Merge branch 'jc/maint-blank-at-eof'
[tgit.git] / builtin-apply.c
blobf667368d161609ea3c93fc6ebe2452b9cd0e557d
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
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"
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.
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
55 static enum ws_error_action {
56 nowarn_ws_error,
57 warn_on_ws_error,
58 die_on_ws_error,
59 correct_ws_error,
60 } ws_error_action = warn_on_ws_error;
61 static int whitespace_error;
62 static int squelch_whitespace_errors = 5;
63 static int applied_after_fixing_ws;
65 static enum ws_ignore {
66 ignore_ws_none,
67 ignore_ws_change,
68 } ws_ignore_action = ignore_ws_none;
71 static const char *patch_input_file;
72 static const char *root;
73 static int root_len;
74 static int read_stdin = 1;
75 static int options;
77 static void parse_whitespace_option(const char *option)
79 if (!option) {
80 ws_error_action = warn_on_ws_error;
81 return;
83 if (!strcmp(option, "warn")) {
84 ws_error_action = warn_on_ws_error;
85 return;
87 if (!strcmp(option, "nowarn")) {
88 ws_error_action = nowarn_ws_error;
89 return;
91 if (!strcmp(option, "error")) {
92 ws_error_action = die_on_ws_error;
93 return;
95 if (!strcmp(option, "error-all")) {
96 ws_error_action = die_on_ws_error;
97 squelch_whitespace_errors = 0;
98 return;
100 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
101 ws_error_action = correct_ws_error;
102 return;
104 die("unrecognized whitespace option '%s'", option);
107 static void parse_ignorewhitespace_option(const char *option)
109 if (!option || !strcmp(option, "no") ||
110 !strcmp(option, "false") || !strcmp(option, "never") ||
111 !strcmp(option, "none")) {
112 ws_ignore_action = ignore_ws_none;
113 return;
115 if (!strcmp(option, "change")) {
116 ws_ignore_action = ignore_ws_change;
117 return;
119 die("unrecognized whitespace ignore option '%s'", option);
122 static void set_default_whitespace_mode(const char *whitespace_option)
124 if (!whitespace_option && !apply_default_whitespace)
125 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
129 * For "diff-stat" like behaviour, we keep track of the biggest change
130 * we've seen, and the longest filename. That allows us to do simple
131 * scaling.
133 static int max_change, max_len;
136 * Various "current state", notably line numbers and what
137 * file (and how) we're patching right now.. The "is_xxxx"
138 * things are flags, where -1 means "don't know yet".
140 static int linenr = 1;
143 * This represents one "hunk" from a patch, starting with
144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
145 * patch text is pointed at by patch, and its byte length
146 * is stored in size. leading and trailing are the number
147 * of context lines.
149 struct fragment {
150 unsigned long leading, trailing;
151 unsigned long oldpos, oldlines;
152 unsigned long newpos, newlines;
153 const char *patch;
154 int size;
155 int rejected;
156 int linenr;
157 struct fragment *next;
161 * When dealing with a binary patch, we reuse "leading" field
162 * to store the type of the binary hunk, either deflated "delta"
163 * or deflated "literal".
165 #define binary_patch_method leading
166 #define BINARY_DELTA_DEFLATED 1
167 #define BINARY_LITERAL_DEFLATED 2
170 * This represents a "patch" to a file, both metainfo changes
171 * such as creation/deletion, filemode and content changes represented
172 * as a series of fragments.
174 struct patch {
175 char *new_name, *old_name, *def_name;
176 unsigned int old_mode, new_mode;
177 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
178 int rejected;
179 unsigned ws_rule;
180 unsigned long deflate_origlen;
181 int lines_added, lines_deleted;
182 int score;
183 unsigned int is_toplevel_relative:1;
184 unsigned int inaccurate_eof:1;
185 unsigned int is_binary:1;
186 unsigned int is_copy:1;
187 unsigned int is_rename:1;
188 unsigned int recount:1;
189 struct fragment *fragments;
190 char *result;
191 size_t resultsize;
192 char old_sha1_prefix[41];
193 char new_sha1_prefix[41];
194 struct patch *next;
198 * A line in a file, len-bytes long (includes the terminating LF,
199 * except for an incomplete line at the end if the file ends with
200 * one), and its contents hashes to 'hash'.
202 struct line {
203 size_t len;
204 unsigned hash : 24;
205 unsigned flag : 8;
206 #define LINE_COMMON 1
210 * This represents a "file", which is an array of "lines".
212 struct image {
213 char *buf;
214 size_t len;
215 size_t nr;
216 size_t alloc;
217 struct line *line_allocated;
218 struct line *line;
222 * Records filenames that have been touched, in order to handle
223 * the case where more than one patches touch the same file.
226 static struct string_list fn_table;
228 static uint32_t hash_line(const char *cp, size_t len)
230 size_t i;
231 uint32_t h;
232 for (i = 0, h = 0; i < len; i++) {
233 if (!isspace(cp[i])) {
234 h = h * 3 + (cp[i] & 0xff);
237 return h;
241 * Compare lines s1 of length n1 and s2 of length n2, ignoring
242 * whitespace difference. Returns 1 if they match, 0 otherwise
244 static int fuzzy_matchlines(const char *s1, size_t n1,
245 const char *s2, size_t n2)
247 const char *last1 = s1 + n1 - 1;
248 const char *last2 = s2 + n2 - 1;
249 int result = 0;
251 if (n1 < 0 || n2 < 0)
252 return 0;
254 /* ignore line endings */
255 while ((*last1 == '\r') || (*last1 == '\n'))
256 last1--;
257 while ((*last2 == '\r') || (*last2 == '\n'))
258 last2--;
260 /* skip leading whitespace */
261 while (isspace(*s1) && (s1 <= last1))
262 s1++;
263 while (isspace(*s2) && (s2 <= last2))
264 s2++;
265 /* early return if both lines are empty */
266 if ((s1 > last1) && (s2 > last2))
267 return 1;
268 while (!result) {
269 result = *s1++ - *s2++;
271 * Skip whitespace inside. We check for whitespace on
272 * both buffers because we don't want "a b" to match
273 * "ab"
275 if (isspace(*s1) && isspace(*s2)) {
276 while (isspace(*s1) && s1 <= last1)
277 s1++;
278 while (isspace(*s2) && s2 <= last2)
279 s2++;
282 * If we reached the end on one side only,
283 * lines don't match
285 if (
286 ((s2 > last2) && (s1 <= last1)) ||
287 ((s1 > last1) && (s2 <= last2)))
288 return 0;
289 if ((s1 > last1) && (s2 > last2))
290 break;
293 return !result;
296 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
298 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
299 img->line_allocated[img->nr].len = len;
300 img->line_allocated[img->nr].hash = hash_line(bol, len);
301 img->line_allocated[img->nr].flag = flag;
302 img->nr++;
305 static void prepare_image(struct image *image, char *buf, size_t len,
306 int prepare_linetable)
308 const char *cp, *ep;
310 memset(image, 0, sizeof(*image));
311 image->buf = buf;
312 image->len = len;
314 if (!prepare_linetable)
315 return;
317 ep = image->buf + image->len;
318 cp = image->buf;
319 while (cp < ep) {
320 const char *next;
321 for (next = cp; next < ep && *next != '\n'; next++)
323 if (next < ep)
324 next++;
325 add_line_info(image, cp, next - cp, 0);
326 cp = next;
328 image->line = image->line_allocated;
331 static void clear_image(struct image *image)
333 free(image->buf);
334 image->buf = NULL;
335 image->len = 0;
338 static void say_patch_name(FILE *output, const char *pre,
339 struct patch *patch, const char *post)
341 fputs(pre, output);
342 if (patch->old_name && patch->new_name &&
343 strcmp(patch->old_name, patch->new_name)) {
344 quote_c_style(patch->old_name, NULL, output, 0);
345 fputs(" => ", output);
346 quote_c_style(patch->new_name, NULL, output, 0);
347 } else {
348 const char *n = patch->new_name;
349 if (!n)
350 n = patch->old_name;
351 quote_c_style(n, NULL, output, 0);
353 fputs(post, output);
356 #define CHUNKSIZE (8192)
357 #define SLOP (16)
359 static void read_patch_file(struct strbuf *sb, int fd)
361 if (strbuf_read(sb, fd, 0) < 0)
362 die_errno("git apply: failed to read");
365 * Make sure that we have some slop in the buffer
366 * so that we can do speculative "memcmp" etc, and
367 * see to it that it is NUL-filled.
369 strbuf_grow(sb, SLOP);
370 memset(sb->buf + sb->len, 0, SLOP);
373 static unsigned long linelen(const char *buffer, unsigned long size)
375 unsigned long len = 0;
376 while (size--) {
377 len++;
378 if (*buffer++ == '\n')
379 break;
381 return len;
384 static int is_dev_null(const char *str)
386 return !memcmp("/dev/null", str, 9) && isspace(str[9]);
389 #define TERM_SPACE 1
390 #define TERM_TAB 2
392 static int name_terminate(const char *name, int namelen, int c, int terminate)
394 if (c == ' ' && !(terminate & TERM_SPACE))
395 return 0;
396 if (c == '\t' && !(terminate & TERM_TAB))
397 return 0;
399 return 1;
402 /* remove double slashes to make --index work with such filenames */
403 static char *squash_slash(char *name)
405 int i = 0, j = 0;
407 while (name[i]) {
408 if ((name[j++] = name[i++]) == '/')
409 while (name[i] == '/')
410 i++;
412 name[j] = '\0';
413 return name;
416 static char *find_name(const char *line, char *def, int p_value, int terminate)
418 int len;
419 const char *start = line;
421 if (*line == '"') {
422 struct strbuf name = STRBUF_INIT;
425 * Proposed "new-style" GNU patch/diff format; see
426 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
428 if (!unquote_c_style(&name, line, NULL)) {
429 char *cp;
431 for (cp = name.buf; p_value; p_value--) {
432 cp = strchr(cp, '/');
433 if (!cp)
434 break;
435 cp++;
437 if (cp) {
438 /* name can later be freed, so we need
439 * to memmove, not just return cp
441 strbuf_remove(&name, 0, cp - name.buf);
442 free(def);
443 if (root)
444 strbuf_insert(&name, 0, root, root_len);
445 return squash_slash(strbuf_detach(&name, NULL));
448 strbuf_release(&name);
451 for (;;) {
452 char c = *line;
454 if (isspace(c)) {
455 if (c == '\n')
456 break;
457 if (name_terminate(start, line-start, c, terminate))
458 break;
460 line++;
461 if (c == '/' && !--p_value)
462 start = line;
464 if (!start)
465 return squash_slash(def);
466 len = line - start;
467 if (!len)
468 return squash_slash(def);
471 * Generally we prefer the shorter name, especially
472 * if the other one is just a variation of that with
473 * something else tacked on to the end (ie "file.orig"
474 * or "file~").
476 if (def) {
477 int deflen = strlen(def);
478 if (deflen < len && !strncmp(start, def, deflen))
479 return squash_slash(def);
480 free(def);
483 if (root) {
484 char *ret = xmalloc(root_len + len + 1);
485 strcpy(ret, root);
486 memcpy(ret + root_len, start, len);
487 ret[root_len + len] = '\0';
488 return squash_slash(ret);
491 return squash_slash(xmemdupz(start, len));
494 static int count_slashes(const char *cp)
496 int cnt = 0;
497 char ch;
499 while ((ch = *cp++))
500 if (ch == '/')
501 cnt++;
502 return cnt;
506 * Given the string after "--- " or "+++ ", guess the appropriate
507 * p_value for the given patch.
509 static int guess_p_value(const char *nameline)
511 char *name, *cp;
512 int val = -1;
514 if (is_dev_null(nameline))
515 return -1;
516 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB);
517 if (!name)
518 return -1;
519 cp = strchr(name, '/');
520 if (!cp)
521 val = 0;
522 else if (prefix) {
524 * Does it begin with "a/$our-prefix" and such? Then this is
525 * very likely to apply to our directory.
527 if (!strncmp(name, prefix, prefix_length))
528 val = count_slashes(prefix);
529 else {
530 cp++;
531 if (!strncmp(cp, prefix, prefix_length))
532 val = count_slashes(prefix) + 1;
535 free(name);
536 return val;
540 * Does the ---/+++ line has the POSIX timestamp after the last HT?
541 * GNU diff puts epoch there to signal a creation/deletion event. Is
542 * this such a timestamp?
544 static int has_epoch_timestamp(const char *nameline)
547 * We are only interested in epoch timestamp; any non-zero
548 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
549 * For the same reason, the date must be either 1969-12-31 or
550 * 1970-01-01, and the seconds part must be "00".
