Merge branch 'bc/master-diff-hunk-header-fix'
[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 #include "string-list.h"
16 #include "dir.h"
17
18 /*
19 * --check turns on checking that the working tree matches the
20 * files that are being modified, but doesn't apply the patch
21 * --stat does just a diffstat, and doesn't actually apply
22 * --numstat does numeric diffstat, and doesn't actually apply
23 * --index-info shows the old and new index info for paths if available.
24 * --index updates the cache as well.
25 * --cached updates only the cache without ever touching the working tree.
26 */
27 static const char *prefix;
28 static int prefix_length = -1;
29 static int newfd = -1;
30
31 static int unidiff_zero;
32 static int p_value = 1;
33 static int p_value_known;
34 static int check_index;
35 static int update_index;
36 static int cached;
37 static int diffstat;
38 static int numstat;
39 static int summary;
40 static int check;
41 static int apply = 1;
42 static int apply_in_reverse;
43 static int apply_with_reject;
44 static int apply_verbosely;
45 static int no_add;
46 static const char *fake_ancestor;
47 static int line_termination = '\n';
48 static unsigned long p_context = ULONG_MAX;
49 static const char apply_usage[] =
50 "git apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|fix|error|error-all>] <patch>...";
51
52 static enum ws_error_action {
53 nowarn_ws_error,
54 warn_on_ws_error,
55 die_on_ws_error,
56 correct_ws_error,
57 } ws_error_action = warn_on_ws_error;
58 static int whitespace_error;
59 static int squelch_whitespace_errors = 5;
60 static int applied_after_fixing_ws;
61 static const char *patch_input_file;
62 static const char *root;
63 static int root_len;
64
65 static void parse_whitespace_option(const char *option)
66 {
67 if (!option) {
68 ws_error_action = warn_on_ws_error;
69 return;
70 }
71 if (!strcmp(option, "warn")) {
72 ws_error_action = warn_on_ws_error;
73 return;
74 }
75 if (!strcmp(option, "nowarn")) {
76 ws_error_action = nowarn_ws_error;
77 return;
78 }
79 if (!strcmp(option, "error")) {
80 ws_error_action = die_on_ws_error;
81 return;
82 }
83 if (!strcmp(option, "error-all")) {
84 ws_error_action = die_on_ws_error;
85 squelch_whitespace_errors = 0;
86 return;
87 }
88 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
89 ws_error_action = correct_ws_error;
90 return;
91 }
92 die("unrecognized whitespace option '%s'", option);
93 }
94
95 static void set_default_whitespace_mode(const char *whitespace_option)
96 {
97 if (!whitespace_option && !apply_default_whitespace)
98 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
99 }
100
101 /*
102 * For "diff-stat" like behaviour, we keep track of the biggest change
103 * we've seen, and the longest filename. That allows us to do simple
104 * scaling.
105 */
106 static int max_change, max_len;
107
108 /*
109 * Various "current state", notably line numbers and what
110 * file (and how) we're patching right now.. The "is_xxxx"
111 * things are flags, where -1 means "don't know yet".
112 */
113 static int linenr = 1;
114
115 /*
116 * This represents one "hunk" from a patch, starting with
117 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
118 * patch text is pointed at by patch, and its byte length
119 * is stored in size. leading and trailing are the number
120 * of context lines.
121 */
122 struct fragment {
123 unsigned long leading, trailing;
124 unsigned long oldpos, oldlines;
125 unsigned long newpos, newlines;
126 const char *patch;
127 int size;
128 int rejected;
129 struct fragment *next;
130 };
131
132 /*
133 * When dealing with a binary patch, we reuse "leading" field
134 * to store the type of the binary hunk, either deflated "delta"
135 * or deflated "literal".
136 */
137 #define binary_patch_method leading
138 #define BINARY_DELTA_DEFLATED 1
139 #define BINARY_LITERAL_DEFLATED 2
140
141 /*
142 * This represents a "patch" to a file, both metainfo changes
143 * such as creation/deletion, filemode and content changes represented
144 * as a series of fragments.
145 */
146 struct patch {
147 char *new_name, *old_name, *def_name;
148 unsigned int old_mode, new_mode;
149 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
150 int rejected;
151 unsigned ws_rule;
152 unsigned long deflate_origlen;
153 int lines_added, lines_deleted;
154 int score;
155 unsigned int is_toplevel_relative:1;
156 unsigned int inaccurate_eof:1;
157 unsigned int is_binary:1;
158 unsigned int is_copy:1;
159 unsigned int is_rename:1;
160 unsigned int recount:1;
161 struct fragment *fragments;
162 char *result;
163 size_t resultsize;
164 char old_sha1_prefix[41];
165 char new_sha1_prefix[41];
166 struct patch *next;
167 };
168
169 /*
170 * A line in a file, len-bytes long (includes the terminating LF,
171 * except for an incomplete line at the end if the file ends with
172 * one), and its contents hashes to 'hash'.
173 */
174 struct line {
175 size_t len;
176 unsigned hash : 24;
177 unsigned flag : 8;
178 #define LINE_COMMON 1
179 };
180
181 /*
182 * This represents a "file", which is an array of "lines".
183 */
184 struct image {
185 char *buf;
186 size_t len;
187 size_t nr;
188 size_t alloc;
189 struct line *line_allocated;
190 struct line *line;
191 };
192
193 /*
194 * Records filenames that have been touched, in order to handle
195 * the case where more than one patches touch the same file.
196 */
197
198 static struct string_list fn_table;
199
200 static uint32_t hash_line(const char *cp, size_t len)
201 {
202 size_t i;
203 uint32_t h;
204 for (i = 0, h = 0; i < len; i++) {
205 if (!isspace(cp[i])) {
206 h = h * 3 + (cp[i] & 0xff);
207 }
208 }
209 return h;
210 }
211
212 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
213 {
214 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
215 img->line_allocated[img->nr].len = len;
216 img->line_allocated[img->nr].hash = hash_line(bol, len);
217 img->line_allocated[img->nr].flag = flag;
218 img->nr++;
219 }
220
221 static void prepare_image(struct image *image, char *buf, size_t len,
222 int prepare_linetable)
223 {
224 const char *cp, *ep;
225
226 memset(image, 0, sizeof(*image));
227 image->buf = buf;
228 image->len = len;
229
230 if (!prepare_linetable)
231 return;
232
233 ep = image->buf + image->len;
234 cp = image->buf;
235 while (cp < ep) {
236 const char *next;
237 for (next = cp; next < ep && *next != '\n'; next++)
238 ;
239 if (next < ep)
240 next++;
241 add_line_info(image, cp, next - cp, 0);
242 cp = next;
243 }
244 image->line = image->line_allocated;
245 }
246
247 static void clear_image(struct image *image)
248 {
249 free(image->buf);
250 image->buf = NULL;
251 image->len = 0;
252 }
253
254 static void say_patch_name(FILE *output, const char *pre,
255 struct patch *patch, const char *post)
256 {
257 fputs(pre, output);
258 if (patch->old_name && patch->new_name &&
259 strcmp(patch->old_name, patch->new_name)) {
260 quote_c_style(patch->old_name, NULL, output, 0);
261 fputs(" => ", output);
262 quote_c_style(patch->new_name, NULL, output, 0);
263 } else {
264 const char *n = patch->new_name;
265 if (!n)
266 n = patch->old_name;
267 quote_c_style(n, NULL, output, 0);
268 }
269 fputs(post, output);
270 }
271
272 #define CHUNKSIZE (8192)
273 #define SLOP (16)
274
275 static void read_patch_file(struct strbuf *sb, int fd)
276 {
277 if (strbuf_read(sb, fd, 0) < 0)
278 die("git apply: read returned %s", strerror(errno));
279
280 /*
281 * Make sure that we have some slop in the buffer
282 * so that we can do speculative "memcmp" etc, and
283 * see to it that it is NUL-filled.
284 */
285 strbuf_grow(sb, SLOP);
286 memset(sb->buf + sb->len, 0, SLOP);
287 }
288
289 static unsigned long linelen(const char *buffer, unsigned long size)
290 {
291 unsigned long len = 0;
292 while (size--) {
293 len++;
294 if (*buffer++ == '\n')
295 break;
296 }
297 return len;
298 }
299
300 static int is_dev_null(const char *str)
301 {
302 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
303 }
304
305 #define TERM_SPACE 1
306 #define TERM_TAB 2
307
308 static int name_terminate(const char *name, int namelen, int c, int terminate)
309 {
310 if (c == ' ' && !(terminate & TERM_SPACE))
311 return 0;
312 if (c == '\t' && !(terminate & TERM_TAB))
313 return 0;
314
315 return 1;
316 }
317
318 static char *find_name(const char *line, char *def, int p_value, int terminate)
319 {
320 int len;
321 const char *start = line;
322
323 if (*line == '"') {
324 struct strbuf name;
325
326 /*
327 * Proposed "new-style" GNU patch/diff format; see
328 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
329 */
330 strbuf_init(&name, 0);
331 if (!unquote_c_style(&name, line, NULL)) {
332 char *cp;
333
334 for (cp = name.buf; p_value; p_value--) {
335 cp = strchr(cp, '/');
336 if (!cp)
337 break;
338 cp++;
339 }
340 if (cp) {
341 /* name can later be freed, so we need
342 * to memmove, not just return cp
343 */
344 strbuf_remove(&name, 0, cp - name.buf);
345 free(def);
346 if (root)
347 strbuf_insert(&name, 0, root, root_len);
348 return strbuf_detach(&name, NULL);
349 }
350 }
351 strbuf_release(&name);
352 }
353
354 for (;;) {
355 char c = *line;
356
357 if (isspace(c)) {
358 if (c == '\n')
359 break;
360 if (name_terminate(start, line-start, c, terminate))
361 break;
362 }
363 line++;
364 if (c == '/' && !--p_value)
365 start = line;
366 }
367 if (!start)
368 return def;
369 len = line - start;
370 if (!len)
371 return def;
372
373 /*
374 * Generally we prefer the shorter name, especially
375 * if the other one is just a variation of that with
376 * something else tacked on to the end (ie "file.orig"
377 * or "file~").
378 */
379 if (def) {
380 int deflen = strlen(def);
381 if (deflen < len && !strncmp(start, def, deflen))
382 return def;
383 free(def);
384 }
385
386 if (root) {
387 char *ret = xmalloc(root_len + len + 1);
388 strcpy(ret, root);
389 memcpy(ret + root_len, start, len);
390 ret[root_len + len] = '\0';
391 return ret;
392 }
393
394 return xmemdupz(start, len);
395 }
396
397 static int count_slashes(const char *cp)
398 {
399 int cnt = 0;
400 char ch;
401
402 while ((ch = *cp++))
403 if (ch == '/')
404 cnt++;
405 return cnt;
406 }
407
408 /*
409 * Given the string after "--- " or "+++ ", guess the appropriate
410 * p_value for the given patch.
411 */
412 static int guess_p_value(const char *nameline)
413 {
414 char *name, *cp;
415 int val = -1;
416
417 if (is_dev_null(nameline))
418 return -1;
419 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
420 if (!name)
421 return -1;
422 cp = strchr(name, '/');
423 if (!cp)
424 val = 0;
425 else if (prefix) {
426 /*
427 * Does it begin with "a/$our-prefix" and such? Then this is
428 * very likely to apply to our directory.
429 */
430 if (!strncmp(name, prefix, prefix_length))
431 val = count_slashes(prefix);
432 else {
433 cp++;
434 if (!strncmp(cp, prefix, prefix_length))
435 val = count_slashes(prefix) + 1;
436 }
437 }
438 free(name);
439 return val;
440 }
441
442 /*
443 * Get the name etc info from the ---/+++ lines of a traditional patch header
444 *
445 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
446 * files, we can happily check the index for a match, but for creating a
447 * new file we should try to match whatever "patch" does. I have no idea.
