![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9 #include "cache.h"
10 #include "cache-tree.h"
11 #include "quote.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "builtin.h"
15 #include "string-list.h"
16 #include "dir.h"
17 #include "parse-options.h"
19 /*
20 * --check turns on checking that the working tree matches the
21 * files that are being modified, but doesn't apply the patch
22 * --stat does just a diffstat, and doesn't actually apply
23 * --numstat does numeric diffstat, and doesn't actually apply
24 * --index-info shows the old and new index info for paths if available.
25 * --index updates the cache as well.
26 * --cached updates only the cache without ever touching the working tree.
27 */
28 static const char *prefix;
29 static int prefix_length = -1;
30 static int newfd = -1;
32 static int unidiff_zero;
33 static int p_value = 1;
34 static int p_value_known;
35 static int check_index;
36 static int update_index;
37 static int cached;
38 static int diffstat;
39 static int numstat;
40 static int summary;
41 static int check;
42 static int apply = 1;
43 static int apply_in_reverse;
44 static int apply_with_reject;
45 static int apply_verbosely;
46 static int no_add;
47 static const char *fake_ancestor;
48 static int line_termination = '\n';
49 static unsigned int p_context = UINT_MAX;
50 static const char * const apply_usage[] = {
51 "git apply [options] [<patch>...]",
52 NULL
53 };
55 static enum ws_error_action {
56 nowarn_ws_error,
57 warn_on_ws_error,
58 die_on_ws_error,
59 correct_ws_error,
60 } ws_error_action = warn_on_ws_error;
61 static int whitespace_error;
62 static int squelch_whitespace_errors = 5;
63 static int applied_after_fixing_ws;
65 static enum ws_ignore {
66 ignore_ws_none,
67 ignore_ws_change,
68 } ws_ignore_action = ignore_ws_none;
71 static const char *patch_input_file;
72 static const char *root;
73 static int root_len;
74 static int read_stdin = 1;
75 static int options;
77 static void parse_whitespace_option(const char *option)
78 {
79 if (!option) {
80 ws_error_action = warn_on_ws_error;
81 return;
82 }
83 if (!strcmp(option, "warn")) {
84 ws_error_action = warn_on_ws_error;
85 return;
86 }
87 if (!strcmp(option, "nowarn")) {
88 ws_error_action = nowarn_ws_error;
89 return;
90 }
91 if (!strcmp(option, "error")) {
92 ws_error_action = die_on_ws_error;
93 return;
94 }
95 if (!strcmp(option, "error-all")) {
96 ws_error_action = die_on_ws_error;
97 squelch_whitespace_errors = 0;
98 return;
99 }
100 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
101 ws_error_action = correct_ws_error;
102 return;
103 }
104 die("unrecognized whitespace option '%s'", option);
105 }
107 static void parse_ignorewhitespace_option(const char *option)
108 {
109 if (!option || !strcmp(option, "no") ||
110 !strcmp(option, "false") || !strcmp(option, "never") ||
111 !strcmp(option, "none")) {
112 ws_ignore_action = ignore_ws_none;
113 return;
114 }
115 if (!strcmp(option, "change")) {
116 ws_ignore_action = ignore_ws_change;
117 return;
118 }
119 die("unrecognized whitespace ignore option '%s'", option);
120 }
122 static void set_default_whitespace_mode(const char *whitespace_option)
123 {
124 if (!whitespace_option && !apply_default_whitespace)
125 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
126 }
128 /*
129 * For "diff-stat" like behaviour, we keep track of the biggest change
130 * we've seen, and the longest filename. That allows us to do simple
131 * scaling.
132 */
133 static int max_change, max_len;
135 /*
136 * Various "current state", notably line numbers and what
137 * file (and how) we're patching right now.. The "is_xxxx"
138 * things are flags, where -1 means "don't know yet".
139 */
140 static int linenr = 1;
142 /*
143 * This represents one "hunk" from a patch, starting with
144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
145 * patch text is pointed at by patch, and its byte length
146 * is stored in size. leading and trailing are the number
147 * of context lines.
148 */
149 struct fragment {
150 unsigned long leading, trailing;
151 unsigned long oldpos, oldlines;
152 unsigned long newpos, newlines;
153 const char *patch;
154 int size;
155 int rejected;
156 int linenr;
157 struct fragment *next;
158 };
160 /*
161 * When dealing with a binary patch, we reuse "leading" field
162 * to store the type of the binary hunk, either deflated "delta"
163 * or deflated "literal".
164 */
165 #define binary_patch_method leading
166 #define BINARY_DELTA_DEFLATED 1
167 #define BINARY_LITERAL_DEFLATED 2
169 /*
170 * This represents a "patch" to a file, both metainfo changes
171 * such as creation/deletion, filemode and content changes represented
172 * as a series of fragments.
173 */
174 struct patch {
175 char *new_name, *old_name, *def_name;
176 unsigned int old_mode, new_mode;
177 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
178 int rejected;
179 unsigned ws_rule;
180 unsigned long deflate_origlen;
181 int lines_added, lines_deleted;
182 int score;
183 unsigned int is_toplevel_relative:1;
184 unsigned int inaccurate_eof:1;
185 unsigned int is_binary:1;
186 unsigned int is_copy:1;
187 unsigned int is_rename:1;
188 unsigned int recount:1;
189 struct fragment *fragments;
190 char *result;
191 size_t resultsize;
192 char old_sha1_prefix[41];
193 char new_sha1_prefix[41];
194 struct patch *next;
195 };
197 /*
198 * A line in a file, len-bytes long (includes the terminating LF,
199 * except for an incomplete line at the end if the file ends with
200 * one), and its contents hashes to 'hash'.
201 */
202 struct line {
203 size_t len;
204 unsigned hash : 24;
205 unsigned flag : 8;
206 #define LINE_COMMON 1
207 };
209 /*
210 * This represents a "file", which is an array of "lines".
211 */
212 struct image {
213 char *buf;
214 size_t len;
215 size_t nr;
216 size_t alloc;
217 struct line *line_allocated;
218 struct line *line;
219 };
221 /*
222 * Records filenames that have been touched, in order to handle
223 * the case where more than one patches touch the same file.
224 */
226 static struct string_list fn_table;
228 static uint32_t hash_line(const char *cp, size_t len)
229 {
230 size_t i;
231 uint32_t h;
232 for (i = 0, h = 0; i < len; i++) {
233 if (!isspace(cp[i])) {
234 h = h * 3 + (cp[i] & 0xff);
235 }
236 }
237 return h;
238 }
240 /*
241 * Compare lines s1 of length n1 and s2 of length n2, ignoring
242 * whitespace difference. Returns 1 if they match, 0 otherwise
243 */
244 static int fuzzy_matchlines(const char *s1, size_t n1,
245 const char *s2, size_t n2)
246 {
247 const char *last1 = s1 + n1 - 1;
248 const char *last2 = s2 + n2 - 1;
249 int result = 0;
251 if (n1 < 0 || n2 < 0)
252 return 0;
254 /* ignore line endings */
255 while ((*last1 == '\r') || (*last1 == '\n'))
256 last1--;
257 while ((*last2 == '\r') || (*last2 == '\n'))
258 last2--;
260 /* skip leading whitespace */
261 while (isspace(*s1) && (s1 <= last1))
262 s1++;
263 while (isspace(*s2) && (s2 <= last2))
264 s2++;
265 /* early return if both lines are empty */
266 if ((s1 > last1) && (s2 > last2))
267 return 1;
268 while (!result) {
269 result = *s1++ - *s2++;
270 /*
271 * Skip whitespace inside. We check for whitespace on
272 * both buffers because we don't want "a b" to match
273 * "ab"
274 */
275 if (isspace(*s1) && isspace(*s2)) {
276 while (isspace(*s1) && s1 <= last1)
277 s1++;
278 while (isspace(*s2) && s2 <= last2)
279 s2++;
280 }
281 /*
282 * If we reached the end on one side only,
283 * lines don't match
284 */
285 if (
286 ((s2 > last2) && (s1 <= last1)) ||
287 ((s1 > last1) && (s2 <= last2)))
288 return 0;
289 if ((s1 > last1) && (s2 > last2))
290 break;
291 }
293 return !result;
294 }
296 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
297 {
298 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
299 img->line_allocated[img->nr].len = len;
300 img->line_allocated[img->nr].hash = hash_line(bol, len);
301 img->line_allocated[img->nr].flag = flag;
302 img->nr++;
303 }
305 static void prepare_image(struct image *image, char *buf, size_t len,
306 int prepare_linetable)
307 {
308 const char *cp, *ep;
310 memset(image, 0, sizeof(*image));
311 image->buf = buf;
312 image->len = len;
314 if (!prepare_linetable)
315 return;
317 ep = image->buf + image->len;
318 cp = image->buf;
319 while (cp < ep) {
320 const char *next;
321 for (next = cp; next < ep && *next != '\n'; next++)
322 ;
323 if (next < ep)
324 next++;
325 add_line_info(image, cp, next - cp, 0);
326 cp = next;
327 }
328 image->line = image->line_allocated;
329 }
331 static void clear_image(struct image *image)
332 {
333 free(image->buf);
334 image->buf = NULL;
335 image->len = 0;
336 }
338 static void say_patch_name(FILE *output, const char *pre,
339 struct patch *patch, const char *post)
340 {
341 fputs(pre, output);
342 if (patch->old_name && patch->new_name &&
343 strcmp(patch->old_name, patch->new_name)) {
344 quote_c_style(patch->old_name, NULL, output, 0);
345 fputs(" => ", output);
346 quote_c_style(patch->new_name, NULL, output, 0);
347 } else {
348 const char *n = patch->new_name;
349 if (!n)
350 n = patch->old_name;
351 quote_c_style(n, NULL, output, 0);
352 }
353 fputs(post, output);
354 }
356 #define CHUNKSIZE (8192)
357 #define SLOP (16)
359 static void read_patch_file(struct strbuf *sb, int fd)
360 {
361 if (strbuf_read(sb, fd, 0) < 0)
362 die_errno("git apply: failed to read");
364 /*
365 * Make sure that we have some slop in the buffer
366 * so that we can do speculative "memcmp" etc, and
367 * see to it that it is NUL-filled.
368 */
369 strbuf_grow(sb, SLOP);
370 memset(sb->buf + sb->len, 0, SLOP);
371 }
373 static unsigned long linelen(const char *buffer, unsigned long size)
374 {
375 unsigned long len = 0;
376 while (size--) {
377 len++;
378 if (*buffer++ == '\n')
379 break;
380 }
381 return len;
382 }
384 static int is_dev_null(const char *str)
385 {
386 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
387 }
389 #define TERM_SPACE 1
390 #define TERM_TAB 2
392 static int name_terminate(const char *name, int namelen, int c, int terminate)
393 {
394 if (c == ' ' && !(terminate & TERM_SPACE))
395 return 0;
396 if (c == '\t' && !(terminate & TERM_TAB))
397 return 0;
399 return 1;
400 }
402 /* remove double slashes to make --index work with such filenames */
403 static char *squash_slash(char *name)
404 {
405 int i = 0, j = 0;
407 while (name[i]) {
408 if ((name[j++] = name[i++]) == '/')
409 while (name[i] == '/')
410 i++;
411 }
412 name[j] = '\0';
413 return name;
414 }
416 static char *find_name(const char *line, char *def, int p_value, int terminate)
417 {
418 int len;
419 const char *start = line;
421 if (*line == '"') {
422 struct strbuf name = STRBUF_INIT;
424 /*
425 * Proposed "new-style" GNU patch/diff format; see
426 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
427 */
428 if (!unquote_c_style(&name, line, NULL)) {
429 char *cp;
431 for (cp = name.buf; p_value; p_value--) {
432 cp = strchr(cp, '/');
433 if (!cp)
434 break;
435 cp++;
436 }
437 if (cp) {
438 /* name can later be freed, so we need
439 * to memmove, not just return cp
440 */
441 strbuf_remove(&name, 0, cp - name.buf);
442 free(def);
443 if (root)
444 strbuf_insert(&name, 0, root, root_len);
445 return squash_slash(strbuf_detach(&name, NULL));
446 }
447 }
448 strbuf_release(&name);
449 }
451 for (;;) {
452 char c = *line;
454 if (isspace(c)) {
455 if (c == '\n')
456 break;
457 if (name_terminate(start, line-start, c, terminate))
458 break;
459 }
460 line++;
461 if (c == '/' && !--p_value)
462 start = line;
463 }
464 if (!start)
465 return squash_slash(def);
466 len = line - start;
467 if (!len)
468 return squash_slash(def);
470 /*
471 * Generally we prefer the shorter name, especially
472 * if the other one is just a variation of that with
473 * something else tacked on to the end (ie "file.orig"
474 * or "file~").
475 */
476 if (def) {
477 int deflen = strlen(def);
478 if (deflen < len && !strncmp(start, def, deflen))
479 return squash_slash(def);
480 free(def);
481 }
483 if (root) {
484 char *ret = xmalloc(root_len + len + 1);
485 strcpy(ret, root);
486 memcpy(ret + root_len, start, len);
487 ret[root_len + len] = '\0';
488 return squash_slash(ret);
489 }
491 return squash_slash(xmemdupz(start, len));
492 }
494 static int count_slashes(const char *cp)
495 {
496 int cnt = 0;
497 char ch;
499 while ((ch = *cp++))
500 if (ch == '/')
501 cnt++;
502 return cnt;
503 }
505 /*
506 * Given the string after "--- " or "+++ ", guess the appropriate
507 * p_value for the given patch.
