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