552 const char stamp_regexp[] =
553 "^(1969-12-31|1970-01-01)"
555 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
557 "([-+][0-2][0-9][0-5][0-9])\n";
558 const char *timestamp = NULL, *cp;
559 static regex_t *stamp;
560 regmatch_t m[10];
561 int zoneoffset;
562 int hourminute;
563 int status;
565 for (cp = nameline; *cp != '\n'; cp++) {
566 if (*cp == '\t')
567 timestamp = cp + 1;
569 if (!timestamp)
570 return 0;
571 if (!stamp) {
572 stamp = xmalloc(sizeof(*stamp));
573 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
574 warning("Cannot prepare timestamp regexp %s",
575 stamp_regexp);
576 return 0;
580 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
581 if (status) {
582 if (status != REG_NOMATCH)
583 warning("regexec returned %d for input: %s",
584 status, timestamp);
585 return 0;
588 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10);
589 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
590 if (timestamp[m[3].rm_so] == '-')
591 zoneoffset = -zoneoffset;
594 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
595 * (west of GMT) or 1970-01-01 (east of GMT)
597 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
598 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
599 return 0;
601 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
602 strtol(timestamp + 14, NULL, 10) -
603 zoneoffset);
605 return ((zoneoffset < 0 && hourminute == 1440) ||
606 (0 <= zoneoffset && !hourminute));
610 * Get the name etc info from the ---/+++ lines of a traditional patch header
612 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
613 * files, we can happily check the index for a match, but for creating a
614 * new file we should try to match whatever "patch" does. I have no idea.
616 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
618 char *name;
620 first += 4; /* skip "--- " */
621 second += 4; /* skip "+++ " */
622 if (!p_value_known) {
623 int p, q;
624 p = guess_p_value(first);
625 q = guess_p_value(second);
626 if (p < 0) p = q;
627 if (0 <= p && p == q) {
628 p_value = p;
629 p_value_known = 1;
632 if (is_dev_null(first)) {
633 patch->is_new = 1;
634 patch->is_delete = 0;
635 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
636 patch->new_name = name;
637 } else if (is_dev_null(second)) {
638 patch->is_new = 0;
639 patch->is_delete = 1;
640 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
641 patch->old_name = name;
642 } else {
643 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
644 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
645 if (has_epoch_timestamp(first)) {
646 patch->is_new = 1;
647 patch->is_delete = 0;
648 patch->new_name = name;
649 } else if (has_epoch_timestamp(second)) {
650 patch->is_new = 0;
651 patch->is_delete = 1;
652 patch->old_name = name;
653 } else {
654 patch->old_name = patch->new_name = name;
657 if (!name)
658 die("unable to find filename in patch at line %d", linenr);
661 static int gitdiff_hdrend(const char *line, struct patch *patch)
663 return -1;
667 * We're anal about diff header consistency, to make
668 * sure that we don't end up having strange ambiguous
669 * patches floating around.
671 * As a result, gitdiff_{old|new}name() will check
672 * their names against any previous information, just
673 * to make sure..
675 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
677 if (!orig_name && !isnull)
678 return find_name(line, NULL, p_value, TERM_TAB);
680 if (orig_name) {
681 int len;
682 const char *name;
683 char *another;
684 name = orig_name;
685 len = strlen(name);
686 if (isnull)
687 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
688 another = find_name(line, NULL, p_value, TERM_TAB);
689 if (!another || memcmp(another, name, len))
690 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
691 free(another);
692 return orig_name;
694 else {
695 /* expect "/dev/null" */
696 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
697 die("git apply: bad git-diff - expected /dev/null on line %d", linenr);
698 return NULL;
702 static int gitdiff_oldname(const char *line, struct patch *patch)
704 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
705 return 0;
708 static int gitdiff_newname(const char *line, struct patch *patch)
710 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
711 return 0;
714 static int gitdiff_oldmode(const char *line, struct patch *patch)
716 patch->old_mode = strtoul(line, NULL, 8);
717 return 0;
720 static int gitdiff_newmode(const char *line, struct patch *patch)
722 patch->new_mode = strtoul(line, NULL, 8);
723 return 0;
726 static int gitdiff_delete(const char *line, struct patch *patch)
728 patch->is_delete = 1;
729 patch->old_name = patch->def_name;
730 return gitdiff_oldmode(line, patch);
733 static int gitdiff_newfile(const char *line, struct patch *patch)
735 patch->is_new = 1;
736 patch->new_name = patch->def_name;
737 return gitdiff_newmode(line, patch);
740 static int gitdiff_copysrc(const char *line, struct patch *patch)
742 patch->is_copy = 1;
743 patch->old_name = find_name(line, NULL, 0, 0);
744 return 0;
747 static int gitdiff_copydst(const char *line, struct patch *patch)
749 patch->is_copy = 1;
750 patch->new_name = find_name(line, NULL, 0, 0);
751 return 0;
754 static int gitdiff_renamesrc(const char *line, struct patch *patch)
756 patch->is_rename = 1;
757 patch->old_name = find_name(line, NULL, 0, 0);
758 return 0;
761 static int gitdiff_renamedst(const char *line, struct patch *patch)
763 patch->is_rename = 1;
764 patch->new_name = find_name(line, NULL, 0, 0);
765 return 0;
768 static int gitdiff_similarity(const char *line, struct patch *patch)
770 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
771 patch->score = 0;
772 return 0;
775 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
777 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
778 patch->score = 0;
779 return 0;
782 static int gitdiff_index(const char *line, struct patch *patch)
785 * index line is N hexadecimal, "..", N hexadecimal,
786 * and optional space with octal mode.
788 const char *ptr, *eol;
789 int len;
791 ptr = strchr(line, '.');
792 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
793 return 0;
794 len = ptr - line;
795 memcpy(patch->old_sha1_prefix, line, len);
796 patch->old_sha1_prefix[len] = 0;
798 line = ptr + 2;
799 ptr = strchr(line, ' ');
800 eol = strchr(line, '\n');
802 if (!ptr || eol < ptr)
803 ptr = eol;
804 len = ptr - line;
806 if (40 < len)
807 return 0;
808 memcpy(patch->new_sha1_prefix, line, len);
809 patch->new_sha1_prefix[len] = 0;
810 if (*ptr == ' ')
811 patch->old_mode = strtoul(ptr+1, NULL, 8);
812 return 0;
816 * This is normal for a diff that doesn't change anything: we'll fall through
817 * into the next diff. Tell the parser to break out.
819 static int gitdiff_unrecognized(const char *line, struct patch *patch)
821 return -1;
824 static const char *stop_at_slash(const char *line, int llen)
826 int i;
828 for (i = 0; i < llen; i++) {
829 int ch = line[i];
830 if (ch == '/')
831 return line + i;
833 return NULL;
837 * This is to extract the same name that appears on "diff --git"
838 * line. We do not find and return anything if it is a rename
839 * patch, and it is OK because we will find the name elsewhere.
840 * We need to reliably find name only when it is mode-change only,
841 * creation or deletion of an empty file. In any of these cases,
842 * both sides are the same name under a/ and b/ respectively.
844 static char *git_header_name(char *line, int llen)
846 const char *name;
847 const char *second = NULL;
848 size_t len;
850 line += strlen("diff --git ");
851 llen -= strlen("diff --git ");
853 if (*line == '"') {
854 const char *cp;
855 struct strbuf first = STRBUF_INIT;
856 struct strbuf sp = STRBUF_INIT;
858 if (unquote_c_style(&first, line, &second))
859 goto free_and_fail1;
861 /* advance to the first slash */
862 cp = stop_at_slash(first.buf, first.len);
863 /* we do not accept absolute paths */
864 if (!cp || cp == first.buf)
865 goto free_and_fail1;
866 strbuf_remove(&first, 0, cp + 1 - first.buf);
869 * second points at one past closing dq of name.
870 * find the second name.
872 while ((second < line + llen) && isspace(*second))
873 second++;
875 if (line + llen <= second)
876 goto free_and_fail1;
877 if (*second == '"') {
878 if (unquote_c_style(&sp, second, NULL))
879 goto free_and_fail1;
880 cp = stop_at_slash(sp.buf, sp.len);
881 if (!cp || cp == sp.buf)
882 goto free_and_fail1;
883 /* They must match, otherwise ignore */
884 if (strcmp(cp + 1, first.buf))
885 goto free_and_fail1;
886 strbuf_release(&sp);
887 return strbuf_detach(&first, NULL);
890 /* unquoted second */
891 cp = stop_at_slash(second, line + llen - second);
892 if (!cp || cp == second)
893 goto free_and_fail1;
894 cp++;
895 if (line + llen - cp != first.len + 1 ||
896 memcmp(first.buf, cp, first.len))
897 goto free_and_fail1;
898 return strbuf_detach(&first, NULL);
900 free_and_fail1:
901 strbuf_release(&first);
902 strbuf_release(&sp);
903 return NULL;
906 /* unquoted first name */
907 name = stop_at_slash(line, llen);
908 if (!name || name == line)
909 return NULL;
910 name++;
913 * since the first name is unquoted, a dq if exists must be
914 * the beginning of the second name.
916 for (second = name; second < line + llen; second++) {
917 if (*second == '"') {
918 struct strbuf sp = STRBUF_INIT;
919 const char *np;
921 if (unquote_c_style(&sp, second, NULL))
922 goto free_and_fail2;
924 np = stop_at_slash(sp.buf, sp.len);
925 if (!np || np == sp.buf)
926 goto free_and_fail2;
927 np++;
929 len = sp.buf + sp.len - np;
930 if (len < second - name &&
931 !strncmp(np, name, len) &&
932 isspace(name[len])) {
933 /* Good */
934 strbuf_remove(&sp, 0, np - sp.buf);
935 return strbuf_detach(&sp, NULL);
938 free_and_fail2:
939 strbuf_release(&sp);
940 return NULL;
945 * Accept a name only if it shows up twice, exactly the same
946 * form.
948 for (len = 0 ; ; len++) {
949 switch (name[len]) {
950 default:
951 continue;
952 case '\n':
953 return NULL;
954 case '\t': case ' ':
955 second = name+len;
956 for (;;) {
957 char c = *second++;
958 if (c == '\n')
959 return NULL;
960 if (c == '/')
961 break;
963 if (second[len] == '\n' && !memcmp(name, second, len)) {
964 return xmemdupz(name, len);
970 /* Verify that we recognize the lines following a git header */
971 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
973 unsigned long offset;
975 /* A git diff has explicit new/delete information, so we don't guess */
976 patch->is_new = 0;
977 patch->is_delete = 0;
980 * Some things may not have the old name in the
981 * rest of the headers anywhere (pure mode changes,
982 * or removing or adding empty files), so we get
983 * the default name from the header.
