82415d93c95095006beeea73f26955cf121cb495
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;
64 static const char *patch_input_file;
65 static const char *root;
66 static int root_len;
67 static int read_stdin = 1;
68 static int options;
70 static void parse_whitespace_option(const char *option)
71 {
72 if (!option) {
73 ws_error_action = warn_on_ws_error;
74 return;
75 }
76 if (!strcmp(option, "warn")) {
77 ws_error_action = warn_on_ws_error;
78 return;
79 }
80 if (!strcmp(option, "nowarn")) {
81 ws_error_action = nowarn_ws_error;
82 return;
83 }
84 if (!strcmp(option, "error")) {
85 ws_error_action = die_on_ws_error;
86 return;
87 }
88 if (!strcmp(option, "error-all")) {
89 ws_error_action = die_on_ws_error;
90 squelch_whitespace_errors = 0;
91 return;
92 }
93 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
94 ws_error_action = correct_ws_error;
95 return;
96 }
97 die("unrecognized whitespace option '%s'", option);
98 }
100 static void set_default_whitespace_mode(const char *whitespace_option)
101 {
102 if (!whitespace_option && !apply_default_whitespace)
103 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
104 }
106 /*
107 * For "diff-stat" like behaviour, we keep track of the biggest change
108 * we've seen, and the longest filename. That allows us to do simple
109 * scaling.
110 */
111 static int max_change, max_len;
113 /*
114 * Various "current state", notably line numbers and what
115 * file (and how) we're patching right now.. The "is_xxxx"
116 * things are flags, where -1 means "don't know yet".
117 */
118 static int linenr = 1;
120 /*
121 * This represents one "hunk" from a patch, starting with
122 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
123 * patch text is pointed at by patch, and its byte length
124 * is stored in size. leading and trailing are the number
125 * of context lines.
126 */
127 struct fragment {
128 unsigned long leading, trailing;
129 unsigned long oldpos, oldlines;
130 unsigned long newpos, newlines;
131 const char *patch;
132 int size;
133 int rejected;
134 int linenr;
135 struct fragment *next;
136 };
138 /*
139 * When dealing with a binary patch, we reuse "leading" field
140 * to store the type of the binary hunk, either deflated "delta"
141 * or deflated "literal".
142 */
143 #define binary_patch_method leading
144 #define BINARY_DELTA_DEFLATED 1
145 #define BINARY_LITERAL_DEFLATED 2
147 /*
148 * This represents a "patch" to a file, both metainfo changes
149 * such as creation/deletion, filemode and content changes represented
150 * as a series of fragments.
151 */
152 struct patch {
153 char *new_name, *old_name, *def_name;
154 unsigned int old_mode, new_mode;
155 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
156 int rejected;
157 unsigned ws_rule;
158 unsigned long deflate_origlen;
159 int lines_added, lines_deleted;
160 int score;
161 unsigned int is_toplevel_relative:1;
162 unsigned int inaccurate_eof:1;
163 unsigned int is_binary:1;
164 unsigned int is_copy:1;
165 unsigned int is_rename:1;
166 unsigned int recount:1;
167 struct fragment *fragments;
168 char *result;
169 size_t resultsize;
170 char old_sha1_prefix[41];
171 char new_sha1_prefix[41];
172 struct patch *next;
173 };
175 /*
176 * A line in a file, len-bytes long (includes the terminating LF,
177 * except for an incomplete line at the end if the file ends with
178 * one), and its contents hashes to 'hash'.
179 */
180 struct line {
181 size_t len;
182 unsigned hash : 24;
183 unsigned flag : 8;
184 #define LINE_COMMON 1
185 };
187 /*
188 * This represents a "file", which is an array of "lines".
189 */
190 struct image {
191 char *buf;
192 size_t len;
193 size_t nr;
194 size_t alloc;
195 struct line *line_allocated;
196 struct line *line;
197 };
199 /*
200 * Records filenames that have been touched, in order to handle
201 * the case where more than one patches touch the same file.
202 */
204 static struct string_list fn_table;
206 static uint32_t hash_line(const char *cp, size_t len)
207 {
208 size_t i;
209 uint32_t h;
210 for (i = 0, h = 0; i < len; i++) {
211 if (!isspace(cp[i])) {
212 h = h * 3 + (cp[i] & 0xff);
213 }
214 }
215 return h;
216 }
218 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
219 {
220 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
221 img->line_allocated[img->nr].len = len;
222 img->line_allocated[img->nr].hash = hash_line(bol, len);
223 img->line_allocated[img->nr].flag = flag;
224 img->nr++;
225 }
227 static void prepare_image(struct image *image, char *buf, size_t len,
228 int prepare_linetable)
229 {
230 const char *cp, *ep;
232 memset(image, 0, sizeof(*image));
233 image->buf = buf;
234 image->len = len;
236 if (!prepare_linetable)
237 return;
239 ep = image->buf + image->len;
240 cp = image->buf;
241 while (cp < ep) {
242 const char *next;
243 for (next = cp; next < ep && *next != '\n'; next++)
244 ;
245 if (next < ep)
246 next++;
247 add_line_info(image, cp, next - cp, 0);
248 cp = next;
249 }
250 image->line = image->line_allocated;
251 }
253 static void clear_image(struct image *image)
254 {
255 free(image->buf);
256 image->buf = NULL;
257 image->len = 0;
258 }
260 static void say_patch_name(FILE *output, const char *pre,
261 struct patch *patch, const char *post)
262 {
263 fputs(pre, output);
264 if (patch->old_name && patch->new_name &&
265 strcmp(patch->old_name, patch->new_name)) {
266 quote_c_style(patch->old_name, NULL, output, 0);
267 fputs(" => ", output);
268 quote_c_style(patch->new_name, NULL, output, 0);
269 } else {
270 const char *n = patch->new_name;
271 if (!n)
272 n = patch->old_name;
273 quote_c_style(n, NULL, output, 0);
274 }
275 fputs(post, output);
276 }
278 #define CHUNKSIZE (8192)
279 #define SLOP (16)
281 static void read_patch_file(struct strbuf *sb, int fd)
282 {
283 if (strbuf_read(sb, fd, 0) < 0)
284 die_errno("git apply: failed to read");
286 /*
287 * Make sure that we have some slop in the buffer
288 * so that we can do speculative "memcmp" etc, and
289 * see to it that it is NUL-filled.
290 */
291 strbuf_grow(sb, SLOP);
292 memset(sb->buf + sb->len, 0, SLOP);
293 }
295 static unsigned long linelen(const char *buffer, unsigned long size)
296 {
297 unsigned long len = 0;
298 while (size--) {
299 len++;
300 if (*buffer++ == '\n')
301 break;
302 }
303 return len;
304 }
306 static int is_dev_null(const char *str)
307 {
308 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
309 }
311 #define TERM_SPACE 1
312 #define TERM_TAB 2
314 static int name_terminate(const char *name, int namelen, int c, int terminate)
315 {
316 if (c == ' ' && !(terminate & TERM_SPACE))
317 return 0;
318 if (c == '\t' && !(terminate & TERM_TAB))
319 return 0;
321 return 1;
322 }
324 /* remove double slashes to make --index work with such filenames */
325 static char *squash_slash(char *name)
326 {
327 int i = 0, j = 0;
329 while (name[i]) {
330 if ((name[j++] = name[i++]) == '/')
331 while (name[i] == '/')
332 i++;
333 }
334 name[j] = '\0';
335 return name;
336 }
338 static char *find_name(const char *line, char *def, int p_value, int terminate)
339 {
340 int len;
341 const char *start = line;
343 if (*line == '"') {
344 struct strbuf name = STRBUF_INIT;
346 /*
347 * Proposed "new-style" GNU patch/diff format; see
348 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
349 */
350 if (!unquote_c_style(&name, line, NULL)) {
351 char *cp;
353 for (cp = name.buf; p_value; p_value--) {
354 cp = strchr(cp, '/');
355 if (!cp)
356 break;
357 cp++;
358 }
359 if (cp) {
360 /* name can later be freed, so we need
361 * to memmove, not just return cp
362 */
363 strbuf_remove(&name, 0, cp - name.buf);
364 free(def);
365 if (root)
366 strbuf_insert(&name, 0, root, root_len);
367 return squash_slash(strbuf_detach(&name, NULL));
368 }
369 }
370 strbuf_release(&name);
371 }
373 for (;;) {
374 char c = *line;
376 if (isspace(c)) {
377 if (c == '\n')
378 break;
379 if (name_terminate(start, line-start, c, terminate))
380 break;
381 }
382 line++;
383 if (c == '/' && !--p_value)
384 start = line;
385 }
386 if (!start)
387 return squash_slash(def);
388 len = line - start;
389 if (!len)
390 return squash_slash(def);
392 /*
393 * Generally we prefer the shorter name, especially
394 * if the other one is just a variation of that with
395 * something else tacked on to the end (ie "file.orig"
396 * or "file~").
397 */
398 if (def) {
399 int deflen = strlen(def);
400 if (deflen < len && !strncmp(start, def, deflen))
401 return squash_slash(def);
402 free(def);
403 }
405 if (root) {
406 char *ret = xmalloc(root_len + len + 1);
407 strcpy(ret, root);
408 memcpy(ret + root_len, start, len);
409 ret[root_len + len] = '\0';
410 return squash_slash(ret);
411 }
413 return squash_slash(xmemdupz(start, len));
414 }
416 static int count_slashes(const char *cp)
417 {
418 int cnt = 0;
419 char ch;
421 while ((ch = *cp++))
422 if (ch == '/')
423 cnt++;
424 return cnt;
425 }
427 /*
428 * Given the string after "--- " or "+++ ", guess the appropriate
429 * p_value for the given patch.
430 */
431 static int guess_p_value(const char *nameline)
432 {
433 char *name, *cp;
434 int val = -1;
436 if (is_dev_null(nameline))
437 return -1;
438 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
439 if (!name)
440 return -1;
441 cp = strchr(name, '/');
442 if (!cp)
443 val = 0;
444 else if (prefix) {
445 /*
446 * Does it begin with "a/$our-prefix" and such? Then this is
447 * very likely to apply to our directory.
448 */
449 if (!strncmp(name, prefix, prefix_length))
450 val = count_slashes(prefix);
451 else {
452 cp++;
453 if (!strncmp(cp, prefix, prefix_length))
454 val = count_slashes(prefix) + 1;
455 }
456 }
457 free(name);
458 return val;
459 }
461 /*
462 * Does the ---/+++ line has the POSIX timestamp after the last HT?
463 * GNU diff puts epoch there to signal a creation/deletion event. Is
464 * this such a timestamp?
465 */
466 static int has_epoch_timestamp(const char *nameline)
467 {
468 /*
469 * We are only interested in epoch timestamp; any non-zero
470 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
471 * For the same reason, the date must be either 1969-12-31 or
472 * 1970-01-01, and the seconds part must be "00".
473 */
474 const char stamp_regexp[] =
475 "^(1969-12-31|1970-01-01)"
476 " "
477 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
478 " "
479 "([-+][0-2][0-9][0-5][0-9])\n";
480 const char *timestamp = NULL, *cp;
481 static regex_t *stamp;
482 regmatch_t m[10];
483 int zoneoffset;
484 int hourminute;
485 int status;
487 for (cp = nameline; *cp != '\n'; cp++) {
488 if (*cp == '\t')
489 timestamp = cp + 1;
490 }
491 if (!timestamp)
492 return 0;
493 if (!stamp) {
494 stamp = xmalloc(sizeof(*stamp));
495 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
496 warning("Cannot prepare timestamp regexp %s",
497 stamp_regexp);
498 return 0;
499 }
500 }
502 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
503 if (status) {
504 if (status != REG_NOMATCH)
505 warning("regexec returned %d for input: %s",
506 status, timestamp);
507 return 0;
508 }
510 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
511 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
512 if (timestamp[m[3].rm_so] == '-')
513 zoneoffset = -zoneoffset;
515 /*
516 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
517 * (west of GMT) or 1970-01-01 (east of GMT)
518 */
519 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
520 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
521 return 0;
523 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
524 strtol(timestamp + 14, NULL, 10) -
525 zoneoffset);
527 return ((zoneoffset < 0 && hourminute == 1440) ||
528 (0 <= zoneoffset && !hourminute));
529 }
531 /*
532 * Get the name etc info from the ---/+++ lines of a traditional patch header
533 *
534 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
535 * files, we can happily check the index for a match, but for creating a
536 * new file we should try to match whatever "patch" does. I have no idea.