448 */
449 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
450 {
451 char *name;
452
453 first += 4; /* skip "--- " */
454 second += 4; /* skip "+++ " */
455 if (!p_value_known) {
456 int p, q;
457 p = guess_p_value(first);
458 q = guess_p_value(second);
459 if (p < 0) p = q;
460 if (0 <= p && p == q) {
461 p_value = p;
462 p_value_known = 1;
463 }
464 }
465 if (is_dev_null(first)) {
466 patch->is_new = 1;
467 patch->is_delete = 0;
468 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
469 patch->new_name = name;
470 } else if (is_dev_null(second)) {
471 patch->is_new = 0;
472 patch->is_delete = 1;
473 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
474 patch->old_name = name;
475 } else {
476 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
477 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
478 patch->old_name = patch->new_name = name;
479 }
480 if (!name)
481 die("unable to find filename in patch at line %d", linenr);
482 }
483
484 static int gitdiff_hdrend(const char *line, struct patch *patch)
485 {
486 return -1;
487 }
488
489 /*
490 * We're anal about diff header consistency, to make
491 * sure that we don't end up having strange ambiguous
492 * patches floating around.
493 *
494 * As a result, gitdiff_{old|new}name() will check
495 * their names against any previous information, just
496 * to make sure..
497 */
498 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
499 {
500 if (!orig_name && !isnull)
501 return find_name(line, NULL, p_value, TERM_TAB);
502
503 if (orig_name) {
504 int len;
505 const char *name;
506 char *another;
507 name = orig_name;
508 len = strlen(name);
509 if (isnull)
510 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
511 another = find_name(line, NULL, p_value, TERM_TAB);
512 if (!another || memcmp(another, name, len))
513 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
514 free(another);
515 return orig_name;
516 }
517 else {
518 /* expect "/dev/null" */
519 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
520 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
521 return NULL;
522 }
523 }
524
525 static int gitdiff_oldname(const char *line, struct patch *patch)
526 {
527 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
528 return 0;
529 }
530
531 static int gitdiff_newname(const char *line, struct patch *patch)
532 {
533 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
534 return 0;
535 }
536
537 static int gitdiff_oldmode(const char *line, struct patch *patch)
538 {
539 patch->old_mode = strtoul(line, NULL, 8);
540 return 0;
541 }
542
543 static int gitdiff_newmode(const char *line, struct patch *patch)
544 {
545 patch->new_mode = strtoul(line, NULL, 8);
546 return 0;
547 }
548
549 static int gitdiff_delete(const char *line, struct patch *patch)
550 {
551 patch->is_delete = 1;
552 patch->old_name = patch->def_name;
553 return gitdiff_oldmode(line, patch);
554 }
555
556 static int gitdiff_newfile(const char *line, struct patch *patch)
557 {
558 patch->is_new = 1;
559 patch->new_name = patch->def_name;
560 return gitdiff_newmode(line, patch);
561 }
562
563 static int gitdiff_copysrc(const char *line, struct patch *patch)
564 {
565 patch->is_copy = 1;
566 patch->old_name = find_name(line, NULL, 0, 0);
567 return 0;
568 }
569
570 static int gitdiff_copydst(const char *line, struct patch *patch)
571 {
572 patch->is_copy = 1;
573 patch->new_name = find_name(line, NULL, 0, 0);
574 return 0;
575 }
576
577 static int gitdiff_renamesrc(const char *line, struct patch *patch)
578 {
579 patch->is_rename = 1;
580 patch->old_name = find_name(line, NULL, 0, 0);
581 return 0;
582 }
583
584 static int gitdiff_renamedst(const char *line, struct patch *patch)
585 {
586 patch->is_rename = 1;
587 patch->new_name = find_name(line, NULL, 0, 0);
588 return 0;
589 }
590
591 static int gitdiff_similarity(const char *line, struct patch *patch)
592 {
593 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
594 patch->score = 0;
595 return 0;
596 }
597
598 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
599 {
600 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
601 patch->score = 0;
602 return 0;
603 }
604
605 static int gitdiff_index(const char *line, struct patch *patch)
606 {
607 /*
608 * index line is N hexadecimal, "..", N hexadecimal,
609 * and optional space with octal mode.
610 */
611 const char *ptr, *eol;
612 int len;
613
614 ptr = strchr(line, '.');
615 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
616 return 0;
617 len = ptr - line;
618 memcpy(patch->old_sha1_prefix, line, len);
619 patch->old_sha1_prefix[len] = 0;
620
621 line = ptr + 2;
622 ptr = strchr(line, ' ');
623 eol = strchr(line, '\n');
624
625 if (!ptr || eol < ptr)
626 ptr = eol;
627 len = ptr - line;
628
629 if (40 < len)
630 return 0;
631 memcpy(patch->new_sha1_prefix, line, len);
632 patch->new_sha1_prefix[len] = 0;
633 if (*ptr == ' ')
634 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
635 return 0;
636 }
637
638 /*
639 * This is normal for a diff that doesn't change anything: we'll fall through
640 * into the next diff. Tell the parser to break out.
641 */
642 static int gitdiff_unrecognized(const char *line, struct patch *patch)
643 {
644 return -1;
645 }
646
647 static const char *stop_at_slash(const char *line, int llen)
648 {
649 int i;
650
651 for (i = 0; i < llen; i++) {
652 int ch = line[i];
653 if (ch == '/')
654 return line + i;
655 }
656 return NULL;
657 }
658
659 /*
660 * This is to extract the same name that appears on "diff --git"
661 * line. We do not find and return anything if it is a rename
662 * patch, and it is OK because we will find the name elsewhere.
663 * We need to reliably find name only when it is mode-change only,
664 * creation or deletion of an empty file. In any of these cases,
665 * both sides are the same name under a/ and b/ respectively.
666 */
667 static char *git_header_name(char *line, int llen)
668 {
669 const char *name;
670 const char *second = NULL;
671 size_t len;
672
673 line += strlen("diff --git ");
674 llen -= strlen("diff --git ");
675
676 if (*line == '"') {
677 const char *cp;
678 struct strbuf first;
679 struct strbuf sp;
680
681 strbuf_init(&first, 0);
682 strbuf_init(&sp, 0);
683
684 if (unquote_c_style(&first, line, &second))
685 goto free_and_fail1;
686
687 /* advance to the first slash */
688 cp = stop_at_slash(first.buf, first.len);
689 /* we do not accept absolute paths */
690 if (!cp || cp == first.buf)
691 goto free_and_fail1;
692 strbuf_remove(&first, 0, cp + 1 - first.buf);
693
694 /*
695 * second points at one past closing dq of name.
696 * find the second name.
697 */
698 while ((second < line + llen) && isspace(*second))
699 second++;
700
701 if (line + llen <= second)
702 goto free_and_fail1;
703 if (*second == '"') {
704 if (unquote_c_style(&sp, second, NULL))
705 goto free_and_fail1;
706 cp = stop_at_slash(sp.buf, sp.len);
707 if (!cp || cp == sp.buf)
708 goto free_and_fail1;
709 /* They must match, otherwise ignore */
710 if (strcmp(cp + 1, first.buf))
711 goto free_and_fail1;
712 strbuf_release(&sp);
713 return strbuf_detach(&first, NULL);
714 }
715
716 /* unquoted second */
717 cp = stop_at_slash(second, line + llen - second);
718 if (!cp || cp == second)
719 goto free_and_fail1;
720 cp++;
721 if (line + llen - cp != first.len + 1 ||
722 memcmp(first.buf, cp, first.len))
723 goto free_and_fail1;
724 return strbuf_detach(&first, NULL);
725
726 free_and_fail1:
727 strbuf_release(&first);
728 strbuf_release(&sp);
729 return NULL;
730 }
731
732 /* unquoted first name */
733 name = stop_at_slash(line, llen);
734 if (!name || name == line)
735 return NULL;
736 name++;
737
738 /*
739 * since the first name is unquoted, a dq if exists must be
740 * the beginning of the second name.
741 */
742 for (second = name; second < line + llen; second++) {
743 if (*second == '"') {
744 struct strbuf sp;
745 const char *np;
746
747 strbuf_init(&sp, 0);
748 if (unquote_c_style(&sp, second, NULL))
749 goto free_and_fail2;
750
751 np = stop_at_slash(sp.buf, sp.len);
752 if (!np || np == sp.buf)
753 goto free_and_fail2;
754 np++;
755
756 len = sp.buf + sp.len - np;
757 if (len < second - name &&
758 !strncmp(np, name, len) &&
759 isspace(name[len])) {
760 /* Good */
761 strbuf_remove(&sp, 0, np - sp.buf);
762 return strbuf_detach(&sp, NULL);
763 }
764
765 free_and_fail2:
766 strbuf_release(&sp);
767 return NULL;
768 }
769 }
770
771 /*
772 * Accept a name only if it shows up twice, exactly the same
773 * form.
774 */
775 for (len = 0 ; ; len++) {
776 switch (name[len]) {
777 default:
778 continue;
779 case '\n':
780 return NULL;
781 case '\t': case ' ':
782 second = name+len;
783 for (;;) {
784 char c = *second++;
785 if (c == '\n')
786 return NULL;
787 if (c == '/')
788 break;
789 }
790 if (second[len] == '\n' && !memcmp(name, second, len)) {
791 return xmemdupz(name, len);
792 }
793 }
794 }
795 }
796
797 /* Verify that we recognize the lines following a git header */
798 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
799 {
800 unsigned long offset;
801
802 /* A git diff has explicit new/delete information, so we don't guess */
803 patch->is_new = 0;
804 patch->is_delete = 0;
805
806 /*
807 * Some things may not have the old name in the
808 * rest of the headers anywhere (pure mode changes,
809 * or removing or adding empty files), so we get
810 * the default name from the header.
811 */
812 patch->def_name = git_header_name(line, len);
813
814 line += len;
815 size -= len;
816 linenr++;
817 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
818 static const struct opentry {
819 const char *str;
820 int (*fn)(const char *, struct patch *);
821 } optable[] = {
822 { "@@ -", gitdiff_hdrend },
823 { "--- ", gitdiff_oldname },
824 { "+++ ", gitdiff_newname },
825 { "old mode ", gitdiff_oldmode },
826 { "new mode ", gitdiff_newmode },
827 { "deleted file mode ", gitdiff_delete },
828 { "new file mode ", gitdiff_newfile },
829 { "copy from ", gitdiff_copysrc },
830 { "copy to ", gitdiff_copydst },
831 { "rename old ", gitdiff_renamesrc },
832 { "rename new ", gitdiff_renamedst },
833 { "rename from ", gitdiff_renamesrc },
834 { "rename to ", gitdiff_renamedst },
835 { "similarity index ", gitdiff_similarity },
836 { "dissimilarity index ", gitdiff_dissimilarity },
837 { "index ", gitdiff_index },
838 { "", gitdiff_unrecognized },
839 };
840 int i;
841
842 len = linelen(line, size);
843 if (!len || line[len-1] != '\n')
844 break;
845 for (i = 0; i < ARRAY_SIZE(optable); i++) {
846 const struct opentry *p = optable + i;
847 int oplen = strlen(p->str);
848 if (len < oplen || memcmp(p->str, line, oplen))
849 continue;
850 if (p->fn(line + oplen, patch) < 0)
851 return offset;
852 break;
853 }
854 }
855
856 return offset;
857 }
858
859 static int parse_num(const char *line, unsigned long *p)
860 {
861 char *ptr;
862
863 if (!isdigit(*line))
864 return 0;
865 *p = strtoul(line, &ptr, 10);
866 return ptr - line;
867 }
868
869 static int parse_range(const char *line, int len, int offset, const char *expect,
870 unsigned long *p1, unsigned long *p2)
871 {
872 int digits, ex;
873
874 if (offset < 0 || offset >= len)
875 return -1;
876 line += offset;
877 len -= offset;
878
879 digits = parse_num(line, p1);
880 if (!digits)
881 return -1;
882
883 offset += digits;
884 line += digits;
885 len -= digits;
886
887 *p2 = 1;
888 if (*line == ',') {
889 digits = parse_num(line+1, p2);
890 if (!digits)
891 return -1;
892
893 offset += digits+1;
894 line += digits+1;
895 len -= digits+1;
896 }
897
898 ex = strlen(expect);
899 if (ex > len)
900 return -1;
901 if (memcmp(line, expect, ex))
902 return -1;
903
904 return offset + ex;
905 }
906
907 static void recount_diff(char *line, int size, struct fragment *fragment)
908 {
909 int oldlines = 0, newlines = 0, ret = 0;
910
911 if (size < 1) {
912 warning("recount: ignore empty hunk");
913 return;
914 }
915
916 for (;;) {
917 int len = linelen(line, size);
918 size -= len;
919 line += len;
920
921 if (size < 1)
922 break;
923
924 switch (*line) {
925 case ' ': case '\n':
926 newlines++;
927 /* fall through */
928 case '-':
929 oldlines++;
930 continue;
931 case '+':
932 newlines++;
933 continue;
934 case '\\':
935 continue;
936 case '@':
937 ret = size < 3 || prefixcmp(line, "@@ ");
938 break;
939 case 'd':
940 ret = size < 5 || prefixcmp(line, "diff ");
941 break;
942 default:
943 ret = -1;
944 break;
945 }
946 if (ret) {
947 warning("recount: unexpected line: %.*s",
948 (int)linelen(line, size), line);
949 return;
950 }
951 break;
952 }
953 fragment->oldlines = oldlines;
954 fragment->newlines = newlines;
955 }
956
957 /*
958 * Parse a unified diff fragment header of the
959 * form "@@ -a,b +c,d @@"
960 */
961 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
962 {
963 int offset;
964
965 if (!len || line[len-1] != '\n')
966 return -1;
967
968 /* Figure out the number of lines in a fragment */
969 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
970 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
971
972 return offset;
973 }
974
975 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
976 {
977 unsigned long offset, len;
978
979 patch->is_toplevel_relative = 0;
980 patch->is_rename = patch->is_copy = 0;
981 patch->is_new = patch->is_delete = -1;
982 patch->old_mode = patch->new_mode = 0;
983 patch->old_name = patch->new_name = NULL;
984 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
985 unsigned long nextlen;
986
987 len = linelen(line, size);
988 if (!len)
989 break;
990
991 /* Testing this early allows us to take a few shortcuts.. */
992 if (len < 6)
993 continue;
994
995 /*
996 * Make sure we don't find any unconnected patch fragments.
