
1330 hunk_size = newsize;
1331 buffer += llen;
1332 size -= llen;
1335 frag = xcalloc(1, sizeof(*frag));
1336 frag->patch = inflate_it(data, hunk_size, origlen);
1337 if (!frag->patch)
1338 goto corrupt;
1339 free(data);
1340 frag->size = origlen;
1341 *buf_p = buffer;
1342 *sz_p = size;
1343 *used_p = used;
1344 frag->binary_patch_method = patch_method;
1345 return frag;
1347 corrupt:
1348 free(data);
1349 *status_p = -1;
1350 error("corrupt binary patch at line %d: %.*s",
1351 linenr-1, llen-1, buffer);
1352 return NULL;
1355 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1358 * We have read "GIT binary patch\n"; what follows is a line
1359 * that says the patch method (currently, either "literal" or
1360 * "delta") and the length of data before deflating; a
1361 * sequence of 'length-byte' followed by base-85 encoded data
1362 * follows.
1364 * When a binary patch is reversible, there is another binary
1365 * hunk in the same format, starting with patch method (either
1366 * "literal" or "delta") with the length of data, and a sequence
1367 * of length-byte + base-85 encoded data, terminated with another
1368 * empty line. This data, when applied to the postimage, produces
1369 * the preimage.
1371 struct fragment *forward;
1372 struct fragment *reverse;
1373 int status;
1374 int used, used_1;
1376 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1377 if (!forward && !status)
1378 /* there has to be one hunk (forward hunk) */
1379 return error("unrecognized binary patch at line %d", linenr-1);
1380 if (status)
1381 /* otherwise we already gave an error message */
1382 return status;
1384 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1385 if (reverse)
1386 used += used_1;
1387 else if (status) {
1389 * Not having reverse hunk is not an error, but having
1390 * a corrupt reverse hunk is.
1392 free((void*) forward->patch);
1393 free(forward);
1394 return status;
1396 forward->next = reverse;
1397 patch->fragments = forward;
1398 patch->is_binary = 1;
1399 return used;
1402 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1404 int hdrsize, patchsize;
1405 int offset = find_header(buffer, size, &hdrsize, patch);
1407 if (offset < 0)
1408 return offset;
1410 patch->ws_rule = whitespace_rule(patch->new_name
1411 ? patch->new_name
1412 : patch->old_name);
1414 patchsize = parse_single_patch(buffer + offset + hdrsize,
1415 size - offset - hdrsize, patch);
1417 if (!patchsize) {
1418 static const char *binhdr[] = {
1419 "Binary files ",
1420 "Files ",
1421 NULL,
1423 static const char git_binary[] = "GIT binary patch\n";
1424 int i;
1425 int hd = hdrsize + offset;
1426 unsigned long llen = linelen(buffer + hd, size - hd);
1428 if (llen == sizeof(git_binary) - 1 &&
1429 !memcmp(git_binary, buffer + hd, llen)) {
1430 int used;
1431 linenr++;
1432 used = parse_binary(buffer + hd + llen,
1433 size - hd - llen, patch);
1434 if (used)
1435 patchsize = used + llen;
1436 else
1437 patchsize = 0;
1439 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1440 for (i = 0; binhdr[i]; i++) {
1441 int len = strlen(binhdr[i]);
1442 if (len < size - hd &&
1443 !memcmp(binhdr[i], buffer + hd, len)) {
1444 linenr++;
1445 patch->is_binary = 1;
1446 patchsize = llen;
1447 break;
1452 /* Empty patch cannot be applied if it is a text patch
1453 * without metadata change. A binary patch appears
1454 * empty to us here.
1456 if ((apply || check) &&
1457 (!patch->is_binary && !metadata_changes(patch)))
1458 die("patch with only garbage at line %d", linenr);
1461 return offset + hdrsize + patchsize;
1464 #define swap(a,b) myswap((a),(b),sizeof(a))
1466 #define myswap(a, b, size) do { \
1467 unsigned char mytmp[size]; \
1468 memcpy(mytmp, &a, size); \
1469 memcpy(&a, &b, size); \
1470 memcpy(&b, mytmp, size); \
1471 } while (0)
1473 static void reverse_patches(struct patch *p)
1475 for (; p; p = p->next) {
1476 struct fragment *frag = p->fragments;
1478 swap(p->new_name, p->old_name);
1479 swap(p->new_mode, p->old_mode);
1480 swap(p->is_new, p->is_delete);
1481 swap(p->lines_added, p->lines_deleted);
1482 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1484 for (; frag; frag = frag->next) {
1485 swap(frag->newpos, frag->oldpos);
1486 swap(frag->newlines, frag->oldlines);
1491 static const char pluses[] =
1492 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1493 static const char minuses[]=
1494 "----------------------------------------------------------------------";
1496 static void show_stats(struct patch *patch)
1498 struct strbuf qname;
1499 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1500 int max, add, del;
1502 strbuf_init(&qname, 0);
1503 quote_c_style(cp, &qname, NULL, 0);
1506 * "scale" the filename
1508 max = max_len;
1509 if (max > 50)
1510 max = 50;
1512 if (qname.len > max) {
1513 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1514 if (!cp)
1515 cp = qname.buf + qname.len + 3 - max;
1516 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1519 if (patch->is_binary) {
1520 printf(" %-*s | Bin\n", max, qname.buf);
1521 strbuf_release(&qname);
1522 return;
1525 printf(" %-*s |", max, qname.buf);
1526 strbuf_release(&qname);
1529 * scale the add/delete
1531 max = max + max_change > 70 ? 70 - max : max_change;
1532 add = patch->lines_added;
1533 del = patch->lines_deleted;
1535 if (max_change > 0) {
1536 int total = ((add + del) * max + max_change / 2) / max_change;
1537 add = (add * max + max_change / 2) / max_change;
1538 del = total - add;
1540 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1541 add, pluses, del, minuses);
1544 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1546 switch (st->st_mode & S_IFMT) {
1547 case S_IFLNK:
1548 strbuf_grow(buf, st->st_size);
1549 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1550 return -1;
1551 strbuf_setlen(buf, st->st_size);
1552 return 0;
1553 case S_IFREG:
1554 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1555 return error("unable to open or read %s", path);
1556 convert_to_git(path, buf->buf, buf->len, buf, 0);
1557 return 0;
1558 default:
1559 return -1;
1563 static void update_pre_post_images(struct image *preimage,
1564 struct image *postimage,
1565 char *buf,
1566 size_t len)
1568 int i, ctx;
1569 char *new, *old, *fixed;
1570 struct image fixed_preimage;
1573 * Update the preimage with whitespace fixes. Note that we
1574 * are not losing preimage->buf -- apply_one_fragment() will
1575 * free "oldlines".
1577 prepare_image(&fixed_preimage, buf, len, 1);
1578 assert(fixed_preimage.nr == preimage->nr);
1579 for (i = 0; i < preimage->nr; i++)
1580 fixed_preimage.line[i].flag = preimage->line[i].flag;
1581 free(preimage->line_allocated);
1582 *preimage = fixed_preimage;
1585 * Adjust the common context lines in postimage, in place.
