1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
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"
16 /*
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.
24 */
25 static const char *prefix;
26 static int prefix_length = -1;
27 static int newfd = -1;
29 static int unidiff_zero;
30 static int p_value = 1;
31 static int p_value_known;
32 static int check_index;
33 static int update_index;
34 static int cached;
35 static int diffstat;
36 static int numstat;
37 static int summary;
38 static int check;
39 static int apply = 1;
40 static int apply_in_reverse;
41 static int apply_with_reject;
42 static int apply_verbosely;
43 static int no_add;
44 static const char *fake_ancestor;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
50 static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55 } ws_error_action = warn_on_ws_error;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
62 {
63 if (!option) {
64 ws_error_action = warn_on_ws_error;
65 return;
66 }
67 if (!strcmp(option, "warn")) {
68 ws_error_action = warn_on_ws_error;
69 return;
70 }
71 if (!strcmp(option, "nowarn")) {
72 ws_error_action = nowarn_ws_error;
73 return;
74 }
75 if (!strcmp(option, "error")) {
76 ws_error_action = die_on_ws_error;
77 return;
78 }
79 if (!strcmp(option, "error-all")) {
80 ws_error_action = die_on_ws_error;
81 squelch_whitespace_errors = 0;
82 return;
83 }
84 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
85 ws_error_action = correct_ws_error;
86 return;
87 }
88 die("unrecognized whitespace option '%s'", option);
89 }
91 static void set_default_whitespace_mode(const char *whitespace_option)
92 {
93 if (!whitespace_option && !apply_default_whitespace)
94 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
95 }
97 /*
98 * For "diff-stat" like behaviour, we keep track of the biggest change
99 * we've seen, and the longest filename. That allows us to do simple
100 * scaling.
101 */
102 static int max_change, max_len;
104 /*
105 * Various "current state", notably line numbers and what
106 * file (and how) we're patching right now.. The "is_xxxx"
107 * things are flags, where -1 means "don't know yet".
108 */
109 static int linenr = 1;
111 /*
112 * This represents one "hunk" from a patch, starting with
113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
114 * patch text is pointed at by patch, and its byte length
115 * is stored in size. leading and trailing are the number
116 * of context lines.
117 */
118 struct fragment {
119 unsigned long leading, trailing;
120 unsigned long oldpos, oldlines;
121 unsigned long newpos, newlines;
122 const char *patch;
123 int size;
124 int rejected;
125 struct fragment *next;
126 };
128 /*
129 * When dealing with a binary patch, we reuse "leading" field
130 * to store the type of the binary hunk, either deflated "delta"
131 * or deflated "literal".
132 */
133 #define binary_patch_method leading
134 #define BINARY_DELTA_DEFLATED 1
135 #define BINARY_LITERAL_DEFLATED 2
137 /*
138 * This represents a "patch" to a file, both metainfo changes
139 * such as creation/deletion, filemode and content changes represented
140 * as a series of fragments.
141 */
142 struct patch {
143 char *new_name, *old_name, *def_name;
144 unsigned int old_mode, new_mode;
145 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
146 int rejected;
147 unsigned long deflate_origlen;
148 int lines_added, lines_deleted;
149 int score;
150 unsigned int is_toplevel_relative:1;
151 unsigned int inaccurate_eof:1;
152 unsigned int is_binary:1;
153 unsigned int is_copy:1;
154 unsigned int is_rename:1;
155 struct fragment *fragments;
156 char *result;
157 size_t resultsize;
158 char old_sha1_prefix[41];
159 char new_sha1_prefix[41];
160 struct patch *next;
161 };
163 static void say_patch_name(FILE *output, const char *pre,
164 struct patch *patch, const char *post)
165 {
166 fputs(pre, output);
167 if (patch->old_name && patch->new_name &&
168 strcmp(patch->old_name, patch->new_name)) {
169 quote_c_style(patch->old_name, NULL, output, 0);
170 fputs(" => ", output);
171 quote_c_style(patch->new_name, NULL, output, 0);
172 } else {
173 const char *n = patch->new_name;
174 if (!n)
175 n = patch->old_name;
176 quote_c_style(n, NULL, output, 0);
177 }
178 fputs(post, output);
179 }
181 #define CHUNKSIZE (8192)
182 #define SLOP (16)
184 static void read_patch_file(struct strbuf *sb, int fd)
185 {
186 if (strbuf_read(sb, fd, 0) < 0)
187 die("git-apply: read returned %s", strerror(errno));
189 /*
190 * Make sure that we have some slop in the buffer
191 * so that we can do speculative "memcmp" etc, and
192 * see to it that it is NUL-filled.
193 */
194 strbuf_grow(sb, SLOP);
195 memset(sb->buf + sb->len, 0, SLOP);
196 }
198 static unsigned long linelen(const char *buffer, unsigned long size)
199 {
200 unsigned long len = 0;
201 while (size--) {
202 len++;
203 if (*buffer++ == '\n')
204 break;
205 }
206 return len;
207 }
209 static int is_dev_null(const char *str)
210 {
211 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
212 }
214 #define TERM_SPACE 1
215 #define TERM_TAB 2
217 static int name_terminate(const char *name, int namelen, int c, int terminate)
218 {
219 if (c == ' ' && !(terminate & TERM_SPACE))
220 return 0;
221 if (c == '\t' && !(terminate & TERM_TAB))
222 return 0;
224 return 1;
225 }
227 static char *find_name(const char *line, char *def, int p_value, int terminate)
228 {
229 int len;
230 const char *start = line;
232 if (*line == '"') {
233 struct strbuf name;
235 /*
236 * Proposed "new-style" GNU patch/diff format; see
237 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
238 */
239 strbuf_init(&name, 0);
240 if (!unquote_c_style(&name, line, NULL)) {
241 char *cp;
243 for (cp = name.buf; p_value; p_value--) {
244 cp = strchr(cp, '/');
245 if (!cp)
246 break;
247 cp++;
248 }
249 if (cp) {
250 /* name can later be freed, so we need
251 * to memmove, not just return cp
252 */
253 strbuf_remove(&name, 0, cp - name.buf);
254 free(def);
255 return strbuf_detach(&name, NULL);
256 }
257 }
258 strbuf_release(&name);
259 }
261 for (;;) {
262 char c = *line;
264 if (isspace(c)) {
265 if (c == '\n')
266 break;
267 if (name_terminate(start, line-start, c, terminate))
268 break;
269 }
270 line++;
271 if (c == '/' && !--p_value)
272 start = line;
273 }
274 if (!start)
275 return def;
276 len = line - start;
277 if (!len)
278 return def;
280 /*
281 * Generally we prefer the shorter name, especially
282 * if the other one is just a variation of that with
283 * something else tacked on to the end (ie "file.orig"
284 * or "file~").
285 */
286 if (def) {
287 int deflen = strlen(def);
288 if (deflen < len && !strncmp(start, def, deflen))
289 return def;
290 free(def);
291 }
293 return xmemdupz(start, len);
294 }
296 static int count_slashes(const char *cp)
297 {
298 int cnt = 0;
299 char ch;
301 while ((ch = *cp++))
302 if (ch == '/')
303 cnt++;
304 return cnt;
305 }
307 /*
308 * Given the string after "--- " or "+++ ", guess the appropriate
309 * p_value for the given patch.
310 */
311 static int guess_p_value(const char *nameline)
312 {
313 char *name, *cp;
314 int val = -1;
316 if (is_dev_null(nameline))
317 return -1;
318 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
319 if (!name)
320 return -1;
321 cp = strchr(name, '/');
322 if (!cp)
323 val = 0;
324 else if (prefix) {
325 /*
326 * Does it begin with "a/$our-prefix" and such? Then this is
327 * very likely to apply to our directory.
328 */
329 if (!strncmp(name, prefix, prefix_length))
330 val = count_slashes(prefix);
331 else {
332 cp++;
333 if (!strncmp(cp, prefix, prefix_length))
334 val = count_slashes(prefix) + 1;
335 }
336 }
337 free(name);
338 return val;
339 }
341 /*
342 * Get the name etc info from the --/+++ lines of a traditional patch header
343 *
344 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
345 * files, we can happily check the index for a match, but for creating a
346 * new file we should try to match whatever "patch" does. I have no idea.
347 */
348 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
349 {
350 char *name;
352 first += 4; /* skip "--- " */
353 second += 4; /* skip "+++ " */
354 if (!p_value_known) {
355 int p, q;
356 p = guess_p_value(first);
357 q = guess_p_value(second);
358 if (p < 0) p = q;
359 if (0 <= p && p == q) {
360 p_value = p;
361 p_value_known = 1;
362 }
363 }
364 if (is_dev_null(first)) {
365 patch->is_new = 1;
366 patch->is_delete = 0;
367 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
368 patch->new_name = name;
369 } else if (is_dev_null(second)) {
370 patch->is_new = 0;
371 patch->is_delete = 1;
372 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
373 patch->old_name = name;
374 } else {
375 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
376 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
377 patch->old_name = patch->new_name = name;
378 }
379 if (!name)
380 die("unable to find filename in patch at line %d", linenr);
381 }
383 static int gitdiff_hdrend(const char *line, struct patch *patch)
384 {
385 return -1;
386 }
388 /*
389 * We're anal about diff header consistency, to make
390 * sure that we don't end up having strange ambiguous
391 * patches floating around.
392 *
393 * As a result, gitdiff_{old|new}name() will check
394 * their names against any previous information, just
395 * to make sure..
