e046e87ad28f47becf5b66abc68476ab2f49e0f2
1/*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9#include "cache.h"
10#include "cache-tree.h"
11#include "quote.h"
12#include "blob.h"
13#include "delta.h"
14#include "builtin.h"
15
16/*
17 * --check turns on checking that the working tree matches the
18 * files that are being modified, but doesn't apply the patch
19 * --stat does just a diffstat, and doesn't actually apply
20 * --numstat does numeric diffstat, and doesn't actually apply
21 * --index-info shows the old and new index info for paths if available.
22 * --index updates the cache as well.
23 * --cached updates only the cache without ever touching the working tree.
24 */
25static const char *prefix;
26static int prefix_length = -1;
27static int newfd = -1;
28
29static int unidiff_zero;
30static int p_value = 1;
31static int p_value_known;
32static int check_index;
33static int update_index;
34static int cached;
35static int diffstat;
36static int numstat;
37static int summary;
38static int check;
39static int apply = 1;
40static int apply_in_reverse;
41static int apply_with_reject;
42static int apply_verbosely;
43static int no_add;
44static const char *fake_ancestor;
45static int line_termination = '\n';
46static unsigned long p_context = ULONG_MAX;
47static const char apply_usage[] =
48"git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|fix|error|error-all>] <patch>...";
49
50static enum ws_error_action {
51 nowarn_ws_error,
52 warn_on_ws_error,
53 die_on_ws_error,
54 correct_ws_error,
55} ws_error_action = warn_on_ws_error;
56static int whitespace_error;
57static int squelch_whitespace_errors = 5;
58static int applied_after_fixing_ws;
59static const char *patch_input_file;
60
61static void parse_whitespace_option(const char *option)
62{
63 if (!option) {
64 ws_error_action = warn_on_ws_error;
65 return;
66 }
67 if (!strcmp(option, "warn")) {
68 ws_error_action = warn_on_ws_error;
69 return;
70 }
71 if (!strcmp(option, "nowarn")) {
72 ws_error_action = nowarn_ws_error;
73 return;
74 }
75 if (!strcmp(option, "error")) {
76 ws_error_action = die_on_ws_error;
77 return;
78 }
79 if (!strcmp(option, "error-all")) {
80 ws_error_action = die_on_ws_error;
81 squelch_whitespace_errors = 0;
82 return;
83 }
84 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
85 ws_error_action = correct_ws_error;
86 return;
87 }
88 die("unrecognized whitespace option '%s'", option);
89}
90
91static void set_default_whitespace_mode(const char *whitespace_option)
92{
93 if (!whitespace_option && !apply_default_whitespace)
94 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
95}
96
97/*
98 * For "diff-stat" like behaviour, we keep track of the biggest change
99 * we've seen, and the longest filename. That allows us to do simple
100 * scaling.
101 */
102static int max_change, max_len;
103
104/*
105 * Various "current state", notably line numbers and what
106 * file (and how) we're patching right now.. The "is_xxxx"
107 * things are flags, where -1 means "don't know yet".
108 */
109static int linenr = 1;
110
111/*
112 * This represents one "hunk" from a patch, starting with
113 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
114 * patch text is pointed at by patch, and its byte length
115 * is stored in size. leading and trailing are the number
116 * of context lines.
117 */
118struct fragment {
119 unsigned long leading, trailing;
120 unsigned long oldpos, oldlines;
121 unsigned long newpos, newlines;
122 const char *patch;
123 int size;
124 int rejected;
125 struct fragment *next;
126};
127
128/*
129 * When dealing with a binary patch, we reuse "leading" field
130 * to store the type of the binary hunk, either deflated "delta"
131 * or deflated "literal".
132 */
133#define binary_patch_method leading
134#define BINARY_DELTA_DEFLATED 1
135#define BINARY_LITERAL_DEFLATED 2
136
137/*
138 * This represents a "patch" to a file, both metainfo changes
139 * such as creation/deletion, filemode and content changes represented
140 * as a series of fragments.
141 */
142struct patch {
143 char *new_name, *old_name, *def_name;
144 unsigned int old_mode, new_mode;
145 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
146 int rejected;
147 unsigned ws_rule;
148 unsigned long deflate_origlen;
149 int lines_added, lines_deleted;
150 int score;
151 unsigned int is_toplevel_relative:1;
152 unsigned int inaccurate_eof:1;
153 unsigned int is_binary:1;
154 unsigned int is_copy:1;
155 unsigned int is_rename:1;
156 struct fragment *fragments;
157 char *result;
158 size_t resultsize;
159 char old_sha1_prefix[41];
160 char new_sha1_prefix[41];
161 struct patch *next;
162};
163
164/*
165 * A line in a file, len-bytes long (includes the terminating LF,
166 * except for an incomplete line at the end if the file ends with
167 * one), and its contents hashes to 'hash'.
168 */
169struct line {
170 size_t len;
171 unsigned hash : 24;
172 unsigned flag : 8;
173};
174
175/*
176 * This represents a "file", which is an array of "lines".
177 */
178struct image {
179 char *buf;
180 size_t len;
181 size_t nr;
182 struct line *line_allocated;
183 struct line *line;
184};
185
186static uint32_t hash_line(const char *cp, size_t len)
187{
188 size_t i;
189 uint32_t h;
190 for (i = 0, h = 0; i < len; i++) {
191 if (!isspace(cp[i])) {
192 h = h * 3 + (cp[i] & 0xff);
193 }
194 }
195 return h;
196}
197
198static void prepare_image(struct image *image, char *buf, size_t len,
199 int prepare_linetable)
200{
201 const char *cp, *ep;
202 int n;
203
204 image->buf = buf;
205 image->len = len;
206
207 if (!prepare_linetable) {
208 image->line = NULL;
209 image->line_allocated = NULL;
210 image->nr = 0;
211 return;
212 }
213
214 ep = image->buf + image->len;
215
216 /* First count lines */
217 cp = image->buf;
218 n = 0;
219 while (cp < ep) {
220 cp = strchrnul(cp, '\n');
221 n++;
222 cp++;
223 }
224
225 image->line_allocated = xcalloc(n, sizeof(struct line));
226 image->line = image->line_allocated;
227 image->nr = n;
228 cp = image->buf;
229 n = 0;
230 while (cp < ep) {
231 const char *next;
232 for (next = cp; next < ep && *next != '\n'; next++)
233 ;
234 if (next < ep)
235 next++;
236 image->line[n].len = next - cp;
237 image->line[n].hash = hash_line(cp, next - cp);
238 cp = next;
239 n++;
240 }
241}
242
243static void clear_image(struct image *image)
244{
245 free(image->buf);
246 image->buf = NULL;
247 image->len = 0;
248}
249
250static void say_patch_name(FILE *output, const char *pre,
251 struct patch *patch, const char *post)
252{
253 fputs(pre, output);
254 if (patch->old_name && patch->new_name &&
255 strcmp(patch->old_name, patch->new_name)) {
256 quote_c_style(patch->old_name, NULL, output, 0);
257 fputs(" => ", output);
258 quote_c_style(patch->new_name, NULL, output, 0);
259 } else {
260 const char *n = patch->new_name;
261 if (!n)
262 n = patch->old_name;
263 quote_c_style(n, NULL, output, 0);
264 }
265 fputs(post, output);
266}
267
268#define CHUNKSIZE (8192)
269#define SLOP (16)
270
271static void read_patch_file(struct strbuf *sb, int fd)
272{
273 if (strbuf_read(sb, fd, 0) < 0)
274 die("git-apply: read returned %s", strerror(errno));
275
276 /*
277 * Make sure that we have some slop in the buffer
278 * so that we can do speculative "memcmp" etc, and
279 * see to it that it is NUL-filled.
280 */
281 strbuf_grow(sb, SLOP);
282 memset(sb->buf + sb->len, 0, SLOP);
283}
284
285static unsigned long linelen(const char *buffer, unsigned long size)
286{
287 unsigned long len = 0;
288 while (size--) {
289 len++;
290 if (*buffer++ == '\n')
291 break;
292 }
293 return len;
294}
295
296static int is_dev_null(const char *str)
297{
298 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
299}
300
301#define TERM_SPACE 1
302#define TERM_TAB 2
303
304static int name_terminate(const char *name, int namelen, int c, int terminate)
305{
306 if (c == ' ' && !(terminate & TERM_SPACE))
307 return 0;
308 if (c == '\t' && !(terminate & TERM_TAB))
309 return 0;
310
311 return 1;
312}
313
314static char *find_name(const char *line, char *def, int p_value, int terminate)
315{
316 int len;
317 const char *start = line;
318
319 if (*line == '"') {
320 struct strbuf name;
321
322 /*
323 * Proposed "new-style" GNU patch/diff format; see
324 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
325 */
326 strbuf_init(&name, 0);
327 if (!unquote_c_style(&name, line, NULL)) {
328 char *cp;
329
330 for (cp = name.buf; p_value; p_value--) {
331 cp = strchr(cp, '/');
332 if (!cp)
333 break;
334 cp++;
335 }
336 if (cp) {
337 /* name can later be freed, so we need
338 * to memmove, not just return cp
339 */
340 strbuf_remove(&name, 0, cp - name.buf);
341 free(def);
342 return strbuf_detach(&name, NULL);
343 }
344 }
345 strbuf_release(&name);
346 }
347
348 for (;;) {
349 char c = *line;
350
351 if (isspace(c)) {
352 if (c == '\n')
353 break;
354 if (name_terminate(start, line-start, c, terminate))
355 break;
356 }
357 line++;
358 if (c == '/' && !--p_value)
359 start = line;
360 }
361 if (!start)
362 return def;
363 len = line - start;
364 if (!len)
365 return def;
366
367 /*
368 * Generally we prefer the shorter name, especially
369 * if the other one is just a variation of that with
370 * something else tacked on to the end (ie "file.orig"
371 * or "file~").
372 */
373 if (def) {
374 int deflen = strlen(def);
375 if (deflen < len && !strncmp(start, def, deflen))
376 return def;
377 free(def);
378 }
379
380 return xmemdupz(start, len);
381}
382
383static int count_slashes(const char *cp)
384{
385 int cnt = 0;
386 char ch;
387
388 while ((ch = *cp++))
389 if (ch == '/')
390 cnt++;
391 return cnt;
392}
393
394/*
395 * Given the string after "--- " or "+++ ", guess the appropriate
396 * p_value for the given patch.
397 */
398static int guess_p_value(const char *nameline)
399{
400 char *name, *cp;
401 int val = -1;
402
403 if (is_dev_null(nameline))
404 return -1;
405 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
406 if (!name)
407 return -1;
408 cp = strchr(name, '/');
409 if (!cp)
410 val = 0;
411 else if (prefix) {
412 /*
413 * Does it begin with "a/$our-prefix" and such? Then this is
414 * very likely to apply to our directory.
415 */
416 if (!strncmp(name, prefix, prefix_length))
417 val = count_slashes(prefix);
418 else {
419 cp++;
420 if (!strncmp(cp, prefix, prefix_length))
421 val = count_slashes(prefix) + 1;
422 }
423 }
424 free(name);
425 return val;
426}
427
428/*
429 * Get the name etc info from the --/+++ lines of a traditional patch header
430 *
431 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
432 * files, we can happily check the index for a match, but for creating a
433 * new file we should try to match whatever "patch" does. I have no idea.
