Make git-prune submodule aware (and fix a SEGFAULT in the process)
[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 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_stripping;
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 write_name_quoted(NULL, 0, patch->old_name, 1, output);
167 fputs(" => ", output);
168 write_name_quoted(NULL, 0, patch->new_name, 1, output);
169 }
170 else {
171 const char *n = patch->new_name;
172 if (!n)
173 n = patch->old_name;
174 write_name_quoted(NULL, 0, n, 1, output);
175 }
176 fputs(post, output);
177 }
178
179 #define CHUNKSIZE (8192)
180 #define SLOP (16)
181
182 static void *read_patch_file(int fd, unsigned long *sizep)
183 {
184 unsigned long size = 0, alloc = CHUNKSIZE;
185 void *buffer = xmalloc(alloc);
186
187 for (;;) {
188 ssize_t nr = alloc - size;
189 if (nr < 1024) {
190 alloc += CHUNKSIZE;
191 buffer = xrealloc(buffer, alloc);
192 nr = alloc - size;
193 }
194 nr = xread(fd, (char *) buffer + size, nr);
195 if (!nr)
196 break;
197 if (nr < 0)
198 die("git-apply: read returned %s", strerror(errno));
199 size += nr;
200 }
201 *sizep = size;
202
203 /*
204 * Make sure that we have some slop in the buffer
205 * so that we can do speculative "memcmp" etc, and
206 * see to it that it is NUL-filled.
207 */
208 if (alloc < size + SLOP)
209 buffer = xrealloc(buffer, size + SLOP);
210 memset((char *) buffer + size, 0, SLOP);
211 return buffer;
212 }
213
214 static unsigned long linelen(const char *buffer, unsigned long size)
215 {
216 unsigned long len = 0;
217 while (size--) {
218 len++;
219 if (*buffer++ == '\n')
220 break;
221 }
222 return len;
223 }
224
225 static int is_dev_null(const char *str)
226 {
227 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
228 }
229
230 #define TERM_SPACE 1
231 #define TERM_TAB 2
232
233 static int name_terminate(const char *name, int namelen, int c, int terminate)
234 {
235 if (c == ' ' && !(terminate & TERM_SPACE))
236 return 0;
237 if (c == '\t' && !(terminate & TERM_TAB))
238 return 0;
239
240 return 1;
241 }
242
243 static char *find_name(const char *line, char *def, int p_value, int terminate)
244 {
245 int len;
246 const char *start = line;
247 char *name;
248
249 if (*line == '"') {
250 /* Proposed "new-style" GNU patch/diff format; see
251 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
252 */
253 name = unquote_c_style(line, NULL);
254 if (name) {
255 char *cp = name;
256 while (p_value) {
257 cp = strchr(name, '/');
258 if (!cp)
259 break;
260 cp++;
261 p_value--;
262 }
263 if (cp) {
264 /* name can later be freed, so we need
265 * to memmove, not just return cp
266 */
267 memmove(name, cp, strlen(cp) + 1);
268 free(def);
269 return name;
270 }
271 else {
272 free(name);
273 name = NULL;
274 }
275 }
276 }
277
278 for (;;) {
279 char c = *line;
280
281 if (isspace(c)) {
282 if (c == '\n')
283 break;
284 if (name_terminate(start, line-start, c, terminate))
285 break;
286 }
287 line++;
288 if (c == '/' && !--p_value)
289 start = line;
290 }
291 if (!start)
292 return def;
293 len = line - start;
294 if (!len)
295 return def;
296
297 /*
298 * Generally we prefer the shorter name, especially
299 * if the other one is just a variation of that with
300 * something else tacked on to the end (ie "file.orig"
301 * or "file~").
302 */
303 if (def) {
304 int deflen = strlen(def);
305 if (deflen < len && !strncmp(start, def, deflen))
306 return def;
307 }
308
309 name = xmalloc(len + 1);
310 memcpy(name, start, len);
311 name[len] = 0;
312 free(def);
313 return name;
314 }
315
316 static int count_slashes(const char *cp)
317 {
318 int cnt = 0;
319 char ch;
320
321 while ((ch = *cp++))
322 if (ch == '/')
323 cnt++;
324 return cnt;
325 }
326
327 /*
328 * Given the string after "--- " or "+++ ", guess the appropriate
329 * p_value for the given patch.
330 */
331 static int guess_p_value(const char *nameline)
332 {
333 char *name, *cp;
334 int val = -1;
335
336 if (is_dev_null(nameline))
337 return -1;
338 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
339 if (!name)
340 return -1;
341 cp = strchr(name, '/');
342 if (!cp)
343 val = 0;
344 else if (prefix) {
345 /*
346 * Does it begin with "a/$our-prefix" and such? Then this is
347 * very likely to apply to our directory.
348 */
349 if (!strncmp(name, prefix, prefix_length))
350 val = count_slashes(prefix);
351 else {
352 cp++;
353 if (!strncmp(cp, prefix, prefix_length))
354 val = count_slashes(prefix) + 1;
355 }
356 }
357 free(name);
358 return val;
359 }
360
361 /*
362 * Get the name etc info from the --/+++ lines of a traditional patch header
363 *
364 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
365 * files, we can happily check the index for a match, but for creating a
366 * new file we should try to match whatever "patch" does. I have no idea.
367 */
368 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
369 {
370 char *name;
371
372 first += 4; /* skip "--- " */
373 second += 4; /* skip "+++ " */
374 if (!p_value_known) {
375 int p, q;
376 p = guess_p_value(first);
377 q = guess_p_value(second);
378 if (p < 0) p = q;
379 if (0 <= p && p == q) {
380 p_value = p;
381 p_value_known = 1;
382 }
383 }
384 if (is_dev_null(first)) {
385 patch->is_new = 1;
386 patch->is_delete = 0;
387 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
388 patch->new_name = name;
389 } else if (is_dev_null(second)) {
390 patch->is_new = 0;
391 patch->is_delete = 1;
392 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
393 patch->old_name = name;
394 } else {
395 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
396 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
397 patch->old_name = patch->new_name = name;
398 }
399 if (!name)
400 die("unable to find filename in patch at line %d", linenr);
401 }
402
403 static int gitdiff_hdrend(const char *line, struct patch *patch)
404 {
405 return -1;
406 }
407
408 /*
409 * We're anal about diff header consistency, to make
410 * sure that we don't end up having strange ambiguous
411 * patches floating around.
412 *
413 * As a result, gitdiff_{old|new}name() will check
414 * their names against any previous information, just
415 * to make sure..
416 */
417 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
418 {
419 if (!orig_name && !isnull)
420 return find_name(line, NULL, p_value, TERM_TAB);
421
422 if (orig_name) {
423 int len;
424 const char *name;
425 char *another;
426 name = orig_name;
427 len = strlen(name);
428 if (isnull)
429 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
430 another = find_name(line, NULL, p_value, TERM_TAB);
431 if (!another || memcmp(another, name, len))
432 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
433 free(another);
434 return orig_name;
435 }
436 else {
437 /* expect "/dev/null" */
438 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
439 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
440 return NULL;
441 }
442 }
443
444 static int gitdiff_oldname(const char *line, struct patch *patch)
445 {
446 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
447 return 0;
448 }
449
450 static int gitdiff_newname(const char *line, struct patch *patch)
451 {
452 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
453 return 0;
454 }
455
456 static int gitdiff_oldmode(const char *line, struct patch *patch)
457 {
458 patch->old_mode = strtoul(line, NULL, 8);
459 return 0;
460 }
461
462 static int gitdiff_newmode(const char *line, struct patch *patch)
463 {
464 patch->new_mode = strtoul(line, NULL, 8);
465 return 0;
466 }
467
468 static int gitdiff_delete(const char *line, struct patch *patch)
469 {
470 patch->is_delete = 1;
471 patch->old_name = patch->def_name;
472 return gitdiff_oldmode(line, patch);
473 }
474
475 static int gitdiff_newfile(const char *line, struct patch *patch)
476 {
477 patch->is_new = 1;
478 patch->new_name = patch->def_name;
479 return gitdiff_newmode(line, patch);
480 }
481
482 static int gitdiff_copysrc(const char *line, struct patch *patch)
483 {
484 patch->is_copy = 1;
485 patch->old_name = find_name(line, NULL, 0, 0);
486 return 0;
487 }
488
489 static int gitdiff_copydst(const char *line, struct patch *patch)
490 {
491 patch->is_copy = 1;
492 patch->new_name = find_name(line, NULL, 0, 0);
493 return 0;
494 }
495
496 static int gitdiff_renamesrc(const char *line, struct patch *patch)
497 {
498 patch->is_rename = 1;
499 patch->old_name = find_name(line, NULL, 0, 0);
500 return 0;
501 }
502
503 static int gitdiff_renamedst(const char *line, struct patch *patch)
504 {
505 patch->is_rename = 1;
506 patch->new_name = find_name(line, NULL, 0, 0);
507 return 0;
508 }
509
510 static int gitdiff_similarity(const char *line, struct patch *patch)
511 {
512 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
513 patch->score = 0;
514 return 0;
515 }
516
517 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
518 {
519 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
520 patch->score = 0;
521 return 0;
522 }
523
524 static int gitdiff_index(const char *line, struct patch *patch)
525 {
526 /* index line is N hexadecimal, "..", N hexadecimal,
527 * and optional space with octal mode.
528 */
529 const char *ptr, *eol;
530 int len;
531
532 ptr = strchr(line, '.');
533 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
534 return 0;
535 len = ptr - line;
536 memcpy(patch->old_sha1_prefix, line, len);
537 patch->old_sha1_prefix[len] = 0;
538
539 line = ptr + 2;
540 ptr = strchr(line, ' ');
541 eol = strchr(line, '\n');
542
543 if (!ptr || eol < ptr)
544 ptr = eol;
545 len = ptr - line;
546
547 if (40 < len)
548 return 0;
549 memcpy(patch->new_sha1_prefix, line, len);
550 patch->new_sha1_prefix[len] = 0;
551 if (*ptr == ' ')
552 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
553 return 0;
554 }
555
556 /*
557 * This is normal for a diff that doesn't change anything: we'll fall through
558 * into the next diff. Tell the parser to break out.
