Merge branch 'rs/unit-tests-test-run'
[git/gitster.git] / apply.c
blob6e1060a952c31769eb7294d8b93e225f53278c78
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
10 #define USE_THE_REPOSITORY_VARIABLE
12 #include "git-compat-util.h"
13 #include "abspath.h"
14 #include "base85.h"
15 #include "config.h"
16 #include "object-store-ll.h"
17 #include "delta.h"
18 #include "diff.h"
19 #include "dir.h"
20 #include "environment.h"
21 #include "gettext.h"
22 #include "hex.h"
23 #include "xdiff-interface.h"
24 #include "merge-ll.h"
25 #include "lockfile.h"
26 #include "name-hash.h"
27 #include "object-name.h"
28 #include "object-file.h"
29 #include "parse-options.h"
30 #include "path.h"
31 #include "quote.h"
32 #include "read-cache.h"
33 #include "rerere.h"
34 #include "apply.h"
35 #include "entry.h"
36 #include "setup.h"
37 #include "symlinks.h"
38 #include "wildmatch.h"
39 #include "ws.h"
41 struct gitdiff_data {
42 struct strbuf *root;
43 int linenr;
44 int p_value;
47 static void git_apply_config(void)
49 git_config_get_string("apply.whitespace", &apply_default_whitespace);
50 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
51 git_config(git_xmerge_config, NULL);
54 static int parse_whitespace_option(struct apply_state *state, const char *option)
56 if (!option) {
57 state->ws_error_action = warn_on_ws_error;
58 return 0;
60 if (!strcmp(option, "warn")) {
61 state->ws_error_action = warn_on_ws_error;
62 return 0;
64 if (!strcmp(option, "nowarn")) {
65 state->ws_error_action = nowarn_ws_error;
66 return 0;
68 if (!strcmp(option, "error")) {
69 state->ws_error_action = die_on_ws_error;
70 return 0;
72 if (!strcmp(option, "error-all")) {
73 state->ws_error_action = die_on_ws_error;
74 state->squelch_whitespace_errors = 0;
75 return 0;
77 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
78 state->ws_error_action = correct_ws_error;
79 return 0;
82 * Please update $__git_whitespacelist in git-completion.bash,
83 * Documentation/git-apply.txt, and Documentation/git-am.txt
84 * when you add new options.
86 return error(_("unrecognized whitespace option '%s'"), option);
89 static int parse_ignorewhitespace_option(struct apply_state *state,
90 const char *option)
92 if (!option || !strcmp(option, "no") ||
93 !strcmp(option, "false") || !strcmp(option, "never") ||
94 !strcmp(option, "none")) {
95 state->ws_ignore_action = ignore_ws_none;
96 return 0;
98 if (!strcmp(option, "change")) {
99 state->ws_ignore_action = ignore_ws_change;
100 return 0;
102 return error(_("unrecognized whitespace ignore option '%s'"), option);
105 int init_apply_state(struct apply_state *state,
106 struct repository *repo,
107 const char *prefix)
109 memset(state, 0, sizeof(*state));
110 state->prefix = prefix;
111 state->repo = repo;
112 state->apply = 1;
113 state->line_termination = '\n';
114 state->p_value = 1;
115 state->p_context = UINT_MAX;
116 state->squelch_whitespace_errors = 5;
117 state->ws_error_action = warn_on_ws_error;
118 state->ws_ignore_action = ignore_ws_none;
119 state->linenr = 1;
120 string_list_init_nodup(&state->fn_table);
121 string_list_init_nodup(&state->limit_by_name);
122 strset_init(&state->removed_symlinks);
123 strset_init(&state->kept_symlinks);
124 strbuf_init(&state->root, 0);
126 git_apply_config();
127 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
128 return -1;
129 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
130 return -1;
131 return 0;
134 void clear_apply_state(struct apply_state *state)
136 string_list_clear(&state->limit_by_name, 0);
137 strset_clear(&state->removed_symlinks);
138 strset_clear(&state->kept_symlinks);
139 strbuf_release(&state->root);
140 FREE_AND_NULL(state->fake_ancestor);
142 /* &state->fn_table is cleared at the end of apply_patch() */
145 static void mute_routine(const char *msg UNUSED, va_list params UNUSED)
147 /* do nothing */
150 int check_apply_state(struct apply_state *state, int force_apply)
152 int is_not_gitdir = !startup_info->have_repository;
154 if (state->apply_with_reject && state->threeway)
155 return error(_("options '%s' and '%s' cannot be used together"), "--reject", "--3way");
156 if (state->threeway) {
157 if (is_not_gitdir)
158 return error(_("'%s' outside a repository"), "--3way");
159 state->check_index = 1;
161 if (state->apply_with_reject) {
162 state->apply = 1;
163 if (state->apply_verbosity == verbosity_normal)
164 state->apply_verbosity = verbosity_verbose;
166 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
167 state->apply = 0;
168 if (state->check_index && is_not_gitdir)
169 return error(_("'%s' outside a repository"), "--index");
170 if (state->cached) {
171 if (is_not_gitdir)
172 return error(_("'%s' outside a repository"), "--cached");
173 state->check_index = 1;
175 if (state->ita_only && (state->check_index || is_not_gitdir))
176 state->ita_only = 0;
177 if (state->check_index)
178 state->unsafe_paths = 0;
180 if (state->apply_verbosity <= verbosity_silent) {
181 state->saved_error_routine = get_error_routine();
182 state->saved_warn_routine = get_warn_routine();
183 set_error_routine(mute_routine);
184 set_warn_routine(mute_routine);
187 return 0;
190 static void set_default_whitespace_mode(struct apply_state *state)
192 if (!state->whitespace_option && !apply_default_whitespace)
193 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
197 * This represents one "hunk" from a patch, starting with
198 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
199 * patch text is pointed at by patch, and its byte length
200 * is stored in size. leading and trailing are the number
201 * of context lines.
203 struct fragment {
204 unsigned long leading, trailing;
205 unsigned long oldpos, oldlines;
206 unsigned long newpos, newlines;
208 * 'patch' is usually borrowed from buf in apply_patch(),
209 * but some codepaths store an allocated buffer.
211 const char *patch;
212 unsigned free_patch:1,
213 rejected:1;
214 int size;
215 int linenr;
216 struct fragment *next;
220 * When dealing with a binary patch, we reuse "leading" field
221 * to store the type of the binary hunk, either deflated "delta"
222 * or deflated "literal".
224 #define binary_patch_method leading
225 #define BINARY_DELTA_DEFLATED 1
226 #define BINARY_LITERAL_DEFLATED 2
228 static void free_fragment_list(struct fragment *list)
230 while (list) {
231 struct fragment *next = list->next;
232 if (list->free_patch)
233 free((char *)list->patch);
234 free(list);
235 list = next;
239 void release_patch(struct patch *patch)
241 free_fragment_list(patch->fragments);
242 free(patch->def_name);
243 free(patch->old_name);
244 free(patch->new_name);
245 free(patch->result);
248 static void free_patch(struct patch *patch)
250 release_patch(patch);
251 free(patch);
254 static void free_patch_list(struct patch *list)
256 while (list) {
257 struct patch *next = list->next;
258 free_patch(list);
259 list = next;
264 * A line in a file, len-bytes long (includes the terminating LF,
265 * except for an incomplete line at the end if the file ends with
266 * one), and its contents hashes to 'hash'.
268 struct line {
269 size_t len;
270 unsigned hash : 24;
271 unsigned flag : 8;
272 #define LINE_COMMON 1
273 #define LINE_PATCHED 2
277 * This represents a "file", which is an array of "lines".
279 struct image {
280 char *buf;
281 size_t len;
282 size_t nr;
283 size_t alloc;
284 struct line *line_allocated;
285 struct line *line;
288 static uint32_t hash_line(const char *cp, size_t len)
290 size_t i;
291 uint32_t h;
292 for (i = 0, h = 0; i < len; i++) {
293 if (!isspace(cp[i])) {
294 h = h * 3 + (cp[i] & 0xff);
297 return h;
301 * Compare lines s1 of length n1 and s2 of length n2, ignoring
302 * whitespace difference. Returns 1 if they match, 0 otherwise
304 static int fuzzy_matchlines(const char *s1, size_t n1,
305 const char *s2, size_t n2)
307 const char *end1 = s1 + n1;
308 const char *end2 = s2 + n2;
310 /* ignore line endings */
311 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
312 end1--;
313 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
314 end2--;
316 while (s1 < end1 && s2 < end2) {
317 if (isspace(*s1)) {
319 * Skip whitespace. We check on both buffers
320 * because we don't want "a b" to match "ab".
322 if (!isspace(*s2))
323 return 0;
324 while (s1 < end1 && isspace(*s1))
325 s1++;
326 while (s2 < end2 && isspace(*s2))
327 s2++;
328 } else if (*s1++ != *s2++)
329 return 0;
332 /* If we reached the end on one side only, lines don't match. */
333 return s1 == end1 && s2 == end2;
336 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
338 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
339 img->line_allocated[img->nr].len = len;
340 img->line_allocated[img->nr].hash = hash_line(bol, len);
341 img->line_allocated[img->nr].flag = flag;
342 img->nr++;
346 * "buf" has the file contents to be patched (read from various sources).
347 * attach it to "image" and add line-based index to it.
348 * "image" now owns the "buf".
350 static void prepare_image(struct image *image, char *buf, size_t len,
351 int prepare_linetable)
353 const char *cp, *ep;
355 memset(image, 0, sizeof(*image));
356 image->buf = buf;
357 image->len = len;
359 if (!prepare_linetable)
360 return;
362 ep = image->buf + image->len;
363 cp = image->buf;
364 while (cp < ep) {
365 const char *next;
366 for (next = cp; next < ep && *next != '\n'; next++)
368 if (next < ep)
369 next++;
370 add_line_info(image, cp, next - cp, 0);
371 cp = next;
373 image->line = image->line_allocated;
376 static void clear_image(struct image *image)
378 free(image->buf);
379 free(image->line_allocated);
380 memset(image, 0, sizeof(*image));
383 /* fmt must contain _one_ %s and no other substitution */
384 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
386 struct strbuf sb = STRBUF_INIT;
388 if (patch->old_name && patch->new_name &&
389 strcmp(patch->old_name, patch->new_name)) {
390 quote_c_style(patch->old_name, &sb, NULL, 0);
391 strbuf_addstr(&sb, " => ");
392 quote_c_style(patch->new_name, &sb, NULL, 0);
393 } else {
394 const char *n = patch->new_name;
395 if (!n)
396 n = patch->old_name;
397 quote_c_style(n, &sb, NULL, 0);
399 fprintf(output, fmt, sb.buf);
400 fputc('\n', output);
401 strbuf_release(&sb);
404 #define SLOP (16)
407 * apply.c isn't equipped to handle arbitrarily large patches, because
408 * it intermingles `unsigned long` with `int` for the type used to store
409 * buffer lengths.
411 * Only process patches that are just shy of 1 GiB large in order to
412 * avoid any truncation or overflow issues.
414 #define MAX_APPLY_SIZE (1024UL * 1024 * 1023)
416 static int read_patch_file(struct strbuf *sb, int fd)
418 if (strbuf_read(sb, fd, 0) < 0)
419 return error_errno(_("failed to read patch"));
420 else if (sb->len >= MAX_APPLY_SIZE)
421 return error(_("patch too large"));
423 * Make sure that we have some slop in the buffer
424 * so that we can do speculative "memcmp" etc, and
425 * see to it that it is NUL-filled.
427 strbuf_grow(sb, SLOP);
428 memset(sb->buf + sb->len, 0, SLOP);
429 return 0;
432 static unsigned long linelen(const char *buffer, unsigned long size)
434 unsigned long len = 0;
435 while (size--) {
436 len++;
437 if (*buffer++ == '\n')
438 break;
440 return len;
443 static int is_dev_null(const char *str)
445 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
448 #define TERM_SPACE 1
449 #define TERM_TAB 2
451 static int name_terminate(int c, int terminate)
453 if (c == ' ' && !(terminate & TERM_SPACE))
454 return 0;
455 if (c == '\t' && !(terminate & TERM_TAB))
456 return 0;
458 return 1;
461 /* remove double slashes to make --index work with such filenames */
462 static char *squash_slash(char *name)
464 int i = 0, j = 0;
466 if (!name)
467 return NULL;
469 while (name[i]) {
470 if ((name[j++] = name[i++]) == '/')
471 while (name[i] == '/')
472 i++;
474 name[j] = '\0';
475 return name;
478 static char *find_name_gnu(struct strbuf *root,
479 const char *line,
480 int p_value)
482 struct strbuf name = STRBUF_INIT;
483 char *cp;
486 * Proposed "new-style" GNU patch/diff format; see
487 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
489 if (unquote_c_style(&name, line, NULL)) {
490 strbuf_release(&name);
491 return NULL;
494 for (cp = name.buf; p_value; p_value--) {
495 cp = strchr(cp, '/');
496 if (!cp) {
497 strbuf_release(&name);
498 return NULL;
500 cp++;
503 strbuf_remove(&name, 0, cp - name.buf);
504 if (root->len)
505 strbuf_insert(&name, 0, root->buf, root->len);
506 return squash_slash(strbuf_detach(&name, NULL));
509 static size_t sane_tz_len(const char *line, size_t len)
511 const char *tz, *p;
513 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
514 return 0;
515 tz = line + len - strlen(" +0500");
517 if (tz[1] != '+' && tz[1] != '-')
518 return 0;
520 for (p = tz + 2; p != line + len; p++)
521 if (!isdigit(*p))
522 return 0;
524 return line + len - tz;
527 static size_t tz_with_colon_len(const char *line, size_t len)
529 const char *tz, *p;
531 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
532 return 0;
533 tz = line + len - strlen(" +08:00");
535 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
536 return 0;
537 p = tz + 2;
538 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
539 !isdigit(*p++) || !isdigit(*p++))
540 return 0;
542 return line + len - tz;
545 static size_t date_len(const char *line, size_t len)
547 const char *date, *p;
549 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
550 return 0;
551 p = date = line + len - strlen("72-02-05");
553 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
554 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
555 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
556 return 0;
558 if (date - line >= strlen("19") &&
559 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
560 date -= strlen("19");
562 return line + len - date;
565 static size_t short_time_len(const char *line, size_t len)
567 const char *time, *p;
569 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
570 return 0;
571 p = time = line + len - strlen(" 07:01:32");
573 /* Permit 1-digit hours? */
574 if (*p++ != ' ' ||
575 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
576 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
577 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
578 return 0;
580 return line + len - time;
583 static size_t fractional_time_len(const char *line, size_t len)
585 const char *p;
586 size_t n;
588 /* Expected format: 19:41:17.620000023 */
589 if (!len || !isdigit(line[len - 1]))
590 return 0;
591 p = line + len - 1;
593 /* Fractional seconds. */
594 while (p > line && isdigit(*p))
595 p--;
596 if (*p != '.')