396 */
397 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
398 {
399 if (!orig_name && !isnull)
400 return find_name(line, NULL, p_value, TERM_TAB);
402 if (orig_name) {
403 int len;
404 const char *name;
405 char *another;
406 name = orig_name;
407 len = strlen(name);
408 if (isnull)
409 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
410 another = find_name(line, NULL, p_value, TERM_TAB);
411 if (!another || memcmp(another, name, len))
412 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
413 free(another);
414 return orig_name;
415 }
416 else {
417 /* expect "/dev/null" */
418 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
419 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
420 return NULL;
421 }
422 }
424 static int gitdiff_oldname(const char *line, struct patch *patch)
425 {
426 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
427 return 0;
428 }
430 static int gitdiff_newname(const char *line, struct patch *patch)
431 {
432 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
433 return 0;
434 }
436 static int gitdiff_oldmode(const char *line, struct patch *patch)
437 {
438 patch->old_mode = strtoul(line, NULL, 8);
439 return 0;
440 }
442 static int gitdiff_newmode(const char *line, struct patch *patch)
443 {
444 patch->new_mode = strtoul(line, NULL, 8);
445 return 0;
446 }
448 static int gitdiff_delete(const char *line, struct patch *patch)
449 {
450 patch->is_delete = 1;
451 patch->old_name = patch->def_name;
452 return gitdiff_oldmode(line, patch);
453 }
455 static int gitdiff_newfile(const char *line, struct patch *patch)
456 {
457 patch->is_new = 1;
458 patch->new_name = patch->def_name;
459 return gitdiff_newmode(line, patch);
460 }
462 static int gitdiff_copysrc(const char *line, struct patch *patch)
463 {
464 patch->is_copy = 1;
465 patch->old_name = find_name(line, NULL, 0, 0);
466 return 0;
467 }
469 static int gitdiff_copydst(const char *line, struct patch *patch)
470 {
471 patch->is_copy = 1;
472 patch->new_name = find_name(line, NULL, 0, 0);
473 return 0;
474 }
476 static int gitdiff_renamesrc(const char *line, struct patch *patch)
477 {
478 patch->is_rename = 1;
479 patch->old_name = find_name(line, NULL, 0, 0);
480 return 0;
481 }
483 static int gitdiff_renamedst(const char *line, struct patch *patch)
484 {
485 patch->is_rename = 1;
486 patch->new_name = find_name(line, NULL, 0, 0);
487 return 0;
488 }
490 static int gitdiff_similarity(const char *line, struct patch *patch)
491 {
492 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
493 patch->score = 0;
494 return 0;
495 }
497 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
498 {
499 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
500 patch->score = 0;
501 return 0;
502 }
504 static int gitdiff_index(const char *line, struct patch *patch)
505 {
506 /*
507 * index line is N hexadecimal, "..", N hexadecimal,
508 * and optional space with octal mode.
509 */
510 const char *ptr, *eol;
511 int len;
513 ptr = strchr(line, '.');
514 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
515 return 0;
516 len = ptr - line;
517 memcpy(patch->old_sha1_prefix, line, len);
518 patch->old_sha1_prefix[len] = 0;
520 line = ptr + 2;
521 ptr = strchr(line, ' ');
522 eol = strchr(line, '\n');
524 if (!ptr || eol < ptr)
525 ptr = eol;
526 len = ptr - line;
528 if (40 < len)
529 return 0;
530 memcpy(patch->new_sha1_prefix, line, len);
531 patch->new_sha1_prefix[len] = 0;
532 if (*ptr == ' ')
533 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
534 return 0;
535 }
537 /*
538 * This is normal for a diff that doesn't change anything: we'll fall through
539 * into the next diff. Tell the parser to break out.
540 */
541 static int gitdiff_unrecognized(const char *line, struct patch *patch)
542 {
543 return -1;
544 }
546 static const char *stop_at_slash(const char *line, int llen)
547 {
548 int i;
550 for (i = 0; i < llen; i++) {
551 int ch = line[i];
552 if (ch == '/')
553 return line + i;
554 }
555 return NULL;
556 }
558 /*
559 * This is to extract the same name that appears on "diff --git"
560 * line. We do not find and return anything if it is a rename
561 * patch, and it is OK because we will find the name elsewhere.
562 * We need to reliably find name only when it is mode-change only,
563 * creation or deletion of an empty file. In any of these cases,
564 * both sides are the same name under a/ and b/ respectively.
565 */
566 static char *git_header_name(char *line, int llen)
567 {
568 const char *name;
569 const char *second = NULL;
570 size_t len;
572 line += strlen("diff --git ");
573 llen -= strlen("diff --git ");
575 if (*line == '"') {
576 const char *cp;
577 struct strbuf first;
578 struct strbuf sp;
580 strbuf_init(&first, 0);
581 strbuf_init(&sp, 0);
583 if (unquote_c_style(&first, line, &second))
584 goto free_and_fail1;
586 /* advance to the first slash */
587 cp = stop_at_slash(first.buf, first.len);
588 /* we do not accept absolute paths */
589 if (!cp || cp == first.buf)
590 goto free_and_fail1;
591 strbuf_remove(&first, 0, cp + 1 - first.buf);
593 /*
594 * second points at one past closing dq of name.
595 * find the second name.
596 */
597 while ((second < line + llen) && isspace(*second))
598 second++;
600 if (line + llen <= second)
601 goto free_and_fail1;
602 if (*second == '"') {
603 if (unquote_c_style(&sp, second, NULL))
604 goto free_and_fail1;
605 cp = stop_at_slash(sp.buf, sp.len);
606 if (!cp || cp == sp.buf)
607 goto free_and_fail1;
608 /* They must match, otherwise ignore */
609 if (strcmp(cp + 1, first.buf))
610 goto free_and_fail1;
611 strbuf_release(&sp);
612 return strbuf_detach(&first, NULL);
613 }
615 /* unquoted second */
616 cp = stop_at_slash(second, line + llen - second);
617 if (!cp || cp == second)
618 goto free_and_fail1;
619 cp++;
620 if (line + llen - cp != first.len + 1 ||
621 memcmp(first.buf, cp, first.len))
622 goto free_and_fail1;
623 return strbuf_detach(&first, NULL);
625 free_and_fail1:
626 strbuf_release(&first);
627 strbuf_release(&sp);
628 return NULL;
629 }
631 /* unquoted first name */
632 name = stop_at_slash(line, llen);
633 if (!name || name == line)
634 return NULL;
635 name++;
637 /*
638 * since the first name is unquoted, a dq if exists must be
639 * the beginning of the second name.
640 */
641 for (second = name; second < line + llen; second++) {
642 if (*second == '"') {
643 struct strbuf sp;
644 const char *np;
646 strbuf_init(&sp, 0);
647 if (unquote_c_style(&sp, second, NULL))
648 goto free_and_fail2;
650 np = stop_at_slash(sp.buf, sp.len);
651 if (!np || np == sp.buf)
652 goto free_and_fail2;
653 np++;
655 len = sp.buf + sp.len - np;
656 if (len < second - name &&
657 !strncmp(np, name, len) &&
658 isspace(name[len])) {
659 /* Good */
660 strbuf_remove(&sp, 0, np - sp.buf);
661 return strbuf_detach(&sp, NULL);
662 }
664 free_and_fail2:
665 strbuf_release(&sp);
666 return NULL;
667 }
668 }
670 /*
671 * Accept a name only if it shows up twice, exactly the same
672 * form.
673 */
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;
688 }
689 if (second[len] == '\n' && !memcmp(name, second, len)) {
690 return xmemdupz(name, len);
691 }
692 }
693 }
694 return NULL;
695 }
697 /* Verify that we recognize the lines following a git header */
698 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
699 {
700 unsigned long offset;
702 /* A git diff has explicit new/delete information, so we don't guess */
703 patch->is_new = 0;
704 patch->is_delete = 0;
706 /*
707 * Some things may not have the old name in the
708 * rest of the headers anywhere (pure mode changes,
709 * or removing or adding empty files), so we get
710 * the default name from the header.
711 */
712 patch->def_name = git_header_name(line, len);
714 line += len;
715 size -= len;
716 linenr++;
717 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
718 static const struct opentry {
719 const char *str;
720 int (*fn)(const char *, struct patch *);
721 } optable[] = {
722 { "@@ -", gitdiff_hdrend },
723 { "--- ", gitdiff_oldname },
724 { "+++ ", gitdiff_newname },
725 { "old mode ", gitdiff_oldmode },
726 { "new mode ", gitdiff_newmode },
727 { "deleted file mode ", gitdiff_delete },
728 { "new file mode ", gitdiff_newfile },
729 { "copy from ", gitdiff_copysrc },
730 { "copy to ", gitdiff_copydst },
731 { "rename old ", gitdiff_renamesrc },
732 { "rename new ", gitdiff_renamedst },
733 { "rename from ", gitdiff_renamesrc },
734 { "rename to ", gitdiff_renamedst },
735 { "similarity index ", gitdiff_similarity },
736 { "dissimilarity index ", gitdiff_dissimilarity },
737 { "index ", gitdiff_index },
738 { "", gitdiff_unrecognized },
739 };
740 int i;
742 len = linelen(line, size);
743 if (!len || line[len-1] != '\n')
744 break;
745 for (i = 0; i < ARRAY_SIZE(optable); i++) {
746 const struct opentry *p = optable + i;
747 int oplen = strlen(p->str);
748 if (len < oplen || memcmp(p->str, line, oplen))
749 continue;
750 if (p->fn(line + oplen, patch) < 0)
751 return offset;
752 break;
753 }
754 }
756 return offset;
757 }
759 static int parse_num(const char *line, unsigned long *p)
760 {
761 char *ptr;
763 if (!isdigit(*line))
764 return 0;
765 *p = strtoul(line, &ptr, 10);
766 return ptr - line;
767 }
769 static int parse_range(const char *line, int len, int offset, const char *expect,
770 unsigned long *p1, unsigned long *p2)
771 {
772 int digits, ex;
774 if (offset < 0 || offset >= len)
775 return -1;
776 line += offset;
777 len -= offset;
779 digits = parse_num(line, p1);
780 if (!digits)
781 return -1;
783 offset += digits;
784 line += digits;
785 len -= digits;
787 *p2 = 1;
788 if (*line == ',') {
789 digits = parse_num(line+1, p2);
790 if (!digits)
791 return -1;
793 offset += digits+1;
794 line += digits+1;
795 len -= digits+1;
796 }
798 ex = strlen(expect);
799 if (ex > len)
800 return -1;
801 if (memcmp(line, expect, ex))
802 return -1;
804 return offset + ex;
805 }
807 /*
808 * Parse a unified diff fragment header of the
809 * form "@@ -a,b +c,d @@"
810 */
811 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
812 {
813 int offset;
815 if (!len || line[len-1] != '\n')
816 return -1;
818 /* Figure out the number of lines in a fragment */
819 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
820 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
822 return offset;
823 }
825 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
826 {
827 unsigned long offset, len;
829 patch->is_toplevel_relative = 0;
830 patch->is_rename = patch->is_copy = 0;
831 patch->is_new = patch->is_delete = -1;
832 patch->old_mode = patch->new_mode = 0;
833 patch->old_name = patch->new_name = NULL;
834 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
835 unsigned long nextlen;
837 len = linelen(line, size);
838 if (!len)
839 break;
841 /* Testing this early allows us to take a few shortcuts.. */
842 if (len < 6)
843 continue;
845 /*
846 * Make sure we don't find any unconnected patch fragments.
