doc: documentation update for the branch track changes
[git/git.git] / builtin-apply.c
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"
15
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;
28
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>...";
49
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;
60
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 }
90
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 }
96
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;
103
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;
110
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 };
127
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
136
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 };
163
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 }
181
182 #define CHUNKSIZE (8192)
183 #define SLOP (16)
184
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));
189
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 }
198
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 }
209
210 static int is_dev_null(const char *str)
211 {
212 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
213 }
214
215 #define TERM_SPACE 1
216 #define TERM_TAB 2
217
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;
224
225 return 1;
226 }
227
228 static char *find_name(const char *line, char *def, int p_value, int terminate)
229 {
230 int len;
231 const char *start = line;
232
233 if (*line == '"') {
234 struct strbuf name;
235
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;
243
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 }
261
262 for (;;) {
263 char c = *line;
264
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;
280
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 }
293
294 return xmemdupz(start, len);
295 }
296
297 static int count_slashes(const char *cp)
298 {
299 int cnt = 0;
300 char ch;
301
302 while ((ch = *cp++))
303 if (ch == '/')
304 cnt++;
305 return cnt;
306 }
307
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;
316
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 }
341
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;
352
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 }
383
384 static int gitdiff_hdrend(const char *line, struct patch *patch)
385 {
386 return -1;
387 }
388
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);
402
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 }
424
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 }
430
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 }
436
437 static int gitdiff_oldmode(const char *line, struct patch *patch)
438 {
439 patch->old_mode = strtoul(line, NULL, 8);
440 return 0;
441 }
442
443 static int gitdiff_newmode(const char *line, struct patch *patch)
444 {
445 patch->new_mode = strtoul(line, NULL, 8);
446 return 0;
447 }
448
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 }
455
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 }
462
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 }
469
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 }
476
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 }
483
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 }
490
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 }
497
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 }
504
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;
513
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;
520
521 line = ptr + 2;
522 ptr = strchr(line, ' ');
523 eol = strchr(line, '\n');
524
525 if (!ptr || eol < ptr)
526 ptr = eol;
527 len = ptr - line;
528
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 }
537
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 }
546
547 static const char *stop_at_slash(const char *line, int llen)
548 {
549 int i;
550
551 for (i = 0; i < llen; i++) {
552 int ch = line[i];
553 if (ch == '/')
554 return line + i;
555 }
556 return NULL;
557 }
558
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;
572
573 line += strlen("diff --git ");
574 llen -= strlen("diff --git ");
575
576 if (*line == '"') {
577 const char *cp;
578 struct strbuf first;
579 struct strbuf sp;
580
581 strbuf_init(&first, 0);
582 strbuf_init(&sp, 0);
583
584 if (unquote_c_style(&first, line, &second))
585 goto free_and_fail1;
586
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);
593
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++;
600
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 }
615
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);
625
626 free_and_fail1:
627 strbuf_release(&first);
628 strbuf_release(&sp);
629 return NULL;
630 }
631
632 /* unquoted first name */
633 name = stop_at_slash(line, llen);
634 if (!name || name == line)
635 return NULL;
636 name++;
637
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;
646
647 strbuf_init(&sp, 0);
648 if (unquote_c_style(&sp, second, NULL))
649 goto free_and_fail2;
650
651 np = stop_at_slash(sp.buf, sp.len);
652 if (!np || np == sp.buf)
653 goto free_and_fail2;
654 np++;
655
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 }
664
665 free_and_fail2:
666 strbuf_release(&sp);
667 return NULL;
668 }
669 }
670
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 }
696
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;
701
702 /* A git diff has explicit new/delete information, so we don't guess */
703 patch->is_new = 0;
704 patch->is_delete = 0;
705
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);
713
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;
741
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 }
755
756 return offset;
757 }
758
759 static int parse_num(const char *line, unsigned long *p)
760 {
761 char *ptr;
762
763 if (!isdigit(*line))
764 return 0;
765 *p = strtoul(line, &ptr, 10);
766 return ptr - line;
767 }
768
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;
773
774 if (offset < 0 || offset >= len)
775 return -1;
776 line += offset;
777 len -= offset;
778
779 digits = parse_num(line, p1);
780 if (!digits)
781 return -1;
782
783 offset += digits;
784 line += digits;
785 len -= digits;
786
787 *p2 = 1;
788 if (*line == ',') {
789 digits = parse_num(line+1, p2);
790 if (!digits)
791 return -1;
792
793 offset += digits+1;
794 line += digits+1;
795 len -= digits+1;
796 }
797
798 ex = strlen(expect);
799 if (ex > len)
800 return -1;
801 if (memcmp(line, expect, ex))
802 return -1;
803
804 return offset + ex;
805 }
806
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;
814
815 if (!len || line[len-1] != '\n')
816 return -1;
817
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);
821
822 return offset;
823 }
824
825 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
826 {
827 unsigned long offset, len;
828
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;
836
837 len = linelen(line, size);
838 if (!len)
839 break;
840
841 /* Testing this early allows us to take a few shortcuts.. */
842 if (len < 6)
843 continue;
844
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 }
857
858 if (size < len + 6)
859 break;
860
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 }
878
879 /* --- followed by +++ ? */
880 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
881 continue;
882
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;
891
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 }
900
901 static void check_whitespace(const char *line, int len, unsigned ws_rule)
902 {
903 char *err;
904 unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
905 NULL, NULL, NULL, NULL);
906 if (!result)
907 return;
908
909 whitespace_error++;
910 if (squelch_whitespace_errors &&
911 squelch_whitespace_errors < whitespace_error)
912 ;
913 else {
914 err = whitespace_error_string(result);
915 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
916 patch_input_file, linenr, err, len - 2, line + 1);
917 free(err);
918 }
919 }
920
921 /*
922 * Parse a unified diff. Note that this really needs to parse each
923 * fragment separately, since the only way to know the difference
924 * between a "---" that is part of a patch, and a "---" that starts
925 * the next patch is to look at the line counts..
926 */
927 static int parse_fragment(char *line, unsigned long size,
928 struct patch *patch, struct fragment *fragment)
929 {
930 int added, deleted;
931 int len = linelen(line, size), offset;
932 unsigned long oldlines, newlines;
933 unsigned long leading, trailing;
934
935 offset = parse_fragment_header(line, len, fragment);
936 if (offset < 0)
937 return -1;
938 oldlines = fragment->oldlines;
939 newlines = fragment->newlines;
940 leading = 0;
941 trailing = 0;
942
943 /* Parse the thing.. */
944 line += len;
945 size -= len;
946 linenr++;
947 added = deleted = 0;
948 for (offset = len;
949 0 < size;
950 offset += len, size -= len, line += len, linenr++) {
951 if (!oldlines && !newlines)
952 break;
953 len = linelen(line, size);
954 if (!len || line[len-1] != '\n')
955 return -1;
956 switch (*line) {
957 default:
958 return -1;
959 case '\n': /* newer GNU diff, an empty context line */
960 case ' ':
961 oldlines--;
962 newlines--;
963 if (!deleted && !added)
964 leading++;
965 trailing++;
966 break;
967 case '-':
968 if (apply_in_reverse &&
969 ws_error_action != nowarn_ws_error)
970 check_whitespace(line, len, patch->ws_rule);
971 deleted++;
972 oldlines--;
973 trailing = 0;
974 break;
975 case '+':
976 if (!apply_in_reverse &&
977 ws_error_action != nowarn_ws_error)
978 check_whitespace(line, len, patch->ws_rule);
979 added++;
980 newlines--;
981 trailing = 0;
982 break;
983
984 /*
985 * We allow "\ No newline at end of file". Depending
986 * on locale settings when the patch was produced we
987 * don't know what this line looks like. The only
988 * thing we do know is that it begins with "\ ".