985 patch->def_name = git_header_name(line, len);
986 if (patch->def_name && root) {
987 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);
988 strcpy(s, root);
989 strcpy(s + root_len, patch->def_name);
990 free(patch->def_name);
991 patch->def_name = s;
994 line += len;
995 size -= len;
996 linenr++;
997 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
998 static const struct opentry {
999 const char *str;
1000 int (*fn)(const char *, struct patch *);
1001 } optable[] = {
1002 { "@@ -", gitdiff_hdrend },
1003 { "--- ", gitdiff_oldname },
1004 { "+++ ", gitdiff_newname },
1005 { "old mode ", gitdiff_oldmode },
1006 { "new mode ", gitdiff_newmode },
1007 { "deleted file mode ", gitdiff_delete },
1008 { "new file mode ", gitdiff_newfile },
1009 { "copy from ", gitdiff_copysrc },
1010 { "copy to ", gitdiff_copydst },
1011 { "rename old ", gitdiff_renamesrc },
1012 { "rename new ", gitdiff_renamedst },
1013 { "rename from ", gitdiff_renamesrc },
1014 { "rename to ", gitdiff_renamedst },
1015 { "similarity index ", gitdiff_similarity },
1016 { "dissimilarity index ", gitdiff_dissimilarity },
1017 { "index ", gitdiff_index },
1018 { "", gitdiff_unrecognized },
1020 int i;
1022 len = linelen(line, size);
1023 if (!len || line[len-1] != '\n')
1024 break;
1025 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1026 const struct opentry *p = optable + i;
1027 int oplen = strlen(p->str);
1028 if (len < oplen || memcmp(p->str, line, oplen))
1029 continue;
1030 if (p->fn(line + oplen, patch) < 0)
1031 return offset;
1032 break;
1036 return offset;
1039 static int parse_num(const char *line, unsigned long *p)
1041 char *ptr;
1043 if (!isdigit(*line))
1044 return 0;
1045 *p = strtoul(line, &ptr, 10);
1046 return ptr - line;
1049 static int parse_range(const char *line, int len, int offset, const char *expect,
1050 unsigned long *p1, unsigned long *p2)
1052 int digits, ex;
1054 if (offset < 0 || offset >= len)
1055 return -1;
1056 line += offset;
1057 len -= offset;
1059 digits = parse_num(line, p1);
1060 if (!digits)
1061 return -1;
1063 offset += digits;
1064 line += digits;
1065 len -= digits;
1067 *p2 = 1;
1068 if (*line == ',') {
1069 digits = parse_num(line+1, p2);
1070 if (!digits)
1071 return -1;
1073 offset += digits+1;
1074 line += digits+1;
1075 len -= digits+1;
1078 ex = strlen(expect);
1079 if (ex > len)
1080 return -1;
1081 if (memcmp(line, expect, ex))
1082 return -1;
1084 return offset + ex;
1087 static void recount_diff(char *line, int size, struct fragment *fragment)
1089 int oldlines = 0, newlines = 0, ret = 0;
1091 if (size < 1) {
1092 warning("recount: ignore empty hunk");
1093 return;
1096 for (;;) {
1097 int len = linelen(line, size);
1098 size -= len;
1099 line += len;
1101 if (size < 1)
1102 break;
1104 switch (*line) {
1105 case ' ': case '\n':
1106 newlines++;
1107 /* fall through */
1108 case '-':
1109 oldlines++;
1110 continue;
1111 case '+':
1112 newlines++;
1113 continue;
1114 case '\\':
1115 continue;
1116 case '@':
1117 ret = size < 3 || prefixcmp(line, "@@ ");
1118 break;
1119 case 'd':
1120 ret = size < 5 || prefixcmp(line, "diff ");
1121 break;
1122 default:
1123 ret = -1;
1124 break;
1126 if (ret) {
1127 warning("recount: unexpected line: %.*s",
1128 (int)linelen(line, size), line);
1129 return;
1131 break;
1133 fragment->oldlines = oldlines;
1134 fragment->newlines = newlines;
1138 * Parse a unified diff fragment header of the
1139 * form "@@ -a,b +c,d @@"
1141 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
1143 int offset;
1145 if (!len || line[len-1] != '\n')
1146 return -1;
1148 /* Figure out the number of lines in a fragment */
1149 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1150 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1152 return offset;
1155 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
1157 unsigned long offset, len;
1159 patch->is_toplevel_relative = 0;
1160 patch->is_rename = patch->is_copy = 0;
1161 patch->is_new = patch->is_delete = -1;
1162 patch->old_mode = patch->new_mode = 0;
1163 patch->old_name = patch->new_name = NULL;
1164 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
1165 unsigned long nextlen;
1167 len = linelen(line, size);
1168 if (!len)
1169 break;
1171 /* Testing this early allows us to take a few shortcuts.. */
1172 if (len < 6)
1173 continue;
1176 * Make sure we don't find any unconnected patch fragments.
1177 * That's a sign that we didn't find a header, and that a
1178 * patch has become corrupted/broken up.
1180 if (!memcmp("@@ -", line, 4)) {
1181 struct fragment dummy;
1182 if (parse_fragment_header(line, len, &dummy) < 0)
1183 continue;
1184 die("patch fragment without header at line %d: %.*s",
1185 linenr, (int)len-1, line);
1188 if (size < len + 6)
1189 break;
1192 * Git patch? It might not have a real patch, just a rename
1193 * or mode change, so we handle that specially
1195 if (!memcmp("diff --git ", line, 11)) {
1196 int git_hdr_len = parse_git_header(line, len, size, patch);
1197 if (git_hdr_len <= len)
1198 continue;
1199 if (!patch->old_name && !patch->new_name) {
1200 if (!patch->def_name)
1201 die("git diff header lacks filename information (line %d)", linenr);
1202 patch->old_name = patch->new_name = patch->def_name;
1204 patch->is_toplevel_relative = 1;
1205 *hdrsize = git_hdr_len;
1206 return offset;
1209 /* --- followed by +++ ? */
1210 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1211 continue;
1214 * We only accept unified patches, so we want it to
1215 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1216 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1218 nextlen = linelen(line + len, size - len);
1219 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1220 continue;
1222 /* Ok, we'll consider it a patch */
1223 parse_traditional_patch(line, line+len, patch);
1224 *hdrsize = len + nextlen;
1225 linenr += 2;
1226 return offset;
1228 return -1;
1231 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1233 char *err;
1235 if (!result)
1236 return;
1238 whitespace_error++;
1239 if (squelch_whitespace_errors &&
1240 squelch_whitespace_errors < whitespace_error)
1241 return;
1243 err = whitespace_error_string(result);
1244 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1245 patch_input_file, linenr, err, len, line);
1246 free(err);
1249 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1251 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1253 record_ws_error(result, line + 1, len - 2, linenr);
1257 * Parse a unified diff. Note that this really needs to parse each
1258 * fragment separately, since the only way to know the difference
1259 * between a "---" that is part of a patch, and a "---" that starts
1260 * the next patch is to look at the line counts..
1262 static int parse_fragment(char *line, unsigned long size,
1263 struct patch *patch, struct fragment *fragment)
1265 int added, deleted;
1266 int len = linelen(line, size), offset;
1267 unsigned long oldlines, newlines;
1268 unsigned long leading, trailing;
1270 offset = parse_fragment_header(line, len, fragment);
1271 if (offset < 0)
1272 return -1;
1273 if (offset > 0 && patch->recount)
1274 recount_diff(line + offset, size - offset, fragment);
1275 oldlines = fragment->oldlines;
1276 newlines = fragment->newlines;
1277 leading = 0;
1278 trailing = 0;
1280 /* Parse the thing.. */
1281 line += len;
1282 size -= len;
1283 linenr++;
1284 added = deleted = 0;
1285 for (offset = len;
1286 0 < size;
1287 offset += len, size -= len, line += len, linenr++) {
1288 if (!oldlines && !newlines)
1289 break;
1290 len = linelen(line, size);
1291 if (!len || line[len-1] != '\n')
1292 return -1;
1293 switch (*line) {
1294 default:
1295 return -1;
1296 case '\n': /* newer GNU diff, an empty context line */
1297 case ' ':
1298 oldlines--;
1299 newlines--;
1300 if (!deleted && !added)
1301 leading++;
1302 trailing++;
1303 break;
1304 case '-':
1305 if (apply_in_reverse &&
1306 ws_error_action != nowarn_ws_error)
1307 check_whitespace(line, len, patch->ws_rule);
1308 deleted++;
1309 oldlines--;
1310 trailing = 0;
1311 break;
1312 case '+':
1313 if (!apply_in_reverse &&
1314 ws_error_action != nowarn_ws_error)
1315 check_whitespace(line, len, patch->ws_rule);
1316 added++;
1317 newlines--;
1318 trailing = 0;
1319 break;
1322 * We allow "\ No newline at end of file". Depending
1323 * on locale settings when the patch was produced we
1324 * don't know what this line looks like. The only
1325 * thing we do know is that it begins with "\ ".
1326 * Checking for 12 is just for sanity check -- any
1327 * l10n of "\ No newline..." is at least that long.
1329 case '\\':
1330 if (len < 12 || memcmp(line, "\\ ", 2))
1331 return -1;
1332 break;
1335 if (oldlines || newlines)
1336 return -1;
1337 fragment->leading = leading;
1338 fragment->trailing = trailing;
1341 * If a fragment ends with an incomplete line, we failed to include
1342 * it in the above loop because we hit oldlines == newlines == 0
1343 * before seeing it.
1345 if (12 < size && !memcmp(line, "\\ ", 2))
1346 offset += linelen(line, size);
1348 patch->lines_added += added;
1349 patch->lines_deleted += deleted;
1351 if (0 < patch->is_new && oldlines)
1352 return error("new file depends on old contents");
1353 if (0 < patch->is_delete && newlines)
1354 return error("deleted file still has contents");
1355 return offset;
1358 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
1360 unsigned long offset = 0;
1361 unsigned long oldlines = 0, newlines = 0, context = 0;
1362 struct fragment **fragp = &patch->fragments;
1364 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1365 struct fragment *fragment;
1366 int len;
1368 fragment = xcalloc(1, sizeof(*fragment));
1369 fragment->linenr = linenr;
1370 len = parse_fragment(line, size, patch, fragment);
1371 if (len <= 0)
1372 die("corrupt patch at line %d", linenr);
1373 fragment->patch = line;
1374 fragment->size = len;
1375 oldlines += fragment->oldlines;
1376 newlines += fragment->newlines;
1377 context += fragment->leading + fragment->trailing;
1379 *fragp = fragment;
1380 fragp = &fragment->next;
1382 offset += len;
1383 line += len;
1384 size -= len;
1388 * If something was removed (i.e. we have old-lines) it cannot
1389 * be creation, and if something was added it cannot be
1390 * deletion. However, the reverse is not true; --unified=0
1391 * patches that only add are not necessarily creation even
1392 * though they do not have any old lines, and ones that only
1393 * delete are not necessarily deletion.
1395 * Unfortunately, a real creation/deletion patch do _not_ have
1396 * any context line by definition, so we cannot safely tell it
1397 * apart with --unified=0 insanity. At least if the patch has
1398 * more than one hunk it is not creation or deletion.
1400 if (patch->is_new < 0 &&
1401 (oldlines || (patch->fragments && patch->fragments->next)))
1402 patch->is_new = 0;
1403 if (patch->is_delete < 0 &&
1404 (newlines || (patch->fragments && patch->fragments->next)))
1405 patch->is_delete = 0;
1407 if (0 < patch->is_new && oldlines)
1408 die("new file %s depends on old contents", patch->new_name);
1409 if (0 < patch->is_delete && newlines)
1410 die("deleted file %s still has contents", patch->old_name);
1411 if (!patch->is_delete && !newlines && context)
1412 fprintf(stderr, "** warning: file %s becomes empty but "
1413 "is not deleted\n", patch->new_name);
1415 return offset;
1418 static inline int metadata_changes(struct patch *patch)
1420 return patch->is_rename > 0 ||
1421 patch->is_copy > 0 ||
1422 patch->is_new > 0 ||
1423 patch->is_delete ||
1424 (patch->old_mode && patch->new_mode &&
1425 patch->old_mode != patch->new_mode);
1428 static char *inflate_it(const void *data, unsigned long size,
1429 unsigned long inflated_size)
1431 z_stream stream;
1432 void *out;
1433 int st;
1435 memset(&stream, 0, sizeof(stream));
1437 stream.next_in = (unsigned char *)data;
1438 stream.avail_in = size;
1439 stream.next_out = out = xmalloc(inflated_size);
1440 stream.avail_out = inflated_size;
1441 git_inflate_init(&stream);
1442 st = git_inflate(&stream, Z_FINISH);
1443 git_inflate_end(&stream);
1444 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1445 free(out);
1446 return NULL;
1448 return out;
1451 static struct fragment *parse_binary_hunk(char **buf_p,
1452 unsigned long *sz_p,
1453 int *status_p,
1454 int *used_p)
1457 * Expect a line that begins with binary patch method ("literal"
1458 * or "delta"), followed by the length of data before deflating.
1459 * a sequence of 'length-byte' followed by base-85 encoded data
1460 * should follow, terminated by a newline.
1462 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1463 * and we would limit the patch line to 66 characters,
1464 * so one line can fit up to 13 groups that would decode
1465 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1466 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1468 int llen, used;
1469 unsigned long size = *sz_p;
1470 char *buffer = *buf_p;
1471 int patch_method;
1472 unsigned long origlen;
1473 char *data = NULL;
1474 int hunk_size = 0;
1475 struct fragment *frag;
1477 llen = linelen(buffer, size);
1478 used = llen;
1480 *status_p = 0;
1482 if (!prefixcmp(buffer, "delta ")) {
1483 patch_method = BINARY_DELTA_DEFLATED;
1484 origlen = strtoul(buffer + 6, NULL, 10);
1486 else if (!prefixcmp(buffer, "literal ")) {
1487 patch_method = BINARY_LITERAL_DEFLATED;
1488 origlen = strtoul(buffer + 8, NULL, 10);
1490 else
1491 return NULL;
1493 linenr++;
1494 buffer += llen;
1495 while (1) {
1496 int byte_length, max_byte_length, newsize;
1497 llen = linelen(buffer, size);
1498 used += llen;
1499 linenr++;
1500 if (llen == 1) {
1501 /* consume the blank line */
1502 buffer++;
1503 size--;
1504 break;
1507 * Minimum line is "A00000\n" which is 7-byte long,
1508 * and the line length must be multiple of 5 plus 2.