537 */
538 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
539 {
540 char *name;
542 first += 4; /* skip "--- " */
543 second += 4; /* skip "+++ " */
544 if (!p_value_known) {
545 int p, q;
546 p = guess_p_value(first);
547 q = guess_p_value(second);
548 if (p < 0) p = q;
549 if (0 <= p && p == q) {
550 p_value = p;
551 p_value_known = 1;
552 }
553 }
554 if (is_dev_null(first)) {
555 patch->is_new = 1;
556 patch->is_delete = 0;
557 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
558 patch->new_name = name;
559 } else if (is_dev_null(second)) {
560 patch->is_new = 0;
561 patch->is_delete = 1;
562 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
563 patch->old_name = name;
564 } else {
565 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
566 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
567 if (has_epoch_timestamp(first)) {
568 patch->is_new = 1;
569 patch->is_delete = 0;
570 patch->new_name = name;
571 } else if (has_epoch_timestamp(second)) {
572 patch->is_new = 0;
573 patch->is_delete = 1;
574 patch->old_name = name;
575 } else {
576 patch->old_name = patch->new_name = name;
577 }
578 }
579 if (!name)
580 die("unable to find filename in patch at line %d", linenr);
581 }
583 static int gitdiff_hdrend(const char *line, struct patch *patch)
584 {
585 return -1;
586 }
588 /*
589 * We're anal about diff header consistency, to make
590 * sure that we don't end up having strange ambiguous
591 * patches floating around.
592 *
593 * As a result, gitdiff_{old|new}name() will check
594 * their names against any previous information, just
595 * to make sure..
596 */
597 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
598 {
599 if (!orig_name && !isnull)
600 return find_name(line, NULL, p_value, TERM_TAB);
602 if (orig_name) {
603 int len;
604 const char *name;
605 char *another;
606 name = orig_name;
607 len = strlen(name);
608 if (isnull)
609 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
610 another = find_name(line, NULL, p_value, TERM_TAB);
611 if (!another || memcmp(another, name, len))
612 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
613 free(another);
614 return orig_name;
615 }
616 else {
617 /* expect "/dev/null" */
618 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
619 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
620 return NULL;
621 }
622 }
624 static int gitdiff_oldname(const char *line, struct patch *patch)
625 {
626 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
627 return 0;
628 }
630 static int gitdiff_newname(const char *line, struct patch *patch)
631 {
632 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
633 return 0;
634 }
636 static int gitdiff_oldmode(const char *line, struct patch *patch)
637 {
638 patch->old_mode = strtoul(line, NULL, 8);
639 return 0;
640 }
642 static int gitdiff_newmode(const char *line, struct patch *patch)
643 {
644 patch->new_mode = strtoul(line, NULL, 8);
645 return 0;
646 }
648 static int gitdiff_delete(const char *line, struct patch *patch)
649 {
650 patch->is_delete = 1;
651 patch->old_name = patch->def_name;
652 return gitdiff_oldmode(line, patch);
653 }
655 static int gitdiff_newfile(const char *line, struct patch *patch)
656 {
657 patch->is_new = 1;
658 patch->new_name = patch->def_name;
659 return gitdiff_newmode(line, patch);
660 }
662 static int gitdiff_copysrc(const char *line, struct patch *patch)
663 {
664 patch->is_copy = 1;
665 patch->old_name = find_name(line, NULL, 0, 0);
666 return 0;
667 }
669 static int gitdiff_copydst(const char *line, struct patch *patch)
670 {
671 patch->is_copy = 1;
672 patch->new_name = find_name(line, NULL, 0, 0);
673 return 0;
674 }
676 static int gitdiff_renamesrc(const char *line, struct patch *patch)
677 {
678 patch->is_rename = 1;
679 patch->old_name = find_name(line, NULL, 0, 0);
680 return 0;
681 }
683 static int gitdiff_renamedst(const char *line, struct patch *patch)
684 {
685 patch->is_rename = 1;
686 patch->new_name = find_name(line, NULL, 0, 0);
687 return 0;
688 }
690 static int gitdiff_similarity(const char *line, struct patch *patch)
691 {
692 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
693 patch->score = 0;
694 return 0;
695 }
697 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
698 {
699 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
700 patch->score = 0;
701 return 0;
702 }
704 static int gitdiff_index(const char *line, struct patch *patch)
705 {
706 /*
707 * index line is N hexadecimal, "..", N hexadecimal,
708 * and optional space with octal mode.
709 */
710 const char *ptr, *eol;
711 int len;
713 ptr = strchr(line, '.');
714 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
715 return 0;
716 len = ptr - line;
717 memcpy(patch->old_sha1_prefix, line, len);
718 patch->old_sha1_prefix[len] = 0;
720 line = ptr + 2;
721 ptr = strchr(line, ' ');
722 eol = strchr(line, '\n');
724 if (!ptr || eol < ptr)
725 ptr = eol;
726 len = ptr - line;
728 if (40 < len)
729 return 0;
730 memcpy(patch->new_sha1_prefix, line, len);
731 patch->new_sha1_prefix[len] = 0;
732 if (*ptr == ' ')
733 patch->old_mode = strtoul(ptr+1, NULL, 8);
734 return 0;
735 }
737 /*
738 * This is normal for a diff that doesn't change anything: we'll fall through
739 * into the next diff. Tell the parser to break out.
740 */
741 static int gitdiff_unrecognized(const char *line, struct patch *patch)
742 {
743 return -1;
744 }
746 static const char *stop_at_slash(const char *line, int llen)
747 {
748 int i;
750 for (i = 0; i < llen; i++) {
751 int ch = line[i];
752 if (ch == '/')
753 return line + i;
754 }
755 return NULL;
756 }
758 /*
759 * This is to extract the same name that appears on "diff --git"
760 * line. We do not find and return anything if it is a rename
761 * patch, and it is OK because we will find the name elsewhere.
762 * We need to reliably find name only when it is mode-change only,
763 * creation or deletion of an empty file. In any of these cases,
764 * both sides are the same name under a/ and b/ respectively.
765 */
766 static char *git_header_name(char *line, int llen)
767 {
768 const char *name;
769 const char *second = NULL;
770 size_t len;
772 line += strlen("diff --git ");
773 llen -= strlen("diff --git ");
775 if (*line == '"') {
776 const char *cp;
777 struct strbuf first = STRBUF_INIT;
778 struct strbuf sp = STRBUF_INIT;
780 if (unquote_c_style(&first, line, &second))
781 goto free_and_fail1;
783 /* advance to the first slash */
784 cp = stop_at_slash(first.buf, first.len);
785 /* we do not accept absolute paths */
786 if (!cp || cp == first.buf)
787 goto free_and_fail1;
788 strbuf_remove(&first, 0, cp + 1 - first.buf);
790 /*
791 * second points at one past closing dq of name.
792 * find the second name.
793 */
794 while ((second < line + llen) && isspace(*second))
795 second++;
797 if (line + llen <= second)
798 goto free_and_fail1;
799 if (*second == '"') {
800 if (unquote_c_style(&sp, second, NULL))
801 goto free_and_fail1;
802 cp = stop_at_slash(sp.buf, sp.len);
803 if (!cp || cp == sp.buf)
804 goto free_and_fail1;
805 /* They must match, otherwise ignore */
806 if (strcmp(cp + 1, first.buf))
807 goto free_and_fail1;
808 strbuf_release(&sp);
809 return strbuf_detach(&first, NULL);
810 }
812 /* unquoted second */
813 cp = stop_at_slash(second, line + llen - second);
814 if (!cp || cp == second)
815 goto free_and_fail1;
816 cp++;
817 if (line + llen - cp != first.len + 1 ||
818 memcmp(first.buf, cp, first.len))
819 goto free_and_fail1;
820 return strbuf_detach(&first, NULL);
822 free_and_fail1:
823 strbuf_release(&first);
824 strbuf_release(&sp);
825 return NULL;
826 }
828 /* unquoted first name */
829 name = stop_at_slash(line, llen);
830 if (!name || name == line)
831 return NULL;
832 name++;
834 /*
835 * since the first name is unquoted, a dq if exists must be
836 * the beginning of the second name.
837 */
838 for (second = name; second < line + llen; second++) {
839 if (*second == '"') {
840 struct strbuf sp = STRBUF_INIT;
841 const char *np;
843 if (unquote_c_style(&sp, second, NULL))
844 goto free_and_fail2;
846 np = stop_at_slash(sp.buf, sp.len);
847 if (!np || np == sp.buf)
848 goto free_and_fail2;
849 np++;
851 len = sp.buf + sp.len - np;
852 if (len < second - name &&
853 !strncmp(np, name, len) &&
854 isspace(name[len])) {
855 /* Good */
856 strbuf_remove(&sp, 0, np - sp.buf);
857 return strbuf_detach(&sp, NULL);
858 }
860 free_and_fail2:
861 strbuf_release(&sp);
862 return NULL;
863 }
864 }
866 /*
867 * Accept a name only if it shows up twice, exactly the same
868 * form.
869 */
870 for (len = 0 ; ; len++) {
871 switch (name[len]) {
872 default:
873 continue;
874 case '\n':
875 return NULL;
876 case '\t': case ' ':
877 second = name+len;
878 for (;;) {
879 char c = *second++;
880 if (c == '\n')
881 return NULL;
882 if (c == '/')
883 break;
884 }
885 if (second[len] == '\n' && !memcmp(name, second, len)) {
886 return xmemdupz(name, len);
887 }
888 }
889 }
890 }
892 /* Verify that we recognize the lines following a git header */
893 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
894 {
895 unsigned long offset;
897 /* A git diff has explicit new/delete information, so we don't guess */
898 patch->is_new = 0;
899 patch->is_delete = 0;
901 /*
902 * Some things may not have the old name in the
903 * rest of the headers anywhere (pure mode changes,
904 * or removing or adding empty files), so we get
905 * the default name from the header.
906 */
907 patch->def_name = git_header_name(line, len);
908 if (patch->def_name && root) {
909 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
910 strcpy(s, root);
911 strcpy(s + root_len, patch->def_name);
912 free(patch->def_name);
913 patch->def_name = s;
914 }
916 line += len;
917 size -= len;
918 linenr++;
919 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
920 static const struct opentry {
921 const char *str;
922 int (*fn)(const char *, struct patch *);
923 } optable[] = {
924 { "@@ -", gitdiff_hdrend },
925 { "--- ", gitdiff_oldname },
926 { "+++ ", gitdiff_newname },
927 { "old mode ", gitdiff_oldmode },
928 { "new mode ", gitdiff_newmode },
929 { "deleted file mode ", gitdiff_delete },
930 { "new file mode ", gitdiff_newfile },
931 { "copy from ", gitdiff_copysrc },
932 { "copy to ", gitdiff_copydst },
933 { "rename old ", gitdiff_renamesrc },
934 { "rename new ", gitdiff_renamedst },
935 { "rename from ", gitdiff_renamesrc },
936 { "rename to ", gitdiff_renamedst },
937 { "similarity index ", gitdiff_similarity },
938 { "dissimilarity index ", gitdiff_dissimilarity },
939 { "index ", gitdiff_index },
940 { "", gitdiff_unrecognized },
941 };
942 int i;
944 len = linelen(line, size);
945 if (!len || line[len-1] != '\n')
946 break;
947 for (i = 0; i < ARRAY_SIZE(optable); i++) {
948 const struct opentry *p = optable + i;
949 int oplen = strlen(p->str);
950 if (len < oplen || memcmp(p->str, line, oplen))
951 continue;
952 if (p->fn(line + oplen, patch) < 0)
953 return offset;
954 break;
955 }
956 }
958 return offset;
959 }
961 static int parse_num(const char *line, unsigned long *p)
962 {
963 char *ptr;
965 if (!isdigit(*line))
966 return 0;
967 *p = strtoul(line, &ptr, 10);
968 return ptr - line;
969 }
971 static int parse_range(const char *line, int len, int offset, const char *expect,
972 unsigned long *p1, unsigned long *p2)
973 {
974 int digits, ex;
976 if (offset < 0 || offset >= len)
977 return -1;
978 line += offset;
979 len -= offset;
981 digits = parse_num(line, p1);
982 if (!digits)
983 return -1;
985 offset += digits;
986 line += digits;
987 len -= digits;
989 *p2 = 1;
990 if (*line == ',') {
991 digits = parse_num(line+1, p2);
992 if (!digits)
993 return -1;
995 offset += digits+1;
996 line += digits+1;
997 len -= digits+1;
998 }
1000 ex = strlen(expect);
1001 if (ex > len)
1002 return -1;
1003 if (memcmp(line, expect, ex))
1004 return -1;
1006 return offset + ex;
1007 }
1009 static void recount_diff(char *line, int size, struct fragment *fragment)
1010 {
1011 int oldlines = 0, newlines = 0, ret = 0;
1013 if (size < 1) {
1014 warning("recount: ignore empty hunk");
1015 return;
1016 }
1018 for (;;) {
1019 int len = linelen(line, size);
1020 size -= len;
1021 line += len;
1023 if (size < 1)
1024 break;
1026 switch (*line) {
1027 case ' ': case '\n':
1028 newlines++;
1029 /* fall through */
1030 case '-':
1031 oldlines++;
1032 continue;
1033 case '+':
1034 newlines++;
1035 continue;
1036 case '\\':
1037 continue;
1038 case '@':
1039 ret = size < 3 || prefixcmp(line, "@@ ");
1040 break;
1041 case 'd':
1042 ret = size < 5 || prefixcmp(line, "diff ");
1043 break;
1044 default:
1045 ret = -1;
1046 break;
1047 }
1048 if (ret) {
1049 warning("recount: unexpected line: %.*s",
1050 (int)linelen(line, size), line);
1051 return;
1052 }
1053 break;
1054 }
1055 fragment->oldlines = oldlines;
1056 fragment->newlines = newlines;
1057 }
1059 /*
1060 * Parse a unified diff fragment header of the
1061 * form "@@ -a,b +c,d @@"
1062 */
1063 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1064 {
1065 int offset;
1067 if (!len || line[len-1] != '\n')
1068 return -1;
1070 /* Figure out the number of lines in a fragment */
1071 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1072 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1074 return offset;
1075 }
1077 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1078 {
1079 unsigned long offset, len;
1081 patch->is_toplevel_relative = 0;
1082 patch->is_rename = patch->is_copy = 0;
1083 patch->is_new = patch->is_delete = -1;
1084 patch->old_mode = patch->new_mode = 0;
1085 patch->old_name = patch->new_name = NULL;
1086 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1087 unsigned long nextlen;
1089 len = linelen(line, size);
1090 if (!len)
1091 break;
1093 /* Testing this early allows us to take a few shortcuts.. */
1094 if (len < 6)
1095 continue;
1097 /*
1098 * Make sure we don't find any unconnected patch fragments.