434 */
435static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
436{
437 char *name;
438
439 first += 4; /* skip "--- " */
440 second += 4; /* skip "+++ " */
441 if (!p_value_known) {
442 int p, q;
443 p = guess_p_value(first);
444 q = guess_p_value(second);
445 if (p < 0) p = q;
446 if (0 <= p && p == q) {
447 p_value = p;
448 p_value_known = 1;
449 }
450 }
451 if (is_dev_null(first)) {
452 patch->is_new = 1;
453 patch->is_delete = 0;
454 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
455 patch->new_name = name;
456 } else if (is_dev_null(second)) {
457 patch->is_new = 0;
458 patch->is_delete = 1;
459 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
460 patch->old_name = name;
461 } else {
462 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
463 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
464 patch->old_name = patch->new_name = name;
465 }
466 if (!name)
467 die("unable to find filename in patch at line %d", linenr);
468}
469
470static int gitdiff_hdrend(const char *line, struct patch *patch)
471{
472 return -1;
473}
474
475/*
476 * We're anal about diff header consistency, to make
477 * sure that we don't end up having strange ambiguous
478 * patches floating around.
479 *
480 * As a result, gitdiff_{old|new}name() will check
481 * their names against any previous information, just
482 * to make sure..
483 */
484static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
485{
486 if (!orig_name && !isnull)
487 return find_name(line, NULL, p_value, TERM_TAB);
488
489 if (orig_name) {
490 int len;
491 const char *name;
492 char *another;
493 name = orig_name;
494 len = strlen(name);
495 if (isnull)
496 die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
497 another = find_name(line, NULL, p_value, TERM_TAB);
498 if (!another || memcmp(another, name, len))
499 die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
500 free(another);
501 return orig_name;
502 }
503 else {
504 /* expect "/dev/null" */
505 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
506 die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
507 return NULL;
508 }
509}
510
511static int gitdiff_oldname(const char *line, struct patch *patch)
512{
513 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
514 return 0;
515}
516
517static int gitdiff_newname(const char *line, struct patch *patch)
518{
519 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
520 return 0;
521}
522
523static int gitdiff_oldmode(const char *line, struct patch *patch)
524{
525 patch->old_mode = strtoul(line, NULL, 8);
526 return 0;
527}
528
529static int gitdiff_newmode(const char *line, struct patch *patch)
530{
531 patch->new_mode = strtoul(line, NULL, 8);
532 return 0;
533}
534
535static int gitdiff_delete(const char *line, struct patch *patch)
536{
537 patch->is_delete = 1;
538 patch->old_name = patch->def_name;
539 return gitdiff_oldmode(line, patch);
540}
541
542static int gitdiff_newfile(const char *line, struct patch *patch)
543{
544 patch->is_new = 1;
545 patch->new_name = patch->def_name;
546 return gitdiff_newmode(line, patch);
547}
548
549static int gitdiff_copysrc(const char *line, struct patch *patch)
550{
551 patch->is_copy = 1;
552 patch->old_name = find_name(line, NULL, 0, 0);
553 return 0;
554}
555
556static int gitdiff_copydst(const char *line, struct patch *patch)
557{
558 patch->is_copy = 1;
559 patch->new_name = find_name(line, NULL, 0, 0);
560 return 0;
561}
562
563static int gitdiff_renamesrc(const char *line, struct patch *patch)
564{
565 patch->is_rename = 1;
566 patch->old_name = find_name(line, NULL, 0, 0);
567 return 0;
568}
569
570static int gitdiff_renamedst(const char *line, struct patch *patch)
571{
572 patch->is_rename = 1;
573 patch->new_name = find_name(line, NULL, 0, 0);
574 return 0;
575}
576
577static int gitdiff_similarity(const char *line, struct patch *patch)
578{
579 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
580 patch->score = 0;
581 return 0;
582}
583
584static int gitdiff_dissimilarity(const char *line, struct patch *patch)
585{
586 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
587 patch->score = 0;
588 return 0;
589}
590
591static int gitdiff_index(const char *line, struct patch *patch)
592{
593 /*
594 * index line is N hexadecimal, "..", N hexadecimal,
595 * and optional space with octal mode.
596 */
597 const char *ptr, *eol;
598 int len;
599
600 ptr = strchr(line, '.');
601 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
602 return 0;
603 len = ptr - line;
604 memcpy(patch->old_sha1_prefix, line, len);
605 patch->old_sha1_prefix[len] = 0;
606
607 line = ptr + 2;
608 ptr = strchr(line, ' ');
609 eol = strchr(line, '\n');
610
611 if (!ptr || eol < ptr)
612 ptr = eol;
613 len = ptr - line;
614
615 if (40 < len)
616 return 0;
617 memcpy(patch->new_sha1_prefix, line, len);
618 patch->new_sha1_prefix[len] = 0;
619 if (*ptr == ' ')
620 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
621 return 0;
622}
623
624/*
625 * This is normal for a diff that doesn't change anything: we'll fall through
626 * into the next diff. Tell the parser to break out.
627 */
628static int gitdiff_unrecognized(const char *line, struct patch *patch)
629{
630 return -1;
631}
632
633static const char *stop_at_slash(const char *line, int llen)
634{
635 int i;
636
637 for (i = 0; i < llen; i++) {
638 int ch = line[i];
639 if (ch == '/')
640 return line + i;
641 }
642 return NULL;
643}
644
645/*
646 * This is to extract the same name that appears on "diff --git"
647 * line. We do not find and return anything if it is a rename
648 * patch, and it is OK because we will find the name elsewhere.
649 * We need to reliably find name only when it is mode-change only,
650 * creation or deletion of an empty file. In any of these cases,
651 * both sides are the same name under a/ and b/ respectively.
652 */
653static char *git_header_name(char *line, int llen)
654{
655 const char *name;
656 const char *second = NULL;
657 size_t len;
658
659 line += strlen("diff --git ");
660 llen -= strlen("diff --git ");
661
662 if (*line == '"') {
663 const char *cp;
664 struct strbuf first;
665 struct strbuf sp;
666
667 strbuf_init(&first, 0);
668 strbuf_init(&sp, 0);
669
670 if (unquote_c_style(&first, line, &second))
671 goto free_and_fail1;
672
673 /* advance to the first slash */
674 cp = stop_at_slash(first.buf, first.len);
675 /* we do not accept absolute paths */
676 if (!cp || cp == first.buf)
677 goto free_and_fail1;
678 strbuf_remove(&first, 0, cp + 1 - first.buf);
679
680 /*
681 * second points at one past closing dq of name.
682 * find the second name.
683 */
684 while ((second < line + llen) && isspace(*second))
685 second++;
686
687 if (line + llen <= second)
688 goto free_and_fail1;
689 if (*second == '"') {
690 if (unquote_c_style(&sp, second, NULL))
691 goto free_and_fail1;
692 cp = stop_at_slash(sp.buf, sp.len);
693 if (!cp || cp == sp.buf)
694 goto free_and_fail1;
695 /* They must match, otherwise ignore */
696 if (strcmp(cp + 1, first.buf))
697 goto free_and_fail1;
698 strbuf_release(&sp);
699 return strbuf_detach(&first, NULL);
700 }
701
702 /* unquoted second */
703 cp = stop_at_slash(second, line + llen - second);
704 if (!cp || cp == second)
705 goto free_and_fail1;
706 cp++;
707 if (line + llen - cp != first.len + 1 ||
708 memcmp(first.buf, cp, first.len))
709 goto free_and_fail1;
710 return strbuf_detach(&first, NULL);
711
712 free_and_fail1:
713 strbuf_release(&first);
714 strbuf_release(&sp);
715 return NULL;
716 }
717
718 /* unquoted first name */
719 name = stop_at_slash(line, llen);
720 if (!name || name == line)
721 return NULL;
722 name++;
723
724 /*
725 * since the first name is unquoted, a dq if exists must be
726 * the beginning of the second name.
727 */
728 for (second = name; second < line + llen; second++) {
729 if (*second == '"') {
730 struct strbuf sp;
731 const char *np;
732
733 strbuf_init(&sp, 0);
734 if (unquote_c_style(&sp, second, NULL))
735 goto free_and_fail2;
736
737 np = stop_at_slash(sp.buf, sp.len);
738 if (!np || np == sp.buf)
739 goto free_and_fail2;
740 np++;
741
742 len = sp.buf + sp.len - np;
743 if (len < second - name &&
744 !strncmp(np, name, len) &&
745 isspace(name[len])) {
746 /* Good */
747 strbuf_remove(&sp, 0, np - sp.buf);
748 return strbuf_detach(&sp, NULL);
749 }
750
751 free_and_fail2:
752 strbuf_release(&sp);
753 return NULL;
754 }
755 }
756
757 /*
758 * Accept a name only if it shows up twice, exactly the same
759 * form.
760 */
761 for (len = 0 ; ; len++) {
762 switch (name[len]) {
763 default:
764 continue;
765 case '\n':
766 return NULL;
767 case '\t': case ' ':
768 second = name+len;
769 for (;;) {
770 char c = *second++;
771 if (c == '\n')
772 return NULL;
773 if (c == '/')
774 break;
775 }
776 if (second[len] == '\n' && !memcmp(name, second, len)) {
777 return xmemdupz(name, len);
778 }
779 }
780 }
781}
782
783/* Verify that we recognize the lines following a git header */
784static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
785{
786 unsigned long offset;
787
788 /* A git diff has explicit new/delete information, so we don't guess */
789 patch->is_new = 0;
790 patch->is_delete = 0;
791
792 /*
793 * Some things may not have the old name in the
794 * rest of the headers anywhere (pure mode changes,
795 * or removing or adding empty files), so we get
796 * the default name from the header.
797 */
798 patch->def_name = git_header_name(line, len);
799
800 line += len;
801 size -= len;
802 linenr++;
803 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
804 static const struct opentry {
805 const char *str;
806 int (*fn)(const char *, struct patch *);
807 } optable[] = {
808 { "@@ -", gitdiff_hdrend },
809 { "--- ", gitdiff_oldname },
810 { "+++ ", gitdiff_newname },
811 { "old mode ", gitdiff_oldmode },
812 { "new mode ", gitdiff_newmode },
813 { "deleted file mode ", gitdiff_delete },
814 { "new file mode ", gitdiff_newfile },
815 { "copy from ", gitdiff_copysrc },
816 { "copy to ", gitdiff_copydst },
817 { "rename old ", gitdiff_renamesrc },
818 { "rename new ", gitdiff_renamedst },
819 { "rename from ", gitdiff_renamesrc },
820 { "rename to ", gitdiff_renamedst },
821 { "similarity index ", gitdiff_similarity },
822 { "dissimilarity index ", gitdiff_dissimilarity },
823 { "index ", gitdiff_index },
824 { "", gitdiff_unrecognized },
825 };
826 int i;
827
828 len = linelen(line, size);
829 if (!len || line[len-1] != '\n')
830 break;
831 for (i = 0; i < ARRAY_SIZE(optable); i++) {
832 const struct opentry *p = optable + i;
833 int oplen = strlen(p->str);
834 if (len < oplen || memcmp(p->str, line, oplen))
835 continue;
836 if (p->fn(line + oplen, patch) < 0)
837 return offset;
838 break;
839 }
840 }
841
842 return offset;
843}
844
845static int parse_num(const char *line, unsigned long *p)
846{
847 char *ptr;
848
849 if (!isdigit(*line))
850 return 0;
851 *p = strtoul(line, &ptr, 10);
852 return ptr - line;
853}
854
855static int parse_range(const char *line, int len, int offset, const char *expect,
856 unsigned long *p1, unsigned long *p2)
857{
858 int digits, ex;
859
860 if (offset < 0 || offset >= len)
861 return -1;
862 line += offset;
863 len -= offset;
864
865 digits = parse_num(line, p1);
866 if (!digits)
867 return -1;
868
869 offset += digits;
870 line += digits;
871 len -= digits;
872
873 *p2 = 1;
874 if (*line == ',') {
875 digits = parse_num(line+1, p2);
876 if (!digits)
877 return -1;
878
879 offset += digits+1;
880 line += digits+1;
881 len -= digits+1;
882 }
883
884 ex = strlen(expect);
885 if (ex > len)
886 return -1;
887 if (memcmp(line, expect, ex))
888 return -1;
889
890 return offset + ex;
891}
892
893/*
894 * Parse a unified diff fragment header of the
895 * form "@@ -a,b +c,d @@"
896 */
897static int parse_fragment_header(char *line, int len, struct fragment *fragment)
898{
899 int offset;
900
901 if (!len || line[len-1] != '\n')
902 return -1;
903
904 /* Figure out the number of lines in a fragment */
905 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
906 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
907
908 return offset;
909}
910
911static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
912{
913 unsigned long offset, len;
914
915 patch->is_toplevel_relative = 0;
916 patch->is_rename = patch->is_copy = 0;
917 patch->is_new = patch->is_delete = -1;
918 patch->old_mode = patch->new_mode = 0;
919 patch->old_name = patch->new_name = NULL;
920 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
921 unsigned long nextlen;
922
923 len = linelen(line, size);
924 if (!len)
925 break;
926
927 /* Testing this early allows us to take a few shortcuts.. */
928 if (len < 6)
929 continue;
930
931 /*
932 * Make sure we don't find any unconnected patch fragments.