1510 if ((llen < 7) || (llen-2) % 5)
1511 goto corrupt;
1512 max_byte_length = (llen - 2) / 5 * 4;
1513 byte_length = *buffer;
1514 if ('A' <= byte_length && byte_length <= 'Z')
1515 byte_length = byte_length - 'A' + 1;
1516 else if ('a' <= byte_length && byte_length <= 'z')
1517 byte_length = byte_length - 'a' + 27;
1518 else
1519 goto corrupt;
1520 /* if the input length was not multiple of 4, we would
1521 * have filler at the end but the filler should never
1522 * exceed 3 bytes
1524 if (max_byte_length < byte_length ||
1525 byte_length <= max_byte_length - 4)
1526 goto corrupt;
1527 newsize = hunk_size + byte_length;
1528 data = xrealloc(data, newsize);
1529 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1530 goto corrupt;
1531 hunk_size = newsize;
1532 buffer += llen;
1533 size -= llen;
1536 frag = xcalloc(1, sizeof(*frag));
1537 frag->patch = inflate_it(data, hunk_size, origlen);
1538 if (!frag->patch)
1539 goto corrupt;
1540 free(data);
1541 frag->size = origlen;
1542 *buf_p = buffer;
1543 *sz_p = size;
1544 *used_p = used;
1545 frag->binary_patch_method = patch_method;
1546 return frag;
1548 corrupt:
1549 free(data);
1550 *status_p = -1;
1551 error("corrupt binary patch at line %d: %.*s",
1552 linenr-1, llen-1, buffer);
1553 return NULL;
1556 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1559 * We have read "GIT binary patch\n"; what follows is a line
1560 * that says the patch method (currently, either "literal" or
1561 * "delta") and the length of data before deflating; a
1562 * sequence of 'length-byte' followed by base-85 encoded data
1563 * follows.
1565 * When a binary patch is reversible, there is another binary
1566 * hunk in the same format, starting with patch method (either
1567 * "literal" or "delta") with the length of data, and a sequence
1568 * of length-byte + base-85 encoded data, terminated with another
1569 * empty line. This data, when applied to the postimage, produces
1570 * the preimage.
1572 struct fragment *forward;
1573 struct fragment *reverse;
1574 int status;
1575 int used, used_1;
1577 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1578 if (!forward && !status)
1579 /* there has to be one hunk (forward hunk) */
1580 return error("unrecognized binary patch at line %d", linenr-1);
1581 if (status)
1582 /* otherwise we already gave an error message */
1583 return status;
1585 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1586 if (reverse)
1587 used += used_1;
1588 else if (status) {
1590 * Not having reverse hunk is not an error, but having
1591 * a corrupt reverse hunk is.
1593 free((void*) forward->patch);
1594 free(forward);
1595 return status;
1597 forward->next = reverse;
1598 patch->fragments = forward;
1599 patch->is_binary = 1;
1600 return used;
1603 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1605 int hdrsize, patchsize;
1606 int offset = find_header(buffer, size, &hdrsize, patch);
1608 if (offset < 0)
1609 return offset;
1611 patch->ws_rule = whitespace_rule(patch->new_name
1612 ? patch->new_name
1613 : patch->old_name);
1615 patchsize = parse_single_patch(buffer + offset + hdrsize,
1616 size - offset - hdrsize, patch);
1618 if (!patchsize) {
1619 static const char *binhdr[] = {
1620 "Binary files ",
1621 "Files ",
1622 NULL,
1624 static const char git_binary[] = "GIT binary patch\n";
1625 int i;
1626 int hd = hdrsize + offset;
1627 unsigned long llen = linelen(buffer + hd, size - hd);
1629 if (llen == sizeof(git_binary) - 1 &&
1630 !memcmp(git_binary, buffer + hd, llen)) {
1631 int used;
1632 linenr++;
1633 used = parse_binary(buffer + hd + llen,
1634 size - hd - llen, patch);
1635 if (used)
1636 patchsize = used + llen;
1637 else
1638 patchsize = 0;
1640 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1641 for (i = 0; binhdr[i]; i++) {
1642 int len = strlen(binhdr[i]);
1643 if (len < size - hd &&
1644 !memcmp(binhdr[i], buffer + hd, len)) {
1645 linenr++;
1646 patch->is_binary = 1;
1647 patchsize = llen;
1648 break;
1653 /* Empty patch cannot be applied if it is a text patch
1654 * without metadata change. A binary patch appears
1655 * empty to us here.
1657 if ((apply || check) &&
1658 (!patch->is_binary && !metadata_changes(patch)))
1659 die("patch with only garbage at line %d", linenr);
1662 return offset + hdrsize + patchsize;
1665 #define swap(a,b) myswap((a),(b),sizeof(a))
1667 #define myswap(a, b, size) do { \
1668 unsigned char mytmp[size]; \
1669 memcpy(mytmp, &a, size); \
1670 memcpy(&a, &b, size); \
1671 memcpy(&b, mytmp, size); \
1672 } while (0)
1674 static void reverse_patches(struct patch *p)
1676 for (; p; p = p->next) {
1677 struct fragment *frag = p->fragments;
1679 swap(p->new_name, p->old_name);
1680 swap(p->new_mode, p->old_mode);
1681 swap(p->is_new, p->is_delete);
1682 swap(p->lines_added, p->lines_deleted);
1683 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1685 for (; frag; frag = frag->next) {
1686 swap(frag->newpos, frag->oldpos);
1687 swap(frag->newlines, frag->oldlines);
1692 static const char pluses[] =
1693 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1694 static const char minuses[]=
1695 "----------------------------------------------------------------------";
1697 static void show_stats(struct patch *patch)
1699 struct strbuf qname = STRBUF_INIT;
1700 char *cp = patch->new_name ? patch->new_name : patch->old_name;
1701 int max, add, del;
1703 quote_c_style(cp, &qname, NULL, 0);
1706 * "scale" the filename
1708 max = max_len;
1709 if (max > 50)
1710 max = 50;
1712 if (qname.len > max) {
1713 cp = strchr(qname.buf + qname.len + 3 - max, '/');
1714 if (!cp)
1715 cp = qname.buf + qname.len + 3 - max;
1716 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
1719 if (patch->is_binary) {
1720 printf(" %-*s | Bin\n", max, qname.buf);
1721 strbuf_release(&qname);
1722 return;
1725 printf(" %-*s |", max, qname.buf);
1726 strbuf_release(&qname);
1729 * scale the add/delete
1731 max = max + max_change > 70 ? 70 - max : max_change;
1732 add = patch->lines_added;
1733 del = patch->lines_deleted;
1735 if (max_change > 0) {
1736 int total = ((add + del) * max + max_change / 2) / max_change;
1737 add = (add * max + max_change / 2) / max_change;
1738 del = total - add;
1740 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
1741 add, pluses, del, minuses);
1744 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
1746 switch (st->st_mode & S_IFMT) {
1747 case S_IFLNK:
1748 if (strbuf_readlink(buf, path, st->st_size) < 0)
1749 return error("unable to read symlink %s", path);
1750 return 0;
1751 case S_IFREG:
1752 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
1753 return error("unable to open or read %s", path);
1754 convert_to_git(path, buf->buf, buf->len, buf, 0);
1755 return 0;
1756 default:
1757 return -1;
1762 * Update the preimage, and the common lines in postimage,
1763 * from buffer buf of length len. If postlen is 0 the postimage
1764 * is updated in place, otherwise it's updated on a new buffer
1765 * of length postlen
1768 static void update_pre_post_images(struct image *preimage,
1769 struct image *postimage,
1770 char *buf,
1771 size_t len, size_t postlen)
1773 int i, ctx;
1774 char *new, *old, *fixed;
1775 struct image fixed_preimage;
1778 * Update the preimage with whitespace fixes. Note that we
1779 * are not losing preimage->buf -- apply_one_fragment() will
1780 * free "oldlines".
1782 prepare_image(&fixed_preimage, buf, len, 1);
1783 assert(fixed_preimage.nr == preimage->nr);
1784 for (i = 0; i < preimage->nr; i++)
1785 fixed_preimage.line[i].flag = preimage->line[i].flag;
1786 free(preimage->line_allocated);
1787 *preimage = fixed_preimage;
1790 * Adjust the common context lines in postimage. This can be
1791 * done in-place when we are just doing whitespace fixing,
1792 * which does not make the string grow, but needs a new buffer
1793 * when ignoring whitespace causes the update, since in this case
1794 * we could have e.g. tabs converted to multiple spaces.
1795 * We trust the caller to tell us if the update can be done
1796 * in place (postlen==0) or not.
1798 old = postimage->buf;
1799 if (postlen)
1800 new = postimage->buf = xmalloc(postlen);
1801 else
1802 new = old;
1803 fixed = preimage->buf;
1804 for (i = ctx = 0; i < postimage->nr; i++) {
1805 size_t len = postimage->line[i].len;
1806 if (!(postimage->line[i].flag & LINE_COMMON)) {
1807 /* an added line -- no counterparts in preimage */
1808 memmove(new, old, len);
1809 old += len;
1810 new += len;
1811 continue;
1814 /* a common context -- skip it in the original postimage */
1815 old += len;
1817 /* and find the corresponding one in the fixed preimage */
1818 while (ctx < preimage->nr &&
1819 !(preimage->line[ctx].flag & LINE_COMMON)) {
1820 fixed += preimage->line[ctx].len;
1821 ctx++;
1823 if (preimage->nr <= ctx)
1824 die("oops");
1826 /* and copy it in, while fixing the line length */
1827 len = preimage->line[ctx].len;
1828 memcpy(new, fixed, len);
1829 new += len;
1830 fixed += len;
1831 postimage->line[i].len = len;
1832 ctx++;
1835 /* Fix the length of the whole thing */
1836 postimage->len = new - postimage->buf;
1839 static int match_fragment(struct image *img,
1840 struct image *preimage,
1841 struct image *postimage,
1842 unsigned long try,
1843 int try_lno,
1844 unsigned ws_rule,
1845 int match_beginning, int match_end)
1847 int i;
1848 char *fixed_buf, *buf, *orig, *target;
1850 if (preimage->nr + try_lno > img->nr)
1851 return 0;
1853 if (match_beginning && try_lno)
1854 return 0;
1856 if (match_end && preimage->nr + try_lno != img->nr)
1857 return 0;
1859 /* Quick hash check */
1860 for (i = 0; i < preimage->nr; i++)
1861 if (preimage->line[i].hash != img->line[try_lno + i].hash)
1862 return 0;
1865 * Do we have an exact match? If we were told to match
1866 * at the end, size must be exactly at try+fragsize,
1867 * otherwise try+fragsize must be still within the preimage,
1868 * and either case, the old piece should match the preimage
1869 * exactly.
1871 if ((match_end
1872 ? (try + preimage->len == img->len)
1873 : (try + preimage->len <= img->len)) &&
1874 !memcmp(img->buf + try, preimage->buf, preimage->len))
1875 return 1;
1878 * No exact match. If we are ignoring whitespace, run a line-by-line
1879 * fuzzy matching. We collect all the line length information because
1880 * we need it to adjust whitespace if we match.
1882 if (ws_ignore_action == ignore_ws_change) {
1883 size_t imgoff = 0;
1884 size_t preoff = 0;
1885 size_t postlen = postimage->len;
1886 for (i = 0; i < preimage->nr; i++) {
1887 size_t prelen = preimage->line[i].len;
1888 size_t imglen = img->line[try_lno+i].len;
1890 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
1891 preimage->buf + preoff, prelen))
1892 return 0;
1893 if (preimage->line[i].flag & LINE_COMMON)
1894 postlen += imglen - prelen;
1895 imgoff += imglen;
1896 preoff += prelen;
1900 * Ok, the preimage matches with whitespace fuzz. Update it and
1901 * the common postimage lines to use the same whitespace as the
1902 * target. imgoff now holds the true length of the target that
1903 * matches the preimage, and we need to update the line lengths
1904 * of the preimage to match the target ones.
1906 fixed_buf = xmalloc(imgoff);
1907 memcpy(fixed_buf, img->buf + try, imgoff);
1908 for (i = 0; i < preimage->nr; i++)
1909 preimage->line[i].len = img->line[try_lno+i].len;
1912 * Update the preimage buffer and the postimage context lines.
1914 update_pre_post_images(preimage, postimage,
1915 fixed_buf, imgoff, postlen);
1916 return 1;
1919 if (ws_error_action != correct_ws_error)
1920 return 0;
1923 * The hunk does not apply byte-by-byte, but the hash says
1924 * it might with whitespace fuzz. We haven't been asked to
1925 * ignore whitespace, we were asked to correct whitespace
1926 * errors, so let's try matching after whitespace correction.