597 return 0;
599 /* Hours, minutes, and whole seconds. */
600 n = short_time_len(line, p - line);
601 if (!n)
602 return 0;
604 return line + len - p + n;
607 static size_t trailing_spaces_len(const char *line, size_t len)
609 const char *p;
611 /* Expected format: ' ' x (1 or more) */
612 if (!len || line[len - 1] != ' ')
613 return 0;
615 p = line + len;
616 while (p != line) {
617 p--;
618 if (*p != ' ')
619 return line + len - (p + 1);
622 /* All spaces! */
623 return len;
626 static size_t diff_timestamp_len(const char *line, size_t len)
628 const char *end = line + len;
629 size_t n;
632 * Posix: 2010-07-05 19:41:17
633 * GNU: 2010-07-05 19:41:17.620000023 -0500
636 if (!isdigit(end[-1]))
637 return 0;
639 n = sane_tz_len(line, end - line);
640 if (!n)
641 n = tz_with_colon_len(line, end - line);
642 end -= n;
644 n = short_time_len(line, end - line);
645 if (!n)
646 n = fractional_time_len(line, end - line);
647 end -= n;
649 n = date_len(line, end - line);
650 if (!n) /* No date. Too bad. */
651 return 0;
652 end -= n;
654 if (end == line) /* No space before date. */
655 return 0;
656 if (end[-1] == '\t') { /* Success! */
657 end--;
658 return line + len - end;
660 if (end[-1] != ' ') /* No space before date. */
661 return 0;
663 /* Whitespace damage. */
664 end -= trailing_spaces_len(line, end - line);
665 return line + len - end;
668 static char *find_name_common(struct strbuf *root,
669 const char *line,
670 const char *def,
671 int p_value,
672 const char *end,
673 int terminate)
675 int len;
676 const char *start = NULL;
678 if (p_value == 0)
679 start = line;
680 while (line != end) {
681 char c = *line;
683 if (!end && isspace(c)) {
684 if (c == '\n')
685 break;
686 if (name_terminate(c, terminate))
687 break;
689 line++;
690 if (c == '/' && !--p_value)
691 start = line;
693 if (!start)
694 return squash_slash(xstrdup_or_null(def));
695 len = line - start;
696 if (!len)
697 return squash_slash(xstrdup_or_null(def));
700 * Generally we prefer the shorter name, especially
701 * if the other one is just a variation of that with
702 * something else tacked on to the end (ie "file.orig"
703 * or "file~").
705 if (def) {
706 int deflen = strlen(def);
707 if (deflen < len && !strncmp(start, def, deflen))
708 return squash_slash(xstrdup(def));
711 if (root->len) {
712 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
713 return squash_slash(ret);
716 return squash_slash(xmemdupz(start, len));
719 static char *find_name(struct strbuf *root,
720 const char *line,
721 char *def,
722 int p_value,
723 int terminate)
725 if (*line == '"') {
726 char *name = find_name_gnu(root, line, p_value);
727 if (name)
728 return name;
731 return find_name_common(root, line, def, p_value, NULL, terminate);
734 static char *find_name_traditional(struct strbuf *root,
735 const char *line,
736 char *def,
737 int p_value)
739 size_t len;
740 size_t date_len;
742 if (*line == '"') {
743 char *name = find_name_gnu(root, line, p_value);
744 if (name)
745 return name;
748 len = strchrnul(line, '\n') - line;
749 date_len = diff_timestamp_len(line, len);
750 if (!date_len)
751 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
752 len -= date_len;
754 return find_name_common(root, line, def, p_value, line + len, 0);
758 * Given the string after "--- " or "+++ ", guess the appropriate
759 * p_value for the given patch.
761 static int guess_p_value(struct apply_state *state, const char *nameline)
763 char *name, *cp;
764 int val = -1;
766 if (is_dev_null(nameline))
767 return -1;
768 name = find_name_traditional(&state->root, nameline, NULL, 0);
769 if (!name)
770 return -1;
771 cp = strchr(name, '/');
772 if (!cp)
773 val = 0;
774 else if (state->prefix) {
776 * Does it begin with "a/$our-prefix" and such? Then this is
777 * very likely to apply to our directory.
779 if (starts_with(name, state->prefix))
780 val = count_slashes(state->prefix);
781 else {
782 cp++;
783 if (starts_with(cp, state->prefix))
784 val = count_slashes(state->prefix) + 1;
787 free(name);
788 return val;
792 * Does the ---/+++ line have the POSIX timestamp after the last HT?
793 * GNU diff puts epoch there to signal a creation/deletion event. Is
794 * this such a timestamp?
796 static int has_epoch_timestamp(const char *nameline)
799 * We are only interested in epoch timestamp; any non-zero
800 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
801 * For the same reason, the date must be either 1969-12-31 or
802 * 1970-01-01, and the seconds part must be "00".
804 const char stamp_regexp[] =
805 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
807 "([-+][0-2][0-9]:?[0-5][0-9])\n";
808 const char *timestamp = NULL, *cp, *colon;
809 static regex_t *stamp;
810 regmatch_t m[10];
811 int zoneoffset, epoch_hour, hour, minute;
812 int status;
814 for (cp = nameline; *cp != '\n'; cp++) {
815 if (*cp == '\t')
816 timestamp = cp + 1;
818 if (!timestamp)
819 return 0;
822 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
823 * (west of GMT) or 1970-01-01 (east of GMT)
825 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
826 epoch_hour = 24;
827 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
828 epoch_hour = 0;
829 else
830 return 0;
832 if (!stamp) {
833 stamp = xmalloc(sizeof(*stamp));
834 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
835 warning(_("Cannot prepare timestamp regexp %s"),
836 stamp_regexp);
837 return 0;
841 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
842 if (status) {
843 if (status != REG_NOMATCH)
844 warning(_("regexec returned %d for input: %s"),
845 status, timestamp);
846 return 0;
849 hour = strtol(timestamp, NULL, 10);
850 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
852 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
853 if (*colon == ':')
854 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
855 else
856 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
857 if (timestamp[m[3].rm_so] == '-')
858 zoneoffset = -zoneoffset;
860 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
864 * Get the name etc info from the ---/+++ lines of a traditional patch header
866 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
867 * files, we can happily check the index for a match, but for creating a
868 * new file we should try to match whatever "patch" does. I have no idea.
870 static int parse_traditional_patch(struct apply_state *state,
871 const char *first,
872 const char *second,
873 struct patch *patch)
875 char *name;
877 first += 4; /* skip "--- " */
878 second += 4; /* skip "+++ " */
879 if (!state->p_value_known) {
880 int p, q;
881 p = guess_p_value(state, first);
882 q = guess_p_value(state, second);
883 if (p < 0) p = q;
884 if (0 <= p && p == q) {
885 state->p_value = p;
886 state->p_value_known = 1;
889 if (is_dev_null(first)) {
890 patch->is_new = 1;
891 patch->is_delete = 0;
892 name = find_name_traditional(&state->root, second, NULL, state->p_value);
893 patch->new_name = name;
894 } else if (is_dev_null(second)) {
895 patch->is_new = 0;
896 patch->is_delete = 1;
897 name = find_name_traditional(&state->root, first, NULL, state->p_value);
898 patch->old_name = name;
899 } else {
900 char *first_name;
901 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
902 name = find_name_traditional(&state->root, second, first_name, state->p_value);
903 free(first_name);
904 if (has_epoch_timestamp(first)) {
905 patch->is_new = 1;
906 patch->is_delete = 0;
907 patch->new_name = name;
908 } else if (has_epoch_timestamp(second)) {
909 patch->is_new = 0;
910 patch->is_delete = 1;
911 patch->old_name = name;
912 } else {
913 patch->old_name = name;
914 patch->new_name = xstrdup_or_null(name);
917 if (!name)
918 return error(_("unable to find filename in patch at line %d"), state->linenr);
920 return 0;
923 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
924 const char *line UNUSED,
925 struct patch *patch UNUSED)
927 return 1;
931 * We're anal about diff header consistency, to make
932 * sure that we don't end up having strange ambiguous
933 * patches floating around.
935 * As a result, gitdiff_{old|new}name() will check
936 * their names against any previous information, just
937 * to make sure..
939 #define DIFF_OLD_NAME 0
940 #define DIFF_NEW_NAME 1
942 static int gitdiff_verify_name(struct gitdiff_data *state,
943 const char *line,
944 int isnull,
945 char **name,
946 int side)
948 if (!*name && !isnull) {
949 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
950 return 0;
953 if (*name) {
954 char *another;
955 if (isnull)
956 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
957 *name, state->linenr);
958 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
959 if (!another || strcmp(another, *name)) {
960 free(another);
961 return error((side == DIFF_NEW_NAME) ?
962 _("git apply: bad git-diff - inconsistent new filename on line %d") :
963 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
965 free(another);
966 } else {
967 if (!is_dev_null(line))
968 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
971 return 0;
974 static int gitdiff_oldname(struct gitdiff_data *state,
975 const char *line,
976 struct patch *patch)
978 return gitdiff_verify_name(state, line,
979 patch->is_new, &patch->old_name,
980 DIFF_OLD_NAME);
983 static int gitdiff_newname(struct gitdiff_data *state,
984 const char *line,
985 struct patch *patch)
987 return gitdiff_verify_name(state, line,
988 patch->is_delete, &patch->new_name,
989 DIFF_NEW_NAME);
992 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
994 char *end;
995 *mode = strtoul(line, &end, 8);
996 if (end == line || !isspace(*end))
997 return error(_("invalid mode on line %d: %s"), linenr, line);
998 *mode = canon_mode(*mode);
999 return 0;
1002 static int gitdiff_oldmode(struct gitdiff_data *state,
1003 const char *line,
1004 struct patch *patch)
1006 return parse_mode_line(line, state->linenr, &patch->old_mode);
1009 static int gitdiff_newmode(struct gitdiff_data *state,
1010 const char *line,
1011 struct patch *patch)
1013 return parse_mode_line(line, state->linenr, &patch->new_mode);
1016 static int gitdiff_delete(struct gitdiff_data *state,
1017 const char *line,
1018 struct patch *patch)
1020 patch->is_delete = 1;
1021 free(patch->old_name);
1022 patch->old_name = xstrdup_or_null(patch->def_name);
1023 return gitdiff_oldmode(state, line, patch);
1026 static int gitdiff_newfile(struct gitdiff_data *state,
1027 const char *line,
1028 struct patch *patch)
1030 patch->is_new = 1;
1031 free(patch->new_name);
1032 patch->new_name = xstrdup_or_null(patch->def_name);
1033 return gitdiff_newmode(state, line, patch);
1036 static int gitdiff_copysrc(struct gitdiff_data *state,
1037 const char *line,
1038 struct patch *patch)
1040 patch->is_copy = 1;
1041 free(patch->old_name);
1042 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1043 return 0;
1046 static int gitdiff_copydst(struct gitdiff_data *state,
1047 const char *line,
1048 struct patch *patch)
1050 patch->is_copy = 1;
1051 free(patch->new_name);
1052 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1053 return 0;
1056 static int gitdiff_renamesrc(struct gitdiff_data *state,
1057 const char *line,
1058 struct patch *patch)
1060 patch->is_rename = 1;
1061 free(patch->old_name);
1062 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1063 return 0;
1066 static int gitdiff_renamedst(struct gitdiff_data *state,
1067 const char *line,
1068 struct patch *patch)
1070 patch->is_rename = 1;
1071 free(patch->new_name);
1072 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1073 return 0;
1076 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1077 const char *line,
1078 struct patch *patch)
1080 unsigned long val = strtoul(line, NULL, 10);
1081 if (val <= 100)
1082 patch->score = val;
1083 return 0;
1086 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1087 const char *line,
1088 struct patch *patch)
1090 unsigned long val = strtoul(line, NULL, 10);
1091 if (val <= 100)
1092 patch->score = val;
1093 return 0;
1096 static int gitdiff_index(struct gitdiff_data *state,
1097 const char *line,
1098 struct patch *patch)
1101 * index line is N hexadecimal, "..", N hexadecimal,
1102 * and optional space with octal mode.
1104 const char *ptr, *eol;
1105 int len;
1106 const unsigned hexsz = the_hash_algo->hexsz;
1108 ptr = strchr(line, '.');
1109 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1110 return 0;
1111 len = ptr - line;
1112 memcpy(patch->old_oid_prefix, line, len);
1113 patch->old_oid_prefix[len] = 0;
1115 line = ptr + 2;
1116 ptr = strchr(line, ' ');
1117 eol = strchrnul(line, '\n');
1119 if (!ptr || eol < ptr)
1120 ptr = eol;
1121 len = ptr - line;
1123 if (hexsz < len)
1124 return 0;
1125 memcpy(patch->new_oid_prefix, line, len);
1126 patch->new_oid_prefix[len] = 0;
1127 if (*ptr == ' ')
1128 return gitdiff_oldmode(state, ptr + 1, patch);
1129 return 0;
1133 * This is normal for a diff that doesn't change anything: we'll fall through
1134 * into the next diff. Tell the parser to break out.
1136 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1137 const char *line UNUSED,
1138 struct patch *patch UNUSED)
1140 return 1;
1144 * Skip p_value leading components from "line"; as we do not accept
1145 * absolute paths, return NULL in that case.