508 */
509 static int guess_p_value(const char *nameline)
510 {
511 char *name, *cp;
512 int val = -1;
514 if (is_dev_null(nameline))
515 return -1;
516 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
517 if (!name)
518 return -1;
519 cp = strchr(name, '/');
520 if (!cp)
521 val = 0;
522 else if (prefix) {
523 /*
524 * Does it begin with "a/$our-prefix" and such? Then this is
525 * very likely to apply to our directory.
526 */
527 if (!strncmp(name, prefix, prefix_length))
528 val = count_slashes(prefix);
529 else {
530 cp++;
531 if (!strncmp(cp, prefix, prefix_length))
532 val = count_slashes(prefix) + 1;
533 }
534 }
535 free(name);
536 return val;
537 }
539 /*
540 * Does the ---/+++ line has the POSIX timestamp after the last HT?
541 * GNU diff puts epoch there to signal a creation/deletion event. Is
542 * this such a timestamp?
543 */
544 static int has_epoch_timestamp(const char *nameline)
545 {
546 /*
547 * We are only interested in epoch timestamp; any non-zero
548 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
549 * For the same reason, the date must be either 1969-12-31 or
550 * 1970-01-01, and the seconds part must be "00".
551 */
552 const char stamp_regexp[] =
553 "^(1969-12-31|1970-01-01)"
554 " "
555 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
556 " "
557 "([-+][0-2][0-9][0-5][0-9])\n";
558 const char *timestamp = NULL, *cp;
559 static regex_t *stamp;
560 regmatch_t m[10];
561 int zoneoffset;
562 int hourminute;
563 int status;
565 for (cp = nameline; *cp != '\n'; cp++) {
566 if (*cp == '\t')
567 timestamp = cp + 1;
568 }
569 if (!timestamp)
570 return 0;
571 if (!stamp) {
572 stamp = xmalloc(sizeof(*stamp));
573 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
574 warning("Cannot prepare timestamp regexp %s",
575 stamp_regexp);
576 return 0;
577 }
578 }
580 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
581 if (status) {
582 if (status != REG_NOMATCH)
583 warning("regexec returned %d for input: %s",
584 status, timestamp);
585 return 0;
586 }
588 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
589 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
590 if (timestamp[m[3].rm_so] == '-')
591 zoneoffset = -zoneoffset;
593 /*
594 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
595 * (west of GMT) or 1970-01-01 (east of GMT)
596 */
597 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
598 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
599 return 0;
601 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
602 strtol(timestamp + 14, NULL, 10) -
603 zoneoffset);
605 return ((zoneoffset < 0 && hourminute == 1440) ||
606 (0 <= zoneoffset && !hourminute));
607 }
609 /*
610 * Get the name etc info from the ---/+++ lines of a traditional patch header
611 *
612 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
613 * files, we can happily check the index for a match, but for creating a
614 * new file we should try to match whatever "patch" does. I have no idea.
615 */
616 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
617 {
618 char *name;
620 first += 4; /* skip "--- " */
621 second += 4; /* skip "+++ " */
622 if (!p_value_known) {
623 int p, q;
624 p = guess_p_value(first);
625 q = guess_p_value(second);
626 if (p < 0) p = q;
627 if (0 <= p && p == q) {
628 p_value = p;
629 p_value_known = 1;
630 }
631 }
632 if (is_dev_null(first)) {
633 patch->is_new = 1;
634 patch->is_delete = 0;
635 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
636 patch->new_name = name;
637 } else if (is_dev_null(second)) {
638 patch->is_new = 0;
639 patch->is_delete = 1;
640 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
641 patch->old_name = name;
642 } else {
643 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
644 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
645 if (has_epoch_timestamp(first)) {
646 patch->is_new = 1;
647 patch->is_delete = 0;
648 patch->new_name = name;
649 } else if (has_epoch_timestamp(second)) {
650 patch->is_new = 0;
651 patch->is_delete = 1;
652 patch->old_name = name;
653 } else {
654 patch->old_name = patch->new_name = name;
655 }
656 }
657 if (!name)
658 die("unable to find filename in patch at line %d", linenr);
659 }
661 static int gitdiff_hdrend(const char *line, struct patch *patch)
662 {
663 return -1;
664 }
666 /*
667 * We're anal about diff header consistency, to make
668 * sure that we don't end up having strange ambiguous
669 * patches floating around.
670 *
671 * As a result, gitdiff_{old|new}name() will check
672 * their names against any previous information, just
673 * to make sure..
674 */
675 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
676 {
677 if (!orig_name && !isnull)
678 return find_name(line, NULL, p_value, TERM_TAB);
680 if (orig_name) {
681 int len;
682 const char *name;
683 char *another;
684 name = orig_name;
685 len = strlen(name);
686 if (isnull)
687 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
688 another = find_name(line, NULL, p_value, TERM_TAB);
689 if (!another || memcmp(another, name, len))
690 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
691 free(another);
692 return orig_name;
693 }
694 else {
695 /* expect "/dev/null" */
696 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
697 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
698 return NULL;
699 }
700 }
702 static int gitdiff_oldname(const char *line, struct patch *patch)
703 {
704 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
705 return 0;
706 }
708 static int gitdiff_newname(const char *line, struct patch *patch)
709 {
710 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
711 return 0;
712 }
714 static int gitdiff_oldmode(const char *line, struct patch *patch)
715 {
716 patch->old_mode = strtoul(line, NULL, 8);
717 return 0;
718 }
720 static int gitdiff_newmode(const char *line, struct patch *patch)
721 {
722 patch->new_mode = strtoul(line, NULL, 8);
723 return 0;
724 }
726 static int gitdiff_delete(const char *line, struct patch *patch)
727 {
728 patch->is_delete = 1;
729 patch->old_name = patch->def_name;
730 return gitdiff_oldmode(line, patch);
731 }
733 static int gitdiff_newfile(const char *line, struct patch *patch)
734 {
735 patch->is_new = 1;
736 patch->new_name = patch->def_name;
737 return gitdiff_newmode(line, patch);
738 }
740 static int gitdiff_copysrc(const char *line, struct patch *patch)
741 {
742 patch->is_copy = 1;
743 patch->old_name = find_name(line, NULL, 0, 0);
744 return 0;
745 }
747 static int gitdiff_copydst(const char *line, struct patch *patch)
748 {
749 patch->is_copy = 1;
750 patch->new_name = find_name(line, NULL, 0, 0);
751 return 0;
752 }
754 static int gitdiff_renamesrc(const char *line, struct patch *patch)
755 {
756 patch->is_rename = 1;
757 patch->old_name = find_name(line, NULL, 0, 0);
758 return 0;
759 }
761 static int gitdiff_renamedst(const char *line, struct patch *patch)
762 {
763 patch->is_rename = 1;
764 patch->new_name = find_name(line, NULL, 0, 0);
765 return 0;
766 }
768 static int gitdiff_similarity(const char *line, struct patch *patch)
769 {
770 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
771 patch->score = 0;
772 return 0;
773 }
775 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
776 {
777 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
778 patch->score = 0;
779 return 0;
780 }
782 static int gitdiff_index(const char *line, struct patch *patch)
783 {
784 /*
785 * index line is N hexadecimal, "..", N hexadecimal,
786 * and optional space with octal mode.
787 */
788 const char *ptr, *eol;
789 int len;
791 ptr = strchr(line, '.');
792 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
793 return 0;
794 len = ptr - line;
795 memcpy(patch->old_sha1_prefix, line, len);
796 patch->old_sha1_prefix[len] = 0;
798 line = ptr + 2;
799 ptr = strchr(line, ' ');
800 eol = strchr(line, '\n');
802 if (!ptr || eol < ptr)
803 ptr = eol;
804 len = ptr - line;
806 if (40 < len)
807 return 0;
808 memcpy(patch->new_sha1_prefix, line, len);
809 patch->new_sha1_prefix[len] = 0;
810 if (*ptr == ' ')
811 patch->old_mode = strtoul(ptr+1, NULL, 8);
812 return 0;
813 }
815 /*
816 * This is normal for a diff that doesn't change anything: we'll fall through
817 * into the next diff. Tell the parser to break out.
818 */
819 static int gitdiff_unrecognized(const char *line, struct patch *patch)
820 {
821 return -1;
822 }
824 static const char *stop_at_slash(const char *line, int llen)
825 {
826 int nslash = p_value;
827 int i;
829 for (i = 0; i < llen; i++) {
830 int ch = line[i];
831 if (ch == '/' && --nslash <= 0)
832 return &line[i];
833 }
834 return NULL;
835 }
837 /*
838 * This is to extract the same name that appears on "diff --git"
839 * line. We do not find and return anything if it is a rename
840 * patch, and it is OK because we will find the name elsewhere.
841 * We need to reliably find name only when it is mode-change only,
842 * creation or deletion of an empty file. In any of these cases,
843 * both sides are the same name under a/ and b/ respectively.
844 */
845 static char *git_header_name(char *line, int llen)
846 {
847 const char *name;
848 const char *second = NULL;
849 size_t len;
851 line += strlen("diff --git ");
852 llen -= strlen("diff --git ");
854 if (*line == '"') {
855 const char *cp;
856 struct strbuf first = STRBUF_INIT;
857 struct strbuf sp = STRBUF_INIT;
859 if (unquote_c_style(&first, line, &second))
860 goto free_and_fail1;
862 /* advance to the first slash */
863 cp = stop_at_slash(first.buf, first.len);
864 /* we do not accept absolute paths */
865 if (!cp || cp == first.buf)
866 goto free_and_fail1;
867 strbuf_remove(&first, 0, cp + 1 - first.buf);
869 /*
870 * second points at one past closing dq of name.
871 * find the second name.
872 */
873 while ((second < line + llen) && isspace(*second))
874 second++;
876 if (line + llen <= second)
877 goto free_and_fail1;
878 if (*second == '"') {
879 if (unquote_c_style(&sp, second, NULL))
880 goto free_and_fail1;
881 cp = stop_at_slash(sp.buf, sp.len);
882 if (!cp || cp == sp.buf)
883 goto free_and_fail1;
884 /* They must match, otherwise ignore */
885 if (strcmp(cp + 1, first.buf))
886 goto free_and_fail1;
887 strbuf_release(&sp);
888 return strbuf_detach(&first, NULL);
889 }
891 /* unquoted second */
892 cp = stop_at_slash(second, line + llen - second);
893 if (!cp || cp == second)
894 goto free_and_fail1;
895 cp++;
896 if (line + llen - cp != first.len + 1 ||
897 memcmp(first.buf, cp, first.len))
898 goto free_and_fail1;
899 return strbuf_detach(&first, NULL);
901 free_and_fail1:
902 strbuf_release(&first);
903 strbuf_release(&sp);
904 return NULL;
905 }
907 /* unquoted first name */
908 name = stop_at_slash(line, llen);
909 if (!name || name == line)
910 return NULL;
911 name++;
913 /*
914 * since the first name is unquoted, a dq if exists must be
915 * the beginning of the second name.
916 */
917 for (second = name; second < line + llen; second++) {
918 if (*second == '"') {
919 struct strbuf sp = STRBUF_INIT;
920 const char *np;
922 if (unquote_c_style(&sp, second, NULL))
923 goto free_and_fail2;
925 np = stop_at_slash(sp.buf, sp.len);
926 if (!np || np == sp.buf)
927 goto free_and_fail2;
928 np++;
930 len = sp.buf + sp.len - np;
931 if (len < second - name &&
932 !strncmp(np, name, len) &&
933 isspace(name[len])) {
934 /* Good */
935 strbuf_remove(&sp, 0, np - sp.buf);
936 return strbuf_detach(&sp, NULL);
937 }
939 free_and_fail2:
940 strbuf_release(&sp);
941 return NULL;
942 }
943 }
945 /*
946 * Accept a name only if it shows up twice, exactly the same
947 * form.
948 */
949 for (len = 0 ; ; len++) {
950 switch (name[len]) {
951 default:
952 continue;
953 case '\n':
954 return NULL;
955 case '\t': case ' ':
956 second = name+len;
957 for (;;) {
958 char c = *second++;
959 if (c == '\n')
960 return NULL;
961 if (c == '/')
962 break;
963 }
964 if (second[len] == '\n' && !memcmp(name, second, len)) {
965 return xmemdupz(name, len);
966 }
967 }
968 }
969 }
971 /* Verify that we recognize the lines following a git header */
972 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
973 {
974 unsigned long offset;
976 /* A git diff has explicit new/delete information, so we don't guess */
977 patch->is_new = 0;
978 patch->is_delete = 0;
980 /*
981 * Some things may not have the old name in the
982 * rest of the headers anywhere (pure mode changes,
983 * or removing or adding empty files), so we get
984 * the default name from the header.