933 * That's a sign that we didn't find a header, and that a
934 * patch has become corrupted/broken up.
935 */
936 if (!memcmp("@@ -", line, 4)) {
937 struct fragment dummy;
938 if (parse_fragment_header(line, len, &dummy) < 0)
939 continue;
940 die("patch fragment without header at line %d: %.*s",
941 linenr, (int)len-1, line);
942 }
943
944 if (size < len + 6)
945 break;
946
947 /*
948 * Git patch? It might not have a real patch, just a rename
949 * or mode change, so we handle that specially
950 */
951 if (!memcmp("diff --git ", line, 11)) {
952 int git_hdr_len = parse_git_header(line, len, size, patch);
953 if (git_hdr_len <= len)
954 continue;
955 if (!patch->old_name && !patch->new_name) {
956 if (!patch->def_name)
957 die("git diff header lacks filename information (line %d)", linenr);
958 patch->old_name = patch->new_name = patch->def_name;
959 }
960 patch->is_toplevel_relative = 1;
961 *hdrsize = git_hdr_len;
962 return offset;
963 }
964
965 /* --- followed by +++ ? */
966 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
967 continue;
968
969 /*
970 * We only accept unified patches, so we want it to
971 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
972 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
973 */
974 nextlen = linelen(line + len, size - len);
975 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
976 continue;
977
978 /* Ok, we'll consider it a patch */
979 parse_traditional_patch(line, line+len, patch);
980 *hdrsize = len + nextlen;
981 linenr += 2;
982 return offset;
983 }
984 return -1;
985}
986
987static void check_whitespace(const char *line, int len, unsigned ws_rule)
988{
989 char *err;
990 unsigned result = check_and_emit_line(line + 1, len - 1, ws_rule,
991 NULL, NULL, NULL, NULL);
992 if (!result)
993 return;
994
995 whitespace_error++;
996 if (squelch_whitespace_errors &&
997 squelch_whitespace_errors < whitespace_error)
998 ;
999 else {
1000 err = whitespace_error_string(result);
1001 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1002 patch_input_file, linenr, err, len - 2, line + 1);
1003 free(err);
1004 }
1005}
1006
1007/*
1008 * Parse a unified diff. Note that this really needs to parse each
1009 * fragment separately, since the only way to know the difference
1010 * between a "---" that is part of a patch, and a "---" that starts
1011 * the next patch is to look at the line counts..
1012 */
1013static int parse_fragment(char *line, unsigned long size,
1014 struct patch *patch, struct fragment *fragment)
1015{
1016 int added, deleted;
1017 int len = linelen(line, size), offset;
1018 unsigned long oldlines, newlines;
1019 unsigned long leading, trailing;
1020
1021 offset = parse_fragment_header(line, len, fragment);
1022 if (offset < 0)
1023 return -1;
1024 oldlines = fragment->oldlines;
1025 newlines = fragment->newlines;
1026 leading = 0;
1027 trailing = 0;
1028
1029 /* Parse the thing.. */
1030 line += len;
1031 size -= len;
1032 linenr++;
1033 added = deleted = 0;
1034 for (offset = len;
1035 0 < size;
1036 offset += len, size -= len, line += len, linenr++) {
1037 if (!oldlines && !newlines)
1038 break;
1039 len = linelen(line, size);
1040 if (!len || line[len-1] != '\n')
1041 return -1;
1042 switch (*line) {
1043 default:
1044 return -1;
1045 case '\n': /* newer GNU diff, an empty context line */
1046 case ' ':
1047 oldlines--;
1048 newlines--;
1049 if (!deleted && !added)
1050 leading++;
1051 trailing++;
1052 break;
1053 case '-':
1054 if (apply_in_reverse &&
1055 ws_error_action != nowarn_ws_error)
1056 check_whitespace(line, len, patch->ws_rule);
1057 deleted++;
1058 oldlines--;
1059 trailing = 0;
1060 break;
1061 case '+':
1062 if (!apply_in_reverse &&
1063 ws_error_action != nowarn_ws_error)
1064 check_whitespace(line, len, patch->ws_rule);
1065 added++;
1066 newlines--;
1067 trailing = 0;
1068 break;
1069
1070 /*
1071 * We allow "\ No newline at end of file". Depending
1072 * on locale settings when the patch was produced we
1073 * don't know what this line looks like. The only
1074 * thing we do know is that it begins with "\ ".
1075 * Checking for 12 is just for sanity check -- any
1076 * l10n of "\ No newline..." is at least that long.
1077 */
1078 case '\\':
1079 if (len < 12 || memcmp(line, "\\ ", 2))
1080 return -1;
1081 break;
1082 }
1083 }
1084 if (oldlines || newlines)
1085 return -1;
1086 fragment->leading = leading;
1087 fragment->trailing = trailing;
1088
1089 /*
1090 * If a fragment ends with an incomplete line, we failed to include
1091 * it in the above loop because we hit oldlines == newlines == 0
1092 * before seeing it.
1093 */
1094 if (12 < size && !memcmp(line, "\\ ", 2))
1095 offset += linelen(line, size);
1096
1097 patch->lines_added += added;
1098 patch->lines_deleted += deleted;
1099
1100 if (0 < patch->is_new && oldlines)
1101 return error("new file depends on old contents");
1102 if (0 < patch->is_delete && newlines)
1103 return error("deleted file still has contents");
1104 return offset;
1105}
1106
1107static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1108{
1109 unsigned long offset = 0;
1110 unsigned long oldlines = 0, newlines = 0, context = 0;
1111 struct fragment **fragp = &patch->fragments;
1112
1113 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1114 struct fragment *fragment;
1115 int len;
1116
1117 fragment = xcalloc(1, sizeof(*fragment));
1118 len = parse_fragment(line, size, patch, fragment);
1119 if (len <= 0)
1120 die("corrupt patch at line %d", linenr);
1121 fragment->patch = line;
1122 fragment->size = len;
1123 oldlines += fragment->oldlines;
1124 newlines += fragment->newlines;
1125 context += fragment->leading + fragment->trailing;
1126
1127 *fragp = fragment;
1128 fragp = &fragment->next;
1129
1130 offset += len;
1131 line += len;
1132 size -= len;
1133 }
1134
1135 /*
1136 * If something was removed (i.e. we have old-lines) it cannot
1137 * be creation, and if something was added it cannot be
1138 * deletion. However, the reverse is not true; --unified=0
1139 * patches that only add are not necessarily creation even
1140 * though they do not have any old lines, and ones that only
1141 * delete are not necessarily deletion.
1142 *
1143 * Unfortunately, a real creation/deletion patch do _not_ have
1144 * any context line by definition, so we cannot safely tell it
1145 * apart with --unified=0 insanity. At least if the patch has
1146 * more than one hunk it is not creation or deletion.
1147 */
1148 if (patch->is_new < 0 &&
1149 (oldlines || (patch->fragments && patch->fragments->next)))
1150 patch->is_new = 0;
1151 if (patch->is_delete < 0 &&
1152 (newlines || (patch->fragments && patch->fragments->next)))
1153 patch->is_delete = 0;
1154 if (!unidiff_zero || context) {
1155 /* If the user says the patch is not generated with
1156 * --unified=0, or if we have seen context lines,
1157 * then not having oldlines means the patch is creation,
1158 * and not having newlines means the patch is deletion.
1159 */
1160 if (patch->is_new < 0 && !oldlines) {
1161 patch->is_new = 1;
1162 patch->old_name = NULL;
1163 }
1164 if (patch->is_delete < 0 && !newlines) {
1165 patch->is_delete = 1;
1166 patch->new_name = NULL;
1167 }
1168 }
1169
1170 if (0 < patch->is_new && oldlines)
1171 die("new file %s depends on old contents", patch->new_name);
1172 if (0 < patch->is_delete && newlines)
1173 die("deleted file %s still has contents", patch->old_name);
1174 if (!patch->is_delete && !newlines && context)
1175 fprintf(stderr, "** warning: file %s becomes empty but "
1176 "is not deleted\n", patch->new_name);
1177
1178 return offset;
1179}
1180
1181static inline int metadata_changes(struct patch *patch)
1182{
1183 return patch->is_rename > 0 ||
1184 patch->is_copy > 0 ||
1185 patch->is_new > 0 ||
1186 patch->is_delete ||
1187 (patch->old_mode && patch->new_mode &&
1188 patch->old_mode != patch->new_mode);
1189}
1190
1191static char *inflate_it(const void *data, unsigned long size,
1192 unsigned long inflated_size)
1193{
1194 z_stream stream;
1195 void *out;
1196 int st;
1197
1198 memset(&stream, 0, sizeof(stream));
1199
1200 stream.next_in = (unsigned char *)data;
1201 stream.avail_in = size;
1202 stream.next_out = out = xmalloc(inflated_size);
1203 stream.avail_out = inflated_size;
1204 inflateInit(&stream);
1205 st = inflate(&stream, Z_FINISH);
1206 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1207 free(out);
1208 return NULL;
1209 }
1210 return out;
1211}
1212
1213static struct fragment *parse_binary_hunk(char **buf_p,
1214 unsigned long *sz_p,
1215 int *status_p,
1216 int *used_p)
1217{
1218 /*
1219 * Expect a line that begins with binary patch method ("literal"
1220 * or "delta"), followed by the length of data before deflating.
1221 * a sequence of 'length-byte' followed by base-85 encoded data
1222 * should follow, terminated by a newline.