1586 * This is possible because whitespace fixing does not make
1587 * the string grow.
1589 new = old = postimage->buf;
1590 fixed = preimage->buf;
1591 for (i = ctx = 0; i < postimage->nr; i++) {
1592 size_t len = postimage->line[i].len;
1593 if (!(postimage->line[i].flag & LINE_COMMON)) {
1594 /* an added line -- no counterparts in preimage */
1595 memmove(new, old, len);
1596 old += len;
1597 new += len;
1598 continue;
1601 /* a common context -- skip it in the original postimage */
1602 old += len;
1604 /* and find the corresponding one in the fixed preimage */
1605 while (ctx < preimage->nr &&
1606 !(preimage->line[ctx].flag & LINE_COMMON)) {
1607 fixed += preimage->line[ctx].len;
1608 ctx++;
1610 if (preimage->nr <= ctx)
1611 die("oops");
1613 /* and copy it in, while fixing the line length */
1614 len = preimage->line[ctx].len;
1615 memcpy(new, fixed, len);
1616 new += len;
1617 fixed += len;
1618 postimage->line[i].len = len;
1619 ctx++;
1622 /* Fix the length of the whole thing */
1623 postimage->len = new - postimage->buf;
1626 static int match_fragment(struct image *img,
1627 struct image *preimage,
1628 struct image *postimage,
1629 unsigned long try,
1630 int try_lno,
1631 unsigned ws_rule,
1632 int match_beginning, int match_end)
1634 int i;
1635 char *fixed_buf, *buf, *orig, *target;
1637 if (preimage->nr + try_lno > img->nr)
1638 return 0;
1640 if (match_beginning && try_lno)
1641 return 0;
1643 if (match_end && preimage->nr + try_lno != img->nr)
1644 return 0;
1646 /* Quick hash check */
1647 for (i = 0; i < preimage->nr; i++)
1648 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1649 return 0;
1652 * Do we have an exact match? If we were told to match
1653 * at the end, size must be exactly at try+fragsize,
1654 * otherwise try+fragsize must be still within the preimage,
1655 * and either case, the old piece should match the preimage
1656 * exactly.
1658 if ((match_end
1659 ? (try + preimage->len == img->len)
1660 : (try + preimage->len <= img->len)) &&
1661 !memcmp(img->buf + try, preimage->buf, preimage->len))
1662 return 1;
1664 if (ws_error_action != correct_ws_error)
1665 return 0;
1668 * The hunk does not apply byte-by-byte, but the hash says
1669 * it might with whitespace fuzz.
1671 fixed_buf = xmalloc(preimage->len + 1);
1672 buf = fixed_buf;
1673 orig = preimage->buf;
1674 target = img->buf + try;
1675 for (i = 0; i < preimage->nr; i++) {
1676 size_t fixlen; /* length after fixing the preimage */
1677 size_t oldlen = preimage->line[i].len;
1678 size_t tgtlen = img->line[try_lno + i].len;
1679 size_t tgtfixlen; /* length after fixing the target line */
1680 char tgtfixbuf[1024], *tgtfix;
1681 int match;
1683 /* Try fixing the line in the preimage */
1684 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1686 /* Try fixing the line in the target */
1687 if (sizeof(tgtfixbuf) < tgtlen)
1688 tgtfix = tgtfixbuf;
1689 else
1690 tgtfix = xmalloc(tgtlen);
1691 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1694 * If they match, either the preimage was based on
1695 * a version before our tree fixed whitespace breakage,
1696 * or we are lacking a whitespace-fix patch the tree
1697 * the preimage was based on already had (i.e. target
1698 * has whitespace breakage, the preimage doesn't).
1699 * In either case, we are fixing the whitespace breakages
1700 * so we might as well take the fix together with their
1701 * real change.
1703 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1705 if (tgtfix != tgtfixbuf)
1706 free(tgtfix);
1707 if (!match)
1708 goto unmatch_exit;
1710 orig += oldlen;
1711 buf += fixlen;
1712 target += tgtlen;
1716 * Yes, the preimage is based on an older version that still
1717 * has whitespace breakages unfixed, and fixing them makes the
1718 * hunk match. Update the context lines in the postimage.
1720 update_pre_post_images(preimage, postimage,
1721 fixed_buf, buf - fixed_buf);
1722 return 1;
1724 unmatch_exit:
1725 free(fixed_buf);
1726 return 0;
1729 static int find_pos(struct image *img,
1730 struct image *preimage,
1731 struct image *postimage,
1732 int line,
1733 unsigned ws_rule,
1734 int match_beginning, int match_end)
1736 int i;
1737 unsigned long backwards, forwards, try;
1738 int backwards_lno, forwards_lno, try_lno;
1740 if (preimage->nr > img->nr)
1741 return -1;
1744 * If match_begining or match_end is specified, there is no
1745 * point starting from a wrong line that will never match and
1746 * wander around and wait for a match at the specified end.
1748 if (match_beginning)
1749 line = 0;
1750 else if (match_end)
1751 line = img->nr - preimage->nr;
1753 if (line > img->nr)
1754 line = img->nr;
1756 try = 0;
1757 for (i = 0; i < line; i++)
1758 try += img->line[i].len;
1761 * There's probably some smart way to do this, but I'll leave
1762 * that to the smart and beautiful people. I'm simple and stupid.
1764 backwards = try;
1765 backwards_lno = line;
1766 forwards = try;
1767 forwards_lno = line;
1768 try_lno = line;
1770 for (i = 0; ; i++) {
1771 if (match_fragment(img, preimage, postimage,
1772 try, try_lno, ws_rule,
1773 match_beginning, match_end))
1774 return try_lno;
1776 again:
1777 if (backwards_lno == 0 && forwards_lno == img->nr)
1778 break;
1780 if (i & 1) {
1781 if (backwards_lno == 0) {
1782 i++;
1783 goto again;
1785 backwards_lno--;
1786 backwards -= img->line[backwards_lno].len;
1787 try = backwards;
1788 try_lno = backwards_lno;
1789 } else {
1790 if (forwards_lno == img->nr) {
1791 i++;
1792 goto again;
1794 forwards += img->line[forwards_lno].len;
1795 forwards_lno++;
1796 try = forwards;
1797 try_lno = forwards_lno;
1801 return -1;
1804 static void remove_first_line(struct image *img)
1806 img->buf += img->line[0].len;
1807 img->len -= img->line[0].len;
1808 img->line++;
1809 img->nr--;
1812 static void remove_last_line(struct image *img)
1814 img->len -= img->line[--img->nr].len;
1817 static void update_image(struct image *img,
1818 int applied_pos,
1819 struct image *preimage,
1820 struct image *postimage)
1823 * remove the copy of preimage at offset in img
1824 * and replace it with postimage
1826 int i, nr;
1827 size_t remove_count, insert_count, applied_at = 0;
1828 char *result;
1830 for (i = 0; i < applied_pos; i++)
1831 applied_at += img->line[i].len;
1833 remove_count = 0;
1834 for (i = 0; i < preimage->nr; i++)
1835 remove_count += img->line[applied_pos + i].len;
1836 insert_count = postimage->len;
1838 /* Adjust the contents */
1839 result = xmalloc(img->len + insert_count - remove_count + 1);
1840 memcpy(result, img->buf, applied_at);
1841 memcpy(result + applied_at, postimage->buf, postimage->len);
1842 memcpy(result + applied_at + postimage->len,
1843 img->buf + (applied_at + remove_count),
1844 img->len - (applied_at + remove_count));
1845 free(img->buf);
1846 img->buf = result;
1847 img->len += insert_count - remove_count;
1848 result[img->len] = '\0';
1850 /* Adjust the line table */
1851 nr = img->nr + postimage->nr - preimage->nr;
1852 if (preimage->nr < postimage->nr) {
1854 * NOTE: this knows that we never call remove_first_line()
1855 * on anything other than pre/post image.