1099 * That's a sign that we didn't find a header, and that a
1100 * patch has become corrupted/broken up.
1101 */
1102 if (!memcmp("@@ -", line, 4)) {
1103 struct fragment dummy;
1104 if (parse_fragment_header(line, len, &dummy) < 0)
1105 continue;
1106 die("patch fragment without header at line %d: %.*s",
1107 linenr, (int)len-1, line);
1108 }
1110 if (size < len + 6)
1111 break;
1113 /*
1114 * Git patch? It might not have a real patch, just a rename
1115 * or mode change, so we handle that specially
1116 */
1117 if (!memcmp("diff --git ", line, 11)) {
1118 int git_hdr_len = parse_git_header(line, len, size, patch);
1119 if (git_hdr_len <= len)
1120 continue;
1121 if (!patch->old_name && !patch->new_name) {
1122 if (!patch->def_name)
1123 die("git diff header lacks filename information (line %d)", linenr);
1124 patch->old_name = patch->new_name = patch->def_name;
1125 }
1126 patch->is_toplevel_relative = 1;
1127 *hdrsize = git_hdr_len;
1128 return offset;
1129 }
1131 /* --- followed by +++ ? */
1132 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1133 continue;
1135 /*
1136 * We only accept unified patches, so we want it to
1137 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1138 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1139 */
1140 nextlen = linelen(line + len, size - len);
1141 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1142 continue;
1144 /* Ok, we'll consider it a patch */
1145 parse_traditional_patch(line, line+len, patch);
1146 *hdrsize = len + nextlen;
1147 linenr += 2;
1148 return offset;
1149 }
1150 return -1;
1151 }
1153 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1154 {
1155 char *err;
1157 if (!result)
1158 return;
1160 whitespace_error++;
1161 if (squelch_whitespace_errors &&
1162 squelch_whitespace_errors < whitespace_error)
1163 return;
1165 err = whitespace_error_string(result);
1166 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1167 patch_input_file, linenr, err, len, line);
1168 free(err);
1169 }
1171 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1172 {
1173 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1175 record_ws_error(result, line + 1, len - 2, linenr);
1176 }
1178 /*
1179 * Parse a unified diff. Note that this really needs to parse each
1180 * fragment separately, since the only way to know the difference
1181 * between a "---" that is part of a patch, and a "---" that starts
1182 * the next patch is to look at the line counts..
1183 */
1184 static int parse_fragment(char *line, unsigned long size,
1185 struct patch *patch, struct fragment *fragment)
1186 {
1187 int added, deleted;
1188 int len = linelen(line, size), offset;
1189 unsigned long oldlines, newlines;
1190 unsigned long leading, trailing;
1192 offset = parse_fragment_header(line, len, fragment);
1193 if (offset < 0)
1194 return -1;
1195 if (offset > 0 && patch->recount)
1196 recount_diff(line + offset, size - offset, fragment);
1197 oldlines = fragment->oldlines;
1198 newlines = fragment->newlines;
1199 leading = 0;
1200 trailing = 0;
1202 /* Parse the thing.. */
1203 line += len;
1204 size -= len;
1205 linenr++;
1206 added = deleted = 0;
1207 for (offset = len;
1208 0 < size;
1209 offset += len, size -= len, line += len, linenr++) {
1210 if (!oldlines && !newlines)
1211 break;
1212 len = linelen(line, size);
1213 if (!len || line[len-1] != '\n')
1214 return -1;
1215 switch (*line) {
1216 default:
1217 return -1;
1218 case '\n': /* newer GNU diff, an empty context line */
1219 case ' ':
1220 oldlines--;
1221 newlines--;
1222 if (!deleted && !added)
1223 leading++;
1224 trailing++;
1225 break;
1226 case '-':
1227 if (apply_in_reverse &&
1228 ws_error_action != nowarn_ws_error)
1229 check_whitespace(line, len, patch->ws_rule);
1230 deleted++;
1231 oldlines--;
1232 trailing = 0;
1233 break;
1234 case '+':
1235 if (!apply_in_reverse &&
1236 ws_error_action != nowarn_ws_error)
1237 check_whitespace(line, len, patch->ws_rule);
1238 added++;
1239 newlines--;
1240 trailing = 0;
1241 break;
1243 /*
1244 * We allow "\ No newline at end of file". Depending
1245 * on locale settings when the patch was produced we
1246 * don't know what this line looks like. The only
1247 * thing we do know is that it begins with "\ ".
1248 * Checking for 12 is just for sanity check -- any
1249 * l10n of "\ No newline..." is at least that long.
1250 */
1251 case '\\':
1252 if (len < 12 || memcmp(line, "\\ ", 2))
1253 return -1;
1254 break;
1255 }
1256 }
1257 if (oldlines || newlines)
1258 return -1;
1259 fragment->leading = leading;
1260 fragment->trailing = trailing;
1262 /*
1263 * If a fragment ends with an incomplete line, we failed to include
1264 * it in the above loop because we hit oldlines == newlines == 0
1265 * before seeing it.
1266 */
1267 if (12 < size && !memcmp(line, "\\ ", 2))
1268 offset += linelen(line, size);
1270 patch->lines_added += added;
1271 patch->lines_deleted += deleted;
1273 if (0 < patch->is_new && oldlines)
1274 return error("new file depends on old contents");
1275 if (0 < patch->is_delete && newlines)
1276 return error("deleted file still has contents");
1277 return offset;
1278 }
1280 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1281 {
1282 unsigned long offset = 0;
1283 unsigned long oldlines = 0, newlines = 0, context = 0;
1284 struct fragment **fragp = &patch->fragments;
1286 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1287 struct fragment *fragment;
1288 int len;
1290 fragment = xcalloc(1, sizeof(*fragment));
1291 fragment->linenr = linenr;
1292 len = parse_fragment(line, size, patch, fragment);
1293 if (len <= 0)
1294 die("corrupt patch at line %d", linenr);
1295 fragment->patch = line;
1296 fragment->size = len;
1297 oldlines += fragment->oldlines;
1298 newlines += fragment->newlines;
1299 context += fragment->leading + fragment->trailing;
1301 *fragp = fragment;
1302 fragp = &fragment->next;
1304 offset += len;
1305 line += len;
1306 size -= len;
1307 }
1309 /*
1310 * If something was removed (i.e. we have old-lines) it cannot
1311 * be creation, and if something was added it cannot be
1312 * deletion. However, the reverse is not true; --unified=0
1313 * patches that only add are not necessarily creation even
1314 * though they do not have any old lines, and ones that only
1315 * delete are not necessarily deletion.
1316 *
1317 * Unfortunately, a real creation/deletion patch do _not_ have
1318 * any context line by definition, so we cannot safely tell it
1319 * apart with --unified=0 insanity. At least if the patch has
1320 * more than one hunk it is not creation or deletion.
1321 */
1322 if (patch->is_new < 0 &&
1323 (oldlines || (patch->fragments && patch->fragments->next)))
1324 patch->is_new = 0;
1325 if (patch->is_delete < 0 &&
1326 (newlines || (patch->fragments && patch->fragments->next)))
1327 patch->is_delete = 0;
1329 if (0 < patch->is_new && oldlines)
1330 die("new file %s depends on old contents", patch->new_name);
1331 if (0 < patch->is_delete && newlines)
1332 die("deleted file %s still has contents", patch->old_name);
1333 if (!patch->is_delete && !newlines && context)
1334 fprintf(stderr, "** warning: file %s becomes empty but "
1335 "is not deleted\n", patch->new_name);
1337 return offset;
1338 }
1340 static inline int metadata_changes(struct patch *patch)
1341 {
1342 return patch->is_rename > 0 ||
1343 patch->is_copy > 0 ||
1344 patch->is_new > 0 ||
1345 patch->is_delete ||
1346 (patch->old_mode && patch->new_mode &&
1347 patch->old_mode != patch->new_mode);
1348 }
1350 static char *inflate_it(const void *data, unsigned long size,
1351 unsigned long inflated_size)
1352 {
1353 z_stream stream;
1354 void *out;
1355 int st;
1357 memset(&stream, 0, sizeof(stream));
1359 stream.next_in = (unsigned char *)data;
1360 stream.avail_in = size;
1361 stream.next_out = out = xmalloc(inflated_size);
1362 stream.avail_out = inflated_size;
1363 git_inflate_init(&stream);
1364 st = git_inflate(&stream, Z_FINISH);
1365 git_inflate_end(&stream);
1366 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1367 free(out);
1368 return NULL;
1369 }
1370 return out;
1371 }
1373 static struct fragment *parse_binary_hunk(char **buf_p,
1374 unsigned long *sz_p,
1375 int *status_p,
1376 int *used_p)
1377 {
1378 /*
1379 * Expect a line that begins with binary patch method ("literal"
1380 * or "delta"), followed by the length of data before deflating.
1381 * a sequence of 'length-byte' followed by base-85 encoded data
1382 * should follow, terminated by a newline.
1383 *
1384 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1385 * and we would limit the patch line to 66 characters,
1386 * so one line can fit up to 13 groups that would decode
1387 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1388 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1389 */
1390 int llen, used;
1391 unsigned long size = *sz_p;
1392 char *buffer = *buf_p;
1393 int patch_method;
1394 unsigned long origlen;
1395 char *data = NULL;
1396 int hunk_size = 0;
1397 struct fragment *frag;
1399 llen = linelen(buffer, size);
1400 used = llen;
1402 *status_p = 0;
1404 if (!prefixcmp(buffer, "delta ")) {
1405 patch_method = BINARY_DELTA_DEFLATED;
1406 origlen = strtoul(buffer + 6, NULL, 10);
1407 }
1408 else if (!prefixcmp(buffer, "literal ")) {
1409 patch_method = BINARY_LITERAL_DEFLATED;
1410 origlen = strtoul(buffer + 8, NULL, 10);
1411 }
1412 else
1413 return NULL;
1415 linenr++;
1416 buffer += llen;
1417 while (1) {
1418 int byte_length, max_byte_length, newsize;
1419 llen = linelen(buffer, size);
1420 used += llen;
1421 linenr++;
1422 if (llen == 1) {
1423 /* consume the blank line */
1424 buffer++;
1425 size--;
1426 break;
1427 }
1428 /*
1429 * Minimum line is "A00000\n" which is 7-byte long,
1430 * and the line length must be multiple of 5 plus 2.