985 */
986 patch->def_name = git_header_name(line, len);
987 if (patch->def_name && root) {
988 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
989 strcpy(s, root);
990 strcpy(s + root_len, patch->def_name);
991 free(patch->def_name);
992 patch->def_name = s;
993 }
995 line += len;
996 size -= len;
997 linenr++;
998 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
999 static const struct opentry {
1000 const char *str;
1001 int (*fn)(const char *, struct patch *);
1002 } optable[] = {
1003 { "@@ -", gitdiff_hdrend },
1004 { "--- ", gitdiff_oldname },
1005 { "+++ ", gitdiff_newname },
1006 { "old mode ", gitdiff_oldmode },
1007 { "new mode ", gitdiff_newmode },
1008 { "deleted file mode ", gitdiff_delete },
1009 { "new file mode ", gitdiff_newfile },
1010 { "copy from ", gitdiff_copysrc },
1011 { "copy to ", gitdiff_copydst },
1012 { "rename old ", gitdiff_renamesrc },
1013 { "rename new ", gitdiff_renamedst },
1014 { "rename from ", gitdiff_renamesrc },
1015 { "rename to ", gitdiff_renamedst },
1016 { "similarity index ", gitdiff_similarity },
1017 { "dissimilarity index ", gitdiff_dissimilarity },
1018 { "index ", gitdiff_index },
1019 { "", gitdiff_unrecognized },
1020 };
1021 int i;
1023 len = linelen(line, size);
1024 if (!len || line[len-1] != '\n')
1025 break;
1026 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1027 const struct opentry *p = optable + i;
1028 int oplen = strlen(p->str);
1029 if (len < oplen || memcmp(p->str, line, oplen))
1030 continue;
1031 if (p->fn(line + oplen, patch) < 0)
1032 return offset;
1033 break;
1034 }
1035 }
1037 return offset;
1038 }
1040 static int parse_num(const char *line, unsigned long *p)
1041 {
1042 char *ptr;
1044 if (!isdigit(*line))
1045 return 0;
1046 *p = strtoul(line, &ptr, 10);
1047 return ptr - line;
1048 }
1050 static int parse_range(const char *line, int len, int offset, const char *expect,
1051 unsigned long *p1, unsigned long *p2)
1052 {
1053 int digits, ex;
1055 if (offset < 0 || offset >= len)
1056 return -1;
1057 line += offset;
1058 len -= offset;
1060 digits = parse_num(line, p1);
1061 if (!digits)
1062 return -1;
1064 offset += digits;
1065 line += digits;
1066 len -= digits;
1068 *p2 = 1;
1069 if (*line == ',') {
1070 digits = parse_num(line+1, p2);
1071 if (!digits)
1072 return -1;
1074 offset += digits+1;
1075 line += digits+1;
1076 len -= digits+1;
1077 }
1079 ex = strlen(expect);
1080 if (ex > len)
1081 return -1;
1082 if (memcmp(line, expect, ex))
1083 return -1;
1085 return offset + ex;
1086 }
1088 static void recount_diff(char *line, int size, struct fragment *fragment)
1089 {
1090 int oldlines = 0, newlines = 0, ret = 0;
1092 if (size < 1) {
1093 warning("recount: ignore empty hunk");
1094 return;
1095 }
1097 for (;;) {
1098 int len = linelen(line, size);
1099 size -= len;
1100 line += len;
1102 if (size < 1)
1103 break;
1105 switch (*line) {
1106 case ' ': case '\n':
1107 newlines++;
1108 /* fall through */
1109 case '-':
1110 oldlines++;
1111 continue;
1112 case '+':
1113 newlines++;
1114 continue;
1115 case '\\':
1116 continue;
1117 case '@':
1118 ret = size < 3 || prefixcmp(line, "@@ ");
1119 break;
1120 case 'd':
1121 ret = size < 5 || prefixcmp(line, "diff ");
1122 break;
1123 default:
1124 ret = -1;
1125 break;
1126 }
1127 if (ret) {
1128 warning("recount: unexpected line: %.*s",
1129 (int)linelen(line, size), line);
1130 return;
1131 }
1132 break;
1133 }
1134 fragment->oldlines = oldlines;
1135 fragment->newlines = newlines;
1136 }
1138 /*
1139 * Parse a unified diff fragment header of the
1140 * form "@@ -a,b +c,d @@"
1141 */
1142 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1143 {
1144 int offset;
1146 if (!len || line[len-1] != '\n')
1147 return -1;
1149 /* Figure out the number of lines in a fragment */
1150 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1151 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1153 return offset;
1154 }
1156 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1157 {
1158 unsigned long offset, len;
1160 patch->is_toplevel_relative = 0;
1161 patch->is_rename = patch->is_copy = 0;
1162 patch->is_new = patch->is_delete = -1;
1163 patch->old_mode = patch->new_mode = 0;
1164 patch->old_name = patch->new_name = NULL;
1165 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1166 unsigned long nextlen;
1168 len = linelen(line, size);
1169 if (!len)
1170 break;
1172 /* Testing this early allows us to take a few shortcuts.. */
1173 if (len < 6)
1174 continue;
1176 /*
1177 * Make sure we don't find any unconnected patch fragments.
1178 * That's a sign that we didn't find a header, and that a
1179 * patch has become corrupted/broken up.
1180 */
1181 if (!memcmp("@@ -", line, 4)) {
1182 struct fragment dummy;
1183 if (parse_fragment_header(line, len, &dummy) < 0)
1184 continue;
1185 die("patch fragment without header at line %d: %.*s",
1186 linenr, (int)len-1, line);
1187 }
1189 if (size < len + 6)
1190 break;
1192 /*
1193 * Git patch? It might not have a real patch, just a rename
1194 * or mode change, so we handle that specially
1195 */
1196 if (!memcmp("diff --git ", line, 11)) {
1197 int git_hdr_len = parse_git_header(line, len, size, patch);
1198 if (git_hdr_len <= len)
1199 continue;
1200 if (!patch->old_name && !patch->new_name) {
1201 if (!patch->def_name)
1202 die("git diff header lacks filename information (line %d)", linenr);
1203 patch->old_name = patch->new_name = patch->def_name;
1204 }
1205 patch->is_toplevel_relative = 1;
1206 *hdrsize = git_hdr_len;
1207 return offset;
1208 }
1210 /* --- followed by +++ ? */
1211 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1212 continue;
1214 /*
1215 * We only accept unified patches, so we want it to
1216 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1217 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1218 */
1219 nextlen = linelen(line + len, size - len);
1220 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1221 continue;
1223 /* Ok, we'll consider it a patch */
1224 parse_traditional_patch(line, line+len, patch);
1225 *hdrsize = len + nextlen;
1226 linenr += 2;
1227 return offset;
1228 }
1229 return -1;
1230 }
1232 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1233 {
1234 char *err;
1236 if (!result)
1237 return;
1239 whitespace_error++;
1240 if (squelch_whitespace_errors &&
1241 squelch_whitespace_errors < whitespace_error)
1242 return;
1244 err = whitespace_error_string(result);
1245 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1246 patch_input_file, linenr, err, len, line);
1247 free(err);
1248 }
1250 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1251 {
1252 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1254 record_ws_error(result, line + 1, len - 2, linenr);
1255 }
1257 /*
1258 * Parse a unified diff. Note that this really needs to parse each
1259 * fragment separately, since the only way to know the difference
1260 * between a "---" that is part of a patch, and a "---" that starts
1261 * the next patch is to look at the line counts..
1262 */
1263 static int parse_fragment(char *line, unsigned long size,
1264 struct patch *patch, struct fragment *fragment)
1265 {
1266 int added, deleted;
1267 int len = linelen(line, size), offset;
1268 unsigned long oldlines, newlines;
1269 unsigned long leading, trailing;
1271 offset = parse_fragment_header(line, len, fragment);
1272 if (offset < 0)
1273 return -1;
1274 if (offset > 0 && patch->recount)
1275 recount_diff(line + offset, size - offset, fragment);
1276 oldlines = fragment->oldlines;
1277 newlines = fragment->newlines;
1278 leading = 0;
1279 trailing = 0;
1281 /* Parse the thing.. */
1282 line += len;
1283 size -= len;
1284 linenr++;
1285 added = deleted = 0;
1286 for (offset = len;
1287 0 < size;
1288 offset += len, size -= len, line += len, linenr++) {
1289 if (!oldlines && !newlines)
1290 break;
1291 len = linelen(line, size);
1292 if (!len || line[len-1] != '\n')
1293 return -1;
1294 switch (*line) {
1295 default:
1296 return -1;
1297 case '\n': /* newer GNU diff, an empty context line */
1298 case ' ':
1299 oldlines--;
1300 newlines--;
1301 if (!deleted && !added)
1302 leading++;
1303 trailing++;
1304 break;
1305 case '-':
1306 if (apply_in_reverse &&
1307 ws_error_action != nowarn_ws_error)
1308 check_whitespace(line, len, patch->ws_rule);
1309 deleted++;
1310 oldlines--;
1311 trailing = 0;
1312 break;
1313 case '+':
1314 if (!apply_in_reverse &&
1315 ws_error_action != nowarn_ws_error)
1316 check_whitespace(line, len, patch->ws_rule);
1317 added++;
1318 newlines--;
1319 trailing = 0;
1320 break;
1322 /*
1323 * We allow "\ No newline at end of file". Depending
1324 * on locale settings when the patch was produced we
1325 * don't know what this line looks like. The only
1326 * thing we do know is that it begins with "\ ".
1327 * Checking for 12 is just for sanity check -- any
1328 * l10n of "\ No newline..." is at least that long.
1329 */
1330 case '\\':
1331 if (len < 12 || memcmp(line, "\\ ", 2))
1332 return -1;
1333 break;
1334 }
1335 }
1336 if (oldlines || newlines)
1337 return -1;
1338 fragment->leading = leading;
1339 fragment->trailing = trailing;
1341 /*
1342 * If a fragment ends with an incomplete line, we failed to include
1343 * it in the above loop because we hit oldlines == newlines == 0
1344 * before seeing it.
1345 */
1346 if (12 < size && !memcmp(line, "\\ ", 2))
1347 offset += linelen(line, size);
1349 patch->lines_added += added;
1350 patch->lines_deleted += deleted;
1352 if (0 < patch->is_new && oldlines)
1353 return error("new file depends on old contents");
1354 if (0 < patch->is_delete && newlines)
1355 return error("deleted file still has contents");
1356 return offset;
1357 }
1359 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1360 {
1361 unsigned long offset = 0;
1362 unsigned long oldlines = 0, newlines = 0, context = 0;
1363 struct fragment **fragp = &patch->fragments;
1365 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1366 struct fragment *fragment;
1367 int len;
1369 fragment = xcalloc(1, sizeof(*fragment));
1370 fragment->linenr = linenr;
1371 len = parse_fragment(line, size, patch, fragment);
1372 if (len <= 0)
1373 die("corrupt patch at line %d", linenr);
1374 fragment->patch = line;
1375 fragment->size = len;
1376 oldlines += fragment->oldlines;
1377 newlines += fragment->newlines;
1378 context += fragment->leading + fragment->trailing;
1380 *fragp = fragment;
1381 fragp = &fragment->next;
1383 offset += len;
1384 line += len;
1385 size -= len;
1386 }
1388 /*
1389 * If something was removed (i.e. we have old-lines) it cannot
1390 * be creation, and if something was added it cannot be
1391 * deletion. However, the reverse is not true; --unified=0
1392 * patches that only add are not necessarily creation even
1393 * though they do not have any old lines, and ones that only
1394 * delete are not necessarily deletion.
1395 *
1396 * Unfortunately, a real creation/deletion patch do _not_ have
1397 * any context line by definition, so we cannot safely tell it
1398 * apart with --unified=0 insanity. At least if the patch has
1399 * more than one hunk it is not creation or deletion.
1400 */
1401 if (patch->is_new < 0 &&
1402 (oldlines || (patch->fragments && patch->fragments->next)))
1403 patch->is_new = 0;
1404 if (patch->is_delete < 0 &&
1405 (newlines || (patch->fragments && patch->fragments->next)))
1406 patch->is_delete = 0;
1408 if (0 < patch->is_new && oldlines)
1409 die("new file %s depends on old contents", patch->new_name);
1410 if (0 < patch->is_delete && newlines)
1411 die("deleted file %s still has contents", patch->old_name);
1412 if (!patch->is_delete && !newlines && context)
1413 fprintf(stderr, "** warning: file %s becomes empty but "
1414 "is not deleted\n", patch->new_name);
1416 return offset;
1417 }
1419 static inline int metadata_changes(struct patch *patch)
1420 {
1421 return patch->is_rename > 0 ||
1422 patch->is_copy > 0 ||
1423 patch->is_new > 0 ||
1424 patch->is_delete ||
1425 (patch->old_mode && patch->new_mode &&
1426 patch->old_mode != patch->new_mode);
1427 }
1429 static char *inflate_it(const void *data, unsigned long size,
1430 unsigned long inflated_size)
1431 {
1432 z_stream stream;
1433 void *out;
1434 int st;
1436 memset(&stream, 0, sizeof(stream));
1438 stream.next_in = (unsigned char *)data;
1439 stream.avail_in = size;
1440 stream.next_out = out = xmalloc(inflated_size);
1441 stream.avail_out = inflated_size;
1442 git_inflate_init(&stream);
1443 st = git_inflate(&stream, Z_FINISH);
1444 git_inflate_end(&stream);
1445 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1446 free(out);
1447 return NULL;
1448 }
1449 return out;
1450 }
1452 static struct fragment *parse_binary_hunk(char **buf_p,
1453 unsigned long *sz_p,
1454 int *status_p,
1455 int *used_p)
1456 {
1457 /*
1458 * Expect a line that begins with binary patch method ("literal"
1459 * or "delta"), followed by the length of data before deflating.
1460 * a sequence of 'length-byte' followed by base-85 encoded data
1461 * should follow, terminated by a newline.