1223 *
1224 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1225 * and we would limit the patch line to 66 characters,
1226 * so one line can fit up to 13 groups that would decode
1227 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1228 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1229 */
1230 int llen, used;
1231 unsigned long size = *sz_p;
1232 char *buffer = *buf_p;
1233 int patch_method;
1234 unsigned long origlen;
1235 char *data = NULL;
1236 int hunk_size = 0;
1237 struct fragment *frag;
1238
1239 llen = linelen(buffer, size);
1240 used = llen;
1241
1242 *status_p = 0;
1243
1244 if (!prefixcmp(buffer, "delta ")) {
1245 patch_method = BINARY_DELTA_DEFLATED;
1246 origlen = strtoul(buffer + 6, NULL, 10);
1247 }
1248 else if (!prefixcmp(buffer, "literal ")) {
1249 patch_method = BINARY_LITERAL_DEFLATED;
1250 origlen = strtoul(buffer + 8, NULL, 10);
1251 }
1252 else
1253 return NULL;
1254
1255 linenr++;
1256 buffer += llen;
1257 while (1) {
1258 int byte_length, max_byte_length, newsize;
1259 llen = linelen(buffer, size);
1260 used += llen;
1261 linenr++;
1262 if (llen == 1) {
1263 /* consume the blank line */
1264 buffer++;
1265 size--;
1266 break;
1267 }
1268 /*
1269 * Minimum line is "A00000\n" which is 7-byte long,
1270 * and the line length must be multiple of 5 plus 2.
1271 */
1272 if ((llen < 7) || (llen-2) % 5)
1273 goto corrupt;
1274 max_byte_length = (llen - 2) / 5 * 4;
1275 byte_length = *buffer;
1276 if ('A' <= byte_length && byte_length <= 'Z')
1277 byte_length = byte_length - 'A' + 1;
1278 else if ('a' <= byte_length && byte_length <= 'z')
1279 byte_length = byte_length - 'a' + 27;
1280 else
1281 goto corrupt;
1282 /* if the input length was not multiple of 4, we would
1283 * have filler at the end but the filler should never
1284 * exceed 3 bytes
1285 */
1286 if (max_byte_length < byte_length ||
1287 byte_length <= max_byte_length - 4)
1288 goto corrupt;
1289 newsize = hunk_size + byte_length;
1290 data = xrealloc(data, newsize);
1291 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1292 goto corrupt;
1293 hunk_size = newsize;
1294 buffer += llen;
1295 size -= llen;
1296 }
1297
1298 frag = xcalloc(1, sizeof(*frag));
1299 frag->patch = inflate_it(data, hunk_size, origlen);
1300 if (!frag->patch)
1301 goto corrupt;
1302 free(data);
1303 frag->size = origlen;
1304 *buf_p = buffer;
1305 *sz_p = size;
1306 *used_p = used;
1307 frag->binary_patch_method = patch_method;
1308 return frag;
1309
1310 corrupt:
1311 free(data);
1312 *status_p = -1;
1313 error("corrupt binary patch at line %d: %.*s",
1314 linenr-1, llen-1, buffer);
1315 return NULL;
1316}
1317
1318static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1319{
1320 /*
1321 * We have read "GIT binary patch\n"; what follows is a line
1322 * that says the patch method (currently, either "literal" or
1323 * "delta") and the length of data before deflating; a
1324 * sequence of 'length-byte' followed by base-85 encoded data
1325 * follows.
1326 *
1327 * When a binary patch is reversible, there is another binary
1328 * hunk in the same format, starting with patch method (either
1329 * "literal" or "delta") with the length of data, and a sequence
1330 * of length-byte + base-85 encoded data, terminated with another
1331 * empty line. This data, when applied to the postimage, produces
1332 * the preimage.
1333 */
1334 struct fragment *forward;
1335 struct fragment *reverse;
1336 int status;
1337 int used, used_1;
1338
1339 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1340 if (!forward && !status)
1341 /* there has to be one hunk (forward hunk) */
1342 return error("unrecognized binary patch at line %d", linenr-1);
1343 if (status)
1344 /* otherwise we already gave an error message */
1345 return status;
1346
1347 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1348 if (reverse)
1349 used += used_1;
1350 else if (status) {
1351 /*
1352 * Not having reverse hunk is not an error, but having
1353 * a corrupt reverse hunk is.
1354 */
1355 free((void*) forward->patch);
1356 free(forward);
1357 return status;
1358 }
1359 forward->next = reverse;
1360 patch->fragments = forward;
1361 patch->is_binary = 1;
1362 return used;
1363}
1364
1365static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1366{
1367 int hdrsize, patchsize;
1368 int offset = find_header(buffer, size, &hdrsize, patch);
1369
1370 if (offset < 0)
1371 return offset;
1372
1373 patch->ws_rule = whitespace_rule(patch->new_name
1374 ? patch->new_name
1375 : patch->old_name);
1376
1377 patchsize = parse_single_patch(buffer + offset + hdrsize,
1378 size - offset - hdrsize, patch);
1379
1380 if (!patchsize) {
1381 static const char *binhdr[] = {
1382 "Binary files ",
1383 "Files ",
1384 NULL,
1385 };
1386 static const char git_binary[] = "GIT binary patch\n";
1387 int i;
1388 int hd = hdrsize + offset;
1389 unsigned long llen = linelen(buffer + hd, size - hd);
1390
1391 if (llen == sizeof(git_binary) - 1 &&
1392 !memcmp(git_binary, buffer + hd, llen)) {
1393 int used;
1394 linenr++;
1395 used = parse_binary(buffer + hd + llen,
1396 size - hd - llen, patch);
1397 if (used)
1398 patchsize = used + llen;
1399 else
1400 patchsize = 0;
1401 }
1402 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1403 for (i = 0; binhdr[i]; i++) {
1404 int len = strlen(binhdr[i]);
1405 if (len < size - hd &&
1406 !memcmp(binhdr[i], buffer + hd, len)) {
1407 linenr++;
1408 patch->is_binary = 1;
1409 patchsize = llen;
1410 break;
1411 }
1412 }
1413 }
1414
1415 /* Empty patch cannot be applied if it is a text patch
1416 * without metadata change. A binary patch appears
1417 * empty to us here.
1418 */
1419 if ((apply || check) &&
1420 (!patch->is_binary && !metadata_changes(patch)))
1421 die("patch with only garbage at line %d", linenr);
1422 }
1423
1424 return offset + hdrsize + patchsize;
1425}
1426
1427#define swap(a,b) myswap((a),(b),sizeof(a))
1428
1429#define myswap(a, b, size) do { \
1430 unsigned char mytmp[size]; \
1431 memcpy(mytmp, &a, size); \
1432 memcpy(&a, &b, size); \
1433 memcpy(&b, mytmp, size); \
1434} while (0)
1435
1436static void reverse_patches(struct patch *p)
1437{
1438 for (; p; p = p->next) {
1439 struct fragment *frag = p->fragments;
1440
1441 swap(p->new_name, p->old_name);
1442 swap(p->new_mode, p->old_mode);
1443 swap(p->is_new, p->is_delete);
1444 swap(p->lines_added, p->lines_deleted);
1445 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1446
1447 for (; frag; frag = frag->next) {
1448 swap(frag->newpos, frag->oldpos);
1449 swap(frag->newlines, frag->oldlines);
1450 }
1451 }
1452}
1453
1454static const char pluses[] =
1455"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1456static const char minuses[]=
1457"----------------------------------------------------------------------";
1458
1459static void show_stats(struct patch *patch)
1460{
1461 struct strbuf qname;
1462 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1463 int max, add, del;
1464
1465 strbuf_init(&qname, 0);
1466 quote_c_style(cp, &qname, NULL, 0);
1467
1468 /*
1469 * "scale" the filename
1470 */
1471 max = max_len;
1472 if (max > 50)
1473 max = 50;
1474
1475 if (qname.len > max) {
1476 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1477 if (!cp)
1478 cp = qname.buf + qname.len + 3 - max;
1479 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1480 }
1481
1482 if (patch->is_binary) {
1483 printf(" %-*s | Bin\n", max, qname.buf);
1484 strbuf_release(&qname);
1485 return;
1486 }
1487
1488 printf(" %-*s |", max, qname.buf);
1489 strbuf_release(&qname);
1490
1491 /*
1492 * scale the add/delete
1493 */
1494 max = max + max_change > 70 ? 70 - max : max_change;
1495 add = patch->lines_added;
1496 del = patch->lines_deleted;
1497
1498 if (max_change > 0) {
1499 int total = ((add + del) * max + max_change / 2) / max_change;
1500 add = (add * max + max_change / 2) / max_change;
1501 del = total - add;
1502 }
1503 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1504 add, pluses, del, minuses);
1505}
1506
1507static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1508{
1509 switch (st->st_mode & S_IFMT) {
1510 case S_IFLNK:
1511 strbuf_grow(buf, st->st_size);
1512 if (readlink(path, buf->buf, st->st_size) != st->st_size)
1513 return -1;
1514 strbuf_setlen(buf, st->st_size);
1515 return 0;
1516 case S_IFREG:
1517 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1518 return error("unable to open or read %s", path);
1519 convert_to_git(path, buf->buf, buf->len, buf);
1520 return 0;
1521 default:
1522 return -1;
1523 }
1524}
1525
1526static int match_fragment(struct image *img,
1527 struct image *preimage,
1528 struct image *postimage,
1529 unsigned long try,
1530 int try_lno,
1531 int match_beginning, int match_end)
1532{
1533 int i;
1534
1535 if (preimage->nr + try_lno > img->nr)
1536 return 0;
1537
1538 if (match_beginning && try_lno)
1539 return 0;
1540
1541 if (match_end && preimage->nr + try_lno != img->nr)
1542 return 0;
1543
1544 /* Quick hash check */
1545 for (i = 0; i < preimage->nr; i++)
1546 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1547 return 0;
1548
1549 /*
1550 * Do we have an exact match? If we were told to match
1551 * at the end, size must be exactly at try+fragsize,
1552 * otherwise try+fragsize must be still within the preimage,
1553 * and either case, the old piece should match the preimage
1554 * exactly.
1555 */
1556 if ((match_end
1557 ? (try + preimage->len == img->len)
1558 : (try + preimage->len <= img->len)) &&
1559 !memcmp(img->buf + try, preimage->buf, preimage->len))
1560 return 1;
1561
1562 /*
1563 * NEEDSWORK: We can optionally match fuzzily here, but
1564 * that is for a later round.
1565 */
1566 return 0;
1567}
1568
1569static int find_pos(struct image *img,
1570 struct image *preimage,
1571 struct image *postimage,
1572 int line,
1573 int match_beginning, int match_end)
1574{
1575 int i;
1576 unsigned long backwards, forwards, try;
1577 int backwards_lno, forwards_lno, try_lno;
1578
1579 if (preimage->nr > img->nr)
1580 return -1;
1581
1582 try = 0;
1583 for (i = 0; i < line; i++)
1584 try += img->line[i].len;
1585
1586 /*
1587 * There's probably some smart way to do this, but I'll leave
1588 * that to the smart and beautiful people. I'm simple and stupid.
1589 */
1590 backwards = try;
1591 backwards_lno = line;
1592 forwards = try;
1593 forwards_lno = line;
1594 try_lno = line;
1595
1596 for (i = 0; ; i++) {
1597 if (match_fragment(img, preimage, postimage,
1598 try, try_lno,
1599 match_beginning, match_end))
1600 return try_lno;
1601
1602 again:
1603 if (backwards_lno == 0 && forwards_lno == img->nr)
1604 break;
1605
1606 if (i & 1) {
1607 if (backwards_lno == 0) {
1608 i++;
1609 goto again;
1610 }
1611 backwards_lno--;
1612 backwards -= img->line[backwards_lno].len;
1613 try = backwards;
1614 try_lno = backwards_lno;
1615 } else {
1616 if (forwards_lno == img->nr) {
1617 i++;
1618 goto again;
1619 }
1620 forwards += img->line[forwards_lno].len;
1621 forwards_lno++;
1622 try = forwards;
1623 try_lno = forwards_lno;
1624 }
1625
1626 }
1627 return -1;
1628}
1629
1630static void remove_first_line(struct image *img)
1631{
1632 img->buf += img->line[0].len;
1633 img->len -= img->line[0].len;
1634 img->line++;
1635 img->nr--;
1636}
1637
1638static void remove_last_line(struct image *img)
1639{
1640 img->len -= img->line[--img->nr].len;
1641}
1642
1643static int apply_line(char *output, const char *patch, int plen,
1644 unsigned ws_rule)
1645{
1646 /*
1647 * plen is number of bytes to be copied from patch,
1648 * starting at patch+1 (patch[0] is '+'). Typically
1649 * patch[plen] is '\n', unless this is the incomplete
1650 * last line.
