]> git.ipfire.org Git - thirdparty/git.git/blob - builtin-apply.c
strbuf_read_file enhancement, and use it.
[thirdparty/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 int show_index_info;
45 static int line_termination = '\n';
46 static unsigned long p_context = ULONG_MAX;
47 static const char apply_usage[] =
48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
49
50 static enum whitespace_eol {
51 nowarn_whitespace,
52 warn_on_whitespace,
53 error_on_whitespace,
54 strip_whitespace,
55 } new_whitespace = warn_on_whitespace;
56 static int whitespace_error;
57 static int squelch_whitespace_errors = 5;
58 static int applied_after_fixing_ws;
59 static const char *patch_input_file;
60
61 static void parse_whitespace_option(const char *option)
62 {
63 if (!option) {
64 new_whitespace = warn_on_whitespace;
65 return;
66 }
67 if (!strcmp(option, "warn")) {
68 new_whitespace = warn_on_whitespace;
69 return;
70 }
71 if (!strcmp(option, "nowarn")) {
72 new_whitespace = nowarn_whitespace;
73 return;
74 }
75 if (!strcmp(option, "error")) {
76 new_whitespace = error_on_whitespace;
77 return;
78 }
79 if (!strcmp(option, "error-all")) {
80 new_whitespace = error_on_whitespace;
81 squelch_whitespace_errors = 0;
82 return;
83 }
84 if (!strcmp(option, "strip")) {
85 new_whitespace = strip_whitespace;
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 new_whitespace = (apply
95 ? warn_on_whitespace
96 : nowarn_whitespace);
97 }
98 }
99
100 /*
101 * For "diff-stat" like behaviour, we keep track of the biggest change
102 * we've seen, and the longest filename. That allows us to do simple
103 * scaling.
104 */
105 static int max_change, max_len;
106
107 /*
108 * Various "current state", notably line numbers and what
109 * file (and how) we're patching right now.. The "is_xxxx"
110 * things are flags, where -1 means "don't know yet".
111 */
112 static int linenr = 1;
113
114 /*
115 * This represents one "hunk" from a patch, starting with
116 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
117 * patch text is pointed at by patch, and its byte length
118 * is stored in size. leading and trailing are the number
119 * of context lines.
120 */
121 struct fragment {
122 unsigned long leading, trailing;
123 unsigned long oldpos, oldlines;
124 unsigned long newpos, newlines;
125 const char *patch;
126 int size;
127 int rejected;
128 struct fragment *next;
129 };
130
131 /*
132 * When dealing with a binary patch, we reuse "leading" field
133 * to store the type of the binary hunk, either deflated "delta"
134 * or deflated "literal".
135 */
136 #define binary_patch_method leading
137 #define BINARY_DELTA_DEFLATED 1
138 #define BINARY_LITERAL_DEFLATED 2
139
140 struct patch {
141 char *new_name, *old_name, *def_name;
142 unsigned int old_mode, new_mode;
143 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
144 int rejected;
145 unsigned long deflate_origlen;
146 int lines_added, lines_deleted;
147 int score;
148 unsigned int is_toplevel_relative:1;
149 unsigned int inaccurate_eof:1;
150 unsigned int is_binary:1;
151 unsigned int is_copy:1;
152 unsigned int is_rename:1;
153 struct fragment *fragments;
154 char *result;
155 unsigned long resultsize;
156 char old_sha1_prefix[41];
157 char new_sha1_prefix[41];
158 struct patch *next;
159 };
160
161 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
162 {
163 fputs(pre, output);
164 if (patch->old_name && patch->new_name &&
165 strcmp(patch->old_name, patch->new_name)) {
166 quote_c_style(patch->old_name, NULL, output, 0);
167 fputs(" => ", output);
168 quote_c_style(patch->new_name, NULL, output, 0);
169 } else {
170 const char *n = patch->new_name;
171 if (!n)
172 n = patch->old_name;
173 quote_c_style(n, NULL, output, 0);
174 }
175 fputs(post, output);
176 }
177
178 #define CHUNKSIZE (8192)
179 #define SLOP (16)
180
181 static void *read_patch_file(int fd, size_t *sizep)
182 {
183 struct strbuf buf;
184
185 strbuf_init(&buf, 0);
186 if (strbuf_read(&buf, fd, 0) < 0)
187 die("git-apply: read returned %s", strerror(errno));
188
189 /*
190 * Make sure that we have some slop in the buffer
191 * so that we can do speculative "memcmp" etc, and
192 * see to it that it is NUL-filled.
193 */
194 strbuf_grow(&buf, SLOP);
195 memset(buf.buf + buf.len, 0, SLOP);
196 return strbuf_detach(&buf, sizep);
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 /* Proposed "new-style" GNU patch/diff format; see
237 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
238 */
239 strbuf_init(&name, 0);
240 if (!unquote_c_style(&name, line, NULL)) {
241 char *cp;
242
243 for (cp = name.buf; p_value; p_value--) {
244 cp = strchr(cp, '/');
245 if (!cp)
246 break;
247 cp++;
248 }
249 if (cp) {
250 /* name can later be freed, so we need
251 * to memmove, not just return cp
252 */
253 strbuf_remove(&name, 0, cp - name.buf);
254 free(def);
255 return strbuf_detach(&name, NULL);
256 }
257 }
258 strbuf_release(&name);
259 }
260
261 for (;;) {
262 char c = *line;
263
264 if (isspace(c)) {
265 if (c == '\n')
266 break;
267 if (name_terminate(start, line-start, c, terminate))
268 break;
269 }
270 line++;
271 if (c == '/' && !--p_value)
272 start = line;
273 }
274 if (!start)
275 return def;
276 len = line - start;
277 if (!len)
278 return def;
279
280 /*
281 * Generally we prefer the shorter name, especially
282 * if the other one is just a variation of that with
283 * something else tacked on to the end (ie "file.orig"
284 * or "file~").
285 */
286 if (def) {
287 int deflen = strlen(def);
288 if (deflen < len && !strncmp(start, def, deflen))
289 return def;
290 free(def);
291 }
292
293 return xmemdupz(start, len);
294 }
295
296 static int count_slashes(const char *cp)
297 {
298 int cnt = 0;
299 char ch;
300
301 while ((ch = *cp++))
302 if (ch == '/')
303 cnt++;
304 return cnt;
305 }
306
307 /*
308 * Given the string after "--- " or "+++ ", guess the appropriate
309 * p_value for the given patch.
310 */
311 static int guess_p_value(const char *nameline)
312 {
313 char *name, *cp;
314 int val = -1;
315
316 if (is_dev_null(nameline))
317 return -1;
318 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
319 if (!name)
320 return -1;
321 cp = strchr(name, '/');
322 if (!cp)
323 val = 0;
324 else if (prefix) {
325 /*
326 * Does it begin with "a/$our-prefix" and such? Then this is
327 * very likely to apply to our directory.
328 */
329 if (!strncmp(name, prefix, prefix_length))
330 val = count_slashes(prefix);
331 else {
332 cp++;
333 if (!strncmp(cp, prefix, prefix_length))
334 val = count_slashes(prefix) + 1;
335 }
336 }
337 free(name);
338 return val;
339 }
340
341 /*
342 * Get the name etc info from the --/+++ lines of a traditional patch header
343 *
344 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
345 * files, we can happily check the index for a match, but for creating a
346 * new file we should try to match whatever "patch" does. I have no idea.
347 */
348 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
349 {
350 char *name;
351
352 first += 4; /* skip "--- " */
353 second += 4; /* skip "+++ " */
354 if (!p_value_known) {
355 int p, q;
356 p = guess_p_value(first);
357 q = guess_p_value(second);
358 if (p < 0) p = q;
359 if (0 <= p && p == q) {
360 p_value = p;
361 p_value_known = 1;
362 }
363 }
364 if (is_dev_null(first)) {
365 patch->is_new = 1;
366 patch->is_delete = 0;
367 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
368 patch->new_name = name;
369 } else if (is_dev_null(second)) {
370 patch->is_new = 0;
371 patch->is_delete = 1;
372 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
373 patch->old_name = name;
374 } else {
375 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
376 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
377 patch->old_name = patch->new_name = name;
378 }
379 if (!name)
380 die("unable to find filename in patch at line %d", linenr);
381 }
382
383 static int gitdiff_hdrend(const char *line, struct patch *patch)
384 {
385 return -1;
386 }
387
388 /*
389 * We're anal about diff header consistency, to make
390 * sure that we don't end up having strange ambiguous
391 * patches floating around.
392 *
393 * As a result, gitdiff_{old|new}name() will check
394 * their names against any previous information, just
395 * to make sure..
396 */
397 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
398 {
399 if (!orig_name && !isnull)
400 return find_name(line, NULL, p_value, TERM_TAB);
401
402 if (orig_name) {
403 int len;
404 const char *name;
405 char *another;
406 name = orig_name;
407 len = strlen(name);
408 if (isnull)
409 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
410 another = find_name(line, NULL, p_value, TERM_TAB);
411 if (!another || memcmp(another, name, len))
412 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
413 free(another);
414 return orig_name;
415 }
416 else {
417 /* expect "/dev/null" */
418 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
419 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
420 return NULL;
421 }
422 }
423
424 static int gitdiff_oldname(const char *line, struct patch *patch)
425 {
426 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
427 return 0;
428 }
429
430 static int gitdiff_newname(const char *line, struct patch *patch)
431 {
432 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
433 return 0;
434 }
435
436 static int gitdiff_oldmode(const char *line, struct patch *patch)
437 {
438 patch->old_mode = strtoul(line, NULL, 8);
439 return 0;
440 }
441
442 static int gitdiff_newmode(const char *line, struct patch *patch)
443 {
444 patch->new_mode = strtoul(line, NULL, 8);
445 return 0;
446 }
447
448 static int gitdiff_delete(const char *line, struct patch *patch)
449 {
450 patch->is_delete = 1;
451 patch->old_name = patch->def_name;
452 return gitdiff_oldmode(line, patch);
453 }
454
455 static int gitdiff_newfile(const char *line, struct patch *patch)
456 {
457 patch->is_new = 1;
458 patch->new_name = patch->def_name;
459 return gitdiff_newmode(line, patch);
460 }
461
462 static int gitdiff_copysrc(const char *line, struct patch *patch)
463 {
464 patch->is_copy = 1;
465 patch->old_name = find_name(line, NULL, 0, 0);
466 return 0;
467 }
468
469 static int gitdiff_copydst(const char *line, struct patch *patch)
470 {
471 patch->is_copy = 1;
472 patch->new_name = find_name(line, NULL, 0, 0);
473 return 0;
474 }
475
476 static int gitdiff_renamesrc(const char *line, struct patch *patch)
477 {
478 patch->is_rename = 1;
479 patch->old_name = find_name(line, NULL, 0, 0);
480 return 0;
481 }
482
483 static int gitdiff_renamedst(const char *line, struct patch *patch)
484 {
485 patch->is_rename = 1;
486 patch->new_name = find_name(line, NULL, 0, 0);
487 return 0;
488 }
489
490 static int gitdiff_similarity(const char *line, struct patch *patch)
491 {
492 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
493 patch->score = 0;
494 return 0;
495 }
496
497 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
498 {
499 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
500 patch->score = 0;
501 return 0;
502 }
503
504 static int gitdiff_index(const char *line, struct patch *patch)
505 {
506 /* index line is N hexadecimal, "..", N hexadecimal,
507 * and optional space with octal mode.