1147 static const char *skip_tree_prefix(int p_value,
1148 const char *line,
1149 int llen)
1151 int nslash;
1152 int i;
1154 if (!p_value)
1155 return (llen && line[0] == '/') ? NULL : line;
1157 nslash = p_value;
1158 for (i = 0; i < llen; i++) {
1159 int ch = line[i];
1160 if (ch == '/' && --nslash <= 0)
1161 return (i == 0) ? NULL : &line[i + 1];
1163 return NULL;
1167 * This is to extract the same name that appears on "diff --git"
1168 * line. We do not find and return anything if it is a rename
1169 * patch, and it is OK because we will find the name elsewhere.
1170 * We need to reliably find name only when it is mode-change only,
1171 * creation or deletion of an empty file. In any of these cases,
1172 * both sides are the same name under a/ and b/ respectively.
1174 static char *git_header_name(int p_value,
1175 const char *line,
1176 int llen)
1178 const char *name;
1179 const char *second = NULL;
1180 size_t len, line_len;
1182 line += strlen("diff --git ");
1183 llen -= strlen("diff --git ");
1185 if (*line == '"') {
1186 const char *cp;
1187 struct strbuf first = STRBUF_INIT;
1188 struct strbuf sp = STRBUF_INIT;
1190 if (unquote_c_style(&first, line, &second))
1191 goto free_and_fail1;
1193 /* strip the a/b prefix including trailing slash */
1194 cp = skip_tree_prefix(p_value, first.buf, first.len);
1195 if (!cp)
1196 goto free_and_fail1;
1197 strbuf_remove(&first, 0, cp - first.buf);
1200 * second points at one past closing dq of name.
1201 * find the second name.
1203 while ((second < line + llen) && isspace(*second))
1204 second++;
1206 if (line + llen <= second)
1207 goto free_and_fail1;
1208 if (*second == '"') {
1209 if (unquote_c_style(&sp, second, NULL))
1210 goto free_and_fail1;
1211 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1212 if (!cp)
1213 goto free_and_fail1;
1214 /* They must match, otherwise ignore */
1215 if (strcmp(cp, first.buf))
1216 goto free_and_fail1;
1217 strbuf_release(&sp);
1218 return strbuf_detach(&first, NULL);
1221 /* unquoted second */
1222 cp = skip_tree_prefix(p_value, second, line + llen - second);
1223 if (!cp)
1224 goto free_and_fail1;
1225 if (line + llen - cp != first.len ||
1226 memcmp(first.buf, cp, first.len))
1227 goto free_and_fail1;
1228 return strbuf_detach(&first, NULL);
1230 free_and_fail1:
1231 strbuf_release(&first);
1232 strbuf_release(&sp);
1233 return NULL;
1236 /* unquoted first name */
1237 name = skip_tree_prefix(p_value, line, llen);
1238 if (!name)
1239 return NULL;
1242 * since the first name is unquoted, a dq if exists must be
1243 * the beginning of the second name.
1245 for (second = name; second < line + llen; second++) {
1246 if (*second == '"') {
1247 struct strbuf sp = STRBUF_INIT;
1248 const char *np;
1250 if (unquote_c_style(&sp, second, NULL))
1251 goto free_and_fail2;
1253 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1254 if (!np)
1255 goto free_and_fail2;
1257 len = sp.buf + sp.len - np;
1258 if (len < second - name &&
1259 !strncmp(np, name, len) &&
1260 isspace(name[len])) {
1261 /* Good */
1262 strbuf_remove(&sp, 0, np - sp.buf);
1263 return strbuf_detach(&sp, NULL);
1266 free_and_fail2:
1267 strbuf_release(&sp);
1268 return NULL;
1273 * Accept a name only if it shows up twice, exactly the same
1274 * form.
1276 second = strchr(name, '\n');
1277 if (!second)
1278 return NULL;
1279 line_len = second - name;
1280 for (len = 0 ; ; len++) {
1281 switch (name[len]) {
1282 default:
1283 continue;
1284 case '\n':
1285 return NULL;
1286 case '\t': case ' ':
1288 * Is this the separator between the preimage
1289 * and the postimage pathname? Again, we are
1290 * only interested in the case where there is
1291 * no rename, as this is only to set def_name
1292 * and a rename patch has the names elsewhere
1293 * in an unambiguous form.
1295 if (!name[len + 1])
1296 return NULL; /* no postimage name */
1297 second = skip_tree_prefix(p_value, name + len + 1,
1298 line_len - (len + 1));
1300 * If we are at the SP at the end of a directory,
1301 * skip_tree_prefix() may return NULL as that makes
1302 * it appears as if we have an absolute path.
1303 * Keep going to find another SP.
1305 if (!second)
1306 continue;
1309 * Does len bytes starting at "name" and "second"
1310 * (that are separated by one HT or SP we just
1311 * found) exactly match?
1313 if (second[len] == '\n' && !strncmp(name, second, len))
1314 return xmemdupz(name, len);
1319 static int check_header_line(int linenr, struct patch *patch)
1321 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1322 (patch->is_rename == 1) + (patch->is_copy == 1);
1323 if (extensions > 1)
1324 return error(_("inconsistent header lines %d and %d"),
1325 patch->extension_linenr, linenr);
1326 if (extensions && !patch->extension_linenr)
1327 patch->extension_linenr = linenr;
1328 return 0;
1331 int parse_git_diff_header(struct strbuf *root,
1332 int *linenr,
1333 int p_value,
1334 const char *line,
1335 int len,
1336 unsigned int size,
1337 struct patch *patch)
1339 unsigned long offset;
1340 struct gitdiff_data parse_hdr_state;
1342 /* A git diff has explicit new/delete information, so we don't guess */
1343 patch->is_new = 0;
1344 patch->is_delete = 0;
1347 * Some things may not have the old name in the
1348 * rest of the headers anywhere (pure mode changes,
1349 * or removing or adding empty files), so we get
1350 * the default name from the header.
1352 patch->def_name = git_header_name(p_value, line, len);
1353 if (patch->def_name && root->len) {
1354 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1355 free(patch->def_name);
1356 patch->def_name = s;
1359 line += len;
1360 size -= len;
1361 (*linenr)++;
1362 parse_hdr_state.root = root;
1363 parse_hdr_state.linenr = *linenr;
1364 parse_hdr_state.p_value = p_value;
1366 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1367 static const struct opentry {
1368 const char *str;
1369 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1370 } optable[] = {
1371 { "@@ -", gitdiff_hdrend },
1372 { "--- ", gitdiff_oldname },
1373 { "+++ ", gitdiff_newname },
1374 { "old mode ", gitdiff_oldmode },
1375 { "new mode ", gitdiff_newmode },
1376 { "deleted file mode ", gitdiff_delete },
1377 { "new file mode ", gitdiff_newfile },
1378 { "copy from ", gitdiff_copysrc },
1379 { "copy to ", gitdiff_copydst },
1380 { "rename old ", gitdiff_renamesrc },
1381 { "rename new ", gitdiff_renamedst },
1382 { "rename from ", gitdiff_renamesrc },
1383 { "rename to ", gitdiff_renamedst },
1384 { "similarity index ", gitdiff_similarity },
1385 { "dissimilarity index ", gitdiff_dissimilarity },
1386 { "index ", gitdiff_index },
1387 { "", gitdiff_unrecognized },
1389 int i;
1391 len = linelen(line, size);
1392 if (!len || line[len-1] != '\n')
1393 break;
1394 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1395 const struct opentry *p = optable + i;
1396 int oplen = strlen(p->str);
1397 int res;
1398 if (len < oplen || memcmp(p->str, line, oplen))
1399 continue;
1400 res = p->fn(&parse_hdr_state, line + oplen, patch);
1401 if (res < 0)
1402 return -1;
1403 if (check_header_line(*linenr, patch))
1404 return -1;
1405 if (res > 0)
1406 goto done;
1407 break;
1411 done:
1412 if (!patch->old_name && !patch->new_name) {
1413 if (!patch->def_name) {
1414 error(Q_("git diff header lacks filename information when removing "
1415 "%d leading pathname component (line %d)",
1416 "git diff header lacks filename information when removing "
1417 "%d leading pathname components (line %d)",
1418 parse_hdr_state.p_value),
1419 parse_hdr_state.p_value, *linenr);
1420 return -128;
1422 patch->old_name = xstrdup(patch->def_name);
1423 patch->new_name = xstrdup(patch->def_name);
1425 if ((!patch->new_name && !patch->is_delete) ||
1426 (!patch->old_name && !patch->is_new)) {
1427 error(_("git diff header lacks filename information "
1428 "(line %d)"), *linenr);
1429 return -128;
1431 patch->is_toplevel_relative = 1;
1432 return offset;
1435 static int parse_num(const char *line, unsigned long *p)
1437 char *ptr;
1439 if (!isdigit(*line))
1440 return 0;
1441 *p = strtoul(line, &ptr, 10);
1442 return ptr - line;
1445 static int parse_range(const char *line, int len, int offset, const char *expect,
1446 unsigned long *p1, unsigned long *p2)
1448 int digits, ex;
1450 if (offset < 0 || offset >= len)
1451 return -1;
1452 line += offset;
1453 len -= offset;
1455 digits = parse_num(line, p1);
1456 if (!digits)
1457 return -1;
1459 offset += digits;
1460 line += digits;
1461 len -= digits;
1463 *p2 = 1;
1464 if (*line == ',') {
1465 digits = parse_num(line+1, p2);
1466 if (!digits)
1467 return -1;
1469 offset += digits+1;
1470 line += digits+1;
1471 len -= digits+1;
1474 ex = strlen(expect);
1475 if (ex > len)
1476 return -1;
1477 if (memcmp(line, expect, ex))
1478 return -1;
1480 return offset + ex;
1483 static void recount_diff(const char *line, int size, struct fragment *fragment)
1485 int oldlines = 0, newlines = 0, ret = 0;
1487 if (size < 1) {
1488 warning("recount: ignore empty hunk");
1489 return;
1492 for (;;) {
1493 int len = linelen(line, size);
1494 size -= len;
1495 line += len;
1497 if (size < 1)
1498 break;
1500 switch (*line) {
1501 case ' ': case '\n':
1502 newlines++;
1503 /* fall through */
1504 case '-':
1505 oldlines++;
1506 continue;
1507 case '+':
1508 newlines++;
1509 continue;
1510 case '\\':
1511 continue;
1512 case '@':
1513 ret = size < 3 || !starts_with(line, "@@ ");
1514 break;
1515 case 'd':
1516 ret = size < 5 || !starts_with(line, "diff ");
1517 break;
1518 default:
1519 ret = -1;
1520 break;
1522 if (ret) {
1523 warning(_("recount: unexpected line: %.*s"),
1524 (int)linelen(line, size), line);
1525 return;
1527 break;
1529 fragment->oldlines = oldlines;
1530 fragment->newlines = newlines;
1534 * Parse a unified diff fragment header of the
1535 * form "@@ -a,b +c,d @@"
1537 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1539 int offset;
1541 if (!len || line[len-1] != '\n')
1542 return -1;
1544 /* Figure out the number of lines in a fragment */
1545 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1546 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1548 return offset;
1552 * Find file diff header
1554 * Returns:
1555 * -1 if no header was found
1556 * -128 in case of error
1557 * the size of the header in bytes (called "offset") otherwise
1559 static int find_header(struct apply_state *state,
1560 const char *line,
1561 unsigned long size,
1562 int *hdrsize,
1563 struct patch *patch)
1565 unsigned long offset, len;
1567 patch->is_toplevel_relative = 0;
1568 patch->is_rename = patch->is_copy = 0;
1569 patch->is_new = patch->is_delete = -1;
1570 patch->old_mode = patch->new_mode = 0;
1571 patch->old_name = patch->new_name = NULL;
1572 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1573 unsigned long nextlen;
1575 len = linelen(line, size);
1576 if (!len)
1577 break;
1579 /* Testing this early allows us to take a few shortcuts.. */
1580 if (len < 6)
1581 continue;
1584 * Make sure we don't find any unconnected patch fragments.
1585 * That's a sign that we didn't find a header, and that a
1586 * patch has become corrupted/broken up.
1588 if (!memcmp("@@ -", line, 4)) {
1589 struct fragment dummy;
1590 if (parse_fragment_header(line, len, &dummy) < 0)
1591 continue;
1592 error(_("patch fragment without header at line %d: %.*s"),
1593 state->linenr, (int)len-1, line);
1594 return -128;
1597 if (size < len + 6)
1598 break;
1601 * Git patch? It might not have a real patch, just a rename
1602 * or mode change, so we handle that specially
1604 if (!memcmp("diff --git ", line, 11)) {
1605 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1606 state->p_value, line, len,
1607 size, patch);
1608 if (git_hdr_len < 0)
1609 return -128;
1610 if (git_hdr_len <= len)
1611 continue;
1612 *hdrsize = git_hdr_len;
1613 return offset;
1616 /* --- followed by +++ ? */
1617 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1618 continue;
1621 * We only accept unified patches, so we want it to
1622 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1623 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1625 nextlen = linelen(line + len, size - len);
1626 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1627 continue;
1629 /* Ok, we'll consider it a patch */
1630 if (parse_traditional_patch(state, line, line+len, patch))
1631 return -128;
1632 *hdrsize = len + nextlen;
1633 state->linenr += 2;
1634 return offset;
1636 return -1;
1639 static void record_ws_error(struct apply_state *state,
1640 unsigned result,
1641 const char *line,
1642 int len,
1643 int linenr)
1645 char *err;
1647 if (!result)
1648 return;
1650 state->whitespace_error++;
1651 if (state->squelch_whitespace_errors &&
1652 state->squelch_whitespace_errors < state->whitespace_error)
1653 return;
1655 err = whitespace_error_string(result);
1656 if (state->apply_verbosity > verbosity_silent)
1657 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1658 state->patch_input_file, linenr, err, len, line);
1659 free(err);
1662 static void check_whitespace(struct apply_state *state,
1663 const char *line,
1664 int len,
1665 unsigned ws_rule)
1667 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1669 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1673 * Check if the patch has context lines with CRLF or
1674 * the patch wants to remove lines with CRLF.