1651 */
1652 int i;
1653 int add_nl_to_tail = 0;
1654 int fixed = 0;
1655 int last_tab_in_indent = 0;
1656 int last_space_in_indent = 0;
1657 int need_fix_leading_space = 0;
1658 char *buf;
1659
1660 if ((ws_error_action != correct_ws_error) || !whitespace_error ||
1661 *patch != '+') {
1662 memcpy(output, patch + 1, plen);
1663 return plen;
1664 }
1665
1666 /*
1667 * Strip trailing whitespace
1668 */
1669 if ((ws_rule & WS_TRAILING_SPACE) &&
1670 (1 < plen && isspace(patch[plen-1]))) {
1671 if (patch[plen] == '\n')
1672 add_nl_to_tail = 1;
1673 plen--;
1674 while (0 < plen && isspace(patch[plen]))
1675 plen--;
1676 fixed = 1;
1677 }
1678
1679 /*
1680 * Check leading whitespaces (indent)
1681 */
1682 for (i = 1; i < plen; i++) {
1683 char ch = patch[i];
1684 if (ch == '\t') {
1685 last_tab_in_indent = i;
1686 if ((ws_rule & WS_SPACE_BEFORE_TAB) &&
1687 0 < last_space_in_indent)
1688 need_fix_leading_space = 1;
1689 } else if (ch == ' ') {
1690 last_space_in_indent = i;
1691 if ((ws_rule & WS_INDENT_WITH_NON_TAB) &&
1692 8 <= i - last_tab_in_indent)
1693 need_fix_leading_space = 1;
1694 }
1695 else
1696 break;
1697 }
1698
1699 buf = output;
1700 if (need_fix_leading_space) {
1701 int consecutive_spaces = 0;
1702 int last = last_tab_in_indent + 1;
1703
1704 if (ws_rule & WS_INDENT_WITH_NON_TAB) {
1705 /* have "last" point at one past the indent */
1706 if (last_tab_in_indent < last_space_in_indent)
1707 last = last_space_in_indent + 1;
1708 else
1709 last = last_tab_in_indent + 1;
1710 }
1711
1712 /*
1713 * between patch[1..last], strip the funny spaces,
1714 * updating them to tab as needed.
1715 */
1716 for (i = 1; i < last; i++, plen--) {
1717 char ch = patch[i];
1718 if (ch != ' ') {
1719 consecutive_spaces = 0;
1720 *output++ = ch;
1721 } else {
1722 consecutive_spaces++;
1723 if (consecutive_spaces == 8) {
1724 *output++ = '\t';
1725 consecutive_spaces = 0;
1726 }
1727 }
1728 }
1729 while (0 < consecutive_spaces--)
1730 *output++ = ' ';
1731 fixed = 1;
1732 i = last;
1733 }
1734 else
1735 i = 1;
1736
1737 memcpy(output, patch + i, plen);
1738 if (add_nl_to_tail)
1739 output[plen++] = '\n';
1740 if (fixed)
1741 applied_after_fixing_ws++;
1742 return output + plen - buf;
1743}
1744
1745static void update_image(struct image *img,
1746 int applied_pos,
1747 struct image *preimage,
1748 struct image *postimage)
1749{
1750 /*
1751 * remove the copy of preimage at offset in img
1752 * and replace it with postimage
1753 */
1754 int i, nr;
1755 size_t remove_count, insert_count, applied_at = 0;
1756 char *result;
1757
1758 for (i = 0; i < applied_pos; i++)
1759 applied_at += img->line[i].len;
1760
1761 remove_count = 0;
1762 for (i = 0; i < preimage->nr; i++)
1763 remove_count += img->line[applied_pos + i].len;
1764 insert_count = postimage->len;
1765
1766 /* Adjust the contents */
1767 result = xmalloc(img->len + insert_count - remove_count + 1);
1768 memcpy(result, img->buf, applied_at);
1769 memcpy(result + applied_at, postimage->buf, postimage->len);
1770 memcpy(result + applied_at + postimage->len,
1771 img->buf + (applied_at + remove_count),
1772 img->len - (applied_at + remove_count));
1773 free(img->buf);
1774 img->buf = result;
1775 img->len += insert_count - remove_count;
1776 result[img->len] = '\0';
1777
1778 /* Adjust the line table */
1779 nr = img->nr + postimage->nr - preimage->nr;
1780 if (preimage->nr < postimage->nr) {
1781 /*
1782 * NOTE: this knows that we never call remove_first_line()
1783 * on anything other than pre/post image.
1784 */
1785 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1786 img->line_allocated = img->line;
1787 }
1788 if (preimage->nr != postimage->nr)
1789 memmove(img->line + applied_pos + postimage->nr,
1790 img->line + applied_pos + preimage->nr,
1791 (img->nr - (applied_pos + preimage->nr)) *
1792 sizeof(*img->line));
1793 memcpy(img->line + applied_pos,
1794 postimage->line,
1795 postimage->nr * sizeof(*img->line));
1796 img->nr = nr;
1797}
1798
1799static int apply_one_fragment(struct image *img, struct fragment *frag,
1800 int inaccurate_eof, unsigned ws_rule)
1801{
1802 int match_beginning, match_end;
1803 const char *patch = frag->patch;
1804 int size = frag->size;
1805 char *old = xmalloc(size);
1806 char *new = xmalloc(size);
1807 char *oldlines, *newlines;
1808 int oldsize = 0, newsize = 0;
1809 int new_blank_lines_at_end = 0;
1810 unsigned long leading, trailing;
1811 int pos, applied_pos;
1812 struct image preimage;
1813 struct image postimage;
1814
1815 while (size > 0) {
1816 char first;
1817 int len = linelen(patch, size);
1818 int plen;
1819 int added_blank_line = 0;
1820
1821 if (!len)
1822 break;
1823
1824 /*
1825 * "plen" is how much of the line we should use for
1826 * the actual patch data. Normally we just remove the
1827 * first character on the line, but if the line is
1828 * followed by "\ No newline", then we also remove the
1829 * last one (which is the newline, of course).
1830 */
1831 plen = len-1;
1832 if (len < size && patch[len] == '\\')
1833 plen--;
1834 first = *patch;
1835 if (apply_in_reverse) {
1836 if (first == '-')
1837 first = '+';
1838 else if (first == '+')
1839 first = '-';
1840 }
1841
1842 switch (first) {
1843 case '\n':
1844 /* Newer GNU diff, empty context line */
1845 if (plen < 0)
1846 /* ... followed by '\No newline'; nothing */
1847 break;
1848 old[oldsize++] = '\n';
1849 new[newsize++] = '\n';
1850 break;
1851 case ' ':
1852 case '-':
1853 memcpy(old + oldsize, patch + 1, plen);
1854 oldsize += plen;
1855 if (first == '-')
1856 break;
1857 /* Fall-through for ' ' */
1858 case '+':
1859 if (first != '+' || !no_add) {
1860 int added = apply_line(new + newsize, patch,
1861 plen, ws_rule);
1862 newsize += added;
1863 if (first == '+' &&
1864 added == 1 && new[newsize-1] == '\n')
1865 added_blank_line = 1;
1866 }
1867 break;
1868 case '@': case '\\':
1869 /* Ignore it, we already handled it */
1870 break;
1871 default:
1872 if (apply_verbosely)
1873 error("invalid start of line: '%c'", first);
1874 return -1;
1875 }
1876 if (added_blank_line)
1877 new_blank_lines_at_end++;
1878 else
1879 new_blank_lines_at_end = 0;
1880 patch += len;
1881 size -= len;
1882 }
1883
1884 if (inaccurate_eof &&
1885 oldsize > 0 && old[oldsize - 1] == '\n' &&
1886 newsize > 0 && new[newsize - 1] == '\n') {
1887 oldsize--;
1888 newsize--;
1889 }
1890
1891 oldlines = old;
1892 newlines = new;
1893 leading = frag->leading;
1894 trailing = frag->trailing;
1895
1896 /*
1897 * If we don't have any leading/trailing data in the patch,
1898 * we want it to match at the beginning/end of the file.
1899 *
1900 * But that would break if the patch is generated with
1901 * --unified=0; sane people wouldn't do that to cause us
1902 * trouble, but we try to please not so sane ones as well.
1903 */
1904 if (unidiff_zero) {
1905 match_beginning = (!leading && !frag->oldpos);
1906 match_end = 0;
1907 }
1908 else {
1909 match_beginning = !leading && (frag->oldpos == 1);
1910 match_end = !trailing;
1911 }
1912
1913 pos = frag->newpos ? (frag->newpos - 1) : 0;
1914 prepare_image(&preimage, oldlines, oldsize, 1);
1915 prepare_image(&postimage, newlines, newsize, 1);
1916 for (;;) {
1917
1918 applied_pos = find_pos(img, &preimage, &postimage,
1919 pos, match_beginning, match_end);
1920
1921 if (applied_pos >= 0)
1922 break;
1923
1924 /* Am I at my context limits? */
1925 if ((leading <= p_context) && (trailing <= p_context))
1926 break;
1927 if (match_beginning || match_end) {
1928 match_beginning = match_end = 0;
1929 continue;
1930 }
1931
1932 /*
1933 * Reduce the number of context lines; reduce both
1934 * leading and trailing if they are equal otherwise
1935 * just reduce the larger context.
1936 */
1937 if (leading >= trailing) {
1938 remove_first_line(&preimage);
1939 remove_first_line(&postimage);
1940 pos--;
1941 leading--;
1942 }
1943 if (trailing > leading) {
1944 remove_last_line(&preimage);
1945 remove_last_line(&postimage);
1946 trailing--;
1947 }
1948 }
1949
1950 if (applied_pos >= 0) {
1951 if (ws_error_action == correct_ws_error &&
1952 new_blank_lines_at_end &&
1953 postimage.nr + applied_pos == img->nr) {
1954 /*
1955 * If the patch application adds blank lines
1956 * at the end, and if the patch applies at the
1957 * end of the image, remove those added blank
1958 * lines.
1959 */
1960 while (new_blank_lines_at_end--)
1961 remove_last_line(&postimage);
1962 }
1963
1964 /*
1965 * Warn if it was necessary to reduce the number
1966 * of context lines.