847 * That's a sign that we didn't find a header, and that a
848 * patch has become corrupted/broken up.
849 */
850 if (!memcmp("@@ -", line, 4)) {
851 struct fragment dummy;
852 if (parse_fragment_header(line, len, &dummy) < 0)
853 continue;
854 die("patch fragment without header at line %d: %.*s",
855 linenr, (int)len-1, line);
856 }
858 if (size < len + 6)
859 break;
861 /*
862 * Git patch? It might not have a real patch, just a rename
863 * or mode change, so we handle that specially
864 */
865 if (!memcmp("diff --git ", line, 11)) {
866 int git_hdr_len = parse_git_header(line, len, size, patch);
867 if (git_hdr_len <= len)
868 continue;
869 if (!patch->old_name && !patch->new_name) {
870 if (!patch->def_name)
871 die("git diff header lacks filename information (line %d)", linenr);
872 patch->old_name = patch->new_name = patch->def_name;
873 }
874 patch->is_toplevel_relative = 1;
875 *hdrsize = git_hdr_len;
876 return offset;
877 }
879 /* --- followed by +++ ? */
880 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
881 continue;
883 /*
884 * We only accept unified patches, so we want it to
885 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
886 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
887 */
888 nextlen = linelen(line + len, size - len);
889 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
890 continue;
892 /* Ok, we'll consider it a patch */
893 parse_traditional_patch(line, line+len, patch);
894 *hdrsize = len + nextlen;
895 linenr += 2;
896 return offset;
897 }
898 return -1;
899 }
901 static void check_whitespace(const char *line, int len)
902 {
903 const char *err = "Adds trailing whitespace";
904 int seen_space = 0;
905 int i;
907 /*
908 * We know len is at least two, since we have a '+' and we
909 * checked that the last character was a '\n' before calling
910 * this function. That is, an addition of an empty line would
911 * check the '+' here. Sneaky...
912 */
913 if (isspace(line[len-2]))
914 goto error;
916 /*
917 * Make sure that there is no space followed by a tab in
918 * indentation.
919 */
920 err = "Space in indent is followed by a tab";
921 for (i = 1; i < len; i++) {
922 if (line[i] == '\t') {
923 if (seen_space)
924 goto error;
925 }
926 else if (line[i] == ' ')
927 seen_space = 1;
928 else
929 break;
930 }
931 return;
933 error:
934 whitespace_error++;
935 if (squelch_whitespace_errors &&
936 squelch_whitespace_errors < whitespace_error)
937 ;
938 else
939 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
940 err, patch_input_file, linenr, len-2, line+1);
941 }
943 /*
944 * Parse a unified diff. Note that this really needs to parse each
945 * fragment separately, since the only way to know the difference
946 * between a "---" that is part of a patch, and a "---" that starts
947 * the next patch is to look at the line counts..
948 */
949 static int parse_fragment(char *line, unsigned long size,
950 struct patch *patch, struct fragment *fragment)
951 {
952 int added, deleted;
953 int len = linelen(line, size), offset;
954 unsigned long oldlines, newlines;
955 unsigned long leading, trailing;
957 offset = parse_fragment_header(line, len, fragment);
958 if (offset < 0)
959 return -1;
960 oldlines = fragment->oldlines;
961 newlines = fragment->newlines;
962 leading = 0;
963 trailing = 0;
965 /* Parse the thing.. */
966 line += len;
967 size -= len;
968 linenr++;
969 added = deleted = 0;
970 for (offset = len;
971 0 < size;
972 offset += len, size -= len, line += len, linenr++) {
973 if (!oldlines && !newlines)
974 break;
975 len = linelen(line, size);
976 if (!len || line[len-1] != '\n')
977 return -1;
978 switch (*line) {
979 default:
980 return -1;
981 case '\n': /* newer GNU diff, an empty context line */
982 case ' ':
983 oldlines--;
984 newlines--;
985 if (!deleted && !added)
986 leading++;
987 trailing++;
988 break;
989 case '-':
990 if (apply_in_reverse &&
991 ws_error_action != nowarn_ws_error)
992 check_whitespace(line, len);
993 deleted++;
994 oldlines--;
995 trailing = 0;
996 break;
997 case '+':
998 if (!apply_in_reverse &&
999 ws_error_action != nowarn_ws_error)
1000 check_whitespace(line, len);
1001 added++;
1002 newlines--;
1003 trailing = 0;
1004 break;
1006 /*
1007 * We allow "\ No newline at end of file". Depending
1008 * on locale settings when the patch was produced we
1009 * don't know what this line looks like. The only
1010 * thing we do know is that it begins with "\ ".
1011 * Checking for 12 is just for sanity check -- any
1012 * l10n of "\ No newline..." is at least that long.
1013 */
1014 case '\\':
1015 if (len < 12 || memcmp(line, "\\ ", 2))
1016 return -1;
1017 break;
1018 }
1019 }
1020 if (oldlines || newlines)
1021 return -1;
1022 fragment->leading = leading;
1023 fragment->trailing = trailing;
1025 /*
1026 * If a fragment ends with an incomplete line, we failed to include
1027 * it in the above loop because we hit oldlines == newlines == 0
1028 * before seeing it.
1029 */
1030 if (12 < size && !memcmp(line, "\\ ", 2))
1031 offset += linelen(line, size);
1033 patch->lines_added += added;
1034 patch->lines_deleted += deleted;
1036 if (0 < patch->is_new && oldlines)
1037 return error("new file depends on old contents");
1038 if (0 < patch->is_delete && newlines)
1039 return error("deleted file still has contents");
1040 return offset;
1041 }
1043 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1044 {
1045 unsigned long offset = 0;
1046 unsigned long oldlines = 0, newlines = 0, context = 0;
1047 struct fragment **fragp = &patch->fragments;
1049 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1050 struct fragment *fragment;
1051 int len;
1053 fragment = xcalloc(1, sizeof(*fragment));
1054 len = parse_fragment(line, size, patch, fragment);
1055 if (len <= 0)
1056 die("corrupt patch at line %d", linenr);
1057 fragment->patch = line;
1058 fragment->size = len;
1059 oldlines += fragment->oldlines;
1060 newlines += fragment->newlines;
1061 context += fragment->leading + fragment->trailing;
1063 *fragp = fragment;
1064 fragp = &fragment->next;
1066 offset += len;
1067 line += len;
1068 size -= len;
1069 }
1071 /*
1072 * If something was removed (i.e. we have old-lines) it cannot
1073 * be creation, and if something was added it cannot be
1074 * deletion. However, the reverse is not true; --unified=0
1075 * patches that only add are not necessarily creation even
1076 * though they do not have any old lines, and ones that only
1077 * delete are not necessarily deletion.
1078 *
1079 * Unfortunately, a real creation/deletion patch do _not_ have
1080 * any context line by definition, so we cannot safely tell it
1081 * apart with --unified=0 insanity. At least if the patch has
1082 * more than one hunk it is not creation or deletion.
1083 */
1084 if (patch->is_new < 0 &&
1085 (oldlines || (patch->fragments && patch->fragments->next)))
1086 patch->is_new = 0;
1087 if (patch->is_delete < 0 &&
1088 (newlines || (patch->fragments && patch->fragments->next)))
1089 patch->is_delete = 0;
1090 if (!unidiff_zero || context) {
1091 /* If the user says the patch is not generated with
1092 * --unified=0, or if we have seen context lines,
1093 * then not having oldlines means the patch is creation,
1094 * and not having newlines means the patch is deletion.
1095 */
1096 if (patch->is_new < 0 && !oldlines) {
1097 patch->is_new = 1;
1098 patch->old_name = NULL;
1099 }
1100 if (patch->is_delete < 0 && !newlines) {
1101 patch->is_delete = 1;
1102 patch->new_name = NULL;
1103 }
1104 }
1106 if (0 < patch->is_new && oldlines)
1107 die("new file %s depends on old contents", patch->new_name);
1108 if (0 < patch->is_delete && newlines)
1109 die("deleted file %s still has contents", patch->old_name);
1110 if (!patch->is_delete && !newlines && context)
1111 fprintf(stderr, "** warning: file %s becomes empty but "
1112 "is not deleted\n", patch->new_name);
1114 return offset;
1115 }
1117 static inline int metadata_changes(struct patch *patch)
1118 {
1119 return patch->is_rename > 0 ||
1120 patch->is_copy > 0 ||
1121 patch->is_new > 0 ||
1122 patch->is_delete ||
1123 (patch->old_mode && patch->new_mode &&
1124 patch->old_mode != patch->new_mode);
1125 }
1127 static char *inflate_it(const void *data, unsigned long size,
1128 unsigned long inflated_size)
1129 {
1130 z_stream stream;
1131 void *out;
1132 int st;
1134 memset(&stream, 0, sizeof(stream));
1136 stream.next_in = (unsigned char *)data;
1137 stream.avail_in = size;
1138 stream.next_out = out = xmalloc(inflated_size);
1139 stream.avail_out = inflated_size;
1140 inflateInit(&stream);
1141 st = inflate(&stream, Z_FINISH);
1142 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1143 free(out);
1144 return NULL;
1145 }
1146 return out;
1147 }
1149 static struct fragment *parse_binary_hunk(char **buf_p,
1150 unsigned long *sz_p,
1151 int *status_p,
1152 int *used_p)
1153 {
1154 /*
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.
1159 *
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.