997 * That's a sign that we didn't find a header, and that a
998 * patch has become corrupted/broken up.
999 */
1000 if (!memcmp("@@ -", line, 4)) {
1001 struct fragment dummy;
1002 if (parse_fragment_header(line, len, &dummy) < 0)
1003 continue;
1004 die("patch fragment without header at line %d: %.*s",
1005 linenr, (int)len-1, line);
1006 }
1007
1008 if (size < len + 6)
1009 break;
1010
1011 /*
1012 * Git patch? It might not have a real patch, just a rename
1013 * or mode change, so we handle that specially
1014 */
1015 if (!memcmp("diff --git ", line, 11)) {
1016 int git_hdr_len = parse_git_header(line, len, size, patch);
1017 if (git_hdr_len <= len)
1018 continue;
1019 if (!patch->old_name && !patch->new_name) {
1020 if (!patch->def_name)
1021 die("git diff header lacks filename information (line %d)", linenr);
1022 patch->old_name = patch->new_name = patch->def_name;
1023 }
1024 patch->is_toplevel_relative = 1;
1025 *hdrsize = git_hdr_len;
1026 return offset;
1027 }
1028
1029 /* --- followed by +++ ? */
1030 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1031 continue;
1032
1033 /*
1034 * We only accept unified patches, so we want it to
1035 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1036 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1037 */
1038 nextlen = linelen(line + len, size - len);
1039 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1040 continue;
1041
1042 /* Ok, we'll consider it a patch */
1043 parse_traditional_patch(line, line+len, patch);
1044 *hdrsize = len + nextlen;
1045 linenr += 2;
1046 return offset;
1047 }
1048 return -1;
1049 }
1050
1051 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1052 {
1053 char *err;
1054 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1055 if (!result)
1056 return;
1057
1058 whitespace_error++;
1059 if (squelch_whitespace_errors &&
1060 squelch_whitespace_errors < whitespace_error)
1061 ;
1062 else {
1063 err = whitespace_error_string(result);
1064 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1065 patch_input_file, linenr, err, len - 2, line + 1);
1066 free(err);
1067 }
1068 }
1069
1070 /*
1071 * Parse a unified diff. Note that this really needs to parse each
1072 * fragment separately, since the only way to know the difference
1073 * between a "---" that is part of a patch, and a "---" that starts
1074 * the next patch is to look at the line counts..
1075 */
1076 static int parse_fragment(char *line, unsigned long size,
1077 struct patch *patch, struct fragment *fragment)
1078 {
1079 int added, deleted;
1080 int len = linelen(line, size), offset;
1081 unsigned long oldlines, newlines;
1082 unsigned long leading, trailing;
1083
1084 offset = parse_fragment_header(line, len, fragment);
1085 if (offset < 0)
1086 return -1;
1087 if (offset > 0 && patch->recount)
1088 recount_diff(line + offset, size - offset, fragment);
1089 oldlines = fragment->oldlines;
1090 newlines = fragment->newlines;
1091 leading = 0;
1092 trailing = 0;
1093
1094 /* Parse the thing.. */
1095 line += len;
1096 size -= len;
1097 linenr++;
1098 added = deleted = 0;
1099 for (offset = len;
1100 0 < size;
1101 offset += len, size -= len, line += len, linenr++) {
1102 if (!oldlines && !newlines)
1103 break;
1104 len = linelen(line, size);
1105 if (!len || line[len-1] != '\n')
1106 return -1;
1107 switch (*line) {
1108 default:
1109 return -1;
1110 case '\n': /* newer GNU diff, an empty context line */
1111 case ' ':
1112 oldlines--;
1113 newlines--;
1114 if (!deleted && !added)
1115 leading++;
1116 trailing++;
1117 break;
1118 case '-':
1119 if (apply_in_reverse &&
1120 ws_error_action != nowarn_ws_error)
1121 check_whitespace(line, len, patch->ws_rule);
1122 deleted++;
1123 oldlines--;
1124 trailing = 0;
1125 break;
1126 case '+':
1127 if (!apply_in_reverse &&
1128 ws_error_action != nowarn_ws_error)
1129 check_whitespace(line, len, patch->ws_rule);
1130 added++;
1131 newlines--;
1132 trailing = 0;
1133 break;
1134
1135 /*
1136 * We allow "\ No newline at end of file". Depending
1137 * on locale settings when the patch was produced we
1138 * don't know what this line looks like. The only
1139 * thing we do know is that it begins with "\ ".
1140 * Checking for 12 is just for sanity check -- any
1141 * l10n of "\ No newline..." is at least that long.
1142 */
1143 case '\\':
1144 if (len < 12 || memcmp(line, "\\ ", 2))
1145 return -1;
1146 break;
1147 }
1148 }
1149 if (oldlines || newlines)
1150 return -1;
1151 fragment->leading = leading;
1152 fragment->trailing = trailing;
1153
1154 /*
1155 * If a fragment ends with an incomplete line, we failed to include
1156 * it in the above loop because we hit oldlines == newlines == 0
1157 * before seeing it.
1158 */
1159 if (12 < size && !memcmp(line, "\\ ", 2))
1160 offset += linelen(line, size);
1161
1162 patch->lines_added += added;
1163 patch->lines_deleted += deleted;
1164
1165 if (0 < patch->is_new && oldlines)
1166 return error("new file depends on old contents");
1167 if (0 < patch->is_delete && newlines)
1168 return error("deleted file still has contents");
1169 return offset;
1170 }
1171
1172 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1173 {
1174 unsigned long offset = 0;
1175 unsigned long oldlines = 0, newlines = 0, context = 0;
1176 struct fragment **fragp = &patch->fragments;
1177
1178 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1179 struct fragment *fragment;
1180 int len;
1181
1182 fragment = xcalloc(1, sizeof(*fragment));
1183 len = parse_fragment(line, size, patch, fragment);
1184 if (len <= 0)
1185 die("corrupt patch at line %d", linenr);
1186 fragment->patch = line;
1187 fragment->size = len;
1188 oldlines += fragment->oldlines;
1189 newlines += fragment->newlines;
1190 context += fragment->leading + fragment->trailing;
1191
1192 *fragp = fragment;
1193 fragp = &fragment->next;
1194
1195 offset += len;
1196 line += len;
1197 size -= len;
1198 }
1199
1200 /*
1201 * If something was removed (i.e. we have old-lines) it cannot
1202 * be creation, and if something was added it cannot be
1203 * deletion. However, the reverse is not true; --unified=0
1204 * patches that only add are not necessarily creation even
1205 * though they do not have any old lines, and ones that only
1206 * delete are not necessarily deletion.
1207 *
1208 * Unfortunately, a real creation/deletion patch do _not_ have
1209 * any context line by definition, so we cannot safely tell it
1210 * apart with --unified=0 insanity. At least if the patch has
1211 * more than one hunk it is not creation or deletion.
1212 */
1213 if (patch->is_new < 0 &&
1214 (oldlines || (patch->fragments && patch->fragments->next)))
1215 patch->is_new = 0;
1216 if (patch->is_delete < 0 &&
1217 (newlines || (patch->fragments && patch->fragments->next)))
1218 patch->is_delete = 0;
1219
1220 if (0 < patch->is_new && oldlines)
1221 die("new file %s depends on old contents", patch->new_name);
1222 if (0 < patch->is_delete && newlines)
1223 die("deleted file %s still has contents", patch->old_name);
1224 if (!patch->is_delete && !newlines && context)
1225 fprintf(stderr, "** warning: file %s becomes empty but "
1226 "is not deleted\n", patch->new_name);
1227
1228 return offset;
1229 }
1230
1231 static inline int metadata_changes(struct patch *patch)
1232 {
1233 return patch->is_rename > 0 ||
1234 patch->is_copy > 0 ||
1235 patch->is_new > 0 ||
1236 patch->is_delete ||
1237 (patch->old_mode && patch->new_mode &&
1238 patch->old_mode != patch->new_mode);
1239 }
1240
1241 static char *inflate_it(const void *data, unsigned long size,
1242 unsigned long inflated_size)
1243 {
1244 z_stream stream;
1245 void *out;
1246 int st;
1247
1248 memset(&stream, 0, sizeof(stream));
1249
1250 stream.next_in = (unsigned char *)data;
1251 stream.avail_in = size;
1252 stream.next_out = out = xmalloc(inflated_size);
1253 stream.avail_out = inflated_size;
1254 inflateInit(&stream);
1255 st = inflate(&stream, Z_FINISH);
1256 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1257 free(out);
1258 return NULL;
1259 }
1260 return out;
1261 }
1262
1263 static struct fragment *parse_binary_hunk(char **buf_p,
1264 unsigned long *sz_p,
1265 int *status_p,
1266 int *used_p)
1267 {
1268 /*
1269 * Expect a line that begins with binary patch method ("literal"
1270 * or "delta"), followed by the length of data before deflating.
1271 * a sequence of 'length-byte' followed by base-85 encoded data
1272 * should follow, terminated by a newline.
1273 *
1274 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1275 * and we would limit the patch line to 66 characters,
1276 * so one line can fit up to 13 groups that would decode
1277 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1278 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1279 */
1280 int llen, used;
1281 unsigned long size = *sz_p;
1282 char *buffer = *buf_p;
1283 int patch_method;
1284 unsigned long origlen;
1285 char *data = NULL;
1286 int hunk_size = 0;
1287 struct fragment *frag;
1288
1289 llen = linelen(buffer, size);
1290 used = llen;
1291
1292 *status_p = 0;
1293
1294 if (!prefixcmp(buffer, "delta ")) {
1295 patch_method = BINARY_DELTA_DEFLATED;
1296 origlen = strtoul(buffer + 6, NULL, 10);
1297 }
1298 else if (!prefixcmp(buffer, "literal ")) {
1299 patch_method = BINARY_LITERAL_DEFLATED;
1300 origlen = strtoul(buffer + 8, NULL, 10);
1301 }
1302 else
1303 return NULL;
1304
1305 linenr++;
1306 buffer += llen;
1307 while (1) {
1308 int byte_length, max_byte_length, newsize;
1309 llen = linelen(buffer, size);
1310 used += llen;
1311 linenr++;
1312 if (llen == 1) {
1313 /* consume the blank line */
1314 buffer++;
1315 size--;
1316 break;
1317 }
1318 /*
1319 * Minimum line is "A00000\n" which is 7-byte long,
1320 * and the line length must be multiple of 5 plus 2.