1431 */
1432 if ((llen < 7) || (llen-2) % 5)
1433 goto corrupt;
1434 max_byte_length = (llen - 2) / 5 * 4;
1435 byte_length = *buffer;
1436 if ('A' <= byte_length && byte_length <= 'Z')
1437 byte_length = byte_length - 'A' + 1;
1438 else if ('a' <= byte_length && byte_length <= 'z')
1439 byte_length = byte_length - 'a' + 27;
1440 else
1441 goto corrupt;
1442 /* if the input length was not multiple of 4, we would
1443 * have filler at the end but the filler should never
1444 * exceed 3 bytes
1445 */
1446 if (max_byte_length < byte_length ||
1447 byte_length <= max_byte_length - 4)
1448 goto corrupt;
1449 newsize = hunk_size + byte_length;
1450 data = xrealloc(data, newsize);
1451 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1452 goto corrupt;
1453 hunk_size = newsize;
1454 buffer += llen;
1455 size -= llen;
1456 }
1458 frag = xcalloc(1, sizeof(*frag));
1459 frag->patch = inflate_it(data, hunk_size, origlen);
1460 if (!frag->patch)
1461 goto corrupt;
1462 free(data);
1463 frag->size = origlen;
1464 *buf_p = buffer;
1465 *sz_p = size;
1466 *used_p = used;
1467 frag->binary_patch_method = patch_method;
1468 return frag;
1470 corrupt:
1471 free(data);
1472 *status_p = -1;
1473 error("corrupt binary patch at line %d: %.*s",
1474 linenr-1, llen-1, buffer);
1475 return NULL;
1476 }
1478 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1479 {
1480 /*
1481 * We have read "GIT binary patch\n"; what follows is a line
1482 * that says the patch method (currently, either "literal" or
1483 * "delta") and the length of data before deflating; a
1484 * sequence of 'length-byte' followed by base-85 encoded data
1485 * follows.
1486 *
1487 * When a binary patch is reversible, there is another binary
1488 * hunk in the same format, starting with patch method (either
1489 * "literal" or "delta") with the length of data, and a sequence
1490 * of length-byte + base-85 encoded data, terminated with another
1491 * empty line. This data, when applied to the postimage, produces
1492 * the preimage.
1493 */
1494 struct fragment *forward;
1495 struct fragment *reverse;
1496 int status;
1497 int used, used_1;
1499 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1500 if (!forward && !status)
1501 /* there has to be one hunk (forward hunk) */
1502 return error("unrecognized binary patch at line %d", linenr-1);
1503 if (status)
1504 /* otherwise we already gave an error message */
1505 return status;
1507 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1508 if (reverse)
1509 used += used_1;
1510 else if (status) {
1511 /*
1512 * Not having reverse hunk is not an error, but having
1513 * a corrupt reverse hunk is.
1514 */
1515 free((void*) forward->patch);
1516 free(forward);
1517 return status;
1518 }
1519 forward->next = reverse;
1520 patch->fragments = forward;
1521 patch->is_binary = 1;
1522 return used;
1523 }
1525 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1526 {
1527 int hdrsize, patchsize;
1528 int offset = find_header(buffer, size, &hdrsize, patch);
1530 if (offset < 0)
1531 return offset;
1533 patch->ws_rule = whitespace_rule(patch->new_name
1534 ? patch->new_name
1535 : patch->old_name);
1537 patchsize = parse_single_patch(buffer + offset + hdrsize,
1538 size - offset - hdrsize, patch);
1540 if (!patchsize) {
1541 static const char *binhdr[] = {
1542 "Binary files ",
1543 "Files ",
1544 NULL,
1545 };
1546 static const char git_binary[] = "GIT binary patch\n";
1547 int i;
1548 int hd = hdrsize + offset;
1549 unsigned long llen = linelen(buffer + hd, size - hd);
1551 if (llen == sizeof(git_binary) - 1 &&
1552 !memcmp(git_binary, buffer + hd, llen)) {
1553 int used;
1554 linenr++;
1555 used = parse_binary(buffer + hd + llen,
1556 size - hd - llen, patch);
1557 if (used)
1558 patchsize = used + llen;
1559 else
1560 patchsize = 0;
1561 }
1562 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1563 for (i = 0; binhdr[i]; i++) {
1564 int len = strlen(binhdr[i]);
1565 if (len < size - hd &&
1566 !memcmp(binhdr[i], buffer + hd, len)) {
1567 linenr++;
1568 patch->is_binary = 1;
1569 patchsize = llen;
1570 break;
1571 }
1572 }
1573 }
1575 /* Empty patch cannot be applied if it is a text patch
1576 * without metadata change. A binary patch appears
1577 * empty to us here.
1578 */
1579 if ((apply || check) &&
1580 (!patch->is_binary && !metadata_changes(patch)))
1581 die("patch with only garbage at line %d", linenr);
1582 }
1584 return offset + hdrsize + patchsize;
1585 }
1587 #define swap(a,b) myswap((a),(b),sizeof(a))
1589 #define myswap(a, b, size) do { \
1590 unsigned char mytmp[size]; \
1591 memcpy(mytmp, &a, size); \
1592 memcpy(&a, &b, size); \
1593 memcpy(&b, mytmp, size); \
1594 } while (0)
1596 static void reverse_patches(struct patch *p)
1597 {
1598 for (; p; p = p->next) {
1599 struct fragment *frag = p->fragments;
1601 swap(p->new_name, p->old_name);
1602 swap(p->new_mode, p->old_mode);
1603 swap(p->is_new, p->is_delete);
1604 swap(p->lines_added, p->lines_deleted);
1605 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1607 for (; frag; frag = frag->next) {
1608 swap(frag->newpos, frag->oldpos);
1609 swap(frag->newlines, frag->oldlines);
1610 }
1611 }
1612 }
1614 static const char pluses[] =
1615 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1616 static const char minuses[]=
1617 "----------------------------------------------------------------------";
1619 static void show_stats(struct patch *patch)
1620 {
1621 struct strbuf qname = STRBUF_INIT;
1622 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1623 int max, add, del;
1625 quote_c_style(cp, &qname, NULL, 0);
1627 /*
1628 * "scale" the filename
1629 */
1630 max = max_len;
1631 if (max > 50)
1632 max = 50;
1634 if (qname.len > max) {
1635 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1636 if (!cp)
1637 cp = qname.buf + qname.len + 3 - max;
1638 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1639 }
1641 if (patch->is_binary) {
1642 printf(" %-*s | Bin\n", max, qname.buf);
1643 strbuf_release(&qname);
1644 return;
1645 }
1647 printf(" %-*s |", max, qname.buf);
1648 strbuf_release(&qname);
1650 /*
1651 * scale the add/delete
1652 */
1653 max = max + max_change > 70 ? 70 - max : max_change;
1654 add = patch->lines_added;
1655 del = patch->lines_deleted;
1657 if (max_change > 0) {
1658 int total = ((add + del) * max + max_change / 2) / max_change;
1659 add = (add * max + max_change / 2) / max_change;
1660 del = total - add;
1661 }
1662 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1663 add, pluses, del, minuses);
1664 }
1666 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1667 {
1668 switch (st->st_mode & S_IFMT) {
1669 case S_IFLNK:
1670 if (strbuf_readlink(buf, path, st->st_size) < 0)
1671 return error("unable to read symlink %s", path);
1672 return 0;
1673 case S_IFREG:
1674 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1675 return error("unable to open or read %s", path);
1676 convert_to_git(path, buf->buf, buf->len, buf, 0);
1677 return 0;
1678 default:
1679 return -1;
1680 }
1681 }
1683 static void update_pre_post_images(struct image *preimage,
1684 struct image *postimage,
1685 char *buf,
1686 size_t len)
1687 {
1688 int i, ctx;
1689 char *new, *old, *fixed;
1690 struct image fixed_preimage;
1692 /*
1693 * Update the preimage with whitespace fixes. Note that we
1694 * are not losing preimage->buf -- apply_one_fragment() will
1695 * free "oldlines".
1696 */
1697 prepare_image(&fixed_preimage, buf, len, 1);
1698 assert(fixed_preimage.nr == preimage->nr);
1699 for (i = 0; i < preimage->nr; i++)
1700 fixed_preimage.line[i].flag = preimage->line[i].flag;
1701 free(preimage->line_allocated);
1702 *preimage = fixed_preimage;
1704 /*
1705 * Adjust the common context lines in postimage, in place.
1706 * This is possible because whitespace fixing does not make
1707 * the string grow.
1708 */
1709 new = old = postimage->buf;
1710 fixed = preimage->buf;
1711 for (i = ctx = 0; i < postimage->nr; i++) {
1712 size_t len = postimage->line[i].len;
1713 if (!(postimage->line[i].flag & LINE_COMMON)) {
1714 /* an added line -- no counterparts in preimage */
1715 memmove(new, old, len);
1716 old += len;
1717 new += len;
1718 continue;
1719 }
1721 /* a common context -- skip it in the original postimage */
1722 old += len;
1724 /* and find the corresponding one in the fixed preimage */
1725 while (ctx < preimage->nr &&
1726 !(preimage->line[ctx].flag & LINE_COMMON)) {
1727 fixed += preimage->line[ctx].len;
1728 ctx++;
1729 }
1730 if (preimage->nr <= ctx)
1731 die("oops");
1733 /* and copy it in, while fixing the line length */
1734 len = preimage->line[ctx].len;
1735 memcpy(new, fixed, len);
1736 new += len;
1737 fixed += len;
1738 postimage->line[i].len = len;
1739 ctx++;
1740 }
1742 /* Fix the length of the whole thing */
1743 postimage->len = new - postimage->buf;
1744 }
1746 static int match_fragment(struct image *img,
1747 struct image *preimage,
1748 struct image *postimage,
1749 unsigned long try,
1750 int try_lno,
1751 unsigned ws_rule,
1752 int match_beginning, int match_end)
1753 {
1754 int i;
1755 char *fixed_buf, *buf, *orig, *target;
1757 if (preimage->nr + try_lno > img->nr)
1758 return 0;
1760 if (match_beginning && try_lno)
1761 return 0;
1763 if (match_end && preimage->nr + try_lno != img->nr)
1764 return 0;
1766 /* Quick hash check */
1767 for (i = 0; i < preimage->nr; i++)
1768 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1769 return 0;
1771 /*
1772 * Do we have an exact match? If we were told to match
1773 * at the end, size must be exactly at try+fragsize,
1774 * otherwise try+fragsize must be still within the preimage,
1775 * and either case, the old piece should match the preimage
1776 * exactly.
1777 */
1778 if ((match_end
1779 ? (try + preimage->len == img->len)
1780 : (try + preimage->len <= img->len)) &&
1781 !memcmp(img->buf + try, preimage->buf, preimage->len))
1782 return 1;
1784 if (ws_error_action != correct_ws_error)
1785 return 0;
1787 /*
1788 * The hunk does not apply byte-by-byte, but the hash says
1789 * it might with whitespace fuzz.
1790 */
1791 fixed_buf = xmalloc(preimage->len + 1);
1792 buf = fixed_buf;
1793 orig = preimage->buf;
1794 target = img->buf + try;
1795 for (i = 0; i < preimage->nr; i++) {
1796 size_t fixlen; /* length after fixing the preimage */
1797 size_t oldlen = preimage->line[i].len;
1798 size_t tgtlen = img->line[try_lno + i].len;
1799 size_t tgtfixlen; /* length after fixing the target line */
1800 char tgtfixbuf[1024], *tgtfix;
1801 int match;
1803 /* Try fixing the line in the preimage */
1804 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1806 /* Try fixing the line in the target */
1807 if (sizeof(tgtfixbuf) > tgtlen)
1808 tgtfix = tgtfixbuf;
1809 else
1810 tgtfix = xmalloc(tgtlen);
1811 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1813 /*
1814 * If they match, either the preimage was based on
1815 * a version before our tree fixed whitespace breakage,
1816 * or we are lacking a whitespace-fix patch the tree
1817 * the preimage was based on already had (i.e. target
1818 * has whitespace breakage, the preimage doesn't).
1819 * In either case, we are fixing the whitespace breakages
1820 * so we might as well take the fix together with their
1821 * real change.
1822 */
1823 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1825 if (tgtfix != tgtfixbuf)
1826 free(tgtfix);
1827 if (!match)
1828 goto unmatch_exit;
1830 orig += oldlen;
1831 buf += fixlen;
1832 target += tgtlen;
1833 }
1835 /*
1836 * Yes, the preimage is based on an older version that still
1837 * has whitespace breakages unfixed, and fixing them makes the
1838 * hunk match. Update the context lines in the postimage.
