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|fix|error|error-all>] <patch>...";
50 static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55 } ws_error_action = warn_on_ws_error;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
61 static void parse_whitespace_option(const char *option)
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 ws_rule;
148 unsigned long deflate_origlen;
149 int lines_added, lines_deleted;
150 int score;
151 unsigned int is_toplevel_relative:1;
152 unsigned int inaccurate_eof:1;
153 unsigned int is_binary:1;
154 unsigned int is_copy:1;
155 unsigned int is_rename:1;
156 struct fragment *fragments;
157 char *result;
158 size_t resultsize;
159 char old_sha1_prefix[41];
160 char new_sha1_prefix[41];
161 struct patch *next;
162 };
164 static void say_patch_name(FILE *output, const char *pre,
165 struct patch *patch, const char *post)
166 {
167 fputs(pre, output);
168 if (patch->old_name && patch->new_name &&
169 strcmp(patch->old_name, patch->new_name)) {
170 quote_c_style(patch->old_name, NULL, output, 0);
171 fputs(" => ", output);
172 quote_c_style(patch->new_name, NULL, output, 0);
173 } else {
174 const char *n = patch->new_name;
175 if (!n)
176 n = patch->old_name;
177 quote_c_style(n, NULL, output, 0);
178 }
179 fputs(post, output);
180 }
182 #define CHUNKSIZE (8192)
183 #define SLOP (16)
185 static void read_patch_file(struct strbuf *sb, int fd)
186 {
187 if (strbuf_read(sb, fd, 0) < 0)
188 die("git-apply: read returned %s", strerror(errno));
190 /*
191 * Make sure that we have some slop in the buffer
192 * so that we can do speculative "memcmp" etc, and
193 * see to it that it is NUL-filled.
194 */
195 strbuf_grow(sb, SLOP);
196 memset(sb->buf + sb->len, 0, SLOP);
197 }
199 static unsigned long linelen(const char *buffer, unsigned long size)
200 {
201 unsigned long len = 0;
202 while (size--) {
203 len++;
204 if (*buffer++ == '\n')
205 break;
206 }
207 return len;
208 }
210 static int is_dev_null(const char *str)
211 {
212 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
213 }
215 #define TERM_SPACE 1
216 #define TERM_TAB 2
218 static int name_terminate(const char *name, int namelen, int c, int terminate)
219 {
220 if (c == ' ' && !(terminate & TERM_SPACE))
221 return 0;
222 if (c == '\t' && !(terminate & TERM_TAB))
223 return 0;
225 return 1;
226 }
228 static char *find_name(const char *line, char *def, int p_value, int terminate)
229 {
230 int len;
231 const char *start = line;
233 if (*line == '"') {
234 struct strbuf name;
236 /*
237 * Proposed "new-style" GNU patch/diff format; see
238 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
239 */
240 strbuf_init(&name, 0);
241 if (!unquote_c_style(&name, line, NULL)) {
242 char *cp;
244 for (cp = name.buf; p_value; p_value--) {
245 cp = strchr(cp, '/');
246 if (!cp)
247 break;
248 cp++;
249 }
250 if (cp) {
251 /* name can later be freed, so we need
252 * to memmove, not just return cp
253 */
254 strbuf_remove(&name, 0, cp - name.buf);
255 free(def);
256 return strbuf_detach(&name, NULL);
257 }
258 }
259 strbuf_release(&name);
260 }
262 for (;;) {
263 char c = *line;
265 if (isspace(c)) {
266 if (c == '\n')
267 break;
268 if (name_terminate(start, line-start, c, terminate))
269 break;
270 }
271 line++;
272 if (c == '/' && !--p_value)
273 start = line;
274 }
275 if (!start)
276 return def;
277 len = line - start;
278 if (!len)
279 return def;
281 /*
282 * Generally we prefer the shorter name, especially
283 * if the other one is just a variation of that with
284 * something else tacked on to the end (ie "file.orig"
285 * or "file~").
286 */
287 if (def) {
288 int deflen = strlen(def);
289 if (deflen < len && !strncmp(start, def, deflen))
290 return def;
291 free(def);
292 }
294 return xmemdupz(start, len);
295 }
297 static int count_slashes(const char *cp)
298 {
299 int cnt = 0;
300 char ch;
302 while ((ch = *cp++))
303 if (ch == '/')
304 cnt++;
305 return cnt;
306 }
308 /*
309 * Given the string after "--- " or "+++ ", guess the appropriate
310 * p_value for the given patch.
311 */
312 static int guess_p_value(const char *nameline)
313 {
314 char *name, *cp;
315 int val = -1;
317 if (is_dev_null(nameline))
318 return -1;
319 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
320 if (!name)
321 return -1;
322 cp = strchr(name, '/');
323 if (!cp)
324 val = 0;
325 else if (prefix) {
326 /*
327 * Does it begin with "a/$our-prefix" and such? Then this is
328 * very likely to apply to our directory.
329 */
330 if (!strncmp(name, prefix, prefix_length))
331 val = count_slashes(prefix);
332 else {
333 cp++;
334 if (!strncmp(cp, prefix, prefix_length))
335 val = count_slashes(prefix) + 1;
336 }
337 }
338 free(name);
339 return val;
340 }
342 /*
343 * Get the name etc info from the --/+++ lines of a traditional patch header
344 *
345 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
346 * files, we can happily check the index for a match, but for creating a
347 * new file we should try to match whatever "patch" does. I have no idea.
348 */
349 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
350 {
351 char *name;
353 first += 4; /* skip "--- " */
354 second += 4; /* skip "+++ " */
355 if (!p_value_known) {
356 int p, q;
357 p = guess_p_value(first);
358 q = guess_p_value(second);
359 if (p < 0) p = q;
360 if (0 <= p && p == q) {
361 p_value = p;
362 p_value_known = 1;
363 }
364 }
365 if (is_dev_null(first)) {
366 patch->is_new = 1;
367 patch->is_delete = 0;
368 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
369 patch->new_name = name;
370 } else if (is_dev_null(second)) {
371 patch->is_new = 0;
372 patch->is_delete = 1;
373 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
374 patch->old_name = name;
375 } else {
376 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
377 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
378 patch->old_name = patch->new_name = name;
379 }
380 if (!name)
381 die("unable to find filename in patch at line %d", linenr);
382 }
384 static int gitdiff_hdrend(const char *line, struct patch *patch)
385 {
386 return -1;
387 }
389 /*
390 * We're anal about diff header consistency, to make
391 * sure that we don't end up having strange ambiguous
392 * patches floating around.
393 *
394 * As a result, gitdiff_{old|new}name() will check
395 * their names against any previous information, just
396 * to make sure..
397 */
398 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
399 {
400 if (!orig_name && !isnull)
401 return find_name(line, NULL, p_value, TERM_TAB);
403 if (orig_name) {
404 int len;
405 const char *name;
406 char *another;
407 name = orig_name;
408 len = strlen(name);
409 if (isnull)
410 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
411 another = find_name(line, NULL, p_value, TERM_TAB);
412 if (!another || memcmp(another, name, len))
413 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
414 free(another);
415 return orig_name;
416 }
417 else {
418 /* expect "/dev/null" */
419 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
420 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
421 return NULL;
422 }
423 }
425 static int gitdiff_oldname(const char *line, struct patch *patch)
426 {
427 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
428 return 0;
429 }
431 static int gitdiff_newname(const char *line, struct patch *patch)
432 {
433 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
434 return 0;
435 }
437 static int gitdiff_oldmode(const char *line, struct patch *patch)
438 {
439 patch->old_mode = strtoul(line, NULL, 8);
440 return 0;
441 }
443 static int gitdiff_newmode(const char *line, struct patch *patch)
444 {
445 patch->new_mode = strtoul(line, NULL, 8);
446 return 0;
447 }
449 static int gitdiff_delete(const char *line, struct patch *patch)
450 {
451 patch->is_delete = 1;
452 patch->old_name = patch->def_name;
453 return gitdiff_oldmode(line, patch);
454 }
456 static int gitdiff_newfile(const char *line, struct patch *patch)
457 {
458 patch->is_new = 1;
459 patch->new_name = patch->def_name;
460 return gitdiff_newmode(line, patch);
461 }
463 static int gitdiff_copysrc(const char *line, struct patch *patch)
464 {
465 patch->is_copy = 1;
466 patch->old_name = find_name(line, NULL, 0, 0);
467 return 0;
468 }
470 static int gitdiff_copydst(const char *line, struct patch *patch)
471 {
472 patch->is_copy = 1;
473 patch->new_name = find_name(line, NULL, 0, 0);
474 return 0;
475 }
477 static int gitdiff_renamesrc(const char *line, struct patch *patch)
478 {
479 patch->is_rename = 1;
480 patch->old_name = find_name(line, NULL, 0, 0);
481 return 0;
482 }
484 static int gitdiff_renamedst(const char *line, struct patch *patch)
485 {
486 patch->is_rename = 1;
487 patch->new_name = find_name(line, NULL, 0, 0);
488 return 0;
489 }
491 static int gitdiff_similarity(const char *line, struct patch *patch)
492 {
493 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
494 patch->score = 0;
495 return 0;
496 }
498 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
499 {
500 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
501 patch->score = 0;
502 return 0;
503 }
505 static int gitdiff_index(const char *line, struct patch *patch)
506 {
507 /*
508 * index line is N hexadecimal, "..", N hexadecimal,
509 * and optional space with octal mode.
510 */
511 const char *ptr, *eol;
512 int len;
514 ptr = strchr(line, '.');
515 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
516 return 0;
517 len = ptr - line;
518 memcpy(patch->old_sha1_prefix, line, len);
519 patch->old_sha1_prefix[len] = 0;
521 line = ptr + 2;
522 ptr = strchr(line, ' ');
523 eol = strchr(line, '\n');
525 if (!ptr || eol < ptr)
526 ptr = eol;
527 len = ptr - line;
529 if (40 < len)
530 return 0;
531 memcpy(patch->new_sha1_prefix, line, len);
532 patch->new_sha1_prefix[len] = 0;
533 if (*ptr == ' ')
534 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
535 return 0;
536 }
538 /*
539 * This is normal for a diff that doesn't change anything: we'll fall through
540 * into the next diff. Tell the parser to break out.