1165 */
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);
1183 }
1184 else if (!prefixcmp(buffer, "literal ")) {
1185 patch_method = BINARY_LITERAL_DEFLATED;
1186 origlen = strtoul(buffer + 8, NULL, 10);
1187 }
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;
1203 }
1204 /*
1205 * Minimum line is "A00000\n" which is 7-byte long,
1206 * and the line length must be multiple of 5 plus 2.
1207 */
1208 if ((llen < 7) || (llen-2) % 5)
1209 goto corrupt;
1210 max_byte_length = (llen - 2) / 5 * 4;
1211 byte_length = *buffer;
1212 if ('A' <= byte_length && byte_length <= 'Z')
1213 byte_length = byte_length - 'A' + 1;
1214 else if ('a' <= byte_length && byte_length <= 'z')
1215 byte_length = byte_length - 'a' + 27;
1216 else
1217 goto corrupt;
1218 /* if the input length was not multiple of 4, we would
1219 * have filler at the end but the filler should never
1220 * exceed 3 bytes
1221 */
1222 if (max_byte_length < byte_length ||
1223 byte_length <= max_byte_length - 4)
1224 goto corrupt;
1225 newsize = hunk_size + byte_length;
1226 data = xrealloc(data, newsize);
1227 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1228 goto corrupt;
1229 hunk_size = newsize;
1230 buffer += llen;
1231 size -= llen;
1232 }
1234 frag = xcalloc(1, sizeof(*frag));
1235 frag->patch = inflate_it(data, hunk_size, origlen);
1236 if (!frag->patch)
1237 goto corrupt;
1238 free(data);
1239 frag->size = origlen;
1240 *buf_p = buffer;
1241 *sz_p = size;
1242 *used_p = used;
1243 frag->binary_patch_method = patch_method;
1244 return frag;
1246 corrupt:
1247 free(data);
1248 *status_p = -1;
1249 error("corrupt binary patch at line %d: %.*s",
1250 linenr-1, llen-1, buffer);
1251 return NULL;
1252 }
1254 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1255 {
1256 /*
1257 * We have read "GIT binary patch\n"; what follows is a line
1258 * that says the patch method (currently, either "literal" or
1259 * "delta") and the length of data before deflating; a
1260 * sequence of 'length-byte' followed by base-85 encoded data
1261 * follows.
1262 *
1263 * When a binary patch is reversible, there is another binary
1264 * hunk in the same format, starting with patch method (either
1265 * "literal" or "delta") with the length of data, and a sequence
1266 * of length-byte + base-85 encoded data, terminated with another
1267 * empty line. This data, when applied to the postimage, produces
1268 * the preimage.
1269 */
1270 struct fragment *forward;
1271 struct fragment *reverse;
1272 int status;
1273 int used, used_1;
1275 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1276 if (!forward && !status)
1277 /* there has to be one hunk (forward hunk) */
1278 return error("unrecognized binary patch at line %d", linenr-1);
1279 if (status)
1280 /* otherwise we already gave an error message */
1281 return status;
1283 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1284 if (reverse)
1285 used += used_1;
1286 else if (status) {
1287 /*
1288 * Not having reverse hunk is not an error, but having
1289 * a corrupt reverse hunk is.
1290 */
1291 free((void*) forward->patch);
1292 free(forward);
1293 return status;
1294 }
1295 forward->next = reverse;
1296 patch->fragments = forward;
1297 patch->is_binary = 1;
1298 return used;
1299 }
1301 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1302 {
1303 int hdrsize, patchsize;
1304 int offset = find_header(buffer, size, &hdrsize, patch);
1306 if (offset < 0)
1307 return offset;
1309 patchsize = parse_single_patch(buffer + offset + hdrsize,
1310 size - offset - hdrsize, patch);
1312 if (!patchsize) {
1313 static const char *binhdr[] = {
1314 "Binary files ",
1315 "Files ",
1316 NULL,
1317 };
1318 static const char git_binary[] = "GIT binary patch\n";
1319 int i;
1320 int hd = hdrsize + offset;
1321 unsigned long llen = linelen(buffer + hd, size - hd);
1323 if (llen == sizeof(git_binary) - 1 &&
1324 !memcmp(git_binary, buffer + hd, llen)) {
1325 int used;
1326 linenr++;
1327 used = parse_binary(buffer + hd + llen,
1328 size - hd - llen, patch);
1329 if (used)
1330 patchsize = used + llen;
1331 else
1332 patchsize = 0;
1333 }
1334 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1335 for (i = 0; binhdr[i]; i++) {
1336 int len = strlen(binhdr[i]);
1337 if (len < size - hd &&
1338 !memcmp(binhdr[i], buffer + hd, len)) {
1339 linenr++;
1340 patch->is_binary = 1;
1341 patchsize = llen;
1342 break;
1343 }
1344 }
1345 }
1347 /* Empty patch cannot be applied if it is a text patch
1348 * without metadata change. A binary patch appears
1349 * empty to us here.
1350 */
1351 if ((apply || check) &&
1352 (!patch->is_binary && !metadata_changes(patch)))
1353 die("patch with only garbage at line %d", linenr);
1354 }
1356 return offset + hdrsize + patchsize;
1357 }
1359 #define swap(a,b) myswap((a),(b),sizeof(a))
1361 #define myswap(a, b, size) do { \
1362 unsigned char mytmp[size]; \
1363 memcpy(mytmp, &a, size); \
1364 memcpy(&a, &b, size); \
1365 memcpy(&b, mytmp, size); \
1366 } while (0)
1368 static void reverse_patches(struct patch *p)
1369 {
1370 for (; p; p = p->next) {
1371 struct fragment *frag = p->fragments;
1373 swap(p->new_name, p->old_name);
1374 swap(p->new_mode, p->old_mode);
1375 swap(p->is_new, p->is_delete);
1376 swap(p->lines_added, p->lines_deleted);
1377 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1379 for (; frag; frag = frag->next) {
1380 swap(frag->newpos, frag->oldpos);
1381 swap(frag->newlines, frag->oldlines);
1382 }
1383 }
1384 }
1386 static const char pluses[] =
1387 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1388 static const char minuses[]=
1389 "----------------------------------------------------------------------";
1391 static void show_stats(struct patch *patch)
1392 {
1393 struct strbuf qname;
1394 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1395 int max, add, del;
1397 strbuf_init(&qname, 0);
1398 quote_c_style(cp, &qname, NULL, 0);
1400 /*
1401 * "scale" the filename
1402 */
1403 max = max_len;
1404 if (max > 50)
1405 max = 50;
1407 if (qname.len > max) {
1408 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1409 if (!cp)
1410 cp = qname.buf + qname.len + 3 - max;
1411 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1412 }
1414 if (patch->is_binary) {
1415 printf(" %-*s | Bin\n", max, qname.buf);
1416 strbuf_release(&qname);
1417 return;
1418 }
1420 printf(" %-*s |", max, qname.buf);
1421 strbuf_release(&qname);
1423 /*
1424 * scale the add/delete
1425 */
1426 max = max + max_change > 70 ? 70 - max : max_change;
1427 add = patch->lines_added;
1428 del = patch->lines_deleted;
1430 if (max_change > 0) {
1431 int total = ((add + del) * max + max_change / 2) / max_change;
1432 add = (add * max + max_change / 2) / max_change;
1433 del = total - add;
1434 }
1435 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1436 add, pluses, del, minuses);
1437 }
1439 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1440 {
1441 switch (st->st_mode & S_IFMT) {
1442 case S_IFLNK:
1443 strbuf_grow(buf, st->st_size);
1444 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1445 return -1;
1446 strbuf_setlen(buf, st->st_size);
1447 return 0;
1448 case S_IFREG:
1449 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1450 return error("unable to open or read %s", path);
1451 convert_to_git(path, buf->buf, buf->len, buf);
1452 return 0;
1453 default:
1454 return -1;
1455 }
1456 }
1458 static int find_offset(const char *buf, unsigned long size,
1459 const char *fragment, unsigned long fragsize,
1460 int line, int *lines)
1461 {
1462 int i;
1463 unsigned long start, backwards, forwards;
1465 if (fragsize > size)
1466 return -1;
1468 start = 0;
1469 if (line > 1) {
1470 unsigned long offset = 0;
1471 i = line-1;
1472 while (offset + fragsize <= size) {
1473 if (buf[offset++] == '\n') {
1474 start = offset;
1475 if (!--i)
1476 break;
1477 }
1478 }
1479 }
1481 /* Exact line number? */
1482 if ((start + fragsize <= size) &&
1483 !memcmp(buf + start, fragment, fragsize))
1484 return start;
1486 /*
1487 * There's probably some smart way to do this, but I'll leave
1488 * that to the smart and beautiful people. I'm simple and stupid.
1489 */
1490 backwards = start;
1491 forwards = start;
1492 for (i = 0; ; i++) {
1493 unsigned long try;
1494 int n;
1496 /* "backward" */
1497 if (i & 1) {
1498 if (!backwards) {
1499 if (forwards + fragsize > size)
1500 break;
1501 continue;
1502 }
1503 do {
1504 --backwards;
1505 } while (backwards && buf[backwards-1] != '\n');
1506 try = backwards;
1507 } else {
1508 while (forwards + fragsize <= size) {
1509 if (buf[forwards++] == '\n')
1510 break;
1511 }
1512 try = forwards;
1513 }
1515 if (try + fragsize > size)
1516 continue;
1517 if (memcmp(buf + try, fragment, fragsize))
1518 continue;
1519 n = (i >> 1)+1;
1520 if (i & 1)
1521 n = -n;
1522 *lines = n;
1523 return try;
1524 }
1526 /*
1527 * We should start searching forward and backward.
1528 */
1529 return -1;
1530 }
1532 static void remove_first_line(const char **rbuf, int *rsize)
1533 {
1534 const char *buf = *rbuf;
1535 int size = *rsize;
1536 unsigned long offset;
1537 offset = 0;
1538 while (offset <= size) {
1539 if (buf[offset++] == '\n')
1540 break;
1541 }
1542 *rsize = size - offset;
1543 *rbuf = buf + offset;
1544 }
1546 static void remove_last_line(const char **rbuf, int *rsize)
1547 {
1548 const char *buf = *rbuf;
1549 int size = *rsize;
1550 unsigned long offset;
1551 offset = size - 1;
1552 while (offset > 0) {
1553 if (buf[--offset] == '\n')
1554 break;
1555 }
1556 *rsize = offset + 1;
1557 }
1559 static int apply_line(char *output, const char *patch, int plen)
1560 {
1561 /*
1562 * plen is number of bytes to be copied from patch,
1563 * starting at patch+1 (patch[0] is '+'). Typically
1564 * patch[plen] is '\n', unless this is the incomplete
1565 * last line.