508 */
509 const char *ptr, *eol;
510 int len;
511
512 ptr = strchr(line, '.');
513 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
514 return 0;
515 len = ptr - line;
516 memcpy(patch->old_sha1_prefix, line, len);
517 patch->old_sha1_prefix[len] = 0;
518
519 line = ptr + 2;
520 ptr = strchr(line, ' ');
521 eol = strchr(line, '\n');
522
523 if (!ptr || eol < ptr)
524 ptr = eol;
525 len = ptr - line;
526
527 if (40 < len)
528 return 0;
529 memcpy(patch->new_sha1_prefix, line, len);
530 patch->new_sha1_prefix[len] = 0;
531 if (*ptr == ' ')
532 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
533 return 0;
534 }
535
536 /*
537 * This is normal for a diff that doesn't change anything: we'll fall through
538 * into the next diff. Tell the parser to break out.
539 */
540 static int gitdiff_unrecognized(const char *line, struct patch *patch)
541 {
542 return -1;
543 }
544
545 static const char *stop_at_slash(const char *line, int llen)
546 {
547 int i;
548
549 for (i = 0; i < llen; i++) {
550 int ch = line[i];
551 if (ch == '/')
552 return line + i;
553 }
554 return NULL;
555 }
556
557 /* This is to extract the same name that appears on "diff --git"
558 * line. We do not find and return anything if it is a rename
559 * patch, and it is OK because we will find the name elsewhere.
560 * We need to reliably find name only when it is mode-change only,
561 * creation or deletion of an empty file. In any of these cases,
562 * both sides are the same name under a/ and b/ respectively.
563 */
564 static char *git_header_name(char *line, int llen)
565 {
566 const char *name;
567 const char *second = NULL;
568 size_t len;
569
570 line += strlen("diff --git ");
571 llen -= strlen("diff --git ");
572
573 if (*line == '"') {
574 const char *cp;
575 struct strbuf first;
576 struct strbuf sp;
577
578 strbuf_init(&first, 0);
579 strbuf_init(&sp, 0);
580
581 if (unquote_c_style(&first, line, &second))
582 goto free_and_fail1;
583
584 /* advance to the first slash */
585 cp = stop_at_slash(first.buf, first.len);
586 /* we do not accept absolute paths */
587 if (!cp || cp == first.buf)
588 goto free_and_fail1;
589 strbuf_remove(&first, 0, cp + 1 - first.buf);
590
591 /* second points at one past closing dq of name.
592 * find the second name.
593 */
594 while ((second < line + llen) && isspace(*second))
595 second++;
596
597 if (line + llen <= second)
598 goto free_and_fail1;
599 if (*second == '"') {
600 if (unquote_c_style(&sp, second, NULL))
601 goto free_and_fail1;
602 cp = stop_at_slash(sp.buf, sp.len);
603 if (!cp || cp == sp.buf)
604 goto free_and_fail1;
605 /* They must match, otherwise ignore */
606 if (strcmp(cp + 1, first.buf))
607 goto free_and_fail1;
608 strbuf_release(&sp);
609 return strbuf_detach(&first, NULL);
610 }
611
612 /* unquoted second */
613 cp = stop_at_slash(second, line + llen - second);
614 if (!cp || cp == second)
615 goto free_and_fail1;
616 cp++;
617 if (line + llen - cp != first.len + 1 ||
618 memcmp(first.buf, cp, first.len))
619 goto free_and_fail1;
620 return strbuf_detach(&first, NULL);
621
622 free_and_fail1:
623 strbuf_release(&first);
624 strbuf_release(&sp);
625 return NULL;
626 }
627
628 /* unquoted first name */
629 name = stop_at_slash(line, llen);
630 if (!name || name == line)
631 return NULL;
632 name++;
633
634 /* since the first name is unquoted, a dq if exists must be
635 * the beginning of the second name.
636 */
637 for (second = name; second < line + llen; second++) {
638 if (*second == '"') {
639 struct strbuf sp;
640 const char *np;
641
642 strbuf_init(&sp, 0);
643 if (unquote_c_style(&sp, second, NULL))
644 goto free_and_fail2;
645
646 np = stop_at_slash(sp.buf, sp.len);
647 if (!np || np == sp.buf)
648 goto free_and_fail2;
649 np++;
650
651 len = sp.buf + sp.len - np;
652 if (len < second - name &&
653 !strncmp(np, name, len) &&
654 isspace(name[len])) {
655 /* Good */
656 strbuf_remove(&sp, 0, np - sp.buf);
657 return strbuf_detach(&sp, NULL);
658 }
659
660 free_and_fail2:
661 strbuf_release(&sp);
662 return NULL;
663 }
664 }
665
666 /*
667 * Accept a name only if it shows up twice, exactly the same
668 * form.
669 */
670 for (len = 0 ; ; len++) {
671 switch (name[len]) {
672 default:
673 continue;
674 case '\n':
675 return NULL;
676 case '\t': case ' ':
677 second = name+len;
678 for (;;) {
679 char c = *second++;
680 if (c == '\n')
681 return NULL;
682 if (c == '/')
683 break;
684 }
685 if (second[len] == '\n' && !memcmp(name, second, len)) {
686 return xmemdupz(name, len);
687 }
688 }
689 }
690 return NULL;
691 }
692
693 /* Verify that we recognize the lines following a git header */
694 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
695 {
696 unsigned long offset;
697
698 /* A git diff has explicit new/delete information, so we don't guess */
699 patch->is_new = 0;
700 patch->is_delete = 0;
701
702 /*
703 * Some things may not have the old name in the
704 * rest of the headers anywhere (pure mode changes,
705 * or removing or adding empty files), so we get
706 * the default name from the header.
707 */
708 patch->def_name = git_header_name(line, len);
709
710 line += len;
711 size -= len;
712 linenr++;
713 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
714 static const struct opentry {
715 const char *str;
716 int (*fn)(const char *, struct patch *);
717 } optable[] = {
718 { "@@ -", gitdiff_hdrend },
719 { "--- ", gitdiff_oldname },
720 { "+++ ", gitdiff_newname },
721 { "old mode ", gitdiff_oldmode },
722 { "new mode ", gitdiff_newmode },
723 { "deleted file mode ", gitdiff_delete },
724 { "new file mode ", gitdiff_newfile },
725 { "copy from ", gitdiff_copysrc },
726 { "copy to ", gitdiff_copydst },
727 { "rename old ", gitdiff_renamesrc },
728 { "rename new ", gitdiff_renamedst },
729 { "rename from ", gitdiff_renamesrc },
730 { "rename to ", gitdiff_renamedst },
731 { "similarity index ", gitdiff_similarity },
732 { "dissimilarity index ", gitdiff_dissimilarity },
733 { "index ", gitdiff_index },
734 { "", gitdiff_unrecognized },
735 };
736 int i;
737
738 len = linelen(line, size);
739 if (!len || line[len-1] != '\n')
740 break;
741 for (i = 0; i < ARRAY_SIZE(optable); i++) {
742 const struct opentry *p = optable + i;
743 int oplen = strlen(p->str);
744 if (len < oplen || memcmp(p->str, line, oplen))
745 continue;
746 if (p->fn(line + oplen, patch) < 0)
747 return offset;
748 break;
749 }
750 }
751
752 return offset;
753 }
754
755 static int parse_num(const char *line, unsigned long *p)
756 {
757 char *ptr;
758
759 if (!isdigit(*line))
760 return 0;
761 *p = strtoul(line, &ptr, 10);
762 return ptr - line;
763 }
764
765 static int parse_range(const char *line, int len, int offset, const char *expect,
766 unsigned long *p1, unsigned long *p2)
767 {
768 int digits, ex;
769
770 if (offset < 0 || offset >= len)
771 return -1;
772 line += offset;
773 len -= offset;
774
775 digits = parse_num(line, p1);
776 if (!digits)
777 return -1;
778
779 offset += digits;
780 line += digits;
781 len -= digits;
782
783 *p2 = 1;
784 if (*line == ',') {
785 digits = parse_num(line+1, p2);
786 if (!digits)
787 return -1;
788
789 offset += digits+1;
790 line += digits+1;
791 len -= digits+1;
792 }
793
794 ex = strlen(expect);
795 if (ex > len)
796 return -1;
797 if (memcmp(line, expect, ex))
798 return -1;
799
800 return offset + ex;
801 }
802
803 /*
804 * Parse a unified diff fragment header of the
805 * form "@@ -a,b +c,d @@"
806 */
807 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
808 {
809 int offset;
810
811 if (!len || line[len-1] != '\n')
812 return -1;
813
814 /* Figure out the number of lines in a fragment */
815 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
816 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
817
818 return offset;
819 }
820
821 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
822 {
823 unsigned long offset, len;
824
825 patch->is_toplevel_relative = 0;
826 patch->is_rename = patch->is_copy = 0;
827 patch->is_new = patch->is_delete = -1;
828 patch->old_mode = patch->new_mode = 0;
829 patch->old_name = patch->new_name = NULL;
830 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
831 unsigned long nextlen;
832
833 len = linelen(line, size);
834 if (!len)
835 break;
836
837 /* Testing this early allows us to take a few shortcuts.. */
838 if (len < 6)
839 continue;
840
841 /*
842 * Make sure we don't find any unconnected patch fragments.
843 * That's a sign that we didn't find a header, and that a
844 * patch has become corrupted/broken up.