1967 */
1968 if ((leading != frag->leading) ||
1969 (trailing != frag->trailing))
1970 fprintf(stderr, "Context reduced to (%ld/%ld)"
1971 " to apply fragment at %d\n",
1972 leading, trailing, applied_pos+1);
1973 update_image(img, applied_pos, &preimage, &postimage);
1974 } else {
1975 if (apply_verbosely)
1976 error("while searching for:\n%.*s", oldsize, oldlines);
1977 }
1978
1979 free(old);
1980 free(new);
1981 free(preimage.line_allocated);
1982 free(postimage.line_allocated);
1983
1984 return (applied_pos < 0);
1985}
1986
1987static int apply_binary_fragment(struct image *img, struct patch *patch)
1988{
1989 struct fragment *fragment = patch->fragments;
1990 unsigned long len;
1991 void *dst;
1992
1993 /* Binary patch is irreversible without the optional second hunk */
1994 if (apply_in_reverse) {
1995 if (!fragment->next)
1996 return error("cannot reverse-apply a binary patch "
1997 "without the reverse hunk to '%s'",
1998 patch->new_name
1999 ? patch->new_name : patch->old_name);
2000 fragment = fragment->next;
2001 }
2002 switch (fragment->binary_patch_method) {
2003 case BINARY_DELTA_DEFLATED:
2004 dst = patch_delta(img->buf, img->len, fragment->patch,
2005 fragment->size, &len);
2006 if (!dst)
2007 return -1;
2008 clear_image(img);
2009 img->buf = dst;
2010 img->len = len;
2011 return 0;
2012 case BINARY_LITERAL_DEFLATED:
2013 clear_image(img);
2014 img->len = fragment->size;
2015 img->buf = xmalloc(img->len+1);
2016 memcpy(img->buf, fragment->patch, img->len);
2017 img->buf[img->len] = '\0';
2018 return 0;
2019 }
2020 return -1;
2021}
2022
2023static int apply_binary(struct image *img, struct patch *patch)
2024{
2025 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2026 unsigned char sha1[20];
2027
2028 /*
2029 * For safety, we require patch index line to contain
2030 * full 40-byte textual SHA1 for old and new, at least for now.
2031 */
2032 if (strlen(patch->old_sha1_prefix) != 40 ||
2033 strlen(patch->new_sha1_prefix) != 40 ||
2034 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2035 get_sha1_hex(patch->new_sha1_prefix, sha1))
2036 return error("cannot apply binary patch to '%s' "
2037 "without full index line", name);
2038
2039 if (patch->old_name) {
2040 /*
2041 * See if the old one matches what the patch
2042 * applies to.
2043 */
2044 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2045 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2046 return error("the patch applies to '%s' (%s), "
2047 "which does not match the "
2048 "current contents.",
2049 name, sha1_to_hex(sha1));
2050 }
2051 else {
2052 /* Otherwise, the old one must be empty. */
2053 if (img->len)
2054 return error("the patch applies to an empty "
2055 "'%s' but it is not empty", name);
2056 }
2057
2058 get_sha1_hex(patch->new_sha1_prefix, sha1);
2059 if (is_null_sha1(sha1)) {
2060 clear_image(img);
2061 return 0; /* deletion patch */
2062 }
2063
2064 if (has_sha1_file(sha1)) {
2065 /* We already have the postimage */
2066 enum object_type type;
2067 unsigned long size;
2068 char *result;
2069
2070 result = read_sha1_file(sha1, &type, &size);
2071 if (!result)
2072 return error("the necessary postimage %s for "
2073 "'%s' cannot be read",
2074 patch->new_sha1_prefix, name);
2075 clear_image(img);
2076 img->buf = result;
2077 img->len = size;
2078 } else {
2079 /*
2080 * We have verified buf matches the preimage;
2081 * apply the patch data to it, which is stored
2082 * in the patch->fragments->{patch,size}.
2083 */
2084 if (apply_binary_fragment(img, patch))
2085 return error("binary patch does not apply to '%s'",
2086 name);
2087
2088 /* verify that the result matches */
2089 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2090 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2091 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2092 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2093 }
2094
2095 return 0;
2096}
2097
2098static int apply_fragments(struct image *img, struct patch *patch)
2099{
2100 struct fragment *frag = patch->fragments;
2101 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2102 unsigned ws_rule = patch->ws_rule;
2103 unsigned inaccurate_eof = patch->inaccurate_eof;
2104
2105 if (patch->is_binary)
2106 return apply_binary(img, patch);
2107
2108 while (frag) {
2109 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2110 error("patch failed: %s:%ld", name, frag->oldpos);
2111 if (!apply_with_reject)
2112 return -1;
2113 frag->rejected = 1;
2114 }
2115 frag = frag->next;
2116 }
2117 return 0;
2118}
2119
2120static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2121{
2122 if (!ce)
2123 return 0;
2124
2125 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2126 strbuf_grow(buf, 100);
2127 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2128 } else {
2129 enum object_type type;
2130 unsigned long sz;
2131 char *result;
2132
2133 result = read_sha1_file(ce->sha1, &type, &sz);
2134 if (!result)
2135 return -1;
2136 /* XXX read_sha1_file NUL-terminates */
2137 strbuf_attach(buf, result, sz, sz + 1);
2138 }
2139 return 0;
2140}
2141
2142static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2143{
2144 struct strbuf buf;
2145 struct image image;
2146 size_t len;
2147 char *img;
2148
2149 strbuf_init(&buf, 0);
2150 if (cached) {
2151 if (read_file_or_gitlink(ce, &buf))
2152 return error("read of %s failed", patch->old_name);
2153 } else if (patch->old_name) {
2154 if (S_ISGITLINK(patch->old_mode)) {
2155 if (ce) {
2156 read_file_or_gitlink(ce, &buf);
2157 } else {
2158 /*
2159 * There is no way to apply subproject
2160 * patch without looking at the index.
2161 */
2162 patch->fragments = NULL;
2163 }
2164 } else {
2165 if (read_old_data(st, patch->old_name, &buf))
2166 return error("read of %s failed", patch->old_name);
2167 }
2168 }
2169
2170 img = strbuf_detach(&buf, &len);
2171 prepare_image(&image, img, len, !patch->is_binary);
2172
2173 if (apply_fragments(&image, patch) < 0)
2174 return -1; /* note with --reject this succeeds. */
2175 patch->result = image.buf;
2176 patch->resultsize = image.len;
2177 free(image.line_allocated);
2178
2179 if (0 < patch->is_delete && patch->resultsize)
2180 return error("removal patch leaves file contents");
2181
2182 return 0;
2183}
2184
2185static int check_to_create_blob(const char *new_name, int ok_if_exists)
2186{
2187 struct stat nst;
2188 if (!lstat(new_name, &nst)) {
2189 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2190 return 0;
2191 /*
2192 * A leading component of new_name might be a symlink
2193 * that is going to be removed with this patch, but
2194 * still pointing at somewhere that has the path.
2195 * In such a case, path "new_name" does not exist as
2196 * far as git is concerned.
2197 */
2198 if (has_symlink_leading_path(new_name, NULL))
2199 return 0;
2200
2201 return error("%s: already exists in working directory", new_name);
2202 }
2203 else if ((errno != ENOENT) && (errno != ENOTDIR))
2204 return error("%s: %s", new_name, strerror(errno));
2205 return 0;
2206}
2207
2208static int verify_index_match(struct cache_entry *ce, struct stat *st)
2209{
2210 if (S_ISGITLINK(ntohl(ce->ce_mode))) {
2211 if (!S_ISDIR(st->st_mode))
2212 return -1;
2213 return 0;
2214 }
2215 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2216}
2217
2218static int check_patch(struct patch *patch, struct patch *prev_patch)
2219{
2220 struct stat st;
2221 const char *old_name = patch->old_name;
2222 const char *new_name = patch->new_name;
2223 const char *name = old_name ? old_name : new_name;
2224 struct cache_entry *ce = NULL;
2225 int ok_if_exists;
2226
2227 patch->rejected = 1; /* we will drop this after we succeed */
2228
2229 /*
2230 * Make sure that we do not have local modifications from the
2231 * index when we are looking at the index. Also make sure
2232 * we have the preimage file to be patched in the work tree,
2233 * unless --cached, which tells git to apply only in the index.
2234 */
2235 if (old_name) {
2236 int stat_ret = 0;
2237 unsigned st_mode = 0;
2238
2239 if (!cached)
2240 stat_ret = lstat(old_name, &st);
2241 if (check_index) {
2242 int pos = cache_name_pos(old_name, strlen(old_name));
2243 if (pos < 0)
2244 return error("%s: does not exist in index",
2245 old_name);
2246 ce = active_cache[pos];
2247 if (stat_ret < 0) {
2248 struct checkout costate;
2249 if (errno != ENOENT)
2250 return error("%s: %s", old_name,
2251 strerror(errno));
2252 /* checkout */
2253 costate.base_dir = "";
2254 costate.base_dir_len = 0;
2255 costate.force = 0;
2256 costate.quiet = 0;
2257 costate.not_new = 0;
2258 costate.refresh_cache = 1;
2259 if (checkout_entry(ce,
2260 &costate,
2261 NULL) ||
2262 lstat(old_name, &st))
2263 return -1;
2264 }
2265 if (!cached && verify_index_match(ce, &st))
2266 return error("%s: does not match index",
2267 old_name);
2268 if (cached)
2269 st_mode = ntohl(ce->ce_mode);
2270 } else if (stat_ret < 0)
2271 return error("%s: %s", old_name, strerror(errno));
2272
2273 if (!cached)
2274 st_mode = ntohl(ce_mode_from_stat(ce, st.st_mode));
2275
2276 if (patch->is_new < 0)
2277 patch->is_new = 0;
2278 if (!patch->old_mode)
2279 patch->old_mode = st_mode;
2280 if ((st_mode ^ patch->old_mode) & S_IFMT)
2281 return error("%s: wrong type", old_name);
2282 if (st_mode != patch->old_mode)
2283 fprintf(stderr, "warning: %s has type %o, expected %o\n",
2284 old_name, st_mode, patch->old_mode);
2285 }
2286
2287 if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2288 !strcmp(prev_patch->old_name, new_name))
2289 /*
2290 * A type-change diff is always split into a patch to
2291 * delete old, immediately followed by a patch to
2292 * create new (see diff.c::run_diff()); in such a case
2293 * it is Ok that the entry to be deleted by the
2294 * previous patch is still in the working tree and in
2295 * the index.
2296 */
2297 ok_if_exists = 1;
2298 else
2299 ok_if_exists = 0;
2300
2301 if (new_name &&
2302 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2303 if (check_index &&
2304 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2305 !ok_if_exists)
2306 return error("%s: already exists in index", new_name);
2307 if (!cached) {
2308 int err = check_to_create_blob(new_name, ok_if_exists);
2309 if (err)
2310 return err;
2311 }
2312 if (!patch->new_mode) {
2313 if (0 < patch->is_new)
2314 patch->new_mode = S_IFREG | 0644;
2315 else
2316 patch->new_mode = patch->old_mode;
2317 }
2318 }
2319
2320 if (new_name && old_name) {
2321 int same = !strcmp(old_name, new_name);
2322 if (!patch->new_mode)
2323 patch->new_mode = patch->old_mode;
2324 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2325 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2326 patch->new_mode, new_name, patch->old_mode,
2327 same ? "" : " of ", same ? "" : old_name);
2328 }
2329
2330 if (apply_data(patch, &st, ce) < 0)
2331 return error("%s: patch does not apply", name);
2332 patch->rejected = 0;
2333 return 0;
2334}
2335
2336static int check_patch_list(struct patch *patch)
2337{
2338 struct patch *prev_patch = NULL;
2339 int err = 0;
2340
2341 for (prev_patch = NULL; patch ; patch = patch->next) {
2342 if (apply_verbosely)
2343 say_patch_name(stderr,
2344 "Checking patch ", patch, "...\n");
2345 err |= check_patch(patch, prev_patch);
2346 prev_patch = patch;
2347 }
2348 return err;
2349}
2350
2351/* This function tries to read the sha1 from the current index */
2352static int get_current_sha1(const char *path, unsigned char *sha1)
2353{
2354 int pos;
2355
2356 if (read_cache() < 0)
2357 return -1;
2358 pos = cache_name_pos(path, strlen(path));
2359 if (pos < 0)
2360 return -1;
2361 hashcpy(sha1, active_cache[pos]->sha1);
2362 return 0;
2363}
2364
2365/* Build an index that contains the just the files needed for a 3way merge */
2366static void build_fake_ancestor(struct patch *list, const char *filename)
2367{
2368 struct patch *patch;
2369 struct index_state result = { 0 };
2370 int fd;
2371
2372 /* Once we start supporting the reverse patch, it may be
2373 * worth showing the new sha1 prefix, but until then...