1928 fixed_buf = xmalloc(preimage->len + 1);
1929 buf = fixed_buf;
1930 orig = preimage->buf;
1931 target = img->buf + try;
1932 for (i = 0; i < preimage->nr; i++) {
1933 size_t fixlen; /* length after fixing the preimage */
1934 size_t oldlen = preimage->line[i].len;
1935 size_t tgtlen = img->line[try_lno + i].len;
1936 size_t tgtfixlen; /* length after fixing the target line */
1937 char tgtfixbuf[1024], *tgtfix;
1938 int match;
1940 /* Try fixing the line in the preimage */
1941 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);
1943 /* Try fixing the line in the target */
1944 if (sizeof(tgtfixbuf) > tgtlen)
1945 tgtfix = tgtfixbuf;
1946 else
1947 tgtfix = xmalloc(tgtlen);
1948 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);
1951 * If they match, either the preimage was based on
1952 * a version before our tree fixed whitespace breakage,
1953 * or we are lacking a whitespace-fix patch the tree
1954 * the preimage was based on already had (i.e. target
1955 * has whitespace breakage, the preimage doesn't).
1956 * In either case, we are fixing the whitespace breakages
1957 * so we might as well take the fix together with their
1958 * real change.
1960 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));
1962 if (tgtfix != tgtfixbuf)
1963 free(tgtfix);
1964 if (!match)
1965 goto unmatch_exit;
1967 orig += oldlen;
1968 buf += fixlen;
1969 target += tgtlen;
1973 * Yes, the preimage is based on an older version that still
1974 * has whitespace breakages unfixed, and fixing them makes the
1975 * hunk match. Update the context lines in the postimage.
1977 update_pre_post_images(preimage, postimage,
1978 fixed_buf, buf - fixed_buf, 0);
1979 return 1;
1981 unmatch_exit:
1982 free(fixed_buf);
1983 return 0;
1986 static int find_pos(struct image *img,
1987 struct image *preimage,
1988 struct image *postimage,
1989 int line,
1990 unsigned ws_rule,
1991 int match_beginning, int match_end)
1993 int i;
1994 unsigned long backwards, forwards, try;
1995 int backwards_lno, forwards_lno, try_lno;
1997 if (preimage->nr > img->nr)
1998 return -1;
2001 * If match_begining or match_end is specified, there is no
2002 * point starting from a wrong line that will never match and
2003 * wander around and wait for a match at the specified end.
2005 if (match_beginning)
2006 line = 0;
2007 else if (match_end)
2008 line = img->nr - preimage->nr;
2010 if (line > img->nr)
2011 line = img->nr;
2013 try = 0;
2014 for (i = 0; i < line; i++)
2015 try += img->line[i].len;
2018 * There's probably some smart way to do this, but I'll leave
2019 * that to the smart and beautiful people. I'm simple and stupid.
2021 backwards = try;
2022 backwards_lno = line;
2023 forwards = try;
2024 forwards_lno = line;
2025 try_lno = line;
2027 for (i = 0; ; i++) {
2028 if (match_fragment(img, preimage, postimage,
2029 try, try_lno, ws_rule,
2030 match_beginning, match_end))
2031 return try_lno;
2033 again:
2034 if (backwards_lno == 0 && forwards_lno == img->nr)
2035 break;
2037 if (i & 1) {
2038 if (backwards_lno == 0) {
2039 i++;
2040 goto again;
2042 backwards_lno--;
2043 backwards -= img->line[backwards_lno].len;
2044 try = backwards;
2045 try_lno = backwards_lno;
2046 } else {
2047 if (forwards_lno == img->nr) {
2048 i++;
2049 goto again;
2051 forwards += img->line[forwards_lno].len;
2052 forwards_lno++;
2053 try = forwards;
2054 try_lno = forwards_lno;
2058 return -1;
2061 static void remove_first_line(struct image *img)
2063 img->buf += img->line[0].len;
2064 img->len -= img->line[0].len;
2065 img->line++;
2066 img->nr--;
2069 static void remove_last_line(struct image *img)
2071 img->len -= img->line[--img->nr].len;
2074 static void update_image(struct image *img,
2075 int applied_pos,
2076 struct image *preimage,
2077 struct image *postimage)
2080 * remove the copy of preimage at offset in img
2081 * and replace it with postimage
2083 int i, nr;
2084 size_t remove_count, insert_count, applied_at = 0;
2085 char *result;
2087 for (i = 0; i < applied_pos; i++)
2088 applied_at += img->line[i].len;
2090 remove_count = 0;
2091 for (i = 0; i < preimage->nr; i++)
2092 remove_count += img->line[applied_pos + i].len;
2093 insert_count = postimage->len;
2095 /* Adjust the contents */
2096 result = xmalloc(img->len + insert_count - remove_count + 1);
2097 memcpy(result, img->buf, applied_at);
2098 memcpy(result + applied_at, postimage->buf, postimage->len);
2099 memcpy(result + applied_at + postimage->len,
2100 img->buf + (applied_at + remove_count),
2101 img->len - (applied_at + remove_count));
2102 free(img->buf);
2103 img->buf = result;
2104 img->len += insert_count - remove_count;
2105 result[img->len] = '\0';
2107 /* Adjust the line table */
2108 nr = img->nr + postimage->nr - preimage->nr;
2109 if (preimage->nr < postimage->nr) {
2111 * NOTE: this knows that we never call remove_first_line()
2112 * on anything other than pre/post image.
2114 img->line = xrealloc(img->line, nr * sizeof(*img->line));
2115 img->line_allocated = img->line;
2117 if (preimage->nr != postimage->nr)
2118 memmove(img->line + applied_pos + postimage->nr,
2119 img->line + applied_pos + preimage->nr,
2120 (img->nr - (applied_pos + preimage->nr)) *
2121 sizeof(*img->line));
2122 memcpy(img->line + applied_pos,
2123 postimage->line,
2124 postimage->nr * sizeof(*img->line));
2125 img->nr = nr;
2128 static int apply_one_fragment(struct image *img, struct fragment *frag,
2129 int inaccurate_eof, unsigned ws_rule)
2131 int match_beginning, match_end;
2132 const char *patch = frag->patch;
2133 int size = frag->size;
2134 char *old, *new, *oldlines, *newlines;
2135 int new_blank_lines_at_end = 0;
2136 unsigned long leading, trailing;
2137 int pos, applied_pos;
2138 struct image preimage;
2139 struct image postimage;
2141 memset(&preimage, 0, sizeof(preimage));
2142 memset(&postimage, 0, sizeof(postimage));
2143 oldlines = xmalloc(size);
2144 newlines = xmalloc(size);
2146 old = oldlines;
2147 new = newlines;
2148 while (size > 0) {
2149 char first;
2150 int len = linelen(patch, size);
2151 int plen, added;
2152 int added_blank_line = 0;
2153 int is_blank_context = 0;
2155 if (!len)
2156 break;
2159 * "plen" is how much of the line we should use for
2160 * the actual patch data. Normally we just remove the
2161 * first character on the line, but if the line is
2162 * followed by "\ No newline", then we also remove the
2163 * last one (which is the newline, of course).
2165 plen = len - 1;
2166 if (len < size && patch[len] == '\\')
2167 plen--;
2168 first = *patch;
2169 if (apply_in_reverse) {
2170 if (first == '-')
2171 first = '+';
2172 else if (first == '+')
2173 first = '-';
2176 switch (first) {
2177 case '\n':
2178 /* Newer GNU diff, empty context line */
2179 if (plen < 0)
2180 /* ... followed by '\No newline'; nothing */
2181 break;
2182 *old++ = '\n';
2183 *new++ = '\n';
2184 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2185 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2186 is_blank_context = 1;
2187 break;
2188 case ' ':
2189 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2190 ws_blank_line(patch + 1, plen, ws_rule))
2191 is_blank_context = 1;
2192 case '-':
2193 memcpy(old, patch + 1, plen);
2194 add_line_info(&preimage, old, plen,
2195 (first == ' ' ? LINE_COMMON : 0));
2196 old += plen;
2197 if (first == '-')
2198 break;
2199 /* Fall-through for ' ' */
2200 case '+':
2201 /* --no-add does not add new lines */
2202 if (first == '+' && no_add)
2203 break;
2205 if (first != '+' ||
2206 !whitespace_error ||
2207 ws_error_action != correct_ws_error) {
2208 memcpy(new, patch + 1, plen);
2209 added = plen;
2211 else {
2212 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2214 add_line_info(&postimage, new, added,
2215 (first == '+' ? 0 : LINE_COMMON));
2216 new += added;
2217 if (first == '+' &&
2218 (ws_rule & WS_BLANK_AT_EOF) &&
2219 ws_blank_line(patch + 1, plen, ws_rule))
2220 added_blank_line = 1;
2221 break;
2222 case '@': case '\\':
2223 /* Ignore it, we already handled it */
2224 break;
2225 default:
2226 if (apply_verbosely)
2227 error("invalid start of line: '%c'", first);
2228 return -1;
2230 if (added_blank_line)
2231 new_blank_lines_at_end++;
2232 else if (is_blank_context)
2234 else
2235 new_blank_lines_at_end = 0;
2236 patch += len;
2237 size -= len;
2239 if (inaccurate_eof &&
2240 old > oldlines && old[-1] == '\n' &&
2241 new > newlines && new[-1] == '\n') {
2242 old--;
2243 new--;
2246 leading = frag->leading;
2247 trailing = frag->trailing;
2250 * A hunk to change lines at the beginning would begin with
2251 * @@ -1,L +N,M @@
2252 * but we need to be careful. -U0 that inserts before the second
2253 * line also has this pattern.
2255 * And a hunk to add to an empty file would begin with
2256 * @@ -0,0 +N,M @@
2258 * In other words, a hunk that is (frag->oldpos <= 1) with or
2259 * without leading context must match at the beginning.
2261 match_beginning = (!frag->oldpos ||
2262 (frag->oldpos == 1 && !unidiff_zero));
2265 * A hunk without trailing lines must match at the end.
2266 * However, we simply cannot tell if a hunk must match end
2267 * from the lack of trailing lines if the patch was generated
2268 * with unidiff without any context.
2270 match_end = !unidiff_zero && !trailing;
2272 pos = frag->newpos ? (frag->newpos - 1) : 0;
2273 preimage.buf = oldlines;
2274 preimage.len = old - oldlines;
2275 postimage.buf = newlines;
2276 postimage.len = new - newlines;
2277 preimage.line = preimage.line_allocated;
2278 postimage.line = postimage.line_allocated;
2280 for (;;) {
2282 applied_pos = find_pos(img, &preimage, &postimage, pos,
2283 ws_rule, match_beginning, match_end);
2285 if (applied_pos >= 0)
2286 break;
2288 /* Am I at my context limits? */
2289 if ((leading <= p_context) && (trailing <= p_context))
2290 break;
2291 if (match_beginning || match_end) {
2292 match_beginning = match_end = 0;
2293 continue;
2297 * Reduce the number of context lines; reduce both
2298 * leading and trailing if they are equal otherwise
2299 * just reduce the larger context.
2301 if (leading >= trailing) {
2302 remove_first_line(&preimage);
2303 remove_first_line(&postimage);
2304 pos--;
2305 leading--;
2307 if (trailing > leading) {
2308 remove_last_line(&preimage);
2309 remove_last_line(&postimage);
2310 trailing--;
2314 if (applied_pos >= 0) {
2315 if (new_blank_lines_at_end &&
2316 preimage.nr + applied_pos == img->nr &&
2317 (ws_rule & WS_BLANK_AT_EOF) &&
2318 ws_error_action != nowarn_ws_error) {
2319 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);
2320 if (ws_error_action == correct_ws_error) {
2321 while (new_blank_lines_at_end--)
2322 remove_last_line(&postimage);
2325 * We would want to prevent write_out_results()
2326 * from taking place in apply_patch() that follows
2327 * the callchain led us here, which is:
2328 * apply_patch->check_patch_list->check_patch->
2329 * apply_data->apply_fragments->apply_one_fragment
2331 if (ws_error_action == die_on_ws_error)
2332 apply = 0;
2336 * Warn if it was necessary to reduce the number
2337 * of context lines.