845 */
846 if (!memcmp("@@ -", line, 4)) {
847 struct fragment dummy;
848 if (parse_fragment_header(line, len, &dummy) < 0)
849 continue;
850 die("patch fragment without header at line %d: %.*s",
851 linenr, (int)len-1, line);
852 }
853
854 if (size < len + 6)
855 break;
856
857 /*
858 * Git patch? It might not have a real patch, just a rename
859 * or mode change, so we handle that specially
860 */
861 if (!memcmp("diff --git ", line, 11)) {
862 int git_hdr_len = parse_git_header(line, len, size, patch);
863 if (git_hdr_len <= len)
864 continue;
865 if (!patch->old_name && !patch->new_name) {
866 if (!patch->def_name)
867 die("git diff header lacks filename information (line %d)", linenr);
868 patch->old_name = patch->new_name = patch->def_name;
869 }
870 patch->is_toplevel_relative = 1;
871 *hdrsize = git_hdr_len;
872 return offset;
873 }
874
875 /** --- followed by +++ ? */
876 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
877 continue;
878
879 /*
880 * We only accept unified patches, so we want it to
881 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
882 * minimum
883 */
884 nextlen = linelen(line + len, size - len);
885 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
886 continue;
887
888 /* Ok, we'll consider it a patch */
889 parse_traditional_patch(line, line+len, patch);
890 *hdrsize = len + nextlen;
891 linenr += 2;
892 return offset;
893 }
894 return -1;
895 }
896
897 static void check_whitespace(const char *line, int len)
898 {
899 const char *err = "Adds trailing whitespace";
900 int seen_space = 0;
901 int i;
902
903 /*
904 * We know len is at least two, since we have a '+' and we
905 * checked that the last character was a '\n' before calling
906 * this function. That is, an addition of an empty line would
907 * check the '+' here. Sneaky...
908 */
909 if (isspace(line[len-2]))
910 goto error;
911
912 /*
913 * Make sure that there is no space followed by a tab in
914 * indentation.
915 */
916 err = "Space in indent is followed by a tab";
917 for (i = 1; i < len; i++) {
918 if (line[i] == '\t') {
919 if (seen_space)
920 goto error;
921 }
922 else if (line[i] == ' ')
923 seen_space = 1;
924 else
925 break;
926 }
927 return;
928
929 error:
930 whitespace_error++;
931 if (squelch_whitespace_errors &&
932 squelch_whitespace_errors < whitespace_error)
933 ;
934 else
935 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
936 err, patch_input_file, linenr, len-2, line+1);
937 }
938
939
940 /*
941 * Parse a unified diff. Note that this really needs to parse each
942 * fragment separately, since the only way to know the difference
943 * between a "---" that is part of a patch, and a "---" that starts
944 * the next patch is to look at the line counts..
945 */
946 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
947 {
948 int added, deleted;
949 int len = linelen(line, size), offset;
950 unsigned long oldlines, newlines;
951 unsigned long leading, trailing;
952
953 offset = parse_fragment_header(line, len, fragment);
954 if (offset < 0)
955 return -1;
956 oldlines = fragment->oldlines;
957 newlines = fragment->newlines;
958 leading = 0;
959 trailing = 0;
960
961 /* Parse the thing.. */
962 line += len;
963 size -= len;
964 linenr++;
965 added = deleted = 0;
966 for (offset = len;
967 0 < size;
968 offset += len, size -= len, line += len, linenr++) {
969 if (!oldlines && !newlines)
970 break;
971 len = linelen(line, size);
972 if (!len || line[len-1] != '\n')
973 return -1;
974 switch (*line) {
975 default:
976 return -1;
977 case '\n': /* newer GNU diff, an empty context line */
978 case ' ':
979 oldlines--;
980 newlines--;
981 if (!deleted && !added)
982 leading++;
983 trailing++;
984 break;
985 case '-':
986 if (apply_in_reverse &&
987 new_whitespace != nowarn_whitespace)
988 check_whitespace(line, len);
989 deleted++;
990 oldlines--;
991 trailing = 0;
992 break;
993 case '+':
994 if (!apply_in_reverse &&
995 new_whitespace != nowarn_whitespace)
996 check_whitespace(line, len);
997 added++;
998 newlines--;
999 trailing = 0;
1000 break;
1001
1002 /* We allow "\ No newline at end of file". Depending
1003 * on locale settings when the patch was produced we
1004 * don't know what this line looks like. The only
1005 * thing we do know is that it begins with "\ ".
1006 * Checking for 12 is just for sanity check -- any
1007 * l10n of "\ No newline..." is at least that long.
1008 */
1009 case '\\':
1010 if (len < 12 || memcmp(line, "\\ ", 2))
1011 return -1;
1012 break;
1013 }
1014 }
1015 if (oldlines || newlines)
1016 return -1;
1017 fragment->leading = leading;
1018 fragment->trailing = trailing;
1019
1020 /* If a fragment ends with an incomplete line, we failed to include
1021 * it in the above loop because we hit oldlines == newlines == 0
1022 * before seeing it.
1023 */
1024 if (12 < size && !memcmp(line, "\\ ", 2))
1025 offset += linelen(line, size);
1026
1027 patch->lines_added += added;
1028 patch->lines_deleted += deleted;
1029
1030 if (0 < patch->is_new && oldlines)
1031 return error("new file depends on old contents");
1032 if (0 < patch->is_delete && newlines)
1033 return error("deleted file still has contents");
1034 return offset;
1035 }
1036
1037 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1038 {
1039 unsigned long offset = 0;
1040 unsigned long oldlines = 0, newlines = 0, context = 0;
1041 struct fragment **fragp = &patch->fragments;
1042
1043 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1044 struct fragment *fragment;
1045 int len;
1046
1047 fragment = xcalloc(1, sizeof(*fragment));
1048 len = parse_fragment(line, size, patch, fragment);
1049 if (len <= 0)
1050 die("corrupt patch at line %d", linenr);
1051 fragment->patch = line;
1052 fragment->size = len;
1053 oldlines += fragment->oldlines;
1054 newlines += fragment->newlines;
1055 context += fragment->leading + fragment->trailing;
1056
1057 *fragp = fragment;
1058 fragp = &fragment->next;
1059
1060 offset += len;
1061 line += len;
1062 size -= len;
1063 }
1064
1065 /*
1066 * If something was removed (i.e. we have old-lines) it cannot
1067 * be creation, and if something was added it cannot be
1068 * deletion. However, the reverse is not true; --unified=0
1069 * patches that only add are not necessarily creation even
1070 * though they do not have any old lines, and ones that only
1071 * delete are not necessarily deletion.
1072 *
1073 * Unfortunately, a real creation/deletion patch do _not_ have
1074 * any context line by definition, so we cannot safely tell it
1075 * apart with --unified=0 insanity. At least if the patch has
1076 * more than one hunk it is not creation or deletion.
1077 */
1078 if (patch->is_new < 0 &&
1079 (oldlines || (patch->fragments && patch->fragments->next)))
1080 patch->is_new = 0;
1081 if (patch->is_delete < 0 &&
1082 (newlines || (patch->fragments && patch->fragments->next)))
1083 patch->is_delete = 0;
1084 if (!unidiff_zero || context) {
1085 /* If the user says the patch is not generated with
1086 * --unified=0, or if we have seen context lines,
1087 * then not having oldlines means the patch is creation,
1088 * and not having newlines means the patch is deletion.
1089 */
1090 if (patch->is_new < 0 && !oldlines) {
1091 patch->is_new = 1;
1092 patch->old_name = NULL;
1093 }
1094 if (patch->is_delete < 0 && !newlines) {
1095 patch->is_delete = 1;
1096 patch->new_name = NULL;
1097 }
1098 }
1099
1100 if (0 < patch->is_new && oldlines)
1101 die("new file %s depends on old contents", patch->new_name);
1102 if (0 < patch->is_delete && newlines)
1103 die("deleted file %s still has contents", patch->old_name);
1104 if (!patch->is_delete && !newlines && context)
1105 fprintf(stderr, "** warning: file %s becomes empty but "
1106 "is not deleted\n", patch->new_name);
1107
1108 return offset;
1109 }
1110
1111 static inline int metadata_changes(struct patch *patch)
1112 {
1113 return patch->is_rename > 0 ||
1114 patch->is_copy > 0 ||
1115 patch->is_new > 0 ||
1116 patch->is_delete ||
1117 (patch->old_mode && patch->new_mode &&
1118 patch->old_mode != patch->new_mode);
1119 }
1120
1121 static char *inflate_it(const void *data, unsigned long size,
1122 unsigned long inflated_size)
1123 {
1124 z_stream stream;
1125 void *out;
1126 int st;
1127
1128 memset(&stream, 0, sizeof(stream));
1129
1130 stream.next_in = (unsigned char *)data;
1131 stream.avail_in = size;
1132 stream.next_out = out = xmalloc(inflated_size);
1133 stream.avail_out = inflated_size;
1134 inflateInit(&stream);
1135 st = inflate(&stream, Z_FINISH);
1136 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1137 free(out);
1138 return NULL;
1139 }
1140 return out;
1141 }
1142
1143 static struct fragment *parse_binary_hunk(char **buf_p,
1144 unsigned long *sz_p,
1145 int *status_p,
1146 int *used_p)
1147 {
1148 /* Expect a line that begins with binary patch method ("literal"
1149 * or "delta"), followed by the length of data before deflating.
1150 * a sequence of 'length-byte' followed by base-85 encoded data
1151 * should follow, terminated by a newline.
1152 *
1153 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1154 * and we would limit the patch line to 66 characters,
1155 * so one line can fit up to 13 groups that would decode
1156 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1157 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1158 */
1159 int llen, used;
1160 unsigned long size = *sz_p;
1161 char *buffer = *buf_p;
1162 int patch_method;
1163 unsigned long origlen;
1164 char *data = NULL;
1165 int hunk_size = 0;
1166 struct fragment *frag;
1167
1168 llen = linelen(buffer, size);
1169 used = llen;
1170
1171 *status_p = 0;
1172
1173 if (!prefixcmp(buffer, "delta ")) {
1174 patch_method = BINARY_DELTA_DEFLATED;
1175 origlen = strtoul(buffer + 6, NULL, 10);
1176 }
1177 else if (!prefixcmp(buffer, "literal ")) {
1178 patch_method = BINARY_LITERAL_DEFLATED;
1179 origlen = strtoul(buffer + 8, NULL, 10);
1180 }
1181 else
1182 return NULL;
1183
1184 linenr++;
1185 buffer += llen;
1186 while (1) {
1187 int byte_length, max_byte_length, newsize;
1188 llen = linelen(buffer, size);
1189 used += llen;
1190 linenr++;
1191 if (llen == 1) {
1192 /* consume the blank line */
1193 buffer++;
1194 size--;
1195 break;
1196 }
1197 /* Minimum line is "A00000\n" which is 7-byte long,
1198 * and the line length must be multiple of 5 plus 2.