541 */
542 static int gitdiff_unrecognized(const char *line, struct patch *patch)
543 {
544 return -1;
545 }
547 static const char *stop_at_slash(const char *line, int llen)
548 {
549 int i;
551 for (i = 0; i < llen; i++) {
552 int ch = line[i];
553 if (ch == '/')
554 return line + i;
555 }
556 return NULL;
557 }
559 /*
560 * This is to extract the same name that appears on "diff --git"
561 * line. We do not find and return anything if it is a rename
562 * patch, and it is OK because we will find the name elsewhere.
563 * We need to reliably find name only when it is mode-change only,
564 * creation or deletion of an empty file. In any of these cases,
565 * both sides are the same name under a/ and b/ respectively.
566 */
567 static char *git_header_name(char *line, int llen)
568 {
569 const char *name;
570 const char *second = NULL;
571 size_t len;
573 line += strlen("diff --git ");
574 llen -= strlen("diff --git ");
576 if (*line == '"') {
577 const char *cp;
578 struct strbuf first;
579 struct strbuf sp;
581 strbuf_init(&first, 0);
582 strbuf_init(&sp, 0);
584 if (unquote_c_style(&first, line, &second))
585 goto free_and_fail1;
587 /* advance to the first slash */
588 cp = stop_at_slash(first.buf, first.len);
589 /* we do not accept absolute paths */
590 if (!cp || cp == first.buf)
591 goto free_and_fail1;
592 strbuf_remove(&first, 0, cp + 1 - first.buf);
594 /*
595 * second points at one past closing dq of name.
596 * find the second name.
597 */
598 while ((second < line + llen) && isspace(*second))
599 second++;
601 if (line + llen <= second)
602 goto free_and_fail1;
603 if (*second == '"') {
604 if (unquote_c_style(&sp, second, NULL))
605 goto free_and_fail1;
606 cp = stop_at_slash(sp.buf, sp.len);
607 if (!cp || cp == sp.buf)
608 goto free_and_fail1;
609 /* They must match, otherwise ignore */
610 if (strcmp(cp + 1, first.buf))
611 goto free_and_fail1;
612 strbuf_release(&sp);
613 return strbuf_detach(&first, NULL);
614 }
616 /* unquoted second */
617 cp = stop_at_slash(second, line + llen - second);
618 if (!cp || cp == second)
619 goto free_and_fail1;
620 cp++;
621 if (line + llen - cp != first.len + 1 ||
622 memcmp(first.buf, cp, first.len))
623 goto free_and_fail1;
624 return strbuf_detach(&first, NULL);
626 free_and_fail1:
627 strbuf_release(&first);
628 strbuf_release(&sp);
629 return NULL;
630 }
632 /* unquoted first name */
633 name = stop_at_slash(line, llen);
634 if (!name || name == line)
635 return NULL;
636 name++;
638 /*
639 * since the first name is unquoted, a dq if exists must be
640 * the beginning of the second name.
641 */
642 for (second = name; second < line + llen; second++) {
643 if (*second == '"') {
644 struct strbuf sp;
645 const char *np;
647 strbuf_init(&sp, 0);
648 if (unquote_c_style(&sp, second, NULL))
649 goto free_and_fail2;
651 np = stop_at_slash(sp.buf, sp.len);
652 if (!np || np == sp.buf)
653 goto free_and_fail2;
654 np++;
656 len = sp.buf + sp.len - np;
657 if (len < second - name &&
658 !strncmp(np, name, len) &&
659 isspace(name[len])) {
660 /* Good */
661 strbuf_remove(&sp, 0, np - sp.buf);
662 return strbuf_detach(&sp, NULL);
663 }
665 free_and_fail2:
666 strbuf_release(&sp);
667 return NULL;
668 }
669 }
671 /*
672 * Accept a name only if it shows up twice, exactly the same
673 * form.
674 */
675 for (len = 0 ; ; len++) {
676 switch (name[len]) {
677 default:
678 continue;
679 case '\n':
680 return NULL;
681 case '\t': case ' ':
682 second = name+len;
683 for (;;) {
684 char c = *second++;
685 if (c == '\n')
686 return NULL;
687 if (c == '/')
688 break;
689 }
690 if (second[len] == '\n' && !memcmp(name, second, len)) {
691 return xmemdupz(name, len);
692 }
693 }
694 }
695 return NULL;
696 }
698 /* Verify that we recognize the lines following a git header */
699 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
700 {
701 unsigned long offset;
703 /* A git diff has explicit new/delete information, so we don't guess */
704 patch->is_new = 0;
705 patch->is_delete = 0;
707 /*
708 * Some things may not have the old name in the
709 * rest of the headers anywhere (pure mode changes,
710 * or removing or adding empty files), so we get
711 * the default name from the header.
712 */
713 patch->def_name = git_header_name(line, len);
715 line += len;
716 size -= len;
717 linenr++;
718 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
719 static const struct opentry {
720 const char *str;
721 int (*fn)(const char *, struct patch *);
722 } optable[] = {
723 { "@@ -", gitdiff_hdrend },
724 { "--- ", gitdiff_oldname },
725 { "+++ ", gitdiff_newname },
726 { "old mode ", gitdiff_oldmode },
727 { "new mode ", gitdiff_newmode },
728 { "deleted file mode ", gitdiff_delete },
729 { "new file mode ", gitdiff_newfile },
730 { "copy from ", gitdiff_copysrc },
731 { "copy to ", gitdiff_copydst },
732 { "rename old ", gitdiff_renamesrc },
733 { "rename new ", gitdiff_renamedst },
734 { "rename from ", gitdiff_renamesrc },
735 { "rename to ", gitdiff_renamedst },
736 { "similarity index ", gitdiff_similarity },
737 { "dissimilarity index ", gitdiff_dissimilarity },
738 { "index ", gitdiff_index },
739 { "", gitdiff_unrecognized },
740 };
741 int i;
743 len = linelen(line, size);
744 if (!len || line[len-1] != '\n')
745 break;
746 for (i = 0; i < ARRAY_SIZE(optable); i++) {
747 const struct opentry *p = optable + i;
748 int oplen = strlen(p->str);
749 if (len < oplen || memcmp(p->str, line, oplen))
750 continue;
751 if (p->fn(line + oplen, patch) < 0)
752 return offset;
753 break;
754 }
755 }
757 return offset;
758 }
760 static int parse_num(const char *line, unsigned long *p)
761 {
762 char *ptr;
764 if (!isdigit(*line))
765 return 0;
766 *p = strtoul(line, &ptr, 10);
767 return ptr - line;
768 }
770 static int parse_range(const char *line, int len, int offset, const char *expect,
771 unsigned long *p1, unsigned long *p2)
772 {
773 int digits, ex;
775 if (offset < 0 || offset >= len)
776 return -1;
777 line += offset;
778 len -= offset;
780 digits = parse_num(line, p1);
781 if (!digits)
782 return -1;
784 offset += digits;
785 line += digits;
786 len -= digits;
788 *p2 = 1;
789 if (*line == ',') {
790 digits = parse_num(line+1, p2);
791 if (!digits)
792 return -1;
794 offset += digits+1;
795 line += digits+1;
796 len -= digits+1;
797 }
799 ex = strlen(expect);
800 if (ex > len)
801 return -1;
802 if (memcmp(line, expect, ex))
803 return -1;
805 return offset + ex;
806 }
808 /*
809 * Parse a unified diff fragment header of the
810 * form "@@ -a,b +c,d @@"
811 */
812 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
813 {
814 int offset;
816 if (!len || line[len-1] != '\n')
817 return -1;
819 /* Figure out the number of lines in a fragment */
820 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
821 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
823 return offset;
824 }
826 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
827 {
828 unsigned long offset, len;
830 patch->is_toplevel_relative = 0;
831 patch->is_rename = patch->is_copy = 0;
832 patch->is_new = patch->is_delete = -1;
833 patch->old_mode = patch->new_mode = 0;
834 patch->old_name = patch->new_name = NULL;
835 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
836 unsigned long nextlen;
838 len = linelen(line, size);
839 if (!len)
840 break;
842 /* Testing this early allows us to take a few shortcuts.. */
843 if (len < 6)
844 continue;
846 /*
847 * Make sure we don't find any unconnected patch fragments.
848 * That's a sign that we didn't find a header, and that a
849 * patch has become corrupted/broken up.
850 */
851 if (!memcmp("@@ -", line, 4)) {
852 struct fragment dummy;
853 if (parse_fragment_header(line, len, &dummy) < 0)
854 continue;
855 die("patch fragment without header at line %d: %.*s",
856 linenr, (int)len-1, line);
857 }
859 if (size < len + 6)
860 break;
862 /*
863 * Git patch? It might not have a real patch, just a rename
864 * or mode change, so we handle that specially
865 */
866 if (!memcmp("diff --git ", line, 11)) {
867 int git_hdr_len = parse_git_header(line, len, size, patch);
868 if (git_hdr_len <= len)
869 continue;
870 if (!patch->old_name && !patch->new_name) {
871 if (!patch->def_name)
872 die("git diff header lacks filename information (line %d)", linenr);
873 patch->old_name = patch->new_name = patch->def_name;
874 }
875 patch->is_toplevel_relative = 1;
876 *hdrsize = git_hdr_len;
877 return offset;
878 }
880 /* --- followed by +++ ? */
881 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
882 continue;
884 /*
885 * We only accept unified patches, so we want it to
886 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
887 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
888 */
889 nextlen = linelen(line + len, size - len);
890 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
891 continue;
893 /* Ok, we'll consider it a patch */
894 parse_traditional_patch(line, line+len, patch);
895 *hdrsize = len + nextlen;
896 linenr += 2;
897 return offset;
898 }
899 return -1;
900 }
902 static void check_whitespace(const char *line, int len, unsigned ws_rule)
903 {
904 const char *err = "Adds trailing whitespace";
905 int seen_space = 0;
906 int i;
908 /*
909 * We know len is at least two, since we have a '+' and we
910 * checked that the last character was a '\n' before calling
911 * this function. That is, an addition of an empty line would
912 * check the '+' here. Sneaky...