559 */
560 static int gitdiff_unrecognized(const char *line, struct patch *patch)
561 {
562 return -1;
563 }
564
565 static const char *stop_at_slash(const char *line, int llen)
566 {
567 int i;
568
569 for (i = 0; i < llen; i++) {
570 int ch = line[i];
571 if (ch == '/')
572 return line + i;
573 }
574 return NULL;
575 }
576
577 /* This is to extract the same name that appears on "diff --git"
578 * line. We do not find and return anything if it is a rename
579 * patch, and it is OK because we will find the name elsewhere.
580 * We need to reliably find name only when it is mode-change only,
581 * creation or deletion of an empty file. In any of these cases,
582 * both sides are the same name under a/ and b/ respectively.
583 */
584 static char *git_header_name(char *line, int llen)
585 {
586 int len;
587 const char *name;
588 const char *second = NULL;
589
590 line += strlen("diff --git ");
591 llen -= strlen("diff --git ");
592
593 if (*line == '"') {
594 const char *cp;
595 char *first = unquote_c_style(line, &second);
596 if (!first)
597 return NULL;
598
599 /* advance to the first slash */
600 cp = stop_at_slash(first, strlen(first));
601 if (!cp || cp == first) {
602 /* we do not accept absolute paths */
603 free_first_and_fail:
604 free(first);
605 return NULL;
606 }
607 len = strlen(cp+1);
608 memmove(first, cp+1, len+1); /* including NUL */
609
610 /* second points at one past closing dq of name.
611 * find the second name.
612 */
613 while ((second < line + llen) && isspace(*second))
614 second++;
615
616 if (line + llen <= second)
617 goto free_first_and_fail;
618 if (*second == '"') {
619 char *sp = unquote_c_style(second, NULL);
620 if (!sp)
621 goto free_first_and_fail;
622 cp = stop_at_slash(sp, strlen(sp));
623 if (!cp || cp == sp) {
624 free_both_and_fail:
625 free(sp);
626 goto free_first_and_fail;
627 }
628 /* They must match, otherwise ignore */
629 if (strcmp(cp+1, first))
630 goto free_both_and_fail;
631 free(sp);
632 return first;
633 }
634
635 /* unquoted second */
636 cp = stop_at_slash(second, line + llen - second);
637 if (!cp || cp == second)
638 goto free_first_and_fail;
639 cp++;
640 if (line + llen - cp != len + 1 ||
641 memcmp(first, cp, len))
642 goto free_first_and_fail;
643 return first;
644 }
645
646 /* unquoted first name */
647 name = stop_at_slash(line, llen);
648 if (!name || name == line)
649 return NULL;
650
651 name++;
652
653 /* since the first name is unquoted, a dq if exists must be
654 * the beginning of the second name.
655 */
656 for (second = name; second < line + llen; second++) {
657 if (*second == '"') {
658 const char *cp = second;
659 const char *np;
660 char *sp = unquote_c_style(second, NULL);
661
662 if (!sp)
663 return NULL;
664 np = stop_at_slash(sp, strlen(sp));
665 if (!np || np == sp) {
666 free_second_and_fail:
667 free(sp);
668 return NULL;
669 }
670 np++;
671 len = strlen(np);
672 if (len < cp - name &&
673 !strncmp(np, name, len) &&
674 isspace(name[len])) {
675 /* Good */
676 memmove(sp, np, len + 1);
677 return sp;
678 }
679 goto free_second_and_fail;
680 }
681 }
682
683 /*
684 * Accept a name only if it shows up twice, exactly the same
685 * form.
686 */
687 for (len = 0 ; ; len++) {
688 switch (name[len]) {
689 default:
690 continue;
691 case '\n':
692 return NULL;
693 case '\t': case ' ':
694 second = name+len;
695 for (;;) {
696 char c = *second++;
697 if (c == '\n')
698 return NULL;
699 if (c == '/')
700 break;
701 }
702 if (second[len] == '\n' && !memcmp(name, second, len)) {
703 char *ret = xmalloc(len + 1);
704 memcpy(ret, name, len);
705 ret[len] = 0;
706 return ret;
707 }
708 }
709 }
710 return NULL;
711 }
712
713 /* Verify that we recognize the lines following a git header */
714 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
715 {
716 unsigned long offset;
717
718 /* A git diff has explicit new/delete information, so we don't guess */
719 patch->is_new = 0;
720 patch->is_delete = 0;
721
722 /*
723 * Some things may not have the old name in the
724 * rest of the headers anywhere (pure mode changes,
725 * or removing or adding empty files), so we get
726 * the default name from the header.
727 */
728 patch->def_name = git_header_name(line, len);
729
730 line += len;
731 size -= len;
732 linenr++;
733 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
734 static const struct opentry {
735 const char *str;
736 int (*fn)(const char *, struct patch *);
737 } optable[] = {
738 { "@@ -", gitdiff_hdrend },
739 { "--- ", gitdiff_oldname },
740 { "+++ ", gitdiff_newname },
741 { "old mode ", gitdiff_oldmode },
742 { "new mode ", gitdiff_newmode },
743 { "deleted file mode ", gitdiff_delete },
744 { "new file mode ", gitdiff_newfile },
745 { "copy from ", gitdiff_copysrc },
746 { "copy to ", gitdiff_copydst },
747 { "rename old ", gitdiff_renamesrc },
748 { "rename new ", gitdiff_renamedst },
749 { "rename from ", gitdiff_renamesrc },
750 { "rename to ", gitdiff_renamedst },
751 { "similarity index ", gitdiff_similarity },
752 { "dissimilarity index ", gitdiff_dissimilarity },
753 { "index ", gitdiff_index },
754 { "", gitdiff_unrecognized },
755 };
756 int i;
757
758 len = linelen(line, size);
759 if (!len || line[len-1] != '\n')
760 break;
761 for (i = 0; i < ARRAY_SIZE(optable); i++) {
762 const struct opentry *p = optable + i;
763 int oplen = strlen(p->str);
764 if (len < oplen || memcmp(p->str, line, oplen))
765 continue;
766 if (p->fn(line + oplen, patch) < 0)
767 return offset;
768 break;
769 }
770 }
771
772 return offset;
773 }
774
775 static int parse_num(const char *line, unsigned long *p)
776 {
777 char *ptr;
778
779 if (!isdigit(*line))
780 return 0;
781 *p = strtoul(line, &ptr, 10);
782 return ptr - line;
783 }
784
785 static int parse_range(const char *line, int len, int offset, const char *expect,
786 unsigned long *p1, unsigned long *p2)
787 {
788 int digits, ex;
789
790 if (offset < 0 || offset >= len)
791 return -1;
792 line += offset;
793 len -= offset;
794
795 digits = parse_num(line, p1);
796 if (!digits)
797 return -1;
798
799 offset += digits;
800 line += digits;
801 len -= digits;
802
803 *p2 = 1;
804 if (*line == ',') {
805 digits = parse_num(line+1, p2);
806 if (!digits)
807 return -1;
808
809 offset += digits+1;
810 line += digits+1;
811 len -= digits+1;
812 }
813
814 ex = strlen(expect);
815 if (ex > len)
816 return -1;
817 if (memcmp(line, expect, ex))
818 return -1;
819
820 return offset + ex;
821 }
822
823 /*
824 * Parse a unified diff fragment header of the
825 * form "@@ -a,b +c,d @@"
826 */
827 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
828 {
829 int offset;
830
831 if (!len || line[len-1] != '\n')
832 return -1;
833
834 /* Figure out the number of lines in a fragment */
835 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
836 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
837
838 return offset;
839 }
840
841 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
842 {
843 unsigned long offset, len;
844
845 patch->is_toplevel_relative = 0;
846 patch->is_rename = patch->is_copy = 0;
847 patch->is_new = patch->is_delete = -1;
848 patch->old_mode = patch->new_mode = 0;
849 patch->old_name = patch->new_name = NULL;
850 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
851 unsigned long nextlen;
852
853 len = linelen(line, size);
854 if (!len)
855 break;
856
857 /* Testing this early allows us to take a few shortcuts.. */
858 if (len < 6)
859 continue;
860
861 /*
862 * Make sure we don't find any unconnected patch fragments.
863 * That's a sign that we didn't find a header, and that a
864 * patch has become corrupted/broken up.
865 */
866 if (!memcmp("@@ -", line, 4)) {
867 struct fragment dummy;
868 if (parse_fragment_header(line, len, &dummy) < 0)
869 continue;
870 die("patch fragment without header at line %d: %.*s",
871 linenr, (int)len-1, line);
872 }
873
874 if (size < len + 6)
875 break;
876
877 /*
878 * Git patch? It might not have a real patch, just a rename
879 * or mode change, so we handle that specially
880 */
881 if (!memcmp("diff --git ", line, 11)) {
882 int git_hdr_len = parse_git_header(line, len, size, patch);
883 if (git_hdr_len <= len)
884 continue;
885 if (!patch->old_name && !patch->new_name) {
886 if (!patch->def_name)
887 die("git diff header lacks filename information (line %d)", linenr);
888 patch->old_name = patch->new_name = patch->def_name;
889 }
890 patch->is_toplevel_relative = 1;
891 *hdrsize = git_hdr_len;
892 return offset;
893 }
894
895 /** --- followed by +++ ? */
896 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
897 continue;
898
899 /*
900 * We only accept unified patches, so we want it to
901 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
902 * minimum
903 */
904 nextlen = linelen(line + len, size - len);
905 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
906 continue;
907
908 /* Ok, we'll consider it a patch */
909 parse_traditional_patch(line, line+len, patch);
910 *hdrsize = len + nextlen;
911 linenr += 2;
912 return offset;
913 }
914 return -1;
915 }
916
917 static void check_whitespace(const char *line, int len)
918 {
919 const char *err = "Adds trailing whitespace";
920 int seen_space = 0;
921 int i;
922
923 /*
924 * We know len is at least two, since we have a '+' and we
925 * checked that the last character was a '\n' before calling
926 * this function. That is, an addition of an empty line would
927 * check the '+' here. Sneaky...