989 * Checking for 12 is just for sanity check -- any
990 * l10n of "\ No newline..." is at least that long.
991 */
992 case '\\':
993 if (len < 12 || memcmp(line, "\\ ", 2))
994 return -1;
995 break;
996 }
997 }
998 if (oldlines || newlines)
999 return -1;
1000 fragment->leading = leading;
1001 fragment->trailing = trailing;
1002
1003 /*
1004 * If a fragment ends with an incomplete line, we failed to include
1005 * it in the above loop because we hit oldlines == newlines == 0
1006 * before seeing it.
1007 */
1008 if (12 < size && !memcmp(line, "\\ ", 2))
1009 offset += linelen(line, size);
1010
1011 patch->lines_added += added;
1012 patch->lines_deleted += deleted;
1013
1014 if (0 < patch->is_new && oldlines)
1015 return error("new file depends on old contents");
1016 if (0 < patch->is_delete && newlines)
1017 return error("deleted file still has contents");
1018 return offset;
1019 }
1020
1021 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1022 {
1023 unsigned long offset = 0;
1024 unsigned long oldlines = 0, newlines = 0, context = 0;
1025 struct fragment **fragp = &patch->fragments;
1026
1027 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1028 struct fragment *fragment;
1029 int len;
1030
1031 fragment = xcalloc(1, sizeof(*fragment));
1032 len = parse_fragment(line, size, patch, fragment);
1033 if (len <= 0)
1034 die("corrupt patch at line %d", linenr);
1035 fragment->patch = line;
1036 fragment->size = len;
1037 oldlines += fragment->oldlines;
1038 newlines += fragment->newlines;
1039 context += fragment->leading + fragment->trailing;
1040
1041 *fragp = fragment;
1042 fragp = &fragment->next;
1043
1044 offset += len;
1045 line += len;
1046 size -= len;
1047 }
1048
1049 /*
1050 * If something was removed (i.e. we have old-lines) it cannot
1051 * be creation, and if something was added it cannot be
1052 * deletion. However, the reverse is not true; --unified=0
1053 * patches that only add are not necessarily creation even
1054 * though they do not have any old lines, and ones that only
1055 * delete are not necessarily deletion.
1056 *
1057 * Unfortunately, a real creation/deletion patch do _not_ have
1058 * any context line by definition, so we cannot safely tell it
1059 * apart with --unified=0 insanity. At least if the patch has
1060 * more than one hunk it is not creation or deletion.
1061 */
1062 if (patch->is_new < 0 &&
1063 (oldlines || (patch->fragments && patch->fragments->next)))
1064 patch->is_new = 0;
1065 if (patch->is_delete < 0 &&
1066 (newlines || (patch->fragments && patch->fragments->next)))
1067 patch->is_delete = 0;
1068 if (!unidiff_zero || context) {
1069 /* If the user says the patch is not generated with
1070 * --unified=0, or if we have seen context lines,
1071 * then not having oldlines means the patch is creation,
1072 * and not having newlines means the patch is deletion.
1073 */
1074 if (patch->is_new < 0 && !oldlines) {
1075 patch->is_new = 1;
1076 patch->old_name = NULL;
1077 }
1078 if (patch->is_delete < 0 && !newlines) {
1079 patch->is_delete = 1;
1080 patch->new_name = NULL;
1081 }
1082 }
1083
1084 if (0 < patch->is_new && oldlines)
1085 die("new file %s depends on old contents", patch->new_name);
1086 if (0 < patch->is_delete && newlines)
1087 die("deleted file %s still has contents", patch->old_name);
1088 if (!patch->is_delete && !newlines && context)
1089 fprintf(stderr, "** warning: file %s becomes empty but "
1090 "is not deleted\n", patch->new_name);
1091
1092 return offset;
1093 }
1094
1095 static inline int metadata_changes(struct patch *patch)
1096 {
1097 return patch->is_rename > 0 ||
1098 patch->is_copy > 0 ||
1099 patch->is_new > 0 ||
1100 patch->is_delete ||
1101 (patch->old_mode && patch->new_mode &&
1102 patch->old_mode != patch->new_mode);
1103 }
1104
1105 static char *inflate_it(const void *data, unsigned long size,
1106 unsigned long inflated_size)
1107 {
1108 z_stream stream;
1109 void *out;
1110 int st;
1111
1112 memset(&stream, 0, sizeof(stream));
1113
1114 stream.next_in = (unsigned char *)data;
1115 stream.avail_in = size;
1116 stream.next_out = out = xmalloc(inflated_size);
1117 stream.avail_out = inflated_size;
1118 inflateInit(&stream);
1119 st = inflate(&stream, Z_FINISH);
1120 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1121 free(out);
1122 return NULL;
1123 }
1124 return out;
1125 }
1126
1127 static struct fragment *parse_binary_hunk(char **buf_p,
1128 unsigned long *sz_p,
1129 int *status_p,
1130 int *used_p)
1131 {
1132 /*
1133 * Expect a line that begins with binary patch method ("literal"
1134 * or "delta"), followed by the length of data before deflating.
1135 * a sequence of 'length-byte' followed by base-85 encoded data
1136 * should follow, terminated by a newline.
1137 *
1138 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1139 * and we would limit the patch line to 66 characters,
1140 * so one line can fit up to 13 groups that would decode
1141 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1142 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1143 */
1144 int llen, used;
1145 unsigned long size = *sz_p;
1146 char *buffer = *buf_p;
1147 int patch_method;
1148 unsigned long origlen;
1149 char *data = NULL;
1150 int hunk_size = 0;
1151 struct fragment *frag;
1152
1153 llen = linelen(buffer, size);
1154 used = llen;
1155
1156 *status_p = 0;
1157
1158 if (!prefixcmp(buffer, "delta ")) {
1159 patch_method = BINARY_DELTA_DEFLATED;
1160 origlen = strtoul(buffer + 6, NULL, 10);
1161 }
1162 else if (!prefixcmp(buffer, "literal ")) {
1163 patch_method = BINARY_LITERAL_DEFLATED;
1164 origlen = strtoul(buffer + 8, NULL, 10);
1165 }
1166 else
1167 return NULL;
1168
1169 linenr++;
1170 buffer += llen;
1171 while (1) {
1172 int byte_length, max_byte_length, newsize;
1173 llen = linelen(buffer, size);
1174 used += llen;
1175 linenr++;
1176 if (llen == 1) {
1177 /* consume the blank line */
1178 buffer++;
1179 size--;
1180 break;
1181 }
1182 /*
1183 * Minimum line is "A00000\n" which is 7-byte long,
1184 * and the line length must be multiple of 5 plus 2.