913 */
914 if ((ws_rule & WS_TRAILING_SPACE) && isspace(line[len-2]))
915 goto error;
917 /*
918 * Make sure that there is no space followed by a tab in
919 * indentation.
920 */
921 if (ws_rule & WS_SPACE_BEFORE_TAB) {
922 err = "Space in indent is followed by a tab";
923 for (i = 1; i < len; i++) {
924 if (line[i] == '\t') {
925 if (seen_space)
926 goto error;
927 }
928 else if (line[i] == ' ')
929 seen_space = 1;
930 else
931 break;
932 }
933 }
935 /*
936 * Make sure that the indentation does not contain more than
937 * 8 spaces.
938 */
939 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
940 (8 < len) && !strncmp("+ ", line, 9)) {
941 err = "Indent more than 8 places with spaces";
942 goto error;
943 }
944 return;
946 error:
947 whitespace_error++;
948 if (squelch_whitespace_errors &&
949 squelch_whitespace_errors < whitespace_error)
950 ;
951 else
952 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
953 err, patch_input_file, linenr, len-2, line+1);
954 }
956 /*
957 * Parse a unified diff. Note that this really needs to parse each
958 * fragment separately, since the only way to know the difference
959 * between a "---" that is part of a patch, and a "---" that starts
960 * the next patch is to look at the line counts..
961 */
962 static int parse_fragment(char *line, unsigned long size,
963 struct patch *patch, struct fragment *fragment)
964 {
965 int added, deleted;
966 int len = linelen(line, size), offset;
967 unsigned long oldlines, newlines;
968 unsigned long leading, trailing;
970 offset = parse_fragment_header(line, len, fragment);
971 if (offset < 0)
972 return -1;
973 oldlines = fragment->oldlines;
974 newlines = fragment->newlines;
975 leading = 0;
976 trailing = 0;
978 /* Parse the thing.. */
979 line += len;
980 size -= len;
981 linenr++;
982 added = deleted = 0;
983 for (offset = len;
984 0 < size;
985 offset += len, size -= len, line += len, linenr++) {
986 if (!oldlines && !newlines)
987 break;
988 len = linelen(line, size);
989 if (!len || line[len-1] != '\n')
990 return -1;
991 switch (*line) {
992 default:
993 return -1;
994 case '\n': /* newer GNU diff, an empty context line */
995 case ' ':
996 oldlines--;
997 newlines--;
998 if (!deleted && !added)
999 leading++;
1000 trailing++;
1001 break;
1002 case '-':
1003 if (apply_in_reverse &&
1004 ws_error_action != nowarn_ws_error)
1005 check_whitespace(line, len, patch->ws_rule);
1006 deleted++;
1007 oldlines--;
1008 trailing = 0;
1009 break;
1010 case '+':
1011 if (!apply_in_reverse &&
1012 ws_error_action != nowarn_ws_error)
1013 check_whitespace(line, len, patch->ws_rule);
1014 added++;
1015 newlines--;
1016 trailing = 0;
1017 break;
1019 /*
1020 * We allow "\ No newline at end of file". Depending
1021 * on locale settings when the patch was produced we
1022 * don't know what this line looks like. The only
1023 * thing we do know is that it begins with "\ ".
1024 * Checking for 12 is just for sanity check -- any
1025 * l10n of "\ No newline..." is at least that long.
1026 */
1027 case '\\':
1028 if (len < 12 || memcmp(line, "\\ ", 2))
1029 return -1;
1030 break;
1031 }
1032 }
1033 if (oldlines || newlines)
1034 return -1;
1035 fragment->leading = leading;
1036 fragment->trailing = trailing;
1038 /*
1039 * If a fragment ends with an incomplete line, we failed to include
1040 * it in the above loop because we hit oldlines == newlines == 0
1041 * before seeing it.
1042 */
1043 if (12 < size && !memcmp(line, "\\ ", 2))
1044 offset += linelen(line, size);
1046 patch->lines_added += added;
1047 patch->lines_deleted += deleted;
1049 if (0 < patch->is_new && oldlines)
1050 return error("new file depends on old contents");
1051 if (0 < patch->is_delete && newlines)
1052 return error("deleted file still has contents");
1053 return offset;
1054 }
1056 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1057 {
1058 unsigned long offset = 0;
1059 unsigned long oldlines = 0, newlines = 0, context = 0;
1060 struct fragment **fragp = &patch->fragments;
1062 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1063 struct fragment *fragment;
1064 int len;
1066 fragment = xcalloc(1, sizeof(*fragment));
1067 len = parse_fragment(line, size, patch, fragment);
1068 if (len <= 0)
1069 die("corrupt patch at line %d", linenr);
1070 fragment->patch = line;
1071 fragment->size = len;
1072 oldlines += fragment->oldlines;
1073 newlines += fragment->newlines;
1074 context += fragment->leading + fragment->trailing;
1076 *fragp = fragment;
1077 fragp = &fragment->next;
1079 offset += len;
1080 line += len;
1081 size -= len;
1082 }
1084 /*
1085 * If something was removed (i.e. we have old-lines) it cannot
1086 * be creation, and if something was added it cannot be
1087 * deletion. However, the reverse is not true; --unified=0
1088 * patches that only add are not necessarily creation even
1089 * though they do not have any old lines, and ones that only
1090 * delete are not necessarily deletion.
1091 *
1092 * Unfortunately, a real creation/deletion patch do _not_ have
1093 * any context line by definition, so we cannot safely tell it
1094 * apart with --unified=0 insanity. At least if the patch has
1095 * more than one hunk it is not creation or deletion.
1096 */
1097 if (patch->is_new < 0 &&
1098 (oldlines || (patch->fragments && patch->fragments->next)))
1099 patch->is_new = 0;
1100 if (patch->is_delete < 0 &&
1101 (newlines || (patch->fragments && patch->fragments->next)))
1102 patch->is_delete = 0;
1103 if (!unidiff_zero || context) {
1104 /* If the user says the patch is not generated with
1105 * --unified=0, or if we have seen context lines,
1106 * then not having oldlines means the patch is creation,
1107 * and not having newlines means the patch is deletion.
1108 */
1109 if (patch->is_new < 0 && !oldlines) {
1110 patch->is_new = 1;
1111 patch->old_name = NULL;
1112 }
1113 if (patch->is_delete < 0 && !newlines) {
1114 patch->is_delete = 1;
1115 patch->new_name = NULL;
1116 }
1117 }
1119 if (0 < patch->is_new && oldlines)
1120 die("new file %s depends on old contents", patch->new_name);
1121 if (0 < patch->is_delete && newlines)
1122 die("deleted file %s still has contents", patch->old_name);
1123 if (!patch->is_delete && !newlines && context)
1124 fprintf(stderr, "** warning: file %s becomes empty but "
1125 "is not deleted\n", patch->new_name);
1127 return offset;
1128 }
1130 static inline int metadata_changes(struct patch *patch)
1131 {
1132 return patch->is_rename > 0 ||
1133 patch->is_copy > 0 ||
1134 patch->is_new > 0 ||
1135 patch->is_delete ||
1136 (patch->old_mode && patch->new_mode &&
1137 patch->old_mode != patch->new_mode);
1138 }
1140 static char *inflate_it(const void *data, unsigned long size,
1141 unsigned long inflated_size)
1142 {
1143 z_stream stream;
1144 void *out;
1145 int st;
1147 memset(&stream, 0, sizeof(stream));
1149 stream.next_in = (unsigned char *)data;
1150 stream.avail_in = size;
1151 stream.next_out = out = xmalloc(inflated_size);
1152 stream.avail_out = inflated_size;
1153 inflateInit(&stream);
1154 st = inflate(&stream, Z_FINISH);
1155 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1156 free(out);
1157 return NULL;
1158 }
1159 return out;
1160 }
1162 static struct fragment *parse_binary_hunk(char **buf_p,
1163 unsigned long *sz_p,
1164 int *status_p,
1165 int *used_p)
1166 {
1167 /*
1168 * Expect a line that begins with binary patch method ("literal"
1169 * or "delta"), followed by the length of data before deflating.
1170 * a sequence of 'length-byte' followed by base-85 encoded data
1171 * should follow, terminated by a newline.
1172 *
1173 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1174 * and we would limit the patch line to 66 characters,
1175 * so one line can fit up to 13 groups that would decode
1176 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1177 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1178 */
1179 int llen, used;
1180 unsigned long size = *sz_p;
1181 char *buffer = *buf_p;
1182 int patch_method;
1183 unsigned long origlen;
1184 char *data = NULL;
1185 int hunk_size = 0;
1186 struct fragment *frag;
1188 llen = linelen(buffer, size);
1189 used = llen;
1191 *status_p = 0;
1193 if (!prefixcmp(buffer, "delta ")) {
1194 patch_method = BINARY_DELTA_DEFLATED;
1195 origlen = strtoul(buffer + 6, NULL, 10);
1196 }
1197 else if (!prefixcmp(buffer, "literal ")) {
1198 patch_method = BINARY_LITERAL_DEFLATED;
1199 origlen = strtoul(buffer + 8, NULL, 10);
1200 }
1201 else
1202 return NULL;
1204 linenr++;
1205 buffer += llen;
1206 while (1) {
1207 int byte_length, max_byte_length, newsize;
1208 llen = linelen(buffer, size);
1209 used += llen;
1210 linenr++;
1211 if (llen == 1) {
1212 /* consume the blank line */
1213 buffer++;
1214 size--;
1215 break;
1216 }
1217 /*
1218 * Minimum line is "A00000\n" which is 7-byte long,
1219 * and the line length must be multiple of 5 plus 2.