1199 */
1200 if ((llen < 7) || (llen-2) % 5)
1201 goto corrupt;
1202 max_byte_length = (llen - 2) / 5 * 4;
1203 byte_length = *buffer;
1204 if ('A' <= byte_length && byte_length <= 'Z')
1205 byte_length = byte_length - 'A' + 1;
1206 else if ('a' <= byte_length && byte_length <= 'z')
1207 byte_length = byte_length - 'a' + 27;
1208 else
1209 goto corrupt;
1210 /* if the input length was not multiple of 4, we would
1211 * have filler at the end but the filler should never
1212 * exceed 3 bytes
1213 */
1214 if (max_byte_length < byte_length ||
1215 byte_length <= max_byte_length - 4)
1216 goto corrupt;
1217 newsize = hunk_size + byte_length;
1218 data = xrealloc(data, newsize);
1219 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1220 goto corrupt;
1221 hunk_size = newsize;
1222 buffer += llen;
1223 size -= llen;
1224 }
1225
1226 frag = xcalloc(1, sizeof(*frag));
1227 frag->patch = inflate_it(data, hunk_size, origlen);
1228 if (!frag->patch)
1229 goto corrupt;
1230 free(data);
1231 frag->size = origlen;
1232 *buf_p = buffer;
1233 *sz_p = size;
1234 *used_p = used;
1235 frag->binary_patch_method = patch_method;
1236 return frag;
1237
1238 corrupt:
1239 free(data);
1240 *status_p = -1;
1241 error("corrupt binary patch at line %d: %.*s",
1242 linenr-1, llen-1, buffer);
1243 return NULL;
1244 }
1245
1246 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1247 {
1248 /* We have read "GIT binary patch\n"; what follows is a line
1249 * that says the patch method (currently, either "literal" or
1250 * "delta") and the length of data before deflating; a
1251 * sequence of 'length-byte' followed by base-85 encoded data
1252 * follows.
1253 *
1254 * When a binary patch is reversible, there is another binary
1255 * hunk in the same format, starting with patch method (either
1256 * "literal" or "delta") with the length of data, and a sequence
1257 * of length-byte + base-85 encoded data, terminated with another
1258 * empty line. This data, when applied to the postimage, produces
1259 * the preimage.
1260 */
1261 struct fragment *forward;
1262 struct fragment *reverse;
1263 int status;
1264 int used, used_1;
1265
1266 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1267 if (!forward && !status)
1268 /* there has to be one hunk (forward hunk) */
1269 return error("unrecognized binary patch at line %d", linenr-1);
1270 if (status)
1271 /* otherwise we already gave an error message */
1272 return status;
1273
1274 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1275 if (reverse)
1276 used += used_1;
1277 else if (status) {
1278 /* not having reverse hunk is not an error, but having
1279 * a corrupt reverse hunk is.
1280 */
1281 free((void*) forward->patch);
1282 free(forward);
1283 return status;
1284 }
1285 forward->next = reverse;
1286 patch->fragments = forward;
1287 patch->is_binary = 1;
1288 return used;
1289 }
1290
1291 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1292 {
1293 int hdrsize, patchsize;
1294 int offset = find_header(buffer, size, &hdrsize, patch);
1295
1296 if (offset < 0)
1297 return offset;
1298
1299 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1300
1301 if (!patchsize) {
1302 static const char *binhdr[] = {
1303 "Binary files ",
1304 "Files ",
1305 NULL,
1306 };
1307 static const char git_binary[] = "GIT binary patch\n";
1308 int i;
1309 int hd = hdrsize + offset;
1310 unsigned long llen = linelen(buffer + hd, size - hd);
1311
1312 if (llen == sizeof(git_binary) - 1 &&
1313 !memcmp(git_binary, buffer + hd, llen)) {
1314 int used;
1315 linenr++;
1316 used = parse_binary(buffer + hd + llen,
1317 size - hd - llen, patch);
1318 if (used)
1319 patchsize = used + llen;
1320 else
1321 patchsize = 0;
1322 }
1323 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1324 for (i = 0; binhdr[i]; i++) {
1325 int len = strlen(binhdr[i]);
1326 if (len < size - hd &&
1327 !memcmp(binhdr[i], buffer + hd, len)) {
1328 linenr++;
1329 patch->is_binary = 1;
1330 patchsize = llen;
1331 break;
1332 }
1333 }
1334 }
1335
1336 /* Empty patch cannot be applied if it is a text patch
1337 * without metadata change. A binary patch appears
1338 * empty to us here.
1339 */
1340 if ((apply || check) &&
1341 (!patch->is_binary && !metadata_changes(patch)))
1342 die("patch with only garbage at line %d", linenr);
1343 }
1344
1345 return offset + hdrsize + patchsize;
1346 }
1347
1348 #define swap(a,b) myswap((a),(b),sizeof(a))
1349
1350 #define myswap(a, b, size) do { \
1351 unsigned char mytmp[size]; \
1352 memcpy(mytmp, &a, size); \
1353 memcpy(&a, &b, size); \
1354 memcpy(&b, mytmp, size); \
1355 } while (0)
1356
1357 static void reverse_patches(struct patch *p)
1358 {
1359 for (; p; p = p->next) {
1360 struct fragment *frag = p->fragments;
1361
1362 swap(p->new_name, p->old_name);
1363 swap(p->new_mode, p->old_mode);
1364 swap(p->is_new, p->is_delete);
1365 swap(p->lines_added, p->lines_deleted);
1366 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1367
1368 for (; frag; frag = frag->next) {
1369 swap(frag->newpos, frag->oldpos);
1370 swap(frag->newlines, frag->oldlines);
1371 }
1372 }
1373 }
1374
1375 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1376 static const char minuses[]= "----------------------------------------------------------------------";
1377
1378 static void show_stats(struct patch *patch)
1379 {
1380 struct strbuf qname;
1381 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1382 int max, add, del;
1383
1384 strbuf_init(&qname, 0);
1385 quote_c_style(cp, &qname, NULL, 0);
1386
1387 /*
1388 * "scale" the filename
1389 */
1390 max = max_len;
1391 if (max > 50)
1392 max = 50;
1393
1394 if (qname.len > max) {
1395 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1396 if (!cp)
1397 cp = qname.buf + qname.len + 3 - max;
1398 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1399 }
1400
1401 if (patch->is_binary) {
1402 printf(" %-*s | Bin\n", max, qname.buf);
1403 strbuf_release(&qname);
1404 return;
1405 }
1406
1407 printf(" %-*s |", max, qname.buf);
1408 strbuf_release(&qname);
1409
1410 /*
1411 * scale the add/delete
1412 */
1413 max = max + max_change > 70 ? 70 - max : max_change;
1414 add = patch->lines_added;
1415 del = patch->lines_deleted;
1416
1417 if (max_change > 0) {
1418 int total = ((add + del) * max + max_change / 2) / max_change;
1419 add = (add * max + max_change / 2) / max_change;
1420 del = total - add;
1421 }
1422 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1423 add, pluses, del, minuses);
1424 }
1425
1426 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1427 {
1428 switch (st->st_mode & S_IFMT) {
1429 case S_IFLNK:
1430 strbuf_grow(buf, st->st_size);
1431 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1432 return -1;
1433 strbuf_setlen(buf, st->st_size);
1434 return 0;
1435 case S_IFREG:
1436 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1437 return error("unable to open or read %s", path);
1438 convert_to_git(path, buf->buf, buf->len, buf);
1439 return 0;
1440 default:
1441 return -1;
1442 }
1443 }
1444
1445 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1446 {
1447 int i;
1448 unsigned long start, backwards, forwards;
1449
1450 if (fragsize > size)
1451 return -1;
1452
1453 start = 0;
1454 if (line > 1) {
1455 unsigned long offset = 0;
1456 i = line-1;
1457 while (offset + fragsize <= size) {
1458 if (buf[offset++] == '\n') {
1459 start = offset;
1460 if (!--i)
1461 break;
1462 }
1463 }
1464 }
1465
1466 /* Exact line number? */
1467 if ((start + fragsize <= size) &&
1468 !memcmp(buf + start, fragment, fragsize))
1469 return start;
1470
1471 /*
1472 * There's probably some smart way to do this, but I'll leave
1473 * that to the smart and beautiful people. I'm simple and stupid.
1474 */
1475 backwards = start;
1476 forwards = start;
1477 for (i = 0; ; i++) {
1478 unsigned long try;
1479 int n;
1480
1481 /* "backward" */
1482 if (i & 1) {
1483 if (!backwards) {
1484 if (forwards + fragsize > size)
1485 break;
1486 continue;
1487 }
1488 do {
1489 --backwards;
1490 } while (backwards && buf[backwards-1] != '\n');
1491 try = backwards;
1492 } else {
1493 while (forwards + fragsize <= size) {
1494 if (buf[forwards++] == '\n')
1495 break;
1496 }
1497 try = forwards;
1498 }
1499
1500 if (try + fragsize > size)
1501 continue;
1502 if (memcmp(buf + try, fragment, fragsize))
1503 continue;
1504 n = (i >> 1)+1;
1505 if (i & 1)
1506 n = -n;
1507 *lines = n;
1508 return try;
1509 }
1510
1511 /*
1512 * We should start searching forward and backward.
1513 */
1514 return -1;
1515 }
1516
1517 static void remove_first_line(const char **rbuf, int *rsize)
1518 {
1519 const char *buf = *rbuf;
1520 int size = *rsize;
1521 unsigned long offset;
1522 offset = 0;
1523 while (offset <= size) {
1524 if (buf[offset++] == '\n')
1525 break;
1526 }
1527 *rsize = size - offset;
1528 *rbuf = buf + offset;
1529 }
1530
1531 static void remove_last_line(const char **rbuf, int *rsize)
1532 {
1533 const char *buf = *rbuf;
1534 int size = *rsize;
1535 unsigned long offset;
1536 offset = size - 1;
1537 while (offset > 0) {
1538 if (buf[--offset] == '\n')
1539 break;
1540 }
1541 *rsize = offset + 1;
1542 }
1543
1544 static int apply_line(char *output, const char *patch, int plen)
1545 {
1546 /* plen is number of bytes to be copied from patch,
1547 * starting at patch+1 (patch[0] is '+'). Typically
1548 * patch[plen] is '\n', unless this is the incomplete
1549 * last line.