1566 */
1567 int i;
1568 int add_nl_to_tail = 0;
1569 int fixed = 0;
1570 int last_tab_in_indent = -1;
1571 int last_space_in_indent = -1;
1572 int need_fix_leading_space = 0;
1573 char *buf;
1575 if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1576 *patch != '+') {
1577 memcpy(output, patch + 1, plen);
1578 return plen;
1579 }
1581 /*
1582 * Strip trailing whitespace
1583 */
1584 if (1 < plen && isspace(patch[plen-1])) {
1585 if (patch[plen] == '\n')
1586 add_nl_to_tail = 1;
1587 plen--;
1588 while (0 < plen && isspace(patch[plen]))
1589 plen--;
1590 fixed = 1;
1591 }
1593 /*
1594 * Check leading whitespaces (indent)
1595 */
1596 for (i = 1; i < plen; i++) {
1597 char ch = patch[i];
1598 if (ch == '\t') {
1599 last_tab_in_indent = i;
1600 if (0 <= last_space_in_indent)
1601 need_fix_leading_space = 1;
1602 }
1603 else if (ch == ' ')
1604 last_space_in_indent = i;
1605 else
1606 break;
1607 }
1609 buf = output;
1610 if (need_fix_leading_space) {
1611 int consecutive_spaces = 0;
1612 /*
1613 * between patch[1..last_tab_in_indent] strip the
1614 * funny spaces, updating them to tab as needed.
1615 */
1616 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1617 char ch = patch[i];
1618 if (ch != ' ') {
1619 consecutive_spaces = 0;
1620 *output++ = ch;
1621 } else {
1622 consecutive_spaces++;
1623 if (consecutive_spaces == 8) {
1624 *output++ = '\t';
1625 consecutive_spaces = 0;
1626 }
1627 }
1628 }
1629 fixed = 1;
1630 i = last_tab_in_indent;
1631 }
1632 else
1633 i = 1;
1635 memcpy(output, patch + i, plen);
1636 if (add_nl_to_tail)
1637 output[plen++] = '\n';
1638 if (fixed)
1639 applied_after_fixing_ws++;
1640 return output + plen - buf;
1641 }
1643 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag,
1644 int inaccurate_eof)
1645 {
1646 int match_beginning, match_end;
1647 const char *patch = frag->patch;
1648 int offset, size = frag->size;
1649 char *old = xmalloc(size);
1650 char *new = xmalloc(size);
1651 const char *oldlines, *newlines;
1652 int oldsize = 0, newsize = 0;
1653 int new_blank_lines_at_end = 0;
1654 unsigned long leading, trailing;
1655 int pos, lines;
1657 while (size > 0) {
1658 char first;
1659 int len = linelen(patch, size);
1660 int plen;
1661 int added_blank_line = 0;
1663 if (!len)
1664 break;
1666 /*
1667 * "plen" is how much of the line we should use for
1668 * the actual patch data. Normally we just remove the
1669 * first character on the line, but if the line is
1670 * followed by "\ No newline", then we also remove the
1671 * last one (which is the newline, of course).
1672 */
1673 plen = len-1;
1674 if (len < size && patch[len] == '\\')
1675 plen--;
1676 first = *patch;
1677 if (apply_in_reverse) {
1678 if (first == '-')
1679 first = '+';
1680 else if (first == '+')
1681 first = '-';
1682 }
1684 switch (first) {
1685 case '\n':
1686 /* Newer GNU diff, empty context line */
1687 if (plen < 0)
1688 /* ... followed by '\No newline'; nothing */
1689 break;
1690 old[oldsize++] = '\n';
1691 new[newsize++] = '\n';
1692 break;
1693 case ' ':
1694 case '-':
1695 memcpy(old + oldsize, patch + 1, plen);
1696 oldsize += plen;
1697 if (first == '-')
1698 break;
1699 /* Fall-through for ' ' */
1700 case '+':
1701 if (first != '+' || !no_add) {
1702 int added = apply_line(new + newsize, patch,
1703 plen);
1704 newsize += added;
1705 if (first == '+' &&
1706 added == 1 && new[newsize-1] == '\n')
1707 added_blank_line = 1;
1708 }
1709 break;
1710 case '@': case '\\':
1711 /* Ignore it, we already handled it */
1712 break;
1713 default:
1714 if (apply_verbosely)
1715 error("invalid start of line: '%c'", first);
1716 return -1;
1717 }
1718 if (added_blank_line)
1719 new_blank_lines_at_end++;
1720 else
1721 new_blank_lines_at_end = 0;
1722 patch += len;
1723 size -= len;
1724 }
1726 if (inaccurate_eof &&
1727 oldsize > 0 && old[oldsize - 1] == '\n' &&
1728 newsize > 0 && new[newsize - 1] == '\n') {
1729 oldsize--;
1730 newsize--;
1731 }
1733 oldlines = old;
1734 newlines = new;
1735 leading = frag->leading;
1736 trailing = frag->trailing;
1738 /*
1739 * If we don't have any leading/trailing data in the patch,
1740 * we want it to match at the beginning/end of the file.
1741 *
1742 * But that would break if the patch is generated with
1743 * --unified=0; sane people wouldn't do that to cause us
1744 * trouble, but we try to please not so sane ones as well.
1745 */
1746 if (unidiff_zero) {
1747 match_beginning = (!leading && !frag->oldpos);
1748 match_end = 0;
1749 }
1750 else {
1751 match_beginning = !leading && (frag->oldpos == 1);
1752 match_end = !trailing;
1753 }
1755 lines = 0;
1756 pos = frag->newpos;
1757 for (;;) {
1758 offset = find_offset(buf->buf, buf->len,
1759 oldlines, oldsize, pos, &lines);
1760 if (match_end && offset + oldsize != buf->len)
1761 offset = -1;
1762 if (match_beginning && offset)
1763 offset = -1;
1764 if (offset >= 0) {
1765 if (ws_error_action == correct_ws_error &&
1766 (buf->len - oldsize - offset == 0)) /* end of file? */
1767 newsize -= new_blank_lines_at_end;
1769 /* Warn if it was necessary to reduce the number
1770 * of context lines.
1771 */
1772 if ((leading != frag->leading) ||
1773 (trailing != frag->trailing))
1774 fprintf(stderr, "Context reduced to (%ld/%ld)"
1775 " to apply fragment at %d\n",
1776 leading, trailing, pos + lines);
1778 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1779 offset = 0;
1780 break;
1781 }
1783 /* Am I at my context limits? */
1784 if ((leading <= p_context) && (trailing <= p_context))
1785 break;
1786 if (match_beginning || match_end) {
1787 match_beginning = match_end = 0;
1788 continue;
1789 }
1790 /*
1791 * Reduce the number of context lines; reduce both
1792 * leading and trailing if they are equal otherwise
1793 * just reduce the larger context.
1794 */
1795 if (leading >= trailing) {
1796 remove_first_line(&oldlines, &oldsize);
1797 remove_first_line(&newlines, &newsize);
1798 pos--;
1799 leading--;
1800 }
1801 if (trailing > leading) {
1802 remove_last_line(&oldlines, &oldsize);
1803 remove_last_line(&newlines, &newsize);
1804 trailing--;
1805 }
1806 }
1808 if (offset && apply_verbosely)
1809 error("while searching for:\n%.*s", oldsize, oldlines);
1811 free(old);
1812 free(new);
1813 return offset;
1814 }
1816 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1817 {
1818 struct fragment *fragment = patch->fragments;
1819 unsigned long len;
1820 void *dst;
1822 /* Binary patch is irreversible without the optional second hunk */
1823 if (apply_in_reverse) {
1824 if (!fragment->next)
1825 return error("cannot reverse-apply a binary patch "
1826 "without the reverse hunk to '%s'",
1827 patch->new_name
1828 ? patch->new_name : patch->old_name);
1829 fragment = fragment->next;
1830 }
1831 switch (fragment->binary_patch_method) {
1832 case BINARY_DELTA_DEFLATED:
1833 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1834 fragment->size, &len);
1835 if (!dst)
1836 return -1;
1837 /* XXX patch_delta NUL-terminates */
1838 strbuf_attach(buf, dst, len, len + 1);
1839 return 0;
1840 case BINARY_LITERAL_DEFLATED:
1841 strbuf_reset(buf);
1842 strbuf_add(buf, fragment->patch, fragment->size);
1843 return 0;
1844 }
1845 return -1;
1846 }
1848 static int apply_binary(struct strbuf *buf, struct patch *patch)
1849 {
1850 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1851 unsigned char sha1[20];
1853 /*
1854 * For safety, we require patch index line to contain
1855 * full 40-byte textual SHA1 for old and new, at least for now.
1856 */
1857 if (strlen(patch->old_sha1_prefix) != 40 ||
1858 strlen(patch->new_sha1_prefix) != 40 ||
1859 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1860 get_sha1_hex(patch->new_sha1_prefix, sha1))
1861 return error("cannot apply binary patch to '%s' "
1862 "without full index line", name);
1864 if (patch->old_name) {
1865 /*
1866 * See if the old one matches what the patch
1867 * applies to.
1868 */
1869 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1870 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1871 return error("the patch applies to '%s' (%s), "
1872 "which does not match the "
1873 "current contents.",
1874 name, sha1_to_hex(sha1));
1875 }
1876 else {
1877 /* Otherwise, the old one must be empty. */
1878 if (buf->len)
1879 return error("the patch applies to an empty "
1880 "'%s' but it is not empty", name);
1881 }
1883 get_sha1_hex(patch->new_sha1_prefix, sha1);
1884 if (is_null_sha1(sha1)) {
1885 strbuf_release(buf);
1886 return 0; /* deletion patch */
1887 }
1889 if (has_sha1_file(sha1)) {
1890 /* We already have the postimage */
1891 enum object_type type;
1892 unsigned long size;
1893 char *result;
1895 result = read_sha1_file(sha1, &type, &size);
1896 if (!result)
1897 return error("the necessary postimage %s for "
1898 "'%s' cannot be read",
1899 patch->new_sha1_prefix, name);
1900 /* XXX read_sha1_file NUL-terminates */
1901 strbuf_attach(buf, result, size, size + 1);
1902 } else {
1903 /*
1904 * We have verified buf matches the preimage;
1905 * apply the patch data to it, which is stored
1906 * in the patch->fragments->{patch,size}.