1462 *
1463 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1464 * and we would limit the patch line to 66 characters,
1465 * so one line can fit up to 13 groups that would decode
1466 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1467 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1468 */
1469 int llen, used;
1470 unsigned long size = *sz_p;
1471 char *buffer = *buf_p;
1472 int patch_method;
1473 unsigned long origlen;
1474 char *data = NULL;
1475 int hunk_size = 0;
1476 struct fragment *frag;
1478 llen = linelen(buffer, size);
1479 used = llen;
1481 *status_p = 0;
1483 if (!prefixcmp(buffer, "delta ")) {
1484 patch_method = BINARY_DELTA_DEFLATED;
1485 origlen = strtoul(buffer + 6, NULL, 10);
1486 }
1487 else if (!prefixcmp(buffer, "literal ")) {
1488 patch_method = BINARY_LITERAL_DEFLATED;
1489 origlen = strtoul(buffer + 8, NULL, 10);
1490 }
1491 else
1492 return NULL;
1494 linenr++;
1495 buffer += llen;
1496 while (1) {
1497 int byte_length, max_byte_length, newsize;
1498 llen = linelen(buffer, size);
1499 used += llen;
1500 linenr++;
1501 if (llen == 1) {
1502 /* consume the blank line */
1503 buffer++;
1504 size--;
1505 break;
1506 }
1507 /*
1508 * Minimum line is "A00000\n" which is 7-byte long,
1509 * and the line length must be multiple of 5 plus 2.
1510 */
1511 if ((llen < 7) || (llen-2) % 5)
1512 goto corrupt;
1513 max_byte_length = (llen - 2) / 5 * 4;
1514 byte_length = *buffer;
1515 if ('A' <= byte_length && byte_length <= 'Z')
1516 byte_length = byte_length - 'A' + 1;
1517 else if ('a' <= byte_length && byte_length <= 'z')
1518 byte_length = byte_length - 'a' + 27;
1519 else
1520 goto corrupt;
1521 /* if the input length was not multiple of 4, we would
1522 * have filler at the end but the filler should never
1523 * exceed 3 bytes
1524 */
1525 if (max_byte_length < byte_length ||
1526 byte_length <= max_byte_length - 4)
1527 goto corrupt;
1528 newsize = hunk_size + byte_length;
1529 data = xrealloc(data, newsize);
1530 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1531 goto corrupt;
1532 hunk_size = newsize;
1533 buffer += llen;
1534 size -= llen;
1535 }
1537 frag = xcalloc(1, sizeof(*frag));
1538 frag->patch = inflate_it(data, hunk_size, origlen);
1539 if (!frag->patch)
1540 goto corrupt;
1541 free(data);
1542 frag->size = origlen;
1543 *buf_p = buffer;
1544 *sz_p = size;
1545 *used_p = used;
1546 frag->binary_patch_method = patch_method;
1547 return frag;
1549 corrupt:
1550 free(data);
1551 *status_p = -1;
1552 error("corrupt binary patch at line %d: %.*s",
1553 linenr-1, llen-1, buffer);
1554 return NULL;
1555 }
1557 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1558 {
1559 /*
1560 * We have read "GIT binary patch\n"; what follows is a line
1561 * that says the patch method (currently, either "literal" or
1562 * "delta") and the length of data before deflating; a
1563 * sequence of 'length-byte' followed by base-85 encoded data
1564 * follows.
1565 *
1566 * When a binary patch is reversible, there is another binary
1567 * hunk in the same format, starting with patch method (either
1568 * "literal" or "delta") with the length of data, and a sequence
1569 * of length-byte + base-85 encoded data, terminated with another
1570 * empty line. This data, when applied to the postimage, produces
1571 * the preimage.
1572 */
1573 struct fragment *forward;
1574 struct fragment *reverse;
1575 int status;
1576 int used, used_1;
1578 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1579 if (!forward && !status)
1580 /* there has to be one hunk (forward hunk) */
1581 return error("unrecognized binary patch at line %d", linenr-1);
1582 if (status)
1583 /* otherwise we already gave an error message */
1584 return status;
1586 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1587 if (reverse)
1588 used += used_1;
1589 else if (status) {
1590 /*
1591 * Not having reverse hunk is not an error, but having
1592 * a corrupt reverse hunk is.
1593 */
1594 free((void*) forward->patch);
1595 free(forward);
1596 return status;
1597 }
1598 forward->next = reverse;
1599 patch->fragments = forward;
1600 patch->is_binary = 1;
1601 return used;
1602 }
1604 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1605 {
1606 int hdrsize, patchsize;
1607 int offset = find_header(buffer, size, &hdrsize, patch);
1609 if (offset < 0)
1610 return offset;
1612 patch->ws_rule = whitespace_rule(patch->new_name
1613 ? patch->new_name
1614 : patch->old_name);
1616 patchsize = parse_single_patch(buffer + offset + hdrsize,
1617 size - offset - hdrsize, patch);
1619 if (!patchsize) {
1620 static const char *binhdr[] = {
1621 "Binary files ",
1622 "Files ",
1623 NULL,
1624 };
1625 static const char git_binary[] = "GIT binary patch\n";
1626 int i;
1627 int hd = hdrsize + offset;
1628 unsigned long llen = linelen(buffer + hd, size - hd);
1630 if (llen == sizeof(git_binary) - 1 &&
1631 !memcmp(git_binary, buffer + hd, llen)) {
1632 int used;
1633 linenr++;
1634 used = parse_binary(buffer + hd + llen,
1635 size - hd - llen, patch);
1636 if (used)
1637 patchsize = used + llen;
1638 else
1639 patchsize = 0;
1640 }
1641 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1642 for (i = 0; binhdr[i]; i++) {
1643 int len = strlen(binhdr[i]);
1644 if (len < size - hd &&
1645 !memcmp(binhdr[i], buffer + hd, len)) {
1646 linenr++;
1647 patch->is_binary = 1;
1648 patchsize = llen;
1649 break;
1650 }
1651 }
1652 }
1654 /* Empty patch cannot be applied if it is a text patch
1655 * without metadata change. A binary patch appears
1656 * empty to us here.
1657 */
1658 if ((apply || check) &&
1659 (!patch->is_binary && !metadata_changes(patch)))
1660 die("patch with only garbage at line %d", linenr);
1661 }
1663 return offset + hdrsize + patchsize;
1664 }
1666 #define swap(a,b) myswap((a),(b),sizeof(a))
1668 #define myswap(a, b, size) do { \
1669 unsigned char mytmp[size]; \
1670 memcpy(mytmp, &a, size); \
1671 memcpy(&a, &b, size); \
1672 memcpy(&b, mytmp, size); \
1673 } while (0)
1675 static void reverse_patches(struct patch *p)
1676 {
1677 for (; p; p = p->next) {
1678 struct fragment *frag = p->fragments;
1680 swap(p->new_name, p->old_name);
1681 swap(p->new_mode, p->old_mode);
1682 swap(p->is_new, p->is_delete);
1683 swap(p->lines_added, p->lines_deleted);
1684 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1686 for (; frag; frag = frag->next) {
1687 swap(frag->newpos, frag->oldpos);
1688 swap(frag->newlines, frag->oldlines);
1689 }
1690 }
1691 }
1693 static const char pluses[] =
1694 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1695 static const char minuses[]=
1696 "----------------------------------------------------------------------";
1698 static void show_stats(struct patch *patch)
1699 {
1700 struct strbuf qname = STRBUF_INIT;
1701 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1702 int max, add, del;
1704 quote_c_style(cp, &qname, NULL, 0);
1706 /*
1707 * "scale" the filename
1708 */
1709 max = max_len;
1710 if (max > 50)
1711 max = 50;
1713 if (qname.len > max) {
1714 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1715 if (!cp)
1716 cp = qname.buf + qname.len + 3 - max;
1717 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1718 }
1720 if (patch->is_binary) {
1721 printf(" %-*s | Bin\n", max, qname.buf);
1722 strbuf_release(&qname);
1723 return;
1724 }
1726 printf(" %-*s |", max, qname.buf);
1727 strbuf_release(&qname);
1729 /*
1730 * scale the add/delete
1731 */
1732 max = max + max_change > 70 ? 70 - max : max_change;
1733 add = patch->lines_added;
1734 del = patch->lines_deleted;
1736 if (max_change > 0) {
1737 int total = ((add + del) * max + max_change / 2) / max_change;
1738 add = (add * max + max_change / 2) / max_change;
1739 del = total - add;
1740 }
1741 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1742 add, pluses, del, minuses);
1743 }
1745 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1746 {
1747 switch (st->st_mode & S_IFMT) {
1748 case S_IFLNK:
1749 if (strbuf_readlink(buf, path, st->st_size) < 0)
1750 return error("unable to read symlink %s", path);
1751 return 0;
1752 case S_IFREG:
1753 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1754 return error("unable to open or read %s", path);
1755 convert_to_git(path, buf->buf, buf->len, buf, 0);
1756 return 0;
1757 default:
1758 return -1;
1759 }
1760 }
1762 /*
1763 * Update the preimage, and the common lines in postimage,
1764 * from buffer buf of length len. If postlen is 0 the postimage
1765 * is updated in place, otherwise it's updated on a new buffer
1766 * of length postlen
1767 */
1769 static void update_pre_post_images(struct image *preimage,
1770 struct image *postimage,
1771 char *buf,
1772 size_t len, size_t postlen)
1773 {
1774 int i, ctx;
1775 char *new, *old, *fixed;
1776 struct image fixed_preimage;
1778 /*
1779 * Update the preimage with whitespace fixes. Note that we
1780 * are not losing preimage->buf -- apply_one_fragment() will
1781 * free "oldlines".
1782 */
1783 prepare_image(&fixed_preimage, buf, len, 1);
1784 assert(fixed_preimage.nr == preimage->nr);
1785 for (i = 0; i < preimage->nr; i++)
1786 fixed_preimage.line[i].flag = preimage->line[i].flag;
1787 free(preimage->line_allocated);
1788 *preimage = fixed_preimage;
1790 /*
1791 * Adjust the common context lines in postimage. This can be
1792 * done in-place when we are just doing whitespace fixing,
1793 * which does not make the string grow, but needs a new buffer
1794 * when ignoring whitespace causes the update, since in this case
1795 * we could have e.g. tabs converted to multiple spaces.
1796 * We trust the caller to tell us if the update can be done
1797 * in place (postlen==0) or not.
1798 */
1799 old = postimage->buf;
1800 if (postlen)
1801 new = postimage->buf = xmalloc(postlen);
1802 else
1803 new = old;
1804 fixed = preimage->buf;
1805 for (i = ctx = 0; i < postimage->nr; i++) {
1806 size_t len = postimage->line[i].len;
1807 if (!(postimage->line[i].flag & LINE_COMMON)) {
1808 /* an added line -- no counterparts in preimage */
1809 memmove(new, old, len);
1810 old += len;
1811 new += len;
1812 continue;
1813 }
1815 /* a common context -- skip it in the original postimage */
1816 old += len;
1818 /* and find the corresponding one in the fixed preimage */
1819 while (ctx < preimage->nr &&
1820 !(preimage->line[ctx].flag & LINE_COMMON)) {
1821 fixed += preimage->line[ctx].len;
1822 ctx++;
1823 }
1824 if (preimage->nr <= ctx)
1825 die("oops");
1827 /* and copy it in, while fixing the line length */
1828 len = preimage->line[ctx].len;
1829 memcpy(new, fixed, len);
1830 new += len;
1831 fixed += len;
1832 postimage->line[i].len = len;
1833 ctx++;
1834 }
1836 /* Fix the length of the whole thing */
1837 postimage->len = new - postimage->buf;
1838 }
1840 static int match_fragment(struct image *img,
1841 struct image *preimage,
1842 struct image *postimage,
1843 unsigned long try,
1844 int try_lno,
1845 unsigned ws_rule,
1846 int match_beginning, int match_end)
1847 {
1848 int i;
1849 char *fixed_buf, *buf, *orig, *target;
1851 if (preimage->nr + try_lno > img->nr)
1852 return 0;
1854 if (match_beginning && try_lno)
1855 return 0;
1857 if (match_end && preimage->nr + try_lno != img->nr)
1858 return 0;
1860 /* Quick hash check */
1861 for (i = 0; i < preimage->nr; i++)
1862 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1863 return 0;
1865 /*
1866 * Do we have an exact match? If we were told to match
1867 * at the end, size must be exactly at try+fragsize,
1868 * otherwise try+fragsize must be still within the preimage,
1869 * and either case, the old piece should match the preimage
1870 * exactly.
1871 */
1872 if ((match_end
1873 ? (try + preimage->len == img->len)
1874 : (try + preimage->len <= img->len)) &&
1875 !memcmp(img->buf + try, preimage->buf, preimage->len))
1876 return 1;
1878 /*
1879 * No exact match. If we are ignoring whitespace, run a line-by-line
1880 * fuzzy matching. We collect all the line length information because
1881 * we need it to adjust whitespace if we match.
1882 */
1883 if (ws_ignore_action == ignore_ws_change) {
1884 size_t imgoff = 0;
1885 size_t preoff = 0;
1886 size_t postlen = postimage->len;
1887 for (i = 0; i < preimage->nr; i++) {
1888 size_t prelen = preimage->line[i].len;
1889 size_t imglen = img->line[try_lno+i].len;
1891 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
1892 preimage->buf + preoff, prelen))
1893 return 0;
1894 if (preimage->line[i].flag & LINE_COMMON)
1895 postlen += imglen - prelen;
1896 imgoff += imglen;
1897 preoff += prelen;
1898 }
1900 /*
1901 * Ok, the preimage matches with whitespace fuzz. Update it and
1902 * the common postimage lines to use the same whitespace as the
1903 * target. imgoff now holds the true length of the target that
1904 * matches the preimage, and we need to update the line lengths
1905 * of the preimage to match the target ones.
