fast-import optimization:
[git/dscho.git] / builtin-apply.c
blob61b51315edb040b19e7e17314c66780bab1be346
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static int show_index_info;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
50 static enum whitespace_eol {
51 nowarn_whitespace,
52 warn_on_whitespace,
53 error_on_whitespace,
54 strip_whitespace,
55 } new_whitespace = warn_on_whitespace;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
63 if (!option) {
64 new_whitespace = warn_on_whitespace;
65 return;
67 if (!strcmp(option, "warn")) {
68 new_whitespace = warn_on_whitespace;
69 return;
71 if (!strcmp(option, "nowarn")) {
72 new_whitespace = nowarn_whitespace;
73 return;
75 if (!strcmp(option, "error")) {
76 new_whitespace = error_on_whitespace;
77 return;
79 if (!strcmp(option, "error-all")) {
80 new_whitespace = error_on_whitespace;
81 squelch_whitespace_errors = 0;
82 return;
84 if (!strcmp(option, "strip")) {
85 new_whitespace = strip_whitespace;
86 return;
88 die("unrecognized whitespace option '%s'", option);
91 static void set_default_whitespace_mode(const char *whitespace_option)
93 if (!whitespace_option && !apply_default_whitespace) {
94 new_whitespace = (apply
95 ? warn_on_whitespace
96 : nowarn_whitespace);
101 * For "diff-stat" like behaviour, we keep track of the biggest change
102 * we've seen, and the longest filename. That allows us to do simple
103 * scaling.
105 static int max_change, max_len;
108 * Various "current state", notably line numbers and what
109 * file (and how) we're patching right now.. The "is_xxxx"
110 * things are flags, where -1 means "don't know yet".
112 static int linenr = 1;
115 * This represents one "hunk" from a patch, starting with
116 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
117 * patch text is pointed at by patch, and its byte length
118 * is stored in size. leading and trailing are the number
119 * of context lines.
121 struct fragment {
122 unsigned long leading, trailing;
123 unsigned long oldpos, oldlines;
124 unsigned long newpos, newlines;
125 const char *patch;
126 int size;
127 int rejected;
128 struct fragment *next;
132 * When dealing with a binary patch, we reuse "leading" field
133 * to store the type of the binary hunk, either deflated "delta"
134 * or deflated "literal".
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED 1
138 #define BINARY_LITERAL_DEFLATED 2
140 struct patch {
141 char *new_name, *old_name, *def_name;
142 unsigned int old_mode, new_mode;
143 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
144 int rejected;
145 unsigned long deflate_origlen;
146 int lines_added, lines_deleted;
147 int score;
148 unsigned int is_toplevel_relative:1;
149 unsigned int inaccurate_eof:1;
150 unsigned int is_binary:1;
151 unsigned int is_copy:1;
152 unsigned int is_rename:1;
153 struct fragment *fragments;
154 char *result;
155 unsigned long resultsize;
156 char old_sha1_prefix[41];
157 char new_sha1_prefix[41];
158 struct patch *next;
161 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
163 fputs(pre, output);
164 if (patch->old_name && patch->new_name &&
165 strcmp(patch->old_name, patch->new_name)) {
166 write_name_quoted(NULL, 0, patch->old_name, 1, output);
167 fputs(" => ", output);
168 write_name_quoted(NULL, 0, patch->new_name, 1, output);
170 else {
171 const char *n = patch->new_name;
172 if (!n)
173 n = patch->old_name;
174 write_name_quoted(NULL, 0, n, 1, output);
176 fputs(post, output);
179 #define CHUNKSIZE (8192)
180 #define SLOP (16)
182 static void *read_patch_file(int fd, unsigned long *sizep)
184 struct strbuf buf;
186 strbuf_init(&buf, 0);
187 if (strbuf_read(&buf, fd, 0) < 0)
188 die("git-apply: read returned %s", strerror(errno));
189 *sizep = buf.len;
192 * Make sure that we have some slop in the buffer
193 * so that we can do speculative "memcmp" etc, and
194 * see to it that it is NUL-filled.
196 strbuf_grow(&buf, SLOP);
197 memset(buf.buf + buf.len, 0, SLOP);
198 return strbuf_detach(&buf);
201 static unsigned long linelen(const char *buffer, unsigned long size)
203 unsigned long len = 0;
204 while (size--) {
205 len++;
206 if (*buffer++ == '\n')
207 break;
209 return len;
212 static int is_dev_null(const char *str)
214 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
217 #define TERM_SPACE 1
218 #define TERM_TAB 2
220 static int name_terminate(const char *name, int namelen, int c, int terminate)
222 if (c == ' ' && !(terminate & TERM_SPACE))
223 return 0;
224 if (c == '\t' && !(terminate & TERM_TAB))
225 return 0;
227 return 1;
230 static char *find_name(const char *line, char *def, int p_value, int terminate)
232 int len;
233 const char *start = line;
234 char *name;
236 if (*line == '"') {
237 /* Proposed "new-style" GNU patch/diff format; see
238 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
240 name = unquote_c_style(line, NULL);
241 if (name) {
242 char *cp = name;
243 while (p_value) {
244 cp = strchr(name, '/');
245 if (!cp)
246 break;
247 cp++;
248 p_value--;
250 if (cp) {
251 /* name can later be freed, so we need
252 * to memmove, not just return cp
254 memmove(name, cp, strlen(cp) + 1);
255 free(def);
256 return name;
258 else {
259 free(name);
260 name = NULL;
265 for (;;) {
266 char c = *line;
268 if (isspace(c)) {
269 if (c == '\n')
270 break;
271 if (name_terminate(start, line-start, c, terminate))
272 break;
274 line++;
275 if (c == '/' && !--p_value)
276 start = line;
278 if (!start)
279 return def;
280 len = line - start;
281 if (!len)
282 return def;
285 * Generally we prefer the shorter name, especially
286 * if the other one is just a variation of that with
287 * something else tacked on to the end (ie "file.orig"
288 * or "file~").
290 if (def) {
291 int deflen = strlen(def);
292 if (deflen < len && !strncmp(start, def, deflen))
293 return def;
296 name = xmalloc(len + 1);
297 memcpy(name, start, len);
298 name[len] = 0;
299 free(def);
300 return name;
303 static int count_slashes(const char *cp)
305 int cnt = 0;
306 char ch;
308 while ((ch = *cp++))
309 if (ch == '/')
310 cnt++;
311 return cnt;
315 * Given the string after "--- " or "+++ ", guess the appropriate
316 * p_value for the given patch.
318 static int guess_p_value(const char *nameline)
320 char *name, *cp;
321 int val = -1;
323 if (is_dev_null(nameline))
324 return -1;
325 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
326 if (!name)
327 return -1;
328 cp = strchr(name, '/');
329 if (!cp)
330 val = 0;
331 else if (prefix) {
333 * Does it begin with "a/$our-prefix" and such? Then this is
334 * very likely to apply to our directory.
336 if (!strncmp(name, prefix, prefix_length))
337 val = count_slashes(prefix);
338 else {
339 cp++;
340 if (!strncmp(cp, prefix, prefix_length))
341 val = count_slashes(prefix) + 1;
344 free(name);
345 return val;
349 * Get the name etc info from the --/+++ lines of a traditional patch header
351 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
352 * files, we can happily check the index for a match, but for creating a
353 * new file we should try to match whatever "patch" does. I have no idea.
355 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
357 char *name;
359 first += 4; /* skip "--- " */
360 second += 4; /* skip "+++ " */
361 if (!p_value_known) {
362 int p, q;
363 p = guess_p_value(first);
364 q = guess_p_value(second);
365 if (p < 0) p = q;
366 if (0 <= p && p == q) {
367 p_value = p;
368 p_value_known = 1;
371 if (is_dev_null(first)) {
372 patch->is_new = 1;
373 patch->is_delete = 0;
374 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
375 patch->new_name = name;
376 } else if (is_dev_null(second)) {
377 patch->is_new = 0;
378 patch->is_delete = 1;
379 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
380 patch->old_name = name;
381 } else {
382 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
383 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
384 patch->old_name = patch->new_name = name;
386 if (!name)
387 die("unable to find filename in patch at line %d", linenr);
390 static int gitdiff_hdrend(const char *line, struct patch *patch)
392 return -1;
396 * We're anal about diff header consistency, to make
397 * sure that we don't end up having strange ambiguous
398 * patches floating around.
400 * As a result, gitdiff_{old|new}name() will check
401 * their names against any previous information, just
402 * to make sure..