1907 */
1908 if (apply_binary_fragment(buf, patch))
1909 return error("binary patch does not apply to '%s'",
1910 name);
1912 /* verify that the result matches */
1913 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1914 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1915 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1916 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1917 }
1919 return 0;
1920 }
1922 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1923 {
1924 struct fragment *frag = patch->fragments;
1925 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1927 if (patch->is_binary)
1928 return apply_binary(buf, patch);
1930 while (frag) {
1931 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1932 error("patch failed: %s:%ld", name, frag->oldpos);
1933 if (!apply_with_reject)
1934 return -1;
1935 frag->rejected = 1;
1936 }
1937 frag = frag->next;
1938 }
1939 return 0;
1940 }
1942 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1943 {
1944 if (!ce)
1945 return 0;
1947 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1948 strbuf_grow(buf, 100);
1949 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1950 } else {
1951 enum object_type type;
1952 unsigned long sz;
1953 char *result;
1955 result = read_sha1_file(ce->sha1, &type, &sz);
1956 if (!result)
1957 return -1;
1958 /* XXX read_sha1_file NUL-terminates */
1959 strbuf_attach(buf, result, sz, sz + 1);
1960 }
1961 return 0;
1962 }
1964 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1965 {
1966 struct strbuf buf;
1968 strbuf_init(&buf, 0);
1969 if (cached) {
1970 if (read_file_or_gitlink(ce, &buf))
1971 return error("read of %s failed", patch->old_name);
1972 } else if (patch->old_name) {
1973 if (S_ISGITLINK(patch->old_mode)) {
1974 if (ce) {
1975 read_file_or_gitlink(ce, &buf);
1976 } else {
1977 /*
1978 * There is no way to apply subproject
1979 * patch without looking at the index.
1980 */
1981 patch->fragments = NULL;
1982 }
1983 } else {
1984 if (read_old_data(st, patch->old_name, &buf))
1985 return error("read of %s failed", patch->old_name);
1986 }
1987 }
1989 if (apply_fragments(&buf, patch) < 0)
1990 return -1; /* note with --reject this succeeds. */
1991 patch->result = strbuf_detach(&buf, &patch->resultsize);
1993 if (0 < patch->is_delete && patch->resultsize)
1994 return error("removal patch leaves file contents");
1996 return 0;
1997 }
1999 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2000 {
2001 struct stat nst;
2002 if (!lstat(new_name, &nst)) {
2003 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2004 return 0;
2005 /*
2006 * A leading component of new_name might be a symlink
2007 * that is going to be removed with this patch, but
2008 * still pointing at somewhere that has the path.
2009 * In such a case, path "new_name" does not exist as
2010 * far as git is concerned.
2011 */
2012 if (has_symlink_leading_path(new_name, NULL))
2013 return 0;
2015 return error("%s: already exists in working directory", new_name);
2016 }
2017 else if ((errno != ENOENT) && (errno != ENOTDIR))
2018 return error("%s: %s", new_name, strerror(errno));
2019 return 0;
2020 }
2022 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2023 {
2024 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2025 if (!S_ISDIR(st->st_mode))
2026 return -1;
2027 return 0;
2028 }
2029 return ce_match_stat(ce, st, 1);
2030 }
2032 static int check_patch(struct patch *patch, struct patch *prev_patch)
2033 {
2034 struct stat st;
2035 const char *old_name = patch->old_name;
2036 const char *new_name = patch->new_name;
2037 const char *name = old_name ? old_name : new_name;
2038 struct cache_entry *ce = NULL;
2039 int ok_if_exists;
2041 patch->rejected = 1; /* we will drop this after we succeed */
2043 /*
2044 * Make sure that we do not have local modifications from the
2045 * index when we are looking at the index. Also make sure
2046 * we have the preimage file to be patched in the work tree,
2047 * unless --cached, which tells git to apply only in the index.
2048 */
2049 if (old_name) {
2050 int stat_ret = 0;
2051 unsigned st_mode = 0;
2053 if (!cached)
2054 stat_ret = lstat(old_name, &st);
2055 if (check_index) {
2056 int pos = cache_name_pos(old_name, strlen(old_name));
2057 if (pos < 0)
2058 return error("%s: does not exist in index",
2059 old_name);
2060 ce = active_cache[pos];
2061 if (stat_ret < 0) {
2062 struct checkout costate;
2063 if (errno != ENOENT)
2064 return error("%s: %s", old_name,
2065 strerror(errno));
2066 /* checkout */
2067 costate.base_dir = "";
2068 costate.base_dir_len = 0;
2069 costate.force = 0;
2070 costate.quiet = 0;
2071 costate.not_new = 0;
2072 costate.refresh_cache = 1;
2073 if (checkout_entry(ce,
2074 &costate,
2075 NULL) ||
2076 lstat(old_name, &st))
2077 return -1;
2078 }
2079 if (!cached && verify_index_match(ce, &st))
2080 return error("%s: does not match index",
2081 old_name);
2082 if (cached)
2083 st_mode = ntohl(ce->ce_mode);
2084 } else if (stat_ret < 0)
2085 return error("%s: %s", old_name, strerror(errno));
2087 if (!cached)
2088 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2090 if (patch->is_new < 0)
2091 patch->is_new = 0;
2092 if (!patch->old_mode)
2093 patch->old_mode = st_mode;
2094 if ((st_mode ^ patch->old_mode) & S_IFMT)
2095 return error("%s: wrong type", old_name);
2096 if (st_mode != patch->old_mode)
2097 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2098 old_name, st_mode, patch->old_mode);
2099 }
2101 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2102 !strcmp(prev_patch->old_name, new_name))
2103 /*
2104 * A type-change diff is always split into a patch to
2105 * delete old, immediately followed by a patch to
2106 * create new (see diff.c::run_diff()); in such a case
2107 * it is Ok that the entry to be deleted by the
2108 * previous patch is still in the working tree and in
2109 * the index.
2110 */
2111 ok_if_exists = 1;
2112 else
2113 ok_if_exists = 0;
2115 if (new_name &&
2116 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2117 if (check_index &&
2118 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2119 !ok_if_exists)
2120 return error("%s: already exists in index", new_name);
2121 if (!cached) {
2122 int err = check_to_create_blob(new_name, ok_if_exists);
2123 if (err)
2124 return err;
2125 }
2126 if (!patch->new_mode) {
2127 if (0 < patch->is_new)
2128 patch->new_mode = S_IFREG | 0644;
2129 else
2130 patch->new_mode = patch->old_mode;
2131 }
2132 }
2134 if (new_name && old_name) {
2135 int same = !strcmp(old_name, new_name);
2136 if (!patch->new_mode)
2137 patch->new_mode = patch->old_mode;
2138 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2139 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2140 patch->new_mode, new_name, patch->old_mode,
2141 same ? "" : " of ", same ? "" : old_name);
2142 }
2144 if (apply_data(patch, &st, ce) < 0)
2145 return error("%s: patch does not apply", name);
2146 patch->rejected = 0;
2147 return 0;
2148 }
2150 static int check_patch_list(struct patch *patch)
2151 {
2152 struct patch *prev_patch = NULL;
2153 int err = 0;
2155 for (prev_patch = NULL; patch ; patch = patch->next) {
2156 if (apply_verbosely)
2157 say_patch_name(stderr,
2158 "Checking patch ", patch, "...\n");
2159 err |= check_patch(patch, prev_patch);
2160 prev_patch = patch;
2161 }
2162 return err;
2163 }
2165 /* This function tries to read the sha1 from the current index */
2166 static int get_current_sha1(const char *path, unsigned char *sha1)
2167 {
2168 int pos;
2170 if (read_cache() < 0)
2171 return -1;
2172 pos = cache_name_pos(path, strlen(path));
2173 if (pos < 0)
2174 return -1;
2175 hashcpy(sha1, active_cache[pos]->sha1);
2176 return 0;
2177 }
2179 /* Build an index that contains the just the files needed for a 3way merge */
2180 static void build_fake_ancestor(struct patch *list, const char *filename)
2181 {
2182 struct patch *patch;
2183 struct index_state result = { 0 };
2184 int fd;
2186 /* Once we start supporting the reverse patch, it may be
2187 * worth showing the new sha1 prefix, but until then...