1839 */
1840 update_pre_post_images(preimage, postimage,
1841 fixed_buf, buf - fixed_buf);
1842 return 1;
1844 unmatch_exit:
1845 free(fixed_buf);
1846 return 0;
1847 }
1849 static int find_pos(struct image *img,
1850 struct image *preimage,
1851 struct image *postimage,
1852 int line,
1853 unsigned ws_rule,
1854 int match_beginning, int match_end)
1855 {
1856 int i;
1857 unsigned long backwards, forwards, try;
1858 int backwards_lno, forwards_lno, try_lno;
1860 if (preimage->nr > img->nr)
1861 return -1;
1863 /*
1864 * If match_begining or match_end is specified, there is no
1865 * point starting from a wrong line that will never match and
1866 * wander around and wait for a match at the specified end.
1867 */
1868 if (match_beginning)
1869 line = 0;
1870 else if (match_end)
1871 line = img->nr - preimage->nr;
1873 if (line > img->nr)
1874 line = img->nr;
1876 try = 0;
1877 for (i = 0; i < line; i++)
1878 try += img->line[i].len;
1880 /*
1881 * There's probably some smart way to do this, but I'll leave
1882 * that to the smart and beautiful people. I'm simple and stupid.
1883 */
1884 backwards = try;
1885 backwards_lno = line;
1886 forwards = try;
1887 forwards_lno = line;
1888 try_lno = line;
1890 for (i = 0; ; i++) {
1891 if (match_fragment(img, preimage, postimage,
1892 try, try_lno, ws_rule,
1893 match_beginning, match_end))
1894 return try_lno;
1896 again:
1897 if (backwards_lno == 0 && forwards_lno == img->nr)
1898 break;
1900 if (i & 1) {
1901 if (backwards_lno == 0) {
1902 i++;
1903 goto again;
1904 }
1905 backwards_lno--;
1906 backwards -= img->line[backwards_lno].len;
1907 try = backwards;
1908 try_lno = backwards_lno;
1909 } else {
1910 if (forwards_lno == img->nr) {
1911 i++;
1912 goto again;
1913 }
1914 forwards += img->line[forwards_lno].len;
1915 forwards_lno++;
1916 try = forwards;
1917 try_lno = forwards_lno;
1918 }
1920 }
1921 return -1;
1922 }
1924 static void remove_first_line(struct image *img)
1925 {
1926 img->buf += img->line[0].len;
1927 img->len -= img->line[0].len;
1928 img->line++;
1929 img->nr--;
1930 }
1932 static void remove_last_line(struct image *img)
1933 {
1934 img->len -= img->line[--img->nr].len;
1935 }
1937 static void update_image(struct image *img,
1938 int applied_pos,
1939 struct image *preimage,
1940 struct image *postimage)
1941 {
1942 /*
1943 * remove the copy of preimage at offset in img
1944 * and replace it with postimage
1945 */
1946 int i, nr;
1947 size_t remove_count, insert_count, applied_at = 0;
1948 char *result;
1950 for (i = 0; i < applied_pos; i++)
1951 applied_at += img->line[i].len;
1953 remove_count = 0;
1954 for (i = 0; i < preimage->nr; i++)
1955 remove_count += img->line[applied_pos + i].len;
1956 insert_count = postimage->len;
1958 /* Adjust the contents */
1959 result = xmalloc(img->len + insert_count - remove_count + 1);
1960 memcpy(result, img->buf, applied_at);
1961 memcpy(result + applied_at, postimage->buf, postimage->len);
1962 memcpy(result + applied_at + postimage->len,
1963 img->buf + (applied_at + remove_count),
1964 img->len - (applied_at + remove_count));
1965 free(img->buf);
1966 img->buf = result;
1967 img->len += insert_count - remove_count;
1968 result[img->len] = '\0';
1970 /* Adjust the line table */
1971 nr = img->nr + postimage->nr - preimage->nr;
1972 if (preimage->nr < postimage->nr) {
1973 /*
1974 * NOTE: this knows that we never call remove_first_line()
1975 * on anything other than pre/post image.
1976 */
1977 img->line = xrealloc(img->line, nr * sizeof(*img->line));
1978 img->line_allocated = img->line;
1979 }
1980 if (preimage->nr != postimage->nr)
1981 memmove(img->line + applied_pos + postimage->nr,
1982 img->line + applied_pos + preimage->nr,
1983 (img->nr - (applied_pos + preimage->nr)) *
1984 sizeof(*img->line));
1985 memcpy(img->line + applied_pos,
1986 postimage->line,
1987 postimage->nr * sizeof(*img->line));
1988 img->nr = nr;
1989 }
1991 static int apply_one_fragment(struct image *img, struct fragment *frag,
1992 int inaccurate_eof, unsigned ws_rule)
1993 {
1994 int match_beginning, match_end;
1995 const char *patch = frag->patch;
1996 int size = frag->size;
1997 char *old, *new, *oldlines, *newlines;
1998 int new_blank_lines_at_end = 0;
1999 unsigned long leading, trailing;
2000 int pos, applied_pos;
2001 struct image preimage;
2002 struct image postimage;
2004 memset(&preimage, 0, sizeof(preimage));
2005 memset(&postimage, 0, sizeof(postimage));
2006 oldlines = xmalloc(size);
2007 newlines = xmalloc(size);
2009 old = oldlines;
2010 new = newlines;
2011 while (size > 0) {
2012 char first;
2013 int len = linelen(patch, size);
2014 int plen, added;
2015 int added_blank_line = 0;
2016 int is_blank_context = 0;
2018 if (!len)
2019 break;
2021 /*
2022 * "plen" is how much of the line we should use for
2023 * the actual patch data. Normally we just remove the
2024 * first character on the line, but if the line is
2025 * followed by "\ No newline", then we also remove the
2026 * last one (which is the newline, of course).
2027 */
2028 plen = len - 1;
2029 if (len < size && patch[len] == '\\')
2030 plen--;
2031 first = *patch;
2032 if (apply_in_reverse) {
2033 if (first == '-')
2034 first = '+';
2035 else if (first == '+')
2036 first = '-';
2037 }
2039 switch (first) {
2040 case '\n':
2041 /* Newer GNU diff, empty context line */
2042 if (plen < 0)
2043 /* ... followed by '\No newline'; nothing */
2044 break;
2045 *old++ = '\n';
2046 *new++ = '\n';
2047 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2048 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2049 is_blank_context = 1;
2050 break;
2051 case ' ':
2052 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2053 ws_blank_line(patch + 1, plen, ws_rule))
2054 is_blank_context = 1;
2055 case '-':
2056 memcpy(old, patch + 1, plen);
2057 add_line_info(&preimage, old, plen,
2058 (first == ' ' ? LINE_COMMON : 0));
2059 old += plen;
2060 if (first == '-')
2061 break;
2062 /* Fall-through for ' ' */
2063 case '+':
2064 /* --no-add does not add new lines */
2065 if (first == '+' && no_add)
2066 break;
2068 if (first != '+' ||
2069 !whitespace_error ||
2070 ws_error_action != correct_ws_error) {
2071 memcpy(new, patch + 1, plen);
2072 added = plen;
2073 }
2074 else {
2075 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2076 }
2077 add_line_info(&postimage, new, added,
2078 (first == '+' ? 0 : LINE_COMMON));
2079 new += added;
2080 if (first == '+' &&
2081 (ws_rule & WS_BLANK_AT_EOF) &&
2082 ws_blank_line(patch + 1, plen, ws_rule))
2083 added_blank_line = 1;
2084 break;
2085 case '@': case '\\':
2086 /* Ignore it, we already handled it */
2087 break;
2088 default:
2089 if (apply_verbosely)
2090 error("invalid start of line: '%c'", first);
2091 return -1;
2092 }
2093 if (added_blank_line)
2094 new_blank_lines_at_end++;
2095 else if (is_blank_context)
2096 ;
2097 else
2098 new_blank_lines_at_end = 0;
2099 patch += len;
2100 size -= len;
2101 }
2102 if (inaccurate_eof &&
2103 old > oldlines && old[-1] == '\n' &&
2104 new > newlines && new[-1] == '\n') {
2105 old--;
2106 new--;
2107 }
2109 leading = frag->leading;
2110 trailing = frag->trailing;
2112 /*
2113 * A hunk to change lines at the beginning would begin with
2114 * @@ -1,L +N,M @@
2115 * but we need to be careful. -U0 that inserts before the second
2116 * line also has this pattern.
2117 *
2118 * And a hunk to add to an empty file would begin with
2119 * @@ -0,0 +N,M @@
2120 *
2121 * In other words, a hunk that is (frag->oldpos <= 1) with or
2122 * without leading context must match at the beginning.
2123 */
2124 match_beginning = (!frag->oldpos ||
2125 (frag->oldpos == 1 && !unidiff_zero));
2127 /*
2128 * A hunk without trailing lines must match at the end.
2129 * However, we simply cannot tell if a hunk must match end
2130 * from the lack of trailing lines if the patch was generated
2131 * with unidiff without any context.
2132 */
2133 match_end = !unidiff_zero && !trailing;
2135 pos = frag->newpos ? (frag->newpos - 1) : 0;
2136 preimage.buf = oldlines;
2137 preimage.len = old - oldlines;
2138 postimage.buf = newlines;
2139 postimage.len = new - newlines;
2140 preimage.line = preimage.line_allocated;
2141 postimage.line = postimage.line_allocated;
2143 for (;;) {
2145 applied_pos = find_pos(img, &preimage, &postimage, pos,
2146 ws_rule, match_beginning, match_end);
2148 if (applied_pos >= 0)
2149 break;
2151 /* Am I at my context limits? */
2152 if ((leading <= p_context) && (trailing <= p_context))
2153 break;
2154 if (match_beginning || match_end) {
2155 match_beginning = match_end = 0;
2156 continue;
2157 }
2159 /*
2160 * Reduce the number of context lines; reduce both
2161 * leading and trailing if they are equal otherwise
2162 * just reduce the larger context.
2163 */
2164 if (leading >= trailing) {
2165 remove_first_line(&preimage);
2166 remove_first_line(&postimage);
2167 pos--;
2168 leading--;
2169 }
2170 if (trailing > leading) {
2171 remove_last_line(&preimage);
2172 remove_last_line(&postimage);
2173 trailing--;
2174 }
2175 }
2177 if (applied_pos >= 0) {
2178 if (new_blank_lines_at_end &&
2179 preimage.nr + applied_pos == img->nr &&
2180 (ws_rule & WS_BLANK_AT_EOF) &&
2181 ws_error_action != nowarn_ws_error) {
2182 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2183 if (ws_error_action == correct_ws_error) {
2184 while (new_blank_lines_at_end--)
2185 remove_last_line(&postimage);
2186 }
2187 /*
2188 * We would want to prevent write_out_results()
2189 * from taking place in apply_patch() that follows
2190 * the callchain led us here, which is:
2191 * apply_patch->check_patch_list->check_patch->
2192 * apply_data->apply_fragments->apply_one_fragment
2193 */
2194 if (ws_error_action == die_on_ws_error)
2195 apply = 0;
2196 }
2198 /*
2199 * Warn if it was necessary to reduce the number
2200 * of context lines.
2201 */
2202 if ((leading != frag->leading) ||
2203 (trailing != frag->trailing))
2204 fprintf(stderr, "Context reduced to (%ld/%ld)"
2205 " to apply fragment at %d\n",
2206 leading, trailing, applied_pos+1);
2207 update_image(img, applied_pos, &preimage, &postimage);
2208 } else {
2209 if (apply_verbosely)
2210 error("while searching for:\n%.*s",
2211 (int)(old - oldlines), oldlines);
2212 }
2214 free(oldlines);
2215 free(newlines);
2216 free(preimage.line_allocated);
2217 free(postimage.line_allocated);
2219 return (applied_pos < 0);
2220 }
2222 static int apply_binary_fragment(struct image *img, struct patch *patch)
2223 {
2224 struct fragment *fragment = patch->fragments;
2225 unsigned long len;
2226 void *dst;
2228 /* Binary patch is irreversible without the optional second hunk */
2229 if (apply_in_reverse) {
2230 if (!fragment->next)
2231 return error("cannot reverse-apply a binary patch "
2232 "without the reverse hunk to '%s'",
2233 patch->new_name
2234 ? patch->new_name : patch->old_name);
2235 fragment = fragment->next;
2236 }
2237 switch (fragment->binary_patch_method) {
2238 case BINARY_DELTA_DEFLATED:
2239 dst = patch_delta(img->buf, img->len, fragment->patch,
2240 fragment->size, &len);
2241 if (!dst)
2242 return -1;
2243 clear_image(img);
2244 img->buf = dst;
2245 img->len = len;
2246 return 0;
2247 case BINARY_LITERAL_DEFLATED:
2248 clear_image(img);
2249 img->len = fragment->size;
2250 img->buf = xmalloc(img->len+1);
2251 memcpy(img->buf, fragment->patch, img->len);
2252 img->buf[img->len] = '\0';
2253 return 0;
2254 }
2255 return -1;
2256 }
2258 static int apply_binary(struct image *img, struct patch *patch)
2259 {
2260 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2261 unsigned char sha1[20];
2263 /*
2264 * For safety, we require patch index line to contain
2265 * full 40-byte textual SHA1 for old and new, at least for now.