1321 */
1322 if ((llen < 7) || (llen-2) % 5)
1323 goto corrupt;
1324 max_byte_length = (llen - 2) / 5 * 4;
1325 byte_length = *buffer;
1326 if ('A' <= byte_length && byte_length <= 'Z')
1327 byte_length = byte_length - 'A' + 1;
1328 else if ('a' <= byte_length && byte_length <= 'z')
1329 byte_length = byte_length - 'a' + 27;
1330 else
1331 goto corrupt;
1332 /* if the input length was not multiple of 4, we would
1333 * have filler at the end but the filler should never
1334 * exceed 3 bytes
1335 */
1336 if (max_byte_length < byte_length ||
1337 byte_length <= max_byte_length - 4)
1338 goto corrupt;
1339 newsize = hunk_size + byte_length;
1340 data = xrealloc(data, newsize);
1341 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1342 goto corrupt;
1343 hunk_size = newsize;
1344 buffer += llen;
1345 size -= llen;
1346 }
1347
1348 frag = xcalloc(1, sizeof(*frag));
1349 frag->patch = inflate_it(data, hunk_size, origlen);
1350 if (!frag->patch)
1351 goto corrupt;
1352 free(data);
1353 frag->size = origlen;
1354 *buf_p = buffer;
1355 *sz_p = size;
1356 *used_p = used;
1357 frag->binary_patch_method = patch_method;
1358 return frag;
1359
1360 corrupt:
1361 free(data);
1362 *status_p = -1;
1363 error("corrupt binary patch at line %d: %.*s",
1364 linenr-1, llen-1, buffer);
1365 return NULL;
1366 }
1367
1368 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1369 {
1370 /*
1371 * We have read "GIT binary patch\n"; what follows is a line
1372 * that says the patch method (currently, either "literal" or
1373 * "delta") and the length of data before deflating; a
1374 * sequence of 'length-byte' followed by base-85 encoded data
1375 * follows.
1376 *
1377 * When a binary patch is reversible, there is another binary
1378 * hunk in the same format, starting with patch method (either
1379 * "literal" or "delta") with the length of data, and a sequence
1380 * of length-byte + base-85 encoded data, terminated with another
1381 * empty line. This data, when applied to the postimage, produces
1382 * the preimage.
1383 */
1384 struct fragment *forward;
1385 struct fragment *reverse;
1386 int status;
1387 int used, used_1;
1388
1389 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1390 if (!forward && !status)
1391 /* there has to be one hunk (forward hunk) */
1392 return error("unrecognized binary patch at line %d", linenr-1);
1393 if (status)
1394 /* otherwise we already gave an error message */
1395 return status;
1396
1397 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1398 if (reverse)
1399 used += used_1;
1400 else if (status) {
1401 /*
1402 * Not having reverse hunk is not an error, but having
1403 * a corrupt reverse hunk is.
1404 */
1405 free((void*) forward->patch);
1406 free(forward);
1407 return status;
1408 }
1409 forward->next = reverse;
1410 patch->fragments = forward;
1411 patch->is_binary = 1;
1412 return used;
1413 }
1414
1415 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1416 {
1417 int hdrsize, patchsize;
1418 int offset = find_header(buffer, size, &hdrsize, patch);
1419
1420 if (offset < 0)
1421 return offset;
1422
1423 patch->ws_rule = whitespace_rule(patch->new_name
1424 ? patch->new_name
1425 : patch->old_name);
1426
1427 patchsize = parse_single_patch(buffer + offset + hdrsize,
1428 size - offset - hdrsize, patch);
1429
1430 if (!patchsize) {
1431 static const char *binhdr[] = {
1432 "Binary files ",
1433 "Files ",
1434 NULL,
1435 };
1436 static const char git_binary[] = "GIT binary patch\n";
1437 int i;
1438 int hd = hdrsize + offset;
1439 unsigned long llen = linelen(buffer + hd, size - hd);
1440
1441 if (llen == sizeof(git_binary) - 1 &&
1442 !memcmp(git_binary, buffer + hd, llen)) {
1443 int used;
1444 linenr++;
1445 used = parse_binary(buffer + hd + llen,
1446 size - hd - llen, patch);
1447 if (used)
1448 patchsize = used + llen;
1449 else
1450 patchsize = 0;
1451 }
1452 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1453 for (i = 0; binhdr[i]; i++) {
1454 int len = strlen(binhdr[i]);
1455 if (len < size - hd &&
1456 !memcmp(binhdr[i], buffer + hd, len)) {
1457 linenr++;
1458 patch->is_binary = 1;
1459 patchsize = llen;
1460 break;
1461 }
1462 }
1463 }
1464
1465 /* Empty patch cannot be applied if it is a text patch
1466 * without metadata change. A binary patch appears
1467 * empty to us here.
1468 */
1469 if ((apply || check) &&
1470 (!patch->is_binary && !metadata_changes(patch)))
1471 die("patch with only garbage at line %d", linenr);
1472 }
1473
1474 return offset + hdrsize + patchsize;
1475 }
1476
1477 #define swap(a,b) myswap((a),(b),sizeof(a))
1478
1479 #define myswap(a, b, size) do { \
1480 unsigned char mytmp[size]; \
1481 memcpy(mytmp, &a, size); \
1482 memcpy(&a, &b, size); \
1483 memcpy(&b, mytmp, size); \
1484 } while (0)
1485
1486 static void reverse_patches(struct patch *p)
1487 {
1488 for (; p; p = p->next) {
1489 struct fragment *frag = p->fragments;
1490
1491 swap(p->new_name, p->old_name);
1492 swap(p->new_mode, p->old_mode);
1493 swap(p->is_new, p->is_delete);
1494 swap(p->lines_added, p->lines_deleted);
1495 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1496
1497 for (; frag; frag = frag->next) {
1498 swap(frag->newpos, frag->oldpos);
1499 swap(frag->newlines, frag->oldlines);
1500 }
1501 }
1502 }
1503
1504 static const char pluses[] =
1505 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1506 static const char minuses[]=
1507 "----------------------------------------------------------------------";
1508
1509 static void show_stats(struct patch *patch)
1510 {
1511 struct strbuf qname;
1512 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1513 int max, add, del;
1514
1515 strbuf_init(&qname, 0);
1516 quote_c_style(cp, &qname, NULL, 0);
1517
1518 /*
1519 * "scale" the filename
1520 */
1521 max = max_len;
1522 if (max > 50)
1523 max = 50;
1524
1525 if (qname.len > max) {
1526 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1527 if (!cp)
1528 cp = qname.buf + qname.len + 3 - max;
1529 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1530 }
1531
1532 if (patch->is_binary) {
1533 printf(" %-*s | Bin\n", max, qname.buf);
1534 strbuf_release(&qname);
1535 return;
1536 }
1537
1538 printf(" %-*s |", max, qname.buf);
1539 strbuf_release(&qname);
1540
1541 /*
1542 * scale the add/delete
1543 */
1544 max = max + max_change > 70 ? 70 - max : max_change;
1545 add = patch->lines_added;
1546 del = patch->lines_deleted;
1547
1548 if (max_change > 0) {
1549 int total = ((add + del) * max + max_change / 2) / max_change;
1550 add = (add * max + max_change / 2) / max_change;
1551 del = total - add;
1552 }
1553 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1554 add, pluses, del, minuses);
1555 }
1556
1557 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1558 {
1559 switch (st->st_mode & S_IFMT) {
1560 case S_IFLNK:
1561 strbuf_grow(buf, st->st_size);
1562 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1563 return -1;
1564 strbuf_setlen(buf, st->st_size);
1565 return 0;
1566 case S_IFREG:
1567 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1568 return error("unable to open or read %s", path);
1569 convert_to_git(path, buf->buf, buf->len, buf, 0);
1570 return 0;
1571 default:
1572 return -1;
1573 }
1574 }
1575
1576 static void update_pre_post_images(struct image *preimage,
1577 struct image *postimage,
1578 char *buf,
1579 size_t len)
1580 {
1581 int i, ctx;
1582 char *new, *old, *fixed;
1583 struct image fixed_preimage;
1584
1585 /*
1586 * Update the preimage with whitespace fixes. Note that we
1587 * are not losing preimage->buf -- apply_one_fragment() will
1588 * free "oldlines".
1589 */
1590 prepare_image(&fixed_preimage, buf, len, 1);
1591 assert(fixed_preimage.nr == preimage->nr);
1592 for (i = 0; i < preimage->nr; i++)
1593 fixed_preimage.line[i].flag = preimage->line[i].flag;
1594 free(preimage->line_allocated);
1595 *preimage = fixed_preimage;
1596
1597 /*
1598 * Adjust the common context lines in postimage, in place.
1599 * This is possible because whitespace fixing does not make
1600 * the string grow.
1601 */
1602 new = old = postimage->buf;
1603 fixed = preimage->buf;
1604 for (i = ctx = 0; i < postimage->nr; i++) {
1605 size_t len = postimage->line[i].len;
1606 if (!(postimage->line[i].flag & LINE_COMMON)) {
1607 /* an added line -- no counterparts in preimage */
1608 memmove(new, old, len);
1609 old += len;
1610 new += len;
1611 continue;
1612 }
1613
1614 /* a common context -- skip it in the original postimage */
1615 old += len;
1616
1617 /* and find the corresponding one in the fixed preimage */
1618 while (ctx < preimage->nr &&
1619 !(preimage->line[ctx].flag & LINE_COMMON)) {
1620 fixed += preimage->line[ctx].len;
1621 ctx++;
1622 }
1623 if (preimage->nr <= ctx)
1624 die("oops");
1625
1626 /* and copy it in, while fixing the line length */
1627 len = preimage->line[ctx].len;
1628 memcpy(new, fixed, len);
1629 new += len;
1630 fixed += len;
1631 postimage->line[i].len = len;
1632 ctx++;
1633 }
1634
1635 /* Fix the length of the whole thing */
1636 postimage->len = new - postimage->buf;
1637 }
1638
1639 static int match_fragment(struct image *img,
1640 struct image *preimage,
1641 struct image *postimage,
1642 unsigned long try,
1643 int try_lno,
1644 unsigned ws_rule,
1645 int match_beginning, int match_end)
1646 {
1647 int i;
1648 char *fixed_buf, *buf, *orig, *target;
1649
1650 if (preimage->nr + try_lno > img->nr)
1651 return 0;
1652
1653 if (match_beginning && try_lno)
1654 return 0;
1655
1656 if (match_end && preimage->nr + try_lno != img->nr)
1657 return 0;
1658
1659 /* Quick hash check */
1660 for (i = 0; i < preimage->nr; i++)
1661 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1662 return 0;
1663
1664 /*
1665 * Do we have an exact match? If we were told to match
1666 * at the end, size must be exactly at try+fragsize,
1667 * otherwise try+fragsize must be still within the preimage,
1668 * and either case, the old piece should match the preimage
1669 * exactly.
1670 */
1671 if ((match_end
1672 ? (try + preimage->len == img->len)
1673 : (try + preimage->len <= img->len)) &&
1674 !memcmp(img->buf + try, preimage->buf, preimage->len))
1675 return 1;
1676
1677 if (ws_error_action != correct_ws_error)
1678 return 0;
1679
1680 /*
1681 * The hunk does not apply byte-by-byte, but the hash says
1682 * it might with whitespace fuzz.
1683 */
1684 fixed_buf = xmalloc(preimage->len + 1);
1685 buf = fixed_buf;
1686 orig = preimage->buf;
1687 target = img->buf + try;
1688 for (i = 0; i < preimage->nr; i++) {
1689 size_t fixlen; /* length after fixing the preimage */
1690 size_t oldlen = preimage->line[i].len;
1691 size_t tgtlen = img->line[try_lno + i].len;
1692 size_t tgtfixlen; /* length after fixing the target line */
1693 char tgtfixbuf[1024], *tgtfix;
1694 int match;
1695
1696 /* Try fixing the line in the preimage */
1697 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1698
1699 /* Try fixing the line in the target */
1700 if (sizeof(tgtfixbuf) < tgtlen)
1701 tgtfix = tgtfixbuf;
1702 else
1703 tgtfix = xmalloc(tgtlen);
1704 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1705
1706 /*
1707 * If they match, either the preimage was based on
1708 * a version before our tree fixed whitespace breakage,
1709 * or we are lacking a whitespace-fix patch the tree
1710 * the preimage was based on already had (i.e. target
1711 * has whitespace breakage, the preimage doesn't).
1712 * In either case, we are fixing the whitespace breakages
1713 * so we might as well take the fix together with their
1714 * real change.
1715 */
1716 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1717
1718 if (tgtfix != tgtfixbuf)
1719 free(tgtfix);
1720 if (!match)
1721 goto unmatch_exit;
1722
1723 orig += oldlen;
1724 buf += fixlen;
1725 target += tgtlen;
1726 }
1727
1728 /*
1729 * Yes, the preimage is based on an older version that still
1730 * has whitespace breakages unfixed, and fixing them makes the
1731 * hunk match. Update the context lines in the postimage.