1220 */
1221 if ((llen < 7) || (llen-2) % 5)
1222 goto corrupt;
1223 max_byte_length = (llen - 2) / 5 * 4;
1224 byte_length = *buffer;
1225 if ('A' <= byte_length && byte_length <= 'Z')
1226 byte_length = byte_length - 'A' + 1;
1227 else if ('a' <= byte_length && byte_length <= 'z')
1228 byte_length = byte_length - 'a' + 27;
1229 else
1230 goto corrupt;
1231 /* if the input length was not multiple of 4, we would
1232 * have filler at the end but the filler should never
1233 * exceed 3 bytes
1234 */
1235 if (max_byte_length < byte_length ||
1236 byte_length <= max_byte_length - 4)
1237 goto corrupt;
1238 newsize = hunk_size + byte_length;
1239 data = xrealloc(data, newsize);
1240 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1241 goto corrupt;
1242 hunk_size = newsize;
1243 buffer += llen;
1244 size -= llen;
1245 }
1247 frag = xcalloc(1, sizeof(*frag));
1248 frag->patch = inflate_it(data, hunk_size, origlen);
1249 if (!frag->patch)
1250 goto corrupt;
1251 free(data);
1252 frag->size = origlen;
1253 *buf_p = buffer;
1254 *sz_p = size;
1255 *used_p = used;
1256 frag->binary_patch_method = patch_method;
1257 return frag;
1259 corrupt:
1260 free(data);
1261 *status_p = -1;
1262 error("corrupt binary patch at line %d: %.*s",
1263 linenr-1, llen-1, buffer);
1264 return NULL;
1265 }
1267 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1268 {
1269 /*
1270 * We have read "GIT binary patch\n"; what follows is a line
1271 * that says the patch method (currently, either "literal" or
1272 * "delta") and the length of data before deflating; a
1273 * sequence of 'length-byte' followed by base-85 encoded data
1274 * follows.
1275 *
1276 * When a binary patch is reversible, there is another binary
1277 * hunk in the same format, starting with patch method (either
1278 * "literal" or "delta") with the length of data, and a sequence
1279 * of length-byte + base-85 encoded data, terminated with another
1280 * empty line. This data, when applied to the postimage, produces
1281 * the preimage.
1282 */
1283 struct fragment *forward;
1284 struct fragment *reverse;
1285 int status;
1286 int used, used_1;
1288 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1289 if (!forward && !status)
1290 /* there has to be one hunk (forward hunk) */
1291 return error("unrecognized binary patch at line %d", linenr-1);
1292 if (status)
1293 /* otherwise we already gave an error message */
1294 return status;
1296 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1297 if (reverse)
1298 used += used_1;
1299 else if (status) {
1300 /*
1301 * Not having reverse hunk is not an error, but having
1302 * a corrupt reverse hunk is.
1303 */
1304 free((void*) forward->patch);
1305 free(forward);
1306 return status;
1307 }
1308 forward->next = reverse;
1309 patch->fragments = forward;
1310 patch->is_binary = 1;
1311 return used;
1312 }
1314 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1315 {
1316 int hdrsize, patchsize;
1317 int offset = find_header(buffer, size, &hdrsize, patch);
1319 if (offset < 0)
1320 return offset;
1322 patch->ws_rule = whitespace_rule(patch->new_name
1323 ? patch->new_name
1324 : patch->old_name);
1326 patchsize = parse_single_patch(buffer + offset + hdrsize,
1327 size - offset - hdrsize, patch);
1329 if (!patchsize) {
1330 static const char *binhdr[] = {
1331 "Binary files ",
1332 "Files ",
1333 NULL,
1334 };
1335 static const char git_binary[] = "GIT binary patch\n";
1336 int i;
1337 int hd = hdrsize + offset;
1338 unsigned long llen = linelen(buffer + hd, size - hd);
1340 if (llen == sizeof(git_binary) - 1 &&
1341 !memcmp(git_binary, buffer + hd, llen)) {
1342 int used;
1343 linenr++;
1344 used = parse_binary(buffer + hd + llen,
1345 size - hd - llen, patch);
1346 if (used)
1347 patchsize = used + llen;
1348 else
1349 patchsize = 0;
1350 }
1351 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1352 for (i = 0; binhdr[i]; i++) {
1353 int len = strlen(binhdr[i]);
1354 if (len < size - hd &&
1355 !memcmp(binhdr[i], buffer + hd, len)) {
1356 linenr++;
1357 patch->is_binary = 1;
1358 patchsize = llen;
1359 break;
1360 }
1361 }
1362 }
1364 /* Empty patch cannot be applied if it is a text patch
1365 * without metadata change. A binary patch appears
1366 * empty to us here.
1367 */
1368 if ((apply || check) &&
1369 (!patch->is_binary && !metadata_changes(patch)))
1370 die("patch with only garbage at line %d", linenr);
1371 }
1373 return offset + hdrsize + patchsize;
1374 }
1376 #define swap(a,b) myswap((a),(b),sizeof(a))
1378 #define myswap(a, b, size) do { \
1379 unsigned char mytmp[size]; \
1380 memcpy(mytmp, &a, size); \
1381 memcpy(&a, &b, size); \
1382 memcpy(&b, mytmp, size); \
1383 } while (0)
1385 static void reverse_patches(struct patch *p)
1386 {
1387 for (; p; p = p->next) {
1388 struct fragment *frag = p->fragments;
1390 swap(p->new_name, p->old_name);
1391 swap(p->new_mode, p->old_mode);
1392 swap(p->is_new, p->is_delete);
1393 swap(p->lines_added, p->lines_deleted);
1394 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1396 for (; frag; frag = frag->next) {
1397 swap(frag->newpos, frag->oldpos);
1398 swap(frag->newlines, frag->oldlines);
1399 }
1400 }
1401 }
1403 static const char pluses[] =
1404 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1405 static const char minuses[]=
1406 "----------------------------------------------------------------------";
1408 static void show_stats(struct patch *patch)
1409 {
1410 struct strbuf qname;
1411 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1412 int max, add, del;
1414 strbuf_init(&qname, 0);
1415 quote_c_style(cp, &qname, NULL, 0);
1417 /*
1418 * "scale" the filename
1419 */
1420 max = max_len;
1421 if (max > 50)
1422 max = 50;
1424 if (qname.len > max) {
1425 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1426 if (!cp)
1427 cp = qname.buf + qname.len + 3 - max;
1428 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1429 }
1431 if (patch->is_binary) {
1432 printf(" %-*s | Bin\n", max, qname.buf);
1433 strbuf_release(&qname);
1434 return;
1435 }
1437 printf(" %-*s |", max, qname.buf);
1438 strbuf_release(&qname);
1440 /*
1441 * scale the add/delete
1442 */
1443 max = max + max_change > 70 ? 70 - max : max_change;
1444 add = patch->lines_added;
1445 del = patch->lines_deleted;
1447 if (max_change > 0) {
1448 int total = ((add + del) * max + max_change / 2) / max_change;
1449 add = (add * max + max_change / 2) / max_change;
1450 del = total - add;
1451 }
1452 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1453 add, pluses, del, minuses);
1454 }
1456 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1457 {
1458 switch (st->st_mode & S_IFMT) {
1459 case S_IFLNK:
1460 strbuf_grow(buf, st->st_size);
1461 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1462 return -1;
1463 strbuf_setlen(buf, st->st_size);
1464 return 0;
1465 case S_IFREG:
1466 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1467 return error("unable to open or read %s", path);
1468 convert_to_git(path, buf->buf, buf->len, buf);
1469 return 0;
1470 default:
1471 return -1;
1472 }
1473 }
1475 static int find_offset(const char *buf, unsigned long size,
1476 const char *fragment, unsigned long fragsize,
1477 int line, int *lines)
1478 {
1479 int i;
1480 unsigned long start, backwards, forwards;
1482 if (fragsize > size)
1483 return -1;
1485 start = 0;
1486 if (line > 1) {
1487 unsigned long offset = 0;
1488 i = line-1;
1489 while (offset + fragsize <= size) {
1490 if (buf[offset++] == '\n') {
1491 start = offset;
1492 if (!--i)
1493 break;
1494 }
1495 }
1496 }
1498 /* Exact line number? */
1499 if ((start + fragsize <= size) &&
1500 !memcmp(buf + start, fragment, fragsize))
1501 return start;
1503 /*
1504 * There's probably some smart way to do this, but I'll leave
1505 * that to the smart and beautiful people. I'm simple and stupid.
1506 */
1507 backwards = start;
1508 forwards = start;
1509 for (i = 0; ; i++) {
1510 unsigned long try;
1511 int n;
1513 /* "backward" */
1514 if (i & 1) {
1515 if (!backwards) {
1516 if (forwards + fragsize > size)
1517 break;
1518 continue;
1519 }
1520 do {
1521 --backwards;
1522 } while (backwards && buf[backwards-1] != '\n');
1523 try = backwards;
1524 } else {
1525 while (forwards + fragsize <= size) {
1526 if (buf[forwards++] == '\n')
1527 break;
1528 }
1529 try = forwards;
1530 }
1532 if (try + fragsize > size)
1533 continue;
1534 if (memcmp(buf + try, fragment, fragsize))
1535 continue;
1536 n = (i >> 1)+1;
1537 if (i & 1)
1538 n = -n;
1539 *lines = n;
1540 return try;
1541 }
1543 /*
1544 * We should start searching forward and backward.
1545 */
1546 return -1;
1547 }
1549 static void remove_first_line(const char **rbuf, int *rsize)
1550 {
1551 const char *buf = *rbuf;
1552 int size = *rsize;
1553 unsigned long offset;
1554 offset = 0;
1555 while (offset <= size) {
1556 if (buf[offset++] == '\n')
1557 break;
1558 }
1559 *rsize = size - offset;
1560 *rbuf = buf + offset;
1561 }
1563 static void remove_last_line(const char **rbuf, int *rsize)
1564 {
1565 const char *buf = *rbuf;
1566 int size = *rsize;
1567 unsigned long offset;
1568 offset = size - 1;
1569 while (offset > 0) {
1570 if (buf[--offset] == '\n')
1571 break;
1572 }
1573 *rsize = offset + 1;
1574 }
1576 static int apply_line(char *output, const char *patch, int plen,
1577 unsigned ws_rule)
1578 {
1579 /*
1580 * plen is number of bytes to be copied from patch,
1581 * starting at patch+1 (patch[0] is '+'). Typically
1582 * patch[plen] is '\n', unless this is the incomplete
1583 * last line.