404 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
406 if (!orig_name && !isnull)
407 return find_name(line, NULL, p_value, TERM_TAB);
409 if (orig_name) {
410 int len;
411 const char *name;
412 char *another;
413 name = orig_name;
414 len = strlen(name);
415 if (isnull)
416 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
417 another = find_name(line, NULL, p_value, TERM_TAB);
418 if (!another || memcmp(another, name, len))
419 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
420 free(another);
421 return orig_name;
423 else {
424 /* expect "/dev/null" */
425 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
426 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
427 return NULL;
431 static int gitdiff_oldname(const char *line, struct patch *patch)
433 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
434 return 0;
437 static int gitdiff_newname(const char *line, struct patch *patch)
439 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
440 return 0;
443 static int gitdiff_oldmode(const char *line, struct patch *patch)
445 patch->old_mode = strtoul(line, NULL, 8);
446 return 0;
449 static int gitdiff_newmode(const char *line, struct patch *patch)
451 patch->new_mode = strtoul(line, NULL, 8);
452 return 0;
455 static int gitdiff_delete(const char *line, struct patch *patch)
457 patch->is_delete = 1;
458 patch->old_name = patch->def_name;
459 return gitdiff_oldmode(line, patch);
462 static int gitdiff_newfile(const char *line, struct patch *patch)
464 patch->is_new = 1;
465 patch->new_name = patch->def_name;
466 return gitdiff_newmode(line, patch);
469 static int gitdiff_copysrc(const char *line, struct patch *patch)
471 patch->is_copy = 1;
472 patch->old_name = find_name(line, NULL, 0, 0);
473 return 0;
476 static int gitdiff_copydst(const char *line, struct patch *patch)
478 patch->is_copy = 1;
479 patch->new_name = find_name(line, NULL, 0, 0);
480 return 0;
483 static int gitdiff_renamesrc(const char *line, struct patch *patch)
485 patch->is_rename = 1;
486 patch->old_name = find_name(line, NULL, 0, 0);
487 return 0;
490 static int gitdiff_renamedst(const char *line, struct patch *patch)
492 patch->is_rename = 1;
493 patch->new_name = find_name(line, NULL, 0, 0);
494 return 0;
497 static int gitdiff_similarity(const char *line, struct patch *patch)
499 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
500 patch->score = 0;
501 return 0;
504 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
506 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
507 patch->score = 0;
508 return 0;
511 static int gitdiff_index(const char *line, struct patch *patch)
513 /* index line is N hexadecimal, "..", N hexadecimal,
514 * and optional space with octal mode.
516 const char *ptr, *eol;
517 int len;
519 ptr = strchr(line, '.');
520 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
521 return 0;
522 len = ptr - line;
523 memcpy(patch->old_sha1_prefix, line, len);
524 patch->old_sha1_prefix[len] = 0;
526 line = ptr + 2;
527 ptr = strchr(line, ' ');
528 eol = strchr(line, '\n');
530 if (!ptr || eol < ptr)
531 ptr = eol;
532 len = ptr - line;
534 if (40 < len)
535 return 0;
536 memcpy(patch->new_sha1_prefix, line, len);
537 patch->new_sha1_prefix[len] = 0;
538 if (*ptr == ' ')
539 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
540 return 0;
544 * This is normal for a diff that doesn't change anything: we'll fall through
545 * into the next diff. Tell the parser to break out.
547 static int gitdiff_unrecognized(const char *line, struct patch *patch)
549 return -1;
552 static const char *stop_at_slash(const char *line, int llen)
554 int i;
556 for (i = 0; i < llen; i++) {
557 int ch = line[i];
558 if (ch == '/')
559 return line + i;
561 return NULL;
564 /* This is to extract the same name that appears on "diff --git"
565 * line. We do not find and return anything if it is a rename
566 * patch, and it is OK because we will find the name elsewhere.
567 * We need to reliably find name only when it is mode-change only,
568 * creation or deletion of an empty file. In any of these cases,
569 * both sides are the same name under a/ and b/ respectively.
571 static char *git_header_name(char *line, int llen)
573 int len;
574 const char *name;
575 const char *second = NULL;
577 line += strlen("diff --git ");
578 llen -= strlen("diff --git ");
580 if (*line == '"') {
581 const char *cp;
582 char *first = unquote_c_style(line, &second);
583 if (!first)
584 return NULL;
586 /* advance to the first slash */
587 cp = stop_at_slash(first, strlen(first));
588 if (!cp || cp == first) {
589 /* we do not accept absolute paths */
590 free_first_and_fail:
591 free(first);
592 return NULL;
594 len = strlen(cp+1);
595 memmove(first, cp+1, len+1); /* including NUL */
597 /* second points at one past closing dq of name.
598 * find the second name.
600 while ((second < line + llen) && isspace(*second))
601 second++;
603 if (line + llen <= second)
604 goto free_first_and_fail;
605 if (*second == '"') {
606 char *sp = unquote_c_style(second, NULL);
607 if (!sp)
608 goto free_first_and_fail;
609 cp = stop_at_slash(sp, strlen(sp));
610 if (!cp || cp == sp) {
611 free_both_and_fail:
612 free(sp);
613 goto free_first_and_fail;
615 /* They must match, otherwise ignore */
616 if (strcmp(cp+1, first))
617 goto free_both_and_fail;
618 free(sp);
619 return first;
622 /* unquoted second */
623 cp = stop_at_slash(second, line + llen - second);
624 if (!cp || cp == second)
625 goto free_first_and_fail;
626 cp++;
627 if (line + llen - cp != len + 1 ||
628 memcmp(first, cp, len))
629 goto free_first_and_fail;
630 return first;
633 /* unquoted first name */
634 name = stop_at_slash(line, llen);
635 if (!name || name == line)
636 return NULL;
638 name++;
640 /* since the first name is unquoted, a dq if exists must be
641 * the beginning of the second name.
643 for (second = name; second < line + llen; second++) {
644 if (*second == '"') {
645 const char *cp = second;
646 const char *np;
647 char *sp = unquote_c_style(second, NULL);
649 if (!sp)
650 return NULL;
651 np = stop_at_slash(sp, strlen(sp));
652 if (!np || np == sp) {
653 free_second_and_fail:
654 free(sp);
655 return NULL;
657 np++;
658 len = strlen(np);
659 if (len < cp - name &&
660 !strncmp(np, name, len) &&
661 isspace(name[len])) {
662 /* Good */
663 memmove(sp, np, len + 1);
664 return sp;
666 goto free_second_and_fail;
671 * Accept a name only if it shows up twice, exactly the same
672 * form.
674 for (len = 0 ; ; len++) {
675 switch (name[len]) {
676 default:
677 continue;
678 case '\n':
679 return NULL;
680 case '\t': case ' ':
681 second = name+len;
682 for (;;) {
683 char c = *second++;
684 if (c == '\n')
685 return NULL;
686 if (c == '/')
687 break;
689 if (second[len] == '\n' && !memcmp(name, second, len)) {
690 char *ret = xmalloc(len + 1);
691 memcpy(ret, name, len);
692 ret[len] = 0;
693 return ret;
697 return NULL;
700 /* Verify that we recognize the lines following a git header */
701 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
703 unsigned long offset;
705 /* A git diff has explicit new/delete information, so we don't guess */
706 patch->is_new = 0;
707 patch->is_delete = 0;
710 * Some things may not have the old name in the
711 * rest of the headers anywhere (pure mode changes,
712 * or removing or adding empty files), so we get
713 * the default name from the header.
715 patch->def_name = git_header_name(line, len);
717 line += len;
718 size -= len;
719 linenr++;
720 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
721 static const struct opentry {
722 const char *str;
723 int (*fn)(const char *, struct patch *);
724 } optable[] = {
725 { "@@ -", gitdiff_hdrend },
726 { "--- ", gitdiff_oldname },
727 { "+++ ", gitdiff_newname },
728 { "old mode ", gitdiff_oldmode },
729 { "new mode ", gitdiff_newmode },
730 { "deleted file mode ", gitdiff_delete },
731 { "new file mode ", gitdiff_newfile },
732 { "copy from ", gitdiff_copysrc },
733 { "copy to ", gitdiff_copydst },
734 { "rename old ", gitdiff_renamesrc },
735 { "rename new ", gitdiff_renamedst },
736 { "rename from ", gitdiff_renamesrc },
737 { "rename to ", gitdiff_renamedst },
738 { "similarity index ", gitdiff_similarity },
739 { "dissimilarity index ", gitdiff_dissimilarity },
740 { "index ", gitdiff_index },
741 { "", gitdiff_unrecognized },
743 int i;
745 len = linelen(line, size);
746 if (!len || line[len-1] != '\n')
747 break;
748 for (i = 0; i < ARRAY_SIZE(optable); i++) {
749 const struct opentry *p = optable + i;
750 int oplen = strlen(p->str);
751 if (len < oplen || memcmp(p->str, line, oplen))
752 continue;
753 if (p->fn(line + oplen, patch) < 0)
754 return offset;
755 break;
759 return offset;
762 static int parse_num(const char *line, unsigned long *p)
764 char *ptr;
766 if (!isdigit(*line))
767 return 0;
768 *p = strtoul(line, &ptr, 10);
769 return ptr - line;
772 static int parse_range(const char *line, int len, int offset, const char *expect,
773 unsigned long *p1, unsigned long *p2)
775 int digits, ex;
777 if (offset < 0 || offset >= len)
778 return -1;
779 line += offset;
780 len -= offset;
782 digits = parse_num(line, p1);
783 if (!digits)
784 return -1;
786 offset += digits;
787 line += digits;
788 len -= digits;
790 *p2 = 1;
791 if (*line == ',') {
792 digits = parse_num(line+1, p2);
793 if (!digits)
794 return -1;
796 offset += digits+1;
797 line += digits+1;
798 len -= digits+1;
801 ex = strlen(expect);
802 if (ex > len)
803 return -1;
804 if (memcmp(line, expect, ex))
805 return -1;
807 return offset + ex;
811 * Parse a unified diff fragment header of the
812 * form "@@ -a,b +c,d @@"
814 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
816 int offset;
818 if (!len || line[len-1] != '\n')
819 return -1;
821 /* Figure out the number of lines in a fragment */
822 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
823 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
825 return offset;
828 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
830 unsigned long offset, len;
832 patch->is_toplevel_relative = 0;
833 patch->is_rename = patch->is_copy = 0;
834 patch->is_new = patch->is_delete = -1;
835 patch->old_mode = patch->new_mode = 0;
836 patch->old_name = patch->new_name = NULL;
837 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
838 unsigned long nextlen;
840 len = linelen(line, size);
841 if (!len)
842 break;
844 /* Testing this early allows us to take a few shortcuts.. */
845 if (len < 6)
846 continue;
849 * Make sure we don't find any unconnected patch fragments.