1732 */
1733 update_pre_post_images(preimage, postimage,
1734 fixed_buf, buf - fixed_buf);
1735 return 1;
1736
1737 unmatch_exit:
1738 free(fixed_buf);
1739 return 0;
1740 }
1741
1742 static int find_pos(struct image *img,
1743 struct image *preimage,
1744 struct image *postimage,
1745 int line,
1746 unsigned ws_rule,
1747 int match_beginning, int match_end)
1748 {
1749 int i;
1750 unsigned long backwards, forwards, try;
1751 int backwards_lno, forwards_lno, try_lno;
1752
1753 if (preimage->nr > img->nr)
1754 return -1;
1755
1756 /*
1757 * If match_begining or match_end is specified, there is no
1758 * point starting from a wrong line that will never match and
1759 * wander around and wait for a match at the specified end.
1760 */
1761 if (match_beginning)
1762 line = 0;
1763 else if (match_end)
1764 line = img->nr - preimage->nr;
1765
1766 if (line > img->nr)
1767 line = img->nr;
1768
1769 try = 0;
1770 for (i = 0; i < line; i++)
1771 try += img->line[i].len;
1772
1773 /*
1774 * There's probably some smart way to do this, but I'll leave
1775 * that to the smart and beautiful people. I'm simple and stupid.
1776 */
1777 backwards = try;
1778 backwards_lno = line;
1779 forwards = try;
1780 forwards_lno = line;
1781 try_lno = line;
1782
1783 for (i = 0; ; i++) {
1784 if (match_fragment(img, preimage, postimage,
1785 try, try_lno, ws_rule,
1786 match_beginning, match_end))
1787 return try_lno;
1788
1789 again:
1790 if (backwards_lno == 0 && forwards_lno == img->nr)
1791 break;
1792
1793 if (i & 1) {
1794 if (backwards_lno == 0) {
1795 i++;
1796 goto again;
1797 }
1798 backwards_lno--;
1799 backwards -= img->line[backwards_lno].len;
1800 try = backwards;
1801 try_lno = backwards_lno;
1802 } else {
1803 if (forwards_lno == img->nr) {
1804 i++;
1805 goto again;
1806 }
1807 forwards += img->line[forwards_lno].len;
1808 forwards_lno++;
1809 try = forwards;
1810 try_lno = forwards_lno;
1811 }
1812
1813 }
1814 return -1;
1815 }
1816
1817 static void remove_first_line(struct image *img)
1818 {
1819 img->buf += img->line[0].len;
1820 img->len -= img->line[0].len;
1821 img->line++;
1822 img->nr--;
1823 }
1824
1825 static void remove_last_line(struct image *img)
1826 {
1827 img->len -= img->line[--img->nr].len;
1828 }
1829
1830 static void update_image(struct image *img,
1831 int applied_pos,
1832 struct image *preimage,
1833 struct image *postimage)
1834 {
1835 /*
1836 * remove the copy of preimage at offset in img
1837 * and replace it with postimage
1838 */
1839 int i, nr;
1840 size_t remove_count, insert_count, applied_at = 0;
1841 char *result;
1842
1843 for (i = 0; i < applied_pos; i++)
1844 applied_at += img->line[i].len;
1845
1846 remove_count = 0;
1847 for (i = 0; i < preimage->nr; i++)
1848 remove_count += img->line[applied_pos + i].len;
1849 insert_count = postimage->len;
1850
1851 /* Adjust the contents */
1852 result = xmalloc(img->len + insert_count - remove_count + 1);
1853 memcpy(result, img->buf, applied_at);
1854 memcpy(result + applied_at, postimage->buf, postimage->len);
1855 memcpy(result + applied_at + postimage->len,
1856 img->buf + (applied_at + remove_count),
1857 img->len - (applied_at + remove_count));
1858 free(img->buf);
1859 img->buf = result;
1860 img->len += insert_count - remove_count;
1861 result[img->len] = '\0';
1862
1863 /* Adjust the line table */
1864 nr = img->nr + postimage->nr - preimage->nr;
1865 if (preimage->nr < postimage->nr) {
1866 /*
1867 * NOTE: this knows that we never call remove_first_line()
1868 * on anything other than pre/post image.
1869 */
1870 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1871 img->line_allocated = img->line;
1872 }
1873 if (preimage->nr != postimage->nr)
1874 memmove(img->line + applied_pos + postimage->nr,
1875 img->line + applied_pos + preimage->nr,
1876 (img->nr - (applied_pos + preimage->nr)) *
1877 sizeof(*img->line));
1878 memcpy(img->line + applied_pos,
1879 postimage->line,
1880 postimage->nr * sizeof(*img->line));
1881 img->nr = nr;
1882 }
1883
1884 static int apply_one_fragment(struct image *img, struct fragment *frag,
1885 int inaccurate_eof, unsigned ws_rule)
1886 {
1887 int match_beginning, match_end;
1888 const char *patch = frag->patch;
1889 int size = frag->size;
1890 char *old, *new, *oldlines, *newlines;
1891 int new_blank_lines_at_end = 0;
1892 unsigned long leading, trailing;
1893 int pos, applied_pos;
1894 struct image preimage;
1895 struct image postimage;
1896
1897 memset(&preimage, 0, sizeof(preimage));
1898 memset(&postimage, 0, sizeof(postimage));
1899 oldlines = xmalloc(size);
1900 newlines = xmalloc(size);
1901
1902 old = oldlines;
1903 new = newlines;
1904 while (size > 0) {
1905 char first;
1906 int len = linelen(patch, size);
1907 int plen, added;
1908 int added_blank_line = 0;
1909
1910 if (!len)
1911 break;
1912
1913 /*
1914 * "plen" is how much of the line we should use for
1915 * the actual patch data. Normally we just remove the
1916 * first character on the line, but if the line is
1917 * followed by "\ No newline", then we also remove the
1918 * last one (which is the newline, of course).
1919 */
1920 plen = len - 1;
1921 if (len < size && patch[len] == '\\')
1922 plen--;
1923 first = *patch;
1924 if (apply_in_reverse) {
1925 if (first == '-')
1926 first = '+';
1927 else if (first == '+')
1928 first = '-';
1929 }
1930
1931 switch (first) {
1932 case '\n':
1933 /* Newer GNU diff, empty context line */
1934 if (plen < 0)
1935 /* ... followed by '\No newline'; nothing */
1936 break;
1937 *old++ = '\n';
1938 *new++ = '\n';
1939 add_line_info(&preimage, "\n", 1, LINE_COMMON);
1940 add_line_info(&postimage, "\n", 1, LINE_COMMON);
1941 break;
1942 case ' ':
1943 case '-':
1944 memcpy(old, patch + 1, plen);
1945 add_line_info(&preimage, old, plen,
1946 (first == ' ' ? LINE_COMMON : 0));
1947 old += plen;
1948 if (first == '-')
1949 break;
1950 /* Fall-through for ' ' */
1951 case '+':
1952 /* --no-add does not add new lines */
1953 if (first == '+' && no_add)
1954 break;
1955
1956 if (first != '+' ||
1957 !whitespace_error ||
1958 ws_error_action != correct_ws_error) {
1959 memcpy(new, patch + 1, plen);
1960 added = plen;
1961 }
1962 else {
1963 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
1964 }
1965 add_line_info(&postimage, new, added,
1966 (first == '+' ? 0 : LINE_COMMON));
1967 new += added;
1968 if (first == '+' &&
1969 added == 1 && new[-1] == '\n')
1970 added_blank_line = 1;
1971 break;
1972 case '@': case '\\':
1973 /* Ignore it, we already handled it */
1974 break;
1975 default:
1976 if (apply_verbosely)
1977 error("invalid start of line: '%c'", first);
1978 return -1;
1979 }
1980 if (added_blank_line)
1981 new_blank_lines_at_end++;
1982 else
1983 new_blank_lines_at_end = 0;
1984 patch += len;
1985 size -= len;
1986 }
1987 if (inaccurate_eof &&
1988 old > oldlines && old[-1] == '\n' &&
1989 new > newlines && new[-1] == '\n') {
1990 old--;
1991 new--;
1992 }
1993
1994 leading = frag->leading;
1995 trailing = frag->trailing;
1996
1997 /*
1998 * A hunk to change lines at the beginning would begin with
1999 * @@ -1,L +N,M @@
2000 * but we need to be careful. -U0 that inserts before the second
2001 * line also has this pattern.
2002 *
2003 * And a hunk to add to an empty file would begin with
2004 * @@ -0,0 +N,M @@
2005 *
2006 * In other words, a hunk that is (frag->oldpos <= 1) with or
2007 * without leading context must match at the beginning.
2008 */
2009 match_beginning = (!frag->oldpos ||
2010 (frag->oldpos == 1 && !unidiff_zero));
2011
2012 /*
2013 * A hunk without trailing lines must match at the end.
2014 * However, we simply cannot tell if a hunk must match end
2015 * from the lack of trailing lines if the patch was generated
2016 * with unidiff without any context.
2017 */
2018 match_end = !unidiff_zero && !trailing;
2019
2020 pos = frag->newpos ? (frag->newpos - 1) : 0;
2021 preimage.buf = oldlines;
2022 preimage.len = old - oldlines;
2023 postimage.buf = newlines;
2024 postimage.len = new - newlines;
2025 preimage.line = preimage.line_allocated;
2026 postimage.line = postimage.line_allocated;
2027
2028 for (;;) {
2029
2030 applied_pos = find_pos(img, &preimage, &postimage, pos,
2031 ws_rule, match_beginning, match_end);
2032
2033 if (applied_pos >= 0)
2034 break;
2035
2036 /* Am I at my context limits? */
2037 if ((leading <= p_context) && (trailing <= p_context))
2038 break;
2039 if (match_beginning || match_end) {
2040 match_beginning = match_end = 0;
2041 continue;
2042 }
2043
2044 /*
2045 * Reduce the number of context lines; reduce both
2046 * leading and trailing if they are equal otherwise
2047 * just reduce the larger context.
2048 */
2049 if (leading >= trailing) {
2050 remove_first_line(&preimage);
2051 remove_first_line(&postimage);
2052 pos--;
2053 leading--;
2054 }
2055 if (trailing > leading) {
2056 remove_last_line(&preimage);
2057 remove_last_line(&postimage);
2058 trailing--;
2059 }
2060 }
2061
2062 if (applied_pos >= 0) {
2063 if (ws_error_action == correct_ws_error &&
2064 new_blank_lines_at_end &&
2065 postimage.nr + applied_pos == img->nr) {
2066 /*
2067 * If the patch application adds blank lines
2068 * at the end, and if the patch applies at the
2069 * end of the image, remove those added blank
2070 * lines.
2071 */
2072 while (new_blank_lines_at_end--)
2073 remove_last_line(&postimage);
2074 }
2075
2076 /*
2077 * Warn if it was necessary to reduce the number
2078 * of context lines.
2079 */
2080 if ((leading != frag->leading) ||
2081 (trailing != frag->trailing))
2082 fprintf(stderr, "Context reduced to (%ld/%ld)"
2083 " to apply fragment at %d\n",
2084 leading, trailing, applied_pos+1);
2085 update_image(img, applied_pos, &preimage, &postimage);
2086 } else {
2087 if (apply_verbosely)
2088 error("while searching for:\n%.*s",
2089 (int)(old - oldlines), oldlines);
2090 }
2091
2092 free(oldlines);
2093 free(newlines);
2094 free(preimage.line_allocated);
2095 free(postimage.line_allocated);
2096
2097 return (applied_pos < 0);
2098 }
2099
2100 static int apply_binary_fragment(struct image *img, struct patch *patch)
2101 {
2102 struct fragment *fragment = patch->fragments;
2103 unsigned long len;
2104 void *dst;
2105
2106 /* Binary patch is irreversible without the optional second hunk */
2107 if (apply_in_reverse) {
2108 if (!fragment->next)
2109 return error("cannot reverse-apply a binary patch "
2110 "without the reverse hunk to '%s'",
2111 patch->new_name
2112 ? patch->new_name : patch->old_name);
2113 fragment = fragment->next;
2114 }
2115 switch (fragment->binary_patch_method) {
2116 case BINARY_DELTA_DEFLATED:
2117 dst = patch_delta(img->buf, img->len, fragment->patch,
2118 fragment->size, &len);
2119 if (!dst)
2120 return -1;
2121 clear_image(img);
2122 img->buf = dst;
2123 img->len = len;
2124 return 0;
2125 case BINARY_LITERAL_DEFLATED:
2126 clear_image(img);
2127 img->len = fragment->size;
2128 img->buf = xmalloc(img->len+1);
2129 memcpy(img->buf, fragment->patch, img->len);
2130 img->buf[img->len] = '\0';
2131 return 0;
2132 }
2133 return -1;
2134 }
2135
2136 static int apply_binary(struct image *img, struct patch *patch)
2137 {
2138 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2139 unsigned char sha1[20];
2140
2141 /*
2142 * For safety, we require patch index line to contain
2143 * full 40-byte textual SHA1 for old and new, at least for now.