1550 */
1551 int i;
1552 int add_nl_to_tail = 0;
1553 int fixed = 0;
1554 int last_tab_in_indent = -1;
1555 int last_space_in_indent = -1;
1556 int need_fix_leading_space = 0;
1557 char *buf;
1558
1559 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1560 *patch != '+') {
1561 memcpy(output, patch + 1, plen);
1562 return plen;
1563 }
1564
1565 if (1 < plen && isspace(patch[plen-1])) {
1566 if (patch[plen] == '\n')
1567 add_nl_to_tail = 1;
1568 plen--;
1569 while (0 < plen && isspace(patch[plen]))
1570 plen--;
1571 fixed = 1;
1572 }
1573
1574 for (i = 1; i < plen; i++) {
1575 char ch = patch[i];
1576 if (ch == '\t') {
1577 last_tab_in_indent = i;
1578 if (0 <= last_space_in_indent)
1579 need_fix_leading_space = 1;
1580 }
1581 else if (ch == ' ')
1582 last_space_in_indent = i;
1583 else
1584 break;
1585 }
1586
1587 buf = output;
1588 if (need_fix_leading_space) {
1589 int consecutive_spaces = 0;
1590 /* between patch[1..last_tab_in_indent] strip the
1591 * funny spaces, updating them to tab as needed.
1592 */
1593 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1594 char ch = patch[i];
1595 if (ch != ' ') {
1596 consecutive_spaces = 0;
1597 *output++ = ch;
1598 } else {
1599 consecutive_spaces++;
1600 if (consecutive_spaces == 8) {
1601 *output++ = '\t';
1602 consecutive_spaces = 0;
1603 }
1604 }
1605 }
1606 fixed = 1;
1607 i = last_tab_in_indent;
1608 }
1609 else
1610 i = 1;
1611
1612 memcpy(output, patch + i, plen);
1613 if (add_nl_to_tail)
1614 output[plen++] = '\n';
1615 if (fixed)
1616 applied_after_fixing_ws++;
1617 return output + plen - buf;
1618 }
1619
1620 static int apply_one_fragment(struct strbuf *buf, struct fragment *frag, int inaccurate_eof)
1621 {
1622 int match_beginning, match_end;
1623 const char *patch = frag->patch;
1624 int offset, size = frag->size;
1625 char *old = xmalloc(size);
1626 char *new = xmalloc(size);
1627 const char *oldlines, *newlines;
1628 int oldsize = 0, newsize = 0;
1629 int new_blank_lines_at_end = 0;
1630 unsigned long leading, trailing;
1631 int pos, lines;
1632
1633 while (size > 0) {
1634 char first;
1635 int len = linelen(patch, size);
1636 int plen;
1637 int added_blank_line = 0;
1638
1639 if (!len)
1640 break;
1641
1642 /*
1643 * "plen" is how much of the line we should use for
1644 * the actual patch data. Normally we just remove the
1645 * first character on the line, but if the line is
1646 * followed by "\ No newline", then we also remove the
1647 * last one (which is the newline, of course).
1648 */
1649 plen = len-1;
1650 if (len < size && patch[len] == '\\')
1651 plen--;
1652 first = *patch;
1653 if (apply_in_reverse) {
1654 if (first == '-')
1655 first = '+';
1656 else if (first == '+')
1657 first = '-';
1658 }
1659
1660 switch (first) {
1661 case '\n':
1662 /* Newer GNU diff, empty context line */
1663 if (plen < 0)
1664 /* ... followed by '\No newline'; nothing */
1665 break;
1666 old[oldsize++] = '\n';
1667 new[newsize++] = '\n';
1668 break;
1669 case ' ':
1670 case '-':
1671 memcpy(old + oldsize, patch + 1, plen);
1672 oldsize += plen;
1673 if (first == '-')
1674 break;
1675 /* Fall-through for ' ' */
1676 case '+':
1677 if (first != '+' || !no_add) {
1678 int added = apply_line(new + newsize, patch,
1679 plen);
1680 newsize += added;
1681 if (first == '+' &&
1682 added == 1 && new[newsize-1] == '\n')
1683 added_blank_line = 1;
1684 }
1685 break;
1686 case '@': case '\\':
1687 /* Ignore it, we already handled it */
1688 break;
1689 default:
1690 if (apply_verbosely)
1691 error("invalid start of line: '%c'", first);
1692 return -1;
1693 }
1694 if (added_blank_line)
1695 new_blank_lines_at_end++;
1696 else
1697 new_blank_lines_at_end = 0;
1698 patch += len;
1699 size -= len;
1700 }
1701
1702 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1703 newsize > 0 && new[newsize - 1] == '\n') {
1704 oldsize--;
1705 newsize--;
1706 }
1707
1708 oldlines = old;
1709 newlines = new;
1710 leading = frag->leading;
1711 trailing = frag->trailing;
1712
1713 /*
1714 * If we don't have any leading/trailing data in the patch,
1715 * we want it to match at the beginning/end of the file.
1716 *
1717 * But that would break if the patch is generated with
1718 * --unified=0; sane people wouldn't do that to cause us
1719 * trouble, but we try to please not so sane ones as well.
1720 */
1721 if (unidiff_zero) {
1722 match_beginning = (!leading && !frag->oldpos);
1723 match_end = 0;
1724 }
1725 else {
1726 match_beginning = !leading && (frag->oldpos == 1);
1727 match_end = !trailing;
1728 }
1729
1730 lines = 0;
1731 pos = frag->newpos;
1732 for (;;) {
1733 offset = find_offset(buf->buf, buf->len,
1734 oldlines, oldsize, pos, &lines);
1735 if (match_end && offset + oldsize != buf->len)
1736 offset = -1;
1737 if (match_beginning && offset)
1738 offset = -1;
1739 if (offset >= 0) {
1740 if (new_whitespace == strip_whitespace &&
1741 (buf->len - oldsize - offset == 0)) /* end of file? */
1742 newsize -= new_blank_lines_at_end;
1743
1744 /* Warn if it was necessary to reduce the number
1745 * of context lines.
1746 */
1747 if ((leading != frag->leading) ||
1748 (trailing != frag->trailing))
1749 fprintf(stderr, "Context reduced to (%ld/%ld)"
1750 " to apply fragment at %d\n",
1751 leading, trailing, pos + lines);
1752
1753 strbuf_splice(buf, offset, oldsize, newlines, newsize);
1754 offset = 0;
1755 break;
1756 }
1757
1758 /* Am I at my context limits? */
1759 if ((leading <= p_context) && (trailing <= p_context))
1760 break;
1761 if (match_beginning || match_end) {
1762 match_beginning = match_end = 0;
1763 continue;
1764 }
1765 /* Reduce the number of context lines
1766 * Reduce both leading and trailing if they are equal
1767 * otherwise just reduce the larger context.
1768 */
1769 if (leading >= trailing) {
1770 remove_first_line(&oldlines, &oldsize);
1771 remove_first_line(&newlines, &newsize);
1772 pos--;
1773 leading--;
1774 }
1775 if (trailing > leading) {
1776 remove_last_line(&oldlines, &oldsize);
1777 remove_last_line(&newlines, &newsize);
1778 trailing--;
1779 }
1780 }
1781
1782 if (offset && apply_verbosely)
1783 error("while searching for:\n%.*s", oldsize, oldlines);
1784
1785 free(old);
1786 free(new);
1787 return offset;
1788 }
1789
1790 static int apply_binary_fragment(struct strbuf *buf, struct patch *patch)
1791 {
1792 struct fragment *fragment = patch->fragments;
1793 unsigned long len;
1794 void *dst;
1795
1796 /* Binary patch is irreversible without the optional second hunk */
1797 if (apply_in_reverse) {
1798 if (!fragment->next)
1799 return error("cannot reverse-apply a binary patch "
1800 "without the reverse hunk to '%s'",
1801 patch->new_name
1802 ? patch->new_name : patch->old_name);
1803 fragment = fragment->next;
1804 }
1805 switch (fragment->binary_patch_method) {
1806 case BINARY_DELTA_DEFLATED:
1807 dst = patch_delta(buf->buf, buf->len, fragment->patch,
1808 fragment->size, &len);
1809 if (!dst)
1810 return -1;
1811 /* XXX patch_delta NUL-terminates */
1812 strbuf_attach(buf, dst, len, len + 1);
1813 return 0;
1814 case BINARY_LITERAL_DEFLATED:
1815 strbuf_reset(buf);
1816 strbuf_add(buf, fragment->patch, fragment->size);
1817 return 0;
1818 }
1819 return -1;
1820 }
1821
1822 static int apply_binary(struct strbuf *buf, struct patch *patch)
1823 {
1824 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1825 unsigned char sha1[20];
1826
1827 /* For safety, we require patch index line to contain
1828 * full 40-byte textual SHA1 for old and new, at least for now.
1829 */
1830 if (strlen(patch->old_sha1_prefix) != 40 ||
1831 strlen(patch->new_sha1_prefix) != 40 ||
1832 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1833 get_sha1_hex(patch->new_sha1_prefix, sha1))
1834 return error("cannot apply binary patch to '%s' "
1835 "without full index line", name);
1836
1837 if (patch->old_name) {
1838 /* See if the old one matches what the patch
1839 * applies to.
1840 */
1841 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1842 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1843 return error("the patch applies to '%s' (%s), "
1844 "which does not match the "
1845 "current contents.",
1846 name, sha1_to_hex(sha1));
1847 }
1848 else {
1849 /* Otherwise, the old one must be empty. */
1850 if (buf->len)
1851 return error("the patch applies to an empty "
1852 "'%s' but it is not empty", name);
1853 }
1854
1855 get_sha1_hex(patch->new_sha1_prefix, sha1);
1856 if (is_null_sha1(sha1)) {
1857 strbuf_release(buf);
1858 return 0; /* deletion patch */
1859 }
1860
1861 if (has_sha1_file(sha1)) {
1862 /* We already have the postimage */
1863 enum object_type type;
1864 unsigned long size;
1865 char *result;
1866
1867 result = read_sha1_file(sha1, &type, &size);
1868 if (!result)
1869 return error("the necessary postimage %s for "
1870 "'%s' cannot be read",
1871 patch->new_sha1_prefix, name);
1872 /* XXX read_sha1_file NUL-terminates */
1873 strbuf_attach(buf, result, size, size + 1);
1874 } else {
1875 /* We have verified buf matches the preimage;
1876 * apply the patch data to it, which is stored
1877 * in the patch->fragments->{patch,size}.