928 */
929 if (isspace(line[len-2]))
930 goto error;
931
932 /*
933 * Make sure that there is no space followed by a tab in
934 * indentation.
935 */
936 err = "Space in indent is followed by a tab";
937 for (i = 1; i < len; i++) {
938 if (line[i] == '\t') {
939 if (seen_space)
940 goto error;
941 }
942 else if (line[i] == ' ')
943 seen_space = 1;
944 else
945 break;
946 }
947 return;
948
949 error:
950 whitespace_error++;
951 if (squelch_whitespace_errors &&
952 squelch_whitespace_errors < whitespace_error)
953 ;
954 else
955 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
956 err, patch_input_file, linenr, len-2, line+1);
957 }
958
959
960 /*
961 * Parse a unified diff. Note that this really needs to parse each
962 * fragment separately, since the only way to know the difference
963 * between a "---" that is part of a patch, and a "---" that starts
964 * the next patch is to look at the line counts..
965 */
966 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
967 {
968 int added, deleted;
969 int len = linelen(line, size), offset;
970 unsigned long oldlines, newlines;
971 unsigned long leading, trailing;
972
973 offset = parse_fragment_header(line, len, fragment);
974 if (offset < 0)
975 return -1;
976 oldlines = fragment->oldlines;
977 newlines = fragment->newlines;
978 leading = 0;
979 trailing = 0;
980
981 /* Parse the thing.. */
982 line += len;
983 size -= len;
984 linenr++;
985 added = deleted = 0;
986 for (offset = len;
987 0 < size;
988 offset += len, size -= len, line += len, linenr++) {
989 if (!oldlines && !newlines)
990 break;
991 len = linelen(line, size);
992 if (!len || line[len-1] != '\n')
993 return -1;
994 switch (*line) {
995 default:
996 return -1;
997 case '\n': /* newer GNU diff, an empty context line */
998 case ' ':
999 oldlines--;
1000 newlines--;
1001 if (!deleted && !added)
1002 leading++;
1003 trailing++;
1004 break;
1005 case '-':
1006 deleted++;
1007 oldlines--;
1008 trailing = 0;
1009 break;
1010 case '+':
1011 if (new_whitespace != nowarn_whitespace)
1012 check_whitespace(line, len);
1013 added++;
1014 newlines--;
1015 trailing = 0;
1016 break;
1017
1018 /* We allow "\ No newline at end of file". Depending
1019 * on locale settings when the patch was produced we
1020 * don't know what this line looks like. The only
1021 * thing we do know is that it begins with "\ ".
1022 * Checking for 12 is just for sanity check -- any
1023 * l10n of "\ No newline..." is at least that long.
1024 */
1025 case '\\':
1026 if (len < 12 || memcmp(line, "\\ ", 2))
1027 return -1;
1028 break;
1029 }
1030 }
1031 if (oldlines || newlines)
1032 return -1;
1033 fragment->leading = leading;
1034 fragment->trailing = trailing;
1035
1036 /* If a fragment ends with an incomplete line, we failed to include
1037 * it in the above loop because we hit oldlines == newlines == 0
1038 * before seeing it.
1039 */
1040 if (12 < size && !memcmp(line, "\\ ", 2))
1041 offset += linelen(line, size);
1042
1043 patch->lines_added += added;
1044 patch->lines_deleted += deleted;
1045
1046 if (0 < patch->is_new && oldlines)
1047 return error("new file depends on old contents");
1048 if (0 < patch->is_delete && newlines)
1049 return error("deleted file still has contents");
1050 return offset;
1051 }
1052
1053 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1054 {
1055 unsigned long offset = 0;
1056 unsigned long oldlines = 0, newlines = 0, context = 0;
1057 struct fragment **fragp = &patch->fragments;
1058
1059 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1060 struct fragment *fragment;
1061 int len;
1062
1063 fragment = xcalloc(1, sizeof(*fragment));
1064 len = parse_fragment(line, size, patch, fragment);
1065 if (len <= 0)
1066 die("corrupt patch at line %d", linenr);
1067 fragment->patch = line;
1068 fragment->size = len;
1069 oldlines += fragment->oldlines;
1070 newlines += fragment->newlines;
1071 context += fragment->leading + fragment->trailing;
1072
1073 *fragp = fragment;
1074 fragp = &fragment->next;
1075
1076 offset += len;
1077 line += len;
1078 size -= len;
1079 }
1080
1081 /*
1082 * If something was removed (i.e. we have old-lines) it cannot
1083 * be creation, and if something was added it cannot be
1084 * deletion. However, the reverse is not true; --unified=0
1085 * patches that only add are not necessarily creation even
1086 * though they do not have any old lines, and ones that only
1087 * delete are not necessarily deletion.
1088 *
1089 * Unfortunately, a real creation/deletion patch do _not_ have
1090 * any context line by definition, so we cannot safely tell it
1091 * apart with --unified=0 insanity. At least if the patch has
1092 * more than one hunk it is not creation or deletion.
1093 */
1094 if (patch->is_new < 0 &&
1095 (oldlines || (patch->fragments && patch->fragments->next)))
1096 patch->is_new = 0;
1097 if (patch->is_delete < 0 &&
1098 (newlines || (patch->fragments && patch->fragments->next)))
1099 patch->is_delete = 0;
1100 if (!unidiff_zero || context) {
1101 /* If the user says the patch is not generated with
1102 * --unified=0, or if we have seen context lines,
1103 * then not having oldlines means the patch is creation,
1104 * and not having newlines means the patch is deletion.
1105 */
1106 if (patch->is_new < 0 && !oldlines) {
1107 patch->is_new = 1;
1108 patch->old_name = NULL;
1109 }
1110 if (patch->is_delete < 0 && !newlines) {
1111 patch->is_delete = 1;
1112 patch->new_name = NULL;
1113 }
1114 }
1115
1116 if (0 < patch->is_new && oldlines)
1117 die("new file %s depends on old contents", patch->new_name);
1118 if (0 < patch->is_delete && newlines)
1119 die("deleted file %s still has contents", patch->old_name);
1120 if (!patch->is_delete && !newlines && context)
1121 fprintf(stderr, "** warning: file %s becomes empty but "
1122 "is not deleted\n", patch->new_name);
1123
1124 return offset;
1125 }
1126
1127 static inline int metadata_changes(struct patch *patch)
1128 {
1129 return patch->is_rename > 0 ||
1130 patch->is_copy > 0 ||
1131 patch->is_new > 0 ||
1132 patch->is_delete ||
1133 (patch->old_mode && patch->new_mode &&
1134 patch->old_mode != patch->new_mode);
1135 }
1136
1137 static char *inflate_it(const void *data, unsigned long size,
1138 unsigned long inflated_size)
1139 {
1140 z_stream stream;
1141 void *out;
1142 int st;
1143
1144 memset(&stream, 0, sizeof(stream));
1145
1146 stream.next_in = (unsigned char *)data;
1147 stream.avail_in = size;
1148 stream.next_out = out = xmalloc(inflated_size);
1149 stream.avail_out = inflated_size;
1150 inflateInit(&stream);
1151 st = inflate(&stream, Z_FINISH);
1152 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1153 free(out);
1154 return NULL;
1155 }
1156 return out;
1157 }
1158
1159 static struct fragment *parse_binary_hunk(char **buf_p,
1160 unsigned long *sz_p,
1161 int *status_p,
1162 int *used_p)
1163 {
1164 /* Expect a line that begins with binary patch method ("literal"
1165 * or "delta"), followed by the length of data before deflating.
1166 * a sequence of 'length-byte' followed by base-85 encoded data
1167 * should follow, terminated by a newline.
1168 *
1169 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1170 * and we would limit the patch line to 66 characters,
1171 * so one line can fit up to 13 groups that would decode
1172 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1173 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1174 */
1175 int llen, used;
1176 unsigned long size = *sz_p;
1177 char *buffer = *buf_p;
1178 int patch_method;
1179 unsigned long origlen;
1180 char *data = NULL;
1181 int hunk_size = 0;
1182 struct fragment *frag;
1183
1184 llen = linelen(buffer, size);
1185 used = llen;
1186
1187 *status_p = 0;
1188
1189 if (!prefixcmp(buffer, "delta ")) {
1190 patch_method = BINARY_DELTA_DEFLATED;
1191 origlen = strtoul(buffer + 6, NULL, 10);
1192 }
1193 else if (!prefixcmp(buffer, "literal ")) {
1194 patch_method = BINARY_LITERAL_DEFLATED;
1195 origlen = strtoul(buffer + 8, NULL, 10);
1196 }
1197 else
1198 return NULL;
1199
1200 linenr++;
1201 buffer += llen;
1202 while (1) {
1203 int byte_length, max_byte_length, newsize;
1204 llen = linelen(buffer, size);
1205 used += llen;
1206 linenr++;
1207 if (llen == 1) {
1208 /* consume the blank line */
1209 buffer++;
1210 size--;
1211 break;
1212 }
1213 /* Minimum line is "A00000\n" which is 7-byte long,
1214 * and the line length must be multiple of 5 plus 2.