2188 */
2189 for (patch = list; patch; patch = patch->next) {
2190 const unsigned char *sha1_ptr;
2191 unsigned char sha1[20];
2192 struct cache_entry *ce;
2193 const char *name;
2195 name = patch->old_name ? patch->old_name : patch->new_name;
2196 if (0 < patch->is_new)
2197 continue;
2198 else if (get_sha1(patch->old_sha1_prefix, sha1))
2199 /* git diff has no index line for mode/type changes */
2200 if (!patch->lines_added && !patch->lines_deleted) {
2201 if (get_current_sha1(patch->new_name, sha1) ||
2202 get_current_sha1(patch->old_name, sha1))
2203 die("mode change for %s, which is not "
2204 "in current HEAD", name);
2205 sha1_ptr = sha1;
2206 } else
2207 die("sha1 information is lacking or useless "
2208 "(%s).", name);
2209 else
2210 sha1_ptr = sha1;
2212 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2213 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2214 die ("Could not add %s to temporary index", name);
2215 }
2217 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2218 if (fd < 0 || write_index(&result, fd) || close(fd))
2219 die ("Could not write temporary index to %s", filename);
2221 discard_index(&result);
2222 }
2224 static void stat_patch_list(struct patch *patch)
2225 {
2226 int files, adds, dels;
2228 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2229 files++;
2230 adds += patch->lines_added;
2231 dels += patch->lines_deleted;
2232 show_stats(patch);
2233 }
2235 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2236 }
2238 static void numstat_patch_list(struct patch *patch)
2239 {
2240 for ( ; patch; patch = patch->next) {
2241 const char *name;
2242 name = patch->new_name ? patch->new_name : patch->old_name;
2243 if (patch->is_binary)
2244 printf("-\t-\t");
2245 else
2246 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2247 write_name_quoted(name, stdout, line_termination);
2248 }
2249 }
2251 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2252 {
2253 if (mode)
2254 printf(" %s mode %06o %s\n", newdelete, mode, name);
2255 else
2256 printf(" %s %s\n", newdelete, name);
2257 }
2259 static void show_mode_change(struct patch *p, int show_name)
2260 {
2261 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2262 if (show_name)
2263 printf(" mode change %06o => %06o %s\n",
2264 p->old_mode, p->new_mode, p->new_name);
2265 else
2266 printf(" mode change %06o => %06o\n",
2267 p->old_mode, p->new_mode);
2268 }
2269 }
2271 static void show_rename_copy(struct patch *p)
2272 {
2273 const char *renamecopy = p->is_rename ? "rename" : "copy";
2274 const char *old, *new;
2276 /* Find common prefix */
2277 old = p->old_name;
2278 new = p->new_name;
2279 while (1) {
2280 const char *slash_old, *slash_new;
2281 slash_old = strchr(old, '/');
2282 slash_new = strchr(new, '/');
2283 if (!slash_old ||
2284 !slash_new ||
2285 slash_old - old != slash_new - new ||
2286 memcmp(old, new, slash_new - new))
2287 break;
2288 old = slash_old + 1;
2289 new = slash_new + 1;
2290 }
2291 /* p->old_name thru old is the common prefix, and old and new
2292 * through the end of names are renames
2293 */
2294 if (old != p->old_name)
2295 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2296 (int)(old - p->old_name), p->old_name,
2297 old, new, p->score);
2298 else
2299 printf(" %s %s => %s (%d%%)\n", renamecopy,
2300 p->old_name, p->new_name, p->score);
2301 show_mode_change(p, 0);
2302 }
2304 static void summary_patch_list(struct patch *patch)
2305 {
2306 struct patch *p;
2308 for (p = patch; p; p = p->next) {
2309 if (p->is_new)
2310 show_file_mode_name("create", p->new_mode, p->new_name);
2311 else if (p->is_delete)
2312 show_file_mode_name("delete", p->old_mode, p->old_name);
2313 else {
2314 if (p->is_rename || p->is_copy)
2315 show_rename_copy(p);
2316 else {
2317 if (p->score) {
2318 printf(" rewrite %s (%d%%)\n",
2319 p->new_name, p->score);
2320 show_mode_change(p, 0);
2321 }
2322 else
2323 show_mode_change(p, 1);
2324 }
2325 }
2326 }
2327 }
2329 static void patch_stats(struct patch *patch)
2330 {
2331 int lines = patch->lines_added + patch->lines_deleted;
2333 if (lines > max_change)
2334 max_change = lines;
2335 if (patch->old_name) {
2336 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2337 if (!len)
2338 len = strlen(patch->old_name);
2339 if (len > max_len)
2340 max_len = len;
2341 }
2342 if (patch->new_name) {
2343 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2344 if (!len)
2345 len = strlen(patch->new_name);
2346 if (len > max_len)
2347 max_len = len;
2348 }
2349 }
2351 static void remove_file(struct patch *patch, int rmdir_empty)
2352 {
2353 if (update_index) {
2354 if (remove_file_from_cache(patch->old_name) < 0)
2355 die("unable to remove %s from index", patch->old_name);
2356 }
2357 if (!cached) {
2358 if (S_ISGITLINK(patch->old_mode)) {
2359 if (rmdir(patch->old_name))
2360 warning("unable to remove submodule %s",
2361 patch->old_name);
2362 } else if (!unlink(patch->old_name) && rmdir_empty) {
2363 char *name = xstrdup(patch->old_name);
2364 char *end = strrchr(name, '/');
2365 while (end) {
2366 *end = 0;
2367 if (rmdir(name))
2368 break;
2369 end = strrchr(name, '/');
2370 }
2371 free(name);
2372 }
2373 }
2374 }
2376 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2377 {
2378 struct stat st;
2379 struct cache_entry *ce;
2380 int namelen = strlen(path);
2381 unsigned ce_size = cache_entry_size(namelen);
2383 if (!update_index)
2384 return;
2386 ce = xcalloc(1, ce_size);
2387 memcpy(ce->name, path, namelen);
2388 ce->ce_mode = create_ce_mode(mode);
2389 ce->ce_flags = htons(namelen);
2390 if (S_ISGITLINK(mode)) {
2391 const char *s = buf;
2393 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2394 die("corrupt patch for subproject %s", path);
2395 } else {
2396 if (!cached) {
2397 if (lstat(path, &st) < 0)
2398 die("unable to stat newly created file %s",
2399 path);
2400 fill_stat_cache_info(ce, &st);
2401 }
2402 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2403 die("unable to create backing store for newly created file %s", path);
2404 }
2405 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2406 die("unable to add cache entry for %s", path);
2407 }
2409 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2410 {
2411 int fd;
2412 struct strbuf nbuf;
2414 if (S_ISGITLINK(mode)) {
2415 struct stat st;
2416 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2417 return 0;
2418 return mkdir(path, 0777);
2419 }
2421 if (has_symlinks && S_ISLNK(mode))
2422 /* Although buf:size is counted string, it also is NUL
2423 * terminated.
2424 */
2425 return symlink(buf, path);
2427 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2428 if (fd < 0)
2429 return -1;
2431 strbuf_init(&nbuf, 0);
2432 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2433 size = nbuf.len;
2434 buf = nbuf.buf;
2435 }
2436 write_or_die(fd, buf, size);
2437 strbuf_release(&nbuf);
2439 if (close(fd) < 0)
2440 die("closing file %s: %s", path, strerror(errno));
2441 return 0;
2442 }
2444 /*
2445 * We optimistically assume that the directories exist,
2446 * which is true 99% of the time anyway. If they don't,
2447 * we create them and try again.
2448 */
2449 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2450 {
2451 if (cached)
2452 return;
2453 if (!try_create_file(path, mode, buf, size))
2454 return;
2456 if (errno == ENOENT) {
2457 if (safe_create_leading_directories(path))
2458 return;
2459 if (!try_create_file(path, mode, buf, size))
2460 return;
2461 }
2463 if (errno == EEXIST || errno == EACCES) {
2464 /* We may be trying to create a file where a directory
2465 * used to be.
2466 */
2467 struct stat st;
2468 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2469 errno = EEXIST;
2470 }
2472 if (errno == EEXIST) {
2473 unsigned int nr = getpid();
2475 for (;;) {
2476 const char *newpath;
2477 newpath = mkpath("%s~%u", path, nr);
2478 if (!try_create_file(newpath, mode, buf, size)) {
2479 if (!rename(newpath, path))
2480 return;
2481 unlink(newpath);
2482 break;
2483 }
2484 if (errno != EEXIST)
2485 break;
2486 ++nr;
2487 }
2488 }
2489 die("unable to write file %s mode %o", path, mode);
2490 }
2492 static void create_file(struct patch *patch)
2493 {
2494 char *path = patch->new_name;
2495 unsigned mode = patch->new_mode;
2496 unsigned long size = patch->resultsize;
2497 char *buf = patch->result;
2499 if (!mode)
2500 mode = S_IFREG | 0644;
2501 create_one_file(path, mode, buf, size);
2502 add_index_file(path, mode, buf, size);
2503 }
2505 /* phase zero is to remove, phase one is to create */
2506 static void write_out_one_result(struct patch *patch, int phase)
2507 {
2508 if (patch->is_delete > 0) {
2509 if (phase == 0)
2510 remove_file(patch, 1);
2511 return;
2512 }
2513 if (patch->is_new > 0 || patch->is_copy) {
2514 if (phase == 1)
2515 create_file(patch);
2516 return;
2517 }
2518 /*
2519 * Rename or modification boils down to the same
2520 * thing: remove the old, write the new
2521 */
2522 if (phase == 0)
2523 remove_file(patch, patch->is_rename);
2524 if (phase == 1)
2525 create_file(patch);
2526 }
2528 static int write_out_one_reject(struct patch *patch)
2529 {
2530 FILE *rej;
2531 char namebuf[PATH_MAX];
2532 struct fragment *frag;
2533 int cnt = 0;
2535 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2536 if (!frag->rejected)
2537 continue;
2538 cnt++;
2539 }
2541 if (!cnt) {
2542 if (apply_verbosely)
2543 say_patch_name(stderr,
2544 "Applied patch ", patch, " cleanly.\n");
2545 return 0;
2546 }
2548 /* This should not happen, because a removal patch that leaves
2549 * contents are marked "rejected" at the patch level.
2550 */
2551 if (!patch->new_name)
2552 die("internal error");
2554 /* Say this even without --verbose */
2555 say_patch_name(stderr, "Applying patch ", patch, " with");
2556 fprintf(stderr, " %d rejects...\n", cnt);
2558 cnt = strlen(patch->new_name);
2559 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2560 cnt = ARRAY_SIZE(namebuf) - 5;
2561 fprintf(stderr,
2562 "warning: truncating .rej filename to %.*s.rej",
2563 cnt - 1, patch->new_name);
2564 }
2565 memcpy(namebuf, patch->new_name, cnt);
2566 memcpy(namebuf + cnt, ".rej", 5);
2568 rej = fopen(namebuf, "w");
2569 if (!rej)
2570 return error("cannot open %s: %s", namebuf, strerror(errno));
2572 /* Normal git tools never deal with .rej, so do not pretend
2573 * this is a git patch by saying --git nor give extended
2574 * headers. While at it, maybe please "kompare" that wants
2575 * the trailing TAB and some garbage at the end of line ;-).