1185 */
1186 if ((llen < 7) || (llen-2) % 5)
1187 goto corrupt;
1188 max_byte_length = (llen - 2) / 5 * 4;
1189 byte_length = *buffer;
1190 if ('A' <= byte_length && byte_length <= 'Z')
1191 byte_length = byte_length - 'A' + 1;
1192 else if ('a' <= byte_length && byte_length <= 'z')
1193 byte_length = byte_length - 'a' + 27;
1194 else
1195 goto corrupt;
1196 /* if the input length was not multiple of 4, we would
1197 * have filler at the end but the filler should never
1198 * exceed 3 bytes
1199 */
1200 if (max_byte_length < byte_length ||
1201 byte_length <= max_byte_length - 4)
1202 goto corrupt;
1203 newsize = hunk_size + byte_length;
1204 data = xrealloc(data, newsize);
1205 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1206 goto corrupt;
1207 hunk_size = newsize;
1208 buffer += llen;
1209 size -= llen;
1210 }
1211
1212 frag = xcalloc(1, sizeof(*frag));
1213 frag->patch = inflate_it(data, hunk_size, origlen);
1214 if (!frag->patch)
1215 goto corrupt;
1216 free(data);
1217 frag->size = origlen;
1218 *buf_p = buffer;
1219 *sz_p = size;
1220 *used_p = used;
1221 frag->binary_patch_method = patch_method;
1222 return frag;
1223
1224 corrupt:
1225 free(data);
1226 *status_p = -1;
1227 error("corrupt binary patch at line %d: %.*s",
1228 linenr-1, llen-1, buffer);
1229 return NULL;
1230 }
1231
1232 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1233 {
1234 /*
1235 * We have read "GIT binary patch\n"; what follows is a line
1236 * that says the patch method (currently, either "literal" or
1237 * "delta") and the length of data before deflating; a
1238 * sequence of 'length-byte' followed by base-85 encoded data
1239 * follows.
1240 *
1241 * When a binary patch is reversible, there is another binary
1242 * hunk in the same format, starting with patch method (either
1243 * "literal" or "delta") with the length of data, and a sequence
1244 * of length-byte + base-85 encoded data, terminated with another
1245 * empty line. This data, when applied to the postimage, produces
1246 * the preimage.
1247 */
1248 struct fragment *forward;
1249 struct fragment *reverse;
1250 int status;
1251 int used, used_1;
1252
1253 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1254 if (!forward && !status)
1255 /* there has to be one hunk (forward hunk) */
1256 return error("unrecognized binary patch at line %d", linenr-1);
1257 if (status)
1258 /* otherwise we already gave an error message */
1259 return status;
1260
1261 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1262 if (reverse)
1263 used += used_1;
1264 else if (status) {
1265 /*
1266 * Not having reverse hunk is not an error, but having
1267 * a corrupt reverse hunk is.
1268 */
1269 free((void*) forward->patch);
1270 free(forward);
1271 return status;
1272 }
1273 forward->next = reverse;
1274 patch->fragments = forward;
1275 patch->is_binary = 1;
1276 return used;
1277 }
1278
1279 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1280 {
1281 int hdrsize, patchsize;
1282 int offset = find_header(buffer, size, &hdrsize, patch);
1283
1284 if (offset < 0)
1285 return offset;
1286
1287 patch->ws_rule = whitespace_rule(patch->new_name
1288 ? patch->new_name
1289 : patch->old_name);
1290
1291 patchsize = parse_single_patch(buffer + offset + hdrsize,
1292 size - offset - hdrsize, patch);
1293
1294 if (!patchsize) {
1295 static const char *binhdr[] = {
1296 "Binary files ",
1297 "Files ",
1298 NULL,
1299 };
1300 static const char git_binary[] = "GIT binary patch\n";
1301 int i;
1302 int hd = hdrsize + offset;
1303 unsigned long llen = linelen(buffer + hd, size - hd);
1304
1305 if (llen == sizeof(git_binary) - 1 &&
1306 !memcmp(git_binary, buffer + hd, llen)) {
1307 int used;
1308 linenr++;
1309 used = parse_binary(buffer + hd + llen,
1310 size - hd - llen, patch);
1311 if (used)
1312 patchsize = used + llen;
1313 else
1314 patchsize = 0;
1315 }
1316 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1317 for (i = 0; binhdr[i]; i++) {
1318 int len = strlen(binhdr[i]);
1319 if (len < size - hd &&
1320 !memcmp(binhdr[i], buffer + hd, len)) {
1321 linenr++;
1322 patch->is_binary = 1;
1323 patchsize = llen;
1324 break;
1325 }
1326 }
1327 }
1328
1329 /* Empty patch cannot be applied if it is a text patch
1330 * without metadata change. A binary patch appears
1331 * empty to us here.
1332 */
1333 if ((apply || check) &&
1334 (!patch->is_binary && !metadata_changes(patch)))
1335 die("patch with only garbage at line %d", linenr);
1336 }
1337
1338 return offset + hdrsize + patchsize;
1339 }
1340
1341 #define swap(a,b) myswap((a),(b),sizeof(a))
1342
1343 #define myswap(a, b, size) do { \
1344 unsigned char mytmp[size]; \
1345 memcpy(mytmp, &a, size); \
1346 memcpy(&a, &b, size); \
1347 memcpy(&b, mytmp, size); \
1348 } while (0)
1349
1350 static void reverse_patches(struct patch *p)
1351 {
1352 for (; p; p = p->next) {
1353 struct fragment *frag = p->fragments;
1354
1355 swap(p->new_name, p->old_name);
1356 swap(p->new_mode, p->old_mode);
1357 swap(p->is_new, p->is_delete);
1358 swap(p->lines_added, p->lines_deleted);
1359 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1360
1361 for (; frag; frag = frag->next) {
1362 swap(frag->newpos, frag->oldpos);
1363 swap(frag->newlines, frag->oldlines);
1364 }
1365 }
1366 }
1367
1368 static const char pluses[] =
1369 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1370 static const char minuses[]=
1371 "----------------------------------------------------------------------";
1372
1373 static void show_stats(struct patch *patch)
1374 {
1375 struct strbuf qname;
1376 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1377 int max, add, del;
1378
1379 strbuf_init(&qname, 0);
1380 quote_c_style(cp, &qname, NULL, 0);
1381
1382 /*
1383 * "scale" the filename
1384 */
1385 max = max_len;
1386 if (max > 50)
1387 max = 50;
1388
1389 if (qname.len > max) {
1390 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1391 if (!cp)
1392 cp = qname.buf + qname.len + 3 - max;
1393 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1394 }
1395
1396 if (patch->is_binary) {
1397 printf(" %-*s | Bin\n", max, qname.buf);
1398 strbuf_release(&qname);
1399 return;
1400 }
1401
1402 printf(" %-*s |", max, qname.buf);
1403 strbuf_release(&qname);
1404
1405 /*
1406 * scale the add/delete
1407 */
1408 max = max + max_change > 70 ? 70 - max : max_change;
1409 add = patch->lines_added;
1410 del = patch->lines_deleted;
1411
1412 if (max_change > 0) {
1413 int total = ((add + del) * max + max_change / 2) / max_change;
1414 add = (add * max + max_change / 2) / max_change;
1415 del = total - add;
1416 }
1417 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1418 add, pluses, del, minuses);
1419 }
1420
1421 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1422 {
1423 switch (st->st_mode & S_IFMT) {
1424 case S_IFLNK:
1425 strbuf_grow(buf, st->st_size);
1426 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1427 return -1;
1428 strbuf_setlen(buf, st->st_size);
1429 return 0;
1430 case S_IFREG:
1431 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1432 return error("unable to open or read %s", path);
1433 convert_to_git(path, buf->buf, buf->len, buf);
1434 return 0;
1435 default:
1436 return -1;
1437 }
1438 }
1439
1440 static int find_offset(const char *buf, unsigned long size,
1441 const char *fragment, unsigned long fragsize,
1442 int line, int *lines)
1443 {
1444 int i;
1445 unsigned long start, backwards, forwards;
1446
1447 if (fragsize > size)
1448 return -1;
1449
1450 start = 0;
1451 if (line > 1) {
1452 unsigned long offset = 0;
1453 i = line-1;
1454 while (offset + fragsize <= size) {
1455 if (buf[offset++] == '\n') {
1456 start = offset;
1457 if (!--i)
1458 break;
1459 }
1460 }
1461 }
1462
1463 /* Exact line number? */
1464 if ((start + fragsize <= size) &&
1465 !memcmp(buf + start, fragment, fragsize))
1466 return start;
1467
1468 /*
1469 * There's probably some smart way to do this, but I'll leave
1470 * that to the smart and beautiful people. I'm simple and stupid.