1215 */
1216 if ((llen < 7) || (llen-2) % 5)
1217 goto corrupt;
1218 max_byte_length = (llen - 2) / 5 * 4;
1219 byte_length = *buffer;
1220 if ('A' <= byte_length && byte_length <= 'Z')
1221 byte_length = byte_length - 'A' + 1;
1222 else if ('a' <= byte_length && byte_length <= 'z')
1223 byte_length = byte_length - 'a' + 27;
1224 else
1225 goto corrupt;
1226 /* if the input length was not multiple of 4, we would
1227 * have filler at the end but the filler should never
1228 * exceed 3 bytes
1229 */
1230 if (max_byte_length < byte_length ||
1231 byte_length <= max_byte_length - 4)
1232 goto corrupt;
1233 newsize = hunk_size + byte_length;
1234 data = xrealloc(data, newsize);
1235 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1236 goto corrupt;
1237 hunk_size = newsize;
1238 buffer += llen;
1239 size -= llen;
1240 }
1241
1242 frag = xcalloc(1, sizeof(*frag));
1243 frag->patch = inflate_it(data, hunk_size, origlen);
1244 if (!frag->patch)
1245 goto corrupt;
1246 free(data);
1247 frag->size = origlen;
1248 *buf_p = buffer;
1249 *sz_p = size;
1250 *used_p = used;
1251 frag->binary_patch_method = patch_method;
1252 return frag;
1253
1254 corrupt:
1255 free(data);
1256 *status_p = -1;
1257 error("corrupt binary patch at line %d: %.*s",
1258 linenr-1, llen-1, buffer);
1259 return NULL;
1260 }
1261
1262 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1263 {
1264 /* We have read "GIT binary patch\n"; what follows is a line
1265 * that says the patch method (currently, either "literal" or
1266 * "delta") and the length of data before deflating; a
1267 * sequence of 'length-byte' followed by base-85 encoded data
1268 * follows.
1269 *
1270 * When a binary patch is reversible, there is another binary
1271 * hunk in the same format, starting with patch method (either
1272 * "literal" or "delta") with the length of data, and a sequence
1273 * of length-byte + base-85 encoded data, terminated with another
1274 * empty line. This data, when applied to the postimage, produces
1275 * the preimage.
1276 */
1277 struct fragment *forward;
1278 struct fragment *reverse;
1279 int status;
1280 int used, used_1;
1281
1282 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1283 if (!forward && !status)
1284 /* there has to be one hunk (forward hunk) */
1285 return error("unrecognized binary patch at line %d", linenr-1);
1286 if (status)
1287 /* otherwise we already gave an error message */
1288 return status;
1289
1290 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1291 if (reverse)
1292 used += used_1;
1293 else if (status) {
1294 /* not having reverse hunk is not an error, but having
1295 * a corrupt reverse hunk is.
1296 */
1297 free((void*) forward->patch);
1298 free(forward);
1299 return status;
1300 }
1301 forward->next = reverse;
1302 patch->fragments = forward;
1303 patch->is_binary = 1;
1304 return used;
1305 }
1306
1307 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1308 {
1309 int hdrsize, patchsize;
1310 int offset = find_header(buffer, size, &hdrsize, patch);
1311
1312 if (offset < 0)
1313 return offset;
1314
1315 patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1316
1317 if (!patchsize) {
1318 static const char *binhdr[] = {
1319 "Binary files ",
1320 "Files ",
1321 NULL,
1322 };
1323 static const char git_binary[] = "GIT binary patch\n";
1324 int i;
1325 int hd = hdrsize + offset;
1326 unsigned long llen = linelen(buffer + hd, size - hd);
1327
1328 if (llen == sizeof(git_binary) - 1 &&
1329 !memcmp(git_binary, buffer + hd, llen)) {
1330 int used;
1331 linenr++;
1332 used = parse_binary(buffer + hd + llen,
1333 size - hd - llen, patch);
1334 if (used)
1335 patchsize = used + llen;
1336 else
1337 patchsize = 0;
1338 }
1339 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1340 for (i = 0; binhdr[i]; i++) {
1341 int len = strlen(binhdr[i]);
1342 if (len < size - hd &&
1343 !memcmp(binhdr[i], buffer + hd, len)) {
1344 linenr++;
1345 patch->is_binary = 1;
1346 patchsize = llen;
1347 break;
1348 }
1349 }
1350 }
1351
1352 /* Empty patch cannot be applied if it is a text patch
1353 * without metadata change. A binary patch appears
1354 * empty to us here.
1355 */
1356 if ((apply || check) &&
1357 (!patch->is_binary && !metadata_changes(patch)))
1358 die("patch with only garbage at line %d", linenr);
1359 }
1360
1361 return offset + hdrsize + patchsize;
1362 }
1363
1364 #define swap(a,b) myswap((a),(b),sizeof(a))
1365
1366 #define myswap(a, b, size) do { \
1367 unsigned char mytmp[size]; \
1368 memcpy(mytmp, &a, size); \
1369 memcpy(&a, &b, size); \
1370 memcpy(&b, mytmp, size); \
1371 } while (0)
1372
1373 static void reverse_patches(struct patch *p)
1374 {
1375 for (; p; p = p->next) {
1376 struct fragment *frag = p->fragments;
1377
1378 swap(p->new_name, p->old_name);
1379 swap(p->new_mode, p->old_mode);
1380 swap(p->is_new, p->is_delete);
1381 swap(p->lines_added, p->lines_deleted);
1382 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1383
1384 for (; frag; frag = frag->next) {
1385 swap(frag->newpos, frag->oldpos);
1386 swap(frag->newlines, frag->oldlines);
1387 }
1388 }
1389 }
1390
1391 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1392 static const char minuses[]= "----------------------------------------------------------------------";
1393
1394 static void show_stats(struct patch *patch)
1395 {
1396 const char *prefix = "";
1397 char *name = patch->new_name;
1398 char *qname = NULL;
1399 int len, max, add, del, total;
1400
1401 if (!name)
1402 name = patch->old_name;
1403
1404 if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1405 qname = xmalloc(len + 1);
1406 quote_c_style(name, qname, NULL, 0);
1407 name = qname;
1408 }
1409
1410 /*
1411 * "scale" the filename
1412 */
1413 len = strlen(name);
1414 max = max_len;
1415 if (max > 50)
1416 max = 50;
1417 if (len > max) {
1418 char *slash;
1419 prefix = "...";
1420 max -= 3;
1421 name += len - max;
1422 slash = strchr(name, '/');
1423 if (slash)
1424 name = slash;
1425 }
1426 len = max;
1427
1428 /*
1429 * scale the add/delete
1430 */
1431 max = max_change;
1432 if (max + len > 70)
1433 max = 70 - len;
1434
1435 add = patch->lines_added;
1436 del = patch->lines_deleted;
1437 total = add + del;
1438
1439 if (max_change > 0) {
1440 total = (total * max + max_change / 2) / max_change;
1441 add = (add * max + max_change / 2) / max_change;
1442 del = total - add;
1443 }
1444 if (patch->is_binary)
1445 printf(" %s%-*s | Bin\n", prefix, len, name);
1446 else
1447 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1448 len, name, patch->lines_added + patch->lines_deleted,
1449 add, pluses, del, minuses);
1450 free(qname);
1451 }
1452
1453 static int read_old_data(struct stat *st, const char *path, char **buf_p, unsigned long *alloc_p, unsigned long *size_p)
1454 {
1455 int fd;
1456 unsigned long got;
1457 unsigned long nsize;
1458 char *nbuf;
1459 unsigned long size = *size_p;
1460 char *buf = *buf_p;
1461
1462 switch (st->st_mode & S_IFMT) {
1463 case S_IFLNK:
1464 return readlink(path, buf, size) != size;
1465 case S_IFREG:
1466 fd = open(path, O_RDONLY);
1467 if (fd < 0)
1468 return error("unable to open %s", path);
1469 got = 0;
1470 for (;;) {
1471 ssize_t ret = xread(fd, buf + got, size - got);
1472 if (ret <= 0)
1473 break;
1474 got += ret;
1475 }
1476 close(fd);
1477 nsize = got;
1478 nbuf = convert_to_git(path, buf, &nsize);
1479 if (nbuf) {
1480 free(buf);
1481 *buf_p = nbuf;
1482 *alloc_p = nsize;
1483 *size_p = nsize;
1484 }
1485 return got != size;
1486 default:
1487 return -1;
1488 }
1489 }
1490
1491 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1492 {
1493 int i;
1494 unsigned long start, backwards, forwards;
1495
1496 if (fragsize > size)
1497 return -1;
1498
1499 start = 0;
1500 if (line > 1) {
1501 unsigned long offset = 0;
1502 i = line-1;
1503 while (offset + fragsize <= size) {
1504 if (buf[offset++] == '\n') {
1505 start = offset;
1506 if (!--i)
1507 break;
1508 }
1509 }
1510 }
1511
1512 /* Exact line number? */
1513 if (!memcmp(buf + start, fragment, fragsize))
1514 return start;
1515
1516 /*
1517 * There's probably some smart way to do this, but I'll leave
1518 * that to the smart and beautiful people. I'm simple and stupid.
1519 */
1520 backwards = start;
1521 forwards = start;
1522 for (i = 0; ; i++) {
1523 unsigned long try;
1524 int n;
1525
1526 /* "backward" */
1527 if (i & 1) {
1528 if (!backwards) {
1529 if (forwards + fragsize > size)
1530 break;
1531 continue;
1532 }
1533 do {
1534 --backwards;
1535 } while (backwards && buf[backwards-1] != '\n');
1536 try = backwards;
1537 } else {
1538 while (forwards + fragsize <= size) {
1539 if (buf[forwards++] == '\n')
1540 break;
1541 }
1542 try = forwards;
1543 }
1544
1545 if (try + fragsize > size)
1546 continue;
1547 if (memcmp(buf + try, fragment, fragsize))
1548 continue;
1549 n = (i >> 1)+1;
1550 if (i & 1)
1551 n = -n;
1552 *lines = n;
1553 return try;
1554 }
1555
1556 /*
1557 * We should start searching forward and backward.