1676 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1678 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1679 patch->ws_rule |= WS_CR_AT_EOL;
1680 patch->crlf_in_old = 1;
1686 * Parse a unified diff. Note that this really needs to parse each
1687 * fragment separately, since the only way to know the difference
1688 * between a "---" that is part of a patch, and a "---" that starts
1689 * the next patch is to look at the line counts..
1691 static int parse_fragment(struct apply_state *state,
1692 const char *line,
1693 unsigned long size,
1694 struct patch *patch,
1695 struct fragment *fragment)
1697 int added, deleted;
1698 int len = linelen(line, size), offset;
1699 unsigned long oldlines, newlines;
1700 unsigned long leading, trailing;
1702 offset = parse_fragment_header(line, len, fragment);
1703 if (offset < 0)
1704 return -1;
1705 if (offset > 0 && patch->recount)
1706 recount_diff(line + offset, size - offset, fragment);
1707 oldlines = fragment->oldlines;
1708 newlines = fragment->newlines;
1709 leading = 0;
1710 trailing = 0;
1712 /* Parse the thing.. */
1713 line += len;
1714 size -= len;
1715 state->linenr++;
1716 added = deleted = 0;
1717 for (offset = len;
1718 0 < size;
1719 offset += len, size -= len, line += len, state->linenr++) {
1720 if (!oldlines && !newlines)
1721 break;
1722 len = linelen(line, size);
1723 if (!len || line[len-1] != '\n')
1724 return -1;
1725 switch (*line) {
1726 default:
1727 return -1;
1728 case '\n': /* newer GNU diff, an empty context line */
1729 case ' ':
1730 oldlines--;
1731 newlines--;
1732 if (!deleted && !added)
1733 leading++;
1734 trailing++;
1735 check_old_for_crlf(patch, line, len);
1736 if (!state->apply_in_reverse &&
1737 state->ws_error_action == correct_ws_error)
1738 check_whitespace(state, line, len, patch->ws_rule);
1739 break;
1740 case '-':
1741 if (!state->apply_in_reverse)
1742 check_old_for_crlf(patch, line, len);
1743 if (state->apply_in_reverse &&
1744 state->ws_error_action != nowarn_ws_error)
1745 check_whitespace(state, line, len, patch->ws_rule);
1746 deleted++;
1747 oldlines--;
1748 trailing = 0;
1749 break;
1750 case '+':
1751 if (state->apply_in_reverse)
1752 check_old_for_crlf(patch, line, len);
1753 if (!state->apply_in_reverse &&
1754 state->ws_error_action != nowarn_ws_error)
1755 check_whitespace(state, line, len, patch->ws_rule);
1756 added++;
1757 newlines--;
1758 trailing = 0;
1759 break;
1762 * We allow "\ No newline at end of file". Depending
1763 * on locale settings when the patch was produced we
1764 * don't know what this line looks like. The only
1765 * thing we do know is that it begins with "\ ".
1766 * Checking for 12 is just for sanity check -- any
1767 * l10n of "\ No newline..." is at least that long.
1769 case '\\':
1770 if (len < 12 || memcmp(line, "\\ ", 2))
1771 return -1;
1772 break;
1775 if (oldlines || newlines)
1776 return -1;
1777 if (!patch->recount && !deleted && !added)
1778 return -1;
1780 fragment->leading = leading;
1781 fragment->trailing = trailing;
1784 * If a fragment ends with an incomplete line, we failed to include
1785 * it in the above loop because we hit oldlines == newlines == 0
1786 * before seeing it.
1788 if (12 < size && !memcmp(line, "\\ ", 2))
1789 offset += linelen(line, size);
1791 patch->lines_added += added;
1792 patch->lines_deleted += deleted;
1794 if (0 < patch->is_new && oldlines)
1795 return error(_("new file depends on old contents"));
1796 if (0 < patch->is_delete && newlines)
1797 return error(_("deleted file still has contents"));
1798 return offset;
1802 * We have seen "diff --git a/... b/..." header (or a traditional patch
1803 * header). Read hunks that belong to this patch into fragments and hang
1804 * them to the given patch structure.
1806 * The (fragment->patch, fragment->size) pair points into the memory given
1807 * by the caller, not a copy, when we return.
1809 * Returns:
1810 * -1 in case of error,
1811 * the number of bytes in the patch otherwise.
1813 static int parse_single_patch(struct apply_state *state,
1814 const char *line,
1815 unsigned long size,
1816 struct patch *patch)
1818 unsigned long offset = 0;
1819 unsigned long oldlines = 0, newlines = 0, context = 0;
1820 struct fragment **fragp = &patch->fragments;
1822 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1823 struct fragment *fragment;
1824 int len;
1826 CALLOC_ARRAY(fragment, 1);
1827 fragment->linenr = state->linenr;
1828 len = parse_fragment(state, line, size, patch, fragment);
1829 if (len <= 0) {
1830 free(fragment);
1831 return error(_("corrupt patch at line %d"), state->linenr);
1833 fragment->patch = line;
1834 fragment->size = len;
1835 oldlines += fragment->oldlines;
1836 newlines += fragment->newlines;
1837 context += fragment->leading + fragment->trailing;
1839 *fragp = fragment;
1840 fragp = &fragment->next;
1842 offset += len;
1843 line += len;
1844 size -= len;
1848 * If something was removed (i.e. we have old-lines) it cannot
1849 * be creation, and if something was added it cannot be
1850 * deletion. However, the reverse is not true; --unified=0
1851 * patches that only add are not necessarily creation even
1852 * though they do not have any old lines, and ones that only
1853 * delete are not necessarily deletion.
1855 * Unfortunately, a real creation/deletion patch do _not_ have
1856 * any context line by definition, so we cannot safely tell it
1857 * apart with --unified=0 insanity. At least if the patch has
1858 * more than one hunk it is not creation or deletion.
1860 if (patch->is_new < 0 &&
1861 (oldlines || (patch->fragments && patch->fragments->next)))
1862 patch->is_new = 0;
1863 if (patch->is_delete < 0 &&
1864 (newlines || (patch->fragments && patch->fragments->next)))
1865 patch->is_delete = 0;
1867 if (0 < patch->is_new && oldlines)
1868 return error(_("new file %s depends on old contents"), patch->new_name);
1869 if (0 < patch->is_delete && newlines)
1870 return error(_("deleted file %s still has contents"), patch->old_name);
1871 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1872 fprintf_ln(stderr,
1873 _("** warning: "
1874 "file %s becomes empty but is not deleted"),
1875 patch->new_name);
1877 return offset;
1880 static inline int metadata_changes(struct patch *patch)
1882 return patch->is_rename > 0 ||
1883 patch->is_copy > 0 ||
1884 patch->is_new > 0 ||
1885 patch->is_delete ||
1886 (patch->old_mode && patch->new_mode &&
1887 patch->old_mode != patch->new_mode);
1890 static char *inflate_it(const void *data, unsigned long size,
1891 unsigned long inflated_size)
1893 git_zstream stream;
1894 void *out;
1895 int st;
1897 memset(&stream, 0, sizeof(stream));
1899 stream.next_in = (unsigned char *)data;
1900 stream.avail_in = size;
1901 stream.next_out = out = xmalloc(inflated_size);
1902 stream.avail_out = inflated_size;
1903 git_inflate_init(&stream);
1904 st = git_inflate(&stream, Z_FINISH);
1905 git_inflate_end(&stream);
1906 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1907 free(out);
1908 return NULL;
1910 return out;
1914 * Read a binary hunk and return a new fragment; fragment->patch
1915 * points at an allocated memory that the caller must free, so
1916 * it is marked as "->free_patch = 1".
1918 static struct fragment *parse_binary_hunk(struct apply_state *state,
1919 char **buf_p,
1920 unsigned long *sz_p,
1921 int *status_p,
1922 int *used_p)
1925 * Expect a line that begins with binary patch method ("literal"
1926 * or "delta"), followed by the length of data before deflating.
1927 * a sequence of 'length-byte' followed by base-85 encoded data
1928 * should follow, terminated by a newline.
1930 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1931 * and we would limit the patch line to 66 characters,
1932 * so one line can fit up to 13 groups that would decode
1933 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1934 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1936 int llen, used;
1937 unsigned long size = *sz_p;
1938 char *buffer = *buf_p;
1939 int patch_method;
1940 unsigned long origlen;
1941 char *data = NULL;
1942 int hunk_size = 0;
1943 struct fragment *frag;
1945 llen = linelen(buffer, size);
1946 used = llen;
1948 *status_p = 0;
1950 if (starts_with(buffer, "delta ")) {
1951 patch_method = BINARY_DELTA_DEFLATED;
1952 origlen = strtoul(buffer + 6, NULL, 10);
1954 else if (starts_with(buffer, "literal ")) {
1955 patch_method = BINARY_LITERAL_DEFLATED;
1956 origlen = strtoul(buffer + 8, NULL, 10);
1958 else
1959 return NULL;
1961 state->linenr++;
1962 buffer += llen;
1963 size -= llen;
1964 while (1) {
1965 int byte_length, max_byte_length, newsize;
1966 llen = linelen(buffer, size);
1967 used += llen;
1968 state->linenr++;
1969 if (llen == 1) {
1970 /* consume the blank line */
1971 buffer++;
1972 size--;
1973 break;
1976 * Minimum line is "A00000\n" which is 7-byte long,
1977 * and the line length must be multiple of 5 plus 2.
1979 if ((llen < 7) || (llen-2) % 5)
1980 goto corrupt;
1981 max_byte_length = (llen - 2) / 5 * 4;
1982 byte_length = *buffer;
1983 if ('A' <= byte_length && byte_length <= 'Z')
1984 byte_length = byte_length - 'A' + 1;
1985 else if ('a' <= byte_length && byte_length <= 'z')
1986 byte_length = byte_length - 'a' + 27;
1987 else
1988 goto corrupt;
1989 /* if the input length was not multiple of 4, we would
1990 * have filler at the end but the filler should never
1991 * exceed 3 bytes
1993 if (max_byte_length < byte_length ||
1994 byte_length <= max_byte_length - 4)
1995 goto corrupt;
1996 newsize = hunk_size + byte_length;
1997 data = xrealloc(data, newsize);
1998 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1999 goto corrupt;
2000 hunk_size = newsize;
2001 buffer += llen;
2002 size -= llen;
2005 CALLOC_ARRAY(frag, 1);
2006 frag->patch = inflate_it(data, hunk_size, origlen);
2007 frag->free_patch = 1;
2008 if (!frag->patch)
2009 goto corrupt;
2010 free(data);
2011 frag->size = origlen;
2012 *buf_p = buffer;
2013 *sz_p = size;
2014 *used_p = used;
2015 frag->binary_patch_method = patch_method;
2016 return frag;
2018 corrupt:
2019 free(data);
2020 *status_p = -1;
2021 error(_("corrupt binary patch at line %d: %.*s"),
2022 state->linenr-1, llen-1, buffer);
2023 return NULL;
2027 * Returns:
2028 * -1 in case of error,
2029 * the length of the parsed binary patch otherwise
2031 static int parse_binary(struct apply_state *state,
2032 char *buffer,
2033 unsigned long size,
2034 struct patch *patch)
2037 * We have read "GIT binary patch\n"; what follows is a line
2038 * that says the patch method (currently, either "literal" or
2039 * "delta") and the length of data before deflating; a
2040 * sequence of 'length-byte' followed by base-85 encoded data
2041 * follows.
2043 * When a binary patch is reversible, there is another binary
2044 * hunk in the same format, starting with patch method (either
2045 * "literal" or "delta") with the length of data, and a sequence
2046 * of length-byte + base-85 encoded data, terminated with another
2047 * empty line. This data, when applied to the postimage, produces
2048 * the preimage.
2050 struct fragment *forward;
2051 struct fragment *reverse;
2052 int status;
2053 int used, used_1;
2055 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2056 if (!forward && !status)
2057 /* there has to be one hunk (forward hunk) */
2058 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2059 if (status)
2060 /* otherwise we already gave an error message */
2061 return status;
2063 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2064 if (reverse)
2065 used += used_1;
2066 else if (status) {
2068 * Not having reverse hunk is not an error, but having
2069 * a corrupt reverse hunk is.
2071 free((void*) forward->patch);
2072 free(forward);
2073 return status;
2075 forward->next = reverse;
2076 patch->fragments = forward;
2077 patch->is_binary = 1;
2078 return used;
2081 static void prefix_one(struct apply_state *state, char **name)
2083 char *old_name = *name;
2084 if (!old_name)
2085 return;
2086 *name = prefix_filename(state->prefix, *name);
2087 free(old_name);
2090 static void prefix_patch(struct apply_state *state, struct patch *p)
2092 if (!state->prefix || p->is_toplevel_relative)
2093 return;
2094 prefix_one(state, &p->new_name);
2095 prefix_one(state, &p->old_name);
2099 * include/exclude
2102 static void add_name_limit(struct apply_state *state,
2103 const char *name,
2104 int exclude)
2106 struct string_list_item *it;
2108 it = string_list_append(&state->limit_by_name, name);
2109 it->util = exclude ? NULL : (void *) 1;
2112 static int use_patch(struct apply_state *state, struct patch *p)
2114 const char *pathname = p->new_name ? p->new_name : p->old_name;
2115 int i;
2117 /* Paths outside are not touched regardless of "--include" */
2118 if (state->prefix && *state->prefix) {
2119 const char *rest;
2120 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2121 return 0;
2124 /* See if it matches any of exclude/include rule */
2125 for (i = 0; i < state->limit_by_name.nr; i++) {
2126 struct string_list_item *it = &state->limit_by_name.items[i];
2127 if (!wildmatch(it->string, pathname, 0))
2128 return (it->util != NULL);
2132 * If we had any include, a path that does not match any rule is
2133 * not used. Otherwise, we saw bunch of exclude rules (or none)
2134 * and such a path is used.
2136 return !state->has_include;
2140 * Read the patch text in "buffer" that extends for "size" bytes; stop
2141 * reading after seeing a single patch (i.e. changes to a single file).
2142 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2144 * Returns:
2145 * -1 if no header was found or parse_binary() failed,
2146 * -128 on another error,
2147 * the number of bytes consumed otherwise,
2148 * so that the caller can call us again for the next patch.