1471 */
1472 backwards = start;
1473 forwards = start;
1474 for (i = 0; ; i++) {
1475 unsigned long try;
1476 int n;
1477
1478 /* "backward" */
1479 if (i & 1) {
1480 if (!backwards) {
1481 if (forwards + fragsize > size)
1482 break;
1483 continue;
1484 }
1485 do {
1486 --backwards;
1487 } while (backwards && buf[backwards-1] != '\n');
1488 try = backwards;
1489 } else {
1490 while (forwards + fragsize <= size) {
1491 if (buf[forwards++] == '\n')
1492 break;
1493 }
1494 try = forwards;
1495 }
1496
1497 if (try + fragsize > size)
1498 continue;
1499 if (memcmp(buf + try, fragment, fragsize))
1500 continue;
1501 n = (i >> 1)+1;
1502 if (i & 1)
1503 n = -n;
1504 *lines = n;
1505 return try;
1506 }
1507
1508 /*
1509 * We should start searching forward and backward.
1510 */
1511 return -1;
1512 }
1513
1514 static void remove_first_line(const char **rbuf, int *rsize)
1515 {
1516 const char *buf = *rbuf;
1517 int size = *rsize;
1518 unsigned long offset;
1519 offset = 0;
1520 while (offset <= size) {
1521 if (buf[offset++] == '\n')
1522 break;
1523 }
1524 *rsize = size - offset;
1525 *rbuf = buf + offset;
1526 }
1527
1528 static void remove_last_line(const char **rbuf, int *rsize)
1529 {
1530 const char *buf = *rbuf;
1531 int size = *rsize;
1532 unsigned long offset;
1533 offset = size - 1;
1534 while (offset > 0) {
1535 if (buf[--offset] == '\n')
1536 break;
1537 }
1538 *rsize = offset + 1;
1539 }
1540
1541 static int apply_line(char *output, const char *patch, int plen,
1542 unsigned ws_rule)
1543 {
1544 /*
1545 * plen is number of bytes to be copied from patch,
1546 * starting at patch+1 (patch[0] is '+'). Typically
1547 * patch[plen] is '\n', unless this is the incomplete
1548 * last line.
1549 */
1550 int i;
1551 int add_nl_to_tail = 0;
1552 int fixed = 0;
1553 int last_tab_in_indent = 0;
1554 int last_space_in_indent = 0;
1555 int need_fix_leading_space = 0;
1556 char *buf;
1557
1558 if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1559 *patch != '+') {
1560 memcpy(output, patch + 1, plen);
1561 return plen;
1562 }
1563
1564 /*
1565 * Strip trailing whitespace
1566 */
1567 if ((ws_rule & WS_TRAILING_SPACE) &&
1568 (1 < plen && isspace(patch[plen-1]))) {
1569 if (patch[plen] == '\n')
1570 add_nl_to_tail = 1;
1571 plen--;
1572 while (0 < plen && isspace(patch[plen]))
1573 plen--;
1574 fixed = 1;
1575 }
1576
1577 /*
1578 * Check leading whitespaces (indent)
1579 */
1580 for (i = 1; i < plen; i++) {
1581 char ch = patch[i];
1582 if (ch == '\t') {
1583 last_tab_in_indent = i;
1584 if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1585 0 < last_space_in_indent)
1586 need_fix_leading_space = 1;
1587 } else if (ch == ' ') {
1588 last_space_in_indent = i;
1589 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1590 8 <= i - last_tab_in_indent)
1591 need_fix_leading_space = 1;
1592 }
1593 else
1594 break;
1595 }
1596
1597 buf = output;
1598 if (need_fix_leading_space) {
1599 int consecutive_spaces = 0;
1600 int last = last_tab_in_indent + 1;
1601
1602 if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1603 /* have "last" point at one past the indent */
1604 if (last_tab_in_indent < last_space_in_indent)
1605 last = last_space_in_indent + 1;
1606 else
1607 last = last_tab_in_indent + 1;
1608 }
1609
1610 /*
1611 * between patch[1..last], strip the funny spaces,
1612 * updating them to tab as needed.
1613 */
1614 for (i = 1; i < last; i++, plen--) {
1615 char ch = patch[i];
1616 if (ch != ' ') {
1617 consecutive_spaces = 0;
1618 *output++ = ch;
1619 } else {
1620 consecutive_spaces++;
1621 if (consecutive_spaces == 8) {
1622 *output++ = '\t';
1623 consecutive_spaces = 0;
1624 }
1625 }
1626 }
1627 while (0 < consecutive_spaces--)
1628 *output++ = ' ';
1629 fixed = 1;
1630 i = last;
1631 }
1632 else
1633 i = 1;
1634
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 }
1642
1643 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag,
1644 int inaccurate_eof, unsigned ws_rule)
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;
1656
1657 while (size > 0) {
1658 char first;
1659 int len = linelen(patch, size);
1660 int plen;
1661 int added_blank_line = 0;
1662
1663 if (!len)
1664 break;
1665
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 }
1683
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, ws_rule);
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 }
1725
1726 if (inaccurate_eof &&
1727 oldsize > 0 && old[oldsize - 1] == '\n' &&
1728 newsize > 0 && new[newsize - 1] == '\n') {
1729 oldsize--;
1730 newsize--;
1731 }
1732
1733 oldlines = old;
1734 newlines = new;
1735 leading = frag->leading;
1736 trailing = frag->trailing;
1737
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 }
1754
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;
1768
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);
1777
1778 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1779 offset = 0;
1780 break;
1781 }
1782
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 }
1807
1808 if (offset && apply_verbosely)
1809 error("while searching for:\n%.*s", oldsize, oldlines);
1810
1811 free(old);
1812 free(new);
1813 return offset;
1814 }
1815
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;
1821
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 }
1847
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];
1852
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);
1863
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 }
1882
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 }
1888
1889 if (has_sha1_file(sha1)) {
1890 /* We already have the postimage */
1891 enum object_type type;
1892 unsigned long size;
1893 char *result;
1894
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);
1911
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 }
1918
1919 return 0;
1920 }
1921
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;
1926 unsigned ws_rule = patch->ws_rule;
1927 unsigned inaccurate_eof = patch->inaccurate_eof;
1928
1929 if (patch->is_binary)
1930 return apply_binary(buf, patch);
1931
1932 while (frag) {
1933 if (apply_one_fragment(buf, frag, inaccurate_eof, ws_rule)) {
1934 error("patch failed: %s:%ld", name, frag->oldpos);
1935 if (!apply_with_reject)
1936 return -1;
1937 frag->rejected = 1;
1938 }
1939 frag = frag->next;
1940 }
1941 return 0;
1942 }
1943
1944 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1945 {
1946 if (!ce)
1947 return 0;
1948
1949 if (S_ISGITLINK(ce->ce_mode)) {
1950 strbuf_grow(buf, 100);
1951 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1952 } else {
1953 enum object_type type;
1954 unsigned long sz;
1955 char *result;
1956
1957 result = read_sha1_file(ce->sha1, &type, &sz);
1958 if (!result)
1959 return -1;
1960 /* XXX read_sha1_file NUL-terminates */
1961 strbuf_attach(buf, result, sz, sz + 1);
1962 }
1963 return 0;
1964 }
1965
1966 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1967 {
1968 struct strbuf buf;
1969
1970 strbuf_init(&buf, 0);
1971 if (cached) {
1972 if (read_file_or_gitlink(ce, &buf))
1973 return error("read of %s failed", patch->old_name);
1974 } else if (patch->old_name) {
1975 if (S_ISGITLINK(patch->old_mode)) {
1976 if (ce) {
1977 read_file_or_gitlink(ce, &buf);
1978 } else {
1979 /*
1980 * There is no way to apply subproject
1981 * patch without looking at the index.