1857 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1858 img->line_allocated = img->line;
1860 if (preimage->nr != postimage->nr)
1861 memmove(img->line + applied_pos + postimage->nr,
1862 img->line + applied_pos + preimage->nr,
1863 (img->nr - (applied_pos + preimage->nr)) *
1864 sizeof(*img->line));
1865 memcpy(img->line + applied_pos,
1866 postimage->line,
1867 postimage->nr * sizeof(*img->line));
1868 img->nr = nr;
1871 static int apply_one_fragment(struct image *img, struct fragment *frag,
1872 int inaccurate_eof, unsigned ws_rule)
1874 int match_beginning, match_end;
1875 const char *patch = frag->patch;
1876 int size = frag->size;
1877 char *old, *new, *oldlines, *newlines;
1878 int new_blank_lines_at_end = 0;
1879 unsigned long leading, trailing;
1880 int pos, applied_pos;
1881 struct image preimage;
1882 struct image postimage;
1884 memset(&preimage, 0, sizeof(preimage));
1885 memset(&postimage, 0, sizeof(postimage));
1886 oldlines = xmalloc(size);
1887 newlines = xmalloc(size);
1889 old = oldlines;
1890 new = newlines;
1891 while (size > 0) {
1892 char first;
1893 int len = linelen(patch, size);
1894 int plen, added;
1895 int added_blank_line = 0;
1897 if (!len)
1898 break;
1901 * "plen" is how much of the line we should use for
1902 * the actual patch data. Normally we just remove the
1903 * first character on the line, but if the line is
1904 * followed by "\ No newline", then we also remove the
1905 * last one (which is the newline, of course).
1907 plen = len - 1;
1908 if (len < size && patch[len] == '\\')
1909 plen--;
1910 first = *patch;
1911 if (apply_in_reverse) {
1912 if (first == '-')
1913 first = '+';
1914 else if (first == '+')
1915 first = '-';
1918 switch (first) {
1919 case '\n':
1920 /* Newer GNU diff, empty context line */
1921 if (plen < 0)
1922 /* ... followed by '\No newline'; nothing */
1923 break;
1924 *old++ = '\n';
1925 *new++ = '\n';
1926 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1927 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1928 break;
1929 case ' ':
1930 case '-':
1931 memcpy(old, patch + 1, plen);
1932 add_line_info(&preimage, old, plen,
1933 (first == ' ' ? LINE_COMMON : 0));
1934 old += plen;
1935 if (first == '-')
1936 break;
1937 /* Fall-through for ' ' */
1938 case '+':
1939 /* --no-add does not add new lines */
1940 if (first == '+' && no_add)
1941 break;
1943 if (first != '+' ||
1944 !whitespace_error ||
1945 ws_error_action != correct_ws_error) {
1946 memcpy(new, patch + 1, plen);
1947 added = plen;
1949 else {
1950 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1952 add_line_info(&postimage, new, added,
1953 (first == '+' ? 0 : LINE_COMMON));
1954 new += added;
1955 if (first == '+' &&
1956 added == 1 && new[-1] == '\n')
1957 added_blank_line = 1;
1958 break;
1959 case '@': case '\\':
1960 /* Ignore it, we already handled it */
1961 break;
1962 default:
1963 if (apply_verbosely)
1964 error("invalid start of line: '%c'", first);
1965 return -1;
1967 if (added_blank_line)
1968 new_blank_lines_at_end++;
1969 else
1970 new_blank_lines_at_end = 0;
1971 patch += len;
1972 size -= len;
1974 if (inaccurate_eof &&
1975 old > oldlines && old[-1] == '\n' &&
1976 new > newlines && new[-1] == '\n') {
1977 old--;
1978 new--;
1981 leading = frag->leading;
1982 trailing = frag->trailing;
1985 * A hunk to change lines at the beginning would begin with
1986 * @@ -1,L +N,M @@
1988 * And a hunk to add to an empty file would begin with
1989 * @@ -0,0 +N,M @@
1991 * In other words, a hunk that is (frag->oldpos <= 1) with or
1992 * without leading context must match at the beginning.
1994 match_beginning = frag->oldpos <= 1;
1997 * A hunk without trailing lines must match at the end.
1998 * However, we simply cannot tell if a hunk must match end
1999 * from the lack of trailing lines if the patch was generated
2000 * with unidiff without any context.
2002 match_end = !unidiff_zero && !trailing;
2004 pos = frag->newpos ? (frag->newpos - 1) : 0;
2005 preimage.buf = oldlines;
2006 preimage.len = old - oldlines;
2007 postimage.buf = newlines;
2008 postimage.len = new - newlines;
2009 preimage.line = preimage.line_allocated;
2010 postimage.line = postimage.line_allocated;
2012 for (;;) {
2014 applied_pos = find_pos(img, &preimage, &postimage, pos,
2015 ws_rule, match_beginning, match_end);
2017 if (applied_pos >= 0)
2018 break;
2020 /* Am I at my context limits? */
2021 if ((leading <= p_context) && (trailing <= p_context))
2022 break;
2023 if (match_beginning || match_end) {
2024 match_beginning = match_end = 0;
2025 continue;
2029 * Reduce the number of context lines; reduce both
2030 * leading and trailing if they are equal otherwise
2031 * just reduce the larger context.
2033 if (leading >= trailing) {
2034 remove_first_line(&preimage);
2035 remove_first_line(&postimage);
2036 pos--;
2037 leading--;
2039 if (trailing > leading) {
2040 remove_last_line(&preimage);
2041 remove_last_line(&postimage);
2042 trailing--;
2046 if (applied_pos >= 0) {
2047 if (ws_error_action == correct_ws_error &&
2048 new_blank_lines_at_end &&
2049 postimage.nr + applied_pos == img->nr) {
2051 * If the patch application adds blank lines
2052 * at the end, and if the patch applies at the
2053 * end of the image, remove those added blank
2054 * lines.