850 * That's a sign that we didn't find a header, and that a
851 * patch has become corrupted/broken up.
853 if (!memcmp("@@ -", line, 4)) {
854 struct fragment dummy;
855 if (parse_fragment_header(line, len, &dummy) < 0)
856 continue;
857 die("patch fragment without header at line %d: %.*s",
858 linenr, (int)len-1, line);
861 if (size < len + 6)
862 break;
865 * Git patch? It might not have a real patch, just a rename
866 * or mode change, so we handle that specially
868 if (!memcmp("diff --git ", line, 11)) {
869 int git_hdr_len = parse_git_header(line, len, size, patch);
870 if (git_hdr_len <= len)
871 continue;
872 if (!patch->old_name && !patch->new_name) {
873 if (!patch->def_name)
874 die("git diff header lacks filename information (line %d)", linenr);
875 patch->old_name = patch->new_name = patch->def_name;
877 patch->is_toplevel_relative = 1;
878 *hdrsize = git_hdr_len;
879 return offset;
882 /** --- followed by +++ ? */
883 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
884 continue;
887 * We only accept unified patches, so we want it to
888 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
889 * minimum
891 nextlen = linelen(line + len, size - len);
892 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
893 continue;
895 /* Ok, we'll consider it a patch */
896 parse_traditional_patch(line, line+len, patch);
897 *hdrsize = len + nextlen;
898 linenr += 2;
899 return offset;
901 return -1;
904 static void check_whitespace(const char *line, int len)
906 const char *err = "Adds trailing whitespace";
907 int seen_space = 0;
908 int i;
911 * We know len is at least two, since we have a '+' and we
912 * checked that the last character was a '\n' before calling
913 * this function. That is, an addition of an empty line would
914 * check the '+' here. Sneaky...
916 if (isspace(line[len-2]))
917 goto error;
920 * Make sure that there is no space followed by a tab in
921 * indentation.
923 err = "Space in indent is followed by a tab";
924 for (i = 1; i < len; i++) {
925 if (line[i] == '\t') {
926 if (seen_space)
927 goto error;
929 else if (line[i] == ' ')
930 seen_space = 1;
931 else
932 break;
934 return;
936 error:
937 whitespace_error++;
938 if (squelch_whitespace_errors &&
939 squelch_whitespace_errors < whitespace_error)
941 else
942 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
943 err, patch_input_file, linenr, len-2, line+1);
948 * Parse a unified diff. Note that this really needs to parse each
949 * fragment separately, since the only way to know the difference
950 * between a "---" that is part of a patch, and a "---" that starts
951 * the next patch is to look at the line counts..
953 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
955 int added, deleted;
956 int len = linelen(line, size), offset;
957 unsigned long oldlines, newlines;
958 unsigned long leading, trailing;
960 offset = parse_fragment_header(line, len, fragment);
961 if (offset < 0)
962 return -1;
963 oldlines = fragment->oldlines;
964 newlines = fragment->newlines;
965 leading = 0;
966 trailing = 0;
968 /* Parse the thing.. */
969 line += len;
970 size -= len;
971 linenr++;
972 added = deleted = 0;
973 for (offset = len;
974 0 < size;
975 offset += len, size -= len, line += len, linenr++) {
976 if (!oldlines && !newlines)
977 break;
978 len = linelen(line, size);
979 if (!len || line[len-1] != '\n')
980 return -1;
981 switch (*line) {
982 default:
983 return -1;
984 case '\n': /* newer GNU diff, an empty context line */
985 case ' ':
986 oldlines--;
987 newlines--;
988 if (!deleted && !added)
989 leading++;
990 trailing++;
991 break;
992 case '-':
993 if (apply_in_reverse &&
994 new_whitespace != nowarn_whitespace)
995 check_whitespace(line, len);
996 deleted++;
997 oldlines--;
998 trailing = 0;
999 break;
1000 case '+':
1001 if (!apply_in_reverse &&
1002 new_whitespace != nowarn_whitespace)
1003 check_whitespace(line, len);
1004 added++;
1005 newlines--;
1006 trailing = 0;
1007 break;
1009 /* We allow "\ No newline at end of file". Depending
1010 * on locale settings when the patch was produced we
1011 * don't know what this line looks like. The only
1012 * thing we do know is that it begins with "\ ".
1013 * Checking for 12 is just for sanity check -- any
1014 * l10n of "\ No newline..." is at least that long.
1016 case '\\':
1017 if (len < 12 || memcmp(line, "\\ ", 2))
1018 return -1;
1019 break;
1022 if (oldlines || newlines)
1023 return -1;
1024 fragment->leading = leading;
1025 fragment->trailing = trailing;
1027 /* If a fragment ends with an incomplete line, we failed to include
1028 * it in the above loop because we hit oldlines == newlines == 0
1029 * before seeing it.
1031 if (12 < size && !memcmp(line, "\\ ", 2))
1032 offset += linelen(line, size);
1034 patch->lines_added += added;
1035 patch->lines_deleted += deleted;
1037 if (0 < patch->is_new && oldlines)
1038 return error("new file depends on old contents");
1039 if (0 < patch->is_delete && newlines)
1040 return error("deleted file still has contents");
1041 return offset;
1044 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1046 unsigned long offset = 0;
1047 unsigned long oldlines = 0, newlines = 0, context = 0;
1048 struct fragment **fragp = &patch->fragments;
1050 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1051 struct fragment *fragment;
1052 int len;
1054 fragment = xcalloc(1, sizeof(*fragment));
1055 len = parse_fragment(line, size, patch, fragment);
1056 if (len <= 0)
1057 die("corrupt patch at line %d", linenr);
1058 fragment->patch = line;
1059 fragment->size = len;
1060 oldlines += fragment->oldlines;
1061 newlines += fragment->newlines;
1062 context += fragment->leading + fragment->trailing;
1064 *fragp = fragment;
1065 fragp = &fragment->next;
1067 offset += len;
1068 line += len;
1069 size -= len;
1073 * If something was removed (i.e. we have old-lines) it cannot
1074 * be creation, and if something was added it cannot be
1075 * deletion. However, the reverse is not true; --unified=0
1076 * patches that only add are not necessarily creation even
1077 * though they do not have any old lines, and ones that only
1078 * delete are not necessarily deletion.
1080 * Unfortunately, a real creation/deletion patch do _not_ have
1081 * any context line by definition, so we cannot safely tell it
1082 * apart with --unified=0 insanity. At least if the patch has
1083 * more than one hunk it is not creation or deletion.
1085 if (patch->is_new < 0 &&
1086 (oldlines || (patch->fragments && patch->fragments->next)))
1087 patch->is_new = 0;
1088 if (patch->is_delete < 0 &&
1089 (newlines || (patch->fragments && patch->fragments->next)))
1090 patch->is_delete = 0;
1091 if (!unidiff_zero || context) {
1092 /* If the user says the patch is not generated with
1093 * --unified=0, or if we have seen context lines,
1094 * then not having oldlines means the patch is creation,
1095 * and not having newlines means the patch is deletion.
1097 if (patch->is_new < 0 && !oldlines) {
1098 patch->is_new = 1;
1099 patch->old_name = NULL;
1101 if (patch->is_delete < 0 && !newlines) {
1102 patch->is_delete = 1;
1103 patch->new_name = NULL;
1107 if (0 < patch->is_new && oldlines)
1108 die("new file %s depends on old contents", patch->new_name);
1109 if (0 < patch->is_delete && newlines)
1110 die("deleted file %s still has contents", patch->old_name);
1111 if (!patch->is_delete && !newlines && context)
1112 fprintf(stderr, "** warning: file %s becomes empty but "
1113 "is not deleted\n", patch->new_name);
1115 return offset;
1118 static inline int metadata_changes(struct patch *patch)
1120 return patch->is_rename > 0 ||
1121 patch->is_copy > 0 ||
1122 patch->is_new > 0 ||
1123 patch->is_delete ||
1124 (patch->old_mode && patch->new_mode &&
1125 patch->old_mode != patch->new_mode);
1128 static char *inflate_it(const void *data, unsigned long size,
1129 unsigned long inflated_size)
1131 z_stream stream;
1132 void *out;
1133 int st;
1135 memset(&stream, 0, sizeof(stream));
1137 stream.next_in = (unsigned char *)data;
1138 stream.avail_in = size;
1139 stream.next_out = out = xmalloc(inflated_size);
1140 stream.avail_out = inflated_size;
1141 inflateInit(&stream);
1142 st = inflate(&stream, Z_FINISH);
1143 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1144 free(out);
1145 return NULL;
1147 return out;
1150 static struct fragment *parse_binary_hunk(char **buf_p,
1151 unsigned long *sz_p,
1152 int *status_p,
1153 int *used_p)
1155 /* Expect a line that begins with binary patch method ("literal"
1156 * or "delta"), followed by the length of data before deflating.
1157 * a sequence of 'length-byte' followed by base-85 encoded data
1158 * should follow, terminated by a newline.