1982 */
1983 patch->fragments = NULL;
1984 }
1985 } else {
1986 if (read_old_data(st, patch->old_name, &buf))
1987 return error("read of %s failed", patch->old_name);
1988 }
1989 }
1990
1991 if (apply_fragments(&buf, patch) < 0)
1992 return -1; /* note with --reject this succeeds. */
1993 patch->result = strbuf_detach(&buf, &patch->resultsize);
1994
1995 if (0 < patch->is_delete && patch->resultsize)
1996 return error("removal patch leaves file contents");
1997
1998 return 0;
1999 }
2000
2001 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2002 {
2003 struct stat nst;
2004 if (!lstat(new_name, &nst)) {
2005 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2006 return 0;
2007 /*
2008 * A leading component of new_name might be a symlink
2009 * that is going to be removed with this patch, but
2010 * still pointing at somewhere that has the path.
2011 * In such a case, path "new_name" does not exist as
2012 * far as git is concerned.
2013 */
2014 if (has_symlink_leading_path(new_name, NULL))
2015 return 0;
2016
2017 return error("%s: already exists in working directory", new_name);
2018 }
2019 else if ((errno != ENOENT) && (errno != ENOTDIR))
2020 return error("%s: %s", new_name, strerror(errno));
2021 return 0;
2022 }
2023
2024 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2025 {
2026 if (S_ISGITLINK(ce->ce_mode)) {
2027 if (!S_ISDIR(st->st_mode))
2028 return -1;
2029 return 0;
2030 }
2031 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2032 }
2033
2034 static int check_patch(struct patch *patch, struct patch *prev_patch)
2035 {
2036 struct stat st;
2037 const char *old_name = patch->old_name;
2038 const char *new_name = patch->new_name;
2039 const char *name = old_name ? old_name : new_name;
2040 struct cache_entry *ce = NULL;
2041 int ok_if_exists;
2042
2043 patch->rejected = 1; /* we will drop this after we succeed */
2044
2045 /*
2046 * Make sure that we do not have local modifications from the
2047 * index when we are looking at the index. Also make sure
2048 * we have the preimage file to be patched in the work tree,
2049 * unless --cached, which tells git to apply only in the index.
2050 */
2051 if (old_name) {
2052 int stat_ret = 0;
2053 unsigned st_mode = 0;
2054
2055 if (!cached)
2056 stat_ret = lstat(old_name, &st);
2057 if (check_index) {
2058 int pos = cache_name_pos(old_name, strlen(old_name));
2059 if (pos < 0)
2060 return error("%s: does not exist in index",
2061 old_name);
2062 ce = active_cache[pos];
2063 if (stat_ret < 0) {
2064 struct checkout costate;
2065 if (errno != ENOENT)
2066 return error("%s: %s", old_name,
2067 strerror(errno));
2068 /* checkout */
2069 costate.base_dir = "";
2070 costate.base_dir_len = 0;
2071 costate.force = 0;
2072 costate.quiet = 0;
2073 costate.not_new = 0;
2074 costate.refresh_cache = 1;
2075 if (checkout_entry(ce,
2076 &costate,
2077 NULL) ||
2078 lstat(old_name, &st))
2079 return -1;
2080 }
2081 if (!cached && verify_index_match(ce, &st))
2082 return error("%s: does not match index",
2083 old_name);
2084 if (cached)
2085 st_mode = ce->ce_mode;
2086 } else if (stat_ret < 0)
2087 return error("%s: %s", old_name, strerror(errno));
2088
2089 if (!cached)
2090 st_mode = ce_mode_from_stat(ce, st.st_mode);
2091
2092 if (patch->is_new < 0)
2093 patch->is_new = 0;
2094 if (!patch->old_mode)
2095 patch->old_mode = st_mode;
2096 if ((st_mode ^ patch->old_mode) & S_IFMT)
2097 return error("%s: wrong type", old_name);
2098 if (st_mode != patch->old_mode)
2099 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2100 old_name, st_mode, patch->old_mode);
2101 }
2102
2103 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2104 !strcmp(prev_patch->old_name, new_name))
2105 /*
2106 * A type-change diff is always split into a patch to
2107 * delete old, immediately followed by a patch to
2108 * create new (see diff.c::run_diff()); in such a case
2109 * it is Ok that the entry to be deleted by the
2110 * previous patch is still in the working tree and in
2111 * the index.
2112 */
2113 ok_if_exists = 1;
2114 else
2115 ok_if_exists = 0;
2116
2117 if (new_name &&
2118 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2119 if (check_index &&
2120 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2121 !ok_if_exists)
2122 return error("%s: already exists in index", new_name);
2123 if (!cached) {
2124 int err = check_to_create_blob(new_name, ok_if_exists);
2125 if (err)
2126 return err;
2127 }
2128 if (!patch->new_mode) {
2129 if (0 < patch->is_new)
2130 patch->new_mode = S_IFREG | 0644;
2131 else
2132 patch->new_mode = patch->old_mode;
2133 }
2134 }
2135
2136 if (new_name && old_name) {
2137 int same = !strcmp(old_name, new_name);
2138 if (!patch->new_mode)
2139 patch->new_mode = patch->old_mode;
2140 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2141 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2142 patch->new_mode, new_name, patch->old_mode,
2143 same ? "" : " of ", same ? "" : old_name);
2144 }
2145
2146 if (apply_data(patch, &st, ce) < 0)
2147 return error("%s: patch does not apply", name);
2148 patch->rejected = 0;
2149 return 0;
2150 }
2151
2152 static int check_patch_list(struct patch *patch)
2153 {
2154 struct patch *prev_patch = NULL;
2155 int err = 0;
2156
2157 for (prev_patch = NULL; patch ; patch = patch->next) {
2158 if (apply_verbosely)
2159 say_patch_name(stderr,
2160 "Checking patch ", patch, "...\n");
2161 err |= check_patch(patch, prev_patch);
2162 prev_patch = patch;
2163 }
2164 return err;
2165 }
2166
2167 /* This function tries to read the sha1 from the current index */
2168 static int get_current_sha1(const char *path, unsigned char *sha1)
2169 {
2170 int pos;
2171
2172 if (read_cache() < 0)
2173 return -1;
2174 pos = cache_name_pos(path, strlen(path));
2175 if (pos < 0)
2176 return -1;
2177 hashcpy(sha1, active_cache[pos]->sha1);
2178 return 0;
2179 }
2180
2181 /* Build an index that contains the just the files needed for a 3way merge */
2182 static void build_fake_ancestor(struct patch *list, const char *filename)
2183 {
2184 struct patch *patch;
2185 struct index_state result = { 0 };
2186 int fd;
2187
2188 /* Once we start supporting the reverse patch, it may be
2189 * worth showing the new sha1 prefix, but until then...