2266 */
2267 if (strlen(patch->old_sha1_prefix) != 40 ||
2268 strlen(patch->new_sha1_prefix) != 40 ||
2269 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2270 get_sha1_hex(patch->new_sha1_prefix, sha1))
2271 return error("cannot apply binary patch to '%s' "
2272 "without full index line", name);
2274 if (patch->old_name) {
2275 /*
2276 * See if the old one matches what the patch
2277 * applies to.
2278 */
2279 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2280 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2281 return error("the patch applies to '%s' (%s), "
2282 "which does not match the "
2283 "current contents.",
2284 name, sha1_to_hex(sha1));
2285 }
2286 else {
2287 /* Otherwise, the old one must be empty. */
2288 if (img->len)
2289 return error("the patch applies to an empty "
2290 "'%s' but it is not empty", name);
2291 }
2293 get_sha1_hex(patch->new_sha1_prefix, sha1);
2294 if (is_null_sha1(sha1)) {
2295 clear_image(img);
2296 return 0; /* deletion patch */
2297 }
2299 if (has_sha1_file(sha1)) {
2300 /* We already have the postimage */
2301 enum object_type type;
2302 unsigned long size;
2303 char *result;
2305 result = read_sha1_file(sha1, &type, &size);
2306 if (!result)
2307 return error("the necessary postimage %s for "
2308 "'%s' cannot be read",
2309 patch->new_sha1_prefix, name);
2310 clear_image(img);
2311 img->buf = result;
2312 img->len = size;
2313 } else {
2314 /*
2315 * We have verified buf matches the preimage;
2316 * apply the patch data to it, which is stored
2317 * in the patch->fragments->{patch,size}.
2318 */
2319 if (apply_binary_fragment(img, patch))
2320 return error("binary patch does not apply to '%s'",
2321 name);
2323 /* verify that the result matches */
2324 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2325 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2326 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2327 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2328 }
2330 return 0;
2331 }
2333 static int apply_fragments(struct image *img, struct patch *patch)
2334 {
2335 struct fragment *frag = patch->fragments;
2336 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2337 unsigned ws_rule = patch->ws_rule;
2338 unsigned inaccurate_eof = patch->inaccurate_eof;
2340 if (patch->is_binary)
2341 return apply_binary(img, patch);
2343 while (frag) {
2344 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2345 error("patch failed: %s:%ld", name, frag->oldpos);
2346 if (!apply_with_reject)
2347 return -1;
2348 frag->rejected = 1;
2349 }
2350 frag = frag->next;
2351 }
2352 return 0;
2353 }
2355 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2356 {
2357 if (!ce)
2358 return 0;
2360 if (S_ISGITLINK(ce->ce_mode)) {
2361 strbuf_grow(buf, 100);
2362 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2363 } else {
2364 enum object_type type;
2365 unsigned long sz;
2366 char *result;
2368 result = read_sha1_file(ce->sha1, &type, &sz);
2369 if (!result)
2370 return -1;
2371 /* XXX read_sha1_file NUL-terminates */
2372 strbuf_attach(buf, result, sz, sz + 1);
2373 }
2374 return 0;
2375 }
2377 static struct patch *in_fn_table(const char *name)
2378 {
2379 struct string_list_item *item;
2381 if (name == NULL)
2382 return NULL;
2384 item = string_list_lookup(name, &fn_table);
2385 if (item != NULL)
2386 return (struct patch *)item->util;
2388 return NULL;
2389 }
2391 /*
2392 * item->util in the filename table records the status of the path.
2393 * Usually it points at a patch (whose result records the contents
2394 * of it after applying it), but it could be PATH_WAS_DELETED for a
2395 * path that a previously applied patch has already removed.
2396 */
2397 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2398 #define PATH_WAS_DELETED ((struct patch *) -1)
2400 static int to_be_deleted(struct patch *patch)
2401 {
2402 return patch == PATH_TO_BE_DELETED;
2403 }
2405 static int was_deleted(struct patch *patch)
2406 {
2407 return patch == PATH_WAS_DELETED;
2408 }
2410 static void add_to_fn_table(struct patch *patch)
2411 {
2412 struct string_list_item *item;
2414 /*
2415 * Always add new_name unless patch is a deletion
2416 * This should cover the cases for normal diffs,
2417 * file creations and copies
2418 */
2419 if (patch->new_name != NULL) {
2420 item = string_list_insert(patch->new_name, &fn_table);
2421 item->util = patch;
2422 }
2424 /*
2425 * store a failure on rename/deletion cases because
2426 * later chunks shouldn't patch old names
2427 */
2428 if ((patch->new_name == NULL) || (patch->is_rename)) {
2429 item = string_list_insert(patch->old_name, &fn_table);
2430 item->util = PATH_WAS_DELETED;
2431 }
2432 }
2434 static void prepare_fn_table(struct patch *patch)
2435 {
2436 /*
2437 * store information about incoming file deletion
2438 */
2439 while (patch) {
2440 if ((patch->new_name == NULL) || (patch->is_rename)) {
2441 struct string_list_item *item;
2442 item = string_list_insert(patch->old_name, &fn_table);
2443 item->util = PATH_TO_BE_DELETED;
2444 }
2445 patch = patch->next;
2446 }
2447 }
2449 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2450 {
2451 struct strbuf buf = STRBUF_INIT;
2452 struct image image;
2453 size_t len;
2454 char *img;
2455 struct patch *tpatch;
2457 if (!(patch->is_copy || patch->is_rename) &&
2458 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2459 if (was_deleted(tpatch)) {
2460 return error("patch %s has been renamed/deleted",
2461 patch->old_name);
2462 }
2463 /* We have a patched copy in memory use that */
2464 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2465 } else if (cached) {
2466 if (read_file_or_gitlink(ce, &buf))
2467 return error("read of %s failed", patch->old_name);
2468 } else if (patch->old_name) {
2469 if (S_ISGITLINK(patch->old_mode)) {
2470 if (ce) {
2471 read_file_or_gitlink(ce, &buf);
2472 } else {
2473 /*
2474 * There is no way to apply subproject
2475 * patch without looking at the index.
2476 */
2477 patch->fragments = NULL;
2478 }
2479 } else {
2480 if (read_old_data(st, patch->old_name, &buf))
2481 return error("read of %s failed", patch->old_name);
2482 }
2483 }
2485 img = strbuf_detach(&buf, &len);
2486 prepare_image(&image, img, len, !patch->is_binary);
2488 if (apply_fragments(&image, patch) < 0)
2489 return -1; /* note with --reject this succeeds. */
2490 patch->result = image.buf;
2491 patch->resultsize = image.len;
2492 add_to_fn_table(patch);
2493 free(image.line_allocated);
2495 if (0 < patch->is_delete && patch->resultsize)
2496 return error("removal patch leaves file contents");
2498 return 0;
2499 }
2501 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2502 {
2503 struct stat nst;
2504 if (!lstat(new_name, &nst)) {
2505 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2506 return 0;
2507 /*
2508 * A leading component of new_name might be a symlink
2509 * that is going to be removed with this patch, but
2510 * still pointing at somewhere that has the path.
2511 * In such a case, path "new_name" does not exist as
2512 * far as git is concerned.
2513 */
2514 if (has_symlink_leading_path(new_name, strlen(new_name)))
2515 return 0;
2517 return error("%s: already exists in working directory", new_name);
2518 }
2519 else if ((errno != ENOENT) && (errno != ENOTDIR))
2520 return error("%s: %s", new_name, strerror(errno));
2521 return 0;
2522 }
2524 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2525 {
2526 if (S_ISGITLINK(ce->ce_mode)) {
2527 if (!S_ISDIR(st->st_mode))
2528 return -1;
2529 return 0;
2530 }
2531 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2532 }
2534 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2535 {
2536 const char *old_name = patch->old_name;
2537 struct patch *tpatch = NULL;
2538 int stat_ret = 0;
2539 unsigned st_mode = 0;
2541 /*
2542 * Make sure that we do not have local modifications from the
2543 * index when we are looking at the index. Also make sure
2544 * we have the preimage file to be patched in the work tree,
2545 * unless --cached, which tells git to apply only in the index.
2546 */
2547 if (!old_name)
2548 return 0;
2550 assert(patch->is_new <= 0);
2552 if (!(patch->is_copy || patch->is_rename) &&
2553 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2554 if (was_deleted(tpatch))
2555 return error("%s: has been deleted/renamed", old_name);
2556 st_mode = tpatch->new_mode;
2557 } else if (!cached) {
2558 stat_ret = lstat(old_name, st);
2559 if (stat_ret && errno != ENOENT)
2560 return error("%s: %s", old_name, strerror(errno));
2561 }
2563 if (to_be_deleted(tpatch))
2564 tpatch = NULL;
2566 if (check_index && !tpatch) {
2567 int pos = cache_name_pos(old_name, strlen(old_name));
2568 if (pos < 0) {
2569 if (patch->is_new < 0)
2570 goto is_new;
2571 return error("%s: does not exist in index", old_name);
2572 }
2573 *ce = active_cache[pos];
2574 if (stat_ret < 0) {
2575 struct checkout costate;
2576 /* checkout */
2577 costate.base_dir = "";
2578 costate.base_dir_len = 0;
2579 costate.force = 0;
2580 costate.quiet = 0;
2581 costate.not_new = 0;
2582 costate.refresh_cache = 1;
2583 if (checkout_entry(*ce, &costate, NULL) ||
2584 lstat(old_name, st))
2585 return -1;
2586 }
2587 if (!cached && verify_index_match(*ce, st))
2588 return error("%s: does not match index", old_name);
2589 if (cached)
2590 st_mode = (*ce)->ce_mode;
2591 } else if (stat_ret < 0) {
2592 if (patch->is_new < 0)
2593 goto is_new;
2594 return error("%s: %s", old_name, strerror(errno));
2595 }
2597 if (!cached && !tpatch)
2598 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2600 if (patch->is_new < 0)
2601 patch->is_new = 0;
2602 if (!patch->old_mode)
2603 patch->old_mode = st_mode;
2604 if ((st_mode ^ patch->old_mode) & S_IFMT)
2605 return error("%s: wrong type", old_name);
2606 if (st_mode != patch->old_mode)
2607 warning("%s has type %o, expected %o",
2608 old_name, st_mode, patch->old_mode);
2609 if (!patch->new_mode && !patch->is_delete)
2610 patch->new_mode = st_mode;
2611 return 0;
2613 is_new:
2614 patch->is_new = 1;
2615 patch->is_delete = 0;
2616 patch->old_name = NULL;
2617 return 0;
2618 }
2620 static int check_patch(struct patch *patch)
2621 {
2622 struct stat st;
2623 const char *old_name = patch->old_name;
2624 const char *new_name = patch->new_name;
2625 const char *name = old_name ? old_name : new_name;
2626 struct cache_entry *ce = NULL;
2627 struct patch *tpatch;
2628 int ok_if_exists;
2629 int status;
2631 patch->rejected = 1; /* we will drop this after we succeed */
2633 status = check_preimage(patch, &ce, &st);
2634 if (status)
2635 return status;
2636 old_name = patch->old_name;
2638 if ((tpatch = in_fn_table(new_name)) &&
2639 (was_deleted(tpatch) || to_be_deleted(tpatch)))
2640 /*
2641 * A type-change diff is always split into a patch to
2642 * delete old, immediately followed by a patch to
2643 * create new (see diff.c::run_diff()); in such a case
2644 * it is Ok that the entry to be deleted by the
2645 * previous patch is still in the working tree and in
2646 * the index.