2056 while (new_blank_lines_at_end--)
2057 remove_last_line(&postimage);
2061 * Warn if it was necessary to reduce the number
2062 * of context lines.
2064 if ((leading != frag->leading) ||
2065 (trailing != frag->trailing))
2066 fprintf(stderr, "Context reduced to (%ld/%ld)"
2067 " to apply fragment at %d\n",
2068 leading, trailing, applied_pos+1);
2069 update_image(img, applied_pos, &preimage, &postimage);
2070 } else {
2071 if (apply_verbosely)
2072 error("while searching for:\n%.*s",
2073 (int)(old - oldlines), oldlines);
2076 free(oldlines);
2077 free(newlines);
2078 free(preimage.line_allocated);
2079 free(postimage.line_allocated);
2081 return (applied_pos < 0);
2084 static int apply_binary_fragment(struct image *img, struct patch *patch)
2086 struct fragment *fragment = patch->fragments;
2087 unsigned long len;
2088 void *dst;
2090 /* Binary patch is irreversible without the optional second hunk */
2091 if (apply_in_reverse) {
2092 if (!fragment->next)
2093 return error("cannot reverse-apply a binary patch "
2094 "without the reverse hunk to '%s'",
2095 patch->new_name
2096 ? patch->new_name : patch->old_name);
2097 fragment = fragment->next;
2099 switch (fragment->binary_patch_method) {
2100 case BINARY_DELTA_DEFLATED:
2101 dst = patch_delta(img->buf, img->len, fragment->patch,
2102 fragment->size, &len);
2103 if (!dst)
2104 return -1;
2105 clear_image(img);
2106 img->buf = dst;
2107 img->len = len;
2108 return 0;
2109 case BINARY_LITERAL_DEFLATED:
2110 clear_image(img);
2111 img->len = fragment->size;
2112 img->buf = xmalloc(img->len+1);
2113 memcpy(img->buf, fragment->patch, img->len);
2114 img->buf[img->len] = '\0';
2115 return 0;
2117 return -1;
2120 static int apply_binary(struct image *img, struct patch *patch)
2122 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2123 unsigned char sha1[20];
2126 * For safety, we require patch index line to contain
2127 * full 40-byte textual SHA1 for old and new, at least for now.
2129 if (strlen(patch->old_sha1_prefix) != 40 ||
2130 strlen(patch->new_sha1_prefix) != 40 ||
2131 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2132 get_sha1_hex(patch->new_sha1_prefix, sha1))
2133 return error("cannot apply binary patch to '%s' "
2134 "without full index line", name);
2136 if (patch->old_name) {
2138 * See if the old one matches what the patch
2139 * applies to.
2141 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2142 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2143 return error("the patch applies to '%s' (%s), "
2144 "which does not match the "
2145 "current contents.",
2146 name, sha1_to_hex(sha1));
2148 else {
2149 /* Otherwise, the old one must be empty. */
2150 if (img->len)
2151 return error("the patch applies to an empty "
2152 "'%s' but it is not empty", name);
2155 get_sha1_hex(patch->new_sha1_prefix, sha1);
2156 if (is_null_sha1(sha1)) {
2157 clear_image(img);
2158 return 0; /* deletion patch */
2161 if (has_sha1_file(sha1)) {
2162 /* We already have the postimage */
2163 enum object_type type;
2164 unsigned long size;
2165 char *result;
2167 result = read_sha1_file(sha1, &type, &size);
2168 if (!result)
2169 return error("the necessary postimage %s for "
2170 "'%s' cannot be read",
2171 patch->new_sha1_prefix, name);
2172 clear_image(img);
2173 img->buf = result;
2174 img->len = size;
2175 } else {
2177 * We have verified buf matches the preimage;
2178 * apply the patch data to it, which is stored
2179 * in the patch->fragments->{patch,size}.
2181 if (apply_binary_fragment(img, patch))
2182 return error("binary patch does not apply to '%s'",
2183 name);
2185 /* verify that the result matches */
2186 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2187 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2188 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2189 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2192 return 0;
2195 static int apply_fragments(struct image *img, struct patch *patch)
2197 struct fragment *frag = patch->fragments;
2198 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2199 unsigned ws_rule = patch->ws_rule;
2200 unsigned inaccurate_eof = patch->inaccurate_eof;
2202 if (patch->is_binary)
2203 return apply_binary(img, patch);
2205 while (frag) {
2206 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2207 error("patch failed: %s:%ld", name, frag->oldpos);
2208 if (!apply_with_reject)
2209 return -1;
2210 frag->rejected = 1;
2212 frag = frag->next;
2214 return 0;
2217 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2219 if (!ce)
2220 return 0;
2222 if (S_ISGITLINK(ce->ce_mode)) {
2223 strbuf_grow(buf, 100);
2224 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2225 } else {
2226 enum object_type type;
2227 unsigned long sz;
2228 char *result;
2230 result = read_sha1_file(ce->sha1, &type, &sz);
2231 if (!result)
2232 return -1;
2233 /* XXX read_sha1_file NUL-terminates */
2234 strbuf_attach(buf, result, sz, sz + 1);
2236 return 0;
2239 static struct patch *in_fn_table(const char *name)
2241 struct path_list_item *item;
2243 if (name == NULL)
2244 return NULL;
2246 item = path_list_lookup(name, &fn_table);
2247 if (item != NULL)
2248 return (struct patch *)item->util;
2250 return NULL;
2253 static void add_to_fn_table(struct patch *patch)
2255 struct path_list_item *item;
2258 * Always add new_name unless patch is a deletion
2259 * This should cover the cases for normal diffs,
2260 * file creations and copies
2262 if (patch->new_name != NULL) {
2263 item = path_list_insert(patch->new_name, &fn_table);
2264 item->util = patch;
2268 * store a failure on rename/deletion cases because
2269 * later chunks shouldn't patch old names
2271 if ((patch->new_name == NULL) || (patch->is_rename)) {
2272 item = path_list_insert(patch->old_name, &fn_table);
2273 item->util = (struct patch *) -1;
2277 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2279 struct strbuf buf;
2280 struct image image;
2281 size_t len;
2282 char *img;
2283 struct patch *tpatch;
2285 strbuf_init(&buf, 0);
2287 if ((tpatch = in_fn_table(patch->old_name)) != NULL) {
2288 if (tpatch == (struct patch *) -1) {
2289 return error("patch %s has been renamed/deleted",
2290 patch->old_name);
2292 /* We have a patched copy in memory use that */
2293 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2294 } else if (cached) {
2295 if (read_file_or_gitlink(ce, &buf))
2296 return error("read of %s failed", patch->old_name);
2297 } else if (patch->old_name) {
2298 if (S_ISGITLINK(patch->old_mode)) {
2299 if (ce) {
2300 read_file_or_gitlink(ce, &buf);
2301 } else {
2303 * There is no way to apply subproject
2304 * patch without looking at the index.