1906 */
1907 fixed_buf = xmalloc(imgoff);
1908 memcpy(fixed_buf, img->buf + try, imgoff);
1909 for (i = 0; i < preimage->nr; i++)
1910 preimage->line[i].len = img->line[try_lno+i].len;
1912 /*
1913 * Update the preimage buffer and the postimage context lines.
1914 */
1915 update_pre_post_images(preimage, postimage,
1916 fixed_buf, imgoff, postlen);
1917 return 1;
1918 }
1920 if (ws_error_action != correct_ws_error)
1921 return 0;
1923 /*
1924 * The hunk does not apply byte-by-byte, but the hash says
1925 * it might with whitespace fuzz. We haven't been asked to
1926 * ignore whitespace, we were asked to correct whitespace
1927 * errors, so let's try matching after whitespace correction.
1928 */
1929 fixed_buf = xmalloc(preimage->len + 1);
1930 buf = fixed_buf;
1931 orig = preimage->buf;
1932 target = img->buf + try;
1933 for (i = 0; i < preimage->nr; i++) {
1934 size_t fixlen; /* length after fixing the preimage */
1935 size_t oldlen = preimage->line[i].len;
1936 size_t tgtlen = img->line[try_lno + i].len;
1937 size_t tgtfixlen; /* length after fixing the target line */
1938 char tgtfixbuf[1024], *tgtfix;
1939 int match;
1941 /* Try fixing the line in the preimage */
1942 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1944 /* Try fixing the line in the target */
1945 if (sizeof(tgtfixbuf) > tgtlen)
1946 tgtfix = tgtfixbuf;
1947 else
1948 tgtfix = xmalloc(tgtlen);
1949 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1951 /*
1952 * If they match, either the preimage was based on
1953 * a version before our tree fixed whitespace breakage,
1954 * or we are lacking a whitespace-fix patch the tree
1955 * the preimage was based on already had (i.e. target
1956 * has whitespace breakage, the preimage doesn't).
1957 * In either case, we are fixing the whitespace breakages
1958 * so we might as well take the fix together with their
1959 * real change.
1960 */
1961 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1963 if (tgtfix != tgtfixbuf)
1964 free(tgtfix);
1965 if (!match)
1966 goto unmatch_exit;
1968 orig += oldlen;
1969 buf += fixlen;
1970 target += tgtlen;
1971 }
1973 /*
1974 * Yes, the preimage is based on an older version that still
1975 * has whitespace breakages unfixed, and fixing them makes the
1976 * hunk match. Update the context lines in the postimage.
1977 */
1978 update_pre_post_images(preimage, postimage,
1979 fixed_buf, buf - fixed_buf, 0);
1980 return 1;
1982 unmatch_exit:
1983 free(fixed_buf);
1984 return 0;
1985 }
1987 static int find_pos(struct image *img,
1988 struct image *preimage,
1989 struct image *postimage,
1990 int line,
1991 unsigned ws_rule,
1992 int match_beginning, int match_end)
1993 {
1994 int i;
1995 unsigned long backwards, forwards, try;
1996 int backwards_lno, forwards_lno, try_lno;
1998 if (preimage->nr > img->nr)
1999 return -1;
2001 /*
2002 * If match_begining or match_end is specified, there is no
2003 * point starting from a wrong line that will never match and
2004 * wander around and wait for a match at the specified end.
2005 */
2006 if (match_beginning)
2007 line = 0;
2008 else if (match_end)
2009 line = img->nr - preimage->nr;
2011 if (line > img->nr)
2012 line = img->nr;
2014 try = 0;
2015 for (i = 0; i < line; i++)
2016 try += img->line[i].len;
2018 /*
2019 * There's probably some smart way to do this, but I'll leave
2020 * that to the smart and beautiful people. I'm simple and stupid.
2021 */
2022 backwards = try;
2023 backwards_lno = line;
2024 forwards = try;
2025 forwards_lno = line;
2026 try_lno = line;
2028 for (i = 0; ; i++) {
2029 if (match_fragment(img, preimage, postimage,
2030 try, try_lno, ws_rule,
2031 match_beginning, match_end))
2032 return try_lno;
2034 again:
2035 if (backwards_lno == 0 && forwards_lno == img->nr)
2036 break;
2038 if (i & 1) {
2039 if (backwards_lno == 0) {
2040 i++;
2041 goto again;
2042 }
2043 backwards_lno--;
2044 backwards -= img->line[backwards_lno].len;
2045 try = backwards;
2046 try_lno = backwards_lno;
2047 } else {
2048 if (forwards_lno == img->nr) {
2049 i++;
2050 goto again;
2051 }
2052 forwards += img->line[forwards_lno].len;
2053 forwards_lno++;
2054 try = forwards;
2055 try_lno = forwards_lno;
2056 }
2058 }
2059 return -1;
2060 }
2062 static void remove_first_line(struct image *img)
2063 {
2064 img->buf += img->line[0].len;
2065 img->len -= img->line[0].len;
2066 img->line++;
2067 img->nr--;
2068 }
2070 static void remove_last_line(struct image *img)
2071 {
2072 img->len -= img->line[--img->nr].len;
2073 }
2075 static void update_image(struct image *img,
2076 int applied_pos,
2077 struct image *preimage,
2078 struct image *postimage)
2079 {
2080 /*
2081 * remove the copy of preimage at offset in img
2082 * and replace it with postimage
2083 */
2084 int i, nr;
2085 size_t remove_count, insert_count, applied_at = 0;
2086 char *result;
2088 for (i = 0; i < applied_pos; i++)
2089 applied_at += img->line[i].len;
2091 remove_count = 0;
2092 for (i = 0; i < preimage->nr; i++)
2093 remove_count += img->line[applied_pos + i].len;
2094 insert_count = postimage->len;
2096 /* Adjust the contents */
2097 result = xmalloc(img->len + insert_count - remove_count + 1);
2098 memcpy(result, img->buf, applied_at);
2099 memcpy(result + applied_at, postimage->buf, postimage->len);
2100 memcpy(result + applied_at + postimage->len,
2101 img->buf + (applied_at + remove_count),
2102 img->len - (applied_at + remove_count));
2103 free(img->buf);
2104 img->buf = result;
2105 img->len += insert_count - remove_count;
2106 result[img->len] = '\0';
2108 /* Adjust the line table */
2109 nr = img->nr + postimage->nr - preimage->nr;
2110 if (preimage->nr < postimage->nr) {
2111 /*
2112 * NOTE: this knows that we never call remove_first_line()
2113 * on anything other than pre/post image.
2114 */
2115 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2116 img->line_allocated = img->line;
2117 }
2118 if (preimage->nr != postimage->nr)
2119 memmove(img->line + applied_pos + postimage->nr,
2120 img->line + applied_pos + preimage->nr,
2121 (img->nr - (applied_pos + preimage->nr)) *
2122 sizeof(*img->line));
2123 memcpy(img->line + applied_pos,
2124 postimage->line,
2125 postimage->nr * sizeof(*img->line));
2126 img->nr = nr;
2127 }
2129 static int apply_one_fragment(struct image *img, struct fragment *frag,
2130 int inaccurate_eof, unsigned ws_rule)
2131 {
2132 int match_beginning, match_end;
2133 const char *patch = frag->patch;
2134 int size = frag->size;
2135 char *old, *new, *oldlines, *newlines;
2136 int new_blank_lines_at_end = 0;
2137 unsigned long leading, trailing;
2138 int pos, applied_pos;
2139 struct image preimage;
2140 struct image postimage;
2142 memset(&preimage, 0, sizeof(preimage));
2143 memset(&postimage, 0, sizeof(postimage));
2144 oldlines = xmalloc(size);
2145 newlines = xmalloc(size);
2147 old = oldlines;
2148 new = newlines;
2149 while (size > 0) {
2150 char first;
2151 int len = linelen(patch, size);
2152 int plen, added;
2153 int added_blank_line = 0;
2154 int is_blank_context = 0;
2156 if (!len)
2157 break;
2159 /*
2160 * "plen" is how much of the line we should use for
2161 * the actual patch data. Normally we just remove the
2162 * first character on the line, but if the line is
2163 * followed by "\ No newline", then we also remove the
2164 * last one (which is the newline, of course).
2165 */
2166 plen = len - 1;
2167 if (len < size && patch[len] == '\\')
2168 plen--;
2169 first = *patch;
2170 if (apply_in_reverse) {
2171 if (first == '-')
2172 first = '+';
2173 else if (first == '+')
2174 first = '-';
2175 }
2177 switch (first) {
2178 case '\n':
2179 /* Newer GNU diff, empty context line */
2180 if (plen < 0)
2181 /* ... followed by '\No newline'; nothing */
2182 break;
2183 *old++ = '\n';
2184 *new++ = '\n';
2185 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2186 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2187 is_blank_context = 1;
2188 break;
2189 case ' ':
2190 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2191 ws_blank_line(patch + 1, plen, ws_rule))
2192 is_blank_context = 1;
2193 case '-':
2194 memcpy(old, patch + 1, plen);
2195 add_line_info(&preimage, old, plen,
2196 (first == ' ' ? LINE_COMMON : 0));
2197 old += plen;
2198 if (first == '-')
2199 break;
2200 /* Fall-through for ' ' */
2201 case '+':
2202 /* --no-add does not add new lines */
2203 if (first == '+' && no_add)
2204 break;
2206 if (first != '+' ||
2207 !whitespace_error ||
2208 ws_error_action != correct_ws_error) {
2209 memcpy(new, patch + 1, plen);
2210 added = plen;
2211 }
2212 else {
2213 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2214 }
2215 add_line_info(&postimage, new, added,
2216 (first == '+' ? 0 : LINE_COMMON));
2217 new += added;
2218 if (first == '+' &&
2219 (ws_rule & WS_BLANK_AT_EOF) &&
2220 ws_blank_line(patch + 1, plen, ws_rule))
2221 added_blank_line = 1;
2222 break;
2223 case '@': case '\\':
2224 /* Ignore it, we already handled it */
2225 break;
2226 default:
2227 if (apply_verbosely)
2228 error("invalid start of line: '%c'", first);
2229 return -1;
2230 }
2231 if (added_blank_line)
2232 new_blank_lines_at_end++;
2233 else if (is_blank_context)
2234 ;
2235 else
2236 new_blank_lines_at_end = 0;
2237 patch += len;
2238 size -= len;
2239 }
2240 if (inaccurate_eof &&
2241 old > oldlines && old[-1] == '\n' &&
2242 new > newlines && new[-1] == '\n') {
2243 old--;
2244 new--;
2245 }
2247 leading = frag->leading;
2248 trailing = frag->trailing;
2250 /*
2251 * A hunk to change lines at the beginning would begin with
2252 * @@ -1,L +N,M @@
2253 * but we need to be careful. -U0 that inserts before the second
2254 * line also has this pattern.
2255 *
2256 * And a hunk to add to an empty file would begin with
2257 * @@ -0,0 +N,M @@
2258 *
2259 * In other words, a hunk that is (frag->oldpos <= 1) with or
2260 * without leading context must match at the beginning.
2261 */
2262 match_beginning = (!frag->oldpos ||
2263 (frag->oldpos == 1 && !unidiff_zero));
2265 /*
2266 * A hunk without trailing lines must match at the end.
2267 * However, we simply cannot tell if a hunk must match end
2268 * from the lack of trailing lines if the patch was generated
2269 * with unidiff without any context.
2270 */
2271 match_end = !unidiff_zero && !trailing;
2273 pos = frag->newpos ? (frag->newpos - 1) : 0;
2274 preimage.buf = oldlines;
2275 preimage.len = old - oldlines;
2276 postimage.buf = newlines;
2277 postimage.len = new - newlines;
2278 preimage.line = preimage.line_allocated;
2279 postimage.line = postimage.line_allocated;
2281 for (;;) {
2283 applied_pos = find_pos(img, &preimage, &postimage, pos,
2284 ws_rule, match_beginning, match_end);
2286 if (applied_pos >= 0)
2287 break;
2289 /* Am I at my context limits? */
2290 if ((leading <= p_context) && (trailing <= p_context))
2291 break;
2292 if (match_beginning || match_end) {
2293 match_beginning = match_end = 0;
2294 continue;
2295 }
2297 /*
2298 * Reduce the number of context lines; reduce both
2299 * leading and trailing if they are equal otherwise
2300 * just reduce the larger context.
2301 */
2302 if (leading >= trailing) {
2303 remove_first_line(&preimage);
2304 remove_first_line(&postimage);
2305 pos--;
2306 leading--;
2307 }
2308 if (trailing > leading) {
2309 remove_last_line(&preimage);
2310 remove_last_line(&postimage);
2311 trailing--;
2312 }
2313 }
2315 if (applied_pos >= 0) {
2316 if (new_blank_lines_at_end &&
2317 preimage.nr + applied_pos == img->nr &&
2318 (ws_rule & WS_BLANK_AT_EOF) &&
2319 ws_error_action != nowarn_ws_error) {
2320 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2321 if (ws_error_action == correct_ws_error) {
2322 while (new_blank_lines_at_end--)
2323 remove_last_line(&postimage);
2324 }
2325 /*
2326 * We would want to prevent write_out_results()
2327 * from taking place in apply_patch() that follows
2328 * the callchain led us here, which is:
2329 * apply_patch->check_patch_list->check_patch->
2330 * apply_data->apply_fragments->apply_one_fragment
2331 */
2332 if (ws_error_action == die_on_ws_error)
2333 apply = 0;
2334 }
2336 /*
2337 * Warn if it was necessary to reduce the number
2338 * of context lines.