2150 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2152 int hdrsize, patchsize;
2153 int offset = find_header(state, buffer, size, &hdrsize, patch);
2155 if (offset < 0)
2156 return offset;
2158 prefix_patch(state, patch);
2160 if (!use_patch(state, patch))
2161 patch->ws_rule = 0;
2162 else if (patch->new_name)
2163 patch->ws_rule = whitespace_rule(state->repo->index,
2164 patch->new_name);
2165 else
2166 patch->ws_rule = whitespace_rule(state->repo->index,
2167 patch->old_name);
2169 patchsize = parse_single_patch(state,
2170 buffer + offset + hdrsize,
2171 size - offset - hdrsize,
2172 patch);
2174 if (patchsize < 0)
2175 return -128;
2177 if (!patchsize) {
2178 static const char git_binary[] = "GIT binary patch\n";
2179 int hd = hdrsize + offset;
2180 unsigned long llen = linelen(buffer + hd, size - hd);
2182 if (llen == sizeof(git_binary) - 1 &&
2183 !memcmp(git_binary, buffer + hd, llen)) {
2184 int used;
2185 state->linenr++;
2186 used = parse_binary(state, buffer + hd + llen,
2187 size - hd - llen, patch);
2188 if (used < 0)
2189 return -1;
2190 if (used)
2191 patchsize = used + llen;
2192 else
2193 patchsize = 0;
2195 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2196 static const char *binhdr[] = {
2197 "Binary files ",
2198 "Files ",
2199 NULL,
2201 int i;
2202 for (i = 0; binhdr[i]; i++) {
2203 int len = strlen(binhdr[i]);
2204 if (len < size - hd &&
2205 !memcmp(binhdr[i], buffer + hd, len)) {
2206 state->linenr++;
2207 patch->is_binary = 1;
2208 patchsize = llen;
2209 break;
2214 /* Empty patch cannot be applied if it is a text patch
2215 * without metadata change. A binary patch appears
2216 * empty to us here.
2218 if ((state->apply || state->check) &&
2219 (!patch->is_binary && !metadata_changes(patch))) {
2220 error(_("patch with only garbage at line %d"), state->linenr);
2221 return -128;
2225 return offset + hdrsize + patchsize;
2228 static void reverse_patches(struct patch *p)
2230 for (; p; p = p->next) {
2231 struct fragment *frag = p->fragments;
2233 SWAP(p->new_name, p->old_name);
2234 if (p->new_mode)
2235 SWAP(p->new_mode, p->old_mode);
2236 SWAP(p->is_new, p->is_delete);
2237 SWAP(p->lines_added, p->lines_deleted);
2238 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2240 for (; frag; frag = frag->next) {
2241 SWAP(frag->newpos, frag->oldpos);
2242 SWAP(frag->newlines, frag->oldlines);
2247 static const char pluses[] =
2248 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2249 static const char minuses[]=
2250 "----------------------------------------------------------------------";
2252 static void show_stats(struct apply_state *state, struct patch *patch)
2254 struct strbuf qname = STRBUF_INIT;
2255 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2256 int max, add, del;
2258 quote_c_style(cp, &qname, NULL, 0);
2261 * "scale" the filename
2263 max = state->max_len;
2264 if (max > 50)
2265 max = 50;
2267 if (qname.len > max) {
2268 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2269 if (!cp)
2270 cp = qname.buf + qname.len + 3 - max;
2271 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2274 if (patch->is_binary) {
2275 printf(" %-*s | Bin\n", max, qname.buf);
2276 strbuf_release(&qname);
2277 return;
2280 printf(" %-*s |", max, qname.buf);
2281 strbuf_release(&qname);
2284 * scale the add/delete
2286 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2287 add = patch->lines_added;
2288 del = patch->lines_deleted;
2290 if (state->max_change > 0) {
2291 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2292 add = (add * max + state->max_change / 2) / state->max_change;
2293 del = total - add;
2295 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2296 add, pluses, del, minuses);
2299 static int read_old_data(struct stat *st, struct patch *patch,
2300 const char *path, struct strbuf *buf)
2302 int conv_flags = patch->crlf_in_old ?
2303 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2304 switch (st->st_mode & S_IFMT) {
2305 case S_IFLNK:
2306 if (strbuf_readlink(buf, path, st->st_size) < 0)
2307 return error(_("unable to read symlink %s"), path);
2308 return 0;
2309 case S_IFREG:
2310 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2311 return error(_("unable to open or read %s"), path);
2313 * "git apply" without "--index/--cached" should never look
2314 * at the index; the target file may not have been added to
2315 * the index yet, and we may not even be in any Git repository.
2316 * Pass NULL to convert_to_git() to stress this; the function
2317 * should never look at the index when explicit crlf option
2318 * is given.
2320 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2321 return 0;
2322 default:
2323 return -1;
2328 * Update the preimage, and the common lines in postimage,
2329 * from buffer buf of length len. If postlen is 0 the postimage
2330 * is updated in place, otherwise it's updated on a new buffer
2331 * of length postlen
2334 static void update_pre_post_images(struct image *preimage,
2335 struct image *postimage,
2336 char *buf,
2337 size_t len, size_t postlen)
2339 int i, ctx, reduced;
2340 char *new_buf, *old_buf, *fixed;
2341 struct image fixed_preimage;
2344 * Update the preimage with whitespace fixes. Note that we
2345 * are not losing preimage->buf -- apply_one_fragment() will
2346 * free "oldlines".
2348 prepare_image(&fixed_preimage, buf, len, 1);
2349 assert(postlen
2350 ? fixed_preimage.nr == preimage->nr
2351 : fixed_preimage.nr <= preimage->nr);
2352 for (i = 0; i < fixed_preimage.nr; i++)
2353 fixed_preimage.line[i].flag = preimage->line[i].flag;
2354 free(preimage->line_allocated);
2355 *preimage = fixed_preimage;
2358 * Adjust the common context lines in postimage. This can be
2359 * done in-place when we are shrinking it with whitespace
2360 * fixing, but needs a new buffer when ignoring whitespace or
2361 * expanding leading tabs to spaces.
2363 * We trust the caller to tell us if the update can be done
2364 * in place (postlen==0) or not.
2366 old_buf = postimage->buf;
2367 if (postlen)
2368 new_buf = postimage->buf = xmalloc(postlen);
2369 else
2370 new_buf = old_buf;
2371 fixed = preimage->buf;
2373 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2374 size_t l_len = postimage->line[i].len;
2375 if (!(postimage->line[i].flag & LINE_COMMON)) {
2376 /* an added line -- no counterparts in preimage */
2377 memmove(new_buf, old_buf, l_len);
2378 old_buf += l_len;
2379 new_buf += l_len;
2380 continue;
2383 /* a common context -- skip it in the original postimage */
2384 old_buf += l_len;
2386 /* and find the corresponding one in the fixed preimage */
2387 while (ctx < preimage->nr &&
2388 !(preimage->line[ctx].flag & LINE_COMMON)) {
2389 fixed += preimage->line[ctx].len;
2390 ctx++;
2394 * preimage is expected to run out, if the caller
2395 * fixed addition of trailing blank lines.
2397 if (preimage->nr <= ctx) {
2398 reduced++;
2399 continue;
2402 /* and copy it in, while fixing the line length */
2403 l_len = preimage->line[ctx].len;
2404 memcpy(new_buf, fixed, l_len);
2405 new_buf += l_len;
2406 fixed += l_len;
2407 postimage->line[i].len = l_len;
2408 ctx++;
2411 if (postlen
2412 ? postlen < new_buf - postimage->buf
2413 : postimage->len < new_buf - postimage->buf)
2414 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2415 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2417 /* Fix the length of the whole thing */
2418 postimage->len = new_buf - postimage->buf;
2419 postimage->nr -= reduced;
2422 static int line_by_line_fuzzy_match(struct image *img,
2423 struct image *preimage,
2424 struct image *postimage,
2425 unsigned long current,
2426 int current_lno,
2427 int preimage_limit)
2429 int i;
2430 size_t imgoff = 0;
2431 size_t preoff = 0;
2432 size_t postlen = postimage->len;
2433 size_t extra_chars;
2434 char *buf;
2435 char *preimage_eof;
2436 char *preimage_end;
2437 struct strbuf fixed;
2438 char *fixed_buf;
2439 size_t fixed_len;
2441 for (i = 0; i < preimage_limit; i++) {
2442 size_t prelen = preimage->line[i].len;
2443 size_t imglen = img->line[current_lno+i].len;
2445 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2446 preimage->buf + preoff, prelen))
2447 return 0;
2448 if (preimage->line[i].flag & LINE_COMMON)
2449 postlen += imglen - prelen;
2450 imgoff += imglen;
2451 preoff += prelen;
2455 * Ok, the preimage matches with whitespace fuzz.
2457 * imgoff now holds the true length of the target that
2458 * matches the preimage before the end of the file.
2460 * Count the number of characters in the preimage that fall
2461 * beyond the end of the file and make sure that all of them
2462 * are whitespace characters. (This can only happen if
2463 * we are removing blank lines at the end of the file.)
2465 buf = preimage_eof = preimage->buf + preoff;
2466 for ( ; i < preimage->nr; i++)
2467 preoff += preimage->line[i].len;
2468 preimage_end = preimage->buf + preoff;
2469 for ( ; buf < preimage_end; buf++)
2470 if (!isspace(*buf))
2471 return 0;
2474 * Update the preimage and the common postimage context
2475 * lines to use the same whitespace as the target.
2476 * If whitespace is missing in the target (i.e.
2477 * if the preimage extends beyond the end of the file),
2478 * use the whitespace from the preimage.
2480 extra_chars = preimage_end - preimage_eof;
2481 strbuf_init(&fixed, imgoff + extra_chars);
2482 strbuf_add(&fixed, img->buf + current, imgoff);
2483 strbuf_add(&fixed, preimage_eof, extra_chars);
2484 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2485 update_pre_post_images(preimage, postimage,
2486 fixed_buf, fixed_len, postlen);
2487 return 1;
2490 static int match_fragment(struct apply_state *state,
2491 struct image *img,
2492 struct image *preimage,
2493 struct image *postimage,
2494 unsigned long current,
2495 int current_lno,
2496 unsigned ws_rule,
2497 int match_beginning, int match_end)
2499 int i;
2500 const char *orig, *target;
2501 struct strbuf fixed = STRBUF_INIT;
2502 size_t postlen;
2503 int preimage_limit;
2504 int ret;
2506 if (preimage->nr + current_lno <= img->nr) {
2508 * The hunk falls within the boundaries of img.
2510 preimage_limit = preimage->nr;
2511 if (match_end && (preimage->nr + current_lno != img->nr)) {
2512 ret = 0;
2513 goto out;
2515 } else if (state->ws_error_action == correct_ws_error &&
2516 (ws_rule & WS_BLANK_AT_EOF)) {
2518 * This hunk extends beyond the end of img, and we are
2519 * removing blank lines at the end of the file. This
2520 * many lines from the beginning of the preimage must
2521 * match with img, and the remainder of the preimage
2522 * must be blank.
2524 preimage_limit = img->nr - current_lno;
2525 } else {
2527 * The hunk extends beyond the end of the img and
2528 * we are not removing blanks at the end, so we
2529 * should reject the hunk at this position.
2531 ret = 0;
2532 goto out;
2535 if (match_beginning && current_lno) {
2536 ret = 0;
2537 goto out;
2540 /* Quick hash check */
2541 for (i = 0; i < preimage_limit; i++) {
2542 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2543 (preimage->line[i].hash != img->line[current_lno + i].hash)) {
2544 ret = 0;
2545 goto out;
2549 if (preimage_limit == preimage->nr) {
2551 * Do we have an exact match? If we were told to match
2552 * at the end, size must be exactly at current+fragsize,
2553 * otherwise current+fragsize must be still within the preimage,
2554 * and either case, the old piece should match the preimage
2555 * exactly.
2557 if ((match_end
2558 ? (current + preimage->len == img->len)
2559 : (current + preimage->len <= img->len)) &&
2560 !memcmp(img->buf + current, preimage->buf, preimage->len)) {
2561 ret = 1;
2562 goto out;
2564 } else {
2566 * The preimage extends beyond the end of img, so
2567 * there cannot be an exact match.
2569 * There must be one non-blank context line that match
2570 * a line before the end of img.
2572 const char *buf, *buf_end;
2574 buf = preimage->buf;
2575 buf_end = buf;
2576 for (i = 0; i < preimage_limit; i++)
2577 buf_end += preimage->line[i].len;
2579 for ( ; buf < buf_end; buf++)
2580 if (!isspace(*buf))
2581 break;
2582 if (buf == buf_end) {
2583 ret = 0;
2584 goto out;
2589 * No exact match. If we are ignoring whitespace, run a line-by-line
2590 * fuzzy matching. We collect all the line length information because
2591 * we need it to adjust whitespace if we match.
2593 if (state->ws_ignore_action == ignore_ws_change) {
2594 ret = line_by_line_fuzzy_match(img, preimage, postimage,
2595 current, current_lno, preimage_limit);
2596 goto out;
2599 if (state->ws_error_action != correct_ws_error) {
2600 ret = 0;
2601 goto out;
2605 * The hunk does not apply byte-by-byte, but the hash says
2606 * it might with whitespace fuzz. We weren't asked to
2607 * ignore whitespace, we were asked to correct whitespace
2608 * errors, so let's try matching after whitespace correction.
2610 * While checking the preimage against the target, whitespace
2611 * errors in both fixed, we count how large the corresponding
2612 * postimage needs to be. The postimage prepared by
2613 * apply_one_fragment() has whitespace errors fixed on added
2614 * lines already, but the common lines were propagated as-is,
2615 * which may become longer when their whitespace errors are
2616 * fixed.
2619 /* First count added lines in postimage */
2620 postlen = 0;
2621 for (i = 0; i < postimage->nr; i++) {
2622 if (!(postimage->line[i].flag & LINE_COMMON))
2623 postlen += postimage->line[i].len;
2627 * The preimage may extend beyond the end of the file,
2628 * but in this loop we will only handle the part of the
2629 * preimage that falls within the file.