2306 patch->fragments = NULL;
2308 } else {
2309 if (read_old_data(st, patch->old_name, &buf))
2310 return error("read of %s failed", patch->old_name);
2314 img = strbuf_detach(&buf, &len);
2315 prepare_image(&image, img, len, !patch->is_binary);
2317 if (apply_fragments(&image, patch) < 0)
2318 return -1; /* note with --reject this succeeds. */
2319 patch->result = image.buf;
2320 patch->resultsize = image.len;
2321 add_to_fn_table(patch);
2322 free(image.line_allocated);
2324 if (0 < patch->is_delete && patch->resultsize)
2325 return error("removal patch leaves file contents");
2327 return 0;
2330 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2332 struct stat nst;
2333 if (!lstat(new_name, &nst)) {
2334 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2335 return 0;
2337 * A leading component of new_name might be a symlink
2338 * that is going to be removed with this patch, but
2339 * still pointing at somewhere that has the path.
2340 * In such a case, path "new_name" does not exist as
2341 * far as git is concerned.
2343 if (has_symlink_leading_path(strlen(new_name), new_name))
2344 return 0;
2346 return error("%s: already exists in working directory", new_name);
2348 else if ((errno != ENOENT) && (errno != ENOTDIR))
2349 return error("%s: %s", new_name, strerror(errno));
2350 return 0;
2353 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2355 if (S_ISGITLINK(ce->ce_mode)) {
2356 if (!S_ISDIR(st->st_mode))
2357 return -1;
2358 return 0;
2360 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2363 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2365 const char *old_name = patch->old_name;
2366 struct patch *tpatch;
2367 int stat_ret = 0;
2368 unsigned st_mode = 0;
2371 * Make sure that we do not have local modifications from the
2372 * index when we are looking at the index. Also make sure
2373 * we have the preimage file to be patched in the work tree,
2374 * unless --cached, which tells git to apply only in the index.
2376 if (!old_name)
2377 return 0;
2379 assert(patch->is_new <= 0);
2380 if ((tpatch = in_fn_table(old_name)) != NULL) {
2381 if (tpatch == (struct patch *) -1) {
2382 return error("%s: has been deleted/renamed", old_name);
2384 st_mode = tpatch->new_mode;
2385 } else if (!cached) {
2386 stat_ret = lstat(old_name, st);
2387 if (stat_ret && errno != ENOENT)
2388 return error("%s: %s", old_name, strerror(errno));
2390 if (check_index && !tpatch) {
2391 int pos = cache_name_pos(old_name, strlen(old_name));
2392 if (pos < 0) {
2393 if (patch->is_new < 0)
2394 goto is_new;
2395 return error("%s: does not exist in index", old_name);
2397 *ce = active_cache[pos];
2398 if (stat_ret < 0) {
2399 struct checkout costate;
2400 /* checkout */
2401 costate.base_dir = "";
2402 costate.base_dir_len = 0;
2403 costate.force = 0;
2404 costate.quiet = 0;
2405 costate.not_new = 0;
2406 costate.refresh_cache = 1;
2407 if (checkout_entry(*ce, &costate, NULL) ||
2408 lstat(old_name, st))
2409 return -1;
2411 if (!cached && verify_index_match(*ce, st))
2412 return error("%s: does not match index", old_name);
2413 if (cached)
2414 st_mode = (*ce)->ce_mode;
2415 } else if (stat_ret < 0) {
2416 if (patch->is_new < 0)
2417 goto is_new;
2418 return error("%s: %s", old_name, strerror(errno));
2421 if (!cached)
2422 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2424 if (patch->is_new < 0)
2425 patch->is_new = 0;
2426 if (!patch->old_mode)
2427 patch->old_mode = st_mode;
2428 if ((st_mode ^ patch->old_mode) & S_IFMT)
2429 return error("%s: wrong type", old_name);
2430 if (st_mode != patch->old_mode)
2431 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2432 old_name, st_mode, patch->old_mode);
2433 return 0;
2435 is_new:
2436 patch->is_new = 1;
2437 patch->is_delete = 0;
2438 patch->old_name = NULL;
2439 return 0;
2442 static int check_patch(struct patch *patch)
2444 struct stat st;
2445 const char *old_name = patch->old_name;
2446 const char *new_name = patch->new_name;
2447 const char *name = old_name ? old_name : new_name;
2448 struct cache_entry *ce = NULL;
2449 int ok_if_exists;
2450 int status;
2452 patch->rejected = 1; /* we will drop this after we succeed */
2454 status = check_preimage(patch, &ce, &st);
2455 if (status)
2456 return status;
2457 old_name = patch->old_name;
2459 if (in_fn_table(new_name) == (struct patch *) -1)
2461 * A type-change diff is always split into a patch to
2462 * delete old, immediately followed by a patch to
2463 * create new (see diff.c::run_diff()); in such a case
2464 * it is Ok that the entry to be deleted by the
2465 * previous patch is still in the working tree and in
2466 * the index.
2468 ok_if_exists = 1;
2469 else
2470 ok_if_exists = 0;
2472 if (new_name &&
2473 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2474 if (check_index &&
2475 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2476 !ok_if_exists)
2477 return error("%s: already exists in index", new_name);
2478 if (!cached) {
2479 int err = check_to_create_blob(new_name, ok_if_exists);
2480 if (err)
2481 return err;
2483 if (!patch->new_mode) {
2484 if (0 < patch->is_new)
2485 patch->new_mode = S_IFREG | 0644;
2486 else
2487 patch->new_mode = patch->old_mode;
2491 if (new_name && old_name) {
2492 int same = !strcmp(old_name, new_name);
2493 if (!patch->new_mode)
2494 patch->new_mode = patch->old_mode;
2495 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2496 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2497 patch->new_mode, new_name, patch->old_mode,
2498 same ? "" : " of ", same ? "" : old_name);
2501 if (apply_data(patch, &st, ce) < 0)
2502 return error("%s: patch does not apply", name);
2503 patch->rejected = 0;
2504 return 0;
2507 static int check_patch_list(struct patch *patch)
2509 int err = 0;
2511 while (patch) {
2512 if (apply_verbosely)
2513 say_patch_name(stderr,
2514 "Checking patch ", patch, "...\n");
2515 err |= check_patch(patch);
2516 patch = patch->next;
2518 return err;
2521 /* This function tries to read the sha1 from the current index */
2522 static int get_current_sha1(const char *path, unsigned char *sha1)
2524 int pos;
2526 if (read_cache() < 0)
2527 return -1;
2528 pos = cache_name_pos(path, strlen(path));
2529 if (pos < 0)
2530 return -1;
2531 hashcpy(sha1, active_cache[pos]->sha1);
2532 return 0;
2535 /* Build an index that contains the just the files needed for a 3way merge */
2536 static void build_fake_ancestor(struct patch *list, const char *filename)
2538 struct patch *patch;
2539 struct index_state result = { 0 };
2540 int fd;
2542 /* Once we start supporting the reverse patch, it may be
2543 * worth showing the new sha1 prefix, but until then...