2576 */
2577 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2578 patch->new_name, patch->new_name);
2579 for (cnt = 1, frag = patch->fragments;
2580 frag;
2581 cnt++, frag = frag->next) {
2582 if (!frag->rejected) {
2583 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2584 continue;
2585 }
2586 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2587 fprintf(rej, "%.*s", frag->size, frag->patch);
2588 if (frag->patch[frag->size-1] != '\n')
2589 fputc('\n', rej);
2590 }
2591 fclose(rej);
2592 return -1;
2593 }
2595 static int write_out_results(struct patch *list, int skipped_patch)
2596 {
2597 int phase;
2598 int errs = 0;
2599 struct patch *l;
2601 if (!list && !skipped_patch)
2602 return error("No changes");
2604 for (phase = 0; phase < 2; phase++) {
2605 l = list;
2606 while (l) {
2607 if (l->rejected)
2608 errs = 1;
2609 else {
2610 write_out_one_result(l, phase);
2611 if (phase == 1 && write_out_one_reject(l))
2612 errs = 1;
2613 }
2614 l = l->next;
2615 }
2616 }
2617 return errs;
2618 }
2620 static struct lock_file lock_file;
2622 static struct excludes {
2623 struct excludes *next;
2624 const char *path;
2625 } *excludes;
2627 static int use_patch(struct patch *p)
2628 {
2629 const char *pathname = p->new_name ? p->new_name : p->old_name;
2630 struct excludes *x = excludes;
2631 while (x) {
2632 if (fnmatch(x->path, pathname, 0) == 0)
2633 return 0;
2634 x = x->next;
2635 }
2636 if (0 < prefix_length) {
2637 int pathlen = strlen(pathname);
2638 if (pathlen <= prefix_length ||
2639 memcmp(prefix, pathname, prefix_length))
2640 return 0;
2641 }
2642 return 1;
2643 }
2645 static void prefix_one(char **name)
2646 {
2647 char *old_name = *name;
2648 if (!old_name)
2649 return;
2650 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2651 free(old_name);
2652 }
2654 static void prefix_patches(struct patch *p)
2655 {
2656 if (!prefix || p->is_toplevel_relative)
2657 return;
2658 for ( ; p; p = p->next) {
2659 if (p->new_name == p->old_name) {
2660 char *prefixed = p->new_name;
2661 prefix_one(&prefixed);
2662 p->new_name = p->old_name = prefixed;
2663 }
2664 else {
2665 prefix_one(&p->new_name);
2666 prefix_one(&p->old_name);
2667 }
2668 }
2669 }
2671 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2672 {
2673 size_t offset;
2674 struct strbuf buf;
2675 struct patch *list = NULL, **listp = &list;
2676 int skipped_patch = 0;
2678 strbuf_init(&buf, 0);
2679 patch_input_file = filename;
2680 read_patch_file(&buf, fd);
2681 offset = 0;
2682 while (offset < buf.len) {
2683 struct patch *patch;
2684 int nr;
2686 patch = xcalloc(1, sizeof(*patch));
2687 patch->inaccurate_eof = inaccurate_eof;
2688 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2689 if (nr < 0)
2690 break;
2691 if (apply_in_reverse)
2692 reverse_patches(patch);
2693 if (prefix)
2694 prefix_patches(patch);
2695 if (use_patch(patch)) {
2696 patch_stats(patch);
2697 *listp = patch;
2698 listp = &patch->next;
2699 }
2700 else {
2701 /* perhaps free it a bit better? */
2702 free(patch);
2703 skipped_patch++;
2704 }
2705 offset += nr;
2706 }
2708 if (whitespace_error && (ws_error_action == die_on_ws_error))
2709 apply = 0;
2711 update_index = check_index && apply;
2712 if (update_index && newfd < 0)
2713 newfd = hold_locked_index(&lock_file, 1);
2715 if (check_index) {
2716 if (read_cache() < 0)
2717 die("unable to read index file");
2718 }
2720 if ((check || apply) &&
2721 check_patch_list(list) < 0 &&
2722 !apply_with_reject)
2723 exit(1);
2725 if (apply && write_out_results(list, skipped_patch))
2726 exit(1);
2728 if (fake_ancestor)
2729 build_fake_ancestor(list, fake_ancestor);
2731 if (diffstat)
2732 stat_patch_list(list);
2734 if (numstat)
2735 numstat_patch_list(list);
2737 if (summary)
2738 summary_patch_list(list);
2740 strbuf_release(&buf);
2741 return 0;
2742 }
2744 static int git_apply_config(const char *var, const char *value)
2745 {
2746 if (!strcmp(var, "apply.whitespace")) {
2747 apply_default_whitespace = xstrdup(value);
2748 return 0;
2749 }
2750 return git_default_config(var, value);
2751 }
2754 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2755 {
2756 int i;
2757 int read_stdin = 1;
2758 int inaccurate_eof = 0;
2759 int errs = 0;
2760 int is_not_gitdir = 0;
2762 const char *whitespace_option = NULL;
2764 prefix = setup_git_directory_gently(&is_not_gitdir);
2765 prefix_length = prefix ? strlen(prefix) : 0;
2766 git_config(git_apply_config);
2767 if (apply_default_whitespace)
2768 parse_whitespace_option(apply_default_whitespace);
2770 for (i = 1; i < argc; i++) {
2771 const char *arg = argv[i];
2772 char *end;
2773 int fd;
2775 if (!strcmp(arg, "-")) {
2776 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2777 read_stdin = 0;
2778 continue;
2779 }
2780 if (!prefixcmp(arg, "--exclude=")) {
2781 struct excludes *x = xmalloc(sizeof(*x));
2782 x->path = arg + 10;
2783 x->next = excludes;
2784 excludes = x;
2785 continue;
2786 }
2787 if (!prefixcmp(arg, "-p")) {
2788 p_value = atoi(arg + 2);
2789 p_value_known = 1;
2790 continue;
2791 }
2792 if (!strcmp(arg, "--no-add")) {
2793 no_add = 1;
2794 continue;
2795 }
2796 if (!strcmp(arg, "--stat")) {
2797 apply = 0;
2798 diffstat = 1;
2799 continue;
2800 }
2801 if (!strcmp(arg, "--allow-binary-replacement") ||
2802 !strcmp(arg, "--binary")) {
2803 continue; /* now no-op */
2804 }
2805 if (!strcmp(arg, "--numstat")) {
2806 apply = 0;
2807 numstat = 1;
2808 continue;
2809 }
2810 if (!strcmp(arg, "--summary")) {
2811 apply = 0;
2812 summary = 1;
2813 continue;
2814 }
2815 if (!strcmp(arg, "--check")) {
2816 apply = 0;
2817 check = 1;
2818 continue;
2819 }
2820 if (!strcmp(arg, "--index")) {
2821 if (is_not_gitdir)
2822 die("--index outside a repository");
2823 check_index = 1;
2824 continue;
2825 }
2826 if (!strcmp(arg, "--cached")) {
2827 if (is_not_gitdir)
2828 die("--cached outside a repository");
2829 check_index = 1;
2830 cached = 1;
2831 continue;
2832 }
2833 if (!strcmp(arg, "--apply")) {
2834 apply = 1;
2835 continue;
2836 }
2837 if (!strcmp(arg, "--build-fake-ancestor")) {
2838 apply = 0;
2839 if (++i >= argc)
2840 die ("need a filename");
2841 fake_ancestor = argv[i];
2842 continue;
2843 }
2844 if (!strcmp(arg, "-z")) {
2845 line_termination = 0;
2846 continue;
2847 }
2848 if (!prefixcmp(arg, "-C")) {
2849 p_context = strtoul(arg + 2, &end, 0);
2850 if (*end != '\0')
2851 die("unrecognized context count '%s'", arg + 2);
2852 continue;
2853 }
2854 if (!prefixcmp(arg, "--whitespace=")) {
2855 whitespace_option = arg + 13;
2856 parse_whitespace_option(arg + 13);
2857 continue;
2858 }
2859 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2860 apply_in_reverse = 1;
2861 continue;
2862 }
2863 if (!strcmp(arg, "--unidiff-zero")) {
2864 unidiff_zero = 1;
2865 continue;
2866 }
2867 if (!strcmp(arg, "--reject")) {
2868 apply = apply_with_reject = apply_verbosely = 1;
2869 continue;
2870 }
2871 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2872 apply_verbosely = 1;
2873 continue;
2874 }
2875 if (!strcmp(arg, "--inaccurate-eof")) {
2876 inaccurate_eof = 1;
2877 continue;
2878 }
2879 if (0 < prefix_length)
2880 arg = prefix_filename(prefix, prefix_length, arg);
2882 fd = open(arg, O_RDONLY);
2883 if (fd < 0)
2884 usage(apply_usage);
2885 read_stdin = 0;
2886 set_default_whitespace_mode(whitespace_option);
2887 errs |= apply_patch(fd, arg, inaccurate_eof);
2888 close(fd);
2889 }
2890 set_default_whitespace_mode(whitespace_option);
2891 if (read_stdin)
2892 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2893 if (whitespace_error) {
2894 if (squelch_whitespace_errors &&
2895 squelch_whitespace_errors < whitespace_error) {
2896 int squelched =
2897 whitespace_error - squelch_whitespace_errors;
2898 fprintf(stderr, "warning: squelched %d "
2899 "whitespace error%s\n",
2900 squelched,
2901 squelched == 1 ? "" : "s");
2902 }
2903 if (ws_error_action == die_on_ws_error)
2904 die("%d line%s add%s whitespace errors.",
2905 whitespace_error,
2906 whitespace_error == 1 ? "" : "s",
2907 whitespace_error == 1 ? "s" : "");
2908 if (applied_after_fixing_ws)
2909 fprintf(stderr, "warning: %d line%s applied after"
2910 " fixing whitespace errors.\n",
2911 applied_after_fixing_ws,
2912 applied_after_fixing_ws == 1 ? "" : "s");
2913 else if (whitespace_error)
2914 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2915 whitespace_error,
2916 whitespace_error == 1 ? "" : "s",
2917 whitespace_error == 1 ? "s" : "");
2918 }
2920 if (update_index) {
2921 if (write_cache(newfd, active_cache, active_nr) ||
2922 close(newfd) || commit_locked_index(&lock_file))
2923 die("Unable to write new index file");
2924 }
2926 return !!errs;
2927 }