2374 */
2375 for (patch = list; patch; patch = patch->next) {
2376 const unsigned char *sha1_ptr;
2377 unsigned char sha1[20];
2378 struct cache_entry *ce;
2379 const char *name;
2380
2381 name = patch->old_name ? patch->old_name : patch->new_name;
2382 if (0 < patch->is_new)
2383 continue;
2384 else if (get_sha1(patch->old_sha1_prefix, sha1))
2385 /* git diff has no index line for mode/type changes */
2386 if (!patch->lines_added && !patch->lines_deleted) {
2387 if (get_current_sha1(patch->new_name, sha1) ||
2388 get_current_sha1(patch->old_name, sha1))
2389 die("mode change for %s, which is not "
2390 "in current HEAD", name);
2391 sha1_ptr = sha1;
2392 } else
2393 die("sha1 information is lacking or useless "
2394 "(%s).", name);
2395 else
2396 sha1_ptr = sha1;
2397
2398 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2399 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2400 die ("Could not add %s to temporary index", name);
2401 }
2402
2403 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2404 if (fd < 0 || write_index(&result, fd) || close(fd))
2405 die ("Could not write temporary index to %s", filename);
2406
2407 discard_index(&result);
2408}
2409
2410static void stat_patch_list(struct patch *patch)
2411{
2412 int files, adds, dels;
2413
2414 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2415 files++;
2416 adds += patch->lines_added;
2417 dels += patch->lines_deleted;
2418 show_stats(patch);
2419 }
2420
2421 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2422}
2423
2424static void numstat_patch_list(struct patch *patch)
2425{
2426 for ( ; patch; patch = patch->next) {
2427 const char *name;
2428 name = patch->new_name ? patch->new_name : patch->old_name;
2429 if (patch->is_binary)
2430 printf("-\t-\t");
2431 else
2432 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2433 write_name_quoted(name, stdout, line_termination);
2434 }
2435}
2436
2437static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2438{
2439 if (mode)
2440 printf(" %s mode %06o %s\n", newdelete, mode, name);
2441 else
2442 printf(" %s %s\n", newdelete, name);
2443}
2444
2445static void show_mode_change(struct patch *p, int show_name)
2446{
2447 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2448 if (show_name)
2449 printf(" mode change %06o => %06o %s\n",
2450 p->old_mode, p->new_mode, p->new_name);
2451 else
2452 printf(" mode change %06o => %06o\n",
2453 p->old_mode, p->new_mode);
2454 }
2455}
2456
2457static void show_rename_copy(struct patch *p)
2458{
2459 const char *renamecopy = p->is_rename ? "rename" : "copy";
2460 const char *old, *new;
2461
2462 /* Find common prefix */
2463 old = p->old_name;
2464 new = p->new_name;
2465 while (1) {
2466 const char *slash_old, *slash_new;
2467 slash_old = strchr(old, '/');
2468 slash_new = strchr(new, '/');
2469 if (!slash_old ||
2470 !slash_new ||
2471 slash_old - old != slash_new - new ||
2472 memcmp(old, new, slash_new - new))
2473 break;
2474 old = slash_old + 1;
2475 new = slash_new + 1;
2476 }
2477 /* p->old_name thru old is the common prefix, and old and new
2478 * through the end of names are renames
2479 */
2480 if (old != p->old_name)
2481 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2482 (int)(old - p->old_name), p->old_name,
2483 old, new, p->score);
2484 else
2485 printf(" %s %s => %s (%d%%)\n", renamecopy,
2486 p->old_name, p->new_name, p->score);
2487 show_mode_change(p, 0);
2488}
2489
2490static void summary_patch_list(struct patch *patch)
2491{
2492 struct patch *p;
2493
2494 for (p = patch; p; p = p->next) {
2495 if (p->is_new)
2496 show_file_mode_name("create", p->new_mode, p->new_name);
2497 else if (p->is_delete)
2498 show_file_mode_name("delete", p->old_mode, p->old_name);
2499 else {
2500 if (p->is_rename || p->is_copy)
2501 show_rename_copy(p);
2502 else {
2503 if (p->score) {
2504 printf(" rewrite %s (%d%%)\n",
2505 p->new_name, p->score);
2506 show_mode_change(p, 0);
2507 }
2508 else
2509 show_mode_change(p, 1);
2510 }
2511 }
2512 }
2513}
2514
2515static void patch_stats(struct patch *patch)
2516{
2517 int lines = patch->lines_added + patch->lines_deleted;
2518
2519 if (lines > max_change)
2520 max_change = lines;
2521 if (patch->old_name) {
2522 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2523 if (!len)
2524 len = strlen(patch->old_name);
2525 if (len > max_len)
2526 max_len = len;
2527 }
2528 if (patch->new_name) {
2529 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2530 if (!len)
2531 len = strlen(patch->new_name);
2532 if (len > max_len)
2533 max_len = len;
2534 }
2535}
2536
2537static void remove_file(struct patch *patch, int rmdir_empty)
2538{
2539 if (update_index) {
2540 if (remove_file_from_cache(patch->old_name) < 0)
2541 die("unable to remove %s from index", patch->old_name);
2542 }
2543 if (!cached) {
2544 if (S_ISGITLINK(patch->old_mode)) {
2545 if (rmdir(patch->old_name))
2546 warning("unable to remove submodule %s",
2547 patch->old_name);
2548 } else if (!unlink(patch->old_name) && rmdir_empty) {
2549 char *name = xstrdup(patch->old_name);
2550 char *end = strrchr(name, '/');
2551 while (end) {
2552 *end = 0;
2553 if (rmdir(name))
2554 break;
2555 end = strrchr(name, '/');
2556 }
2557 free(name);
2558 }
2559 }
2560}
2561
2562static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2563{
2564 struct stat st;
2565 struct cache_entry *ce;
2566 int namelen = strlen(path);
2567 unsigned ce_size = cache_entry_size(namelen);
2568
2569 if (!update_index)
2570 return;
2571
2572 ce = xcalloc(1, ce_size);
2573 memcpy(ce->name, path, namelen);
2574 ce->ce_mode = create_ce_mode(mode);
2575 ce->ce_flags = htons(namelen);
2576 if (S_ISGITLINK(mode)) {
2577 const char *s = buf;
2578
2579 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2580 die("corrupt patch for subproject %s", path);
2581 } else {
2582 if (!cached) {
2583 if (lstat(path, &st) < 0)
2584 die("unable to stat newly created file %s",
2585 path);
2586 fill_stat_cache_info(ce, &st);
2587 }
2588 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2589 die("unable to create backing store for newly created file %s", path);
2590 }
2591 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2592 die("unable to add cache entry for %s", path);
2593}
2594
2595static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2596{
2597 int fd;
2598 struct strbuf nbuf;
2599
2600 if (S_ISGITLINK(mode)) {
2601 struct stat st;
2602 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2603 return 0;
2604 return mkdir(path, 0777);
2605 }
2606
2607 if (has_symlinks && S_ISLNK(mode))
2608 /* Although buf:size is counted string, it also is NUL
2609 * terminated.
2610 */
2611 return symlink(buf, path);
2612
2613 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2614 if (fd < 0)
2615 return -1;
2616
2617 strbuf_init(&nbuf, 0);
2618 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2619 size = nbuf.len;
2620 buf = nbuf.buf;
2621 }
2622 write_or_die(fd, buf, size);
2623 strbuf_release(&nbuf);
2624
2625 if (close(fd) < 0)
2626 die("closing file %s: %s", path, strerror(errno));
2627 return 0;
2628}
2629
2630/*
2631 * We optimistically assume that the directories exist,
2632 * which is true 99% of the time anyway. If they don't,
2633 * we create them and try again.
2634 */
2635static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2636{
2637 if (cached)
2638 return;
2639 if (!try_create_file(path, mode, buf, size))
2640 return;
2641
2642 if (errno == ENOENT) {
2643 if (safe_create_leading_directories(path))
2644 return;
2645 if (!try_create_file(path, mode, buf, size))
2646 return;
2647 }
2648
2649 if (errno == EEXIST || errno == EACCES) {
2650 /* We may be trying to create a file where a directory
2651 * used to be.
2652 */
2653 struct stat st;
2654 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2655 errno = EEXIST;
2656 }
2657
2658 if (errno == EEXIST) {
2659 unsigned int nr = getpid();
2660
2661 for (;;) {
2662 const char *newpath;
2663 newpath = mkpath("%s~%u", path, nr);
2664 if (!try_create_file(newpath, mode, buf, size)) {
2665 if (!rename(newpath, path))
2666 return;
2667 unlink(newpath);
2668 break;
2669 }
2670 if (errno != EEXIST)
2671 break;
2672 ++nr;
2673 }
2674 }
2675 die("unable to write file %s mode %o", path, mode);
2676}
2677
2678static void create_file(struct patch *patch)
2679{
2680 char *path = patch->new_name;
2681 unsigned mode = patch->new_mode;
2682 unsigned long size = patch->resultsize;
2683 char *buf = patch->result;
2684
2685 if (!mode)
2686 mode = S_IFREG | 0644;
2687 create_one_file(path, mode, buf, size);
2688 add_index_file(path, mode, buf, size);
2689}
2690
2691/* phase zero is to remove, phase one is to create */
2692static void write_out_one_result(struct patch *patch, int phase)
2693{
2694 if (patch->is_delete > 0) {
2695 if (phase == 0)
2696 remove_file(patch, 1);
2697 return;
2698 }
2699 if (patch->is_new > 0 || patch->is_copy) {
2700 if (phase == 1)
2701 create_file(patch);
2702 return;
2703 }
2704 /*
2705 * Rename or modification boils down to the same
2706 * thing: remove the old, write the new
2707 */
2708 if (phase == 0)
2709 remove_file(patch, patch->is_rename);
2710 if (phase == 1)
2711 create_file(patch);
2712}
2713
2714static int write_out_one_reject(struct patch *patch)
2715{
2716 FILE *rej;
2717 char namebuf[PATH_MAX];
2718 struct fragment *frag;
2719 int cnt = 0;
2720
2721 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2722 if (!frag->rejected)
2723 continue;
2724 cnt++;
2725 }
2726
2727 if (!cnt) {
2728 if (apply_verbosely)
2729 say_patch_name(stderr,
2730 "Applied patch ", patch, " cleanly.\n");
2731 return 0;
2732 }
2733
2734 /* This should not happen, because a removal patch that leaves
2735 * contents are marked "rejected" at the patch level.