1160 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1161 * and we would limit the patch line to 66 characters,
1162 * so one line can fit up to 13 groups that would decode
1163 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1164 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1166 int llen, used;
1167 unsigned long size = *sz_p;
1168 char *buffer = *buf_p;
1169 int patch_method;
1170 unsigned long origlen;
1171 char *data = NULL;
1172 int hunk_size = 0;
1173 struct fragment *frag;
1175 llen = linelen(buffer, size);
1176 used = llen;
1178 *status_p = 0;
1180 if (!prefixcmp(buffer, "delta ")) {
1181 patch_method = BINARY_DELTA_DEFLATED;
1182 origlen = strtoul(buffer + 6, NULL, 10);
1184 else if (!prefixcmp(buffer, "literal ")) {
1185 patch_method = BINARY_LITERAL_DEFLATED;
1186 origlen = strtoul(buffer + 8, NULL, 10);
1188 else
1189 return NULL;
1191 linenr++;
1192 buffer += llen;
1193 while (1) {
1194 int byte_length, max_byte_length, newsize;
1195 llen = linelen(buffer, size);
1196 used += llen;
1197 linenr++;
1198 if (llen == 1) {
1199 /* consume the blank line */
1200 buffer++;
1201 size--;
1202 break;
1204 /* Minimum line is "A00000\n" which is 7-byte long,
1205 * and the line length must be multiple of 5 plus 2.
1207 if ((llen < 7) || (llen-2) % 5)
1208 goto corrupt;
1209 max_byte_length = (llen - 2) / 5 * 4;
1210 byte_length = *buffer;
1211 if ('A' <= byte_length && byte_length <= 'Z')
1212 byte_length = byte_length - 'A' + 1;
1213 else if ('a' <= byte_length && byte_length <= 'z')
1214 byte_length = byte_length - 'a' + 27;
1215 else
1216 goto corrupt;
1217 /* if the input length was not multiple of 4, we would
1218 * have filler at the end but the filler should never
1219 * exceed 3 bytes
1221 if (max_byte_length < byte_length ||
1222 byte_length <= max_byte_length - 4)
1223 goto corrupt;
1224 newsize = hunk_size + byte_length;
1225 data = xrealloc(data, newsize);
1226 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1227 goto corrupt;
1228 hunk_size = newsize;
1229 buffer += llen;
1230 size -= llen;
1233 frag = xcalloc(1, sizeof(*frag));
1234 frag->patch = inflate_it(data, hunk_size, origlen);
1235 if (!frag->patch)
1236 goto corrupt;
1237 free(data);
1238 frag->size = origlen;
1239 *buf_p = buffer;
1240 *sz_p = size;
1241 *used_p = used;
1242 frag->binary_patch_method = patch_method;
1243 return frag;
1245 corrupt:
1246 free(data);
1247 *status_p = -1;
1248 error("corrupt binary patch at line %d: %.*s",
1249 linenr-1, llen-1, buffer);
1250 return NULL;
1253 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1255 /* We have read "GIT binary patch\n"; what follows is a line
1256 * that says the patch method (currently, either "literal" or
1257 * "delta") and the length of data before deflating; a
1258 * sequence of 'length-byte' followed by base-85 encoded data
1259 * follows.
1261 * When a binary patch is reversible, there is another binary
1262 * hunk in the same format, starting with patch method (either
1263 * "literal" or "delta") with the length of data, and a sequence
1264 * of length-byte + base-85 encoded data, terminated with another
1265 * empty line. This data, when applied to the postimage, produces
1266 * the preimage.
1268 struct fragment *forward;
1269 struct fragment *reverse;
1270 int status;
1271 int used, used_1;
1273 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1274 if (!forward && !status)
1275 /* there has to be one hunk (forward hunk) */
1276 return error("unrecognized binary patch at line %d", linenr-1);
1277 if (status)
1278 /* otherwise we already gave an error message */
1279 return status;
1281 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1282 if (reverse)
1283 used += used_1;
1284 else if (status) {
1285 /* not having reverse hunk is not an error, but having
1286 * a corrupt reverse hunk is.
1288 free((void*) forward->patch);
1289 free(forward);
1290 return status;
1292 forward->next = reverse;
1293 patch->fragments = forward;
1294 patch->is_binary = 1;
1295 return used;
1298 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1300 int hdrsize, patchsize;
1301 int offset = find_header(buffer, size, &hdrsize, patch);
1303 if (offset < 0)
1304 return offset;
1306 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1308 if (!patchsize) {
1309 static const char *binhdr[] = {
1310 "Binary files ",
1311 "Files ",
1312 NULL,
1314 static const char git_binary[] = "GIT binary patch\n";
1315 int i;
1316 int hd = hdrsize + offset;
1317 unsigned long llen = linelen(buffer + hd, size - hd);
1319 if (llen == sizeof(git_binary) - 1 &&
1320 !memcmp(git_binary, buffer + hd, llen)) {
1321 int used;
1322 linenr++;
1323 used = parse_binary(buffer + hd + llen,
1324 size - hd - llen, patch);
1325 if (used)
1326 patchsize = used + llen;
1327 else
1328 patchsize = 0;
1330 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1331 for (i = 0; binhdr[i]; i++) {
1332 int len = strlen(binhdr[i]);
1333 if (len < size - hd &&
1334 !memcmp(binhdr[i], buffer + hd, len)) {
1335 linenr++;
1336 patch->is_binary = 1;
1337 patchsize = llen;
1338 break;
1343 /* Empty patch cannot be applied if it is a text patch
1344 * without metadata change. A binary patch appears
1345 * empty to us here.
1347 if ((apply || check) &&
1348 (!patch->is_binary && !metadata_changes(patch)))
1349 die("patch with only garbage at line %d", linenr);
1352 return offset + hdrsize + patchsize;
1355 #define swap(a,b) myswap((a),(b),sizeof(a))
1357 #define myswap(a, b, size) do { \
1358 unsigned char mytmp[size]; \
1359 memcpy(mytmp, &a, size); \
1360 memcpy(&a, &b, size); \
1361 memcpy(&b, mytmp, size); \
1362 } while (0)
1364 static void reverse_patches(struct patch *p)
1366 for (; p; p = p->next) {
1367 struct fragment *frag = p->fragments;
1369 swap(p->new_name, p->old_name);
1370 swap(p->new_mode, p->old_mode);
1371 swap(p->is_new, p->is_delete);
1372 swap(p->lines_added, p->lines_deleted);
1373 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1375 for (; frag; frag = frag->next) {
1376 swap(frag->newpos, frag->oldpos);
1377 swap(frag->newlines, frag->oldlines);
1382 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1383 static const char minuses[]= "----------------------------------------------------------------------";
1385 static void show_stats(struct patch *patch)
1387 const char *prefix = "";
1388 char *name = patch->new_name;
1389 char *qname = NULL;
1390 int len, max, add, del, total;
1392 if (!name)
1393 name = patch->old_name;
1395 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1396 qname = xmalloc(len + 1);
1397 quote_c_style(name, qname, NULL, 0);
1398 name = qname;
1402 * "scale" the filename
1404 len = strlen(name);
1405 max = max_len;
1406 if (max > 50)
1407 max = 50;
1408 if (len > max) {
1409 char *slash;
1410 prefix = "...";
1411 max -= 3;
1412 name += len - max;
1413 slash = strchr(name, '/');
1414 if (slash)
1415 name = slash;
1417 len = max;
1420 * scale the add/delete
1422 max = max_change;
1423 if (max + len > 70)
1424 max = 70 - len;
1426 add = patch->lines_added;
1427 del = patch->lines_deleted;
1428 total = add + del;
1430 if (max_change > 0) {
1431 total = (total * max + max_change / 2) / max_change;
1432 add = (add * max + max_change / 2) / max_change;
1433 del = total - add;
1435 if (patch->is_binary)
1436 printf(" %s%-*s | Bin\n", prefix, len, name);
1437 else
1438 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1439 len, name, patch->lines_added + patch->lines_deleted,
1440 add, pluses, del, minuses);
1441 free(qname);
1444 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1446 int fd;
1448 switch (st->st_mode & S_IFMT) {
1449 case S_IFLNK:
1450 strbuf_grow(buf, st->st_size);
1451 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1452 return -1;
1453 strbuf_setlen(buf, st->st_size);
1454 return 0;
1455 case S_IFREG:
1456 fd = open(path, O_RDONLY);
1457 if (fd < 0)
1458 return error("unable to open %s", path);
1459 if (strbuf_read(buf, fd, st->st_size) < 0) {
1460 close(fd);
1461 return -1;
1463 close(fd);
1464 convert_to_git(path, buf->buf, buf->len, buf);
1465 return 0;
1466 default:
1467 return -1;
1471 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1473 int i;
1474 unsigned long start, backwards, forwards;
1476 if (fragsize > size)
1477 return -1;
1479 start = 0;
1480 if (line > 1) {
1481 unsigned long offset = 0;
1482 i = line-1;
1483 while (offset + fragsize <= size) {
1484 if (buf[offset++] == '\n') {
1485 start = offset;
1486 if (!--i)
1487 break;
1492 /* Exact line number? */
1493 if ((start + fragsize <= size) &&
1494 !memcmp(buf + start, fragment, fragsize))
1495 return start;
1498 * There's probably some smart way to do this, but I'll leave
1499 * that to the smart and beautiful people. I'm simple and stupid.