2545 for (patch = list; patch; patch = patch->next) {
2546 const unsigned char *sha1_ptr;
2547 unsigned char sha1[20];
2548 struct cache_entry *ce;
2549 const char *name;
2551 name = patch->old_name ? patch->old_name : patch->new_name;
2552 if (0 < patch->is_new)
2553 continue;
2554 else if (get_sha1(patch->old_sha1_prefix, sha1))
2555 /* git diff has no index line for mode/type changes */
2556 if (!patch->lines_added && !patch->lines_deleted) {
2557 if (get_current_sha1(patch->new_name, sha1) ||
2558 get_current_sha1(patch->old_name, sha1))
2559 die("mode change for %s, which is not "
2560 "in current HEAD", name);
2561 sha1_ptr = sha1;
2562 } else
2563 die("sha1 information is lacking or useless "
2564 "(%s).", name);
2565 else
2566 sha1_ptr = sha1;
2568 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2569 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2570 die ("Could not add %s to temporary index", name);
2573 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2574 if (fd < 0 || write_index(&result, fd) || close(fd))
2575 die ("Could not write temporary index to %s", filename);
2577 discard_index(&result);
2580 static void stat_patch_list(struct patch *patch)
2582 int files, adds, dels;
2584 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2585 files++;
2586 adds += patch->lines_added;
2587 dels += patch->lines_deleted;
2588 show_stats(patch);
2591 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2594 static void numstat_patch_list(struct patch *patch)
2596 for ( ; patch; patch = patch->next) {
2597 const char *name;
2598 name = patch->new_name ? patch->new_name : patch->old_name;
2599 if (patch->is_binary)
2600 printf("-\t-\t");
2601 else
2602 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2603 write_name_quoted(name, stdout, line_termination);
2607 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2609 if (mode)
2610 printf(" %s mode %06o %s\n", newdelete, mode, name);
2611 else
2612 printf(" %s %s\n", newdelete, name);
2615 static void show_mode_change(struct patch *p, int show_name)
2617 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2618 if (show_name)
2619 printf(" mode change %06o => %06o %s\n",
2620 p->old_mode, p->new_mode, p->new_name);
2621 else
2622 printf(" mode change %06o => %06o\n",
2623 p->old_mode, p->new_mode);
2627 static void show_rename_copy(struct patch *p)
2629 const char *renamecopy = p->is_rename ? "rename" : "copy";
2630 const char *old, *new;
2632 /* Find common prefix */
2633 old = p->old_name;
2634 new = p->new_name;
2635 while (1) {
2636 const char *slash_old, *slash_new;
2637 slash_old = strchr(old, '/');
2638 slash_new = strchr(new, '/');
2639 if (!slash_old ||
2640 !slash_new ||
2641 slash_old - old != slash_new - new ||
2642 memcmp(old, new, slash_new - new))
2643 break;
2644 old = slash_old + 1;
2645 new = slash_new + 1;
2647 /* p->old_name thru old is the common prefix, and old and new
2648 * through the end of names are renames
2650 if (old != p->old_name)
2651 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2652 (int)(old - p->old_name), p->old_name,
2653 old, new, p->score);
2654 else
2655 printf(" %s %s => %s (%d%%)\n", renamecopy,
2656 p->old_name, p->new_name, p->score);
2657 show_mode_change(p, 0);
2660 static void summary_patch_list(struct patch *patch)
2662 struct patch *p;
2664 for (p = patch; p; p = p->next) {
2665 if (p->is_new)
2666 show_file_mode_name("create", p->new_mode, p->new_name);
2667 else if (p->is_delete)
2668 show_file_mode_name("delete", p->old_mode, p->old_name);
2669 else {
2670 if (p->is_rename || p->is_copy)
2671 show_rename_copy(p);
2672 else {
2673 if (p->score) {
2674 printf(" rewrite %s (%d%%)\n",
2675 p->new_name, p->score);
2676 show_mode_change(p, 0);
2678 else
2679 show_mode_change(p, 1);
2685 static void patch_stats(struct patch *patch)
2687 int lines = patch->lines_added + patch->lines_deleted;
2689 if (lines > max_change)
2690 max_change = lines;
2691 if (patch->old_name) {
2692 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2693 if (!len)
2694 len = strlen(patch->old_name);
2695 if (len > max_len)
2696 max_len = len;
2698 if (patch->new_name) {
2699 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2700 if (!len)
2701 len = strlen(patch->new_name);
2702 if (len > max_len)
2703 max_len = len;
2707 static void remove_file(struct patch *patch, int rmdir_empty)
2709 if (update_index) {
2710 if (remove_file_from_cache(patch->old_name) < 0)
2711 die("unable to remove %s from index", patch->old_name);
2713 if (!cached) {
2714 if (S_ISGITLINK(patch->old_mode)) {
2715 if (rmdir(patch->old_name))
2716 warning("unable to remove submodule %s",
2717 patch->old_name);
2718 } else if (!unlink(patch->old_name) && rmdir_empty) {
2719 char *name = xstrdup(patch->old_name);
2720 char *end = strrchr(name, '/');
2721 while (end) {
2722 *end = 0;
2723 if (rmdir(name))
2724 break;
2725 end = strrchr(name, '/');
2727 free(name);
2732 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2734 struct stat st;
2735 struct cache_entry *ce;
2736 int namelen = strlen(path);
2737 unsigned ce_size = cache_entry_size(namelen);
2739 if (!update_index)
2740 return;
2742 ce = xcalloc(1, ce_size);
2743 memcpy(ce->name, path, namelen);
2744 ce->ce_mode = create_ce_mode(mode);
2745 ce->ce_flags = namelen;
2746 if (S_ISGITLINK(mode)) {
2747 const char *s = buf;
2749 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2750 die("corrupt patch for subproject %s", path);
2751 } else {
2752 if (!cached) {
2753 if (lstat(path, &st) < 0)
2754 die("unable to stat newly created file %s",
2755 path);
2756 fill_stat_cache_info(ce, &st);
2758 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2759 die("unable to create backing store for newly created file %s", path);
2761 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2762 die("unable to add cache entry for %s", path);
2765 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2767 int fd;
2768 struct strbuf nbuf;
2770 if (S_ISGITLINK(mode)) {
2771 struct stat st;
2772 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2773 return 0;
2774 return mkdir(path, 0777);
2777 if (has_symlinks && S_ISLNK(mode))
2778 /* Although buf:size is counted string, it also is NUL
2779 * terminated.
2781 return symlink(buf, path);
2783 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2784 if (fd < 0)
2785 return -1;
2787 strbuf_init(&nbuf, 0);
2788 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2789 size = nbuf.len;
2790 buf = nbuf.buf;
2792 write_or_die(fd, buf, size);
2793 strbuf_release(&nbuf);
2795 if (close(fd) < 0)
2796 die("closing file %s: %s", path, strerror(errno));
2797 return 0;
2801 * We optimistically assume that the directories exist,
2802 * which is true 99% of the time anyway. If they don't,
2803 * we create them and try again.