1878 */
1879 if (apply_binary_fragment(buf, patch))
1880 return error("binary patch does not apply to '%s'",
1881 name);
1882
1883 /* verify that the result matches */
1884 hash_sha1_file(buf->buf, buf->len, blob_type, sha1);
1885 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1886 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
1887 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1888 }
1889
1890 return 0;
1891 }
1892
1893 static int apply_fragments(struct strbuf *buf, struct patch *patch)
1894 {
1895 struct fragment *frag = patch->fragments;
1896 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1897
1898 if (patch->is_binary)
1899 return apply_binary(buf, patch);
1900
1901 while (frag) {
1902 if (apply_one_fragment(buf, frag, patch->inaccurate_eof)) {
1903 error("patch failed: %s:%ld", name, frag->oldpos);
1904 if (!apply_with_reject)
1905 return -1;
1906 frag->rejected = 1;
1907 }
1908 frag = frag->next;
1909 }
1910 return 0;
1911 }
1912
1913 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
1914 {
1915 if (!ce)
1916 return 0;
1917
1918 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1919 strbuf_grow(buf, 100);
1920 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
1921 } else {
1922 enum object_type type;
1923 unsigned long sz;
1924 char *result;
1925
1926 result = read_sha1_file(ce->sha1, &type, &sz);
1927 if (!result)
1928 return -1;
1929 /* XXX read_sha1_file NUL-terminates */
1930 strbuf_attach(buf, result, sz, sz + 1);
1931 }
1932 return 0;
1933 }
1934
1935 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1936 {
1937 struct strbuf buf;
1938
1939 strbuf_init(&buf, 0);
1940 if (cached) {
1941 if (read_file_or_gitlink(ce, &buf))
1942 return error("read of %s failed", patch->old_name);
1943 } else if (patch->old_name) {
1944 if (S_ISGITLINK(patch->old_mode)) {
1945 if (ce) {
1946 read_file_or_gitlink(ce, &buf);
1947 } else {
1948 /*
1949 * There is no way to apply subproject
1950 * patch without looking at the index.
1951 */
1952 patch->fragments = NULL;
1953 }
1954 } else {
1955 if (read_old_data(st, patch->old_name, &buf))
1956 return error("read of %s failed", patch->old_name);
1957 }
1958 }
1959
1960 if (apply_fragments(&buf, patch) < 0)
1961 return -1; /* note with --reject this succeeds. */
1962 patch->result = strbuf_detach(&buf, &patch->resultsize);
1963
1964 if (0 < patch->is_delete && patch->resultsize)
1965 return error("removal patch leaves file contents");
1966
1967 return 0;
1968 }
1969
1970 static int check_to_create_blob(const char *new_name, int ok_if_exists)
1971 {
1972 struct stat nst;
1973 if (!lstat(new_name, &nst)) {
1974 if (S_ISDIR(nst.st_mode) || ok_if_exists)
1975 return 0;
1976 /*
1977 * A leading component of new_name might be a symlink
1978 * that is going to be removed with this patch, but
1979 * still pointing at somewhere that has the path.
1980 * In such a case, path "new_name" does not exist as
1981 * far as git is concerned.
1982 */
1983 if (has_symlink_leading_path(new_name, NULL))
1984 return 0;
1985
1986 return error("%s: already exists in working directory", new_name);
1987 }
1988 else if ((errno != ENOENT) && (errno != ENOTDIR))
1989 return error("%s: %s", new_name, strerror(errno));
1990 return 0;
1991 }
1992
1993 static int verify_index_match(struct cache_entry *ce, struct stat *st)
1994 {
1995 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
1996 if (!S_ISDIR(st->st_mode))
1997 return -1;
1998 return 0;
1999 }
2000 return ce_match_stat(ce, st, 1);
2001 }
2002
2003 static int check_patch(struct patch *patch, struct patch *prev_patch)
2004 {
2005 struct stat st;
2006 const char *old_name = patch->old_name;
2007 const char *new_name = patch->new_name;
2008 const char *name = old_name ? old_name : new_name;
2009 struct cache_entry *ce = NULL;
2010 int ok_if_exists;
2011
2012 patch->rejected = 1; /* we will drop this after we succeed */
2013
2014 /*
2015 * Make sure that we do not have local modifications from the
2016 * index when we are looking at the index. Also make sure
2017 * we have the preimage file to be patched in the work tree,
2018 * unless --cached, which tells git to apply only in the index.
2019 */
2020 if (old_name) {
2021 int stat_ret = 0;
2022 unsigned st_mode = 0;
2023
2024 if (!cached)
2025 stat_ret = lstat(old_name, &st);
2026 if (check_index) {
2027 int pos = cache_name_pos(old_name, strlen(old_name));
2028 if (pos < 0)
2029 return error("%s: does not exist in index",
2030 old_name);
2031 ce = active_cache[pos];
2032 if (stat_ret < 0) {
2033 struct checkout costate;
2034 if (errno != ENOENT)
2035 return error("%s: %s", old_name,
2036 strerror(errno));
2037 /* checkout */
2038 costate.base_dir = "";
2039 costate.base_dir_len = 0;
2040 costate.force = 0;
2041 costate.quiet = 0;
2042 costate.not_new = 0;
2043 costate.refresh_cache = 1;
2044 if (checkout_entry(ce,
2045 &costate,
2046 NULL) ||
2047 lstat(old_name, &st))
2048 return -1;
2049 }
2050 if (!cached && verify_index_match(ce, &st))
2051 return error("%s: does not match index",
2052 old_name);
2053 if (cached)
2054 st_mode = ntohl(ce->ce_mode);
2055 } else if (stat_ret < 0)
2056 return error("%s: %s", old_name, strerror(errno));
2057
2058 if (!cached)
2059 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2060
2061 if (patch->is_new < 0)
2062 patch->is_new = 0;
2063 if (!patch->old_mode)
2064 patch->old_mode = st_mode;
2065 if ((st_mode ^ patch->old_mode) & S_IFMT)
2066 return error("%s: wrong type", old_name);
2067 if (st_mode != patch->old_mode)
2068 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2069 old_name, st_mode, patch->old_mode);
2070 }
2071
2072 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2073 !strcmp(prev_patch->old_name, new_name))
2074 /* A type-change diff is always split into a patch to
2075 * delete old, immediately followed by a patch to
2076 * create new (see diff.c::run_diff()); in such a case
2077 * it is Ok that the entry to be deleted by the
2078 * previous patch is still in the working tree and in
2079 * the index.
2080 */
2081 ok_if_exists = 1;
2082 else
2083 ok_if_exists = 0;
2084
2085 if (new_name &&
2086 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2087 if (check_index &&
2088 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2089 !ok_if_exists)
2090 return error("%s: already exists in index", new_name);
2091 if (!cached) {
2092 int err = check_to_create_blob(new_name, ok_if_exists);
2093 if (err)
2094 return err;
2095 }
2096 if (!patch->new_mode) {
2097 if (0 < patch->is_new)
2098 patch->new_mode = S_IFREG | 0644;
2099 else
2100 patch->new_mode = patch->old_mode;
2101 }
2102 }
2103
2104 if (new_name && old_name) {
2105 int same = !strcmp(old_name, new_name);
2106 if (!patch->new_mode)
2107 patch->new_mode = patch->old_mode;
2108 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2109 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2110 patch->new_mode, new_name, patch->old_mode,
2111 same ? "" : " of ", same ? "" : old_name);
2112 }
2113
2114 if (apply_data(patch, &st, ce) < 0)
2115 return error("%s: patch does not apply", name);
2116 patch->rejected = 0;
2117 return 0;
2118 }
2119
2120 static int check_patch_list(struct patch *patch)
2121 {
2122 struct patch *prev_patch = NULL;
2123 int err = 0;
2124
2125 for (prev_patch = NULL; patch ; patch = patch->next) {
2126 if (apply_verbosely)
2127 say_patch_name(stderr,
2128 "Checking patch ", patch, "...\n");
2129 err |= check_patch(patch, prev_patch);
2130 prev_patch = patch;
2131 }
2132 return err;
2133 }
2134
2135 /* This function tries to read the sha1 from the current index */
2136 static int get_current_sha1(const char *path, unsigned char *sha1)
2137 {
2138 int pos;
2139
2140 if (read_cache() < 0)
2141 return -1;
2142 pos = cache_name_pos(path, strlen(path));
2143 if (pos < 0)
2144 return -1;
2145 hashcpy(sha1, active_cache[pos]->sha1);
2146 return 0;
2147 }
2148
2149 static void show_index_list(struct patch *list)
2150 {
2151 struct patch *patch;
2152
2153 /* Once we start supporting the reverse patch, it may be
2154 * worth showing the new sha1 prefix, but until then...