1501 backwards = start;
1502 forwards = start;
1503 for (i = 0; ; i++) {
1504 unsigned long try;
1505 int n;
1507 /* "backward" */
1508 if (i & 1) {
1509 if (!backwards) {
1510 if (forwards + fragsize > size)
1511 break;
1512 continue;
1514 do {
1515 --backwards;
1516 } while (backwards && buf[backwards-1] != '\n');
1517 try = backwards;
1518 } else {
1519 while (forwards + fragsize <= size) {
1520 if (buf[forwards++] == '\n')
1521 break;
1523 try = forwards;
1526 if (try + fragsize > size)
1527 continue;
1528 if (memcmp(buf + try, fragment, fragsize))
1529 continue;
1530 n = (i >> 1)+1;
1531 if (i & 1)
1532 n = -n;
1533 *lines = n;
1534 return try;
1538 * We should start searching forward and backward.
1540 return -1;
1543 static void remove_first_line(const char **rbuf, int *rsize)
1545 const char *buf = *rbuf;
1546 int size = *rsize;
1547 unsigned long offset;
1548 offset = 0;
1549 while (offset <= size) {
1550 if (buf[offset++] == '\n')
1551 break;
1553 *rsize = size - offset;
1554 *rbuf = buf + offset;
1557 static void remove_last_line(const char **rbuf, int *rsize)
1559 const char *buf = *rbuf;
1560 int size = *rsize;
1561 unsigned long offset;
1562 offset = size - 1;
1563 while (offset > 0) {
1564 if (buf[--offset] == '\n')
1565 break;
1567 *rsize = offset + 1;
1570 static int apply_line(char *output, const char *patch, int plen)
1572 /* plen is number of bytes to be copied from patch,
1573 * starting at patch+1 (patch[0] is '+'). Typically
1574 * patch[plen] is '\n', unless this is the incomplete
1575 * last line.
1577 int i;
1578 int add_nl_to_tail = 0;
1579 int fixed = 0;
1580 int last_tab_in_indent = -1;
1581 int last_space_in_indent = -1;
1582 int need_fix_leading_space = 0;
1583 char *buf;
1585 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1586 *patch != '+') {
1587 memcpy(output, patch + 1, plen);
1588 return plen;
1591 if (1 < plen && isspace(patch[plen-1])) {
1592 if (patch[plen] == '\n')
1593 add_nl_to_tail = 1;
1594 plen--;
1595 while (0 < plen && isspace(patch[plen]))
1596 plen--;
1597 fixed = 1;
1600 for (i = 1; i < plen; i++) {
1601 char ch = patch[i];
1602 if (ch == '\t') {
1603 last_tab_in_indent = i;
1604 if (0 <= last_space_in_indent)
1605 need_fix_leading_space = 1;
1607 else if (ch == ' ')
1608 last_space_in_indent = i;
1609 else
1610 break;
1613 buf = output;
1614 if (need_fix_leading_space) {
1615 /* between patch[1..last_tab_in_indent] strip the
1616 * funny spaces, updating them to tab as needed.
1618 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1619 char ch = patch[i];
1620 if (ch != ' ')
1621 *output++ = ch;
1622 else if ((i % 8) == 0)
1623 *output++ = '\t';
1625 fixed = 1;
1626 i = last_tab_in_indent;
1628 else
1629 i = 1;
1631 memcpy(output, patch + i, plen);
1632 if (add_nl_to_tail)
1633 output[plen++] = '\n';
1634 if (fixed)
1635 applied_after_fixing_ws++;
1636 return output + plen - buf;
1639 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag, int inaccurate_eof)
1641 int match_beginning, match_end;
1642 const char *patch = frag->patch;
1643 int offset, size = frag->size;
1644 char *old = xmalloc(size);
1645 char *new = xmalloc(size);
1646 const char *oldlines, *newlines;
1647 int oldsize = 0, newsize = 0;
1648 int new_blank_lines_at_end = 0;
1649 unsigned long leading, trailing;
1650 int pos, lines;
1652 while (size > 0) {
1653 char first;
1654 int len = linelen(patch, size);
1655 int plen;
1656 int added_blank_line = 0;
1658 if (!len)
1659 break;
1662 * "plen" is how much of the line we should use for
1663 * the actual patch data. Normally we just remove the
1664 * first character on the line, but if the line is
1665 * followed by "\ No newline", then we also remove the
1666 * last one (which is the newline, of course).
1668 plen = len-1;
1669 if (len < size && patch[len] == '\\')
1670 plen--;
1671 first = *patch;
1672 if (apply_in_reverse) {
1673 if (first == '-')
1674 first = '+';
1675 else if (first == '+')
1676 first = '-';
1679 switch (first) {
1680 case '\n':
1681 /* Newer GNU diff, empty context line */
1682 if (plen < 0)
1683 /* ... followed by '\No newline'; nothing */
1684 break;
1685 old[oldsize++] = '\n';
1686 new[newsize++] = '\n';
1687 break;
1688 case ' ':
1689 case '-':
1690 memcpy(old + oldsize, patch + 1, plen);
1691 oldsize += plen;
1692 if (first == '-')
1693 break;
1694 /* Fall-through for ' ' */
1695 case '+':
1696 if (first != '+' || !no_add) {
1697 int added = apply_line(new + newsize, patch,
1698 plen);
1699 newsize += added;
1700 if (first == '+' &&
1701 added == 1 && new[newsize-1] == '\n')
1702 added_blank_line = 1;
1704 break;
1705 case '@': case '\\':
1706 /* Ignore it, we already handled it */
1707 break;
1708 default:
1709 if (apply_verbosely)
1710 error("invalid start of line: '%c'", first);
1711 return -1;
1713 if (added_blank_line)
1714 new_blank_lines_at_end++;
1715 else
1716 new_blank_lines_at_end = 0;
1717 patch += len;
1718 size -= len;
1721 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1722 newsize > 0 && new[newsize - 1] == '\n') {
1723 oldsize--;
1724 newsize--;
1727 oldlines = old;
1728 newlines = new;
1729 leading = frag->leading;
1730 trailing = frag->trailing;
1733 * If we don't have any leading/trailing data in the patch,
1734 * we want it to match at the beginning/end of the file.
1736 * But that would break if the patch is generated with
1737 * --unified=0; sane people wouldn't do that to cause us
1738 * trouble, but we try to please not so sane ones as well.
1740 if (unidiff_zero) {
1741 match_beginning = (!leading && !frag->oldpos);
1742 match_end = 0;
1744 else {
1745 match_beginning = !leading && (frag->oldpos == 1);
1746 match_end = !trailing;
1749 lines = 0;
1750 pos = frag->newpos;
1751 for (;;) {
1752 offset = find_offset(buf->buf, buf->len,
1753 oldlines, oldsize, pos, &lines);
1754 if (match_end && offset + oldsize != buf->len)
1755 offset = -1;
1756 if (match_beginning && offset)
1757 offset = -1;
1758 if (offset >= 0) {
1759 if (new_whitespace == strip_whitespace &&
1760 (buf->len - oldsize - offset == 0)) /* end of file? */
1761 newsize -= new_blank_lines_at_end;
1763 /* Warn if it was necessary to reduce the number
1764 * of context lines.
1766 if ((leading != frag->leading) ||
1767 (trailing != frag->trailing))
1768 fprintf(stderr, "Context reduced to (%ld/%ld)"
1769 " to apply fragment at %d\n",
1770 leading, trailing, pos + lines);
1772 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1773 offset = 0;
1774 break;
1777 /* Am I at my context limits? */
1778 if ((leading <= p_context) && (trailing <= p_context))
1779 break;
1780 if (match_beginning || match_end) {
1781 match_beginning = match_end = 0;
1782 continue;
1784 /* Reduce the number of context lines
1785 * Reduce both leading and trailing if they are equal
1786 * otherwise just reduce the larger context.
1788 if (leading >= trailing) {
1789 remove_first_line(&oldlines, &oldsize);
1790 remove_first_line(&newlines, &newsize);
1791 pos--;
1792 leading--;
1794 if (trailing > leading) {
1795 remove_last_line(&oldlines, &oldsize);
1796 remove_last_line(&newlines, &newsize);
1797 trailing--;
1801 if (offset && apply_verbosely)
1802 error("while searching for:\n%.*s", oldsize, oldlines);
1804 free(old);
1805 free(new);
1806 return offset;
1809 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1811 struct fragment *fragment = patch->fragments;
1812 unsigned long len;
1813 void *dst;
1815 /* Binary patch is irreversible without the optional second hunk */
1816 if (apply_in_reverse) {
1817 if (!fragment->next)
1818 return error("cannot reverse-apply a binary patch "
1819 "without the reverse hunk to '%s'",
1820 patch->new_name
1821 ? patch->new_name : patch->old_name);
1822 fragment = fragment->next;
1824 switch (fragment->binary_patch_method) {
1825 case BINARY_DELTA_DEFLATED:
1826 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1827 fragment->size, &len);
1828 if (!dst)
1829 return -1;
1830 /* XXX patch_delta NUL-terminates */
1831 strbuf_attach(buf, dst, len, len + 1);
1832 return 0;
1833 case BINARY_LITERAL_DEFLATED:
1834 strbuf_reset(buf);
1835 strbuf_add(buf, fragment->patch, fragment->size);
1836 return 0;
1838 return -1;
1841 static int apply_binary(struct strbuf *buf, struct patch *patch)
1843 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1844 unsigned char sha1[20];
1846 /* For safety, we require patch index line to contain
1847 * full 40-byte textual SHA1 for old and new, at least for now.
1849 if (strlen(patch->old_sha1_prefix) != 40 ||
1850 strlen(patch->new_sha1_prefix) != 40 ||
1851 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1852 get_sha1_hex(patch->new_sha1_prefix, sha1))
1853 return error("cannot apply binary patch to '%s' "
1854 "without full index line", name);
1856 if (patch->old_name) {
1857 /* See if the old one matches what the patch
1858 * applies to.