1558 */
1559 return -1;
1560 }
1561
1562 static void remove_first_line(const char **rbuf, int *rsize)
1563 {
1564 const char *buf = *rbuf;
1565 int size = *rsize;
1566 unsigned long offset;
1567 offset = 0;
1568 while (offset <= size) {
1569 if (buf[offset++] == '\n')
1570 break;
1571 }
1572 *rsize = size - offset;
1573 *rbuf = buf + offset;
1574 }
1575
1576 static void remove_last_line(const char **rbuf, int *rsize)
1577 {
1578 const char *buf = *rbuf;
1579 int size = *rsize;
1580 unsigned long offset;
1581 offset = size - 1;
1582 while (offset > 0) {
1583 if (buf[--offset] == '\n')
1584 break;
1585 }
1586 *rsize = offset + 1;
1587 }
1588
1589 struct buffer_desc {
1590 char *buffer;
1591 unsigned long size;
1592 unsigned long alloc;
1593 };
1594
1595 static int apply_line(char *output, const char *patch, int plen)
1596 {
1597 /* plen is number of bytes to be copied from patch,
1598 * starting at patch+1 (patch[0] is '+'). Typically
1599 * patch[plen] is '\n', unless this is the incomplete
1600 * last line.
1601 */
1602 int i;
1603 int add_nl_to_tail = 0;
1604 int fixed = 0;
1605 int last_tab_in_indent = -1;
1606 int last_space_in_indent = -1;
1607 int need_fix_leading_space = 0;
1608 char *buf;
1609
1610 if ((new_whitespace != strip_whitespace) || !whitespace_error ||
1611 *patch != '+') {
1612 memcpy(output, patch + 1, plen);
1613 return plen;
1614 }
1615
1616 if (1 < plen && isspace(patch[plen-1])) {
1617 if (patch[plen] == '\n')
1618 add_nl_to_tail = 1;
1619 plen--;
1620 while (0 < plen && isspace(patch[plen]))
1621 plen--;
1622 fixed = 1;
1623 }
1624
1625 for (i = 1; i < plen; i++) {
1626 char ch = patch[i];
1627 if (ch == '\t') {
1628 last_tab_in_indent = i;
1629 if (0 <= last_space_in_indent)
1630 need_fix_leading_space = 1;
1631 }
1632 else if (ch == ' ')
1633 last_space_in_indent = i;
1634 else
1635 break;
1636 }
1637
1638 buf = output;
1639 if (need_fix_leading_space) {
1640 /* between patch[1..last_tab_in_indent] strip the
1641 * funny spaces, updating them to tab as needed.
1642 */
1643 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1644 char ch = patch[i];
1645 if (ch != ' ')
1646 *output++ = ch;
1647 else if ((i % 8) == 0)
1648 *output++ = '\t';
1649 }
1650 fixed = 1;
1651 i = last_tab_in_indent;
1652 }
1653 else
1654 i = 1;
1655
1656 memcpy(output, patch + i, plen);
1657 if (add_nl_to_tail)
1658 output[plen++] = '\n';
1659 if (fixed)
1660 applied_after_stripping++;
1661 return output + plen - buf;
1662 }
1663
1664 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag, int inaccurate_eof)
1665 {
1666 int match_beginning, match_end;
1667 char *buf = desc->buffer;
1668 const char *patch = frag->patch;
1669 int offset, size = frag->size;
1670 char *old = xmalloc(size);
1671 char *new = xmalloc(size);
1672 const char *oldlines, *newlines;
1673 int oldsize = 0, newsize = 0;
1674 unsigned long leading, trailing;
1675 int pos, lines;
1676
1677 while (size > 0) {
1678 char first;
1679 int len = linelen(patch, size);
1680 int plen;
1681
1682 if (!len)
1683 break;
1684
1685 /*
1686 * "plen" is how much of the line we should use for
1687 * the actual patch data. Normally we just remove the
1688 * first character on the line, but if the line is
1689 * followed by "\ No newline", then we also remove the
1690 * last one (which is the newline, of course).
1691 */
1692 plen = len-1;
1693 if (len < size && patch[len] == '\\')
1694 plen--;
1695 first = *patch;
1696 if (apply_in_reverse) {
1697 if (first == '-')
1698 first = '+';
1699 else if (first == '+')
1700 first = '-';
1701 }
1702 switch (first) {
1703 case '\n':
1704 /* Newer GNU diff, empty context line */
1705 if (plen < 0)
1706 /* ... followed by '\No newline'; nothing */
1707 break;
1708 old[oldsize++] = '\n';
1709 new[newsize++] = '\n';
1710 break;
1711 case ' ':
1712 case '-':
1713 memcpy(old + oldsize, patch + 1, plen);
1714 oldsize += plen;
1715 if (first == '-')
1716 break;
1717 /* Fall-through for ' ' */
1718 case '+':
1719 if (first != '+' || !no_add)
1720 newsize += apply_line(new + newsize, patch,
1721 plen);
1722 break;
1723 case '@': case '\\':
1724 /* Ignore it, we already handled it */
1725 break;
1726 default:
1727 if (apply_verbosely)
1728 error("invalid start of line: '%c'", first);
1729 return -1;
1730 }
1731 patch += len;
1732 size -= len;
1733 }
1734
1735 if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1736 newsize > 0 && new[newsize - 1] == '\n') {
1737 oldsize--;
1738 newsize--;
1739 }
1740
1741 oldlines = old;
1742 newlines = new;
1743 leading = frag->leading;
1744 trailing = frag->trailing;
1745
1746 /*
1747 * If we don't have any leading/trailing data in the patch,
1748 * we want it to match at the beginning/end of the file.
1749 *
1750 * But that would break if the patch is generated with
1751 * --unified=0; sane people wouldn't do that to cause us
1752 * trouble, but we try to please not so sane ones as well.
1753 */
1754 if (unidiff_zero) {
1755 match_beginning = (!leading && !frag->oldpos);
1756 match_end = 0;
1757 }
1758 else {
1759 match_beginning = !leading && (frag->oldpos == 1);
1760 match_end = !trailing;
1761 }
1762
1763 lines = 0;
1764 pos = frag->newpos;
1765 for (;;) {
1766 offset = find_offset(buf, desc->size,
1767 oldlines, oldsize, pos, &lines);
1768 if (match_end && offset + oldsize != desc->size)
1769 offset = -1;
1770 if (match_beginning && offset)
1771 offset = -1;
1772 if (offset >= 0) {
1773 int diff = newsize - oldsize;
1774 unsigned long size = desc->size + diff;
1775 unsigned long alloc = desc->alloc;
1776
1777 /* Warn if it was necessary to reduce the number
1778 * of context lines.
1779 */
1780 if ((leading != frag->leading) ||
1781 (trailing != frag->trailing))
1782 fprintf(stderr, "Context reduced to (%ld/%ld)"
1783 " to apply fragment at %d\n",
1784 leading, trailing, pos + lines);
1785
1786 if (size > alloc) {
1787 alloc = size + 8192;
1788 desc->alloc = alloc;
1789 buf = xrealloc(buf, alloc);
1790 desc->buffer = buf;
1791 }
1792 desc->size = size;
1793 memmove(buf + offset + newsize,
1794 buf + offset + oldsize,
1795 size - offset - newsize);
1796 memcpy(buf + offset, newlines, newsize);
1797 offset = 0;
1798
1799 break;
1800 }
1801
1802 /* Am I at my context limits? */
1803 if ((leading <= p_context) && (trailing <= p_context))
1804 break;
1805 if (match_beginning || match_end) {
1806 match_beginning = match_end = 0;
1807 continue;
1808 }
1809 /* Reduce the number of context lines
1810 * Reduce both leading and trailing if they are equal
1811 * otherwise just reduce the larger context.
1812 */
1813 if (leading >= trailing) {
1814 remove_first_line(&oldlines, &oldsize);
1815 remove_first_line(&newlines, &newsize);
1816 pos--;
1817 leading--;
1818 }
1819 if (trailing > leading) {
1820 remove_last_line(&oldlines, &oldsize);
1821 remove_last_line(&newlines, &newsize);
1822 trailing--;
1823 }
1824 }
1825
1826 if (offset && apply_verbosely)
1827 error("while searching for:\n%.*s", oldsize, oldlines);
1828
1829 free(old);
1830 free(new);
1831 return offset;
1832 }
1833
1834 static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1835 {
1836 unsigned long dst_size;
1837 struct fragment *fragment = patch->fragments;
1838 void *data;
1839 void *result;
1840
1841 /* Binary patch is irreversible without the optional second hunk */
1842 if (apply_in_reverse) {
1843 if (!fragment->next)
1844 return error("cannot reverse-apply a binary patch "
1845 "without the reverse hunk to '%s'",
1846 patch->new_name
1847 ? patch->new_name : patch->old_name);
1848 fragment = fragment->next;
1849 }
1850 data = (void*) fragment->patch;
1851 switch (fragment->binary_patch_method) {
1852 case BINARY_DELTA_DEFLATED:
1853 result = patch_delta(desc->buffer, desc->size,
1854 data,
1855 fragment->size,
1856 &dst_size);
1857 free(desc->buffer);
1858 desc->buffer = result;
1859 break;
1860 case BINARY_LITERAL_DEFLATED:
1861 free(desc->buffer);
1862 desc->buffer = data;
1863 dst_size = fragment->size;
1864 break;
1865 }
1866 if (!desc->buffer)
1867 return -1;
1868 desc->size = desc->alloc = dst_size;
1869 return 0;
1870 }
1871
1872 static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1873 {
1874 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1875 unsigned char sha1[20];
1876
1877 /* For safety, we require patch index line to contain
1878 * full 40-byte textual SHA1 for old and new, at least for now.
1879 */
1880 if (strlen(patch->old_sha1_prefix) != 40 ||
1881 strlen(patch->new_sha1_prefix) != 40 ||
1882 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1883 get_sha1_hex(patch->new_sha1_prefix, sha1))
1884 return error("cannot apply binary patch to '%s' "
1885 "without full index line", name);
1886
1887 if (patch->old_name) {
1888 /* See if the old one matches what the patch
1889 * applies to.