2339 */
2340 if ((leading != frag->leading) ||
2341 (trailing != frag->trailing))
2342 fprintf(stderr, "Context reduced to (%ld/%ld)"
2343 " to apply fragment at %d\n",
2344 leading, trailing, applied_pos+1);
2345 update_image(img, applied_pos, &preimage, &postimage);
2346 } else {
2347 if (apply_verbosely)
2348 error("while searching for:\n%.*s",
2349 (int)(old - oldlines), oldlines);
2350 }
2352 free(oldlines);
2353 free(newlines);
2354 free(preimage.line_allocated);
2355 free(postimage.line_allocated);
2357 return (applied_pos < 0);
2358 }
2360 static int apply_binary_fragment(struct image *img, struct patch *patch)
2361 {
2362 struct fragment *fragment = patch->fragments;
2363 unsigned long len;
2364 void *dst;
2366 /* Binary patch is irreversible without the optional second hunk */
2367 if (apply_in_reverse) {
2368 if (!fragment->next)
2369 return error("cannot reverse-apply a binary patch "
2370 "without the reverse hunk to '%s'",
2371 patch->new_name
2372 ? patch->new_name : patch->old_name);
2373 fragment = fragment->next;
2374 }
2375 switch (fragment->binary_patch_method) {
2376 case BINARY_DELTA_DEFLATED:
2377 dst = patch_delta(img->buf, img->len, fragment->patch,
2378 fragment->size, &len);
2379 if (!dst)
2380 return -1;
2381 clear_image(img);
2382 img->buf = dst;
2383 img->len = len;
2384 return 0;
2385 case BINARY_LITERAL_DEFLATED:
2386 clear_image(img);
2387 img->len = fragment->size;
2388 img->buf = xmalloc(img->len+1);
2389 memcpy(img->buf, fragment->patch, img->len);
2390 img->buf[img->len] = '\0';
2391 return 0;
2392 }
2393 return -1;
2394 }
2396 static int apply_binary(struct image *img, struct patch *patch)
2397 {
2398 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2399 unsigned char sha1[20];
2401 /*
2402 * For safety, we require patch index line to contain
2403 * full 40-byte textual SHA1 for old and new, at least for now.
2404 */
2405 if (strlen(patch->old_sha1_prefix) != 40 ||
2406 strlen(patch->new_sha1_prefix) != 40 ||
2407 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2408 get_sha1_hex(patch->new_sha1_prefix, sha1))
2409 return error("cannot apply binary patch to '%s' "
2410 "without full index line", name);
2412 if (patch->old_name) {
2413 /*
2414 * See if the old one matches what the patch
2415 * applies to.
2416 */
2417 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2418 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2419 return error("the patch applies to '%s' (%s), "
2420 "which does not match the "
2421 "current contents.",
2422 name, sha1_to_hex(sha1));
2423 }
2424 else {
2425 /* Otherwise, the old one must be empty. */
2426 if (img->len)
2427 return error("the patch applies to an empty "
2428 "'%s' but it is not empty", name);
2429 }
2431 get_sha1_hex(patch->new_sha1_prefix, sha1);
2432 if (is_null_sha1(sha1)) {
2433 clear_image(img);
2434 return 0; /* deletion patch */
2435 }
2437 if (has_sha1_file(sha1)) {
2438 /* We already have the postimage */
2439 enum object_type type;
2440 unsigned long size;
2441 char *result;
2443 result = read_sha1_file(sha1, &type, &size);
2444 if (!result)
2445 return error("the necessary postimage %s for "
2446 "'%s' cannot be read",
2447 patch->new_sha1_prefix, name);
2448 clear_image(img);
2449 img->buf = result;
2450 img->len = size;
2451 } else {
2452 /*
2453 * We have verified buf matches the preimage;
2454 * apply the patch data to it, which is stored
2455 * in the patch->fragments->{patch,size}.
2456 */
2457 if (apply_binary_fragment(img, patch))
2458 return error("binary patch does not apply to '%s'",
2459 name);
2461 /* verify that the result matches */
2462 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2463 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2464 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2465 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2466 }
2468 return 0;
2469 }
2471 static int apply_fragments(struct image *img, struct patch *patch)
2472 {
2473 struct fragment *frag = patch->fragments;
2474 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2475 unsigned ws_rule = patch->ws_rule;
2476 unsigned inaccurate_eof = patch->inaccurate_eof;
2478 if (patch->is_binary)
2479 return apply_binary(img, patch);
2481 while (frag) {
2482 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2483 error("patch failed: %s:%ld", name, frag->oldpos);
2484 if (!apply_with_reject)
2485 return -1;
2486 frag->rejected = 1;
2487 }
2488 frag = frag->next;
2489 }
2490 return 0;
2491 }
2493 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2494 {
2495 if (!ce)
2496 return 0;
2498 if (S_ISGITLINK(ce->ce_mode)) {
2499 strbuf_grow(buf, 100);
2500 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2501 } else {
2502 enum object_type type;
2503 unsigned long sz;
2504 char *result;
2506 result = read_sha1_file(ce->sha1, &type, &sz);
2507 if (!result)
2508 return -1;
2509 /* XXX read_sha1_file NUL-terminates */
2510 strbuf_attach(buf, result, sz, sz + 1);
2511 }
2512 return 0;
2513 }
2515 static struct patch *in_fn_table(const char *name)
2516 {
2517 struct string_list_item *item;
2519 if (name == NULL)
2520 return NULL;
2522 item = string_list_lookup(name, &fn_table);
2523 if (item != NULL)
2524 return (struct patch *)item->util;
2526 return NULL;
2527 }
2529 /*
2530 * item->util in the filename table records the status of the path.
2531 * Usually it points at a patch (whose result records the contents
2532 * of it after applying it), but it could be PATH_WAS_DELETED for a
2533 * path that a previously applied patch has already removed.
2534 */
2535 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2536 #define PATH_WAS_DELETED ((struct patch *) -1)
2538 static int to_be_deleted(struct patch *patch)
2539 {
2540 return patch == PATH_TO_BE_DELETED;
2541 }
2543 static int was_deleted(struct patch *patch)
2544 {
2545 return patch == PATH_WAS_DELETED;
2546 }
2548 static void add_to_fn_table(struct patch *patch)
2549 {
2550 struct string_list_item *item;
2552 /*
2553 * Always add new_name unless patch is a deletion
2554 * This should cover the cases for normal diffs,
2555 * file creations and copies
2556 */
2557 if (patch->new_name != NULL) {
2558 item = string_list_insert(patch->new_name, &fn_table);
2559 item->util = patch;
2560 }
2562 /*
2563 * store a failure on rename/deletion cases because
2564 * later chunks shouldn't patch old names
2565 */
2566 if ((patch->new_name == NULL) || (patch->is_rename)) {
2567 item = string_list_insert(patch->old_name, &fn_table);
2568 item->util = PATH_WAS_DELETED;
2569 }
2570 }
2572 static void prepare_fn_table(struct patch *patch)
2573 {
2574 /*
2575 * store information about incoming file deletion
2576 */
2577 while (patch) {
2578 if ((patch->new_name == NULL) || (patch->is_rename)) {
2579 struct string_list_item *item;
2580 item = string_list_insert(patch->old_name, &fn_table);
2581 item->util = PATH_TO_BE_DELETED;
2582 }
2583 patch = patch->next;
2584 }
2585 }
2587 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2588 {
2589 struct strbuf buf = STRBUF_INIT;
2590 struct image image;
2591 size_t len;
2592 char *img;
2593 struct patch *tpatch;
2595 if (!(patch->is_copy || patch->is_rename) &&
2596 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2597 if (was_deleted(tpatch)) {
2598 return error("patch %s has been renamed/deleted",
2599 patch->old_name);
2600 }
2601 /* We have a patched copy in memory use that */
2602 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2603 } else if (cached) {
2604 if (read_file_or_gitlink(ce, &buf))
2605 return error("read of %s failed", patch->old_name);
2606 } else if (patch->old_name) {
2607 if (S_ISGITLINK(patch->old_mode)) {
2608 if (ce) {
2609 read_file_or_gitlink(ce, &buf);
2610 } else {
2611 /*
2612 * There is no way to apply subproject
2613 * patch without looking at the index.
2614 */
2615 patch->fragments = NULL;
2616 }
2617 } else {
2618 if (read_old_data(st, patch->old_name, &buf))
2619 return error("read of %s failed", patch->old_name);
2620 }
2621 }
2623 img = strbuf_detach(&buf, &len);
2624 prepare_image(&image, img, len, !patch->is_binary);
2626 if (apply_fragments(&image, patch) < 0)
2627 return -1; /* note with --reject this succeeds. */
2628 patch->result = image.buf;
2629 patch->resultsize = image.len;
2630 add_to_fn_table(patch);
2631 free(image.line_allocated);
2633 if (0 < patch->is_delete && patch->resultsize)
2634 return error("removal patch leaves file contents");
2636 return 0;
2637 }
2639 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2640 {
2641 struct stat nst;
2642 if (!lstat(new_name, &nst)) {
2643 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2644 return 0;
2645 /*
2646 * A leading component of new_name might be a symlink
2647 * that is going to be removed with this patch, but
2648 * still pointing at somewhere that has the path.
2649 * In such a case, path "new_name" does not exist as
2650 * far as git is concerned.
2651 */
2652 if (has_symlink_leading_path(new_name, strlen(new_name)))
2653 return 0;
2655 return error("%s: already exists in working directory", new_name);
2656 }
2657 else if ((errno != ENOENT) && (errno != ENOTDIR))
2658 return error("%s: %s", new_name, strerror(errno));
2659 return 0;
2660 }
2662 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2663 {
2664 if (S_ISGITLINK(ce->ce_mode)) {
2665 if (!S_ISDIR(st->st_mode))
2666 return -1;
2667 return 0;
2668 }
2669 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2670 }
2672 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2673 {
2674 const char *old_name = patch->old_name;
2675 struct patch *tpatch = NULL;
2676 int stat_ret = 0;
2677 unsigned st_mode = 0;
2679 /*
2680 * Make sure that we do not have local modifications from the
2681 * index when we are looking at the index. Also make sure
2682 * we have the preimage file to be patched in the work tree,
2683 * unless --cached, which tells git to apply only in the index.
2684 */
2685 if (!old_name)
2686 return 0;
2688 assert(patch->is_new <= 0);
2690 if (!(patch->is_copy || patch->is_rename) &&
2691 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2692 if (was_deleted(tpatch))
2693 return error("%s: has been deleted/renamed", old_name);
2694 st_mode = tpatch->new_mode;
2695 } else if (!cached) {
2696 stat_ret = lstat(old_name, st);
2697 if (stat_ret && errno != ENOENT)
2698 return error("%s: %s", old_name, strerror(errno));
2699 }
2701 if (to_be_deleted(tpatch))
2702 tpatch = NULL;
2704 if (check_index && !tpatch) {
2705 int pos = cache_name_pos(old_name, strlen(old_name));
2706 if (pos < 0) {
2707 if (patch->is_new < 0)
2708 goto is_new;
2709 return error("%s: does not exist in index", old_name);
2710 }
2711 *ce = active_cache[pos];
2712 if (stat_ret < 0) {
2713 struct checkout costate;
2714 /* checkout */
2715 costate.base_dir = "";
2716 costate.base_dir_len = 0;
2717 costate.force = 0;
2718 costate.quiet = 0;
2719 costate.not_new = 0;
2720 costate.refresh_cache = 1;
2721 if (checkout_entry(*ce, &costate, NULL) ||
2722 lstat(old_name, st))
2723 return -1;
2724 }
2725 if (!cached && verify_index_match(*ce, st))
2726 return error("%s: does not match index", old_name);
2727 if (cached)
2728 st_mode = (*ce)->ce_mode;
2729 } else if (stat_ret < 0) {
2730 if (patch->is_new < 0)
2731 goto is_new;
2732 return error("%s: %s", old_name, strerror(errno));
2733 }
2735 if (!cached && !tpatch)
2736 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2738 if (patch->is_new < 0)
2739 patch->is_new = 0;
2740 if (!patch->old_mode)
2741 patch->old_mode = st_mode;
2742 if ((st_mode ^ patch->old_mode) & S_IFMT)
2743 return error("%s: wrong type", old_name);
2744 if (st_mode != patch->old_mode)
2745 warning("%s has type %o, expected %o",
2746 old_name, st_mode, patch->old_mode);
2747 if (!patch->new_mode && !patch->is_delete)
2748 patch->new_mode = st_mode;
2749 return 0;
2751 is_new:
2752 patch->is_new = 1;
2753 patch->is_delete = 0;
2754 patch->old_name = NULL;
2755 return 0;
2756 }
2758 static int check_patch(struct patch *patch)
2759 {
2760 struct stat st;
2761 const char *old_name = patch->old_name;
2762 const char *new_name = patch->new_name;
2763 const char *name = old_name ? old_name : new_name;
2764 struct cache_entry *ce = NULL;
2765 struct patch *tpatch;
2766 int ok_if_exists;
2767 int status;
2769 patch->rejected = 1; /* we will drop this after we succeed */
2771 status = check_preimage(patch, &ce, &st);
2772 if (status)
2773 return status;
2774 old_name = patch->old_name;
2776 if ((tpatch = in_fn_table(new_name)) &&
2777 (was_deleted(tpatch) || to_be_deleted(tpatch)))
2778 /*
2779 * A type-change diff is always split into a patch to
2780 * delete old, immediately followed by a patch to
2781 * create new (see diff.c::run_diff()); in such a case
2782 * it is Ok that the entry to be deleted by the
2783 * previous patch is still in the working tree and in
2784 * the index.