1860 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1861 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1862 return error("the patch applies to '%s' (%s), "
1863 "which does not match the "
1864 "current contents.",
1865 name, sha1_to_hex(sha1));
1867 else {
1868 /* Otherwise, the old one must be empty. */
1869 if (buf->len)
1870 return error("the patch applies to an empty "
1871 "'%s' but it is not empty", name);
1874 get_sha1_hex(patch->new_sha1_prefix, sha1);
1875 if (is_null_sha1(sha1)) {
1876 strbuf_release(buf);
1877 return 0; /* deletion patch */
1880 if (has_sha1_file(sha1)) {
1881 /* We already have the postimage */
1882 enum object_type type;
1883 unsigned long size;
1884 char *result;
1886 result = read_sha1_file(sha1, &type, &size);
1887 if (!result)
1888 return error("the necessary postimage %s for "
1889 "'%s' cannot be read",
1890 patch->new_sha1_prefix, name);
1891 /* XXX read_sha1_file NUL-terminates */
1892 strbuf_attach(buf, result, size, size + 1);
1893 } else {
1894 /* We have verified buf matches the preimage;
1895 * apply the patch data to it, which is stored
1896 * in the patch->fragments->{patch,size}.
1898 if (apply_binary_fragment(buf, patch))
1899 return error("binary patch does not apply to '%s'",
1900 name);
1902 /* verify that the result matches */
1903 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1904 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1905 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1906 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1909 return 0;
1912 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1914 struct fragment *frag = patch->fragments;
1915 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1917 if (patch->is_binary)
1918 return apply_binary(buf, patch);
1920 while (frag) {
1921 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1922 error("patch failed: %s:%ld", name, frag->oldpos);
1923 if (!apply_with_reject)
1924 return -1;
1925 frag->rejected = 1;
1927 frag = frag->next;
1929 return 0;
1932 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1934 if (!ce)
1935 return 0;
1937 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1938 strbuf_grow(buf, 100);
1939 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1940 } else {
1941 enum object_type type;
1942 unsigned long sz;
1943 char *result;
1945 result = read_sha1_file(ce->sha1, &type, &sz);
1946 if (!result)
1947 return -1;
1948 /* XXX read_sha1_file NUL-terminates */
1949 strbuf_attach(buf, result, sz, sz + 1);
1951 return 0;
1954 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1956 struct strbuf buf;
1958 strbuf_init(&buf, 0);
1959 if (cached) {
1960 if (read_file_or_gitlink(ce, &buf))
1961 return error("read of %s failed", patch->old_name);
1962 } else if (patch->old_name) {
1963 if (S_ISGITLINK(patch->old_mode)) {
1964 if (ce) {
1965 read_file_or_gitlink(ce, &buf);
1966 } else {
1968 * There is no way to apply subproject
1969 * patch without looking at the index.
1971 patch->fragments = NULL;
1973 } else {
1974 if (read_old_data(st, patch->old_name, &buf))
1975 return error("read of %s failed", patch->old_name);
1979 if (apply_fragments(&buf, patch) < 0)
1980 return -1; /* note with --reject this succeeds. */
1981 patch->result = buf.buf;
1982 patch->resultsize = buf.len;
1984 if (0 < patch->is_delete && patch->resultsize)
1985 return error("removal patch leaves file contents");
1987 return 0;
1990 static int check_to_create_blob(const char *new_name, int ok_if_exists)
1992 struct stat nst;
1993 if (!lstat(new_name, &nst)) {
1994 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1995 return 0;
1997 * A leading component of new_name might be a symlink
1998 * that is going to be removed with this patch, but
1999 * still pointing at somewhere that has the path.
2000 * In such a case, path "new_name" does not exist as
2001 * far as git is concerned.
2003 if (has_symlink_leading_path(new_name, NULL))
2004 return 0;
2006 return error("%s: already exists in working directory", new_name);
2008 else if ((errno != ENOENT) && (errno != ENOTDIR))
2009 return error("%s: %s", new_name, strerror(errno));
2010 return 0;
2013 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2015 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2016 if (!S_ISDIR(st->st_mode))
2017 return -1;
2018 return 0;
2020 return ce_match_stat(ce, st, 1);
2023 static int check_patch(struct patch *patch, struct patch *prev_patch)
2025 struct stat st;
2026 const char *old_name = patch->old_name;
2027 const char *new_name = patch->new_name;
2028 const char *name = old_name ? old_name : new_name;
2029 struct cache_entry *ce = NULL;
2030 int ok_if_exists;
2032 patch->rejected = 1; /* we will drop this after we succeed */
2035 * Make sure that we do not have local modifications from the
2036 * index when we are looking at the index. Also make sure
2037 * we have the preimage file to be patched in the work tree,
2038 * unless --cached, which tells git to apply only in the index.
2040 if (old_name) {
2041 int stat_ret = 0;
2042 unsigned st_mode = 0;
2044 if (!cached)
2045 stat_ret = lstat(old_name, &st);
2046 if (check_index) {
2047 int pos = cache_name_pos(old_name, strlen(old_name));
2048 if (pos < 0)
2049 return error("%s: does not exist in index",
2050 old_name);
2051 ce = active_cache[pos];
2052 if (stat_ret < 0) {
2053 struct checkout costate;
2054 if (errno != ENOENT)
2055 return error("%s: %s", old_name,
2056 strerror(errno));
2057 /* checkout */
2058 costate.base_dir = "";
2059 costate.base_dir_len = 0;
2060 costate.force = 0;
2061 costate.quiet = 0;
2062 costate.not_new = 0;
2063 costate.refresh_cache = 1;
2064 if (checkout_entry(ce,
2065 &costate,
2066 NULL) ||
2067 lstat(old_name, &st))
2068 return -1;
2070 if (!cached && verify_index_match(ce, &st))
2071 return error("%s: does not match index",
2072 old_name);
2073 if (cached)
2074 st_mode = ntohl(ce->ce_mode);
2075 } else if (stat_ret < 0)
2076 return error("%s: %s", old_name, strerror(errno));
2078 if (!cached)
2079 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2081 if (patch->is_new < 0)
2082 patch->is_new = 0;
2083 if (!patch->old_mode)
2084 patch->old_mode = st_mode;
2085 if ((st_mode ^ patch->old_mode) & S_IFMT)
2086 return error("%s: wrong type", old_name);
2087 if (st_mode != patch->old_mode)
2088 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2089 old_name, st_mode, patch->old_mode);
2092 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2093 !strcmp(prev_patch->old_name, new_name))
2094 /* A type-change diff is always split into a patch to
2095 * delete old, immediately followed by a patch to
2096 * create new (see diff.c::run_diff()); in such a case
2097 * it is Ok that the entry to be deleted by the
2098 * previous patch is still in the working tree and in
2099 * the index.
2101 ok_if_exists = 1;
2102 else
2103 ok_if_exists = 0;
2105 if (new_name &&
2106 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2107 if (check_index &&
2108 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2109 !ok_if_exists)
2110 return error("%s: already exists in index", new_name);
2111 if (!cached) {
2112 int err = check_to_create_blob(new_name, ok_if_exists);
2113 if (err)
2114 return err;
2116 if (!patch->new_mode) {
2117 if (0 < patch->is_new)
2118 patch->new_mode = S_IFREG | 0644;
2119 else
2120 patch->new_mode = patch->old_mode;
2124 if (new_name && old_name) {
2125 int same = !strcmp(old_name, new_name);
2126 if (!patch->new_mode)
2127 patch->new_mode = patch->old_mode;
2128 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2129 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2130 patch->new_mode, new_name, patch->old_mode,
2131 same ? "" : " of ", same ? "" : old_name);
2134 if (apply_data(patch, &st, ce) < 0)
2135 return error("%s: patch does not apply", name);
2136 patch->rejected = 0;
2137 return 0;
2140 static int check_patch_list(struct patch *patch)
2142 struct patch *prev_patch = NULL;
2143 int err = 0;
2145 for (prev_patch = NULL; patch ; patch = patch->next) {
2146 if (apply_verbosely)
2147 say_patch_name(stderr,
2148 "Checking patch ", patch, "...\n");
2149 err |= check_patch(patch, prev_patch);
2150 prev_patch = patch;
2152 return err;
2155 static void show_index_list(struct patch *list)
2157 struct patch *patch;
2159 /* Once we start supporting the reverse patch, it may be
2160 * worth showing the new sha1 prefix, but until then...