1890 */
1891 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1892 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1893 return error("the patch applies to '%s' (%s), "
1894 "which does not match the "
1895 "current contents.",
1896 name, sha1_to_hex(sha1));
1897 }
1898 else {
1899 /* Otherwise, the old one must be empty. */
1900 if (desc->size)
1901 return error("the patch applies to an empty "
1902 "'%s' but it is not empty", name);
1903 }
1904
1905 get_sha1_hex(patch->new_sha1_prefix, sha1);
1906 if (is_null_sha1(sha1)) {
1907 free(desc->buffer);
1908 desc->alloc = desc->size = 0;
1909 desc->buffer = NULL;
1910 return 0; /* deletion patch */
1911 }
1912
1913 if (has_sha1_file(sha1)) {
1914 /* We already have the postimage */
1915 enum object_type type;
1916 unsigned long size;
1917
1918 free(desc->buffer);
1919 desc->buffer = read_sha1_file(sha1, &type, &size);
1920 if (!desc->buffer)
1921 return error("the necessary postimage %s for "
1922 "'%s' cannot be read",
1923 patch->new_sha1_prefix, name);
1924 desc->alloc = desc->size = size;
1925 }
1926 else {
1927 /* We have verified desc matches the preimage;
1928 * apply the patch data to it, which is stored
1929 * in the patch->fragments->{patch,size}.
1930 */
1931 if (apply_binary_fragment(desc, patch))
1932 return error("binary patch does not apply to '%s'",
1933 name);
1934
1935 /* verify that the result matches */
1936 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1937 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1938 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)", name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1939 }
1940
1941 return 0;
1942 }
1943
1944 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1945 {
1946 struct fragment *frag = patch->fragments;
1947 const char *name = patch->old_name ? patch->old_name : patch->new_name;
1948
1949 if (patch->is_binary)
1950 return apply_binary(desc, patch);
1951
1952 while (frag) {
1953 if (apply_one_fragment(desc, frag, patch->inaccurate_eof)) {
1954 error("patch failed: %s:%ld", name, frag->oldpos);
1955 if (!apply_with_reject)
1956 return -1;
1957 frag->rejected = 1;
1958 }
1959 frag = frag->next;
1960 }
1961 return 0;
1962 }
1963
1964 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1965 {
1966 char *buf;
1967 unsigned long size, alloc;
1968 struct buffer_desc desc;
1969
1970 size = 0;
1971 alloc = 0;
1972 buf = NULL;
1973 if (cached) {
1974 if (ce) {
1975 enum object_type type;
1976 buf = read_sha1_file(ce->sha1, &type, &size);
1977 if (!buf)
1978 return error("read of %s failed",
1979 patch->old_name);
1980 alloc = size;
1981 }
1982 }
1983 else if (patch->old_name) {
1984 size = xsize_t(st->st_size);
1985 alloc = size + 8192;
1986 buf = xmalloc(alloc);
1987 if (read_old_data(st, patch->old_name, &buf, &alloc, &size))
1988 return error("read of %s failed", patch->old_name);
1989 }
1990
1991 desc.size = size;
1992 desc.alloc = alloc;
1993 desc.buffer = buf;
1994
1995 if (apply_fragments(&desc, patch) < 0)
1996 return -1; /* note with --reject this succeeds. */
1997
1998 /* NUL terminate the result */
1999 if (desc.alloc <= desc.size)
2000 desc.buffer = xrealloc(desc.buffer, desc.size + 1);
2001 desc.buffer[desc.size] = 0;
2002
2003 patch->result = desc.buffer;
2004 patch->resultsize = desc.size;
2005
2006 if (0 < patch->is_delete && patch->resultsize)
2007 return error("removal patch leaves file contents");
2008
2009 return 0;
2010 }
2011
2012 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2013 {
2014 struct stat nst;
2015 if (!lstat(new_name, &nst)) {
2016 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2017 return 0;
2018 /*
2019 * A leading component of new_name might be a symlink
2020 * that is going to be removed with this patch, but
2021 * still pointing at somewhere that has the path.
2022 * In such a case, path "new_name" does not exist as
2023 * far as git is concerned.
2024 */
2025 if (has_symlink_leading_path(new_name, NULL))
2026 return 0;
2027
2028 return error("%s: already exists in working directory", new_name);
2029 }
2030 else if ((errno != ENOENT) && (errno != ENOTDIR))
2031 return error("%s: %s", new_name, strerror(errno));
2032 return 0;
2033 }
2034
2035 static int check_patch(struct patch *patch, struct patch *prev_patch)
2036 {
2037 struct stat st;
2038 const char *old_name = patch->old_name;
2039 const char *new_name = patch->new_name;
2040 const char *name = old_name ? old_name : new_name;
2041 struct cache_entry *ce = NULL;
2042 int ok_if_exists;
2043
2044 patch->rejected = 1; /* we will drop this after we succeed */
2045 if (old_name) {
2046 int changed = 0;
2047 int stat_ret = 0;
2048 unsigned st_mode = 0;
2049
2050 if (!cached)
2051 stat_ret = lstat(old_name, &st);
2052 if (check_index) {
2053 int pos = cache_name_pos(old_name, strlen(old_name));
2054 if (pos < 0)
2055 return error("%s: does not exist in index",
2056 old_name);
2057 ce = active_cache[pos];
2058 if (stat_ret < 0) {
2059 struct checkout costate;
2060 if (errno != ENOENT)
2061 return error("%s: %s", old_name,
2062 strerror(errno));
2063 /* checkout */
2064 costate.base_dir = "";
2065 costate.base_dir_len = 0;
2066 costate.force = 0;
2067 costate.quiet = 0;
2068 costate.not_new = 0;
2069 costate.refresh_cache = 1;
2070 if (checkout_entry(ce,
2071 &costate,
2072 NULL) ||
2073 lstat(old_name, &st))
2074 return -1;
2075 }
2076 if (!cached)
2077 changed = ce_match_stat(ce, &st, 1);
2078 if (changed)
2079 return error("%s: does not match index",
2080 old_name);
2081 if (cached)
2082 st_mode = ntohl(ce->ce_mode);
2083 }
2084 else if (stat_ret < 0)
2085 return error("%s: %s", old_name, strerror(errno));
2086
2087 if (!cached)
2088 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2089
2090 if (patch->is_new < 0)
2091 patch->is_new = 0;
2092 if (!patch->old_mode)
2093 patch->old_mode = st_mode;
2094 if ((st_mode ^ patch->old_mode) & S_IFMT)
2095 return error("%s: wrong type", old_name);
2096 if (st_mode != patch->old_mode)
2097 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2098 old_name, st_mode, patch->old_mode);
2099 }
2100
2101 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2102 !strcmp(prev_patch->old_name, new_name))
2103 /* A type-change diff is always split into a patch to
2104 * delete old, immediately followed by a patch to
2105 * create new (see diff.c::run_diff()); in such a case
2106 * it is Ok that the entry to be deleted by the
2107 * previous patch is still in the working tree and in
2108 * the index.
2109 */
2110 ok_if_exists = 1;
2111 else
2112 ok_if_exists = 0;
2113
2114 if (new_name &&
2115 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2116 if (check_index &&
2117 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2118 !ok_if_exists)
2119 return error("%s: already exists in index", new_name);
2120 if (!cached) {
2121 int err = check_to_create_blob(new_name, ok_if_exists);
2122 if (err)
2123 return err;
2124 }
2125 if (!patch->new_mode) {
2126 if (0 < patch->is_new)
2127 patch->new_mode = S_IFREG | 0644;
2128 else
2129 patch->new_mode = patch->old_mode;
2130 }
2131 }
2132
2133 if (new_name && old_name) {
2134 int same = !strcmp(old_name, new_name);
2135 if (!patch->new_mode)
2136 patch->new_mode = patch->old_mode;
2137 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2138 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2139 patch->new_mode, new_name, patch->old_mode,
2140 same ? "" : " of ", same ? "" : old_name);
2141 }
2142
2143 if (apply_data(patch, &st, ce) < 0)
2144 return error("%s: patch does not apply", name);
2145 patch->rejected = 0;
2146 return 0;
2147 }
2148
2149 static int check_patch_list(struct patch *patch)
2150 {
2151 struct patch *prev_patch = NULL;
2152 int err = 0;
2153
2154 for (prev_patch = NULL; patch ; patch = patch->next) {
2155 if (apply_verbosely)
2156 say_patch_name(stderr,
2157 "Checking patch ", patch, "...\n");
2158 err |= check_patch(patch, prev_patch);
2159 prev_patch = patch;
2160 }
2161 return err;
2162 }
2163
2164 static void show_index_list(struct patch *list)
2165 {
2166 struct patch *patch;
2167
2168 /* Once we start supporting the reverse patch, it may be
2169 * worth showing the new sha1 prefix, but until then...