2736 */
2737 if (!patch->new_name)
2738 die("internal error");
2739
2740 /* Say this even without --verbose */
2741 say_patch_name(stderr, "Applying patch ", patch, " with");
2742 fprintf(stderr, " %d rejects...\n", cnt);
2743
2744 cnt = strlen(patch->new_name);
2745 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2746 cnt = ARRAY_SIZE(namebuf) - 5;
2747 fprintf(stderr,
2748 "warning: truncating .rej filename to %.*s.rej",
2749 cnt - 1, patch->new_name);
2750 }
2751 memcpy(namebuf, patch->new_name, cnt);
2752 memcpy(namebuf + cnt, ".rej", 5);
2753
2754 rej = fopen(namebuf, "w");
2755 if (!rej)
2756 return error("cannot open %s: %s", namebuf, strerror(errno));
2757
2758 /* Normal git tools never deal with .rej, so do not pretend
2759 * this is a git patch by saying --git nor give extended
2760 * headers. While at it, maybe please "kompare" that wants
2761 * the trailing TAB and some garbage at the end of line ;-).
2762 */
2763 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2764 patch->new_name, patch->new_name);
2765 for (cnt = 1, frag = patch->fragments;
2766 frag;
2767 cnt++, frag = frag->next) {
2768 if (!frag->rejected) {
2769 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2770 continue;
2771 }
2772 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2773 fprintf(rej, "%.*s", frag->size, frag->patch);
2774 if (frag->patch[frag->size-1] != '\n')
2775 fputc('\n', rej);
2776 }
2777 fclose(rej);
2778 return -1;
2779}
2780
2781static int write_out_results(struct patch *list, int skipped_patch)
2782{
2783 int phase;
2784 int errs = 0;
2785 struct patch *l;
2786
2787 if (!list && !skipped_patch)
2788 return error("No changes");
2789
2790 for (phase = 0; phase < 2; phase++) {
2791 l = list;
2792 while (l) {
2793 if (l->rejected)
2794 errs = 1;
2795 else {
2796 write_out_one_result(l, phase);
2797 if (phase == 1 && write_out_one_reject(l))
2798 errs = 1;
2799 }
2800 l = l->next;
2801 }
2802 }
2803 return errs;
2804}
2805
2806static struct lock_file lock_file;
2807
2808static struct excludes {
2809 struct excludes *next;
2810 const char *path;
2811} *excludes;
2812
2813static int use_patch(struct patch *p)
2814{
2815 const char *pathname = p->new_name ? p->new_name : p->old_name;
2816 struct excludes *x = excludes;
2817 while (x) {
2818 if (fnmatch(x->path, pathname, 0) == 0)
2819 return 0;
2820 x = x->next;
2821 }
2822 if (0 < prefix_length) {
2823 int pathlen = strlen(pathname);
2824 if (pathlen <= prefix_length ||
2825 memcmp(prefix, pathname, prefix_length))
2826 return 0;
2827 }
2828 return 1;
2829}
2830
2831static void prefix_one(char **name)
2832{
2833 char *old_name = *name;
2834 if (!old_name)
2835 return;
2836 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
2837 free(old_name);
2838}
2839
2840static void prefix_patches(struct patch *p)
2841{
2842 if (!prefix || p->is_toplevel_relative)
2843 return;
2844 for ( ; p; p = p->next) {
2845 if (p->new_name == p->old_name) {
2846 char *prefixed = p->new_name;
2847 prefix_one(&prefixed);
2848 p->new_name = p->old_name = prefixed;
2849 }
2850 else {
2851 prefix_one(&p->new_name);
2852 prefix_one(&p->old_name);
2853 }
2854 }
2855}
2856
2857static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2858{
2859 size_t offset;
2860 struct strbuf buf;
2861 struct patch *list = NULL, **listp = &list;
2862 int skipped_patch = 0;
2863
2864 strbuf_init(&buf, 0);
2865 patch_input_file = filename;
2866 read_patch_file(&buf, fd);
2867 offset = 0;
2868 while (offset < buf.len) {
2869 struct patch *patch;
2870 int nr;
2871
2872 patch = xcalloc(1, sizeof(*patch));
2873 patch->inaccurate_eof = inaccurate_eof;
2874 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
2875 if (nr < 0)
2876 break;
2877 if (apply_in_reverse)
2878 reverse_patches(patch);
2879 if (prefix)
2880 prefix_patches(patch);
2881 if (use_patch(patch)) {
2882 patch_stats(patch);
2883 *listp = patch;
2884 listp = &patch->next;
2885 }
2886 else {
2887 /* perhaps free it a bit better? */
2888 free(patch);
2889 skipped_patch++;
2890 }
2891 offset += nr;
2892 }
2893
2894 if (whitespace_error && (ws_error_action == die_on_ws_error))
2895 apply = 0;
2896
2897 update_index = check_index && apply;
2898 if (update_index && newfd < 0)
2899 newfd = hold_locked_index(&lock_file, 1);
2900
2901 if (check_index) {
2902 if (read_cache() < 0)
2903 die("unable to read index file");
2904 }
2905
2906 if ((check || apply) &&
2907 check_patch_list(list) < 0 &&
2908 !apply_with_reject)
2909 exit(1);
2910
2911 if (apply && write_out_results(list, skipped_patch))
2912 exit(1);
2913
2914 if (fake_ancestor)
2915 build_fake_ancestor(list, fake_ancestor);
2916
2917 if (diffstat)
2918 stat_patch_list(list);
2919
2920 if (numstat)
2921 numstat_patch_list(list);
2922
2923 if (summary)
2924 summary_patch_list(list);
2925
2926 strbuf_release(&buf);
2927 return 0;
2928}
2929
2930static int git_apply_config(const char *var, const char *value)
2931{
2932 if (!strcmp(var, "apply.whitespace")) {
2933 apply_default_whitespace = xstrdup(value);
2934 return 0;
2935 }
2936 return git_default_config(var, value);
2937}
2938
2939
2940int cmd_apply(int argc, const char **argv, const char *unused_prefix)
2941{
2942 int i;
2943 int read_stdin = 1;
2944 int inaccurate_eof = 0;
2945 int errs = 0;
2946 int is_not_gitdir = 0;
2947
2948 const char *whitespace_option = NULL;
2949
2950 prefix = setup_git_directory_gently(&is_not_gitdir);
2951 prefix_length = prefix ? strlen(prefix) : 0;
2952 git_config(git_apply_config);
2953 if (apply_default_whitespace)
2954 parse_whitespace_option(apply_default_whitespace);
2955
2956 for (i = 1; i < argc; i++) {
2957 const char *arg = argv[i];
2958 char *end;
2959 int fd;
2960
2961 if (!strcmp(arg, "-")) {
2962 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2963 read_stdin = 0;
2964 continue;
2965 }
2966 if (!prefixcmp(arg, "--exclude=")) {
2967 struct excludes *x = xmalloc(sizeof(*x));
2968 x->path = arg + 10;
2969 x->next = excludes;
2970 excludes = x;
2971 continue;
2972 }
2973 if (!prefixcmp(arg, "-p")) {
2974 p_value = atoi(arg + 2);
2975 p_value_known = 1;
2976 continue;
2977 }
2978 if (!strcmp(arg, "--no-add")) {
2979 no_add = 1;
2980 continue;
2981 }
2982 if (!strcmp(arg, "--stat")) {
2983 apply = 0;
2984 diffstat = 1;
2985 continue;
2986 }
2987 if (!strcmp(arg, "--allow-binary-replacement") ||
2988 !strcmp(arg, "--binary")) {
2989 continue; /* now no-op */
2990 }
2991 if (!strcmp(arg, "--numstat")) {
2992 apply = 0;
2993 numstat = 1;
2994 continue;
2995 }
2996 if (!strcmp(arg, "--summary")) {
2997 apply = 0;
2998 summary = 1;
2999 continue;
3000 }
3001 if (!strcmp(arg, "--check")) {
3002 apply = 0;
3003 check = 1;
3004 continue;
3005 }
3006 if (!strcmp(arg, "--index")) {
3007 if (is_not_gitdir)
3008 die("--index outside a repository");
3009 check_index = 1;
3010 continue;
3011 }
3012 if (!strcmp(arg, "--cached")) {
3013 if (is_not_gitdir)
3014 die("--cached outside a repository");
3015 check_index = 1;
3016 cached = 1;
3017 continue;
3018 }
3019 if (!strcmp(arg, "--apply")) {
3020 apply = 1;
3021 continue;
3022 }
3023 if (!strcmp(arg, "--build-fake-ancestor")) {
3024 apply = 0;
3025 if (++i >= argc)
3026 die ("need a filename");
3027 fake_ancestor = argv[i];
3028 continue;
3029 }
3030 if (!strcmp(arg, "-z")) {
3031 line_termination = 0;
3032 continue;
3033 }
3034 if (!prefixcmp(arg, "-C")) {
3035 p_context = strtoul(arg + 2, &end, 0);
3036 if (*end != '\0')
3037 die("unrecognized context count '%s'", arg + 2);
3038 continue;
3039 }
3040 if (!prefixcmp(arg, "--whitespace=")) {
3041 whitespace_option = arg + 13;
3042 parse_whitespace_option(arg + 13);
3043 continue;
3044 }
3045 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
3046 apply_in_reverse = 1;
3047 continue;
3048 }
3049 if (!strcmp(arg, "--unidiff-zero")) {
3050 unidiff_zero = 1;
3051 continue;
3052 }
3053 if (!strcmp(arg, "--reject")) {
3054 apply = apply_with_reject = apply_verbosely = 1;
3055 continue;
3056 }
3057 if (!strcmp(arg, "-v") || !strcmp(arg, "--verbose")) {
3058 apply_verbosely = 1;
3059 continue;
3060 }
3061 if (!strcmp(arg, "--inaccurate-eof")) {
3062 inaccurate_eof = 1;
3063 continue;
3064 }
3065 if (0 < prefix_length)
3066 arg = prefix_filename(prefix, prefix_length, arg);
3067
3068 fd = open(arg, O_RDONLY);
3069 if (fd < 0)
3070 usage(apply_usage);
3071 read_stdin = 0;
3072 set_default_whitespace_mode(whitespace_option);
3073 errs |= apply_patch(fd, arg, inaccurate_eof);
3074 close(fd);
3075 }
3076 set_default_whitespace_mode(whitespace_option);
3077 if (read_stdin)
3078 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
3079 if (whitespace_error) {
3080 if (squelch_whitespace_errors &&
3081 squelch_whitespace_errors < whitespace_error) {
3082 int squelched =
3083 whitespace_error - squelch_whitespace_errors;
3084 fprintf(stderr, "warning: squelched %d "
3085 "whitespace error%s\n",
3086 squelched,
3087 squelched == 1 ? "" : "s");
3088 }
3089 if (ws_error_action == die_on_ws_error)
3090 die("%d line%s add%s whitespace errors.",
3091 whitespace_error,
3092 whitespace_error == 1 ? "" : "s",
3093 whitespace_error == 1 ? "s" : "");
3094 if (applied_after_fixing_ws && apply)
3095 fprintf(stderr, "warning: %d line%s applied after"
3096 " fixing whitespace errors.\n",
3097 applied_after_fixing_ws,
3098 applied_after_fixing_ws == 1 ? "" : "s");
3099 else if (whitespace_error)
3100 fprintf(stderr, "warning: %d line%s add%s whitespace errors.\n",
3101 whitespace_error,
3102 whitespace_error == 1 ? "" : "s",
3103 whitespace_error == 1 ? "s" : "");
3104 }
3105
3106 if (update_index) {
3107 if (write_cache(newfd, active_cache, active_nr) ||
3108 commit_locked_index(&lock_file))
3109 die("Unable to write new index file");
3110 }
3111
3112 return !!errs;
3113}