2162 for (patch = list; patch; patch = patch->next) {
2163 const unsigned char *sha1_ptr;
2164 unsigned char sha1[20];
2165 const char *name;
2167 name = patch->old_name ? patch->old_name : patch->new_name;
2168 if (0 < patch->is_new)
2169 sha1_ptr = null_sha1;
2170 else if (get_sha1(patch->old_sha1_prefix, sha1))
2171 die("sha1 information is lacking or useless (%s).",
2172 name);
2173 else
2174 sha1_ptr = sha1;
2176 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2177 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2178 quote_c_style(name, NULL, stdout, 0);
2179 else
2180 fputs(name, stdout);
2181 putchar(line_termination);
2185 static void stat_patch_list(struct patch *patch)
2187 int files, adds, dels;
2189 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2190 files++;
2191 adds += patch->lines_added;
2192 dels += patch->lines_deleted;
2193 show_stats(patch);
2196 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2199 static void numstat_patch_list(struct patch *patch)
2201 for ( ; patch; patch = patch->next) {
2202 const char *name;
2203 name = patch->new_name ? patch->new_name : patch->old_name;
2204 if (patch->is_binary)
2205 printf("-\t-\t");
2206 else
2207 printf("%d\t%d\t",
2208 patch->lines_added, patch->lines_deleted);
2209 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2210 quote_c_style(name, NULL, stdout, 0);
2211 else
2212 fputs(name, stdout);
2213 putchar(line_termination);
2217 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2219 if (mode)
2220 printf(" %s mode %06o %s\n", newdelete, mode, name);
2221 else
2222 printf(" %s %s\n", newdelete, name);
2225 static void show_mode_change(struct patch *p, int show_name)
2227 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2228 if (show_name)
2229 printf(" mode change %06o => %06o %s\n",
2230 p->old_mode, p->new_mode, p->new_name);
2231 else
2232 printf(" mode change %06o => %06o\n",
2233 p->old_mode, p->new_mode);
2237 static void show_rename_copy(struct patch *p)
2239 const char *renamecopy = p->is_rename ? "rename" : "copy";
2240 const char *old, *new;
2242 /* Find common prefix */
2243 old = p->old_name;
2244 new = p->new_name;
2245 while (1) {
2246 const char *slash_old, *slash_new;
2247 slash_old = strchr(old, '/');
2248 slash_new = strchr(new, '/');
2249 if (!slash_old ||
2250 !slash_new ||
2251 slash_old - old != slash_new - new ||
2252 memcmp(old, new, slash_new - new))
2253 break;
2254 old = slash_old + 1;
2255 new = slash_new + 1;
2257 /* p->old_name thru old is the common prefix, and old and new
2258 * through the end of names are renames
2260 if (old != p->old_name)
2261 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2262 (int)(old - p->old_name), p->old_name,
2263 old, new, p->score);
2264 else
2265 printf(" %s %s => %s (%d%%)\n", renamecopy,
2266 p->old_name, p->new_name, p->score);
2267 show_mode_change(p, 0);
2270 static void summary_patch_list(struct patch *patch)
2272 struct patch *p;
2274 for (p = patch; p; p = p->next) {
2275 if (p->is_new)
2276 show_file_mode_name("create", p->new_mode, p->new_name);
2277 else if (p->is_delete)
2278 show_file_mode_name("delete", p->old_mode, p->old_name);
2279 else {
2280 if (p->is_rename || p->is_copy)
2281 show_rename_copy(p);
2282 else {
2283 if (p->score) {
2284 printf(" rewrite %s (%d%%)\n",
2285 p->new_name, p->score);
2286 show_mode_change(p, 0);
2288 else
2289 show_mode_change(p, 1);
2295 static void patch_stats(struct patch *patch)
2297 int lines = patch->lines_added + patch->lines_deleted;
2299 if (lines > max_change)
2300 max_change = lines;
2301 if (patch->old_name) {
2302 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2303 if (!len)
2304 len = strlen(patch->old_name);
2305 if (len > max_len)
2306 max_len = len;
2308 if (patch->new_name) {
2309 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2310 if (!len)
2311 len = strlen(patch->new_name);
2312 if (len > max_len)
2313 max_len = len;
2317 static void remove_file(struct patch *patch, int rmdir_empty)
2319 if (update_index) {
2320 if (remove_file_from_cache(patch->old_name) < 0)
2321 die("unable to remove %s from index", patch->old_name);
2322 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2324 if (!cached) {
2325 if (S_ISGITLINK(patch->old_mode)) {
2326 if (rmdir(patch->old_name))
2327 warning("unable to remove submodule %s",
2328 patch->old_name);
2329 } else if (!unlink(patch->old_name) && rmdir_empty) {
2330 char *name = xstrdup(patch->old_name);
2331 char *end = strrchr(name, '/');
2332 while (end) {
2333 *end = 0;
2334 if (rmdir(name))
2335 break;
2336 end = strrchr(name, '/');
2338 free(name);
2343 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2345 struct stat st;
2346 struct cache_entry *ce;
2347 int namelen = strlen(path);
2348 unsigned ce_size = cache_entry_size(namelen);
2350 if (!update_index)
2351 return;
2353 ce = xcalloc(1, ce_size);
2354 memcpy(ce->name, path, namelen);
2355 ce->ce_mode = create_ce_mode(mode);
2356 ce->ce_flags = htons(namelen);
2357 if (S_ISGITLINK(mode)) {
2358 const char *s = buf;
2360 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2361 die("corrupt patch for subproject %s", path);
2362 } else {
2363 if (!cached) {
2364 if (lstat(path, &st) < 0)
2365 die("unable to stat newly created file %s",
2366 path);
2367 fill_stat_cache_info(ce, &st);
2369 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2370 die("unable to create backing store for newly created file %s", path);
2372 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2373 die("unable to add cache entry for %s", path);
2376 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2378 int fd;
2379 struct strbuf nbuf;
2381 if (S_ISGITLINK(mode)) {
2382 struct stat st;
2383 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2384 return 0;
2385 return mkdir(path, 0777);
2388 if (has_symlinks && S_ISLNK(mode))
2389 /* Although buf:size is counted string, it also is NUL
2390 * terminated.
2392 return symlink(buf, path);
2394 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2395 if (fd < 0)
2396 return -1;
2398 strbuf_init(&nbuf, 0);
2399 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2400 size = nbuf.len;
2401 buf = nbuf.buf;
2403 write_or_die(fd, buf, size);
2404 strbuf_release(&nbuf);
2406 if (close(fd) < 0)
2407 die("closing file %s: %s", path, strerror(errno));
2408 return 0;
2412 * We optimistically assume that the directories exist,
2413 * which is true 99% of the time anyway. If they don't,
2414 * we create them and try again.
2416 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2418 if (cached)
2419 return;
2420 if (!try_create_file(path, mode, buf, size))
2421 return;
2423 if (errno == ENOENT) {
2424 if (safe_create_leading_directories(path))
2425 return;
2426 if (!try_create_file(path, mode, buf, size))
2427 return;
2430 if (errno == EEXIST || errno == EACCES) {
2431 /* We may be trying to create a file where a directory
2432 * used to be.
2434 struct stat st;
2435 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2436 errno = EEXIST;
2439 if (errno == EEXIST) {
2440 unsigned int nr = getpid();
2442 for (;;) {
2443 const char *newpath;
2444 newpath = mkpath("%s~%u", path, nr);
2445 if (!try_create_file(newpath, mode, buf, size)) {
2446 if (!rename(newpath, path))
2447 return;
2448 unlink(newpath);
2449 break;
2451 if (errno != EEXIST)
2452 break;
2453 ++nr;
2456 die("unable to write file %s mode %o", path, mode);
2459 static void create_file(struct patch *patch)
2461 char *path = patch->new_name;
2462 unsigned mode = patch->new_mode;
2463 unsigned long size = patch->resultsize;
2464 char *buf = patch->result;
2466 if (!mode)
2467 mode = S_IFREG | 0644;
2468 create_one_file(path, mode, buf, size);
2469 add_index_file(path, mode, buf, size);
2470 cache_tree_invalidate_path(active_cache_tree, path);
2473 /* phase zero is to remove, phase one is to create */
2474 static void write_out_one_result(struct patch *patch, int phase)
2476 if (patch->is_delete > 0) {
2477 if (phase == 0)
2478 remove_file(patch, 1);
2479 return;
2481 if (patch->is_new > 0 || patch->is_copy) {
2482 if (phase == 1)
2483 create_file(patch);
2484 return;
2487 * Rename or modification boils down to the same
2488 * thing: remove the old, write the new
2490 if (phase == 0)
2491 remove_file(patch, patch->is_rename);
2492 if (phase == 1)
2493 create_file(patch);
2496 static int write_out_one_reject(struct patch *patch)
2498 FILE *rej;
2499 char namebuf[PATH_MAX];
2500 struct fragment *frag;
2501 int cnt = 0;
2503 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2504 if (!frag->rejected)
2505 continue;
2506 cnt++;
2509 if (!cnt) {
2510 if (apply_verbosely)
2511 say_patch_name(stderr,
2512 "Applied patch ", patch, " cleanly.\n");
2513 return 0;
2516 /* This should not happen, because a removal patch that leaves
2517 * contents are marked "rejected" at the patch level.
2519 if (!patch->new_name)
2520 die("internal error");
2522 /* Say this even without --verbose */
2523 say_patch_name(stderr, "Applying patch ", patch, " with");
2524 fprintf(stderr, " %d rejects...\n", cnt);
2526 cnt = strlen(patch->new_name);
2527 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2528 cnt = ARRAY_SIZE(namebuf) - 5;
2529 fprintf(stderr,
2530 "warning: truncating .rej filename to %.*s.rej",
2531 cnt - 1, patch->new_name);
2533 memcpy(namebuf, patch->new_name, cnt);
2534 memcpy(namebuf + cnt, ".rej", 5);
2536 rej = fopen(namebuf, "w");
2537 if (!rej)
2538 return error("cannot open %s: %s", namebuf, strerror(errno));
2540 /* Normal git tools never deal with .rej, so do not pretend
2541 * this is a git patch by saying --git nor give extended
2542 * headers. While at it, maybe please "kompare" that wants
2543 * the trailing TAB and some garbage at the end of line ;-).