1584 */
1585 int i;
1586 int add_nl_to_tail = 0;
1587 int fixed = 0;
1588 int last_tab_in_indent = -1;
1589 int last_space_in_indent = -1;
1590 int need_fix_leading_space = 0;
1591 char *buf;
1593 if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1594 *patch != '+') {
1595 memcpy(output, patch + 1, plen);
1596 return plen;
1597 }
1599 /*
1600 * Strip trailing whitespace
1601 */
1602 if ((ws_rule & WS_TRAILING_SPACE) &&
1603 (1 < plen && isspace(patch[plen-1]))) {
1604 if (patch[plen] == '\n')
1605 add_nl_to_tail = 1;
1606 plen--;
1607 while (0 < plen && isspace(patch[plen]))
1608 plen--;
1609 fixed = 1;
1610 }
1612 /*
1613 * Check leading whitespaces (indent)
1614 */
1615 for (i = 1; i < plen; i++) {
1616 char ch = patch[i];
1617 if (ch == '\t') {
1618 last_tab_in_indent = i;
1619 if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1620 0 <= last_space_in_indent)
1621 need_fix_leading_space = 1;
1622 } else if (ch == ' ') {
1623 last_space_in_indent = i;
1624 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1625 last_tab_in_indent < 0 &&
1626 8 <= i)
1627 need_fix_leading_space = 1;
1628 }
1629 else
1630 break;
1631 }
1633 buf = output;
1634 if (need_fix_leading_space) {
1635 int consecutive_spaces = 0;
1636 int last = last_tab_in_indent + 1;
1638 if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1639 /* have "last" point at one past the indent */
1640 if (last_tab_in_indent < last_space_in_indent)
1641 last = last_space_in_indent + 1;
1642 else
1643 last = last_tab_in_indent + 1;
1644 }
1646 /*
1647 * between patch[1..last], strip the funny spaces,
1648 * updating them to tab as needed.
1649 */
1650 for (i = 1; i < last; i++, plen--) {
1651 char ch = patch[i];
1652 if (ch != ' ') {
1653 consecutive_spaces = 0;
1654 *output++ = ch;
1655 } else {
1656 consecutive_spaces++;
1657 if (consecutive_spaces == 8) {
1658 *output++ = '\t';
1659 consecutive_spaces = 0;
1660 }
1661 }
1662 }
1663 while (0 < consecutive_spaces--)
1664 *output++ = ' ';
1665 fixed = 1;
1666 i = last;
1667 }
1668 else
1669 i = 1;
1671 memcpy(output, patch + i, plen);
1672 if (add_nl_to_tail)
1673 output[plen++] = '\n';
1674 if (fixed)
1675 applied_after_fixing_ws++;
1676 return output + plen - buf;
1677 }
1679 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag,
1680 int inaccurate_eof, unsigned ws_rule)
1681 {
1682 int match_beginning, match_end;
1683 const char *patch = frag->patch;
1684 int offset, size = frag->size;
1685 char *old = xmalloc(size);
1686 char *new = xmalloc(size);
1687 const char *oldlines, *newlines;
1688 int oldsize = 0, newsize = 0;
1689 int new_blank_lines_at_end = 0;
1690 unsigned long leading, trailing;
1691 int pos, lines;
1693 while (size > 0) {
1694 char first;
1695 int len = linelen(patch, size);
1696 int plen;
1697 int added_blank_line = 0;
1699 if (!len)
1700 break;
1702 /*
1703 * "plen" is how much of the line we should use for
1704 * the actual patch data. Normally we just remove the
1705 * first character on the line, but if the line is
1706 * followed by "\ No newline", then we also remove the
1707 * last one (which is the newline, of course).
1708 */
1709 plen = len-1;
1710 if (len < size && patch[len] == '\\')
1711 plen--;
1712 first = *patch;
1713 if (apply_in_reverse) {
1714 if (first == '-')
1715 first = '+';
1716 else if (first == '+')
1717 first = '-';
1718 }
1720 switch (first) {
1721 case '\n':
1722 /* Newer GNU diff, empty context line */
1723 if (plen < 0)
1724 /* ... followed by '\No newline'; nothing */
1725 break;
1726 old[oldsize++] = '\n';
1727 new[newsize++] = '\n';
1728 break;
1729 case ' ':
1730 case '-':
1731 memcpy(old + oldsize, patch + 1, plen);
1732 oldsize += plen;
1733 if (first == '-')
1734 break;
1735 /* Fall-through for ' ' */
1736 case '+':
1737 if (first != '+' || !no_add) {
1738 int added = apply_line(new + newsize, patch,
1739 plen, ws_rule);
1740 newsize += added;
1741 if (first == '+' &&
1742 added == 1 && new[newsize-1] == '\n')
1743 added_blank_line = 1;
1744 }
1745 break;
1746 case '@': case '\\':
1747 /* Ignore it, we already handled it */
1748 break;
1749 default:
1750 if (apply_verbosely)
1751 error("invalid start of line: '%c'", first);
1752 return -1;
1753 }
1754 if (added_blank_line)
1755 new_blank_lines_at_end++;
1756 else
1757 new_blank_lines_at_end = 0;
1758 patch += len;
1759 size -= len;
1760 }
1762 if (inaccurate_eof &&
1763 oldsize > 0 && old[oldsize - 1] == '\n' &&
1764 newsize > 0 && new[newsize - 1] == '\n') {
1765 oldsize--;
1766 newsize--;
1767 }
1769 oldlines = old;
1770 newlines = new;
1771 leading = frag->leading;
1772 trailing = frag->trailing;
1774 /*
1775 * If we don't have any leading/trailing data in the patch,
1776 * we want it to match at the beginning/end of the file.
1777 *
1778 * But that would break if the patch is generated with
1779 * --unified=0; sane people wouldn't do that to cause us
1780 * trouble, but we try to please not so sane ones as well.
1781 */
1782 if (unidiff_zero) {
1783 match_beginning = (!leading && !frag->oldpos);
1784 match_end = 0;
1785 }
1786 else {
1787 match_beginning = !leading && (frag->oldpos == 1);
1788 match_end = !trailing;
1789 }
1791 lines = 0;
1792 pos = frag->newpos;
1793 for (;;) {
1794 offset = find_offset(buf->buf, buf->len,
1795 oldlines, oldsize, pos, &lines);
1796 if (match_end && offset + oldsize != buf->len)
1797 offset = -1;
1798 if (match_beginning && offset)
1799 offset = -1;
1800 if (offset >= 0) {
1801 if (ws_error_action == correct_ws_error &&
1802 (buf->len - oldsize - offset == 0)) /* end of file? */
1803 newsize -= new_blank_lines_at_end;
1805 /* Warn if it was necessary to reduce the number
1806 * of context lines.
1807 */
1808 if ((leading != frag->leading) ||
1809 (trailing != frag->trailing))
1810 fprintf(stderr, "Context reduced to (%ld/%ld)"
1811 " to apply fragment at %d\n",
1812 leading, trailing, pos + lines);
1814 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1815 offset = 0;
1816 break;
1817 }
1819 /* Am I at my context limits? */
1820 if ((leading <= p_context) && (trailing <= p_context))
1821 break;
1822 if (match_beginning || match_end) {
1823 match_beginning = match_end = 0;
1824 continue;
1825 }
1826 /*
1827 * Reduce the number of context lines; reduce both
1828 * leading and trailing if they are equal otherwise
1829 * just reduce the larger context.
1830 */
1831 if (leading >= trailing) {
1832 remove_first_line(&oldlines, &oldsize);
1833 remove_first_line(&newlines, &newsize);
1834 pos--;
1835 leading--;
1836 }
1837 if (trailing > leading) {
1838 remove_last_line(&oldlines, &oldsize);
1839 remove_last_line(&newlines, &newsize);
1840 trailing--;
1841 }
1842 }
1844 if (offset && apply_verbosely)
1845 error("while searching for:\n%.*s", oldsize, oldlines);
1847 free(old);
1848 free(new);
1849 return offset;
1850 }
1852 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1853 {
1854 struct fragment *fragment = patch->fragments;
1855 unsigned long len;
1856 void *dst;
1858 /* Binary patch is irreversible without the optional second hunk */
1859 if (apply_in_reverse) {
1860 if (!fragment->next)
1861 return error("cannot reverse-apply a binary patch "
1862 "without the reverse hunk to '%s'",
1863 patch->new_name
1864 ? patch->new_name : patch->old_name);
1865 fragment = fragment->next;
1866 }
1867 switch (fragment->binary_patch_method) {
1868 case BINARY_DELTA_DEFLATED:
1869 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1870 fragment->size, &len);
1871 if (!dst)
1872 return -1;
1873 /* XXX patch_delta NUL-terminates */
1874 strbuf_attach(buf, dst, len, len + 1);
1875 return 0;
1876 case BINARY_LITERAL_DEFLATED:
1877 strbuf_reset(buf);
1878 strbuf_add(buf, fragment->patch, fragment->size);
1879 return 0;
1880 }
1881 return -1;
1882 }
1884 static int apply_binary(struct strbuf *buf, struct patch *patch)
1885 {
1886 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1887 unsigned char sha1[20];
1889 /*
1890 * For safety, we require patch index line to contain
1891 * full 40-byte textual SHA1 for old and new, at least for now.
1892 */
1893 if (strlen(patch->old_sha1_prefix) != 40 ||
1894 strlen(patch->new_sha1_prefix) != 40 ||
1895 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1896 get_sha1_hex(patch->new_sha1_prefix, sha1))
1897 return error("cannot apply binary patch to '%s' "
1898 "without full index line", name);
1900 if (patch->old_name) {
1901 /*
1902 * See if the old one matches what the patch
1903 * applies to.
1904 */
1905 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1906 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1907 return error("the patch applies to '%s' (%s), "
1908 "which does not match the "
1909 "current contents.",
1910 name, sha1_to_hex(sha1));
1911 }
1912 else {
1913 /* Otherwise, the old one must be empty. */
1914 if (buf->len)
1915 return error("the patch applies to an empty "
1916 "'%s' but it is not empty", name);
1917 }
1919 get_sha1_hex(patch->new_sha1_prefix, sha1);
1920 if (is_null_sha1(sha1)) {
1921 strbuf_release(buf);
1922 return 0; /* deletion patch */
1923 }
1925 if (has_sha1_file(sha1)) {
1926 /* We already have the postimage */
1927 enum object_type type;
1928 unsigned long size;
1929 char *result;
1931 result = read_sha1_file(sha1, &type, &size);
1932 if (!result)
1933 return error("the necessary postimage %s for "
1934 "'%s' cannot be read",
1935 patch->new_sha1_prefix, name);
1936 /* XXX read_sha1_file NUL-terminates */
1937 strbuf_attach(buf, result, size, size + 1);
1938 } else {
1939 /*
1940 * We have verified buf matches the preimage;
1941 * apply the patch data to it, which is stored
1942 * in the patch->fragments->{patch,size}.