2805 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2807 if (cached)
2808 return;
2809 if (!try_create_file(path, mode, buf, size))
2810 return;
2812 if (errno == ENOENT) {
2813 if (safe_create_leading_directories(path))
2814 return;
2815 if (!try_create_file(path, mode, buf, size))
2816 return;
2819 if (errno == EEXIST || errno == EACCES) {
2820 /* We may be trying to create a file where a directory
2821 * used to be.
2823 struct stat st;
2824 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2825 errno = EEXIST;
2828 if (errno == EEXIST) {
2829 unsigned int nr = getpid();
2831 for (;;) {
2832 const char *newpath;
2833 newpath = mkpath("%s~%u", path, nr);
2834 if (!try_create_file(newpath, mode, buf, size)) {
2835 if (!rename(newpath, path))
2836 return;
2837 unlink(newpath);
2838 break;
2840 if (errno != EEXIST)
2841 break;
2842 ++nr;
2845 die("unable to write file %s mode %o", path, mode);
2848 static void create_file(struct patch *patch)
2850 char *path = patch->new_name;
2851 unsigned mode = patch->new_mode;
2852 unsigned long size = patch->resultsize;
2853 char *buf = patch->result;
2855 if (!mode)
2856 mode = S_IFREG | 0644;
2857 create_one_file(path, mode, buf, size);
2858 add_index_file(path, mode, buf, size);
2861 /* phase zero is to remove, phase one is to create */
2862 static void write_out_one_result(struct patch *patch, int phase)
2864 if (patch->is_delete > 0) {
2865 if (phase == 0)
2866 remove_file(patch, 1);
2867 return;
2869 if (patch->is_new > 0 || patch->is_copy) {
2870 if (phase == 1)
2871 create_file(patch);
2872 return;
2875 * Rename or modification boils down to the same
2876 * thing: remove the old, write the new
2878 if (phase == 0)
2879 remove_file(patch, patch->is_rename);
2880 if (phase == 1)
2881 create_file(patch);
2884 static int write_out_one_reject(struct patch *patch)
2886 FILE *rej;
2887 char namebuf[PATH_MAX];
2888 struct fragment *frag;
2889 int cnt = 0;
2891 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2892 if (!frag->rejected)
2893 continue;
2894 cnt++;
2897 if (!cnt) {
2898 if (apply_verbosely)
2899 say_patch_name(stderr,
2900 "Applied patch ", patch, " cleanly.\n");
2901 return 0;
2904 /* This should not happen, because a removal patch that leaves
2905 * contents are marked "rejected" at the patch level.
2907 if (!patch->new_name)
2908 die("internal error");
2910 /* Say this even without --verbose */
2911 say_patch_name(stderr, "Applying patch ", patch, " with");
2912 fprintf(stderr, " %d rejects...\n", cnt);
2914 cnt = strlen(patch->new_name);
2915 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2916 cnt = ARRAY_SIZE(namebuf) - 5;
2917 fprintf(stderr,
2918 "warning: truncating .rej filename to %.*s.rej",
2919 cnt - 1, patch->new_name);
2921 memcpy(namebuf, patch->new_name, cnt);
2922 memcpy(namebuf + cnt, ".rej", 5);
2924 rej = fopen(namebuf, "w");
2925 if (!rej)
2926 return error("cannot open %s: %s", namebuf, strerror(errno));
2928 /* Normal git tools never deal with .rej, so do not pretend
2929 * this is a git patch by saying --git nor give extended
2930 * headers. While at it, maybe please "kompare" that wants
2931 * the trailing TAB and some garbage at the end of line ;-).
2933 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2934 patch->new_name, patch->new_name);
2935 for (cnt = 1, frag = patch->fragments;
2936 frag;
2937 cnt++, frag = frag->next) {
2938 if (!frag->rejected) {
2939 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2940 continue;
2942 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2943 fprintf(rej, "%.*s", frag->size, frag->patch);
2944 if (frag->patch[frag->size-1] != '\n')
2945 fputc('\n', rej);
2947 fclose(rej);
2948 return -1;
2951 static int write_out_results(struct patch *list, int skipped_patch)
2953 int phase;
2954 int errs = 0;
2955 struct patch *l;
2957 if (!list && !skipped_patch)
2958 return error("No changes");
2960 for (phase = 0; phase < 2; phase++) {
2961 l = list;
2962 while (l) {
2963 if (l->rejected)
2964 errs = 1;
2965 else {
2966 write_out_one_result(l, phase);
2967 if (phase == 1 && write_out_one_reject(l))
2968 errs = 1;
2970 l = l->next;
2973 return errs;
2976 static struct lock_file lock_file;
2978 static struct excludes {
2979 struct excludes *next;
2980 const char *path;
2981 } *excludes;
2983 static int use_patch(struct patch *p)
2985 const char *pathname = p->new_name ? p->new_name : p->old_name;
2986 struct excludes *x = excludes;
2987 while (x) {
2988 if (fnmatch(x->path, pathname, 0) == 0)
2989 return 0;
2990 x = x->next;
2992 if (0 < prefix_length) {
2993 int pathlen = strlen(pathname);
2994 if (pathlen <= prefix_length ||
2995 memcmp(prefix, pathname, prefix_length))
2996 return 0;
2998 return 1;
3001 static void prefix_one(char **name)
3003 char *old_name = *name;
3004 if (!old_name)
3005 return;
3006 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3007 free(old_name);
3010 static void prefix_patches(struct patch *p)
3012 if (!prefix || p->is_toplevel_relative)
3013 return;
3014 for ( ; p; p = p->next) {
3015 if (p->new_name == p->old_name) {
3016 char *prefixed = p->new_name;
3017 prefix_one(&prefixed);
3018 p->new_name = p->old_name = prefixed;
3020 else {
3021 prefix_one(&p->new_name);
3022 prefix_one(&p->old_name);
3027 #define INACCURATE_EOF (1<<0)
3028 #define RECOUNT (1<<1)
3030 static int apply_patch(int fd, const char *filename, int options)
3032 size_t offset;
3033 struct strbuf buf;
3034 struct patch *list = NULL, **listp = &list;
3035 int skipped_patch = 0;
3037 /* FIXME - memory leak when using multiple patch files as inputs */
3038 memset(&fn_table, 0, sizeof(struct path_list));
3039 strbuf_init(&buf, 0);
3040 patch_input_file = filename;
3041 read_patch_file(&buf, fd);
3042 offset = 0;
3043 while (offset < buf.len) {
3044 struct patch *patch;
3045 int nr;
3047 patch = xcalloc(1, sizeof(*patch));