2144 */
2145 if (strlen(patch->old_sha1_prefix) != 40 ||
2146 strlen(patch->new_sha1_prefix) != 40 ||
2147 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2148 get_sha1_hex(patch->new_sha1_prefix, sha1))
2149 return error("cannot apply binary patch to '%s' "
2150 "without full index line", name);
2151
2152 if (patch->old_name) {
2153 /*
2154 * See if the old one matches what the patch
2155 * applies to.
2156 */
2157 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2158 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2159 return error("the patch applies to '%s' (%s), "
2160 "which does not match the "
2161 "current contents.",
2162 name, sha1_to_hex(sha1));
2163 }
2164 else {
2165 /* Otherwise, the old one must be empty. */
2166 if (img->len)
2167 return error("the patch applies to an empty "
2168 "'%s' but it is not empty", name);
2169 }
2170
2171 get_sha1_hex(patch->new_sha1_prefix, sha1);
2172 if (is_null_sha1(sha1)) {
2173 clear_image(img);
2174 return 0; /* deletion patch */
2175 }
2176
2177 if (has_sha1_file(sha1)) {
2178 /* We already have the postimage */
2179 enum object_type type;
2180 unsigned long size;
2181 char *result;
2182
2183 result = read_sha1_file(sha1, &type, &size);
2184 if (!result)
2185 return error("the necessary postimage %s for "
2186 "'%s' cannot be read",
2187 patch->new_sha1_prefix, name);
2188 clear_image(img);
2189 img->buf = result;
2190 img->len = size;
2191 } else {
2192 /*
2193 * We have verified buf matches the preimage;
2194 * apply the patch data to it, which is stored
2195 * in the patch->fragments->{patch,size}.
2196 */
2197 if (apply_binary_fragment(img, patch))
2198 return error("binary patch does not apply to '%s'",
2199 name);
2200
2201 /* verify that the result matches */
2202 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2203 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2204 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2205 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2206 }
2207
2208 return 0;
2209 }
2210
2211 static int apply_fragments(struct image *img, struct patch *patch)
2212 {
2213 struct fragment *frag = patch->fragments;
2214 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2215 unsigned ws_rule = patch->ws_rule;
2216 unsigned inaccurate_eof = patch->inaccurate_eof;
2217
2218 if (patch->is_binary)
2219 return apply_binary(img, patch);
2220
2221 while (frag) {
2222 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2223 error("patch failed: %s:%ld", name, frag->oldpos);
2224 if (!apply_with_reject)
2225 return -1;
2226 frag->rejected = 1;
2227 }
2228 frag = frag->next;
2229 }
2230 return 0;
2231 }
2232
2233 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2234 {
2235 if (!ce)
2236 return 0;
2237
2238 if (S_ISGITLINK(ce->ce_mode)) {
2239 strbuf_grow(buf, 100);
2240 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2241 } else {
2242 enum object_type type;
2243 unsigned long sz;
2244 char *result;
2245
2246 result = read_sha1_file(ce->sha1, &type, &sz);
2247 if (!result)
2248 return -1;
2249 /* XXX read_sha1_file NUL-terminates */
2250 strbuf_attach(buf, result, sz, sz + 1);
2251 }
2252 return 0;
2253 }
2254
2255 static struct patch *in_fn_table(const char *name)
2256 {
2257 struct string_list_item *item;
2258
2259 if (name == NULL)
2260 return NULL;
2261
2262 item = string_list_lookup(name, &fn_table);
2263 if (item != NULL)
2264 return (struct patch *)item->util;
2265
2266 return NULL;
2267 }
2268
2269 static void add_to_fn_table(struct patch *patch)
2270 {
2271 struct string_list_item *item;
2272
2273 /*
2274 * Always add new_name unless patch is a deletion
2275 * This should cover the cases for normal diffs,
2276 * file creations and copies
2277 */
2278 if (patch->new_name != NULL) {
2279 item = string_list_insert(patch->new_name, &fn_table);
2280 item->util = patch;
2281 }
2282
2283 /*
2284 * store a failure on rename/deletion cases because
2285 * later chunks shouldn't patch old names
2286 */
2287 if ((patch->new_name == NULL) || (patch->is_rename)) {
2288 item = string_list_insert(patch->old_name, &fn_table);
2289 item->util = (struct patch *) -1;
2290 }
2291 }
2292
2293 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2294 {
2295 struct strbuf buf;
2296 struct image image;
2297 size_t len;
2298 char *img;
2299 struct patch *tpatch;
2300
2301 strbuf_init(&buf, 0);
2302
2303 if (!(patch->is_copy || patch->is_rename) &&
2304 ((tpatch = in_fn_table(patch->old_name)) != NULL)) {
2305 if (tpatch == (struct patch *) -1) {
2306 return error("patch %s has been renamed/deleted",
2307 patch->old_name);
2308 }
2309 /* We have a patched copy in memory use that */
2310 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2311 } else if (cached) {
2312 if (read_file_or_gitlink(ce, &buf))
2313 return error("read of %s failed", patch->old_name);
2314 } else if (patch->old_name) {
2315 if (S_ISGITLINK(patch->old_mode)) {
2316 if (ce) {
2317 read_file_or_gitlink(ce, &buf);
2318 } else {
2319 /*
2320 * There is no way to apply subproject
2321 * patch without looking at the index.
2322 */
2323 patch->fragments = NULL;
2324 }
2325 } else {
2326 if (read_old_data(st, patch->old_name, &buf))
2327 return error("read of %s failed", patch->old_name);
2328 }
2329 }
2330
2331 img = strbuf_detach(&buf, &len);
2332 prepare_image(&image, img, len, !patch->is_binary);
2333
2334 if (apply_fragments(&image, patch) < 0)
2335 return -1; /* note with --reject this succeeds. */
2336 patch->result = image.buf;
2337 patch->resultsize = image.len;
2338 add_to_fn_table(patch);
2339 free(image.line_allocated);
2340
2341 if (0 < patch->is_delete && patch->resultsize)
2342 return error("removal patch leaves file contents");
2343
2344 return 0;
2345 }
2346
2347 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2348 {
2349 struct stat nst;
2350 if (!lstat(new_name, &nst)) {
2351 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2352 return 0;
2353 /*
2354 * A leading component of new_name might be a symlink
2355 * that is going to be removed with this patch, but
2356 * still pointing at somewhere that has the path.
2357 * In such a case, path "new_name" does not exist as
2358 * far as git is concerned.
2359 */
2360 if (has_symlink_leading_path(strlen(new_name), new_name))
2361 return 0;
2362
2363 return error("%s: already exists in working directory", new_name);
2364 }
2365 else if ((errno != ENOENT) && (errno != ENOTDIR))
2366 return error("%s: %s", new_name, strerror(errno));
2367 return 0;
2368 }
2369
2370 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2371 {
2372 if (S_ISGITLINK(ce->ce_mode)) {
2373 if (!S_ISDIR(st->st_mode))
2374 return -1;
2375 return 0;
2376 }
2377 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2378 }
2379
2380 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2381 {
2382 const char *old_name = patch->old_name;
2383 struct patch *tpatch = NULL;
2384 int stat_ret = 0;
2385 unsigned st_mode = 0;
2386
2387 /*
2388 * Make sure that we do not have local modifications from the
2389 * index when we are looking at the index. Also make sure
2390 * we have the preimage file to be patched in the work tree,
2391 * unless --cached, which tells git to apply only in the index.
2392 */
2393 if (!old_name)
2394 return 0;
2395
2396 assert(patch->is_new <= 0);
2397
2398 if (!(patch->is_copy || patch->is_rename) &&
2399 (tpatch = in_fn_table(old_name)) != NULL) {
2400 if (tpatch == (struct patch *) -1) {
2401 return error("%s: has been deleted/renamed", old_name);
2402 }
2403 st_mode = tpatch->new_mode;
2404 } else if (!cached) {
2405 stat_ret = lstat(old_name, st);
2406 if (stat_ret && errno != ENOENT)
2407 return error("%s: %s", old_name, strerror(errno));
2408 }
2409
2410 if (check_index && !tpatch) {
2411 int pos = cache_name_pos(old_name, strlen(old_name));
2412 if (pos < 0) {
2413 if (patch->is_new < 0)
2414 goto is_new;
2415 return error("%s: does not exist in index", old_name);
2416 }
2417 *ce = active_cache[pos];
2418 if (stat_ret < 0) {
2419 struct checkout costate;
2420 /* checkout */
2421 costate.base_dir = "";
2422 costate.base_dir_len = 0;
2423 costate.force = 0;
2424 costate.quiet = 0;
2425 costate.not_new = 0;
2426 costate.refresh_cache = 1;
2427 if (checkout_entry(*ce, &costate, NULL) ||
2428 lstat(old_name, st))
2429 return -1;
2430 }
2431 if (!cached && verify_index_match(*ce, st))
2432 return error("%s: does not match index", old_name);
2433 if (cached)
2434 st_mode = (*ce)->ce_mode;
2435 } else if (stat_ret < 0) {
2436 if (patch->is_new < 0)
2437 goto is_new;
2438 return error("%s: %s", old_name, strerror(errno));
2439 }
2440
2441 if (!cached)
2442 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2443
2444 if (patch->is_new < 0)
2445 patch->is_new = 0;
2446 if (!patch->old_mode)
2447 patch->old_mode = st_mode;
2448 if ((st_mode ^ patch->old_mode) & S_IFMT)
2449 return error("%s: wrong type", old_name);
2450 if (st_mode != patch->old_mode)
2451 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2452 old_name, st_mode, patch->old_mode);
2453 return 0;
2454
2455 is_new:
2456 patch->is_new = 1;
2457 patch->is_delete = 0;
2458 patch->old_name = NULL;
2459 return 0;
2460 }
2461
2462 static int check_patch(struct patch *patch)
2463 {
2464 struct stat st;
2465 const char *old_name = patch->old_name;
2466 const char *new_name = patch->new_name;
2467 const char *name = old_name ? old_name : new_name;
2468 struct cache_entry *ce = NULL;
2469 int ok_if_exists;
2470 int status;
2471
2472 patch->rejected = 1; /* we will drop this after we succeed */
2473
2474 status = check_preimage(patch, &ce, &st);
2475 if (status)
2476 return status;
2477 old_name = patch->old_name;
2478
2479 if (in_fn_table(new_name) == (struct patch *) -1)
2480 /*
2481 * A type-change diff is always split into a patch to
2482 * delete old, immediately followed by a patch to
2483 * create new (see diff.c::run_diff()); in such a case
2484 * it is Ok that the entry to be deleted by the
2485 * previous patch is still in the working tree and in
2486 * the index.