2545 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2546 patch->new_name, patch->new_name);
2547 for (cnt = 1, frag = patch->fragments;
2548 frag;
2549 cnt++, frag = frag->next) {
2550 if (!frag->rejected) {
2551 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2552 continue;
2554 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2555 fprintf(rej, "%.*s", frag->size, frag->patch);
2556 if (frag->patch[frag->size-1] != '\n')
2557 fputc('\n', rej);
2559 fclose(rej);
2560 return -1;
2563 static int write_out_results(struct patch *list, int skipped_patch)
2565 int phase;
2566 int errs = 0;
2567 struct patch *l;
2569 if (!list && !skipped_patch)
2570 return error("No changes");
2572 for (phase = 0; phase < 2; phase++) {
2573 l = list;
2574 while (l) {
2575 if (l->rejected)
2576 errs = 1;
2577 else {
2578 write_out_one_result(l, phase);
2579 if (phase == 1 && write_out_one_reject(l))
2580 errs = 1;
2582 l = l->next;
2585 return errs;
2588 static struct lock_file lock_file;
2590 static struct excludes {
2591 struct excludes *next;
2592 const char *path;
2593 } *excludes;
2595 static int use_patch(struct patch *p)
2597 const char *pathname = p->new_name ? p->new_name : p->old_name;
2598 struct excludes *x = excludes;
2599 while (x) {
2600 if (fnmatch(x->path, pathname, 0) == 0)
2601 return 0;
2602 x = x->next;
2604 if (0 < prefix_length) {
2605 int pathlen = strlen(pathname);
2606 if (pathlen <= prefix_length ||
2607 memcmp(prefix, pathname, prefix_length))
2608 return 0;
2610 return 1;
2613 static void prefix_one(char **name)
2615 char *old_name = *name;
2616 if (!old_name)
2617 return;
2618 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2619 free(old_name);
2622 static void prefix_patches(struct patch *p)
2624 if (!prefix || p->is_toplevel_relative)
2625 return;
2626 for ( ; p; p = p->next) {
2627 if (p->new_name == p->old_name) {
2628 char *prefixed = p->new_name;
2629 prefix_one(&prefixed);
2630 p->new_name = p->old_name = prefixed;
2632 else {
2633 prefix_one(&p->new_name);
2634 prefix_one(&p->old_name);
2639 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2641 unsigned long offset, size;
2642 char *buffer = read_patch_file(fd, &size);
2643 struct patch *list = NULL, **listp = &list;
2644 int skipped_patch = 0;
2646 patch_input_file = filename;
2647 if (!buffer)
2648 return -1;
2649 offset = 0;
2650 while (size > 0) {
2651 struct patch *patch;
2652 int nr;
2654 patch = xcalloc(1, sizeof(*patch));
2655 patch->inaccurate_eof = inaccurate_eof;
2656 nr = parse_chunk(buffer + offset, size, patch);
2657 if (nr < 0)
2658 break;
2659 if (apply_in_reverse)
2660 reverse_patches(patch);
2661 if (prefix)
2662 prefix_patches(patch);
2663 if (use_patch(patch)) {
2664 patch_stats(patch);
2665 *listp = patch;
2666 listp = &patch->next;
2668 else {
2669 /* perhaps free it a bit better? */
2670 free(patch);
2671 skipped_patch++;
2673 offset += nr;
2674 size -= nr;
2677 if (whitespace_error && (new_whitespace == error_on_whitespace))
2678 apply = 0;
2680 update_index = check_index && apply;
2681 if (update_index && newfd < 0)
2682 newfd = hold_locked_index(&lock_file, 1);
2684 if (check_index) {
2685 if (read_cache() < 0)
2686 die("unable to read index file");
2689 if ((check || apply) &&
2690 check_patch_list(list) < 0 &&
2691 !apply_with_reject)
2692 exit(1);
2694 if (apply && write_out_results(list, skipped_patch))
2695 exit(1);
2697 if (show_index_info)
2698 show_index_list(list);
2700 if (diffstat)
2701 stat_patch_list(list);
2703 if (numstat)
2704 numstat_patch_list(list);
2706 if (summary)
2707 summary_patch_list(list);
2709 free(buffer);
2710 return 0;
2713 static int git_apply_config(const char *var, const char *value)
2715 if (!strcmp(var, "apply.whitespace")) {
2716 apply_default_whitespace = xstrdup(value);
2717 return 0;
2719 return git_default_config(var, value);
2723 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2725 int i;
2726 int read_stdin = 1;
2727 int inaccurate_eof = 0;
2728 int errs = 0;
2729 int is_not_gitdir = 0;
2731 const char *whitespace_option = NULL;
2733 prefix = setup_git_directory_gently(&is_not_gitdir);
2734 prefix_length = prefix ? strlen(prefix) : 0;
2735 git_config(git_apply_config);
2736 if (apply_default_whitespace)
2737 parse_whitespace_option(apply_default_whitespace);
2739 for (i = 1; i < argc; i++) {
2740 const char *arg = argv[i];
2741 char *end;
2742 int fd;
2744 if (!strcmp(arg, "-")) {
2745 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2746 read_stdin = 0;
2747 continue;
2749 if (!prefixcmp(arg, "--exclude=")) {
2750 struct excludes *x = xmalloc(sizeof(*x));
2751 x->path = arg + 10;
2752 x->next = excludes;
2753 excludes = x;
2754 continue;
2756 if (!prefixcmp(arg, "-p")) {
2757 p_value = atoi(arg + 2);
2758 p_value_known = 1;
2759 continue;
2761 if (!strcmp(arg, "--no-add")) {
2762 no_add = 1;
2763 continue;
2765 if (!strcmp(arg, "--stat")) {
2766 apply = 0;
2767 diffstat = 1;
2768 continue;
2770 if (!strcmp(arg, "--allow-binary-replacement") ||
2771 !strcmp(arg, "--binary")) {
2772 continue; /* now no-op */
2774 if (!strcmp(arg, "--numstat")) {
2775 apply = 0;
2776 numstat = 1;
2777 continue;
2779 if (!strcmp(arg, "--summary")) {
2780 apply = 0;
2781 summary = 1;
2782 continue;
2784 if (!strcmp(arg, "--check")) {
2785 apply = 0;
2786 check = 1;
2787 continue;
2789 if (!strcmp(arg, "--index")) {
2790 if (is_not_gitdir)
2791 die("--index outside a repository");
2792 check_index = 1;
2793 continue;
2795 if (!strcmp(arg, "--cached")) {
2796 if (is_not_gitdir)
2797 die("--cached outside a repository");
2798 check_index = 1;
2799 cached = 1;
2800 continue;
2802 if (!strcmp(arg, "--apply")) {
2803 apply = 1;
2804 continue;
2806 if (!strcmp(arg, "--index-info")) {
2807 apply = 0;
2808 show_index_info = 1;
2809 continue;
2811 if (!strcmp(arg, "-z")) {
2812 line_termination = 0;
2813 continue;
2815 if (!prefixcmp(arg, "-C")) {
2816 p_context = strtoul(arg + 2, &end, 0);
2817 if (*end != '\0')
2818 die("unrecognized context count '%s'", arg + 2);
2819 continue;
2821 if (!prefixcmp(arg, "--whitespace=")) {
2822 whitespace_option = arg + 13;
2823 parse_whitespace_option(arg + 13);
2824 continue;
2826 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2827 apply_in_reverse = 1;
2828 continue;
2830 if (!strcmp(arg, "--unidiff-zero")) {
2831 unidiff_zero = 1;
2832 continue;
2834 if (!strcmp(arg, "--reject")) {
2835 apply = apply_with_reject = apply_verbosely = 1;
2836 continue;
2838 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2839 apply_verbosely = 1;
2840 continue;
2842 if (!strcmp(arg, "--inaccurate-eof")) {
2843 inaccurate_eof = 1;
2844 continue;
2846 if (0 < prefix_length)
2847 arg = prefix_filename(prefix, prefix_length, arg);
2849 fd = open(arg, O_RDONLY);
2850 if (fd < 0)
2851 usage(apply_usage);
2852 read_stdin = 0;
2853 set_default_whitespace_mode(whitespace_option);
2854 errs |= apply_patch(fd, arg, inaccurate_eof);
2855 close(fd);
2857 set_default_whitespace_mode(whitespace_option);
2858 if (read_stdin)
2859 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2860 if (whitespace_error) {
2861 if (squelch_whitespace_errors &&
2862 squelch_whitespace_errors < whitespace_error) {
2863 int squelched =
2864 whitespace_error - squelch_whitespace_errors;
2865 fprintf(stderr, "warning: squelched %d "
2866 "whitespace error%s\n",
2867 squelched,
2868 squelched == 1 ? "" : "s");
2870 if (new_whitespace == error_on_whitespace)
2871 die("%d line%s add%s whitespace errors.",
2872 whitespace_error,
2873 whitespace_error == 1 ? "" : "s",
2874 whitespace_error == 1 ? "s" : "");
2875 if (applied_after_fixing_ws)
2876 fprintf(stderr, "warning: %d line%s applied after"
2877 " fixing whitespace errors.\n",
2878 applied_after_fixing_ws,
2879 applied_after_fixing_ws == 1 ? "" : "s");
2880 else if (whitespace_error)
2881 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2882 whitespace_error,
2883 whitespace_error == 1 ? "" : "s",
2884 whitespace_error == 1 ? "s" : "");
2887 if (update_index) {
2888 if (write_cache(newfd, active_cache, active_nr) ||
2889 close(newfd) || commit_locked_index(&lock_file))
2890 die("Unable to write new index file");
2893 return !!errs;