2170 */
2171 for (patch = list; patch; patch = patch->next) {
2172 const unsigned char *sha1_ptr;
2173 unsigned char sha1[20];
2174 const char *name;
2175
2176 name = patch->old_name ? patch->old_name : patch->new_name;
2177 if (0 < patch->is_new)
2178 sha1_ptr = null_sha1;
2179 else if (get_sha1(patch->old_sha1_prefix, sha1))
2180 die("sha1 information is lacking or useless (%s).",
2181 name);
2182 else
2183 sha1_ptr = sha1;
2184
2185 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2186 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2187 quote_c_style(name, NULL, stdout, 0);
2188 else
2189 fputs(name, stdout);
2190 putchar(line_termination);
2191 }
2192 }
2193
2194 static void stat_patch_list(struct patch *patch)
2195 {
2196 int files, adds, dels;
2197
2198 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2199 files++;
2200 adds += patch->lines_added;
2201 dels += patch->lines_deleted;
2202 show_stats(patch);
2203 }
2204
2205 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2206 }
2207
2208 static void numstat_patch_list(struct patch *patch)
2209 {
2210 for ( ; patch; patch = patch->next) {
2211 const char *name;
2212 name = patch->new_name ? patch->new_name : patch->old_name;
2213 if (patch->is_binary)
2214 printf("-\t-\t");
2215 else
2216 printf("%d\t%d\t",
2217 patch->lines_added, patch->lines_deleted);
2218 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2219 quote_c_style(name, NULL, stdout, 0);
2220 else
2221 fputs(name, stdout);
2222 putchar(line_termination);
2223 }
2224 }
2225
2226 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2227 {
2228 if (mode)
2229 printf(" %s mode %06o %s\n", newdelete, mode, name);
2230 else
2231 printf(" %s %s\n", newdelete, name);
2232 }
2233
2234 static void show_mode_change(struct patch *p, int show_name)
2235 {
2236 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2237 if (show_name)
2238 printf(" mode change %06o => %06o %s\n",
2239 p->old_mode, p->new_mode, p->new_name);
2240 else
2241 printf(" mode change %06o => %06o\n",
2242 p->old_mode, p->new_mode);
2243 }
2244 }
2245
2246 static void show_rename_copy(struct patch *p)
2247 {
2248 const char *renamecopy = p->is_rename ? "rename" : "copy";
2249 const char *old, *new;
2250
2251 /* Find common prefix */
2252 old = p->old_name;
2253 new = p->new_name;
2254 while (1) {
2255 const char *slash_old, *slash_new;
2256 slash_old = strchr(old, '/');
2257 slash_new = strchr(new, '/');
2258 if (!slash_old ||
2259 !slash_new ||
2260 slash_old - old != slash_new - new ||
2261 memcmp(old, new, slash_new - new))
2262 break;
2263 old = slash_old + 1;
2264 new = slash_new + 1;
2265 }
2266 /* p->old_name thru old is the common prefix, and old and new
2267 * through the end of names are renames
2268 */
2269 if (old != p->old_name)
2270 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2271 (int)(old - p->old_name), p->old_name,
2272 old, new, p->score);
2273 else
2274 printf(" %s %s => %s (%d%%)\n", renamecopy,
2275 p->old_name, p->new_name, p->score);
2276 show_mode_change(p, 0);
2277 }
2278
2279 static void summary_patch_list(struct patch *patch)
2280 {
2281 struct patch *p;
2282
2283 for (p = patch; p; p = p->next) {
2284 if (p->is_new)
2285 show_file_mode_name("create", p->new_mode, p->new_name);
2286 else if (p->is_delete)
2287 show_file_mode_name("delete", p->old_mode, p->old_name);
2288 else {
2289 if (p->is_rename || p->is_copy)
2290 show_rename_copy(p);
2291 else {
2292 if (p->score) {
2293 printf(" rewrite %s (%d%%)\n",
2294 p->new_name, p->score);
2295 show_mode_change(p, 0);
2296 }
2297 else
2298 show_mode_change(p, 1);
2299 }
2300 }
2301 }
2302 }
2303
2304 static void patch_stats(struct patch *patch)
2305 {
2306 int lines = patch->lines_added + patch->lines_deleted;
2307
2308 if (lines > max_change)
2309 max_change = lines;
2310 if (patch->old_name) {
2311 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2312 if (!len)
2313 len = strlen(patch->old_name);
2314 if (len > max_len)
2315 max_len = len;
2316 }
2317 if (patch->new_name) {
2318 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2319 if (!len)
2320 len = strlen(patch->new_name);
2321 if (len > max_len)
2322 max_len = len;
2323 }
2324 }
2325
2326 static void remove_file(struct patch *patch, int rmdir_empty)
2327 {
2328 if (update_index) {
2329 if (remove_file_from_cache(patch->old_name) < 0)
2330 die("unable to remove %s from index", patch->old_name);
2331 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2332 }
2333 if (!cached) {
2334 if (!unlink(patch->old_name) && rmdir_empty) {
2335 char *name = xstrdup(patch->old_name);
2336 char *end = strrchr(name, '/');
2337 while (end) {
2338 *end = 0;
2339 if (rmdir(name))
2340 break;
2341 end = strrchr(name, '/');
2342 }
2343 free(name);
2344 }
2345 }
2346 }
2347
2348 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2349 {
2350 struct stat st;
2351 struct cache_entry *ce;
2352 int namelen = strlen(path);
2353 unsigned ce_size = cache_entry_size(namelen);
2354
2355 if (!update_index)
2356 return;
2357
2358 ce = xcalloc(1, ce_size);
2359 memcpy(ce->name, path, namelen);
2360 ce->ce_mode = create_ce_mode(mode);
2361 ce->ce_flags = htons(namelen);
2362 if (!cached) {
2363 if (lstat(path, &st) < 0)
2364 die("unable to stat newly created file %s", path);
2365 fill_stat_cache_info(ce, &st);
2366 }
2367 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2368 die("unable to create backing store for newly created file %s", path);
2369 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2370 die("unable to add cache entry for %s", path);
2371 }
2372
2373 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2374 {
2375 int fd;
2376 char *nbuf;
2377
2378 if (has_symlinks && S_ISLNK(mode))
2379 /* Although buf:size is counted string, it also is NUL
2380 * terminated.
2381 */
2382 return symlink(buf, path);
2383
2384 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2385 if (fd < 0)
2386 return -1;
2387
2388 nbuf = convert_to_working_tree(path, buf, &size);
2389 if (nbuf)
2390 buf = nbuf;
2391
2392 while (size) {
2393 int written = xwrite(fd, buf, size);
2394 if (written < 0)
2395 die("writing file %s: %s", path, strerror(errno));
2396 if (!written)
2397 die("out of space writing file %s", path);
2398 buf += written;
2399 size -= written;
2400 }
2401 if (close(fd) < 0)
2402 die("closing file %s: %s", path, strerror(errno));
2403 if (nbuf)
2404 free(nbuf);
2405 return 0;
2406 }
2407
2408 /*
2409 * We optimistically assume that the directories exist,
2410 * which is true 99% of the time anyway. If they don't,
2411 * we create them and try again.
2412 */
2413 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2414 {
2415 if (cached)
2416 return;
2417 if (!try_create_file(path, mode, buf, size))
2418 return;
2419
2420 if (errno == ENOENT) {
2421 if (safe_create_leading_directories(path))
2422 return;
2423 if (!try_create_file(path, mode, buf, size))
2424 return;
2425 }
2426
2427 if (errno == EEXIST || errno == EACCES) {
2428 /* We may be trying to create a file where a directory
2429 * used to be.
2430 */
2431 struct stat st;
2432 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2433 errno = EEXIST;
2434 }
2435
2436 if (errno == EEXIST) {
2437 unsigned int nr = getpid();
2438
2439 for (;;) {
2440 const char *newpath;
2441 newpath = mkpath("%s~%u", path, nr);
2442 if (!try_create_file(newpath, mode, buf, size)) {
2443 if (!rename(newpath, path))
2444 return;
2445 unlink(newpath);
2446 break;
2447 }
2448 if (errno != EEXIST)
2449 break;
2450 ++nr;
2451 }
2452 }
2453 die("unable to write file %s mode %o", path, mode);
2454 }
2455
2456 static void create_file(struct patch *patch)
2457 {
2458 char *path = patch->new_name;
2459 unsigned mode = patch->new_mode;
2460 unsigned long size = patch->resultsize;
2461 char *buf = patch->result;
2462
2463 if (!mode)
2464 mode = S_IFREG | 0644;
2465 create_one_file(path, mode, buf, size);
2466 add_index_file(path, mode, buf, size);
2467 cache_tree_invalidate_path(active_cache_tree, path);
2468 }
2469
2470 /* phase zero is to remove, phase one is to create */
2471 static void write_out_one_result(struct patch *patch, int phase)
2472 {
2473 if (patch->is_delete > 0) {
2474 if (phase == 0)
2475 remove_file(patch, 1);
2476 return;
2477 }
2478 if (patch->is_new > 0 || patch->is_copy) {
2479 if (phase == 1)
2480 create_file(patch);
2481 return;
2482 }
2483 /*
2484 * Rename or modification boils down to the same
2485 * thing: remove the old, write the new
2486 */
2487 if (phase == 0)
2488 remove_file(patch, 0);
2489 if (phase == 1)
2490 create_file(patch);
2491 }
2492
2493 static int write_out_one_reject(struct patch *patch)
2494 {
2495 FILE *rej;
2496 char namebuf[PATH_MAX];
2497 struct fragment *frag;
2498 int cnt = 0;
2499
2500 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2501 if (!frag->rejected)
2502 continue;
2503 cnt++;
2504 }
2505
2506 if (!cnt) {
2507 if (apply_verbosely)
2508 say_patch_name(stderr,
2509 "Applied patch ", patch, " cleanly.\n");
2510 return 0;
2511 }
2512
2513 /* This should not happen, because a removal patch that leaves
2514 * contents are marked "rejected" at the patch level.
2515 */
2516 if (!patch->new_name)
2517 die("internal error");
2518
2519 /* Say this even without --verbose */
2520 say_patch_name(stderr, "Applying patch ", patch, " with");
2521 fprintf(stderr, " %d rejects...\n", cnt);
2522
2523 cnt = strlen(patch->new_name);
2524 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2525 cnt = ARRAY_SIZE(namebuf) - 5;
2526 fprintf(stderr,
2527 "warning: truncating .rej filename to %.*s.rej",
2528 cnt - 1, patch->new_name);
2529 }
2530 memcpy(namebuf, patch->new_name, cnt);
2531 memcpy(namebuf + cnt, ".rej", 5);
2532
2533 rej = fopen(namebuf, "w");
2534 if (!rej)
2535 return error("cannot open %s: %s", namebuf, strerror(errno));
2536
2537 /* Normal git tools never deal with .rej, so do not pretend
2538 * this is a git patch by saying --git nor give extended
2539 * headers. While at it, maybe please "kompare" that wants
2540 * the trailing TAB and some garbage at the end of line ;-).