2631 strbuf_grow(&fixed, preimage->len + 1);
2632 orig = preimage->buf;
2633 target = img->buf + current;
2634 for (i = 0; i < preimage_limit; i++) {
2635 size_t oldlen = preimage->line[i].len;
2636 size_t tgtlen = img->line[current_lno + i].len;
2637 size_t fixstart = fixed.len;
2638 struct strbuf tgtfix;
2639 int match;
2641 /* Try fixing the line in the preimage */
2642 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2644 /* Try fixing the line in the target */
2645 strbuf_init(&tgtfix, tgtlen);
2646 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2649 * If they match, either the preimage was based on
2650 * a version before our tree fixed whitespace breakage,
2651 * or we are lacking a whitespace-fix patch the tree
2652 * the preimage was based on already had (i.e. target
2653 * has whitespace breakage, the preimage doesn't).
2654 * In either case, we are fixing the whitespace breakages
2655 * so we might as well take the fix together with their
2656 * real change.
2658 match = (tgtfix.len == fixed.len - fixstart &&
2659 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2660 fixed.len - fixstart));
2662 /* Add the length if this is common with the postimage */
2663 if (preimage->line[i].flag & LINE_COMMON)
2664 postlen += tgtfix.len;
2666 strbuf_release(&tgtfix);
2667 if (!match) {
2668 ret = 0;
2669 goto out;
2672 orig += oldlen;
2673 target += tgtlen;
2678 * Now handle the lines in the preimage that falls beyond the
2679 * end of the file (if any). They will only match if they are
2680 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2681 * false).
2683 for ( ; i < preimage->nr; i++) {
2684 size_t fixstart = fixed.len; /* start of the fixed preimage */
2685 size_t oldlen = preimage->line[i].len;
2686 int j;
2688 /* Try fixing the line in the preimage */
2689 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2691 for (j = fixstart; j < fixed.len; j++) {
2692 if (!isspace(fixed.buf[j])) {
2693 ret = 0;
2694 goto out;
2699 orig += oldlen;
2703 * Yes, the preimage is based on an older version that still
2704 * has whitespace breakages unfixed, and fixing them makes the
2705 * hunk match. Update the context lines in the postimage.
2707 if (postlen < postimage->len)
2708 postlen = 0;
2709 update_pre_post_images(preimage, postimage,
2710 fixed.buf, fixed.len, postlen);
2712 ret = 1;
2714 out:
2715 strbuf_release(&fixed);
2716 return ret;
2719 static int find_pos(struct apply_state *state,
2720 struct image *img,
2721 struct image *preimage,
2722 struct image *postimage,
2723 int line,
2724 unsigned ws_rule,
2725 int match_beginning, int match_end)
2727 int i;
2728 unsigned long backwards, forwards, current;
2729 int backwards_lno, forwards_lno, current_lno;
2732 * When running with --allow-overlap, it is possible that a hunk is
2733 * seen that pretends to start at the beginning (but no longer does),
2734 * and that *still* needs to match the end. So trust `match_end` more
2735 * than `match_beginning`.
2737 if (state->allow_overlap && match_beginning && match_end &&
2738 img->nr - preimage->nr != 0)
2739 match_beginning = 0;
2742 * If match_beginning or match_end is specified, there is no
2743 * point starting from a wrong line that will never match and
2744 * wander around and wait for a match at the specified end.
2746 if (match_beginning)
2747 line = 0;
2748 else if (match_end)
2749 line = img->nr - preimage->nr;
2752 * Because the comparison is unsigned, the following test
2753 * will also take care of a negative line number that can
2754 * result when match_end and preimage is larger than the target.
2756 if ((size_t) line > img->nr)
2757 line = img->nr;
2759 current = 0;
2760 for (i = 0; i < line; i++)
2761 current += img->line[i].len;
2764 * There's probably some smart way to do this, but I'll leave
2765 * that to the smart and beautiful people. I'm simple and stupid.
2767 backwards = current;
2768 backwards_lno = line;
2769 forwards = current;
2770 forwards_lno = line;
2771 current_lno = line;
2773 for (i = 0; ; i++) {
2774 if (match_fragment(state, img, preimage, postimage,
2775 current, current_lno, ws_rule,
2776 match_beginning, match_end))
2777 return current_lno;
2779 again:
2780 if (backwards_lno == 0 && forwards_lno == img->nr)
2781 break;
2783 if (i & 1) {
2784 if (backwards_lno == 0) {
2785 i++;
2786 goto again;
2788 backwards_lno--;
2789 backwards -= img->line[backwards_lno].len;
2790 current = backwards;
2791 current_lno = backwards_lno;
2792 } else {
2793 if (forwards_lno == img->nr) {
2794 i++;
2795 goto again;
2797 forwards += img->line[forwards_lno].len;
2798 forwards_lno++;
2799 current = forwards;
2800 current_lno = forwards_lno;
2804 return -1;
2807 static void remove_first_line(struct image *img)
2809 img->buf += img->line[0].len;
2810 img->len -= img->line[0].len;
2811 img->line++;
2812 img->nr--;
2815 static void remove_last_line(struct image *img)
2817 img->len -= img->line[--img->nr].len;
2821 * The change from "preimage" and "postimage" has been found to
2822 * apply at applied_pos (counts in line numbers) in "img".
2823 * Update "img" to remove "preimage" and replace it with "postimage".
2825 static void update_image(struct apply_state *state,
2826 struct image *img,
2827 int applied_pos,
2828 struct image *preimage,
2829 struct image *postimage)
2832 * remove the copy of preimage at offset in img
2833 * and replace it with postimage
2835 int i, nr;
2836 size_t remove_count, insert_count, applied_at = 0;
2837 char *result;
2838 int preimage_limit;
2841 * If we are removing blank lines at the end of img,
2842 * the preimage may extend beyond the end.
2843 * If that is the case, we must be careful only to
2844 * remove the part of the preimage that falls within
2845 * the boundaries of img. Initialize preimage_limit
2846 * to the number of lines in the preimage that falls
2847 * within the boundaries.
2849 preimage_limit = preimage->nr;
2850 if (preimage_limit > img->nr - applied_pos)
2851 preimage_limit = img->nr - applied_pos;
2853 for (i = 0; i < applied_pos; i++)
2854 applied_at += img->line[i].len;
2856 remove_count = 0;
2857 for (i = 0; i < preimage_limit; i++)
2858 remove_count += img->line[applied_pos + i].len;
2859 insert_count = postimage->len;
2861 /* Adjust the contents */
2862 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2863 memcpy(result, img->buf, applied_at);
2864 memcpy(result + applied_at, postimage->buf, postimage->len);
2865 memcpy(result + applied_at + postimage->len,
2866 img->buf + (applied_at + remove_count),
2867 img->len - (applied_at + remove_count));
2868 free(img->buf);
2869 img->buf = result;
2870 img->len += insert_count - remove_count;
2871 result[img->len] = '\0';
2873 /* Adjust the line table */
2874 nr = img->nr + postimage->nr - preimage_limit;
2875 if (preimage_limit < postimage->nr) {
2877 * NOTE: this knows that we never call remove_first_line()
2878 * on anything other than pre/post image.
2880 REALLOC_ARRAY(img->line, nr);
2881 img->line_allocated = img->line;
2883 if (preimage_limit != postimage->nr)
2884 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2885 img->line + applied_pos + preimage_limit,
2886 img->nr - (applied_pos + preimage_limit));
2887 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2888 if (!state->allow_overlap)
2889 for (i = 0; i < postimage->nr; i++)
2890 img->line[applied_pos + i].flag |= LINE_PATCHED;
2891 img->nr = nr;
2895 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2896 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2897 * replace the part of "img" with "postimage" text.
2899 static int apply_one_fragment(struct apply_state *state,
2900 struct image *img, struct fragment *frag,
2901 int inaccurate_eof, unsigned ws_rule,
2902 int nth_fragment)
2904 int match_beginning, match_end;
2905 const char *patch = frag->patch;
2906 int size = frag->size;
2907 char *old, *oldlines;
2908 struct strbuf newlines;
2909 int new_blank_lines_at_end = 0;
2910 int found_new_blank_lines_at_end = 0;
2911 int hunk_linenr = frag->linenr;
2912 unsigned long leading, trailing;
2913 int pos, applied_pos;
2914 struct image preimage;
2915 struct image postimage;
2917 memset(&preimage, 0, sizeof(preimage));
2918 memset(&postimage, 0, sizeof(postimage));
2919 oldlines = xmalloc(size);
2920 strbuf_init(&newlines, size);
2922 old = oldlines;
2923 while (size > 0) {
2924 char first;
2925 int len = linelen(patch, size);
2926 int plen;
2927 int added_blank_line = 0;
2928 int is_blank_context = 0;
2929 size_t start;
2931 if (!len)
2932 break;
2935 * "plen" is how much of the line we should use for
2936 * the actual patch data. Normally we just remove the
2937 * first character on the line, but if the line is
2938 * followed by "\ No newline", then we also remove the
2939 * last one (which is the newline, of course).
2941 plen = len - 1;
2942 if (len < size && patch[len] == '\\')
2943 plen--;
2944 first = *patch;
2945 if (state->apply_in_reverse) {
2946 if (first == '-')
2947 first = '+';
2948 else if (first == '+')
2949 first = '-';
2952 switch (first) {
2953 case '\n':
2954 /* Newer GNU diff, empty context line */
2955 if (plen < 0)
2956 /* ... followed by '\No newline'; nothing */
2957 break;
2958 *old++ = '\n';
2959 strbuf_addch(&newlines, '\n');
2960 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2961 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2962 is_blank_context = 1;
2963 break;
2964 case ' ':
2965 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2966 ws_blank_line(patch + 1, plen))
2967 is_blank_context = 1;
2968 /* fallthrough */
2969 case '-':
2970 memcpy(old, patch + 1, plen);
2971 add_line_info(&preimage, old, plen,
2972 (first == ' ' ? LINE_COMMON : 0));
2973 old += plen;
2974 if (first == '-')
2975 break;
2976 /* fallthrough */
2977 case '+':
2978 /* --no-add does not add new lines */
2979 if (first == '+' && state->no_add)
2980 break;
2982 start = newlines.len;
2983 if (first != '+' ||
2984 !state->whitespace_error ||
2985 state->ws_error_action != correct_ws_error) {
2986 strbuf_add(&newlines, patch + 1, plen);
2988 else {
2989 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2991 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2992 (first == '+' ? 0 : LINE_COMMON));
2993 if (first == '+' &&
2994 (ws_rule & WS_BLANK_AT_EOF) &&
2995 ws_blank_line(patch + 1, plen))
2996 added_blank_line = 1;
2997 break;
2998 case '@': case '\\':
2999 /* Ignore it, we already handled it */
3000 break;
3001 default:
3002 if (state->apply_verbosity > verbosity_normal)
3003 error(_("invalid start of line: '%c'"), first);
3004 applied_pos = -1;
3005 goto out;
3007 if (added_blank_line) {
3008 if (!new_blank_lines_at_end)
3009 found_new_blank_lines_at_end = hunk_linenr;
3010 new_blank_lines_at_end++;
3012 else if (is_blank_context)
3014 else
3015 new_blank_lines_at_end = 0;
3016 patch += len;
3017 size -= len;
3018 hunk_linenr++;
3020 if (inaccurate_eof &&
3021 old > oldlines && old[-1] == '\n' &&
3022 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
3023 old--;
3024 strbuf_setlen(&newlines, newlines.len - 1);
3025 preimage.line_allocated[preimage.nr - 1].len--;
3026 postimage.line_allocated[postimage.nr - 1].len--;
3029 leading = frag->leading;
3030 trailing = frag->trailing;
3033 * A hunk to change lines at the beginning would begin with
3034 * @@ -1,L +N,M @@
3035 * but we need to be careful. -U0 that inserts before the second
3036 * line also has this pattern.
3038 * And a hunk to add to an empty file would begin with
3039 * @@ -0,0 +N,M @@
3041 * In other words, a hunk that is (frag->oldpos <= 1) with or
3042 * without leading context must match at the beginning.
3044 match_beginning = (!frag->oldpos ||
3045 (frag->oldpos == 1 && !state->unidiff_zero));
3048 * A hunk without trailing lines must match at the end.
3049 * However, we simply cannot tell if a hunk must match end
3050 * from the lack of trailing lines if the patch was generated
3051 * with unidiff without any context.
3053 match_end = !state->unidiff_zero && !trailing;
3055 pos = frag->newpos ? (frag->newpos - 1) : 0;
3056 preimage.buf = oldlines;
3057 preimage.len = old - oldlines;
3058 postimage.buf = newlines.buf;
3059 postimage.len = newlines.len;
3060 preimage.line = preimage.line_allocated;
3061 postimage.line = postimage.line_allocated;
3063 for (;;) {
3065 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3066 ws_rule, match_beginning, match_end);
3068 if (applied_pos >= 0)
3069 break;
3071 /* Am I at my context limits? */
3072 if ((leading <= state->p_context) && (trailing <= state->p_context))
3073 break;
3074 if (match_beginning || match_end) {
3075 match_beginning = match_end = 0;
3076 continue;
3080 * Reduce the number of context lines; reduce both
3081 * leading and trailing if they are equal otherwise
3082 * just reduce the larger context.
3084 if (leading >= trailing) {
3085 remove_first_line(&preimage);
3086 remove_first_line(&postimage);
3087 pos--;
3088 leading--;
3090 if (trailing > leading) {
3091 remove_last_line(&preimage);
3092 remove_last_line(&postimage);
3093 trailing--;
3097 if (applied_pos >= 0) {
3098 if (new_blank_lines_at_end &&
3099 preimage.nr + applied_pos >= img->nr &&
3100 (ws_rule & WS_BLANK_AT_EOF) &&
3101 state->ws_error_action != nowarn_ws_error) {
3102 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3103 found_new_blank_lines_at_end);
3104 if (state->ws_error_action == correct_ws_error) {
3105 while (new_blank_lines_at_end--)
3106 remove_last_line(&postimage);
3109 * We would want to prevent write_out_results()
3110 * from taking place in apply_patch() that follows
3111 * the callchain led us here, which is:
3112 * apply_patch->check_patch_list->check_patch->
3113 * apply_data->apply_fragments->apply_one_fragment
3115 if (state->ws_error_action == die_on_ws_error)
3116 state->apply = 0;
3119 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3120 int offset = applied_pos - pos;
3121 if (state->apply_in_reverse)
3122 offset = 0 - offset;
3123 fprintf_ln(stderr,
3124 Q_("Hunk #%d succeeded at %d (offset %d line).",
3125 "Hunk #%d succeeded at %d (offset %d lines).",
3126 offset),
3127 nth_fragment, applied_pos + 1, offset);
3131 * Warn if it was necessary to reduce the number
3132 * of context lines.