2339 if ((leading != frag->leading) ||
2340 (trailing != frag->trailing))
2341 fprintf(stderr, "Context reduced to (%ld/%ld)"
2342 " to apply fragment at %d\n",
2343 leading, trailing, applied_pos+1);
2344 update_image(img, applied_pos, &preimage, &postimage);
2345 } else {
2346 if (apply_verbosely)
2347 error("while searching for:\n%.*s",
2348 (int)(old - oldlines), oldlines);
2351 free(oldlines);
2352 free(newlines);
2353 free(preimage.line_allocated);
2354 free(postimage.line_allocated);
2356 return (applied_pos < 0);
2359 static int apply_binary_fragment(struct image *img, struct patch *patch)
2361 struct fragment *fragment = patch->fragments;
2362 unsigned long len;
2363 void *dst;
2365 /* Binary patch is irreversible without the optional second hunk */
2366 if (apply_in_reverse) {
2367 if (!fragment->next)
2368 return error("cannot reverse-apply a binary patch "
2369 "without the reverse hunk to '%s'",
2370 patch->new_name
2371 ? patch->new_name : patch->old_name);
2372 fragment = fragment->next;
2374 switch (fragment->binary_patch_method) {
2375 case BINARY_DELTA_DEFLATED:
2376 dst = patch_delta(img->buf, img->len, fragment->patch,
2377 fragment->size, &len);
2378 if (!dst)
2379 return -1;
2380 clear_image(img);
2381 img->buf = dst;
2382 img->len = len;
2383 return 0;
2384 case BINARY_LITERAL_DEFLATED:
2385 clear_image(img);
2386 img->len = fragment->size;
2387 img->buf = xmalloc(img->len+1);
2388 memcpy(img->buf, fragment->patch, img->len);
2389 img->buf[img->len] = '\0';
2390 return 0;
2392 return -1;
2395 static int apply_binary(struct image *img, struct patch *patch)
2397 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2398 unsigned char sha1[20];
2401 * For safety, we require patch index line to contain
2402 * full 40-byte textual SHA1 for old and new, at least for now.
2404 if (strlen(patch->old_sha1_prefix) != 40 ||
2405 strlen(patch->new_sha1_prefix) != 40 ||
2406 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
2407 get_sha1_hex(patch->new_sha1_prefix, sha1))
2408 return error("cannot apply binary patch to '%s' "
2409 "without full index line", name);
2411 if (patch->old_name) {
2413 * See if the old one matches what the patch
2414 * applies to.
2416 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2417 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
2418 return error("the patch applies to '%s' (%s), "
2419 "which does not match the "
2420 "current contents.",
2421 name, sha1_to_hex(sha1));
2423 else {
2424 /* Otherwise, the old one must be empty. */
2425 if (img->len)
2426 return error("the patch applies to an empty "
2427 "'%s' but it is not empty", name);
2430 get_sha1_hex(patch->new_sha1_prefix, sha1);
2431 if (is_null_sha1(sha1)) {
2432 clear_image(img);
2433 return 0; /* deletion patch */
2436 if (has_sha1_file(sha1)) {
2437 /* We already have the postimage */
2438 enum object_type type;
2439 unsigned long size;
2440 char *result;
2442 result = read_sha1_file(sha1, &type, &size);
2443 if (!result)
2444 return error("the necessary postimage %s for "
2445 "'%s' cannot be read",
2446 patch->new_sha1_prefix, name);
2447 clear_image(img);
2448 img->buf = result;
2449 img->len = size;
2450 } else {
2452 * We have verified buf matches the preimage;
2453 * apply the patch data to it, which is stored
2454 * in the patch->fragments->{patch,size}.
2456 if (apply_binary_fragment(img, patch))
2457 return error("binary patch does not apply to '%s'",
2458 name);
2460 /* verify that the result matches */
2461 hash_sha1_file(img->buf, img->len, blob_type, sha1);
2462 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
2463 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",
2464 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
2467 return 0;
2470 static int apply_fragments(struct image *img, struct patch *patch)
2472 struct fragment *frag = patch->fragments;
2473 const char *name = patch->old_name ? patch->old_name : patch->new_name;
2474 unsigned ws_rule = patch->ws_rule;
2475 unsigned inaccurate_eof = patch->inaccurate_eof;
2477 if (patch->is_binary)
2478 return apply_binary(img, patch);
2480 while (frag) {
2481 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {
2482 error("patch failed: %s:%ld", name, frag->oldpos);
2483 if (!apply_with_reject)
2484 return -1;
2485 frag->rejected = 1;
2487 frag = frag->next;
2489 return 0;
2492 static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)
2494 if (!ce)
2495 return 0;
2497 if (S_ISGITLINK(ce->ce_mode)) {
2498 strbuf_grow(buf, 100);
2499 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));
2500 } else {
2501 enum object_type type;
2502 unsigned long sz;
2503 char *result;
2505 result = read_sha1_file(ce->sha1, &type, &sz);
2506 if (!result)
2507 return -1;
2508 /* XXX read_sha1_file NUL-terminates */
2509 strbuf_attach(buf, result, sz, sz + 1);
2511 return 0;
2514 static struct patch *in_fn_table(const char *name)
2516 struct string_list_item *item;
2518 if (name == NULL)
2519 return NULL;
2521 item = string_list_lookup(name, &fn_table);
2522 if (item != NULL)
2523 return (struct patch *)item->util;
2525 return NULL;
2529 * item->util in the filename table records the status of the path.
2530 * Usually it points at a patch (whose result records the contents
2531 * of it after applying it), but it could be PATH_WAS_DELETED for a
2532 * path that a previously applied patch has already removed.
2534 #define PATH_TO_BE_DELETED ((struct patch *) -2)
2535 #define PATH_WAS_DELETED ((struct patch *) -1)
2537 static int to_be_deleted(struct patch *patch)
2539 return patch == PATH_TO_BE_DELETED;
2542 static int was_deleted(struct patch *patch)
2544 return patch == PATH_WAS_DELETED;
2547 static void add_to_fn_table(struct patch *patch)
2549 struct string_list_item *item;
2552 * Always add new_name unless patch is a deletion
2553 * This should cover the cases for normal diffs,
2554 * file creations and copies
2556 if (patch->new_name != NULL) {
2557 item = string_list_insert(patch->new_name, &fn_table);
2558 item->util = patch;
2562 * store a failure on rename/deletion cases because
2563 * later chunks shouldn't patch old names
2565 if ((patch->new_name == NULL) || (patch->is_rename)) {
2566 item = string_list_insert(patch->old_name, &fn_table);
2567 item->util = PATH_WAS_DELETED;
2571 static void prepare_fn_table(struct patch *patch)
2574 * store information about incoming file deletion
2576 while (patch) {
2577 if ((patch->new_name == NULL) || (patch->is_rename)) {
2578 struct string_list_item *item;
2579 item = string_list_insert(patch->old_name, &fn_table);
2580 item->util = PATH_TO_BE_DELETED;
2582 patch = patch->next;
2586 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
2588 struct strbuf buf = STRBUF_INIT;
2589 struct image image;
2590 size_t len;
2591 char *img;
2592 struct patch *tpatch;
2594 if (!(patch->is_copy || patch->is_rename) &&
2595 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {
2596 if (was_deleted(tpatch)) {
2597 return error("patch %s has been renamed/deleted",
2598 patch->old_name);
2600 /* We have a patched copy in memory use that */
2601 strbuf_add(&buf, tpatch->result, tpatch->resultsize);
2602 } else if (cached) {
2603 if (read_file_or_gitlink(ce, &buf))
2604 return error("read of %s failed", patch->old_name);
2605 } else if (patch->old_name) {
2606 if (S_ISGITLINK(patch->old_mode)) {
2607 if (ce) {
2608 read_file_or_gitlink(ce, &buf);
2609 } else {
2611 * There is no way to apply subproject
2612 * patch without looking at the index.
2614 patch->fragments = NULL;
2616 } else {
2617 if (read_old_data(st, patch->old_name, &buf))
2618 return error("read of %s failed", patch->old_name);
2622 img = strbuf_detach(&buf, &len);
2623 prepare_image(&image, img, len, !patch->is_binary);
2625 if (apply_fragments(&image, patch) < 0)
2626 return -1; /* note with --reject this succeeds. */
2627 patch->result = image.buf;
2628 patch->resultsize = image.len;
2629 add_to_fn_table(patch);
2630 free(image.line_allocated);
2632 if (0 < patch->is_delete && patch->resultsize)
2633 return error("removal patch leaves file contents");
2635 return 0;
2638 static int check_to_create_blob(const char *new_name, int ok_if_exists)
2640 struct stat nst;
2641 if (!lstat(new_name, &nst)) {
2642 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2643 return 0;
2645 * A leading component of new_name might be a symlink
2646 * that is going to be removed with this patch, but
2647 * still pointing at somewhere that has the path.
2648 * In such a case, path "new_name" does not exist as
2649 * far as git is concerned.
2651 if (has_symlink_leading_path(new_name, strlen(new_name)))
2652 return 0;
2654 return error("%s: already exists in working directory", new_name);
2656 else if ((errno != ENOENT) && (errno != ENOTDIR))
2657 return error("%s: %s", new_name, strerror(errno));
2658 return 0;
2661 static int verify_index_match(struct cache_entry *ce, struct stat *st)
2663 if (S_ISGITLINK(ce->ce_mode)) {
2664 if (!S_ISDIR(st->st_mode))
2665 return -1;
2666 return 0;
2668 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID);
2671 static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)
2673 const char *old_name = patch->old_name;
2674 struct patch *tpatch = NULL;
2675 int stat_ret = 0;
2676 unsigned st_mode = 0;
2679 * Make sure that we do not have local modifications from the
2680 * index when we are looking at the index. Also make sure
2681 * we have the preimage file to be patched in the work tree,
2682 * unless --cached, which tells git to apply only in the index.
2684 if (!old_name)
2685 return 0;
2687 assert(patch->is_new <= 0);
2689 if (!(patch->is_copy || patch->is_rename) &&
2690 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {
2691 if (was_deleted(tpatch))
2692 return error("%s: has been deleted/renamed", old_name);
2693 st_mode = tpatch->new_mode;
2694 } else if (!cached) {
2695 stat_ret = lstat(old_name, st);
2696 if (stat_ret && errno != ENOENT)
2697 return error("%s: %s", old_name, strerror(errno));
2700 if (to_be_deleted(tpatch))
2701 tpatch = NULL;
2703 if (check_index && !tpatch) {
2704 int pos = cache_name_pos(old_name, strlen(old_name));
2705 if (pos < 0) {
2706 if (patch->is_new < 0)
2707 goto is_new;
2708 return error("%s: does not exist in index", old_name);
2710 *ce = active_cache[pos];
2711 if (stat_ret < 0) {
2712 struct checkout costate;
2713 /* checkout */
2714 costate.base_dir = "";
2715 costate.base_dir_len = 0;
2716 costate.force = 0;
2717 costate.quiet = 0;
2718 costate.not_new = 0;
2719 costate.refresh_cache = 1;
2720 if (checkout_entry(*ce, &costate, NULL) ||
2721 lstat(old_name, st))
2722 return -1;
2724 if (!cached && verify_index_match(*ce, st))
2725 return error("%s: does not match index", old_name);
2726 if (cached)
2727 st_mode = (*ce)->ce_mode;
2728 } else if (stat_ret < 0) {
2729 if (patch->is_new < 0)
2730 goto is_new;
2731 return error("%s: %s", old_name, strerror(errno));
2734 if (!cached && !tpatch)
2735 st_mode = ce_mode_from_stat(*ce, st->st_mode);
2737 if (patch->is_new < 0)
2738 patch->is_new = 0;
2739 if (!patch->old_mode)
2740 patch->old_mode = st_mode;
2741 if ((st_mode ^ patch->old_mode) & S_IFMT)
2742 return error("%s: wrong type", old_name);
2743 if (st_mode != patch->old_mode)
2744 warning("%s has type %o, expected %o",
2745 old_name, st_mode, patch->old_mode);
2746 if (!patch->new_mode && !patch->is_delete)
2747 patch->new_mode = st_mode;
2748 return 0;
2750 is_new:
2751 patch->is_new = 1;
2752 patch->is_delete = 0;
2753 patch->old_name = NULL;
2754 return 0;
2757 static int check_patch(struct patch *patch)
2759 struct stat st;
2760 const char *old_name = patch->old_name;
2761 const char *new_name = patch->new_name;
2762 const char *name = old_name ? old_name : new_name;
2763 struct cache_entry *ce = NULL;
2764 struct patch *tpatch;
2765 int ok_if_exists;
2766 int status;
2768 patch->rejected = 1; /* we will drop this after we succeed */
2770 status = check_preimage(patch, &ce, &st);
2771 if (status)
2772 return status;
2773 old_name = patch->old_name;
2775 if ((tpatch = in_fn_table(new_name)) &&
2776 (was_deleted(tpatch) || to_be_deleted(tpatch)))
2778 * A type-change diff is always split into a patch to
2779 * delete old, immediately followed by a patch to
2780 * create new (see diff.c::run_diff()); in such a case
2781 * it is Ok that the entry to be deleted by the
2782 * previous patch is still in the working tree and in
2783 * the index.