2541 */
2542 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2543 patch->new_name, patch->new_name);
2544 for (cnt = 1, frag = patch->fragments;
2545 frag;
2546 cnt++, frag = frag->next) {
2547 if (!frag->rejected) {
2548 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2549 continue;
2550 }
2551 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2552 fprintf(rej, "%.*s", frag->size, frag->patch);
2553 if (frag->patch[frag->size-1] != '\n')
2554 fputc('\n', rej);
2555 }
2556 fclose(rej);
2557 return -1;
2558 }
2559
2560 static int write_out_results(struct patch *list, int skipped_patch)
2561 {
2562 int phase;
2563 int errs = 0;
2564 struct patch *l;
2565
2566 if (!list && !skipped_patch)
2567 return error("No changes");
2568
2569 for (phase = 0; phase < 2; phase++) {
2570 l = list;
2571 while (l) {
2572 if (l->rejected)
2573 errs = 1;
2574 else {
2575 write_out_one_result(l, phase);
2576 if (phase == 1 && write_out_one_reject(l))
2577 errs = 1;
2578 }
2579 l = l->next;
2580 }
2581 }
2582 return errs;
2583 }
2584
2585 static struct lock_file lock_file;
2586
2587 static struct excludes {
2588 struct excludes *next;
2589 const char *path;
2590 } *excludes;
2591
2592 static int use_patch(struct patch *p)
2593 {
2594 const char *pathname = p->new_name ? p->new_name : p->old_name;
2595 struct excludes *x = excludes;
2596 while (x) {
2597 if (fnmatch(x->path, pathname, 0) == 0)
2598 return 0;
2599 x = x->next;
2600 }
2601 if (0 < prefix_length) {
2602 int pathlen = strlen(pathname);
2603 if (pathlen <= prefix_length ||
2604 memcmp(prefix, pathname, prefix_length))
2605 return 0;
2606 }
2607 return 1;
2608 }
2609
2610 static void prefix_one(char **name)
2611 {
2612 char *old_name = *name;
2613 if (!old_name)
2614 return;
2615 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2616 free(old_name);
2617 }
2618
2619 static void prefix_patches(struct patch *p)
2620 {
2621 if (!prefix || p->is_toplevel_relative)
2622 return;
2623 for ( ; p; p = p->next) {
2624 if (p->new_name == p->old_name) {
2625 char *prefixed = p->new_name;
2626 prefix_one(&prefixed);
2627 p->new_name = p->old_name = prefixed;
2628 }
2629 else {
2630 prefix_one(&p->new_name);
2631 prefix_one(&p->old_name);
2632 }
2633 }
2634 }
2635
2636 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2637 {
2638 unsigned long offset, size;
2639 char *buffer = read_patch_file(fd, &size);
2640 struct patch *list = NULL, **listp = &list;
2641 int skipped_patch = 0;
2642
2643 patch_input_file = filename;
2644 if (!buffer)
2645 return -1;
2646 offset = 0;
2647 while (size > 0) {
2648 struct patch *patch;
2649 int nr;
2650
2651 patch = xcalloc(1, sizeof(*patch));
2652 patch->inaccurate_eof = inaccurate_eof;
2653 nr = parse_chunk(buffer + offset, size, patch);
2654 if (nr < 0)
2655 break;
2656 if (apply_in_reverse)
2657 reverse_patches(patch);
2658 if (prefix)
2659 prefix_patches(patch);
2660 if (use_patch(patch)) {
2661 patch_stats(patch);
2662 *listp = patch;
2663 listp = &patch->next;
2664 }
2665 else {
2666 /* perhaps free it a bit better? */
2667 free(patch);
2668 skipped_patch++;
2669 }
2670 offset += nr;
2671 size -= nr;
2672 }
2673
2674 if (whitespace_error && (new_whitespace == error_on_whitespace))
2675 apply = 0;
2676
2677 update_index = check_index && apply;
2678 if (update_index && newfd < 0)
2679 newfd = hold_locked_index(&lock_file, 1);
2680
2681 if (check_index) {
2682 if (read_cache() < 0)
2683 die("unable to read index file");
2684 }
2685
2686 if ((check || apply) &&
2687 check_patch_list(list) < 0 &&
2688 !apply_with_reject)
2689 exit(1);
2690
2691 if (apply && write_out_results(list, skipped_patch))
2692 exit(1);
2693
2694 if (show_index_info)
2695 show_index_list(list);
2696
2697 if (diffstat)
2698 stat_patch_list(list);
2699
2700 if (numstat)
2701 numstat_patch_list(list);
2702
2703 if (summary)
2704 summary_patch_list(list);
2705
2706 free(buffer);
2707 return 0;
2708 }
2709
2710 static int git_apply_config(const char *var, const char *value)
2711 {
2712 if (!strcmp(var, "apply.whitespace")) {
2713 apply_default_whitespace = xstrdup(value);
2714 return 0;
2715 }
2716 return git_default_config(var, value);
2717 }
2718
2719
2720 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2721 {
2722 int i;
2723 int read_stdin = 1;
2724 int inaccurate_eof = 0;
2725 int errs = 0;
2726 int is_not_gitdir = 0;
2727
2728 const char *whitespace_option = NULL;
2729
2730 prefix = setup_git_directory_gently(&is_not_gitdir);
2731 prefix_length = prefix ? strlen(prefix) : 0;
2732 git_config(git_apply_config);
2733 if (apply_default_whitespace)
2734 parse_whitespace_option(apply_default_whitespace);
2735
2736 for (i = 1; i < argc; i++) {
2737 const char *arg = argv[i];
2738 char *end;
2739 int fd;
2740
2741 if (!strcmp(arg, "-")) {
2742 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2743 read_stdin = 0;
2744 continue;
2745 }
2746 if (!prefixcmp(arg, "--exclude=")) {
2747 struct excludes *x = xmalloc(sizeof(*x));
2748 x->path = arg + 10;
2749 x->next = excludes;
2750 excludes = x;
2751 continue;
2752 }
2753 if (!prefixcmp(arg, "-p")) {
2754 p_value = atoi(arg + 2);
2755 p_value_known = 1;
2756 continue;
2757 }
2758 if (!strcmp(arg, "--no-add")) {
2759 no_add = 1;
2760 continue;
2761 }
2762 if (!strcmp(arg, "--stat")) {
2763 apply = 0;
2764 diffstat = 1;
2765 continue;
2766 }
2767 if (!strcmp(arg, "--allow-binary-replacement") ||
2768 !strcmp(arg, "--binary")) {
2769 continue; /* now no-op */
2770 }
2771 if (!strcmp(arg, "--numstat")) {
2772 apply = 0;
2773 numstat = 1;
2774 continue;
2775 }
2776 if (!strcmp(arg, "--summary")) {
2777 apply = 0;
2778 summary = 1;
2779 continue;
2780 }
2781 if (!strcmp(arg, "--check")) {
2782 apply = 0;
2783 check = 1;
2784 continue;
2785 }
2786 if (!strcmp(arg, "--index")) {
2787 if (is_not_gitdir)
2788 die("--index outside a repository");
2789 check_index = 1;
2790 continue;
2791 }
2792 if (!strcmp(arg, "--cached")) {
2793 if (is_not_gitdir)
2794 die("--cached outside a repository");
2795 check_index = 1;
2796 cached = 1;
2797 continue;
2798 }
2799 if (!strcmp(arg, "--apply")) {
2800 apply = 1;
2801 continue;
2802 }
2803 if (!strcmp(arg, "--index-info")) {
2804 apply = 0;
2805 show_index_info = 1;
2806 continue;
2807 }
2808 if (!strcmp(arg, "-z")) {
2809 line_termination = 0;
2810 continue;
2811 }
2812 if (!prefixcmp(arg, "-C")) {
2813 p_context = strtoul(arg + 2, &end, 0);
2814 if (*end != '\0')
2815 die("unrecognized context count '%s'", arg + 2);
2816 continue;
2817 }
2818 if (!prefixcmp(arg, "--whitespace=")) {
2819 whitespace_option = arg + 13;
2820 parse_whitespace_option(arg + 13);
2821 continue;
2822 }
2823 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2824 apply_in_reverse = 1;
2825 continue;
2826 }
2827 if (!strcmp(arg, "--unidiff-zero")) {
2828 unidiff_zero = 1;
2829 continue;
2830 }
2831 if (!strcmp(arg, "--reject")) {
2832 apply = apply_with_reject = apply_verbosely = 1;
2833 continue;
2834 }
2835 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
2836 apply_verbosely = 1;
2837 continue;
2838 }
2839 if (!strcmp(arg, "--inaccurate-eof")) {
2840 inaccurate_eof = 1;
2841 continue;
2842 }
2843 if (0 < prefix_length)
2844 arg = prefix_filename(prefix, prefix_length, arg);
2845
2846 fd = open(arg, O_RDONLY);
2847 if (fd < 0)
2848 usage(apply_usage);
2849 read_stdin = 0;
2850 set_default_whitespace_mode(whitespace_option);
2851 errs |= apply_patch(fd, arg, inaccurate_eof);
2852 close(fd);
2853 }
2854 set_default_whitespace_mode(whitespace_option);
2855 if (read_stdin)
2856 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2857 if (whitespace_error) {
2858 if (squelch_whitespace_errors &&
2859 squelch_whitespace_errors < whitespace_error) {
2860 int squelched =
2861 whitespace_error - squelch_whitespace_errors;
2862 fprintf(stderr, "warning: squelched %d "
2863 "whitespace error%s\n",
2864 squelched,
2865 squelched == 1 ? "" : "s");
2866 }
2867 if (new_whitespace == error_on_whitespace)
2868 die("%d line%s add%s trailing whitespaces.",
2869 whitespace_error,
2870 whitespace_error == 1 ? "" : "s",
2871 whitespace_error == 1 ? "s" : "");
2872 if (applied_after_stripping)
2873 fprintf(stderr, "warning: %d line%s applied after"
2874 " stripping trailing whitespaces.\n",
2875 applied_after_stripping,
2876 applied_after_stripping == 1 ? "" : "s");
2877 else if (whitespace_error)
2878 fprintf(stderr, "warning: %d line%s add%s trailing"
2879 " whitespaces.\n",
2880 whitespace_error,
2881 whitespace_error == 1 ? "" : "s",
2882 whitespace_error == 1 ? "s" : "");
2883 }
2884
2885 if (update_index) {
2886 if (write_cache(newfd, active_cache, active_nr) ||
2887 close(newfd) || commit_locked_index(&lock_file))
2888 die("Unable to write new index file");
2889 }
2890
2891 return !!errs;
2892 }