3134 if ((leading != frag->leading ||
3135 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3136 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3137 " to apply fragment at %d"),
3138 leading, trailing, applied_pos+1);
3139 update_image(state, img, applied_pos, &preimage, &postimage);
3140 } else {
3141 if (state->apply_verbosity > verbosity_normal)
3142 error(_("while searching for:\n%.*s"),
3143 (int)(old - oldlines), oldlines);
3146 out:
3147 free(oldlines);
3148 strbuf_release(&newlines);
3149 free(preimage.line_allocated);
3150 free(postimage.line_allocated);
3152 return (applied_pos < 0);
3155 static int apply_binary_fragment(struct apply_state *state,
3156 struct image *img,
3157 struct patch *patch)
3159 struct fragment *fragment = patch->fragments;
3160 unsigned long len;
3161 void *dst;
3163 if (!fragment)
3164 return error(_("missing binary patch data for '%s'"),
3165 patch->new_name ?
3166 patch->new_name :
3167 patch->old_name);
3169 /* Binary patch is irreversible without the optional second hunk */
3170 if (state->apply_in_reverse) {
3171 if (!fragment->next)
3172 return error(_("cannot reverse-apply a binary patch "
3173 "without the reverse hunk to '%s'"),
3174 patch->new_name
3175 ? patch->new_name : patch->old_name);
3176 fragment = fragment->next;
3178 switch (fragment->binary_patch_method) {
3179 case BINARY_DELTA_DEFLATED:
3180 dst = patch_delta(img->buf, img->len, fragment->patch,
3181 fragment->size, &len);
3182 if (!dst)
3183 return -1;
3184 clear_image(img);
3185 img->buf = dst;
3186 img->len = len;
3187 return 0;
3188 case BINARY_LITERAL_DEFLATED:
3189 clear_image(img);
3190 img->len = fragment->size;
3191 img->buf = xmemdupz(fragment->patch, img->len);
3192 return 0;
3194 return -1;
3198 * Replace "img" with the result of applying the binary patch.
3199 * The binary patch data itself in patch->fragment is still kept
3200 * but the preimage prepared by the caller in "img" is freed here
3201 * or in the helper function apply_binary_fragment() this calls.
3203 static int apply_binary(struct apply_state *state,
3204 struct image *img,
3205 struct patch *patch)
3207 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3208 struct object_id oid;
3209 const unsigned hexsz = the_hash_algo->hexsz;
3212 * For safety, we require patch index line to contain
3213 * full hex textual object ID for old and new, at least for now.
3215 if (strlen(patch->old_oid_prefix) != hexsz ||
3216 strlen(patch->new_oid_prefix) != hexsz ||
3217 get_oid_hex(patch->old_oid_prefix, &oid) ||
3218 get_oid_hex(patch->new_oid_prefix, &oid))
3219 return error(_("cannot apply binary patch to '%s' "
3220 "without full index line"), name);
3222 if (patch->old_name) {
3224 * See if the old one matches what the patch
3225 * applies to.
3227 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3228 &oid);
3229 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3230 return error(_("the patch applies to '%s' (%s), "
3231 "which does not match the "
3232 "current contents."),
3233 name, oid_to_hex(&oid));
3235 else {
3236 /* Otherwise, the old one must be empty. */
3237 if (img->len)
3238 return error(_("the patch applies to an empty "
3239 "'%s' but it is not empty"), name);
3242 get_oid_hex(patch->new_oid_prefix, &oid);
3243 if (is_null_oid(&oid)) {
3244 clear_image(img);
3245 return 0; /* deletion patch */
3248 if (has_object(the_repository, &oid, 0)) {
3249 /* We already have the postimage */
3250 enum object_type type;
3251 unsigned long size;
3252 char *result;
3254 result = repo_read_object_file(the_repository, &oid, &type,
3255 &size);
3256 if (!result)
3257 return error(_("the necessary postimage %s for "
3258 "'%s' cannot be read"),
3259 patch->new_oid_prefix, name);
3260 clear_image(img);
3261 img->buf = result;
3262 img->len = size;
3263 } else {
3265 * We have verified buf matches the preimage;
3266 * apply the patch data to it, which is stored
3267 * in the patch->fragments->{patch,size}.
3269 if (apply_binary_fragment(state, img, patch))
3270 return error(_("binary patch does not apply to '%s'"),
3271 name);
3273 /* verify that the result matches */
3274 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3275 &oid);
3276 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3277 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3278 name, patch->new_oid_prefix, oid_to_hex(&oid));
3281 return 0;
3284 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3286 struct fragment *frag = patch->fragments;
3287 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3288 unsigned ws_rule = patch->ws_rule;
3289 unsigned inaccurate_eof = patch->inaccurate_eof;
3290 int nth = 0;
3292 if (patch->is_binary)
3293 return apply_binary(state, img, patch);
3295 while (frag) {
3296 nth++;
3297 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3298 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3299 if (!state->apply_with_reject)
3300 return -1;
3301 frag->rejected = 1;
3303 frag = frag->next;
3305 return 0;
3308 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3310 if (S_ISGITLINK(mode)) {
3311 strbuf_grow(buf, 100);
3312 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3313 } else {
3314 enum object_type type;
3315 unsigned long sz;
3316 char *result;
3318 result = repo_read_object_file(the_repository, oid, &type,
3319 &sz);
3320 if (!result)
3321 return -1;
3322 /* XXX read_sha1_file NUL-terminates */
3323 strbuf_attach(buf, result, sz, sz + 1);
3325 return 0;
3328 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3330 if (!ce)
3331 return 0;
3332 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3335 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3337 struct string_list_item *item;
3339 if (!name)
3340 return NULL;
3342 item = string_list_lookup(&state->fn_table, name);
3343 if (item)
3344 return (struct patch *)item->util;
3346 return NULL;
3350 * item->util in the filename table records the status of the path.
3351 * Usually it points at a patch (whose result records the contents
3352 * of it after applying it), but it could be PATH_WAS_DELETED for a
3353 * path that a previously applied patch has already removed, or
3354 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3356 * The latter is needed to deal with a case where two paths A and B
3357 * are swapped by first renaming A to B and then renaming B to A;
3358 * moving A to B should not be prevented due to presence of B as we
3359 * will remove it in a later patch.
3361 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3362 #define PATH_WAS_DELETED ((struct patch *) -1)
3364 static int to_be_deleted(struct patch *patch)
3366 return patch == PATH_TO_BE_DELETED;
3369 static int was_deleted(struct patch *patch)
3371 return patch == PATH_WAS_DELETED;
3374 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3376 struct string_list_item *item;
3379 * Always add new_name unless patch is a deletion
3380 * This should cover the cases for normal diffs,
3381 * file creations and copies
3383 if (patch->new_name) {
3384 item = string_list_insert(&state->fn_table, patch->new_name);
3385 item->util = patch;
3389 * store a failure on rename/deletion cases because
3390 * later chunks shouldn't patch old names
3392 if ((patch->new_name == NULL) || (patch->is_rename)) {
3393 item = string_list_insert(&state->fn_table, patch->old_name);
3394 item->util = PATH_WAS_DELETED;
3398 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3401 * store information about incoming file deletion
3403 while (patch) {
3404 if ((patch->new_name == NULL) || (patch->is_rename)) {
3405 struct string_list_item *item;
3406 item = string_list_insert(&state->fn_table, patch->old_name);
3407 item->util = PATH_TO_BE_DELETED;
3409 patch = patch->next;
3413 static int checkout_target(struct index_state *istate,
3414 struct cache_entry *ce, struct stat *st)
3416 struct checkout costate = CHECKOUT_INIT;
3418 costate.refresh_cache = 1;
3419 costate.istate = istate;
3420 if (checkout_entry(ce, &costate, NULL, NULL) ||
3421 lstat(ce->name, st))
3422 return error(_("cannot checkout %s"), ce->name);
3423 return 0;
3426 static struct patch *previous_patch(struct apply_state *state,
3427 struct patch *patch,
3428 int *gone)
3430 struct patch *previous;
3432 *gone = 0;
3433 if (patch->is_copy || patch->is_rename)
3434 return NULL; /* "git" patches do not depend on the order */
3436 previous = in_fn_table(state, patch->old_name);
3437 if (!previous)
3438 return NULL;
3440 if (to_be_deleted(previous))
3441 return NULL; /* the deletion hasn't happened yet */
3443 if (was_deleted(previous))
3444 *gone = 1;
3446 return previous;
3449 static int verify_index_match(struct apply_state *state,
3450 const struct cache_entry *ce,
3451 struct stat *st)
3453 if (S_ISGITLINK(ce->ce_mode)) {
3454 if (!S_ISDIR(st->st_mode))
3455 return -1;
3456 return 0;
3458 return ie_match_stat(state->repo->index, ce, st,
3459 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3462 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3464 static int load_patch_target(struct apply_state *state,
3465 struct strbuf *buf,
3466 const struct cache_entry *ce,
3467 struct stat *st,
3468 struct patch *patch,
3469 const char *name,
3470 unsigned expected_mode)
3472 if (state->cached || state->check_index) {
3473 if (read_file_or_gitlink(ce, buf))
3474 return error(_("failed to read %s"), name);
3475 } else if (name) {
3476 if (S_ISGITLINK(expected_mode)) {
3477 if (ce)
3478 return read_file_or_gitlink(ce, buf);
3479 else
3480 return SUBMODULE_PATCH_WITHOUT_INDEX;
3481 } else if (has_symlink_leading_path(name, strlen(name))) {
3482 return error(_("reading from '%s' beyond a symbolic link"), name);
3483 } else {
3484 if (read_old_data(st, patch, name, buf))
3485 return error(_("failed to read %s"), name);
3488 return 0;
3492 * We are about to apply "patch"; populate the "image" with the
3493 * current version we have, from the working tree or from the index,
3494 * depending on the situation e.g. --cached/--index. If we are
3495 * applying a non-git patch that incrementally updates the tree,
3496 * we read from the result of a previous diff.
3498 static int load_preimage(struct apply_state *state,
3499 struct image *image,
3500 struct patch *patch, struct stat *st,
3501 const struct cache_entry *ce)
3503 struct strbuf buf = STRBUF_INIT;
3504 size_t len;
3505 char *img;
3506 struct patch *previous;
3507 int status;
3509 previous = previous_patch(state, patch, &status);
3510 if (status)
3511 return error(_("path %s has been renamed/deleted"),
3512 patch->old_name);
3513 if (previous) {
3514 /* We have a patched copy in memory; use that. */
3515 strbuf_add(&buf, previous->result, previous->resultsize);
3516 } else {
3517 status = load_patch_target(state, &buf, ce, st, patch,
3518 patch->old_name, patch->old_mode);
3519 if (status < 0)
3520 return status;
3521 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3523 * There is no way to apply subproject
3524 * patch without looking at the index.
3525 * NEEDSWORK: shouldn't this be flagged
3526 * as an error???
3528 free_fragment_list(patch->fragments);
3529 patch->fragments = NULL;
3530 } else if (status) {
3531 return error(_("failed to read %s"), patch->old_name);
3535 img = strbuf_detach(&buf, &len);
3536 prepare_image(image, img, len, !patch->is_binary);
3537 return 0;
3540 static int resolve_to(struct image *image, const struct object_id *result_id)
3542 unsigned long size;
3543 enum object_type type;
3545 clear_image(image);
3547 image->buf = repo_read_object_file(the_repository, result_id, &type,
3548 &size);
3549 if (!image->buf || type != OBJ_BLOB)
3550 die("unable to read blob object %s", oid_to_hex(result_id));
3551 image->len = size;
3553 return 0;
3556 static int three_way_merge(struct apply_state *state,
3557 struct image *image,
3558 char *path,
3559 const struct object_id *base,
3560 const struct object_id *ours,
3561 const struct object_id *theirs)
3563 mmfile_t base_file, our_file, their_file;
3564 mmbuffer_t result = { NULL };
3565 enum ll_merge_result status;
3567 /* resolve trivial cases first */
3568 if (oideq(base, ours))
3569 return resolve_to(image, theirs);
3570 else if (oideq(base, theirs) || oideq(ours, theirs))
3571 return resolve_to(image, ours);
3573 read_mmblob(&base_file, base);
3574 read_mmblob(&our_file, ours);
3575 read_mmblob(&their_file, theirs);
3576 status = ll_merge(&result, path,
3577 &base_file, "base",
3578 &our_file, "ours",
3579 &their_file, "theirs",
3580 state->repo->index,
3581 NULL);
3582 if (status == LL_MERGE_BINARY_CONFLICT)
3583 warning("Cannot merge binary files: %s (%s vs. %s)",
3584 path, "ours", "theirs");
3585 free(base_file.ptr);
3586 free(our_file.ptr);
3587 free(their_file.ptr);
3588 if (status < 0 || !result.ptr) {
3589 free(result.ptr);
3590 return -1;
3592 clear_image(image);
3593 image->buf = result.ptr;
3594 image->len = result.size;
3596 return status;
3600 * When directly falling back to add/add three-way merge, we read from
3601 * the current contents of the new_name. In no cases other than that
3602 * this function will be called.