2155 */
2156 for (patch = list; patch; patch = patch->next) {
2157 const unsigned char *sha1_ptr;
2158 unsigned char sha1[20];
2159 const char *name;
2160
2161 name = patch->old_name ? patch->old_name : patch->new_name;
2162 if (0 < patch->is_new)
2163 sha1_ptr = null_sha1;
2164 else if (get_sha1(patch->old_sha1_prefix, sha1))
2165 /* git diff has no index line for mode/type changes */
2166 if (!patch->lines_added && !patch->lines_deleted) {
2167 if (get_current_sha1(patch->new_name, sha1) ||
2168 get_current_sha1(patch->old_name, sha1))
2169 die("mode change for %s, which is not "
2170 "in current HEAD", name);
2171 sha1_ptr = sha1;
2172 } else
2173 die("sha1 information is lacking or useless "
2174 "(%s).", name);
2175 else
2176 sha1_ptr = sha1;
2177
2178 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2179 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2180 quote_c_style(name, NULL, stdout, 0);
2181 else
2182 fputs(name, stdout);
2183 putchar(line_termination);
2184 }
2185 }
2186
2187 static void stat_patch_list(struct patch *patch)
2188 {
2189 int files, adds, dels;
2190
2191 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2192 files++;
2193 adds += patch->lines_added;
2194 dels += patch->lines_deleted;
2195 show_stats(patch);
2196 }
2197
2198 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2199 }
2200
2201 static void numstat_patch_list(struct patch *patch)
2202 {
2203 for ( ; patch; patch = patch->next) {
2204 const char *name;
2205 name = patch->new_name ? patch->new_name : patch->old_name;
2206 if (patch->is_binary)
2207 printf("-\t-\t");
2208 else
2209 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2210 write_name_quoted(name, stdout, line_termination);
2211 }
2212 }
2213
2214 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2215 {
2216 if (mode)
2217 printf(" %s mode %06o %s\n", newdelete, mode, name);
2218 else
2219 printf(" %s %s\n", newdelete, name);
2220 }
2221
2222 static void show_mode_change(struct patch *p, int show_name)
2223 {
2224 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2225 if (show_name)
2226 printf(" mode change %06o => %06o %s\n",
2227 p->old_mode, p->new_mode, p->new_name);
2228 else
2229 printf(" mode change %06o => %06o\n",
2230 p->old_mode, p->new_mode);
2231 }
2232 }
2233
2234 static void show_rename_copy(struct patch *p)
2235 {
2236 const char *renamecopy = p->is_rename ? "rename" : "copy";
2237 const char *old, *new;
2238
2239 /* Find common prefix */
2240 old = p->old_name;
2241 new = p->new_name;
2242 while (1) {
2243 const char *slash_old, *slash_new;
2244 slash_old = strchr(old, '/');
2245 slash_new = strchr(new, '/');
2246 if (!slash_old ||
2247 !slash_new ||
2248 slash_old - old != slash_new - new ||
2249 memcmp(old, new, slash_new - new))
2250 break;
2251 old = slash_old + 1;
2252 new = slash_new + 1;
2253 }
2254 /* p->old_name thru old is the common prefix, and old and new
2255 * through the end of names are renames
2256 */
2257 if (old != p->old_name)
2258 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2259 (int)(old - p->old_name), p->old_name,
2260 old, new, p->score);
2261 else
2262 printf(" %s %s => %s (%d%%)\n", renamecopy,
2263 p->old_name, p->new_name, p->score);
2264 show_mode_change(p, 0);
2265 }
2266
2267 static void summary_patch_list(struct patch *patch)
2268 {
2269 struct patch *p;
2270
2271 for (p = patch; p; p = p->next) {
2272 if (p->is_new)
2273 show_file_mode_name("create", p->new_mode, p->new_name);
2274 else if (p->is_delete)
2275 show_file_mode_name("delete", p->old_mode, p->old_name);
2276 else {
2277 if (p->is_rename || p->is_copy)
2278 show_rename_copy(p);
2279 else {
2280 if (p->score) {
2281 printf(" rewrite %s (%d%%)\n",
2282 p->new_name, p->score);
2283 show_mode_change(p, 0);
2284 }
2285 else
2286 show_mode_change(p, 1);
2287 }
2288 }
2289 }
2290 }
2291
2292 static void patch_stats(struct patch *patch)
2293 {
2294 int lines = patch->lines_added + patch->lines_deleted;
2295
2296 if (lines > max_change)
2297 max_change = lines;
2298 if (patch->old_name) {
2299 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2300 if (!len)
2301 len = strlen(patch->old_name);
2302 if (len > max_len)
2303 max_len = len;
2304 }
2305 if (patch->new_name) {
2306 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2307 if (!len)
2308 len = strlen(patch->new_name);
2309 if (len > max_len)
2310 max_len = len;
2311 }
2312 }
2313
2314 static void remove_file(struct patch *patch, int rmdir_empty)
2315 {
2316 if (update_index) {
2317 if (remove_file_from_cache(patch->old_name) < 0)
2318 die("unable to remove %s from index", patch->old_name);
2319 }
2320 if (!cached) {
2321 if (S_ISGITLINK(patch->old_mode)) {
2322 if (rmdir(patch->old_name))
2323 warning("unable to remove submodule %s",
2324 patch->old_name);
2325 } else if (!unlink(patch->old_name) && rmdir_empty) {
2326 char *name = xstrdup(patch->old_name);
2327 char *end = strrchr(name, '/');
2328 while (end) {
2329 *end = 0;
2330 if (rmdir(name))
2331 break;
2332 end = strrchr(name, '/');
2333 }
2334 free(name);
2335 }
2336 }
2337 }
2338
2339 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2340 {
2341 struct stat st;
2342 struct cache_entry *ce;
2343 int namelen = strlen(path);
2344 unsigned ce_size = cache_entry_size(namelen);
2345
2346 if (!update_index)
2347 return;
2348
2349 ce = xcalloc(1, ce_size);
2350 memcpy(ce->name, path, namelen);
2351 ce->ce_mode = create_ce_mode(mode);
2352 ce->ce_flags = htons(namelen);
2353 if (S_ISGITLINK(mode)) {
2354 const char *s = buf;
2355
2356 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2357 die("corrupt patch for subproject %s", path);
2358 } else {
2359 if (!cached) {
2360 if (lstat(path, &st) < 0)
2361 die("unable to stat newly created file %s",
2362 path);
2363 fill_stat_cache_info(ce, &st);
2364 }
2365 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2366 die("unable to create backing store for newly created file %s", path);
2367 }
2368 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2369 die("unable to add cache entry for %s", path);
2370 }
2371
2372 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2373 {
2374 int fd;
2375 struct strbuf nbuf;
2376
2377 if (S_ISGITLINK(mode)) {
2378 struct stat st;
2379 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2380 return 0;
2381 return mkdir(path, 0777);
2382 }
2383
2384 if (has_symlinks && S_ISLNK(mode))
2385 /* Although buf:size is counted string, it also is NUL
2386 * terminated.
2387 */
2388 return symlink(buf, path);
2389
2390 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2391 if (fd < 0)
2392 return -1;
2393
2394 strbuf_init(&nbuf, 0);
2395 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2396 size = nbuf.len;
2397 buf = nbuf.buf;
2398 }
2399 write_or_die(fd, buf, size);
2400 strbuf_release(&nbuf);
2401
2402 if (close(fd) < 0)
2403 die("closing file %s: %s", path, strerror(errno));
2404 return 0;
2405 }
2406
2407 /*
2408 * We optimistically assume that the directories exist,
2409 * which is true 99% of the time anyway. If they don't,
2410 * we create them and try again.
2411 */
2412 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2413 {
2414 if (cached)
2415 return;
2416 if (!try_create_file(path, mode, buf, size))
2417 return;
2418
2419 if (errno == ENOENT) {
2420 if (safe_create_leading_directories(path))
2421 return;
2422 if (!try_create_file(path, mode, buf, size))
2423 return;
2424 }
2425
2426 if (errno == EEXIST || errno == EACCES) {
2427 /* We may be trying to create a file where a directory
2428 * used to be.
2429 */
2430 struct stat st;
2431 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2432 errno = EEXIST;
2433 }
2434
2435 if (errno == EEXIST) {
2436 unsigned int nr = getpid();
2437
2438 for (;;) {
2439 const char *newpath;
2440 newpath = mkpath("%s~%u", path, nr);
2441 if (!try_create_file(newpath, mode, buf, size)) {
2442 if (!rename(newpath, path))
2443 return;
2444 unlink(newpath);
2445 break;
2446 }
2447 if (errno != EEXIST)
2448 break;
2449 ++nr;
2450 }
2451 }
2452 die("unable to write file %s mode %o", path, mode);
2453 }
2454
2455 static void create_file(struct patch *patch)
2456 {
2457 char *path = patch->new_name;
2458 unsigned mode = patch->new_mode;
2459 unsigned long size = patch->resultsize;
2460 char *buf = patch->result;
2461
2462 if (!mode)
2463 mode = S_IFREG | 0644;
2464 create_one_file(path, mode, buf, size);
2465 add_index_file(path, mode, buf, size);
2466 }
2467
2468 /* phase zero is to remove, phase one is to create */
2469 static void write_out_one_result(struct patch *patch, int phase)
2470 {
2471 if (patch->is_delete > 0) {
2472 if (phase == 0)
2473 remove_file(patch, 1);
2474 return;
2475 }
2476 if (patch->is_new > 0 || patch->is_copy) {
2477 if (phase == 1)
2478 create_file(patch);
2479 return;
2480 }
2481 /*
2482 * Rename or modification boils down to the same
2483 * thing: remove the old, write the new
2484 */
2485 if (phase == 0)
2486 remove_file(patch, patch->is_rename);
2487 if (phase == 1)
2488 create_file(patch);
2489 }
2490
2491 static int write_out_one_reject(struct patch *patch)
2492 {
2493 FILE *rej;
2494 char namebuf[PATH_MAX];
2495 struct fragment *frag;
2496 int cnt = 0;
2497
2498 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2499 if (!frag->rejected)
2500 continue;
2501 cnt++;
2502 }
2503
2504 if (!cnt) {
2505 if (apply_verbosely)
2506 say_patch_name(stderr,
2507 "Applied patch ", patch, " cleanly.\n");
2508 return 0;
2509 }
2510
2511 /* This should not happen, because a removal patch that leaves
2512 * contents are marked "rejected" at the patch level.
2513 */
2514 if (!patch->new_name)
2515 die("internal error");
2516
2517 /* Say this even without --verbose */
2518 say_patch_name(stderr, "Applying patch ", patch, " with");
2519 fprintf(stderr, " %d rejects...\n", cnt);
2520
2521 cnt = strlen(patch->new_name);
2522 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2523 cnt = ARRAY_SIZE(namebuf) - 5;
2524 fprintf(stderr,
2525 "warning: truncating .rej filename to %.*s.rej",
2526 cnt - 1, patch->new_name);
2527 }
2528 memcpy(namebuf, patch->new_name, cnt);
2529 memcpy(namebuf + cnt, ".rej", 5);
2530
2531 rej = fopen(namebuf, "w");
2532 if (!rej)
2533 return error("cannot open %s: %s", namebuf, strerror(errno));
2534
2535 /* Normal git tools never deal with .rej, so do not pretend
2536 * this is a git patch by saying --git nor give extended
2537 * headers. While at it, maybe please "kompare" that wants
2538 * the trailing TAB and some garbage at the end of line ;-).