2190 */
2191 for (patch = list; patch; patch = patch->next) {
2192 const unsigned char *sha1_ptr;
2193 unsigned char sha1[20];
2194 struct cache_entry *ce;
2195 const char *name;
2196
2197 name = patch->old_name ? patch->old_name : patch->new_name;
2198 if (0 < patch->is_new)
2199 continue;
2200 else if (get_sha1(patch->old_sha1_prefix, sha1))
2201 /* git diff has no index line for mode/type changes */
2202 if (!patch->lines_added && !patch->lines_deleted) {
2203 if (get_current_sha1(patch->new_name, sha1) ||
2204 get_current_sha1(patch->old_name, sha1))
2205 die("mode change for %s, which is not "
2206 "in current HEAD", name);
2207 sha1_ptr = sha1;
2208 } else
2209 die("sha1 information is lacking or useless "
2210 "(%s).", name);
2211 else
2212 sha1_ptr = sha1;
2213
2214 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2215 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2216 die ("Could not add %s to temporary index", name);
2217 }
2218
2219 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2220 if (fd < 0 || write_index(&result, fd) || close(fd))
2221 die ("Could not write temporary index to %s", filename);
2222
2223 discard_index(&result);
2224 }
2225
2226 static void stat_patch_list(struct patch *patch)
2227 {
2228 int files, adds, dels;
2229
2230 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2231 files++;
2232 adds += patch->lines_added;
2233 dels += patch->lines_deleted;
2234 show_stats(patch);
2235 }
2236
2237 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2238 }
2239
2240 static void numstat_patch_list(struct patch *patch)
2241 {
2242 for ( ; patch; patch = patch->next) {
2243 const char *name;
2244 name = patch->new_name ? patch->new_name : patch->old_name;
2245 if (patch->is_binary)
2246 printf("-\t-\t");
2247 else
2248 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2249 write_name_quoted(name, stdout, line_termination);
2250 }
2251 }
2252
2253 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2254 {
2255 if (mode)
2256 printf(" %s mode %06o %s\n", newdelete, mode, name);
2257 else
2258 printf(" %s %s\n", newdelete, name);
2259 }
2260
2261 static void show_mode_change(struct patch *p, int show_name)
2262 {
2263 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2264 if (show_name)
2265 printf(" mode change %06o => %06o %s\n",
2266 p->old_mode, p->new_mode, p->new_name);
2267 else
2268 printf(" mode change %06o => %06o\n",
2269 p->old_mode, p->new_mode);
2270 }
2271 }
2272
2273 static void show_rename_copy(struct patch *p)
2274 {
2275 const char *renamecopy = p->is_rename ? "rename" : "copy";
2276 const char *old, *new;
2277
2278 /* Find common prefix */
2279 old = p->old_name;
2280 new = p->new_name;
2281 while (1) {
2282 const char *slash_old, *slash_new;
2283 slash_old = strchr(old, '/');
2284 slash_new = strchr(new, '/');
2285 if (!slash_old ||
2286 !slash_new ||
2287 slash_old - old != slash_new - new ||
2288 memcmp(old, new, slash_new - new))
2289 break;
2290 old = slash_old + 1;
2291 new = slash_new + 1;
2292 }
2293 /* p->old_name thru old is the common prefix, and old and new
2294 * through the end of names are renames
2295 */
2296 if (old != p->old_name)
2297 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2298 (int)(old - p->old_name), p->old_name,
2299 old, new, p->score);
2300 else
2301 printf(" %s %s => %s (%d%%)\n", renamecopy,
2302 p->old_name, p->new_name, p->score);
2303 show_mode_change(p, 0);
2304 }
2305
2306 static void summary_patch_list(struct patch *patch)
2307 {
2308 struct patch *p;
2309
2310 for (p = patch; p; p = p->next) {
2311 if (p->is_new)
2312 show_file_mode_name("create", p->new_mode, p->new_name);
2313 else if (p->is_delete)
2314 show_file_mode_name("delete", p->old_mode, p->old_name);
2315 else {
2316 if (p->is_rename || p->is_copy)
2317 show_rename_copy(p);
2318 else {
2319 if (p->score) {
2320 printf(" rewrite %s (%d%%)\n",
2321 p->new_name, p->score);
2322 show_mode_change(p, 0);
2323 }
2324 else
2325 show_mode_change(p, 1);
2326 }
2327 }
2328 }
2329 }
2330
2331 static void patch_stats(struct patch *patch)
2332 {
2333 int lines = patch->lines_added + patch->lines_deleted;
2334
2335 if (lines > max_change)
2336 max_change = lines;
2337 if (patch->old_name) {
2338 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2339 if (!len)
2340 len = strlen(patch->old_name);
2341 if (len > max_len)
2342 max_len = len;
2343 }
2344 if (patch->new_name) {
2345 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2346 if (!len)
2347 len = strlen(patch->new_name);
2348 if (len > max_len)
2349 max_len = len;
2350 }
2351 }
2352
2353 static void remove_file(struct patch *patch, int rmdir_empty)
2354 {
2355 if (update_index) {
2356 if (remove_file_from_cache(patch->old_name) < 0)
2357 die("unable to remove %s from index", patch->old_name);
2358 }
2359 if (!cached) {
2360 if (S_ISGITLINK(patch->old_mode)) {
2361 if (rmdir(patch->old_name))
2362 warning("unable to remove submodule %s",
2363 patch->old_name);
2364 } else if (!unlink(patch->old_name) && rmdir_empty) {
2365 char *name = xstrdup(patch->old_name);
2366 char *end = strrchr(name, '/');
2367 while (end) {
2368 *end = 0;
2369 if (rmdir(name))
2370 break;
2371 end = strrchr(name, '/');
2372 }
2373 free(name);
2374 }
2375 }
2376 }
2377
2378 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2379 {
2380 struct stat st;
2381 struct cache_entry *ce;
2382 int namelen = strlen(path);
2383 unsigned ce_size = cache_entry_size(namelen);
2384
2385 if (!update_index)
2386 return;
2387
2388 ce = xcalloc(1, ce_size);
2389 memcpy(ce->name, path, namelen);
2390 ce->ce_mode = create_ce_mode(mode);
2391 ce->ce_flags = namelen;
2392 if (S_ISGITLINK(mode)) {
2393 const char *s = buf;
2394
2395 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2396 die("corrupt patch for subproject %s", path);
2397 } else {
2398 if (!cached) {
2399 if (lstat(path, &st) < 0)
2400 die("unable to stat newly created file %s",
2401 path);
2402 fill_stat_cache_info(ce, &st);
2403 }
2404 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2405 die("unable to create backing store for newly created file %s", path);
2406 }
2407 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2408 die("unable to add cache entry for %s", path);
2409 }
2410
2411 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2412 {
2413 int fd;
2414 struct strbuf nbuf;
2415
2416 if (S_ISGITLINK(mode)) {
2417 struct stat st;
2418 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2419 return 0;
2420 return mkdir(path, 0777);
2421 }
2422
2423 if (has_symlinks && S_ISLNK(mode))
2424 /* Although buf:size is counted string, it also is NUL
2425 * terminated.