3048 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3049 patch->recount = !!(options & RECOUNT);
3050 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3051 if (nr < 0)
3052 break;
3053 if (apply_in_reverse)
3054 reverse_patches(patch);
3055 if (prefix)
3056 prefix_patches(patch);
3057 if (use_patch(patch)) {
3058 patch_stats(patch);
3059 *listp = patch;
3060 listp = &patch->next;
3062 else {
3063 /* perhaps free it a bit better? */
3064 free(patch);
3065 skipped_patch++;
3067 offset += nr;
3070 if (whitespace_error && (ws_error_action == die_on_ws_error))
3071 apply = 0;
3073 update_index = check_index && apply;
3074 if (update_index && newfd < 0)
3075 newfd = hold_locked_index(&lock_file, 1);
3077 if (check_index) {
3078 if (read_cache() < 0)
3079 die("unable to read index file");
3082 if ((check || apply) &&
3083 check_patch_list(list) < 0 &&
3084 !apply_with_reject)
3085 exit(1);
3087 if (apply && write_out_results(list, skipped_patch))
3088 exit(1);
3090 if (fake_ancestor)
3091 build_fake_ancestor(list, fake_ancestor);
3093 if (diffstat)
3094 stat_patch_list(list);
3096 if (numstat)
3097 numstat_patch_list(list);
3099 if (summary)
3100 summary_patch_list(list);
3102 strbuf_release(&buf);
3103 return 0;
3106 static int git_apply_config(const char *var, const char *value, void *cb)
3108 if (!strcmp(var, "apply.whitespace"))
3109 return git_config_string(&apply_default_whitespace, var, value);
3110 return git_default_config(var, value, cb);
3114 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3116 int i;
3117 int read_stdin = 1;
3118 int options = 0;
3119 int errs = 0;
3120 int is_not_gitdir;
3122 const char *whitespace_option = NULL;
3124 prefix = setup_git_directory_gently(&is_not_gitdir);
3125 prefix_length = prefix ? strlen(prefix) : 0;
3126 git_config(git_apply_config, NULL);
3127 if (apply_default_whitespace)
3128 parse_whitespace_option(apply_default_whitespace);
3130 for (i = 1; i < argc; i++) {
3131 const char *arg = argv[i];
3132 char *end;
3133 int fd;
3135 if (!strcmp(arg, "-")) {
3136 errs |= apply_patch(0, "<stdin>", options);
3137 read_stdin = 0;
3138 continue;
3140 if (!prefixcmp(arg, "--exclude=")) {
3141 struct excludes *x = xmalloc(sizeof(*x));
3142 x->path = arg + 10;
3143 x->next = excludes;
3144 excludes = x;
3145 continue;
3147 if (!prefixcmp(arg, "-p")) {
3148 p_value = atoi(arg + 2);
3149 p_value_known = 1;
3150 continue;
3152 if (!strcmp(arg, "--no-add")) {
3153 no_add = 1;
3154 continue;
3156 if (!strcmp(arg, "--stat")) {
3157 apply = 0;
3158 diffstat = 1;
3159 continue;
3161 if (!strcmp(arg, "--allow-binary-replacement") ||
3162 !strcmp(arg, "--binary")) {
3163 continue; /* now no-op */
3165 if (!strcmp(arg, "--numstat")) {
3166 apply = 0;
3167 numstat = 1;
3168 continue;
3170 if (!strcmp(arg, "--summary")) {
3171 apply = 0;
3172 summary = 1;
3173 continue;
3175 if (!strcmp(arg, "--check")) {
3176 apply = 0;
3177 check = 1;
3178 continue;
3180 if (!strcmp(arg, "--index")) {
3181 if (is_not_gitdir)
3182 die("--index outside a repository");
3183 check_index = 1;
3184 continue;
3186 if (!strcmp(arg, "--cached")) {
3187 if (is_not_gitdir)
3188 die("--cached outside a repository");
3189 check_index = 1;
3190 cached = 1;
3191 continue;
3193 if (!strcmp(arg, "--apply")) {
3194 apply = 1;
3195 continue;
3197 if (!strcmp(arg, "--build-fake-ancestor")) {
3198 apply = 0;
3199 if (++i >= argc)
3200 die ("need a filename");
3201 fake_ancestor = argv[i];
3202 continue;
3204 if (!strcmp(arg, "-z")) {
3205 line_termination = 0;
3206 continue;
3208 if (!prefixcmp(arg, "-C")) {
3209 p_context = strtoul(arg + 2, &end, 0);
3210 if (*end != '\0')
3211 die("unrecognized context count '%s'", arg + 2);
3212 continue;
3214 if (!prefixcmp(arg, "--whitespace=")) {
3215 whitespace_option = arg + 13;
3216 parse_whitespace_option(arg + 13);
3217 continue;
3219 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3220 apply_in_reverse = 1;
3221 continue;
3223 if (!strcmp(arg, "--unidiff-zero")) {
3224 unidiff_zero = 1;
3225 continue;
3227 if (!strcmp(arg, "--reject")) {
3228 apply = apply_with_reject = apply_verbosely = 1;
3229 continue;
3231 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3232 apply_verbosely = 1;
3233 continue;
3235 if (!strcmp(arg, "--inaccurate-eof")) {
3236 options |= INACCURATE_EOF;
3237 continue;
3239 if (!strcmp(arg, "--recount")) {
3240 options |= RECOUNT;
3241 continue;
3243 if (0 < prefix_length)
3244 arg = prefix_filename(prefix, prefix_length, arg);
3246 fd = open(arg, O_RDONLY);
3247 if (fd < 0)
3248 die("can't open patch '%s': %s", arg, strerror(errno));
3249 read_stdin = 0;
3250 set_default_whitespace_mode(whitespace_option);
3251 errs |= apply_patch(fd, arg, options);
3252 close(fd);
3254 set_default_whitespace_mode(whitespace_option);
3255 if (read_stdin)
3256 errs |= apply_patch(0, "<stdin>", options);
3257 if (whitespace_error) {
3258 if (squelch_whitespace_errors &&
3259 squelch_whitespace_errors < whitespace_error) {
3260 int squelched =
3261 whitespace_error - squelch_whitespace_errors;
3262 fprintf(stderr, "warning: squelched %d "
3263 "whitespace error%s\n",
3264 squelched,
3265 squelched == 1 ? "" : "s");
3267 if (ws_error_action == die_on_ws_error)
3268 die("%d line%s add%s whitespace errors.",
3269 whitespace_error,
3270 whitespace_error == 1 ? "" : "s",
3271 whitespace_error == 1 ? "s" : "");
3272 if (applied_after_fixing_ws && apply)
3273 fprintf(stderr, "warning: %d line%s applied after"
3274 " fixing whitespace errors.\n",
3275 applied_after_fixing_ws,
3276 applied_after_fixing_ws == 1 ? "" : "s");
3277 else if (whitespace_error)
3278 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3279 whitespace_error,
3280 whitespace_error == 1 ? "" : "s",
3281 whitespace_error == 1 ? "s" : "");
3284 if (update_index) {
3285 if (write_cache(newfd, active_cache, active_nr) ||
3286 commit_locked_index(&lock_file))
3287 die("Unable to write new index file");
3290 return !!errs;