2487 */
2488 ok_if_exists = 1;
2489 else
2490 ok_if_exists = 0;
2491
2492 if (new_name &&
2493 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2494 if (check_index &&
2495 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2496 !ok_if_exists)
2497 return error("%s: already exists in index", new_name);
2498 if (!cached) {
2499 int err = check_to_create_blob(new_name, ok_if_exists);
2500 if (err)
2501 return err;
2502 }
2503 if (!patch->new_mode) {
2504 if (0 < patch->is_new)
2505 patch->new_mode = S_IFREG | 0644;
2506 else
2507 patch->new_mode = patch->old_mode;
2508 }
2509 }
2510
2511 if (new_name && old_name) {
2512 int same = !strcmp(old_name, new_name);
2513 if (!patch->new_mode)
2514 patch->new_mode = patch->old_mode;
2515 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2516 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2517 patch->new_mode, new_name, patch->old_mode,
2518 same ? "" : " of ", same ? "" : old_name);
2519 }
2520
2521 if (apply_data(patch, &st, ce) < 0)
2522 return error("%s: patch does not apply", name);
2523 patch->rejected = 0;
2524 return 0;
2525 }
2526
2527 static int check_patch_list(struct patch *patch)
2528 {
2529 int err = 0;
2530
2531 while (patch) {
2532 if (apply_verbosely)
2533 say_patch_name(stderr,
2534 "Checking patch ", patch, "...\n");
2535 err |= check_patch(patch);
2536 patch = patch->next;
2537 }
2538 return err;
2539 }
2540
2541 /* This function tries to read the sha1 from the current index */
2542 static int get_current_sha1(const char *path, unsigned char *sha1)
2543 {
2544 int pos;
2545
2546 if (read_cache() < 0)
2547 return -1;
2548 pos = cache_name_pos(path, strlen(path));
2549 if (pos < 0)
2550 return -1;
2551 hashcpy(sha1, active_cache[pos]->sha1);
2552 return 0;
2553 }
2554
2555 /* Build an index that contains the just the files needed for a 3way merge */
2556 static void build_fake_ancestor(struct patch *list, const char *filename)
2557 {
2558 struct patch *patch;
2559 struct index_state result = { 0 };
2560 int fd;
2561
2562 /* Once we start supporting the reverse patch, it may be
2563 * worth showing the new sha1 prefix, but until then...
2564 */
2565 for (patch = list; patch; patch = patch->next) {
2566 const unsigned char *sha1_ptr;
2567 unsigned char sha1[20];
2568 struct cache_entry *ce;
2569 const char *name;
2570
2571 name = patch->old_name ? patch->old_name : patch->new_name;
2572 if (0 < patch->is_new)
2573 continue;
2574 else if (get_sha1(patch->old_sha1_prefix, sha1))
2575 /* git diff has no index line for mode/type changes */
2576 if (!patch->lines_added && !patch->lines_deleted) {
2577 if (get_current_sha1(patch->new_name, sha1) ||
2578 get_current_sha1(patch->old_name, sha1))
2579 die("mode change for %s, which is not "
2580 "in current HEAD", name);
2581 sha1_ptr = sha1;
2582 } else
2583 die("sha1 information is lacking or useless "
2584 "(%s).", name);
2585 else
2586 sha1_ptr = sha1;
2587
2588 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2589 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2590 die ("Could not add %s to temporary index", name);
2591 }
2592
2593 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2594 if (fd < 0 || write_index(&result, fd) || close(fd))
2595 die ("Could not write temporary index to %s", filename);
2596
2597 discard_index(&result);
2598 }
2599
2600 static void stat_patch_list(struct patch *patch)
2601 {
2602 int files, adds, dels;
2603
2604 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2605 files++;
2606 adds += patch->lines_added;
2607 dels += patch->lines_deleted;
2608 show_stats(patch);
2609 }
2610
2611 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2612 }
2613
2614 static void numstat_patch_list(struct patch *patch)
2615 {
2616 for ( ; patch; patch = patch->next) {
2617 const char *name;
2618 name = patch->new_name ? patch->new_name : patch->old_name;
2619 if (patch->is_binary)
2620 printf("-\t-\t");
2621 else
2622 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2623 write_name_quoted(name, stdout, line_termination);
2624 }
2625 }
2626
2627 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2628 {
2629 if (mode)
2630 printf(" %s mode %06o %s\n", newdelete, mode, name);
2631 else
2632 printf(" %s %s\n", newdelete, name);
2633 }
2634
2635 static void show_mode_change(struct patch *p, int show_name)
2636 {
2637 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2638 if (show_name)
2639 printf(" mode change %06o => %06o %s\n",
2640 p->old_mode, p->new_mode, p->new_name);
2641 else
2642 printf(" mode change %06o => %06o\n",
2643 p->old_mode, p->new_mode);
2644 }
2645 }
2646
2647 static void show_rename_copy(struct patch *p)
2648 {
2649 const char *renamecopy = p->is_rename ? "rename" : "copy";
2650 const char *old, *new;
2651
2652 /* Find common prefix */
2653 old = p->old_name;
2654 new = p->new_name;
2655 while (1) {
2656 const char *slash_old, *slash_new;
2657 slash_old = strchr(old, '/');
2658 slash_new = strchr(new, '/');
2659 if (!slash_old ||
2660 !slash_new ||
2661 slash_old - old != slash_new - new ||
2662 memcmp(old, new, slash_new - new))
2663 break;
2664 old = slash_old + 1;
2665 new = slash_new + 1;
2666 }
2667 /* p->old_name thru old is the common prefix, and old and new
2668 * through the end of names are renames
2669 */
2670 if (old != p->old_name)
2671 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2672 (int)(old - p->old_name), p->old_name,
2673 old, new, p->score);
2674 else
2675 printf(" %s %s => %s (%d%%)\n", renamecopy,
2676 p->old_name, p->new_name, p->score);
2677 show_mode_change(p, 0);
2678 }
2679
2680 static void summary_patch_list(struct patch *patch)
2681 {
2682 struct patch *p;
2683
2684 for (p = patch; p; p = p->next) {
2685 if (p->is_new)
2686 show_file_mode_name("create", p->new_mode, p->new_name);
2687 else if (p->is_delete)
2688 show_file_mode_name("delete", p->old_mode, p->old_name);
2689 else {
2690 if (p->is_rename || p->is_copy)
2691 show_rename_copy(p);
2692 else {
2693 if (p->score) {
2694 printf(" rewrite %s (%d%%)\n",
2695 p->new_name, p->score);
2696 show_mode_change(p, 0);
2697 }
2698 else
2699 show_mode_change(p, 1);
2700 }
2701 }
2702 }
2703 }
2704
2705 static void patch_stats(struct patch *patch)
2706 {
2707 int lines = patch->lines_added + patch->lines_deleted;
2708
2709 if (lines > max_change)
2710 max_change = lines;
2711 if (patch->old_name) {
2712 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2713 if (!len)
2714 len = strlen(patch->old_name);
2715 if (len > max_len)
2716 max_len = len;
2717 }
2718 if (patch->new_name) {
2719 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2720 if (!len)
2721 len = strlen(patch->new_name);
2722 if (len > max_len)
2723 max_len = len;
2724 }
2725 }
2726
2727 static void remove_file(struct patch *patch, int rmdir_empty)
2728 {
2729 if (update_index) {
2730 if (remove_file_from_cache(patch->old_name) < 0)
2731 die("unable to remove %s from index", patch->old_name);
2732 }
2733 if (!cached) {
2734 if (S_ISGITLINK(patch->old_mode)) {
2735 if (rmdir(patch->old_name))
2736 warning("unable to remove submodule %s",
2737 patch->old_name);
2738 } else if (!unlink(patch->old_name) && rmdir_empty) {
2739 remove_path(patch->old_name);
2740 }
2741 }
2742 }
2743
2744 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2745 {
2746 struct stat st;
2747 struct cache_entry *ce;
2748 int namelen = strlen(path);
2749 unsigned ce_size = cache_entry_size(namelen);
2750
2751 if (!update_index)
2752 return;
2753
2754 ce = xcalloc(1, ce_size);
2755 memcpy(ce->name, path, namelen);
2756 ce->ce_mode = create_ce_mode(mode);
2757 ce->ce_flags = namelen;
2758 if (S_ISGITLINK(mode)) {
2759 const char *s = buf;
2760
2761 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2762 die("corrupt patch for subproject %s", path);
2763 } else {
2764 if (!cached) {
2765 if (lstat(path, &st) < 0)
2766 die("unable to stat newly created file %s",
2767 path);
2768 fill_stat_cache_info(ce, &st);
2769 }
2770 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2771 die("unable to create backing store for newly created file %s", path);
2772 }
2773 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2774 die("unable to add cache entry for %s", path);
2775 }
2776
2777 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2778 {
2779 int fd;
2780 struct strbuf nbuf;
2781
2782 if (S_ISGITLINK(mode)) {
2783 struct stat st;
2784 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2785 return 0;
2786 return mkdir(path, 0777);
2787 }
2788
2789 if (has_symlinks && S_ISLNK(mode))
2790 /* Although buf:size is counted string, it also is NUL
2791 * terminated.
2792 */
2793 return symlink(buf, path);
2794
2795 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2796 if (fd < 0)
2797 return -1;
2798
2799 strbuf_init(&nbuf, 0);
2800 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2801 size = nbuf.len;
2802 buf = nbuf.buf;
2803 }
2804 write_or_die(fd, buf, size);
2805 strbuf_release(&nbuf);
2806
2807 if (close(fd) < 0)
2808 die("closing file %s: %s", path, strerror(errno));
2809 return 0;
2810 }
2811
2812 /*
2813 * We optimistically assume that the directories exist,
2814 * which is true 99% of the time anyway. If they don't,
2815 * we create them and try again.
2816 */
2817 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2818 {
2819 if (cached)
2820 return;
2821 if (!try_create_file(path, mode, buf, size))
2822 return;
2823
2824 if (errno == ENOENT) {
2825 if (safe_create_leading_directories(path))
2826 return;
2827 if (!try_create_file(path, mode, buf, size))
2828 return;
2829 }
2830
2831 if (errno == EEXIST || errno == EACCES) {
2832 /* We may be trying to create a file where a directory
2833 * used to be.
2834 */
2835 struct stat st;
2836 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2837 errno = EEXIST;
2838 }
2839
2840 if (errno == EEXIST) {
2841 unsigned int nr = getpid();
2842
2843 for (;;) {
2844 const char *newpath;
2845 newpath = mkpath("%s~%u", path, nr);
2846 if (!try_create_file(newpath, mode, buf, size)) {
2847 if (!rename(newpath, path))
2848 return;
2849 unlink(newpath);
2850 break;
2851 }
2852 if (errno != EEXIST)
2853 break;
2854 ++nr;
2855 }
2856 }
2857 die("unable to write file %s mode %o", path, mode);
2858 }
2859
2860 static void create_file(struct patch *patch)
2861 {
2862 char *path = patch->new_name;
2863 unsigned mode = patch->new_mode;
2864 unsigned long size = patch->resultsize;
2865 char *buf = patch->result;
2866
2867 if (!mode)
2868 mode = S_IFREG | 0644;
2869 create_one_file(path, mode, buf, size);
2870 add_index_file(path, mode, buf, size);
2871 }
2872
2873 /* phase zero is to remove, phase one is to create */
2874 static void write_out_one_result(struct patch *patch, int phase)
2875 {
2876 if (patch->is_delete > 0) {
2877 if (phase == 0)
2878 remove_file(patch, 1);
2879 return;
2880 }
2881 if (patch->is_new > 0 || patch->is_copy) {
2882 if (phase == 1)
2883 create_file(patch);
2884 return;
2885 }
2886 /*
2887 * Rename or modification boils down to the same
2888 * thing: remove the old, write the new
2889 */
2890 if (phase == 0)
2891 remove_file(patch, patch->is_rename);
2892 if (phase == 1)
2893 create_file(patch);
2894 }
2895
2896 static int write_out_one_reject(struct patch *patch)
2897 {
2898 FILE *rej;
2899 char namebuf[PATH_MAX];
2900 struct fragment *frag;
2901 int cnt = 0;
2902
2903 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2904 if (!frag->rejected)
2905 continue;
2906 cnt++;
2907 }
2908
2909 if (!cnt) {
2910 if (apply_verbosely)
2911 say_patch_name(stderr,
2912 "Applied patch ", patch, " cleanly.\n");
2913 return 0;
2914 }
2915
2916 /* This should not happen, because a removal patch that leaves
2917 * contents are marked "rejected" at the patch level.
2918 */
2919 if (!patch->new_name)
2920 die("internal error");
2921
2922 /* Say this even without --verbose */
2923 say_patch_name(stderr, "Applying patch ", patch, " with");
2924 fprintf(stderr, " %d rejects...\n", cnt);
2925
2926 cnt = strlen(patch->new_name);
2927 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2928 cnt = ARRAY_SIZE(namebuf) - 5;
2929 fprintf(stderr,
2930 "warning: truncating .rej filename to %.*s.rej",
2931 cnt - 1, patch->new_name);
2932 }
2933 memcpy(namebuf, patch->new_name, cnt);
2934 memcpy(namebuf + cnt, ".rej", 5);
2935
2936 rej = fopen(namebuf, "w");
2937 if (!rej)
2938 return error("cannot open %s: %s", namebuf, strerror(errno));
2939
2940 /* Normal git tools never deal with .rej, so do not pretend
2941 * this is a git patch by saying --git nor give extended
2942 * headers. While at it, maybe please "kompare" that wants
2943 * the trailing TAB and some garbage at the end of line ;-).