1943 */
1944 if (apply_binary_fragment(buf, patch))
1945 return error("binary patch does not apply to '%s'",
1946 name);
1948 /* verify that the result matches */
1949 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1950 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1951 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1952 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1953 }
1955 return 0;
1956 }
1958 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1959 {
1960 struct fragment *frag = patch->fragments;
1961 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1962 unsigned ws_rule = patch->ws_rule;
1963 unsigned inaccurate_eof = patch->inaccurate_eof;
1965 if (patch->is_binary)
1966 return apply_binary(buf, patch);
1968 while (frag) {
1969 if (apply_one_fragment(buf, frag, inaccurate_eof, ws_rule)) {
1970 error("patch failed: %s:%ld", name, frag->oldpos);
1971 if (!apply_with_reject)
1972 return -1;
1973 frag->rejected = 1;
1974 }
1975 frag = frag->next;
1976 }
1977 return 0;
1978 }
1980 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1981 {
1982 if (!ce)
1983 return 0;
1985 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1986 strbuf_grow(buf, 100);
1987 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1988 } else {
1989 enum object_type type;
1990 unsigned long sz;
1991 char *result;
1993 result = read_sha1_file(ce->sha1, &type, &sz);
1994 if (!result)
1995 return -1;
1996 /* XXX read_sha1_file NUL-terminates */
1997 strbuf_attach(buf, result, sz, sz + 1);
1998 }
1999 return 0;
2000 }
2002 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2003 {
2004 struct strbuf buf;
2006 strbuf_init(&buf, 0);
2007 if (cached) {
2008 if (read_file_or_gitlink(ce, &buf))
2009 return error("read of %s failed", patch->old_name);
2010 } else if (patch->old_name) {
2011 if (S_ISGITLINK(patch->old_mode)) {
2012 if (ce) {
2013 read_file_or_gitlink(ce, &buf);
2014 } else {
2015 /*
2016 * There is no way to apply subproject
2017 * patch without looking at the index.
2018 */
2019 patch->fragments = NULL;
2020 }
2021 } else {
2022 if (read_old_data(st, patch->old_name, &buf))
2023 return error("read of %s failed", patch->old_name);
2024 }
2025 }
2027 if (apply_fragments(&buf, patch) < 0)
2028 return -1; /* note with --reject this succeeds. */
2029 patch->result = strbuf_detach(&buf, &patch->resultsize);
2031 if (0 < patch->is_delete && patch->resultsize)
2032 return error("removal patch leaves file contents");
2034 return 0;
2035 }
2037 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2038 {
2039 struct stat nst;
2040 if (!lstat(new_name, &nst)) {
2041 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2042 return 0;
2043 /*
2044 * A leading component of new_name might be a symlink
2045 * that is going to be removed with this patch, but
2046 * still pointing at somewhere that has the path.
2047 * In such a case, path "new_name" does not exist as
2048 * far as git is concerned.
2049 */
2050 if (has_symlink_leading_path(new_name, NULL))
2051 return 0;
2053 return error("%s: already exists in working directory", new_name);
2054 }
2055 else if ((errno != ENOENT) && (errno != ENOTDIR))
2056 return error("%s: %s", new_name, strerror(errno));
2057 return 0;
2058 }
2060 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2061 {
2062 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2063 if (!S_ISDIR(st->st_mode))
2064 return -1;
2065 return 0;
2066 }
2067 return ce_match_stat(ce, st, 1);
2068 }
2070 static int check_patch(struct patch *patch, struct patch *prev_patch)
2071 {
2072 struct stat st;
2073 const char *old_name = patch->old_name;
2074 const char *new_name = patch->new_name;
2075 const char *name = old_name ? old_name : new_name;
2076 struct cache_entry *ce = NULL;
2077 int ok_if_exists;
2079 patch->rejected = 1; /* we will drop this after we succeed */
2081 /*
2082 * Make sure that we do not have local modifications from the
2083 * index when we are looking at the index. Also make sure
2084 * we have the preimage file to be patched in the work tree,
2085 * unless --cached, which tells git to apply only in the index.
2086 */
2087 if (old_name) {
2088 int stat_ret = 0;
2089 unsigned st_mode = 0;
2091 if (!cached)
2092 stat_ret = lstat(old_name, &st);
2093 if (check_index) {
2094 int pos = cache_name_pos(old_name, strlen(old_name));
2095 if (pos < 0)
2096 return error("%s: does not exist in index",
2097 old_name);
2098 ce = active_cache[pos];
2099 if (stat_ret < 0) {
2100 struct checkout costate;
2101 if (errno != ENOENT)
2102 return error("%s: %s", old_name,
2103 strerror(errno));
2104 /* checkout */
2105 costate.base_dir = "";
2106 costate.base_dir_len = 0;
2107 costate.force = 0;
2108 costate.quiet = 0;
2109 costate.not_new = 0;
2110 costate.refresh_cache = 1;
2111 if (checkout_entry(ce,
2112 &costate,
2113 NULL) ||
2114 lstat(old_name, &st))
2115 return -1;
2116 }
2117 if (!cached && verify_index_match(ce, &st))
2118 return error("%s: does not match index",
2119 old_name);
2120 if (cached)
2121 st_mode = ntohl(ce->ce_mode);
2122 } else if (stat_ret < 0)
2123 return error("%s: %s", old_name, strerror(errno));
2125 if (!cached)
2126 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2128 if (patch->is_new < 0)
2129 patch->is_new = 0;
2130 if (!patch->old_mode)
2131 patch->old_mode = st_mode;
2132 if ((st_mode ^ patch->old_mode) & S_IFMT)
2133 return error("%s: wrong type", old_name);
2134 if (st_mode != patch->old_mode)
2135 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2136 old_name, st_mode, patch->old_mode);
2137 }
2139 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2140 !strcmp(prev_patch->old_name, new_name))
2141 /*
2142 * A type-change diff is always split into a patch to
2143 * delete old, immediately followed by a patch to
2144 * create new (see diff.c::run_diff()); in such a case
2145 * it is Ok that the entry to be deleted by the
2146 * previous patch is still in the working tree and in
2147 * the index.
2148 */
2149 ok_if_exists = 1;
2150 else
2151 ok_if_exists = 0;
2153 if (new_name &&
2154 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2155 if (check_index &&
2156 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2157 !ok_if_exists)
2158 return error("%s: already exists in index", new_name);
2159 if (!cached) {
2160 int err = check_to_create_blob(new_name, ok_if_exists);
2161 if (err)
2162 return err;
2163 }
2164 if (!patch->new_mode) {
2165 if (0 < patch->is_new)
2166 patch->new_mode = S_IFREG | 0644;
2167 else
2168 patch->new_mode = patch->old_mode;
2169 }
2170 }
2172 if (new_name && old_name) {
2173 int same = !strcmp(old_name, new_name);
2174 if (!patch->new_mode)
2175 patch->new_mode = patch->old_mode;
2176 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2177 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2178 patch->new_mode, new_name, patch->old_mode,
2179 same ? "" : " of ", same ? "" : old_name);
2180 }
2182 if (apply_data(patch, &st, ce) < 0)
2183 return error("%s: patch does not apply", name);
2184 patch->rejected = 0;
2185 return 0;
2186 }
2188 static int check_patch_list(struct patch *patch)
2189 {
2190 struct patch *prev_patch = NULL;
2191 int err = 0;
2193 for (prev_patch = NULL; patch ; patch = patch->next) {
2194 if (apply_verbosely)
2195 say_patch_name(stderr,
2196 "Checking patch ", patch, "...\n");
2197 err |= check_patch(patch, prev_patch);
2198 prev_patch = patch;
2199 }
2200 return err;
2201 }
2203 /* This function tries to read the sha1 from the current index */
2204 static int get_current_sha1(const char *path, unsigned char *sha1)
2205 {
2206 int pos;
2208 if (read_cache() < 0)
2209 return -1;
2210 pos = cache_name_pos(path, strlen(path));
2211 if (pos < 0)
2212 return -1;
2213 hashcpy(sha1, active_cache[pos]->sha1);
2214 return 0;
2215 }
2217 /* Build an index that contains the just the files needed for a 3way merge */
2218 static void build_fake_ancestor(struct patch *list, const char *filename)
2219 {
2220 struct patch *patch;
2221 struct index_state result = { 0 };
2222 int fd;
2224 /* Once we start supporting the reverse patch, it may be
2225 * worth showing the new sha1 prefix, but until then...