2647 */
2648 ok_if_exists = 1;
2649 else
2650 ok_if_exists = 0;
2652 if (new_name &&
2653 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2654 if (check_index &&
2655 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2656 !ok_if_exists)
2657 return error("%s: already exists in index", new_name);
2658 if (!cached) {
2659 int err = check_to_create_blob(new_name, ok_if_exists);
2660 if (err)
2661 return err;
2662 }
2663 if (!patch->new_mode) {
2664 if (0 < patch->is_new)
2665 patch->new_mode = S_IFREG | 0644;
2666 else
2667 patch->new_mode = patch->old_mode;
2668 }
2669 }
2671 if (new_name && old_name) {
2672 int same = !strcmp(old_name, new_name);
2673 if (!patch->new_mode)
2674 patch->new_mode = patch->old_mode;
2675 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2676 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2677 patch->new_mode, new_name, patch->old_mode,
2678 same ? "" : " of ", same ? "" : old_name);
2679 }
2681 if (apply_data(patch, &st, ce) < 0)
2682 return error("%s: patch does not apply", name);
2683 patch->rejected = 0;
2684 return 0;
2685 }
2687 static int check_patch_list(struct patch *patch)
2688 {
2689 int err = 0;
2691 prepare_fn_table(patch);
2692 while (patch) {
2693 if (apply_verbosely)
2694 say_patch_name(stderr,
2695 "Checking patch ", patch, "...\n");
2696 err |= check_patch(patch);
2697 patch = patch->next;
2698 }
2699 return err;
2700 }
2702 /* This function tries to read the sha1 from the current index */
2703 static int get_current_sha1(const char *path, unsigned char *sha1)
2704 {
2705 int pos;
2707 if (read_cache() < 0)
2708 return -1;
2709 pos = cache_name_pos(path, strlen(path));
2710 if (pos < 0)
2711 return -1;
2712 hashcpy(sha1, active_cache[pos]->sha1);
2713 return 0;
2714 }
2716 /* Build an index that contains the just the files needed for a 3way merge */
2717 static void build_fake_ancestor(struct patch *list, const char *filename)
2718 {
2719 struct patch *patch;
2720 struct index_state result = { NULL };
2721 int fd;
2723 /* Once we start supporting the reverse patch, it may be
2724 * worth showing the new sha1 prefix, but until then...
2725 */
2726 for (patch = list; patch; patch = patch->next) {
2727 const unsigned char *sha1_ptr;
2728 unsigned char sha1[20];
2729 struct cache_entry *ce;
2730 const char *name;
2732 name = patch->old_name ? patch->old_name : patch->new_name;
2733 if (0 < patch->is_new)
2734 continue;
2735 else if (get_sha1(patch->old_sha1_prefix, sha1))
2736 /* git diff has no index line for mode/type changes */
2737 if (!patch->lines_added && !patch->lines_deleted) {
2738 if (get_current_sha1(patch->new_name, sha1) ||
2739 get_current_sha1(patch->old_name, sha1))
2740 die("mode change for %s, which is not "
2741 "in current HEAD", name);
2742 sha1_ptr = sha1;
2743 } else
2744 die("sha1 information is lacking or useless "
2745 "(%s).", name);
2746 else
2747 sha1_ptr = sha1;
2749 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2750 if (!ce)
2751 die("make_cache_entry failed for path '%s'", name);
2752 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2753 die ("Could not add %s to temporary index", name);
2754 }
2756 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2757 if (fd < 0 || write_index(&result, fd) || close(fd))
2758 die ("Could not write temporary index to %s", filename);
2760 discard_index(&result);
2761 }
2763 static void stat_patch_list(struct patch *patch)
2764 {
2765 int files, adds, dels;
2767 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2768 files++;
2769 adds += patch->lines_added;
2770 dels += patch->lines_deleted;
2771 show_stats(patch);
2772 }
2774 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2775 }
2777 static void numstat_patch_list(struct patch *patch)
2778 {
2779 for ( ; patch; patch = patch->next) {
2780 const char *name;
2781 name = patch->new_name ? patch->new_name : patch->old_name;
2782 if (patch->is_binary)
2783 printf("-\t-\t");
2784 else
2785 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2786 write_name_quoted(name, stdout, line_termination);
2787 }
2788 }
2790 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2791 {
2792 if (mode)
2793 printf(" %s mode %06o %s\n", newdelete, mode, name);
2794 else
2795 printf(" %s %s\n", newdelete, name);
2796 }
2798 static void show_mode_change(struct patch *p, int show_name)
2799 {
2800 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2801 if (show_name)
2802 printf(" mode change %06o => %06o %s\n",
2803 p->old_mode, p->new_mode, p->new_name);
2804 else
2805 printf(" mode change %06o => %06o\n",
2806 p->old_mode, p->new_mode);
2807 }
2808 }
2810 static void show_rename_copy(struct patch *p)
2811 {
2812 const char *renamecopy = p->is_rename ? "rename" : "copy";
2813 const char *old, *new;
2815 /* Find common prefix */
2816 old = p->old_name;
2817 new = p->new_name;
2818 while (1) {
2819 const char *slash_old, *slash_new;
2820 slash_old = strchr(old, '/');
2821 slash_new = strchr(new, '/');
2822 if (!slash_old ||
2823 !slash_new ||
2824 slash_old - old != slash_new - new ||
2825 memcmp(old, new, slash_new - new))
2826 break;
2827 old = slash_old + 1;
2828 new = slash_new + 1;
2829 }
2830 /* p->old_name thru old is the common prefix, and old and new
2831 * through the end of names are renames
2832 */
2833 if (old != p->old_name)
2834 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2835 (int)(old - p->old_name), p->old_name,
2836 old, new, p->score);
2837 else
2838 printf(" %s %s => %s (%d%%)\n", renamecopy,
2839 p->old_name, p->new_name, p->score);
2840 show_mode_change(p, 0);
2841 }
2843 static void summary_patch_list(struct patch *patch)
2844 {
2845 struct patch *p;
2847 for (p = patch; p; p = p->next) {
2848 if (p->is_new)
2849 show_file_mode_name("create", p->new_mode, p->new_name);
2850 else if (p->is_delete)
2851 show_file_mode_name("delete", p->old_mode, p->old_name);
2852 else {
2853 if (p->is_rename || p->is_copy)
2854 show_rename_copy(p);
2855 else {
2856 if (p->score) {
2857 printf(" rewrite %s (%d%%)\n",
2858 p->new_name, p->score);
2859 show_mode_change(p, 0);
2860 }
2861 else
2862 show_mode_change(p, 1);
2863 }
2864 }
2865 }
2866 }
2868 static void patch_stats(struct patch *patch)
2869 {
2870 int lines = patch->lines_added + patch->lines_deleted;
2872 if (lines > max_change)
2873 max_change = lines;
2874 if (patch->old_name) {
2875 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2876 if (!len)
2877 len = strlen(patch->old_name);
2878 if (len > max_len)
2879 max_len = len;
2880 }
2881 if (patch->new_name) {
2882 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2883 if (!len)
2884 len = strlen(patch->new_name);
2885 if (len > max_len)
2886 max_len = len;
2887 }
2888 }
2890 static void remove_file(struct patch *patch, int rmdir_empty)
2891 {
2892 if (update_index) {
2893 if (remove_file_from_cache(patch->old_name) < 0)
2894 die("unable to remove %s from index", patch->old_name);
2895 }
2896 if (!cached) {
2897 if (S_ISGITLINK(patch->old_mode)) {
2898 if (rmdir(patch->old_name))
2899 warning("unable to remove submodule %s",
2900 patch->old_name);
2901 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
2902 remove_path(patch->old_name);
2903 }
2904 }
2905 }
2907 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2908 {
2909 struct stat st;
2910 struct cache_entry *ce;
2911 int namelen = strlen(path);
2912 unsigned ce_size = cache_entry_size(namelen);
2914 if (!update_index)
2915 return;
2917 ce = xcalloc(1, ce_size);
2918 memcpy(ce->name, path, namelen);
2919 ce->ce_mode = create_ce_mode(mode);
2920 ce->ce_flags = namelen;
2921 if (S_ISGITLINK(mode)) {
2922 const char *s = buf;
2924 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
2925 die("corrupt patch for subproject %s", path);
2926 } else {
2927 if (!cached) {
2928 if (lstat(path, &st) < 0)
2929 die_errno("unable to stat newly created file '%s'",
2930 path);
2931 fill_stat_cache_info(ce, &st);
2932 }
2933 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2934 die("unable to create backing store for newly created file %s", path);
2935 }
2936 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2937 die("unable to add cache entry for %s", path);
2938 }
2940 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2941 {
2942 int fd;
2943 struct strbuf nbuf = STRBUF_INIT;
2945 if (S_ISGITLINK(mode)) {
2946 struct stat st;
2947 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
2948 return 0;
2949 return mkdir(path, 0777);
2950 }
2952 if (has_symlinks && S_ISLNK(mode))
2953 /* Although buf:size is counted string, it also is NUL
2954 * terminated.
2955 */
2956 return symlink(buf, path);
2958 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2959 if (fd < 0)
2960 return -1;
2962 if (convert_to_working_tree(path, buf, size, &nbuf)) {
2963 size = nbuf.len;
2964 buf = nbuf.buf;
2965 }
2966 write_or_die(fd, buf, size);
2967 strbuf_release(&nbuf);
2969 if (close(fd) < 0)
2970 die_errno("closing file '%s'", path);
2971 return 0;
2972 }
2974 /*
2975 * We optimistically assume that the directories exist,
2976 * which is true 99% of the time anyway. If they don't,
2977 * we create them and try again.
2978 */
2979 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2980 {
2981 if (cached)
2982 return;
2983 if (!try_create_file(path, mode, buf, size))
2984 return;
2986 if (errno == ENOENT) {
2987 if (safe_create_leading_directories(path))
2988 return;
2989 if (!try_create_file(path, mode, buf, size))
2990 return;
2991 }
2993 if (errno == EEXIST || errno == EACCES) {
2994 /* We may be trying to create a file where a directory
2995 * used to be.
2996 */
2997 struct stat st;
2998 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
2999 errno = EEXIST;
3000 }
3002 if (errno == EEXIST) {
3003 unsigned int nr = getpid();
3005 for (;;) {
3006 char newpath[PATH_MAX];
3007 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3008 if (!try_create_file(newpath, mode, buf, size)) {
3009 if (!rename(newpath, path))
3010 return;
3011 unlink_or_warn(newpath);
3012 break;
3013 }
3014 if (errno != EEXIST)
3015 break;
3016 ++nr;
3017 }
3018 }
3019 die_errno("unable to write file '%s' mode %o", path, mode);
3020 }
3022 static void create_file(struct patch *patch)
3023 {
3024 char *path = patch->new_name;
3025 unsigned mode = patch->new_mode;
3026 unsigned long size = patch->resultsize;
3027 char *buf = patch->result;
3029 if (!mode)
3030 mode = S_IFREG | 0644;
3031 create_one_file(path, mode, buf, size);
3032 add_index_file(path, mode, buf, size);
3033 }
3035 /* phase zero is to remove, phase one is to create */
3036 static void write_out_one_result(struct patch *patch, int phase)
3037 {
3038 if (patch->is_delete > 0) {
3039 if (phase == 0)
3040 remove_file(patch, 1);
3041 return;
3042 }
3043 if (patch->is_new > 0 || patch->is_copy) {
3044 if (phase == 1)
3045 create_file(patch);
3046 return;
3047 }
3048 /*
3049 * Rename or modification boils down to the same
3050 * thing: remove the old, write the new
3051 */
3052 if (phase == 0)
3053 remove_file(patch, patch->is_rename);
3054 if (phase == 1)
3055 create_file(patch);
3056 }
3058 static int write_out_one_reject(struct patch *patch)
3059 {
3060 FILE *rej;
3061 char namebuf[PATH_MAX];
3062 struct fragment *frag;
3063 int cnt = 0;
3065 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3066 if (!frag->rejected)
3067 continue;
3068 cnt++;
3069 }
3071 if (!cnt) {
3072 if (apply_verbosely)
3073 say_patch_name(stderr,
3074 "Applied patch ", patch, " cleanly.\n");
3075 return 0;
3076 }
3078 /* This should not happen, because a removal patch that leaves
3079 * contents are marked "rejected" at the patch level.
3080 */
3081 if (!patch->new_name)
3082 die("internal error");
3084 /* Say this even without --verbose */
3085 say_patch_name(stderr, "Applying patch ", patch, " with");
3086 fprintf(stderr, " %d rejects...\n", cnt);
3088 cnt = strlen(patch->new_name);
3089 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3090 cnt = ARRAY_SIZE(namebuf) - 5;
3091 warning("truncating .rej filename to %.*s.rej",
3092 cnt - 1, patch->new_name);
3093 }
3094 memcpy(namebuf, patch->new_name, cnt);
3095 memcpy(namebuf + cnt, ".rej", 5);
3097 rej = fopen(namebuf, "w");
3098 if (!rej)
3099 return error("cannot open %s: %s", namebuf, strerror(errno));
3101 /* Normal git tools never deal with .rej, so do not pretend
3102 * this is a git patch by saying --git nor give extended
3103 * headers. While at it, maybe please "kompare" that wants
3104 * the trailing TAB and some garbage at the end of line ;-).