2785 ok_if_exists = 1;
2786 else
2787 ok_if_exists = 0;
2789 if (new_name &&
2790 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2791 if (check_index &&
2792 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2793 !ok_if_exists)
2794 return error("%s: already exists in index", new_name);
2795 if (!cached) {
2796 int err = check_to_create_blob(new_name, ok_if_exists);
2797 if (err)
2798 return err;
2800 if (!patch->new_mode) {
2801 if (0 < patch->is_new)
2802 patch->new_mode = S_IFREG | 0644;
2803 else
2804 patch->new_mode = patch->old_mode;
2808 if (new_name && old_name) {
2809 int same = !strcmp(old_name, new_name);
2810 if (!patch->new_mode)
2811 patch->new_mode = patch->old_mode;
2812 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2813 return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2814 patch->new_mode, new_name, patch->old_mode,
2815 same ? "" : " of ", same ? "" : old_name);
2818 if (apply_data(patch, &st, ce) < 0)
2819 return error("%s: patch does not apply", name);
2820 patch->rejected = 0;
2821 return 0;
2824 static int check_patch_list(struct patch *patch)
2826 int err = 0;
2828 prepare_fn_table(patch);
2829 while (patch) {
2830 if (apply_verbosely)
2831 say_patch_name(stderr,
2832 "Checking patch ", patch, "...\n");
2833 err |= check_patch(patch);
2834 patch = patch->next;
2836 return err;
2839 /* This function tries to read the sha1 from the current index */
2840 static int get_current_sha1(const char *path, unsigned char *sha1)
2842 int pos;
2844 if (read_cache() < 0)
2845 return -1;
2846 pos = cache_name_pos(path, strlen(path));
2847 if (pos < 0)
2848 return -1;
2849 hashcpy(sha1, active_cache[pos]->sha1);
2850 return 0;
2853 /* Build an index that contains the just the files needed for a 3way merge */
2854 static void build_fake_ancestor(struct patch *list, const char *filename)
2856 struct patch *patch;
2857 struct index_state result = { NULL };
2858 int fd;
2860 /* Once we start supporting the reverse patch, it may be
2861 * worth showing the new sha1 prefix, but until then...
2863 for (patch = list; patch; patch = patch->next) {
2864 const unsigned char *sha1_ptr;
2865 unsigned char sha1[20];
2866 struct cache_entry *ce;
2867 const char *name;
2869 name = patch->old_name ? patch->old_name : patch->new_name;
2870 if (0 < patch->is_new)
2871 continue;
2872 else if (get_sha1(patch->old_sha1_prefix, sha1))
2873 /* git diff has no index line for mode/type changes */
2874 if (!patch->lines_added && !patch->lines_deleted) {
2875 if (get_current_sha1(patch->new_name, sha1) ||
2876 get_current_sha1(patch->old_name, sha1))
2877 die("mode change for %s, which is not "
2878 "in current HEAD", name);
2879 sha1_ptr = sha1;
2880 } else
2881 die("sha1 information is lacking or useless "
2882 "(%s).", name);
2883 else
2884 sha1_ptr = sha1;
2886 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);
2887 if (!ce)
2888 die("make_cache_entry failed for path '%s'", name);
2889 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
2890 die ("Could not add %s to temporary index", name);
2893 fd = open(filename, O_WRONLY | O_CREAT, 0666);
2894 if (fd < 0 || write_index(&result, fd) || close(fd))
2895 die ("Could not write temporary index to %s", filename);
2897 discard_index(&result);
2900 static void stat_patch_list(struct patch *patch)
2902 int files, adds, dels;
2904 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2905 files++;
2906 adds += patch->lines_added;
2907 dels += patch->lines_deleted;
2908 show_stats(patch);
2911 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2914 static void numstat_patch_list(struct patch *patch)
2916 for ( ; patch; patch = patch->next) {
2917 const char *name;
2918 name = patch->new_name ? patch->new_name : patch->old_name;
2919 if (patch->is_binary)
2920 printf("-\t-\t");
2921 else
2922 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2923 write_name_quoted(name, stdout, line_termination);
2927 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2929 if (mode)
2930 printf(" %s mode %06o %s\n", newdelete, mode, name);
2931 else
2932 printf(" %s %s\n", newdelete, name);
2935 static void show_mode_change(struct patch *p, int show_name)
2937 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2938 if (show_name)
2939 printf(" mode change %06o => %06o %s\n",
2940 p->old_mode, p->new_mode, p->new_name);
2941 else
2942 printf(" mode change %06o => %06o\n",
2943 p->old_mode, p->new_mode);
2947 static void show_rename_copy(struct patch *p)
2949 const char *renamecopy = p->is_rename ? "rename" : "copy";
2950 const char *old, *new;
2952 /* Find common prefix */
2953 old = p->old_name;
2954 new = p->new_name;
2955 while (1) {
2956 const char *slash_old, *slash_new;
2957 slash_old = strchr(old, '/');
2958 slash_new = strchr(new, '/');
2959 if (!slash_old ||
2960 !slash_new ||
2961 slash_old - old != slash_new - new ||
2962 memcmp(old, new, slash_new - new))
2963 break;
2964 old = slash_old + 1;
2965 new = slash_new + 1;
2967 /* p->old_name thru old is the common prefix, and old and new
2968 * through the end of names are renames
2970 if (old != p->old_name)
2971 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2972 (int)(old - p->old_name), p->old_name,
2973 old, new, p->score);
2974 else
2975 printf(" %s %s => %s (%d%%)\n", renamecopy,
2976 p->old_name, p->new_name, p->score);
2977 show_mode_change(p, 0);
2980 static void summary_patch_list(struct patch *patch)
2982 struct patch *p;
2984 for (p = patch; p; p = p->next) {
2985 if (p->is_new)
2986 show_file_mode_name("create", p->new_mode, p->new_name);
2987 else if (p->is_delete)
2988 show_file_mode_name("delete", p->old_mode, p->old_name);
2989 else {
2990 if (p->is_rename || p->is_copy)
2991 show_rename_copy(p);
2992 else {
2993 if (p->score) {
2994 printf(" rewrite %s (%d%%)\n",
2995 p->new_name, p->score);
2996 show_mode_change(p, 0);
2998 else
2999 show_mode_change(p, 1);
3005 static void patch_stats(struct patch *patch)
3007 int lines = patch->lines_added + patch->lines_deleted;
3009 if (lines > max_change)
3010 max_change = lines;
3011 if (patch->old_name) {
3012 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
3013 if (!len)
3014 len = strlen(patch->old_name);
3015 if (len > max_len)
3016 max_len = len;
3018 if (patch->new_name) {
3019 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
3020 if (!len)
3021 len = strlen(patch->new_name);
3022 if (len > max_len)
3023 max_len = len;
3027 static void remove_file(struct patch *patch, int rmdir_empty)
3029 if (update_index) {
3030 if (remove_file_from_cache(patch->old_name) < 0)
3031 die("unable to remove %s from index", patch->old_name);
3033 if (!cached) {
3034 if (S_ISGITLINK(patch->old_mode)) {
3035 if (rmdir(patch->old_name))
3036 warning("unable to remove submodule %s",
3037 patch->old_name);
3038 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {
3039 remove_path(patch->old_name);
3044 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
3046 struct stat st;
3047 struct cache_entry *ce;
3048 int namelen = strlen(path);
3049 unsigned ce_size = cache_entry_size(namelen);
3051 if (!update_index)
3052 return;
3054 ce = xcalloc(1, ce_size);
3055 memcpy(ce->name, path, namelen);
3056 ce->ce_mode = create_ce_mode(mode);
3057 ce->ce_flags = namelen;
3058 if (S_ISGITLINK(mode)) {
3059 const char *s = buf;
3061 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))
3062 die("corrupt patch for subproject %s", path);
3063 } else {
3064 if (!cached) {
3065 if (lstat(path, &st) < 0)
3066 die_errno("unable to stat newly created file '%s'",
3067 path);
3068 fill_stat_cache_info(ce, &st);
3070 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
3071 die("unable to create backing store for newly created file %s", path);
3073 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
3074 die("unable to add cache entry for %s", path);
3077 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
3079 int fd;
3080 struct strbuf nbuf = STRBUF_INIT;
3082 if (S_ISGITLINK(mode)) {
3083 struct stat st;
3084 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
3085 return 0;
3086 return mkdir(path, 0777);
3089 if (has_symlinks && S_ISLNK(mode))
3090 /* Although buf:size is counted string, it also is NUL
3091 * terminated.
3093 return symlink(buf, path);
3095 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
3096 if (fd < 0)
3097 return -1;
3099 if (convert_to_working_tree(path, buf, size, &nbuf)) {
3100 size = nbuf.len;
3101 buf = nbuf.buf;
3103 write_or_die(fd, buf, size);
3104 strbuf_release(&nbuf);
3106 if (close(fd) < 0)
3107 die_errno("closing file '%s'", path);
3108 return 0;
3112 * We optimistically assume that the directories exist,
3113 * which is true 99% of the time anyway. If they don't,
3114 * we create them and try again.
3116 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
3118 if (cached)
3119 return;
3120 if (!try_create_file(path, mode, buf, size))
3121 return;
3123 if (errno == ENOENT) {
3124 if (safe_create_leading_directories(path))
3125 return;
3126 if (!try_create_file(path, mode, buf, size))
3127 return;
3130 if (errno == EEXIST || errno == EACCES) {
3131 /* We may be trying to create a file where a directory
3132 * used to be.
3134 struct stat st;
3135 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
3136 errno = EEXIST;
3139 if (errno == EEXIST) {
3140 unsigned int nr = getpid();
3142 for (;;) {
3143 char newpath[PATH_MAX];
3144 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
3145 if (!try_create_file(newpath, mode, buf, size)) {
3146 if (!rename(newpath, path))
3147 return;
3148 unlink_or_warn(newpath);
3149 break;
3151 if (errno != EEXIST)
3152 break;
3153 ++nr;
3156 die_errno("unable to write file '%s' mode %o", path, mode);
3159 static void create_file(struct patch *patch)
3161 char *path = patch->new_name;
3162 unsigned mode = patch->new_mode;
3163 unsigned long size = patch->resultsize;
3164 char *buf = patch->result;
3166 if (!mode)
3167 mode = S_IFREG | 0644;
3168 create_one_file(path, mode, buf, size);
3169 add_index_file(path, mode, buf, size);
3172 /* phase zero is to remove, phase one is to create */
3173 static void write_out_one_result(struct patch *patch, int phase)
3175 if (patch->is_delete > 0) {
3176 if (phase == 0)
3177 remove_file(patch, 1);
3178 return;
3180 if (patch->is_new > 0 || patch->is_copy) {
3181 if (phase == 1)
3182 create_file(patch);
3183 return;
3186 * Rename or modification boils down to the same
3187 * thing: remove the old, write the new
3189 if (phase == 0)
3190 remove_file(patch, patch->is_rename);
3191 if (phase == 1)
3192 create_file(patch);
3195 static int write_out_one_reject(struct patch *patch)
3197 FILE *rej;
3198 char namebuf[PATH_MAX];
3199 struct fragment *frag;
3200 int cnt = 0;
3202 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
3203 if (!frag->rejected)
3204 continue;
3205 cnt++;
3208 if (!cnt) {
3209 if (apply_verbosely)
3210 say_patch_name(stderr,
3211 "Applied patch ", patch, " cleanly.\n");
3212 return 0;
3215 /* This should not happen, because a removal patch that leaves
3216 * contents are marked "rejected" at the patch level.
3218 if (!patch->new_name)
3219 die("internal error");
3221 /* Say this even without --verbose */
3222 say_patch_name(stderr, "Applying patch ", patch, " with");
3223 fprintf(stderr, " %d rejects...\n", cnt);
3225 cnt = strlen(patch->new_name);
3226 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
3227 cnt = ARRAY_SIZE(namebuf) - 5;
3228 warning("truncating .rej filename to %.*s.rej",
3229 cnt - 1, patch->new_name);
3231 memcpy(namebuf, patch->new_name, cnt);
3232 memcpy(namebuf + cnt, ".rej", 5);
3234 rej = fopen(namebuf, "w");
3235 if (!rej)
3236 return error("cannot open %s: %s", namebuf, strerror(errno));
3238 /* Normal git tools never deal with .rej, so do not pretend
3239 * this is a git patch by saying --git nor give extended
3240 * headers. While at it, maybe please "kompare" that wants
3241 * the trailing TAB and some garbage at the end of line ;-).