2944 */
2945 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2946 patch->new_name, patch->new_name);
2947 for (cnt = 1, frag = patch->fragments;
2948 frag;
2949 cnt++, frag = frag->next) {
2950 if (!frag->rejected) {
2951 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2952 continue;
2953 }
2954 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2955 fprintf(rej, "%.*s", frag->size, frag->patch);
2956 if (frag->patch[frag->size-1] != '\n')
2957 fputc('\n', rej);
2958 }
2959 fclose(rej);
2960 return -1;
2961 }
2962
2963 static int write_out_results(struct patch *list, int skipped_patch)
2964 {
2965 int phase;
2966 int errs = 0;
2967 struct patch *l;
2968
2969 if (!list && !skipped_patch)
2970 return error("No changes");
2971
2972 for (phase = 0; phase < 2; phase++) {
2973 l = list;
2974 while (l) {
2975 if (l->rejected)
2976 errs = 1;
2977 else {
2978 write_out_one_result(l, phase);
2979 if (phase == 1 && write_out_one_reject(l))
2980 errs = 1;
2981 }
2982 l = l->next;
2983 }
2984 }
2985 return errs;
2986 }
2987
2988 static struct lock_file lock_file;
2989
2990 static struct string_list limit_by_name;
2991 static int has_include;
2992 static void add_name_limit(const char *name, int exclude)
2993 {
2994 struct string_list_item *it;
2995
2996 it = string_list_append(name, &limit_by_name);
2997 it->util = exclude ? NULL : (void *) 1;
2998 }
2999
3000 static int use_patch(struct patch *p)
3001 {
3002 const char *pathname = p->new_name ? p->new_name : p->old_name;
3003 int i;
3004
3005 /* Paths outside are not touched regardless of "--include" */
3006 if (0 < prefix_length) {
3007 int pathlen = strlen(pathname);
3008 if (pathlen <= prefix_length ||
3009 memcmp(prefix, pathname, prefix_length))
3010 return 0;
3011 }
3012
3013 /* See if it matches any of exclude/include rule */
3014 for (i = 0; i < limit_by_name.nr; i++) {
3015 struct string_list_item *it = &limit_by_name.items[i];
3016 if (!fnmatch(it->string, pathname, 0))
3017 return (it->util != NULL);
3018 }
3019
3020 /*
3021 * If we had any include, a path that does not match any rule is
3022 * not used. Otherwise, we saw bunch of exclude rules (or none)
3023 * and such a path is used.
3024 */
3025 return !has_include;
3026 }
3027
3028
3029 static void prefix_one(char **name)
3030 {
3031 char *old_name = *name;
3032 if (!old_name)
3033 return;
3034 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3035 free(old_name);
3036 }
3037
3038 static void prefix_patches(struct patch *p)
3039 {
3040 if (!prefix || p->is_toplevel_relative)
3041 return;
3042 for ( ; p; p = p->next) {
3043 if (p->new_name == p->old_name) {
3044 char *prefixed = p->new_name;
3045 prefix_one(&prefixed);
3046 p->new_name = p->old_name = prefixed;
3047 }
3048 else {
3049 prefix_one(&p->new_name);
3050 prefix_one(&p->old_name);
3051 }
3052 }
3053 }
3054
3055 #define INACCURATE_EOF (1<<0)
3056 #define RECOUNT (1<<1)
3057
3058 static int apply_patch(int fd, const char *filename, int options)
3059 {
3060 size_t offset;
3061 struct strbuf buf;
3062 struct patch *list = NULL, **listp = &list;
3063 int skipped_patch = 0;
3064
3065 /* FIXME - memory leak when using multiple patch files as inputs */
3066 memset(&fn_table, 0, sizeof(struct string_list));
3067 strbuf_init(&buf, 0);
3068 patch_input_file = filename;
3069 read_patch_file(&buf, fd);
3070 offset = 0;
3071 while (offset < buf.len) {
3072 struct patch *patch;
3073 int nr;
3074
3075 patch = xcalloc(1, sizeof(*patch));
3076 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3077 patch->recount = !!(options & RECOUNT);
3078 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3079 if (nr < 0)
3080 break;
3081 if (apply_in_reverse)
3082 reverse_patches(patch);
3083 if (prefix)
3084 prefix_patches(patch);
3085 if (use_patch(patch)) {
3086 patch_stats(patch);
3087 *listp = patch;
3088 listp = &patch->next;
3089 }
3090 else {
3091 /* perhaps free it a bit better? */
3092 free(patch);
3093 skipped_patch++;
3094 }
3095 offset += nr;
3096 }
3097
3098 if (whitespace_error && (ws_error_action == die_on_ws_error))
3099 apply = 0;
3100
3101 update_index = check_index && apply;
3102 if (update_index && newfd < 0)
3103 newfd = hold_locked_index(&lock_file, 1);
3104
3105 if (check_index) {
3106 if (read_cache() < 0)
3107 die("unable to read index file");
3108 }
3109
3110 if ((check || apply) &&
3111 check_patch_list(list) < 0 &&
3112 !apply_with_reject)
3113 exit(1);
3114
3115 if (apply && write_out_results(list, skipped_patch))
3116 exit(1);
3117
3118 if (fake_ancestor)
3119 build_fake_ancestor(list, fake_ancestor);
3120
3121 if (diffstat)
3122 stat_patch_list(list);
3123
3124 if (numstat)
3125 numstat_patch_list(list);
3126
3127 if (summary)
3128 summary_patch_list(list);
3129
3130 strbuf_release(&buf);
3131 return 0;
3132 }
3133
3134 static int git_apply_config(const char *var, const char *value, void *cb)
3135 {
3136 if (!strcmp(var, "apply.whitespace"))
3137 return git_config_string(&apply_default_whitespace, var, value);
3138 return git_default_config(var, value, cb);
3139 }
3140
3141
3142 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3143 {
3144 int i;
3145 int read_stdin = 1;
3146 int options = 0;
3147 int errs = 0;
3148 int is_not_gitdir;
3149
3150 const char *whitespace_option = NULL;
3151
3152 prefix = setup_git_directory_gently(&is_not_gitdir);
3153 prefix_length = prefix ? strlen(prefix) : 0;
3154 git_config(git_apply_config, NULL);
3155 if (apply_default_whitespace)
3156 parse_whitespace_option(apply_default_whitespace);
3157
3158 for (i = 1; i < argc; i++) {
3159 const char *arg = argv[i];
3160 char *end;
3161 int fd;
3162
3163 if (!strcmp(arg, "-")) {
3164 errs |= apply_patch(0, "<stdin>", options);
3165 read_stdin = 0;
3166 continue;
3167 }
3168 if (!prefixcmp(arg, "--exclude=")) {
3169 add_name_limit(arg + 10, 1);
3170 continue;
3171 }
3172 if (!prefixcmp(arg, "--include=")) {
3173 add_name_limit(arg + 10, 0);
3174 has_include = 1;
3175 continue;
3176 }
3177 if (!prefixcmp(arg, "-p")) {
3178 p_value = atoi(arg + 2);
3179 p_value_known = 1;
3180 continue;
3181 }
3182 if (!strcmp(arg, "--no-add")) {
3183 no_add = 1;
3184 continue;
3185 }
3186 if (!strcmp(arg, "--stat")) {
3187 apply = 0;
3188 diffstat = 1;
3189 continue;
3190 }
3191 if (!strcmp(arg, "--allow-binary-replacement") ||
3192 !strcmp(arg, "--binary")) {
3193 continue; /* now no-op */
3194 }
3195 if (!strcmp(arg, "--numstat")) {
3196 apply = 0;
3197 numstat = 1;
3198 continue;
3199 }
3200 if (!strcmp(arg, "--summary")) {
3201 apply = 0;
3202 summary = 1;
3203 continue;
3204 }
3205 if (!strcmp(arg, "--check")) {
3206 apply = 0;
3207 check = 1;
3208 continue;
3209 }
3210 if (!strcmp(arg, "--index")) {
3211 if (is_not_gitdir)
3212 die("--index outside a repository");
3213 check_index = 1;
3214 continue;
3215 }
3216 if (!strcmp(arg, "--cached")) {
3217 if (is_not_gitdir)
3218 die("--cached outside a repository");
3219 check_index = 1;
3220 cached = 1;
3221 continue;
3222 }
3223 if (!strcmp(arg, "--apply")) {
3224 apply = 1;
3225 continue;
3226 }
3227 if (!strcmp(arg, "--build-fake-ancestor")) {
3228 apply = 0;
3229 if (++i >= argc)
3230 die ("need a filename");
3231 fake_ancestor = argv[i];
3232 continue;
3233 }
3234 if (!strcmp(arg, "-z")) {
3235 line_termination = 0;
3236 continue;
3237 }
3238 if (!prefixcmp(arg, "-C")) {
3239 p_context = strtoul(arg + 2, &end, 0);
3240 if (*end != '\0')
3241 die("unrecognized context count '%s'", arg + 2);
3242 continue;
3243 }
3244 if (!prefixcmp(arg, "--whitespace=")) {
3245 whitespace_option = arg + 13;
3246 parse_whitespace_option(arg + 13);
3247 continue;
3248 }
3249 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3250 apply_in_reverse = 1;
3251 continue;
3252 }
3253 if (!strcmp(arg, "--unidiff-zero")) {
3254 unidiff_zero = 1;
3255 continue;
3256 }
3257 if (!strcmp(arg, "--reject")) {
3258 apply = apply_with_reject = apply_verbosely = 1;
3259 continue;
3260 }
3261 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3262 apply_verbosely = 1;
3263 continue;
3264 }
3265 if (!strcmp(arg, "--inaccurate-eof")) {
3266 options |= INACCURATE_EOF;
3267 continue;
3268 }
3269 if (!strcmp(arg, "--recount")) {
3270 options |= RECOUNT;
3271 continue;
3272 }
3273 if (!prefixcmp(arg, "--directory=")) {
3274 arg += strlen("--directory=");
3275 root_len = strlen(arg);
3276 if (root_len && arg[root_len - 1] != '/') {
3277 char *new_root;
3278 root = new_root = xmalloc(root_len + 2);
3279 strcpy(new_root, arg);
3280 strcpy(new_root + root_len++, "/");
3281 } else
3282 root = arg;
3283 continue;
3284 }
3285 if (0 < prefix_length)
3286 arg = prefix_filename(prefix, prefix_length, arg);
3287
3288 fd = open(arg, O_RDONLY);
3289 if (fd < 0)
3290 die("can't open patch '%s': %s", arg, strerror(errno));
3291 read_stdin = 0;
3292 set_default_whitespace_mode(whitespace_option);
3293 errs |= apply_patch(fd, arg, options);
3294 close(fd);
3295 }
3296 set_default_whitespace_mode(whitespace_option);
3297 if (read_stdin)
3298 errs |= apply_patch(0, "<stdin>", options);
3299 if (whitespace_error) {
3300 if (squelch_whitespace_errors &&
3301 squelch_whitespace_errors < whitespace_error) {
3302 int squelched =
3303 whitespace_error - squelch_whitespace_errors;
3304 fprintf(stderr, "warning: squelched %d "
3305 "whitespace error%s\n",
3306 squelched,
3307 squelched == 1 ? "" : "s");
3308 }
3309 if (ws_error_action == die_on_ws_error)
3310 die("%d line%s add%s whitespace errors.",
3311 whitespace_error,
3312 whitespace_error == 1 ? "" : "s",
3313 whitespace_error == 1 ? "s" : "");
3314 if (applied_after_fixing_ws && apply)
3315 fprintf(stderr, "warning: %d line%s applied after"
3316 " fixing whitespace errors.\n",
3317 applied_after_fixing_ws,
3318 applied_after_fixing_ws == 1 ? "" : "s");
3319 else if (whitespace_error)
3320 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3321 whitespace_error,
3322 whitespace_error == 1 ? "" : "s",
3323 whitespace_error == 1 ? "s" : "");
3324 }
3325
3326 if (update_index) {
3327 if (write_cache(newfd, active_cache, active_nr) ||
3328 commit_locked_index(&lock_file))
3329 die("Unable to write new index file");
3330 }
3331
3332 return !!errs;
3333 }