3105 */
3106 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3107 patch->new_name, patch->new_name);
3108 for (cnt = 1, frag = patch->fragments;
3109 frag;
3110 cnt++, frag = frag->next) {
3111 if (!frag->rejected) {
3112 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3113 continue;
3114 }
3115 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3116 fprintf(rej, "%.*s", frag->size, frag->patch);
3117 if (frag->patch[frag->size-1] != '\n')
3118 fputc('\n', rej);
3119 }
3120 fclose(rej);
3121 return -1;
3122 }
3124 static int write_out_results(struct patch *list, int skipped_patch)
3125 {
3126 int phase;
3127 int errs = 0;
3128 struct patch *l;
3130 if (!list && !skipped_patch)
3131 return error("No changes");
3133 for (phase = 0; phase < 2; phase++) {
3134 l = list;
3135 while (l) {
3136 if (l->rejected)
3137 errs = 1;
3138 else {
3139 write_out_one_result(l, phase);
3140 if (phase == 1 && write_out_one_reject(l))
3141 errs = 1;
3142 }
3143 l = l->next;
3144 }
3145 }
3146 return errs;
3147 }
3149 static struct lock_file lock_file;
3151 static struct string_list limit_by_name;
3152 static int has_include;
3153 static void add_name_limit(const char *name, int exclude)
3154 {
3155 struct string_list_item *it;
3157 it = string_list_append(name, &limit_by_name);
3158 it->util = exclude ? NULL : (void *) 1;
3159 }
3161 static int use_patch(struct patch *p)
3162 {
3163 const char *pathname = p->new_name ? p->new_name : p->old_name;
3164 int i;
3166 /* Paths outside are not touched regardless of "--include" */
3167 if (0 < prefix_length) {
3168 int pathlen = strlen(pathname);
3169 if (pathlen <= prefix_length ||
3170 memcmp(prefix, pathname, prefix_length))
3171 return 0;
3172 }
3174 /* See if it matches any of exclude/include rule */
3175 for (i = 0; i < limit_by_name.nr; i++) {
3176 struct string_list_item *it = &limit_by_name.items[i];
3177 if (!fnmatch(it->string, pathname, 0))
3178 return (it->util != NULL);
3179 }
3181 /*
3182 * If we had any include, a path that does not match any rule is
3183 * not used. Otherwise, we saw bunch of exclude rules (or none)
3184 * and such a path is used.
3185 */
3186 return !has_include;
3187 }
3190 static void prefix_one(char **name)
3191 {
3192 char *old_name = *name;
3193 if (!old_name)
3194 return;
3195 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3196 free(old_name);
3197 }
3199 static void prefix_patches(struct patch *p)
3200 {
3201 if (!prefix || p->is_toplevel_relative)
3202 return;
3203 for ( ; p; p = p->next) {
3204 if (p->new_name == p->old_name) {
3205 char *prefixed = p->new_name;
3206 prefix_one(&prefixed);
3207 p->new_name = p->old_name = prefixed;
3208 }
3209 else {
3210 prefix_one(&p->new_name);
3211 prefix_one(&p->old_name);
3212 }
3213 }
3214 }
3216 #define INACCURATE_EOF (1<<0)
3217 #define RECOUNT (1<<1)
3219 static int apply_patch(int fd, const char *filename, int options)
3220 {
3221 size_t offset;
3222 struct strbuf buf = STRBUF_INIT;
3223 struct patch *list = NULL, **listp = &list;
3224 int skipped_patch = 0;
3226 /* FIXME - memory leak when using multiple patch files as inputs */
3227 memset(&fn_table, 0, sizeof(struct string_list));
3228 patch_input_file = filename;
3229 read_patch_file(&buf, fd);
3230 offset = 0;
3231 while (offset < buf.len) {
3232 struct patch *patch;
3233 int nr;
3235 patch = xcalloc(1, sizeof(*patch));
3236 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3237 patch->recount = !!(options & RECOUNT);
3238 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3239 if (nr < 0)
3240 break;
3241 if (apply_in_reverse)
3242 reverse_patches(patch);
3243 if (prefix)
3244 prefix_patches(patch);
3245 if (use_patch(patch)) {
3246 patch_stats(patch);
3247 *listp = patch;
3248 listp = &patch->next;
3249 }
3250 else {
3251 /* perhaps free it a bit better? */
3252 free(patch);
3253 skipped_patch++;
3254 }
3255 offset += nr;
3256 }
3258 if (whitespace_error && (ws_error_action == die_on_ws_error))
3259 apply = 0;
3261 update_index = check_index && apply;
3262 if (update_index && newfd < 0)
3263 newfd = hold_locked_index(&lock_file, 1);
3265 if (check_index) {
3266 if (read_cache() < 0)
3267 die("unable to read index file");
3268 }
3270 if ((check || apply) &&
3271 check_patch_list(list) < 0 &&
3272 !apply_with_reject)
3273 exit(1);
3275 if (apply && write_out_results(list, skipped_patch))
3276 exit(1);
3278 if (fake_ancestor)
3279 build_fake_ancestor(list, fake_ancestor);
3281 if (diffstat)
3282 stat_patch_list(list);
3284 if (numstat)
3285 numstat_patch_list(list);
3287 if (summary)
3288 summary_patch_list(list);
3290 strbuf_release(&buf);
3291 return 0;
3292 }
3294 static int git_apply_config(const char *var, const char *value, void *cb)
3295 {
3296 if (!strcmp(var, "apply.whitespace"))
3297 return git_config_string(&apply_default_whitespace, var, value);
3298 return git_default_config(var, value, cb);
3299 }
3301 static int option_parse_exclude(const struct option *opt,
3302 const char *arg, int unset)
3303 {
3304 add_name_limit(arg, 1);
3305 return 0;
3306 }
3308 static int option_parse_include(const struct option *opt,
3309 const char *arg, int unset)
3310 {
3311 add_name_limit(arg, 0);
3312 has_include = 1;
3313 return 0;
3314 }
3316 static int option_parse_p(const struct option *opt,
3317 const char *arg, int unset)
3318 {
3319 p_value = atoi(arg);
3320 p_value_known = 1;
3321 return 0;
3322 }
3324 static int option_parse_z(const struct option *opt,
3325 const char *arg, int unset)
3326 {
3327 if (unset)
3328 line_termination = '\n';
3329 else
3330 line_termination = 0;
3331 return 0;
3332 }
3334 static int option_parse_whitespace(const struct option *opt,
3335 const char *arg, int unset)
3336 {
3337 const char **whitespace_option = opt->value;
3339 *whitespace_option = arg;
3340 parse_whitespace_option(arg);
3341 return 0;
3342 }
3344 static int option_parse_directory(const struct option *opt,
3345 const char *arg, int unset)
3346 {
3347 root_len = strlen(arg);
3348 if (root_len && arg[root_len - 1] != '/') {
3349 char *new_root;
3350 root = new_root = xmalloc(root_len + 2);
3351 strcpy(new_root, arg);
3352 strcpy(new_root + root_len++, "/");
3353 } else
3354 root = arg;
3355 return 0;
3356 }
3358 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3359 {
3360 int i;
3361 int errs = 0;
3362 int is_not_gitdir;
3363 int binary;
3364 int force_apply = 0;
3366 const char *whitespace_option = NULL;
3368 struct option builtin_apply_options[] = {
3369 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3370 "don't apply changes matching the given path",
3371 0, option_parse_exclude },
3372 { OPTION_CALLBACK, 0, "include", NULL, "path",
3373 "apply changes matching the given path",
3374 0, option_parse_include },
3375 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3376 "remove <num> leading slashes from traditional diff paths",
3377 0, option_parse_p },
3378 OPT_BOOLEAN(0, "no-add", &no_add,
3379 "ignore additions made by the patch"),
3380 OPT_BOOLEAN(0, "stat", &diffstat,
3381 "instead of applying the patch, output diffstat for the input"),
3382 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3383 NULL, "old option, now no-op",
3384 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3385 { OPTION_BOOLEAN, 0, "binary", &binary,
3386 NULL, "old option, now no-op",
3387 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3388 OPT_BOOLEAN(0, "numstat", &numstat,
3389 "shows number of added and deleted lines in decimal notation"),
3390 OPT_BOOLEAN(0, "summary", &summary,
3391 "instead of applying the patch, output a summary for the input"),
3392 OPT_BOOLEAN(0, "check", &check,
3393 "instead of applying the patch, see if the patch is applicable"),
3394 OPT_BOOLEAN(0, "index", &check_index,
3395 "make sure the patch is applicable to the current index"),
3396 OPT_BOOLEAN(0, "cached", &cached,
3397 "apply a patch without touching the working tree"),
3398 OPT_BOOLEAN(0, "apply", &force_apply,
3399 "also apply the patch (use with --stat/--summary/--check)"),
3400 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3401 "build a temporary index based on embedded index information"),
3402 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3403 "paths are separated with NUL character",
3404 PARSE_OPT_NOARG, option_parse_z },
3405 OPT_INTEGER('C', NULL, &p_context,
3406 "ensure at least <n> lines of context match"),
3407 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3408 "detect new or modified lines that have whitespace errors",
3409 0, option_parse_whitespace },
3410 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3411 "apply the patch in reverse"),
3412 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3413 "don't expect at least one line of context"),
3414 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3415 "leave the rejected hunks in corresponding *.rej files"),
3416 OPT__VERBOSE(&apply_verbosely),
3417 OPT_BIT(0, "inaccurate-eof", &options,
3418 "tolerate incorrectly detected missing new-line at the end of file",
3419 INACCURATE_EOF),
3420 OPT_BIT(0, "recount", &options,
3421 "do not trust the line counts in the hunk headers",
3422 RECOUNT),
3423 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3424 "prepend <root> to all filenames",
3425 0, option_parse_directory },
3426 OPT_END()
3427 };
3429 prefix = setup_git_directory_gently(&is_not_gitdir);
3430 prefix_length = prefix ? strlen(prefix) : 0;
3431 git_config(git_apply_config, NULL);
3432 if (apply_default_whitespace)
3433 parse_whitespace_option(apply_default_whitespace);
3435 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3436 apply_usage, 0);
3438 if (apply_with_reject)
3439 apply = apply_verbosely = 1;
3440 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3441 apply = 0;
3442 if (check_index && is_not_gitdir)
3443 die("--index outside a repository");
3444 if (cached) {
3445 if (is_not_gitdir)
3446 die("--cached outside a repository");
3447 check_index = 1;
3448 }
3449 for (i = 0; i < argc; i++) {
3450 const char *arg = argv[i];
3451 int fd;
3453 if (!strcmp(arg, "-")) {
3454 errs |= apply_patch(0, "<stdin>", options);
3455 read_stdin = 0;
3456 continue;
3457 } else if (0 < prefix_length)
3458 arg = prefix_filename(prefix, prefix_length, arg);
3460 fd = open(arg, O_RDONLY);
3461 if (fd < 0)
3462 die_errno("can't open patch '%s'", arg);
3463 read_stdin = 0;
3464 set_default_whitespace_mode(whitespace_option);
3465 errs |= apply_patch(fd, arg, options);
3466 close(fd);
3467 }
3468 set_default_whitespace_mode(whitespace_option);
3469 if (read_stdin)
3470 errs |= apply_patch(0, "<stdin>", options);
3471 if (whitespace_error) {
3472 if (squelch_whitespace_errors &&
3473 squelch_whitespace_errors < whitespace_error) {
3474 int squelched =
3475 whitespace_error - squelch_whitespace_errors;
3476 warning("squelched %d "
3477 "whitespace error%s",
3478 squelched,
3479 squelched == 1 ? "" : "s");
3480 }
3481 if (ws_error_action == die_on_ws_error)
3482 die("%d line%s add%s whitespace errors.",
3483 whitespace_error,
3484 whitespace_error == 1 ? "" : "s",
3485 whitespace_error == 1 ? "s" : "");
3486 if (applied_after_fixing_ws && apply)
3487 warning("%d line%s applied after"
3488 " fixing whitespace errors.",
3489 applied_after_fixing_ws,
3490 applied_after_fixing_ws == 1 ? "" : "s");
3491 else if (whitespace_error)
3492 warning("%d line%s add%s whitespace errors.",
3493 whitespace_error,
3494 whitespace_error == 1 ? "" : "s",
3495 whitespace_error == 1 ? "s" : "");
3496 }
3498 if (update_index) {
3499 if (write_cache(newfd, active_cache, active_nr) ||
3500 commit_locked_index(&lock_file))
3501 die("Unable to write new index file");
3502 }
3504 return !!errs;
3505 }