2226 */
2227 for (patch = list; patch; patch = patch->next) {
2228 const unsigned char *sha1_ptr;
2229 unsigned char sha1[20];
2230 struct cache_entry *ce;
2231 const char *name;
2233 name = patch->old_name ? patch->old_name : patch->new_name;
2234 if (0 < patch->is_new)
2235 continue;
2236 else if (get_sha1(patch->old_sha1_prefix, sha1))
2237 /* git diff has no index line for mode/type changes */
2238 if (!patch->lines_added && !patch->lines_deleted) {
2239 if (get_current_sha1(patch->new_name, sha1) ||
2240 get_current_sha1(patch->old_name, sha1))
2241 die("mode change for %s, which is not "
2242 "in current HEAD", name);
2243 sha1_ptr = sha1;
2244 } else
2245 die("sha1 information is lacking or useless "
2246 "(%s).", name);
2247 else
2248 sha1_ptr = sha1;
2250 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2251 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2252 die ("Could not add %s to temporary index", name);
2253 }
2255 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2256 if (fd < 0 || write_index(&result, fd) || close(fd))
2257 die ("Could not write temporary index to %s", filename);
2259 discard_index(&result);
2260 }
2262 static void stat_patch_list(struct patch *patch)
2263 {
2264 int files, adds, dels;
2266 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2267 files++;
2268 adds += patch->lines_added;
2269 dels += patch->lines_deleted;
2270 show_stats(patch);
2271 }
2273 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2274 }
2276 static void numstat_patch_list(struct patch *patch)
2277 {
2278 for ( ; patch; patch = patch->next) {
2279 const char *name;
2280 name = patch->new_name ? patch->new_name : patch->old_name;
2281 if (patch->is_binary)
2282 printf("-\t-\t");
2283 else
2284 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2285 write_name_quoted(name, stdout, line_termination);
2286 }
2287 }
2289 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2290 {
2291 if (mode)
2292 printf(" %s mode %06o %s\n", newdelete, mode, name);
2293 else
2294 printf(" %s %s\n", newdelete, name);
2295 }
2297 static void show_mode_change(struct patch *p, int show_name)
2298 {
2299 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2300 if (show_name)
2301 printf(" mode change %06o => %06o %s\n",
2302 p->old_mode, p->new_mode, p->new_name);
2303 else
2304 printf(" mode change %06o => %06o\n",
2305 p->old_mode, p->new_mode);
2306 }
2307 }
2309 static void show_rename_copy(struct patch *p)
2310 {
2311 const char *renamecopy = p->is_rename ? "rename" : "copy";
2312 const char *old, *new;
2314 /* Find common prefix */
2315 old = p->old_name;
2316 new = p->new_name;
2317 while (1) {
2318 const char *slash_old, *slash_new;
2319 slash_old = strchr(old, '/');
2320 slash_new = strchr(new, '/');
2321 if (!slash_old ||
2322 !slash_new ||
2323 slash_old - old != slash_new - new ||
2324 memcmp(old, new, slash_new - new))
2325 break;
2326 old = slash_old + 1;
2327 new = slash_new + 1;
2328 }
2329 /* p->old_name thru old is the common prefix, and old and new
2330 * through the end of names are renames
2331 */
2332 if (old != p->old_name)
2333 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2334 (int)(old - p->old_name), p->old_name,
2335 old, new, p->score);
2336 else
2337 printf(" %s %s => %s (%d%%)\n", renamecopy,
2338 p->old_name, p->new_name, p->score);
2339 show_mode_change(p, 0);
2340 }
2342 static void summary_patch_list(struct patch *patch)
2343 {
2344 struct patch *p;
2346 for (p = patch; p; p = p->next) {
2347 if (p->is_new)
2348 show_file_mode_name("create", p->new_mode, p->new_name);
2349 else if (p->is_delete)
2350 show_file_mode_name("delete", p->old_mode, p->old_name);
2351 else {
2352 if (p->is_rename || p->is_copy)
2353 show_rename_copy(p);
2354 else {
2355 if (p->score) {
2356 printf(" rewrite %s (%d%%)\n",
2357 p->new_name, p->score);
2358 show_mode_change(p, 0);
2359 }
2360 else
2361 show_mode_change(p, 1);
2362 }
2363 }
2364 }
2365 }
2367 static void patch_stats(struct patch *patch)
2368 {
2369 int lines = patch->lines_added + patch->lines_deleted;
2371 if (lines > max_change)
2372 max_change = lines;
2373 if (patch->old_name) {
2374 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2375 if (!len)
2376 len = strlen(patch->old_name);
2377 if (len > max_len)
2378 max_len = len;
2379 }
2380 if (patch->new_name) {
2381 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2382 if (!len)
2383 len = strlen(patch->new_name);
2384 if (len > max_len)
2385 max_len = len;
2386 }
2387 }
2389 static void remove_file(struct patch *patch, int rmdir_empty)
2390 {
2391 if (update_index) {
2392 if (remove_file_from_cache(patch->old_name) < 0)
2393 die("unable to remove %s from index", patch->old_name);
2394 }
2395 if (!cached) {
2396 if (S_ISGITLINK(patch->old_mode)) {
2397 if (rmdir(patch->old_name))
2398 warning("unable to remove submodule %s",
2399 patch->old_name);
2400 } else if (!unlink(patch->old_name) && rmdir_empty) {
2401 char *name = xstrdup(patch->old_name);
2402 char *end = strrchr(name, '/');
2403 while (end) {
2404 *end = 0;
2405 if (rmdir(name))
2406 break;
2407 end = strrchr(name, '/');
2408 }
2409 free(name);
2410 }
2411 }
2412 }
2414 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2415 {
2416 struct stat st;
2417 struct cache_entry *ce;
2418 int namelen = strlen(path);
2419 unsigned ce_size = cache_entry_size(namelen);
2421 if (!update_index)
2422 return;
2424 ce = xcalloc(1, ce_size);
2425 memcpy(ce->name, path, namelen);
2426 ce->ce_mode = create_ce_mode(mode);
2427 ce->ce_flags = htons(namelen);
2428 if (S_ISGITLINK(mode)) {
2429 const char *s = buf;
2431 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2432 die("corrupt patch for subproject %s", path);
2433 } else {
2434 if (!cached) {
2435 if (lstat(path, &st) < 0)
2436 die("unable to stat newly created file %s",
2437 path);
2438 fill_stat_cache_info(ce, &st);
2439 }
2440 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2441 die("unable to create backing store for newly created file %s", path);
2442 }
2443 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2444 die("unable to add cache entry for %s", path);
2445 }
2447 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2448 {
2449 int fd;
2450 struct strbuf nbuf;
2452 if (S_ISGITLINK(mode)) {
2453 struct stat st;
2454 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2455 return 0;
2456 return mkdir(path, 0777);
2457 }
2459 if (has_symlinks && S_ISLNK(mode))
2460 /* Although buf:size is counted string, it also is NUL
2461 * terminated.
2462 */
2463 return symlink(buf, path);
2465 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2466 if (fd < 0)
2467 return -1;
2469 strbuf_init(&nbuf, 0);
2470 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2471 size = nbuf.len;
2472 buf = nbuf.buf;
2473 }
2474 write_or_die(fd, buf, size);
2475 strbuf_release(&nbuf);
2477 if (close(fd) < 0)
2478 die("closing file %s: %s", path, strerror(errno));
2479 return 0;
2480 }
2482 /*
2483 * We optimistically assume that the directories exist,
2484 * which is true 99% of the time anyway. If they don't,
2485 * we create them and try again.
2486 */
2487 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2488 {
2489 if (cached)
2490 return;
2491 if (!try_create_file(path, mode, buf, size))
2492 return;
2494 if (errno == ENOENT) {
2495 if (safe_create_leading_directories(path))
2496 return;
2497 if (!try_create_file(path, mode, buf, size))
2498 return;
2499 }
2501 if (errno == EEXIST || errno == EACCES) {
2502 /* We may be trying to create a file where a directory
2503 * used to be.
2504 */
2505 struct stat st;
2506 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2507 errno = EEXIST;
2508 }
2510 if (errno == EEXIST) {
2511 unsigned int nr = getpid();
2513 for (;;) {
2514 const char *newpath;
2515 newpath = mkpath("%s~%u", path, nr);
2516 if (!try_create_file(newpath, mode, buf, size)) {
2517 if (!rename(newpath, path))
2518 return;
2519 unlink(newpath);
2520 break;
2521 }
2522 if (errno != EEXIST)
2523 break;
2524 ++nr;
2525 }
2526 }
2527 die("unable to write file %s mode %o", path, mode);
2528 }
2530 static void create_file(struct patch *patch)
2531 {
2532 char *path = patch->new_name;
2533 unsigned mode = patch->new_mode;
2534 unsigned long size = patch->resultsize;
2535 char *buf = patch->result;
2537 if (!mode)
2538 mode = S_IFREG | 0644;
2539 create_one_file(path, mode, buf, size);
2540 add_index_file(path, mode, buf, size);
2541 }
2543 /* phase zero is to remove, phase one is to create */
2544 static void write_out_one_result(struct patch *patch, int phase)
2545 {
2546 if (patch->is_delete > 0) {
2547 if (phase == 0)
2548 remove_file(patch, 1);
2549 return;
2550 }
2551 if (patch->is_new > 0 || patch->is_copy) {
2552 if (phase == 1)
2553 create_file(patch);
2554 return;
2555 }
2556 /*
2557 * Rename or modification boils down to the same
2558 * thing: remove the old, write the new
2559 */
2560 if (phase == 0)
2561 remove_file(patch, patch->is_rename);
2562 if (phase == 1)
2563 create_file(patch);
2564 }
2566 static int write_out_one_reject(struct patch *patch)
2567 {
2568 FILE *rej;
2569 char namebuf[PATH_MAX];
2570 struct fragment *frag;
2571 int cnt = 0;
2573 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2574 if (!frag->rejected)
2575 continue;
2576 cnt++;
2577 }
2579 if (!cnt) {
2580 if (apply_verbosely)
2581 say_patch_name(stderr,
2582 "Applied patch ", patch, " cleanly.\n");
2583 return 0;
2584 }
2586 /* This should not happen, because a removal patch that leaves
2587 * contents are marked "rejected" at the patch level.
2588 */
2589 if (!patch->new_name)
2590 die("internal error");
2592 /* Say this even without --verbose */
2593 say_patch_name(stderr, "Applying patch ", patch, " with");
2594 fprintf(stderr, " %d rejects...\n", cnt);
2596 cnt = strlen(patch->new_name);
2597 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2598 cnt = ARRAY_SIZE(namebuf) - 5;
2599 fprintf(stderr,
2600 "warning: truncating .rej filename to %.*s.rej",
2601 cnt - 1, patch->new_name);
2602 }
2603 memcpy(namebuf, patch->new_name, cnt);
2604 memcpy(namebuf + cnt, ".rej", 5);
2606 rej = fopen(namebuf, "w");
2607 if (!rej)
2608 return error("cannot open %s: %s", namebuf, strerror(errno));
2610 /* Normal git tools never deal with .rej, so do not pretend
2611 * this is a git patch by saying --git nor give extended
2612 * headers. While at it, maybe please "kompare" that wants
2613 * the trailing TAB and some garbage at the end of line ;-).