2426 */
2427 return symlink(buf, path);
2428
2429 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2430 if (fd < 0)
2431 return -1;
2432
2433 strbuf_init(&nbuf, 0);
2434 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2435 size = nbuf.len;
2436 buf = nbuf.buf;
2437 }
2438 write_or_die(fd, buf, size);
2439 strbuf_release(&nbuf);
2440
2441 if (close(fd) < 0)
2442 die("closing file %s: %s", path, strerror(errno));
2443 return 0;
2444 }
2445
2446 /*
2447 * We optimistically assume that the directories exist,
2448 * which is true 99% of the time anyway. If they don't,
2449 * we create them and try again.
2450 */
2451 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2452 {
2453 if (cached)
2454 return;
2455 if (!try_create_file(path, mode, buf, size))
2456 return;
2457
2458 if (errno == ENOENT) {
2459 if (safe_create_leading_directories(path))
2460 return;
2461 if (!try_create_file(path, mode, buf, size))
2462 return;
2463 }
2464
2465 if (errno == EEXIST || errno == EACCES) {
2466 /* We may be trying to create a file where a directory
2467 * used to be.
2468 */
2469 struct stat st;
2470 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2471 errno = EEXIST;
2472 }
2473
2474 if (errno == EEXIST) {
2475 unsigned int nr = getpid();
2476
2477 for (;;) {
2478 const char *newpath;
2479 newpath = mkpath("%s~%u", path, nr);
2480 if (!try_create_file(newpath, mode, buf, size)) {
2481 if (!rename(newpath, path))
2482 return;
2483 unlink(newpath);
2484 break;
2485 }
2486 if (errno != EEXIST)
2487 break;
2488 ++nr;
2489 }
2490 }
2491 die("unable to write file %s mode %o", path, mode);
2492 }
2493
2494 static void create_file(struct patch *patch)
2495 {
2496 char *path = patch->new_name;
2497 unsigned mode = patch->new_mode;
2498 unsigned long size = patch->resultsize;
2499 char *buf = patch->result;
2500
2501 if (!mode)
2502 mode = S_IFREG | 0644;
2503 create_one_file(path, mode, buf, size);
2504 add_index_file(path, mode, buf, size);
2505 }
2506
2507 /* phase zero is to remove, phase one is to create */
2508 static void write_out_one_result(struct patch *patch, int phase)
2509 {
2510 if (patch->is_delete > 0) {
2511 if (phase == 0)
2512 remove_file(patch, 1);
2513 return;
2514 }
2515 if (patch->is_new > 0 || patch->is_copy) {
2516 if (phase == 1)
2517 create_file(patch);
2518 return;
2519 }
2520 /*
2521 * Rename or modification boils down to the same
2522 * thing: remove the old, write the new
2523 */
2524 if (phase == 0)
2525 remove_file(patch, patch->is_rename);
2526 if (phase == 1)
2527 create_file(patch);
2528 }
2529
2530 static int write_out_one_reject(struct patch *patch)
2531 {
2532 FILE *rej;
2533 char namebuf[PATH_MAX];
2534 struct fragment *frag;
2535 int cnt = 0;
2536
2537 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2538 if (!frag->rejected)
2539 continue;
2540 cnt++;
2541 }
2542
2543 if (!cnt) {
2544 if (apply_verbosely)
2545 say_patch_name(stderr,
2546 "Applied patch ", patch, " cleanly.\n");
2547 return 0;
2548 }
2549
2550 /* This should not happen, because a removal patch that leaves
2551 * contents are marked "rejected" at the patch level.
2552 */
2553 if (!patch->new_name)
2554 die("internal error");
2555
2556 /* Say this even without --verbose */
2557 say_patch_name(stderr, "Applying patch ", patch, " with");
2558 fprintf(stderr, " %d rejects...\n", cnt);
2559
2560 cnt = strlen(patch->new_name);
2561 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2562 cnt = ARRAY_SIZE(namebuf) - 5;
2563 fprintf(stderr,
2564 "warning: truncating .rej filename to %.*s.rej",
2565 cnt - 1, patch->new_name);
2566 }
2567 memcpy(namebuf, patch->new_name, cnt);
2568 memcpy(namebuf + cnt, ".rej", 5);
2569
2570 rej = fopen(namebuf, "w");
2571 if (!rej)
2572 return error("cannot open %s: %s", namebuf, strerror(errno));
2573
2574 /* Normal git tools never deal with .rej, so do not pretend
2575 * this is a git patch by saying --git nor give extended
2576 * headers. While at it, maybe please "kompare" that wants
2577 * the trailing TAB and some garbage at the end of line ;-).
2578 */
2579 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2580 patch->new_name, patch->new_name);
2581 for (cnt = 1, frag = patch->fragments;
2582 frag;
2583 cnt++, frag = frag->next) {
2584 if (!frag->rejected) {
2585 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2586 continue;
2587 }
2588 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2589 fprintf(rej, "%.*s", frag->size, frag->patch);
2590 if (frag->patch[frag->size-1] != '\n')
2591 fputc('\n', rej);
2592 }
2593 fclose(rej);
2594 return -1;
2595 }
2596
2597 static int write_out_results(struct patch *list, int skipped_patch)
2598 {
2599 int phase;
2600 int errs = 0;
2601 struct patch *l;
2602
2603 if (!list && !skipped_patch)
2604 return error("No changes");
2605
2606 for (phase = 0; phase < 2; phase++) {
2607 l = list;
2608 while (l) {
2609 if (l->rejected)
2610 errs = 1;
2611 else {
2612 write_out_one_result(l, phase);
2613 if (phase == 1 && write_out_one_reject(l))
2614 errs = 1;
2615 }
2616 l = l->next;
2617 }
2618 }
2619 return errs;
2620 }
2621
2622 static struct lock_file lock_file;
2623
2624 static struct excludes {
2625 struct excludes *next;
2626 const char *path;
2627 } *excludes;
2628
2629 static int use_patch(struct patch *p)
2630 {
2631 const char *pathname = p->new_name ? p->new_name : p->old_name;
2632 struct excludes *x = excludes;
2633 while (x) {
2634 if (fnmatch(x->path, pathname, 0) == 0)
2635 return 0;
2636 x = x->next;
2637 }
2638 if (0 < prefix_length) {
2639 int pathlen = strlen(pathname);
2640 if (pathlen <= prefix_length ||
2641 memcmp(prefix, pathname, prefix_length))
2642 return 0;
2643 }
2644 return 1;
2645 }
2646
2647 static void prefix_one(char **name)
2648 {
2649 char *old_name = *name;
2650 if (!old_name)
2651 return;
2652 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2653 free(old_name);
2654 }
2655
2656 static void prefix_patches(struct patch *p)
2657 {
2658 if (!prefix || p->is_toplevel_relative)
2659 return;
2660 for ( ; p; p = p->next) {
2661 if (p->new_name == p->old_name) {
2662 char *prefixed = p->new_name;
2663 prefix_one(&prefixed);
2664 p->new_name = p->old_name = prefixed;
2665 }
2666 else {
2667 prefix_one(&p->new_name);
2668 prefix_one(&p->old_name);
2669 }
2670 }
2671 }
2672
2673 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2674 {
2675 size_t offset;
2676 struct strbuf buf;