3243 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
3244 patch->new_name, patch->new_name);
3245 for (cnt = 1, frag = patch->fragments;
3246 frag;
3247 cnt++, frag = frag->next) {
3248 if (!frag->rejected) {
3249 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
3250 continue;
3252 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
3253 fprintf(rej, "%.*s", frag->size, frag->patch);
3254 if (frag->patch[frag->size-1] != '\n')
3255 fputc('\n', rej);
3257 fclose(rej);
3258 return -1;
3261 static int write_out_results(struct patch *list, int skipped_patch)
3263 int phase;
3264 int errs = 0;
3265 struct patch *l;
3267 if (!list && !skipped_patch)
3268 return error("No changes");
3270 for (phase = 0; phase < 2; phase++) {
3271 l = list;
3272 while (l) {
3273 if (l->rejected)
3274 errs = 1;
3275 else {
3276 write_out_one_result(l, phase);
3277 if (phase == 1 && write_out_one_reject(l))
3278 errs = 1;
3280 l = l->next;
3283 return errs;
3286 static struct lock_file lock_file;
3288 static struct string_list limit_by_name;
3289 static int has_include;
3290 static void add_name_limit(const char *name, int exclude)
3292 struct string_list_item *it;
3294 it = string_list_append(name, &limit_by_name);
3295 it->util = exclude ? NULL : (void *) 1;
3298 static int use_patch(struct patch *p)
3300 const char *pathname = p->new_name ? p->new_name : p->old_name;
3301 int i;
3303 /* Paths outside are not touched regardless of "--include" */
3304 if (0 < prefix_length) {
3305 int pathlen = strlen(pathname);
3306 if (pathlen <= prefix_length ||
3307 memcmp(prefix, pathname, prefix_length))
3308 return 0;
3311 /* See if it matches any of exclude/include rule */
3312 for (i = 0; i < limit_by_name.nr; i++) {
3313 struct string_list_item *it = &limit_by_name.items[i];
3314 if (!fnmatch(it->string, pathname, 0))
3315 return (it->util != NULL);
3319 * If we had any include, a path that does not match any rule is
3320 * not used. Otherwise, we saw bunch of exclude rules (or none)
3321 * and such a path is used.
3323 return !has_include;
3327 static void prefix_one(char **name)
3329 char *old_name = *name;
3330 if (!old_name)
3331 return;
3332 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));
3333 free(old_name);
3336 static void prefix_patches(struct patch *p)
3338 if (!prefix || p->is_toplevel_relative)
3339 return;
3340 for ( ; p; p = p->next) {
3341 if (p->new_name == p->old_name) {
3342 char *prefixed = p->new_name;
3343 prefix_one(&prefixed);
3344 p->new_name = p->old_name = prefixed;
3346 else {
3347 prefix_one(&p->new_name);
3348 prefix_one(&p->old_name);
3353 #define INACCURATE_EOF (1<<0)
3354 #define RECOUNT (1<<1)
3356 static int apply_patch(int fd, const char *filename, int options)
3358 size_t offset;
3359 struct strbuf buf = STRBUF_INIT;
3360 struct patch *list = NULL, **listp = &list;
3361 int skipped_patch = 0;
3363 /* FIXME - memory leak when using multiple patch files as inputs */
3364 memset(&fn_table, 0, sizeof(struct string_list));
3365 patch_input_file = filename;
3366 read_patch_file(&buf, fd);
3367 offset = 0;
3368 while (offset < buf.len) {
3369 struct patch *patch;
3370 int nr;
3372 patch = xcalloc(1, sizeof(*patch));
3373 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
3374 patch->recount = !!(options & RECOUNT);
3375 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);
3376 if (nr < 0)
3377 break;
3378 if (apply_in_reverse)
3379 reverse_patches(patch);
3380 if (prefix)
3381 prefix_patches(patch);
3382 if (use_patch(patch)) {
3383 patch_stats(patch);
3384 *listp = patch;
3385 listp = &patch->next;
3387 else {
3388 /* perhaps free it a bit better? */
3389 free(patch);
3390 skipped_patch++;
3392 offset += nr;
3395 if (whitespace_error && (ws_error_action == die_on_ws_error))
3396 apply = 0;
3398 update_index = check_index && apply;
3399 if (update_index && newfd < 0)
3400 newfd = hold_locked_index(&lock_file, 1);
3402 if (check_index) {
3403 if (read_cache() < 0)
3404 die("unable to read index file");
3407 if ((check || apply) &&
3408 check_patch_list(list) < 0 &&
3409 !apply_with_reject)
3410 exit(1);
3412 if (apply && write_out_results(list, skipped_patch))
3413 exit(1);
3415 if (fake_ancestor)
3416 build_fake_ancestor(list, fake_ancestor);
3418 if (diffstat)
3419 stat_patch_list(list);
3421 if (numstat)
3422 numstat_patch_list(list);
3424 if (summary)
3425 summary_patch_list(list);
3427 strbuf_release(&buf);
3428 return 0;
3431 static int git_apply_config(const char *var, const char *value, void *cb)
3433 if (!strcmp(var, "apply.whitespace"))
3434 return git_config_string(&apply_default_whitespace, var, value);
3435 else if (!strcmp(var, "apply.ignorewhitespace"))
3436 return git_config_string(&apply_default_ignorewhitespace, var, value);
3437 return git_default_config(var, value, cb);
3440 static int option_parse_exclude(const struct option *opt,
3441 const char *arg, int unset)
3443 add_name_limit(arg, 1);
3444 return 0;
3447 static int option_parse_include(const struct option *opt,
3448 const char *arg, int unset)
3450 add_name_limit(arg, 0);
3451 has_include = 1;
3452 return 0;
3455 static int option_parse_p(const struct option *opt,
3456 const char *arg, int unset)
3458 p_value = atoi(arg);
3459 p_value_known = 1;
3460 return 0;
3463 static int option_parse_z(const struct option *opt,
3464 const char *arg, int unset)
3466 if (unset)
3467 line_termination = '\n';
3468 else
3469 line_termination = 0;
3470 return 0;
3473 static int option_parse_space_change(const struct option *opt,
3474 const char *arg, int unset)
3476 if (unset)
3477 ws_ignore_action = ignore_ws_none;
3478 else
3479 ws_ignore_action = ignore_ws_change;
3480 return 0;
3483 static int option_parse_whitespace(const struct option *opt,
3484 const char *arg, int unset)
3486 const char **whitespace_option = opt->value;
3488 *whitespace_option = arg;
3489 parse_whitespace_option(arg);
3490 return 0;
3493 static int option_parse_directory(const struct option *opt,
3494 const char *arg, int unset)
3496 root_len = strlen(arg);
3497 if (root_len && arg[root_len - 1] != '/') {
3498 char *new_root;
3499 root = new_root = xmalloc(root_len + 2);
3500 strcpy(new_root, arg);
3501 strcpy(new_root + root_len++, "/");
3502 } else
3503 root = arg;
3504 return 0;
3507 int cmd_apply(int argc, const char **argv, const char *unused_prefix)
3509 int i;
3510 int errs = 0;
3511 int is_not_gitdir;
3512 int binary;
3513 int force_apply = 0;
3515 const char *whitespace_option = NULL;
3517 struct option builtin_apply_options[] = {
3518 { OPTION_CALLBACK, 0, "exclude", NULL, "path",
3519 "don't apply changes matching the given path",
3520 0, option_parse_exclude },
3521 { OPTION_CALLBACK, 0, "include", NULL, "path",
3522 "apply changes matching the given path",
3523 0, option_parse_include },
3524 { OPTION_CALLBACK, 'p', NULL, NULL, "num",
3525 "remove <num> leading slashes from traditional diff paths",
3526 0, option_parse_p },
3527 OPT_BOOLEAN(0, "no-add", &no_add,
3528 "ignore additions made by the patch"),
3529 OPT_BOOLEAN(0, "stat", &diffstat,
3530 "instead of applying the patch, output diffstat for the input"),
3531 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,
3532 NULL, "old option, now no-op",
3533 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3534 { OPTION_BOOLEAN, 0, "binary", &binary,
3535 NULL, "old option, now no-op",
3536 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },
3537 OPT_BOOLEAN(0, "numstat", &numstat,
3538 "shows number of added and deleted lines in decimal notation"),
3539 OPT_BOOLEAN(0, "summary", &summary,
3540 "instead of applying the patch, output a summary for the input"),
3541 OPT_BOOLEAN(0, "check", &check,
3542 "instead of applying the patch, see if the patch is applicable"),
3543 OPT_BOOLEAN(0, "index", &check_index,
3544 "make sure the patch is applicable to the current index"),
3545 OPT_BOOLEAN(0, "cached", &cached,
3546 "apply a patch without touching the working tree"),
3547 OPT_BOOLEAN(0, "apply", &force_apply,
3548 "also apply the patch (use with --stat/--summary/--check)"),
3549 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
3550 "build a temporary index based on embedded index information"),
3551 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,
3552 "paths are separated with NUL character",
3553 PARSE_OPT_NOARG, option_parse_z },
3554 OPT_INTEGER('C', NULL, &p_context,
3555 "ensure at least <n> lines of context match"),
3556 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",
3557 "detect new or modified lines that have whitespace errors",
3558 0, option_parse_whitespace },
3559 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
3560 "ignore changes in whitespace when finding context",
3561 PARSE_OPT_NOARG, option_parse_space_change },
3562 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
3563 "ignore changes in whitespace when finding context",
3564 PARSE_OPT_NOARG, option_parse_space_change },
3565 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,
3566 "apply the patch in reverse"),
3567 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,
3568 "don't expect at least one line of context"),
3569 OPT_BOOLEAN(0, "reject", &apply_with_reject,
3570 "leave the rejected hunks in corresponding *.rej files"),
3571 OPT__VERBOSE(&apply_verbosely),
3572 OPT_BIT(0, "inaccurate-eof", &options,
3573 "tolerate incorrectly detected missing new-line at the end of file",
3574 INACCURATE_EOF),
3575 OPT_BIT(0, "recount", &options,
3576 "do not trust the line counts in the hunk headers",
3577 RECOUNT),
3578 { OPTION_CALLBACK, 0, "directory", NULL, "root",
3579 "prepend <root> to all filenames",
3580 0, option_parse_directory },
3581 OPT_END()
3584 prefix = setup_git_directory_gently(&is_not_gitdir);
3585 prefix_length = prefix ? strlen(prefix) : 0;
3586 git_config(git_apply_config, NULL);
3587 if (apply_default_whitespace)
3588 parse_whitespace_option(apply_default_whitespace);
3589 if (apply_default_ignorewhitespace)
3590 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
3592 argc = parse_options(argc, argv, prefix, builtin_apply_options,
3593 apply_usage, 0);
3595 if (apply_with_reject)
3596 apply = apply_verbosely = 1;
3597 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))
3598 apply = 0;
3599 if (check_index && is_not_gitdir)
3600 die("--index outside a repository");
3601 if (cached) {
3602 if (is_not_gitdir)
3603 die("--cached outside a repository");
3604 check_index = 1;
3606 for (i = 0; i < argc; i++) {
3607 const char *arg = argv[i];
3608 int fd;
3610 if (!strcmp(arg, "-")) {
3611 errs |= apply_patch(0, "<stdin>", options);
3612 read_stdin = 0;
3613 continue;
3614 } else if (0 < prefix_length)
3615 arg = prefix_filename(prefix, prefix_length, arg);
3617 fd = open(arg, O_RDONLY);
3618 if (fd < 0)
3619 die_errno("can't open patch '%s'", arg);
3620 read_stdin = 0;
3621 set_default_whitespace_mode(whitespace_option);
3622 errs |= apply_patch(fd, arg, options);
3623 close(fd);
3625 set_default_whitespace_mode(whitespace_option);
3626 if (read_stdin)
3627 errs |= apply_patch(0, "<stdin>", options);
3628 if (whitespace_error) {
3629 if (squelch_whitespace_errors &&
3630 squelch_whitespace_errors < whitespace_error) {
3631 int squelched =
3632 whitespace_error - squelch_whitespace_errors;
3633 warning("squelched %d "
3634 "whitespace error%s",
3635 squelched,
3636 squelched == 1 ? "" : "s");
3638 if (ws_error_action == die_on_ws_error)
3639 die("%d line%s add%s whitespace errors.",
3640 whitespace_error,
3641 whitespace_error == 1 ? "" : "s",
3642 whitespace_error == 1 ? "s" : "");
3643 if (applied_after_fixing_ws && apply)
3644 warning("%d line%s applied after"
3645 " fixing whitespace errors.",
3646 applied_after_fixing_ws,
3647 applied_after_fixing_ws == 1 ? "" : "s");
3648 else if (whitespace_error)
3649 warning("%d line%s add%s whitespace errors.",
3650 whitespace_error,
3651 whitespace_error == 1 ? "" : "s",
3652 whitespace_error == 1 ? "s" : "");
3655 if (update_index) {
3656 if (write_cache(newfd, active_cache, active_nr) ||
3657 commit_locked_index(&lock_file))
3658 die("Unable to write new index file");
3661 return !!errs;