2614 */
2615 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2616 patch->new_name, patch->new_name);
2617 for (cnt = 1, frag = patch->fragments;
2618 frag;
2619 cnt++, frag = frag->next) {
2620 if (!frag->rejected) {
2621 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2622 continue;
2623 }
2624 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2625 fprintf(rej, "%.*s", frag->size, frag->patch);
2626 if (frag->patch[frag->size-1] != '\n')
2627 fputc('\n', rej);
2628 }
2629 fclose(rej);
2630 return -1;
2631 }
2633 static int write_out_results(struct patch *list, int skipped_patch)
2634 {
2635 int phase;
2636 int errs = 0;
2637 struct patch *l;
2639 if (!list && !skipped_patch)
2640 return error("No changes");
2642 for (phase = 0; phase < 2; phase++) {
2643 l = list;
2644 while (l) {
2645 if (l->rejected)
2646 errs = 1;
2647 else {
2648 write_out_one_result(l, phase);
2649 if (phase == 1 && write_out_one_reject(l))
2650 errs = 1;
2651 }
2652 l = l->next;
2653 }
2654 }
2655 return errs;
2656 }
2658 static struct lock_file lock_file;
2660 static struct excludes {
2661 struct excludes *next;
2662 const char *path;
2663 } *excludes;
2665 static int use_patch(struct patch *p)
2666 {
2667 const char *pathname = p->new_name ? p->new_name : p->old_name;
2668 struct excludes *x = excludes;
2669 while (x) {
2670 if (fnmatch(x->path, pathname, 0) == 0)
2671 return 0;
2672 x = x->next;
2673 }
2674 if (0 < prefix_length) {
2675 int pathlen = strlen(pathname);
2676 if (pathlen <= prefix_length ||
2677 memcmp(prefix, pathname, prefix_length))
2678 return 0;
2679 }
2680 return 1;
2681 }
2683 static void prefix_one(char **name)
2684 {
2685 char *old_name = *name;
2686 if (!old_name)
2687 return;
2688 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2689 free(old_name);
2690 }
2692 static void prefix_patches(struct patch *p)
2693 {
2694 if (!prefix || p->is_toplevel_relative)
2695 return;
2696 for ( ; p; p = p->next) {
2697 if (p->new_name == p->old_name) {
2698 char *prefixed = p->new_name;
2699 prefix_one(&prefixed);
2700 p->new_name = p->old_name = prefixed;
2701 }
2702 else {
2703 prefix_one(&p->new_name);
2704 prefix_one(&p->old_name);
2705 }
2706 }
2707 }
2709 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2710 {
2711 size_t offset;
2712 struct strbuf buf;
2713 struct patch *list = NULL, **listp = &list;
2714 int skipped_patch = 0;
2716 strbuf_init(&buf, 0);
2717 patch_input_file = filename;
2718 read_patch_file(&buf, fd);
2719 offset = 0;
2720 while (offset < buf.len) {
2721 struct patch *patch;
2722 int nr;
2724 patch = xcalloc(1, sizeof(*patch));
2725 patch->inaccurate_eof = inaccurate_eof;
2726 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2727 if (nr < 0)
2728 break;
2729 if (apply_in_reverse)
2730 reverse_patches(patch);
2731 if (prefix)
2732 prefix_patches(patch);
2733 if (use_patch(patch)) {
2734 patch_stats(patch);
2735 *listp = patch;
2736 listp = &patch->next;
2737 }
2738 else {
2739 /* perhaps free it a bit better? */
2740 free(patch);
2741 skipped_patch++;
2742 }
2743 offset += nr;
2744 }
2746 if (whitespace_error && (ws_error_action == die_on_ws_error))
2747 apply = 0;
2749 update_index = check_index && apply;
2750 if (update_index && newfd < 0)
2751 newfd = hold_locked_index(&lock_file, 1);
2753 if (check_index) {
2754 if (read_cache() < 0)
2755 die("unable to read index file");
2756 }
2758 if ((check || apply) &&
2759 check_patch_list(list) < 0 &&
2760 !apply_with_reject)
2761 exit(1);
2763 if (apply && write_out_results(list, skipped_patch))
2764 exit(1);
2766 if (fake_ancestor)
2767 build_fake_ancestor(list, fake_ancestor);
2769 if (diffstat)
2770 stat_patch_list(list);
2772 if (numstat)
2773 numstat_patch_list(list);
2775 if (summary)
2776 summary_patch_list(list);
2778 strbuf_release(&buf);
2779 return 0;
2780 }
2782 static int git_apply_config(const char *var, const char *value)
2783 {
2784 if (!strcmp(var, "apply.whitespace")) {
2785 apply_default_whitespace = xstrdup(value);
2786 return 0;
2787 }
2788 return git_default_config(var, value);
2789 }
2792 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2793 {
2794 int i;
2795 int read_stdin = 1;
2796 int inaccurate_eof = 0;
2797 int errs = 0;
2798 int is_not_gitdir = 0;
2800 const char *whitespace_option = NULL;
2802 prefix = setup_git_directory_gently(&is_not_gitdir);
2803 prefix_length = prefix ? strlen(prefix) : 0;
2804 git_config(git_apply_config);
2805 if (apply_default_whitespace)
2806 parse_whitespace_option(apply_default_whitespace);
2808 for (i = 1; i < argc; i++) {
2809 const char *arg = argv[i];
2810 char *end;
2811 int fd;
2813 if (!strcmp(arg, "-")) {
2814 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2815 read_stdin = 0;
2816 continue;
2817 }
2818 if (!prefixcmp(arg, "--exclude=")) {
2819 struct excludes *x = xmalloc(sizeof(*x));
2820 x->path = arg + 10;
2821 x->next = excludes;
2822 excludes = x;
2823 continue;
2824 }
2825 if (!prefixcmp(arg, "-p")) {
2826 p_value = atoi(arg + 2);
2827 p_value_known = 1;
2828 continue;
2829 }
2830 if (!strcmp(arg, "--no-add")) {
2831 no_add = 1;
2832 continue;
2833 }
2834 if (!strcmp(arg, "--stat")) {
2835 apply = 0;
2836 diffstat = 1;
2837 continue;
2838 }
2839 if (!strcmp(arg, "--allow-binary-replacement") ||
2840 !strcmp(arg, "--binary")) {
2841 continue; /* now no-op */
2842 }
2843 if (!strcmp(arg, "--numstat")) {
2844 apply = 0;
2845 numstat = 1;
2846 continue;
2847 }
2848 if (!strcmp(arg, "--summary")) {
2849 apply = 0;
2850 summary = 1;
2851 continue;
2852 }
2853 if (!strcmp(arg, "--check")) {
2854 apply = 0;
2855 check = 1;
2856 continue;
2857 }
2858 if (!strcmp(arg, "--index")) {
2859 if (is_not_gitdir)
2860 die("--index outside a repository");
2861 check_index = 1;
2862 continue;
2863 }
2864 if (!strcmp(arg, "--cached")) {
2865 if (is_not_gitdir)
2866 die("--cached outside a repository");
2867 check_index = 1;
2868 cached = 1;
2869 continue;
2870 }
2871 if (!strcmp(arg, "--apply")) {
2872 apply = 1;
2873 continue;
2874 }
2875 if (!strcmp(arg, "--build-fake-ancestor")) {
2876 apply = 0;
2877 if (++i >= argc)
2878 die ("need a filename");
2879 fake_ancestor = argv[i];
2880 continue;
2881 }
2882 if (!strcmp(arg, "-z")) {
2883 line_termination = 0;
2884 continue;
2885 }
2886 if (!prefixcmp(arg, "-C")) {
2887 p_context = strtoul(arg + 2, &end, 0);
2888 if (*end != '\0')
2889 die("unrecognized context count '%s'", arg + 2);
2890 continue;
2891 }
2892 if (!prefixcmp(arg, "--whitespace=")) {
2893 whitespace_option = arg + 13;
2894 parse_whitespace_option(arg + 13);
2895 continue;
2896 }
2897 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2898 apply_in_reverse = 1;
2899 continue;
2900 }
2901 if (!strcmp(arg, "--unidiff-zero")) {
2902 unidiff_zero = 1;
2903 continue;
2904 }
2905 if (!strcmp(arg, "--reject")) {
2906 apply = apply_with_reject = apply_verbosely = 1;
2907 continue;
2908 }
2909 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2910 apply_verbosely = 1;
2911 continue;
2912 }
2913 if (!strcmp(arg, "--inaccurate-eof")) {
2914 inaccurate_eof = 1;
2915 continue;
2916 }
2917 if (0 < prefix_length)
2918 arg = prefix_filename(prefix, prefix_length, arg);
2920 fd = open(arg, O_RDONLY);
2921 if (fd < 0)
2922 usage(apply_usage);
2923 read_stdin = 0;
2924 set_default_whitespace_mode(whitespace_option);
2925 errs |= apply_patch(fd, arg, inaccurate_eof);
2926 close(fd);
2927 }
2928 set_default_whitespace_mode(whitespace_option);
2929 if (read_stdin)
2930 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2931 if (whitespace_error) {
2932 if (squelch_whitespace_errors &&
2933 squelch_whitespace_errors < whitespace_error) {
2934 int squelched =
2935 whitespace_error - squelch_whitespace_errors;
2936 fprintf(stderr, "warning: squelched %d "
2937 "whitespace error%s\n",
2938 squelched,
2939 squelched == 1 ? "" : "s");
2940 }
2941 if (ws_error_action == die_on_ws_error)
2942 die("%d line%s add%s whitespace errors.",
2943 whitespace_error,
2944 whitespace_error == 1 ? "" : "s",
2945 whitespace_error == 1 ? "s" : "");
2946 if (applied_after_fixing_ws)
2947 fprintf(stderr, "warning: %d line%s applied after"
2948 " fixing whitespace errors.\n",
2949 applied_after_fixing_ws,
2950 applied_after_fixing_ws == 1 ? "" : "s");
2951 else if (whitespace_error)
2952 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2953 whitespace_error,
2954 whitespace_error == 1 ? "" : "s",
2955 whitespace_error == 1 ? "s" : "");
2956 }
2958 if (update_index) {
2959 if (write_cache(newfd, active_cache, active_nr) ||
2960 close(newfd) || commit_locked_index(&lock_file))
2961 die("Unable to write new index file");
2962 }
2964 return !!errs;
2965 }