2539 */
2540 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2541 patch->new_name, patch->new_name);
2542 for (cnt = 1, frag = patch->fragments;
2543 frag;
2544 cnt++, frag = frag->next) {
2545 if (!frag->rejected) {
2546 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2547 continue;
2548 }
2549 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2550 fprintf(rej, "%.*s", frag->size, frag->patch);
2551 if (frag->patch[frag->size-1] != '\n')
2552 fputc('\n', rej);
2553 }
2554 fclose(rej);
2555 return -1;
2556 }
2557
2558 static int write_out_results(struct patch *list, int skipped_patch)
2559 {
2560 int phase;
2561 int errs = 0;
2562 struct patch *l;
2563
2564 if (!list && !skipped_patch)
2565 return error("No changes");
2566
2567 for (phase = 0; phase < 2; phase++) {
2568 l = list;
2569 while (l) {
2570 if (l->rejected)
2571 errs = 1;
2572 else {
2573 write_out_one_result(l, phase);
2574 if (phase == 1 && write_out_one_reject(l))
2575 errs = 1;
2576 }
2577 l = l->next;
2578 }
2579 }
2580 return errs;
2581 }
2582
2583 static struct lock_file lock_file;
2584
2585 static struct excludes {
2586 struct excludes *next;
2587 const char *path;
2588 } *excludes;
2589
2590 static int use_patch(struct patch *p)
2591 {
2592 const char *pathname = p->new_name ? p->new_name : p->old_name;
2593 struct excludes *x = excludes;
2594 while (x) {
2595 if (fnmatch(x->path, pathname, 0) == 0)
2596 return 0;
2597 x = x->next;
2598 }
2599 if (0 < prefix_length) {
2600 int pathlen = strlen(pathname);
2601 if (pathlen <= prefix_length ||
2602 memcmp(prefix, pathname, prefix_length))
2603 return 0;
2604 }
2605 return 1;
2606 }
2607
2608 static void prefix_one(char **name)
2609 {
2610 char *old_name = *name;
2611 if (!old_name)
2612 return;
2613 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2614 free(old_name);
2615 }
2616
2617 static void prefix_patches(struct patch *p)
2618 {
2619 if (!prefix || p->is_toplevel_relative)
2620 return;
2621 for ( ; p; p = p->next) {
2622 if (p->new_name == p->old_name) {
2623 char *prefixed = p->new_name;
2624 prefix_one(&prefixed);
2625 p->new_name = p->old_name = prefixed;
2626 }
2627 else {
2628 prefix_one(&p->new_name);
2629 prefix_one(&p->old_name);
2630 }
2631 }
2632 }
2633
2634 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2635 {
2636 unsigned long offset, size;
2637 char *buffer = read_patch_file(fd, &size);
2638 struct patch *list = NULL, **listp = &list;
2639 int skipped_patch = 0;
2640
2641 patch_input_file = filename;
2642 if (!buffer)
2643 return -1;
2644 offset = 0;
2645 while (size > 0) {
2646 struct patch *patch;
2647 int nr;
2648
2649 patch = xcalloc(1, sizeof(*patch));
2650 patch->inaccurate_eof = inaccurate_eof;
2651 nr = parse_chunk(buffer + offset, size, patch);
2652 if (nr < 0)
2653 break;
2654 if (apply_in_reverse)
2655 reverse_patches(patch);
2656 if (prefix)
2657 prefix_patches(patch);
2658 if (use_patch(patch)) {
2659 patch_stats(patch);
2660 *listp = patch;
2661 listp = &patch->next;
2662 }
2663 else {
2664 /* perhaps free it a bit better? */
2665 free(patch);
2666 skipped_patch++;
2667 }
2668 offset += nr;
2669 size -= nr;
2670 }
2671
2672 if (whitespace_error && (new_whitespace == error_on_whitespace))
2673 apply = 0;
2674
2675 update_index = check_index && apply;
2676 if (update_index && newfd < 0)
2677 newfd = hold_locked_index(&lock_file, 1);
2678
2679 if (check_index) {
2680 if (read_cache() < 0)
2681 die("unable to read index file");
2682 }
2683
2684 if ((check || apply) &&
2685 check_patch_list(list) < 0 &&
2686 !apply_with_reject)
2687 exit(1);
2688
2689 if (apply && write_out_results(list, skipped_patch))
2690 exit(1);
2691
2692 if (show_index_info)
2693 show_index_list(list);
2694
2695 if (diffstat)
2696 stat_patch_list(list);
2697
2698 if (numstat)
2699 numstat_patch_list(list);
2700
2701 if (summary)
2702 summary_patch_list(list);
2703
2704 free(buffer);
2705 return 0;
2706 }
2707
2708 static int git_apply_config(const char *var, const char *value)
2709 {
2710 if (!strcmp(var, "apply.whitespace")) {
2711 apply_default_whitespace = xstrdup(value);
2712 return 0;
2713 }
2714 return git_default_config(var, value);
2715 }
2716
2717
2718 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2719 {
2720 int i;
2721 int read_stdin = 1;
2722 int inaccurate_eof = 0;
2723 int errs = 0;
2724 int is_not_gitdir = 0;
2725
2726 const char *whitespace_option = NULL;
2727
2728 prefix = setup_git_directory_gently(&is_not_gitdir);
2729 prefix_length = prefix ? strlen(prefix) : 0;
2730 git_config(git_apply_config);
2731 if (apply_default_whitespace)
2732 parse_whitespace_option(apply_default_whitespace);
2733
2734 for (i = 1; i < argc; i++) {
2735 const char *arg = argv[i];
2736 char *end;
2737 int fd;
2738
2739 if (!strcmp(arg, "-")) {
2740 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2741 read_stdin = 0;
2742 continue;
2743 }
2744 if (!prefixcmp(arg, "--exclude=")) {
2745 struct excludes *x = xmalloc(sizeof(*x));
2746 x->path = arg + 10;
2747 x->next = excludes;
2748 excludes = x;
2749 continue;
2750 }
2751 if (!prefixcmp(arg, "-p")) {
2752 p_value = atoi(arg + 2);
2753 p_value_known = 1;
2754 continue;
2755 }
2756 if (!strcmp(arg, "--no-add")) {
2757 no_add = 1;
2758 continue;
2759 }
2760 if (!strcmp(arg, "--stat")) {
2761 apply = 0;
2762 diffstat = 1;
2763 continue;
2764 }
2765 if (!strcmp(arg, "--allow-binary-replacement") ||
2766 !strcmp(arg, "--binary")) {
2767 continue; /* now no-op */
2768 }
2769 if (!strcmp(arg, "--numstat")) {
2770 apply = 0;
2771 numstat = 1;
2772 continue;
2773 }
2774 if (!strcmp(arg, "--summary")) {
2775 apply = 0;
2776 summary = 1;
2777 continue;
2778 }
2779 if (!strcmp(arg, "--check")) {
2780 apply = 0;
2781 check = 1;
2782 continue;
2783 }
2784 if (!strcmp(arg, "--index")) {
2785 if (is_not_gitdir)
2786 die("--index outside a repository");
2787 check_index = 1;
2788 continue;
2789 }
2790 if (!strcmp(arg, "--cached")) {
2791 if (is_not_gitdir)
2792 die("--cached outside a repository");
2793 check_index = 1;
2794 cached = 1;
2795 continue;
2796 }
2797 if (!strcmp(arg, "--apply")) {
2798 apply = 1;
2799 continue;
2800 }
2801 if (!strcmp(arg, "--index-info")) {
2802 apply = 0;
2803 show_index_info = 1;
2804 continue;
2805 }
2806 if (!strcmp(arg, "-z")) {
2807 line_termination = 0;
2808 continue;
2809 }
2810 if (!prefixcmp(arg, "-C")) {
2811 p_context = strtoul(arg + 2, &end, 0);
2812 if (*end != '\0')
2813 die("unrecognized context count '%s'", arg + 2);
2814 continue;
2815 }
2816 if (!prefixcmp(arg, "--whitespace=")) {
2817 whitespace_option = arg + 13;
2818 parse_whitespace_option(arg + 13);
2819 continue;
2820 }
2821 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2822 apply_in_reverse = 1;
2823 continue;
2824 }
2825 if (!strcmp(arg, "--unidiff-zero")) {
2826 unidiff_zero = 1;
2827 continue;
2828 }
2829 if (!strcmp(arg, "--reject")) {
2830 apply = apply_with_reject = apply_verbosely = 1;
2831 continue;
2832 }
2833 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2834 apply_verbosely = 1;
2835 continue;
2836 }
2837 if (!strcmp(arg, "--inaccurate-eof")) {
2838 inaccurate_eof = 1;
2839 continue;
2840 }
2841 if (0 < prefix_length)
2842 arg = prefix_filename(prefix, prefix_length, arg);
2843
2844 fd = open(arg, O_RDONLY);
2845 if (fd < 0)
2846 usage(apply_usage);
2847 read_stdin = 0;
2848 set_default_whitespace_mode(whitespace_option);
2849 errs |= apply_patch(fd, arg, inaccurate_eof);
2850 close(fd);
2851 }
2852 set_default_whitespace_mode(whitespace_option);
2853 if (read_stdin)
2854 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2855 if (whitespace_error) {
2856 if (squelch_whitespace_errors &&
2857 squelch_whitespace_errors < whitespace_error) {
2858 int squelched =
2859 whitespace_error - squelch_whitespace_errors;
2860 fprintf(stderr, "warning: squelched %d "
2861 "whitespace error%s\n",
2862 squelched,
2863 squelched == 1 ? "" : "s");
2864 }
2865 if (new_whitespace == error_on_whitespace)
2866 die("%d line%s add%s whitespace errors.",
2867 whitespace_error,
2868 whitespace_error == 1 ? "" : "s",
2869 whitespace_error == 1 ? "s" : "");
2870 if (applied_after_fixing_ws)
2871 fprintf(stderr, "warning: %d line%s applied after"
2872 " fixing whitespace errors.\n",
2873 applied_after_fixing_ws,
2874 applied_after_fixing_ws == 1 ? "" : "s");
2875 else if (whitespace_error)
2876 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
2877 whitespace_error,
2878 whitespace_error == 1 ? "" : "s",
2879 whitespace_error == 1 ? "s" : "");
2880 }
2881
2882 if (update_index) {
2883 if (write_cache(newfd, active_cache, active_nr) ||
2884 close(newfd) || commit_locked_index(&lock_file))
2885 die("Unable to write new index file");
2886 }
2887
2888 return !!errs;
2889 }