2785 */
2786 ok_if_exists = 1;
2787 else
2788 ok_if_exists = 0;
2790 if (new_name &&
2791 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2792 if (check_index &&
2793 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2794 !ok_if_exists)
2795 return error("%s: already exists in index", new_name);
2796 if (!cached) {
2797 int err = check_to_create_blob(new_name, ok_if_exists);
2798 if (err)
2799 return err;
2800 }
2801 if (!patch->new_mode) {
2802 if (0 < patch->is_new)
2803 patch->new_mode = S_IFREG | 0644;
2804 else
2805 patch->new_mode = patch->old_mode;
2806 }
2807 }
2809 if (new_name && old_name) {
2810 int same = !strcmp(old_name, new_name);
2811 if (!patch->new_mode)
2812 patch->new_mode = patch->old_mode;
2813 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2814 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2815 patch->new_mode, new_name, patch->old_mode,
2816 same ? "" : " of ", same ? "" : old_name);
2817 }
2819 if (apply_data(patch, &st, ce) < 0)
2820 return error("%s: patch does not apply", name);
2821 patch->rejected = 0;
2822 return 0;
2823 }
2825 static int check_patch_list(struct patch *patch)
2826 {
2827 int err = 0;
2829 prepare_fn_table(patch);
2830 while (patch) {
2831 if (apply_verbosely)
2832 say_patch_name(stderr,
2833 "Checking patch ", patch, "...\n");
2834 err |= check_patch(patch);
2835 patch = patch->next;
2836 }
2837 return err;
2838 }
2840 /* This function tries to read the sha1 from the current index */
2841 static int get_current_sha1(const char *path, unsigned char *sha1)
2842 {
2843 int pos;
2845 if (read_cache() < 0)
2846 return -1;
2847 pos = cache_name_pos(path, strlen(path));
2848 if (pos < 0)
2849 return -1;
2850 hashcpy(sha1, active_cache[pos]->sha1);
2851 return 0;
2852 }
2854 /* Build an index that contains the just the files needed for a 3way merge */
2855 static void build_fake_ancestor(struct patch *list, const char *filename)
2856 {
2857 struct patch *patch;
2858 struct index_state result = { NULL };
2859 int fd;
2861 /* Once we start supporting the reverse patch, it may be
2862 * worth showing the new sha1 prefix, but until then...
2863 */
2864 for (patch = list; patch; patch = patch->next) {
2865 const unsigned char *sha1_ptr;
2866 unsigned char sha1[20];
2867 struct cache_entry *ce;
2868 const char *name;
2870 name = patch->old_name ? patch->old_name : patch->new_name;
2871 if (0 < patch->is_new)
2872 continue;
2873 else if (get_sha1(patch->old_sha1_prefix, sha1))
2874 /* git diff has no index line for mode/type changes */
2875 if (!patch->lines_added && !patch->lines_deleted) {
2876 if (get_current_sha1(patch->new_name, sha1) ||
2877 get_current_sha1(patch->old_name, sha1))
2878 die("mode change for %s, which is not "
2879 "in current HEAD", name);
2880 sha1_ptr = sha1;
2881 } else
2882 die("sha1 information is lacking or useless "
2883 "(%s).", name);
2884 else
2885 sha1_ptr = sha1;
2887 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2888 if (!ce)
2889 die("make_cache_entry failed for path '%s'", name);
2890 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2891 die ("Could not add %s to temporary index", name);
2892 }
2894 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2895 if (fd < 0 || write_index(&result, fd) || close(fd))
2896 die ("Could not write temporary index to %s", filename);
2898 discard_index(&result);
2899 }
2901 static void stat_patch_list(struct patch *patch)
2902 {
2903 int files, adds, dels;
2905 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2906 files++;
2907 adds += patch->lines_added;
2908 dels += patch->lines_deleted;
2909 show_stats(patch);
2910 }
2912 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2913 }
2915 static void numstat_patch_list(struct patch *patch)
2916 {
2917 for ( ; patch; patch = patch->next) {
2918 const char *name;
2919 name = patch->new_name ? patch->new_name : patch->old_name;
2920 if (patch->is_binary)
2921 printf("-\t-\t");
2922 else
2923 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2924 write_name_quoted(name, stdout, line_termination);
2925 }
2926 }
2928 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2929 {
2930 if (mode)
2931 printf(" %s mode %06o %s\n", newdelete, mode, name);
2932 else
2933 printf(" %s %s\n", newdelete, name);
2934 }
2936 static void show_mode_change(struct patch *p, int show_name)
2937 {
2938 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2939 if (show_name)
2940 printf(" mode change %06o => %06o %s\n",
2941 p->old_mode, p->new_mode, p->new_name);
2942 else
2943 printf(" mode change %06o => %06o\n",
2944 p->old_mode, p->new_mode);
2945 }
2946 }
2948 static void show_rename_copy(struct patch *p)
2949 {
2950 const char *renamecopy = p->is_rename ? "rename" : "copy";
2951 const char *old, *new;
2953 /* Find common prefix */
2954 old = p->old_name;
2955 new = p->new_name;
2956 while (1) {
2957 const char *slash_old, *slash_new;
2958 slash_old = strchr(old, '/');
2959 slash_new = strchr(new, '/');
2960 if (!slash_old ||
2961 !slash_new ||
2962 slash_old - old != slash_new - new ||
2963 memcmp(old, new, slash_new - new))
2964 break;
2965 old = slash_old + 1;
2966 new = slash_new + 1;
2967 }
2968 /* p->old_name thru old is the common prefix, and old and new
2969 * through the end of names are renames
2970 */
2971 if (old != p->old_name)
2972 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2973 (int)(old - p->old_name), p->old_name,
2974 old, new, p->score);
2975 else
2976 printf(" %s %s => %s (%d%%)\n", renamecopy,
2977 p->old_name, p->new_name, p->score);
2978 show_mode_change(p, 0);
2979 }
2981 static void summary_patch_list(struct patch *patch)
2982 {
2983 struct patch *p;
2985 for (p = patch; p; p = p->next) {
2986 if (p->is_new)
2987 show_file_mode_name("create", p->new_mode, p->new_name);
2988 else if (p->is_delete)
2989 show_file_mode_name("delete", p->old_mode, p->old_name);
2990 else {
2991 if (p->is_rename || p->is_copy)
2992 show_rename_copy(p);
2993 else {
2994 if (p->score) {
2995 printf(" rewrite %s (%d%%)\n",
2996 p->new_name, p->score);
2997 show_mode_change(p, 0);
2998 }
2999 else
3000 show_mode_change(p, 1);
3001 }
3002 }
3003 }
3004 }
3006 static void patch_stats(struct patch *patch)
3007 {
3008 int lines = patch->lines_added + patch->lines_deleted;
3010 if (lines > max_change)
3011 max_change = lines;
3012 if (patch->old_name) {
3013 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3014 if (!len)
3015 len = strlen(patch->old_name);
3016 if (len > max_len)
3017 max_len = len;
3018 }
3019 if (patch->new_name) {
3020 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3021 if (!len)
3022 len = strlen(patch->new_name);
3023 if (len > max_len)
3024 max_len = len;
3025 }
3026 }
3028 static void remove_file(struct patch *patch, int rmdir_empty)
3029 {
3030 if (update_index) {
3031 if (remove_file_from_cache(patch->old_name) < 0)
3032 die("unable to remove %s from index", patch->old_name);
3033 }
3034 if (!cached) {
3035 if (S_ISGITLINK(patch->old_mode)) {
3036 if (rmdir(patch->old_name))
3037 warning("unable to remove submodule %s",
3038 patch->old_name);
3039 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
3040 remove_path(patch->old_name);
3041 }
3042 }
3043 }
3045 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3046 {
3047 struct stat st;
3048 struct cache_entry *ce;
3049 int namelen = strlen(path);
3050 unsigned ce_size = cache_entry_size(namelen);
3052 if (!update_index)
3053 return;
3055 ce = xcalloc(1, ce_size);
3056 memcpy(ce->name, path, namelen);
3057 ce->ce_mode = create_ce_mode(mode);
3058 ce->ce_flags = namelen;
3059 if (S_ISGITLINK(mode)) {
3060 const char *s = buf;
3062 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3063 die("corrupt patch for subproject %s", path);
3064 } else {
3065 if (!cached) {
3066 if (lstat(path, &st) < 0)
3067 die_errno("unable to stat newly created file '%s'",
3068 path);
3069 fill_stat_cache_info(ce, &st);
3070 }
3071 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3072 die("unable to create backing store for newly created file %s", path);
3073 }
3074 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3075 die("unable to add cache entry for %s", path);
3076 }
3078 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3079 {
3080 int fd;
3081 struct strbuf nbuf = STRBUF_INIT;
3083 if (S_ISGITLINK(mode)) {
3084 struct stat st;
3085 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3086 return 0;
3087 return mkdir(path, 0777);
3088 }
3090 if (has_symlinks && S_ISLNK(mode))
3091 /* Although buf:size is counted string, it also is NUL
3092 * terminated.
3093 */
3094 return symlink(buf, path);
3096 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3097 if (fd < 0)
3098 return -1;
3100 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3101 size = nbuf.len;
3102 buf = nbuf.buf;
3103 }
3104 write_or_die(fd, buf, size);
3105 strbuf_release(&nbuf);
3107 if (close(fd) < 0)
3108 die_errno("closing file '%s'", path);
3109 return 0;
3110 }
3112 /*
3113 * We optimistically assume that the directories exist,
3114 * which is true 99% of the time anyway. If they don't,
3115 * we create them and try again.
3116 */
3117 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3118 {
3119 if (cached)
3120 return;
3121 if (!try_create_file(path, mode, buf, size))
3122 return;
3124 if (errno == ENOENT) {
3125 if (safe_create_leading_directories(path))
3126 return;
3127 if (!try_create_file(path, mode, buf, size))
3128 return;
3129 }
3131 if (errno == EEXIST || errno == EACCES) {
3132 /* We may be trying to create a file where a directory
3133 * used to be.
3134 */
3135 struct stat st;
3136 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3137 errno = EEXIST;
3138 }
3140 if (errno == EEXIST) {
3141 unsigned int nr = getpid();
3143 for (;;) {
3144 char newpath[PATH_MAX];
3145 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3146 if (!try_create_file(newpath, mode, buf, size)) {
3147 if (!rename(newpath, path))
3148 return;
3149 unlink_or_warn(newpath);
3150 break;
3151 }
3152 if (errno != EEXIST)
3153 break;
3154 ++nr;
3155 }
3156 }
3157 die_errno("unable to write file '%s' mode %o", path, mode);
3158 }
3160 static void create_file(struct patch *patch)
3161 {
3162 char *path = patch->new_name;
3163 unsigned mode = patch->new_mode;
3164 unsigned long size = patch->resultsize;
3165 char *buf = patch->result;
3167 if (!mode)
3168 mode = S_IFREG | 0644;
3169 create_one_file(path, mode, buf, size);
3170 add_index_file(path, mode, buf, size);
3171 }
3173 /* phase zero is to remove, phase one is to create */
3174 static void write_out_one_result(struct patch *patch, int phase)
3175 {
3176 if (patch->is_delete > 0) {
3177 if (phase == 0)
3178 remove_file(patch, 1);
3179 return;
3180 }
3181 if (patch->is_new > 0 || patch->is_copy) {
3182 if (phase == 1)
3183 create_file(patch);
3184 return;
3185 }
3186 /*
3187 * Rename or modification boils down to the same
3188 * thing: remove the old, write the new
3189 */
3190 if (phase == 0)
3191 remove_file(patch, patch->is_rename);
3192 if (phase == 1)
3193 create_file(patch);
3194 }
3196 static int write_out_one_reject(struct patch *patch)
3197 {
3198 FILE *rej;
3199 char namebuf[PATH_MAX];
3200 struct fragment *frag;
3201 int cnt = 0;
3203 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3204 if (!frag->rejected)
3205 continue;
3206 cnt++;
3207 }
3209 if (!cnt) {
3210 if (apply_verbosely)
3211 say_patch_name(stderr,
3212 "Applied patch ", patch, " cleanly.\n");
3213 return 0;
3214 }
3216 /* This should not happen, because a removal patch that leaves
3217 * contents are marked "rejected" at the patch level.
3218 */
3219 if (!patch->new_name)
3220 die("internal error");
3222 /* Say this even without --verbose */
3223 say_patch_name(stderr, "Applying patch ", patch, " with");
3224 fprintf(stderr, " %d rejects...\n", cnt);
3226 cnt = strlen(patch->new_name);
3227 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3228 cnt = ARRAY_SIZE(namebuf) - 5;
3229 warning("truncating .rej filename to %.*s.rej",
3230 cnt - 1, patch->new_name);
3231 }
3232 memcpy(namebuf, patch->new_name, cnt);
3233 memcpy(namebuf + cnt, ".rej", 5);
3235 rej = fopen(namebuf, "w");
3236 if (!rej)
3237 return error("cannot open %s: %s", namebuf, strerror(errno));
3239 /* Normal git tools never deal with .rej, so do not pretend
3240 * this is a git patch by saying --git nor give extended
3241 * headers. While at it, maybe please "kompare" that wants
3242 * the trailing TAB and some garbage at the end of line ;-).