2677 struct patch *list = NULL, **listp = &list;
2678 int skipped_patch = 0;
2679
2680 strbuf_init(&buf, 0);
2681 patch_input_file = filename;
2682 read_patch_file(&buf, fd);
2683 offset = 0;
2684 while (offset < buf.len) {
2685 struct patch *patch;
2686 int nr;
2687
2688 patch = xcalloc(1, sizeof(*patch));
2689 patch->inaccurate_eof = inaccurate_eof;
2690 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2691 if (nr < 0)
2692 break;
2693 if (apply_in_reverse)
2694 reverse_patches(patch);
2695 if (prefix)
2696 prefix_patches(patch);
2697 if (use_patch(patch)) {
2698 patch_stats(patch);
2699 *listp = patch;
2700 listp = &patch->next;
2701 }
2702 else {
2703 /* perhaps free it a bit better? */
2704 free(patch);
2705 skipped_patch++;
2706 }
2707 offset += nr;
2708 }
2709
2710 if (whitespace_error && (ws_error_action == die_on_ws_error))
2711 apply = 0;
2712
2713 update_index = check_index && apply;
2714 if (update_index && newfd < 0)
2715 newfd = hold_locked_index(&lock_file, 1);
2716
2717 if (check_index) {
2718 if (read_cache() < 0)
2719 die("unable to read index file");
2720 }
2721
2722 if ((check || apply) &&
2723 check_patch_list(list) < 0 &&
2724 !apply_with_reject)
2725 exit(1);
2726
2727 if (apply && write_out_results(list, skipped_patch))
2728 exit(1);
2729
2730 if (fake_ancestor)
2731 build_fake_ancestor(list, fake_ancestor);
2732
2733 if (diffstat)
2734 stat_patch_list(list);
2735
2736 if (numstat)
2737 numstat_patch_list(list);
2738
2739 if (summary)
2740 summary_patch_list(list);
2741
2742 strbuf_release(&buf);
2743 return 0;
2744 }
2745
2746 static int git_apply_config(const char *var, const char *value)
2747 {
2748 if (!strcmp(var, "apply.whitespace")) {
2749 apply_default_whitespace = xstrdup(value);
2750 return 0;
2751 }
2752 return git_default_config(var, value);
2753 }
2754
2755
2756 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2757 {
2758 int i;
2759 int read_stdin = 1;
2760 int inaccurate_eof = 0;
2761 int errs = 0;
2762 int is_not_gitdir = 0;
2763
2764 const char *whitespace_option = NULL;
2765
2766 prefix = setup_git_directory_gently(&is_not_gitdir);
2767 prefix_length = prefix ? strlen(prefix) : 0;
2768 git_config(git_apply_config);
2769 if (apply_default_whitespace)
2770 parse_whitespace_option(apply_default_whitespace);
2771
2772 for (i = 1; i < argc; i++) {
2773 const char *arg = argv[i];
2774 char *end;
2775 int fd;
2776
2777 if (!strcmp(arg, "-")) {
2778 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2779 read_stdin = 0;
2780 continue;
2781 }
2782 if (!prefixcmp(arg, "--exclude=")) {
2783 struct excludes *x = xmalloc(sizeof(*x));
2784 x->path = arg + 10;
2785 x->next = excludes;
2786 excludes = x;
2787 continue;
2788 }
2789 if (!prefixcmp(arg, "-p")) {
2790 p_value = atoi(arg + 2);
2791 p_value_known = 1;
2792 continue;
2793 }
2794 if (!strcmp(arg, "--no-add")) {
2795 no_add = 1;
2796 continue;
2797 }
2798 if (!strcmp(arg, "--stat")) {
2799 apply = 0;
2800 diffstat = 1;
2801 continue;
2802 }
2803 if (!strcmp(arg, "--allow-binary-replacement") ||
2804 !strcmp(arg, "--binary")) {
2805 continue; /* now no-op */
2806 }
2807 if (!strcmp(arg, "--numstat")) {
2808 apply = 0;
2809 numstat = 1;
2810 continue;
2811 }
2812 if (!strcmp(arg, "--summary")) {
2813 apply = 0;
2814 summary = 1;
2815 continue;
2816 }
2817 if (!strcmp(arg, "--check")) {
2818 apply = 0;
2819 check = 1;
2820 continue;
2821 }
2822 if (!strcmp(arg, "--index")) {
2823 if (is_not_gitdir)
2824 die("--index outside a repository");
2825 check_index = 1;
2826 continue;
2827 }
2828 if (!strcmp(arg, "--cached")) {
2829 if (is_not_gitdir)
2830 die("--cached outside a repository");
2831 check_index = 1;
2832 cached = 1;
2833 continue;
2834 }
2835 if (!strcmp(arg, "--apply")) {
2836 apply = 1;
2837 continue;
2838 }
2839 if (!strcmp(arg, "--build-fake-ancestor")) {
2840 apply = 0;
2841 if (++i >= argc)
2842 die ("need a filename");
2843 fake_ancestor = argv[i];
2844 continue;
2845 }
2846 if (!strcmp(arg, "-z")) {
2847 line_termination = 0;
2848 continue;
2849 }
2850 if (!prefixcmp(arg, "-C")) {
2851 p_context = strtoul(arg + 2, &end, 0);
2852 if (*end != '\0')
2853 die("unrecognized context count '%s'", arg + 2);
2854 continue;
2855 }
2856 if (!prefixcmp(arg, "--whitespace=")) {
2857 whitespace_option = arg + 13;
2858 parse_whitespace_option(arg + 13);
2859 continue;
2860 }
2861 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2862 apply_in_reverse = 1;
2863 continue;
2864 }
2865 if (!strcmp(arg, "--unidiff-zero")) {
2866 unidiff_zero = 1;
2867 continue;
2868 }
2869 if (!strcmp(arg, "--reject")) {
2870 apply = apply_with_reject = apply_verbosely = 1;
2871 continue;
2872 }
2873 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2874 apply_verbosely = 1;
2875 continue;
2876 }
2877 if (!strcmp(arg, "--inaccurate-eof")) {
2878 inaccurate_eof = 1;
2879 continue;
2880 }
2881 if (0 < prefix_length)
2882 arg = prefix_filename(prefix, prefix_length, arg);
2883
2884 fd = open(arg, O_RDONLY);
2885 if (fd < 0)
2886 usage(apply_usage);
2887 read_stdin = 0;
2888 set_default_whitespace_mode(whitespace_option);
2889 errs |= apply_patch(fd, arg, inaccurate_eof);
2890 close(fd);
2891 }
2892 set_default_whitespace_mode(whitespace_option);
2893 if (read_stdin)
2894 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2895 if (whitespace_error) {
2896 if (squelch_whitespace_errors &&
2897 squelch_whitespace_errors < whitespace_error) {
2898 int squelched =
2899 whitespace_error - squelch_whitespace_errors;
2900 fprintf(stderr, "warning: squelched %d "
2901 "whitespace error%s\n",
2902 squelched,
2903 squelched == 1 ? "" : "s");
2904 }
2905 if (ws_error_action == die_on_ws_error)
2906 die("%d line%s add%s whitespace errors.",
2907 whitespace_error,
2908 whitespace_error == 1 ? "" : "s",
2909 whitespace_error == 1 ? "s" : "");
2910 if (applied_after_fixing_ws && apply)
2911 fprintf(stderr, "warning: %d line%s applied after"
2912 " fixing whitespace errors.\n",
2913 applied_after_fixing_ws,
2914 applied_after_fixing_ws == 1 ? "" : "s");
2915 else if (whitespace_error)
2916 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2917 whitespace_error,
2918 whitespace_error == 1 ? "" : "s",
2919 whitespace_error == 1 ? "s" : "");
2920 }
2921
2922 if (update_index) {
2923 if (write_cache(newfd, active_cache, active_nr) ||
2924 commit_locked_index(&lock_file))
2925 die("Unable to write new index file");
2926 }
2927
2928 return !!errs;
2929 }