3243 */
3244 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3245 patch->new_name, patch->new_name);
3246 for (cnt = 1, frag = patch->fragments;
3247 frag;
3248 cnt++, frag = frag->next) {
3249 if (!frag->rejected) {
3250 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3251 continue;
3252 }
3253 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3254 fprintf(rej, "%.*s", frag->size, frag->patch);
3255 if (frag->patch[frag->size-1] != '\n')
3256 fputc('\n', rej);
3257 }
3258 fclose(rej);
3259 return -1;
3260 }
3262 static int write_out_results(struct patch *list, int skipped_patch)
3263 {
3264 int phase;
3265 int errs = 0;
3266 struct patch *l;
3268 if (!list && !skipped_patch)
3269 return error("No changes");
3271 for (phase = 0; phase < 2; phase++) {
3272 l = list;
3273 while (l) {
3274 if (l->rejected)
3275 errs = 1;
3276 else {
3277 write_out_one_result(l, phase);
3278 if (phase == 1 && write_out_one_reject(l))
3279 errs = 1;
3280 }
3281 l = l->next;
3282 }
3283 }
3284 return errs;
3285 }
3287 static struct lock_file lock_file;
3289 static struct string_list limit_by_name;
3290 static int has_include;
3291 static void add_name_limit(const char *name, int exclude)
3292 {
3293 struct string_list_item *it;
3295 it = string_list_append(name, &limit_by_name);
3296 it->util = exclude ? NULL : (void *) 1;
3297 }
3299 static int use_patch(struct patch *p)
3300 {
3301 const char *pathname = p->new_name ? p->new_name : p->old_name;
3302 int i;
3304 /* Paths outside are not touched regardless of "--include" */
3305 if (0 < prefix_length) {
3306 int pathlen = strlen(pathname);
3307 if (pathlen <= prefix_length ||
3308 memcmp(prefix, pathname, prefix_length))
3309 return 0;
3310 }
3312 /* See if it matches any of exclude/include rule */
3313 for (i = 0; i < limit_by_name.nr; i++) {
3314 struct string_list_item *it = &limit_by_name.items[i];
3315 if (!fnmatch(it->string, pathname, 0))
3316 return (it->util != NULL);
3317 }
3319 /*
3320 * If we had any include, a path that does not match any rule is
3321 * not used. Otherwise, we saw bunch of exclude rules (or none)
3322 * and such a path is used.
3323 */
3324 return !has_include;
3325 }
3328 static void prefix_one(char **name)
3329 {
3330 char *old_name = *name;
3331 if (!old_name)
3332 return;
3333 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3334 free(old_name);
3335 }
3337 static void prefix_patches(struct patch *p)
3338 {
3339 if (!prefix || p->is_toplevel_relative)
3340 return;
3341 for ( ; p; p = p->next) {
3342 if (p->new_name == p->old_name) {
3343 char *prefixed = p->new_name;
3344 prefix_one(&prefixed);
3345 p->new_name = p->old_name = prefixed;
3346 }
3347 else {
3348 prefix_one(&p->new_name);
3349 prefix_one(&p->old_name);
3350 }
3351 }
3352 }
3354 #define INACCURATE_EOF (1<<0)
3355 #define RECOUNT (1<<1)
3357 static int apply_patch(int fd, const char *filename, int options)
3358 {
3359 size_t offset;
3360 struct strbuf buf = STRBUF_INIT;
3361 struct patch *list = NULL, **listp = &list;
3362 int skipped_patch = 0;
3364 /* FIXME - memory leak when using multiple patch files as inputs */
3365 memset(&fn_table, 0, sizeof(struct string_list));
3366 patch_input_file = filename;
3367 read_patch_file(&buf, fd);
3368 offset = 0;
3369 while (offset < buf.len) {
3370 struct patch *patch;
3371 int nr;
3373 patch = xcalloc(1, sizeof(*patch));
3374 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3375 patch->recount = !!(options & RECOUNT);
3376 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3377 if (nr < 0)
3378 break;
3379 if (apply_in_reverse)
3380 reverse_patches(patch);
3381 if (prefix)
3382 prefix_patches(patch);
3383 if (use_patch(patch)) {
3384 patch_stats(patch);
3385 *listp = patch;
3386 listp = &patch->next;
3387 }
3388 else {
3389 /* perhaps free it a bit better? */
3390 free(patch);
3391 skipped_patch++;
3392 }
3393 offset += nr;
3394 }
3396 if (whitespace_error && (ws_error_action == die_on_ws_error))
3397 apply = 0;
3399 update_index = check_index && apply;
3400 if (update_index && newfd < 0)
3401 newfd = hold_locked_index(&lock_file, 1);
3403 if (check_index) {
3404 if (read_cache() < 0)
3405 die("unable to read index file");
3406 }
3408 if ((check || apply) &&
3409 check_patch_list(list) < 0 &&
3410 !apply_with_reject)
3411 exit(1);
3413 if (apply && write_out_results(list, skipped_patch))
3414 exit(1);
3416 if (fake_ancestor)
3417 build_fake_ancestor(list, fake_ancestor);
3419 if (diffstat)
3420 stat_patch_list(list);
3422 if (numstat)
3423 numstat_patch_list(list);
3425 if (summary)
3426 summary_patch_list(list);
3428 strbuf_release(&buf);
3429 return 0;
3430 }
3432 static int git_apply_config(const char *var, const char *value, void *cb)
3433 {
3434 if (!strcmp(var, "apply.whitespace"))
3435 return git_config_string(&apply_default_whitespace, var, value);
3436 else if (!strcmp(var, "apply.ignorewhitespace"))
3437 return git_config_string(&apply_default_ignorewhitespace, var, value);
3438 return git_default_config(var, value, cb);
3439 }
3441 static int option_parse_exclude(const struct option *opt,
3442 const char *arg, int unset)
3443 {
3444 add_name_limit(arg, 1);
3445 return 0;
3446 }
3448 static int option_parse_include(const struct option *opt,
3449 const char *arg, int unset)
3450 {
3451 add_name_limit(arg, 0);
3452 has_include = 1;
3453 return 0;
3454 }
3456 static int option_parse_p(const struct option *opt,
3457 const char *arg, int unset)
3458 {
3459 p_value = atoi(arg);
3460 p_value_known = 1;
3461 return 0;
3462 }
3464 static int option_parse_z(const struct option *opt,
3465 const char *arg, int unset)
3466 {
3467 if (unset)
3468 line_termination = '\n';
3469 else
3470 line_termination = 0;
3471 return 0;
3472 }
3474 static int option_parse_space_change(const struct option *opt,
3475 const char *arg, int unset)
3476 {
3477 if (unset)
3478 ws_ignore_action = ignore_ws_none;
3479 else
3480 ws_ignore_action = ignore_ws_change;
3481 return 0;
3482 }
3484 static int option_parse_whitespace(const struct option *opt,
3485 const char *arg, int unset)
3486 {
3487 const char **whitespace_option = opt->value;
3489 *whitespace_option = arg;
3490 parse_whitespace_option(arg);
3491 return 0;
3492 }
3494 static int option_parse_directory(const struct option *opt,
3495 const char *arg, int unset)
3496 {
3497 root_len = strlen(arg);
3498 if (root_len && arg[root_len - 1] != '/') {
3499 char *new_root;
3500 root = new_root = xmalloc(root_len + 2);
3501 strcpy(new_root, arg);
3502 strcpy(new_root + root_len++, "/");
3503 } else
3504 root = arg;
3505 return 0;
3506 }
3508 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3509 {
3510 int i;
3511 int errs = 0;
3512 int is_not_gitdir;
3513 int binary;
3514 int force_apply = 0;
3516 const char *whitespace_option = NULL;
3518 struct option builtin_apply_options[] = {
3519 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3520 "don't apply changes matching the given path",
3521 0, option_parse_exclude },
3522 { OPTION_CALLBACK, 0, "include", NULL, "path",
3523 "apply changes matching the given path",
3524 0, option_parse_include },
3525 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3526 "remove <num> leading slashes from traditional diff paths",
3527 0, option_parse_p },
3528 OPT_BOOLEAN(0, "no-add", &no_add,
3529 "ignore additions made by the patch"),
3530 OPT_BOOLEAN(0, "stat", &diffstat,
3531 "instead of applying the patch, output diffstat for the input"),
3532 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3533 NULL, "old option, now no-op",
3534 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3535 { OPTION_BOOLEAN, 0, "binary", &binary,
3536 NULL, "old option, now no-op",
3537 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3538 OPT_BOOLEAN(0, "numstat", &numstat,
3539 "shows number of added and deleted lines in decimal notation"),
3540 OPT_BOOLEAN(0, "summary", &summary,
3541 "instead of applying the patch, output a summary for the input"),
3542 OPT_BOOLEAN(0, "check", &check,
3543 "instead of applying the patch, see if the patch is applicable"),
3544 OPT_BOOLEAN(0, "index", &check_index,
3545 "make sure the patch is applicable to the current index"),
3546 OPT_BOOLEAN(0, "cached", &cached,
3547 "apply a patch without touching the working tree"),
3548 OPT_BOOLEAN(0, "apply", &force_apply,
3549 "also apply the patch (use with --stat/--summary/--check)"),
3550 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3551 "build a temporary index based on embedded index information"),
3552 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3553 "paths are separated with NUL character",
3554 PARSE_OPT_NOARG, option_parse_z },
3555 OPT_INTEGER('C', NULL, &p_context,
3556 "ensure at least <n> lines of context match"),
3557 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3558 "detect new or modified lines that have whitespace errors",
3559 0, option_parse_whitespace },
3560 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3561 "ignore changes in whitespace when finding context",
3562 PARSE_OPT_NOARG, option_parse_space_change },
3563 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3564 "ignore changes in whitespace when finding context",
3565 PARSE_OPT_NOARG, option_parse_space_change },
3566 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3567 "apply the patch in reverse"),
3568 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3569 "don't expect at least one line of context"),
3570 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3571 "leave the rejected hunks in corresponding *.rej files"),
3572 OPT__VERBOSE(&apply_verbosely),
3573 OPT_BIT(0, "inaccurate-eof", &options,
3574 "tolerate incorrectly detected missing new-line at the end of file",
3575 INACCURATE_EOF),
3576 OPT_BIT(0, "recount", &options,
3577 "do not trust the line counts in the hunk headers",
3578 RECOUNT),
3579 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3580 "prepend <root> to all filenames",
3581 0, option_parse_directory },
3582 OPT_END()
3583 };
3585 prefix = setup_git_directory_gently(&is_not_gitdir);
3586 prefix_length = prefix ? strlen(prefix) : 0;
3587 git_config(git_apply_config, NULL);
3588 if (apply_default_whitespace)
3589 parse_whitespace_option(apply_default_whitespace);
3590 if (apply_default_ignorewhitespace)
3591 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3593 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3594 apply_usage, 0);
3596 if (apply_with_reject)
3597 apply = apply_verbosely = 1;
3598 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3599 apply = 0;
3600 if (check_index && is_not_gitdir)
3601 die("--index outside a repository");
3602 if (cached) {
3603 if (is_not_gitdir)
3604 die("--cached outside a repository");
3605 check_index = 1;
3606 }
3607 for (i = 0; i < argc; i++) {
3608 const char *arg = argv[i];
3609 int fd;
3611 if (!strcmp(arg, "-")) {
3612 errs |= apply_patch(0, "<stdin>", options);
3613 read_stdin = 0;
3614 continue;
3615 } else if (0 < prefix_length)
3616 arg = prefix_filename(prefix, prefix_length, arg);
3618 fd = open(arg, O_RDONLY);
3619 if (fd < 0)
3620 die_errno("can't open patch '%s'", arg);
3621 read_stdin = 0;
3622 set_default_whitespace_mode(whitespace_option);
3623 errs |= apply_patch(fd, arg, options);
3624 close(fd);
3625 }
3626 set_default_whitespace_mode(whitespace_option);
3627 if (read_stdin)
3628 errs |= apply_patch(0, "<stdin>", options);
3629 if (whitespace_error) {
3630 if (squelch_whitespace_errors &&
3631 squelch_whitespace_errors < whitespace_error) {
3632 int squelched =
3633 whitespace_error - squelch_whitespace_errors;
3634 warning("squelched %d "
3635 "whitespace error%s",
3636 squelched,
3637 squelched == 1 ? "" : "s");
3638 }
3639 if (ws_error_action == die_on_ws_error)
3640 die("%d line%s add%s whitespace errors.",
3641 whitespace_error,
3642 whitespace_error == 1 ? "" : "s",
3643 whitespace_error == 1 ? "s" : "");
3644 if (applied_after_fixing_ws && apply)
3645 warning("%d line%s applied after"
3646 " fixing whitespace errors.",
3647 applied_after_fixing_ws,
3648 applied_after_fixing_ws == 1 ? "" : "s");
3649 else if (whitespace_error)
3650 warning("%d line%s add%s whitespace errors.",
3651 whitespace_error,
3652 whitespace_error == 1 ? "" : "s",
3653 whitespace_error == 1 ? "s" : "");
3654 }
3656 if (update_index) {
3657 if (write_cache(newfd, active_cache, active_nr) ||
3658 commit_locked_index(&lock_file))
3659 die("Unable to write new index file");
3660 }
3662 return !!errs;
3663 }