3604 static int load_current(struct apply_state *state,
3605 struct image *image,
3606 struct patch *patch)
3608 struct strbuf buf = STRBUF_INIT;
3609 int status, pos;
3610 size_t len;
3611 char *img;
3612 struct stat st;
3613 struct cache_entry *ce;
3614 char *name = patch->new_name;
3615 unsigned mode = patch->new_mode;
3617 if (!patch->is_new)
3618 BUG("patch to %s is not a creation", patch->old_name);
3620 pos = index_name_pos(state->repo->index, name, strlen(name));
3621 if (pos < 0)
3622 return error(_("%s: does not exist in index"), name);
3623 ce = state->repo->index->cache[pos];
3624 if (lstat(name, &st)) {
3625 if (errno != ENOENT)
3626 return error_errno("%s", name);
3627 if (checkout_target(state->repo->index, ce, &st))
3628 return -1;
3630 if (verify_index_match(state, ce, &st))
3631 return error(_("%s: does not match index"), name);
3633 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3634 if (status < 0)
3635 return status;
3636 else if (status)
3637 return -1;
3638 img = strbuf_detach(&buf, &len);
3639 prepare_image(image, img, len, !patch->is_binary);
3640 return 0;
3643 static int try_threeway(struct apply_state *state,
3644 struct image *image,
3645 struct patch *patch,
3646 struct stat *st,
3647 const struct cache_entry *ce)
3649 struct object_id pre_oid, post_oid, our_oid;
3650 struct strbuf buf = STRBUF_INIT;
3651 size_t len;
3652 int status;
3653 char *img;
3654 struct image tmp_image;
3656 /* No point falling back to 3-way merge in these cases */
3657 if (patch->is_delete ||
3658 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3659 (patch->is_new && !patch->direct_to_threeway) ||
3660 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3661 return -1;
3663 /* Preimage the patch was prepared for */
3664 if (patch->is_new)
3665 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3666 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3667 read_blob_object(&buf, &pre_oid, patch->old_mode))
3668 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3670 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3671 fprintf(stderr, _("Performing three-way merge...\n"));
3673 img = strbuf_detach(&buf, &len);
3674 prepare_image(&tmp_image, img, len, 1);
3675 /* Apply the patch to get the post image */
3676 if (apply_fragments(state, &tmp_image, patch) < 0) {
3677 clear_image(&tmp_image);
3678 return -1;
3680 /* post_oid is theirs */
3681 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3682 clear_image(&tmp_image);
3684 /* our_oid is ours */
3685 if (patch->is_new) {
3686 if (load_current(state, &tmp_image, patch))
3687 return error(_("cannot read the current contents of '%s'"),
3688 patch->new_name);
3689 } else {
3690 if (load_preimage(state, &tmp_image, patch, st, ce))
3691 return error(_("cannot read the current contents of '%s'"),
3692 patch->old_name);
3694 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3695 clear_image(&tmp_image);
3697 /* in-core three-way merge between post and our using pre as base */
3698 status = three_way_merge(state, image, patch->new_name,
3699 &pre_oid, &our_oid, &post_oid);
3700 if (status < 0) {
3701 if (state->apply_verbosity > verbosity_silent)
3702 fprintf(stderr,
3703 _("Failed to perform three-way merge...\n"));
3704 return status;
3707 if (status) {
3708 patch->conflicted_threeway = 1;
3709 if (patch->is_new)
3710 oidclr(&patch->threeway_stage[0], the_repository->hash_algo);
3711 else
3712 oidcpy(&patch->threeway_stage[0], &pre_oid);
3713 oidcpy(&patch->threeway_stage[1], &our_oid);
3714 oidcpy(&patch->threeway_stage[2], &post_oid);
3715 if (state->apply_verbosity > verbosity_silent)
3716 fprintf(stderr,
3717 _("Applied patch to '%s' with conflicts.\n"),
3718 patch->new_name);
3719 } else {
3720 if (state->apply_verbosity > verbosity_silent)
3721 fprintf(stderr,
3722 _("Applied patch to '%s' cleanly.\n"),
3723 patch->new_name);
3725 return 0;
3728 static int apply_data(struct apply_state *state, struct patch *patch,
3729 struct stat *st, const struct cache_entry *ce)
3731 struct image image;
3733 if (load_preimage(state, &image, patch, st, ce) < 0)
3734 return -1;
3736 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3737 if (state->apply_verbosity > verbosity_silent &&
3738 state->threeway && !patch->direct_to_threeway)
3739 fprintf(stderr, _("Falling back to direct application...\n"));
3741 /* Note: with --reject, apply_fragments() returns 0 */
3742 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0) {
3743 clear_image(&image);
3744 return -1;
3747 patch->result = image.buf;
3748 patch->resultsize = image.len;
3749 add_to_fn_table(state, patch);
3750 free(image.line_allocated);
3752 if (0 < patch->is_delete && patch->resultsize)
3753 return error(_("removal patch leaves file contents"));
3755 return 0;
3759 * If "patch" that we are looking at modifies or deletes what we have,
3760 * we would want it not to lose any local modification we have, either
3761 * in the working tree or in the index.
3763 * This also decides if a non-git patch is a creation patch or a
3764 * modification to an existing empty file. We do not check the state
3765 * of the current tree for a creation patch in this function; the caller
3766 * check_patch() separately makes sure (and errors out otherwise) that
3767 * the path the patch creates does not exist in the current tree.
3769 static int check_preimage(struct apply_state *state,
3770 struct patch *patch,
3771 struct cache_entry **ce,
3772 struct stat *st)
3774 const char *old_name = patch->old_name;
3775 struct patch *previous = NULL;
3776 int stat_ret = 0, status;
3777 unsigned st_mode = 0;
3779 if (!old_name)
3780 return 0;
3782 assert(patch->is_new <= 0);
3783 previous = previous_patch(state, patch, &status);
3785 if (status)
3786 return error(_("path %s has been renamed/deleted"), old_name);
3787 if (previous) {
3788 st_mode = previous->new_mode;
3789 } else if (!state->cached) {
3790 stat_ret = lstat(old_name, st);
3791 if (stat_ret && errno != ENOENT)
3792 return error_errno("%s", old_name);
3795 if (state->check_index && !previous) {
3796 int pos = index_name_pos(state->repo->index, old_name,
3797 strlen(old_name));
3798 if (pos < 0) {
3799 if (patch->is_new < 0)
3800 goto is_new;
3801 return error(_("%s: does not exist in index"), old_name);
3803 *ce = state->repo->index->cache[pos];
3804 if (stat_ret < 0) {
3805 if (checkout_target(state->repo->index, *ce, st))
3806 return -1;
3808 if (!state->cached && verify_index_match(state, *ce, st))
3809 return error(_("%s: does not match index"), old_name);
3810 if (state->cached)
3811 st_mode = (*ce)->ce_mode;
3812 } else if (stat_ret < 0) {
3813 if (patch->is_new < 0)
3814 goto is_new;
3815 return error_errno("%s", old_name);
3818 if (!state->cached && !previous) {
3819 if (*ce && !(*ce)->ce_mode)
3820 BUG("ce_mode == 0 for path '%s'", old_name);
3822 if (trust_executable_bit)
3823 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3824 else if (*ce)
3825 st_mode = (*ce)->ce_mode;
3826 else
3827 st_mode = patch->old_mode;
3830 if (patch->is_new < 0)
3831 patch->is_new = 0;
3832 if (!patch->old_mode)
3833 patch->old_mode = st_mode;
3834 if ((st_mode ^ patch->old_mode) & S_IFMT)
3835 return error(_("%s: wrong type"), old_name);
3836 if (st_mode != patch->old_mode)
3837 warning(_("%s has type %o, expected %o"),
3838 old_name, st_mode, patch->old_mode);
3839 if (!patch->new_mode && !patch->is_delete)
3840 patch->new_mode = st_mode;
3841 return 0;
3843 is_new:
3844 patch->is_new = 1;
3845 patch->is_delete = 0;
3846 FREE_AND_NULL(patch->old_name);
3847 return 0;
3851 #define EXISTS_IN_INDEX 1
3852 #define EXISTS_IN_WORKTREE 2
3853 #define EXISTS_IN_INDEX_AS_ITA 3
3855 static int check_to_create(struct apply_state *state,
3856 const char *new_name,
3857 int ok_if_exists)
3859 struct stat nst;
3861 if (state->check_index && (!ok_if_exists || !state->cached)) {
3862 int pos;
3864 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3865 if (pos >= 0) {
3866 struct cache_entry *ce = state->repo->index->cache[pos];
3868 /* allow ITA, as they do not yet exist in the index */
3869 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3870 return EXISTS_IN_INDEX;
3872 /* ITA entries can never match working tree files */
3873 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3874 return EXISTS_IN_INDEX_AS_ITA;
3878 if (state->cached)
3879 return 0;
3881 if (!lstat(new_name, &nst)) {
3882 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3883 return 0;
3885 * A leading component of new_name might be a symlink
3886 * that is going to be removed with this patch, but
3887 * still pointing at somewhere that has the path.
3888 * In such a case, path "new_name" does not exist as
3889 * far as git is concerned.
3891 if (has_symlink_leading_path(new_name, strlen(new_name)))
3892 return 0;
3894 return EXISTS_IN_WORKTREE;
3895 } else if (!is_missing_file_error(errno)) {
3896 return error_errno("%s", new_name);
3898 return 0;
3901 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3903 for ( ; patch; patch = patch->next) {
3904 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3905 (patch->is_rename || patch->is_delete))
3906 /* the symlink at patch->old_name is removed */
3907 strset_add(&state->removed_symlinks, patch->old_name);
3909 if (patch->new_name && S_ISLNK(patch->new_mode))
3910 /* the symlink at patch->new_name is created or remains */
3911 strset_add(&state->kept_symlinks, patch->new_name);
3915 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3917 do {
3918 while (--name->len && name->buf[name->len] != '/')
3919 ; /* scan backwards */
3920 if (!name->len)
3921 break;
3922 name->buf[name->len] = '\0';
3923 if (strset_contains(&state->kept_symlinks, name->buf))
3924 return 1;
3925 if (strset_contains(&state->removed_symlinks, name->buf))
3927 * This cannot be "return 0", because we may
3928 * see a new one created at a higher level.
3930 continue;
3932 /* otherwise, check the preimage */
3933 if (state->check_index) {
3934 struct cache_entry *ce;
3936 ce = index_file_exists(state->repo->index, name->buf,
3937 name->len, ignore_case);
3938 if (ce && S_ISLNK(ce->ce_mode))
3939 return 1;
3940 } else {
3941 struct stat st;
3942 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3943 return 1;
3945 } while (1);
3946 return 0;
3949 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3951 int ret;
3952 struct strbuf name = STRBUF_INIT;
3954 assert(*name_ != '\0');
3955 strbuf_addstr(&name, name_);
3956 ret = path_is_beyond_symlink_1(state, &name);
3957 strbuf_release(&name);
3959 return ret;
3962 static int check_unsafe_path(struct patch *patch)
3964 const char *old_name = NULL;
3965 const char *new_name = NULL;
3966 if (patch->is_delete)
3967 old_name = patch->old_name;
3968 else if (!patch->is_new && !patch->is_copy)
3969 old_name = patch->old_name;
3970 if (!patch->is_delete)
3971 new_name = patch->new_name;
3973 if (old_name && !verify_path(old_name, patch->old_mode))
3974 return error(_("invalid path '%s'"), old_name);
3975 if (new_name && !verify_path(new_name, patch->new_mode))
3976 return error(_("invalid path '%s'"), new_name);
3977 return 0;
3981 * Check and apply the patch in-core; leave the result in patch->result
3982 * for the caller to write it out to the final destination.
3984 static int check_patch(struct apply_state *state, struct patch *patch)
3986 struct stat st;
3987 const char *old_name = patch->old_name;
3988 const char *new_name = patch->new_name;
3989 const char *name = old_name ? old_name : new_name;
3990 struct cache_entry *ce = NULL;
3991 struct patch *tpatch;
3992 int ok_if_exists;
3993 int status;
3995 patch->rejected = 1; /* we will drop this after we succeed */
3997 status = check_preimage(state, patch, &ce, &st);
3998 if (status)
3999 return status;
4000 old_name = patch->old_name;
4003 * A type-change diff is always split into a patch to delete
4004 * old, immediately followed by a patch to create new (see
4005 * diff.c::run_diff()); in such a case it is Ok that the entry
4006 * to be deleted by the previous patch is still in the working
4007 * tree and in the index.
4009 * A patch to swap-rename between A and B would first rename A
4010 * to B and then rename B to A. While applying the first one,
4011 * the presence of B should not stop A from getting renamed to
4012 * B; ask to_be_deleted() about the later rename. Removal of
4013 * B and rename from A to B is handled the same way by asking
4014 * was_deleted().
4016 if ((tpatch = in_fn_table(state, new_name)) &&
4017 (was_deleted(tpatch) || to_be_deleted(tpatch)))
4018 ok_if_exists = 1;
4019 else
4020 ok_if_exists = 0;
4022 if (new_name &&
4023 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
4024 int err = check_to_create(state, new_name, ok_if_exists);
4026 if (err && state->threeway) {
4027 patch->direct_to_threeway = 1;
4028 } else switch (err) {
4029 case 0:
4030 break; /* happy */
4031 case EXISTS_IN_INDEX:
4032 return error(_("%s: already exists in index"), new_name);
4033 case EXISTS_IN_INDEX_AS_ITA:
4034 return error(_("%s: does not match index"), new_name);
4035 case EXISTS_IN_WORKTREE:
4036 return error(_("%s: already exists in working directory"),
4037 new_name);
4038 default:
4039 return err;
4042 if (!patch->new_mode) {
4043 if (0 < patch->is_new)
4044 patch->new_mode = S_IFREG | 0644;
4045 else
4046 patch->new_mode = patch->old_mode;
4050 if (new_name && old_name) {
4051 int same = !strcmp(old_name, new_name);
4052 if (!patch->new_mode)
4053 patch->new_mode = patch->old_mode;
4054 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4055 if (same)
4056 return error(_("new mode (%o) of %s does not "
4057 "match old mode (%o)"),
4058 patch->new_mode, new_name,
4059 patch->old_mode);
4060 else
4061 return error(_("new mode (%o) of %s does not "
4062 "match old mode (%o) of %s"),
4063 patch->new_mode, new_name,
4064 patch->old_mode, old_name);
4068 if (!state->unsafe_paths && check_unsafe_path(patch))
4069 return -128;
4072 * An attempt to read from or delete a path that is beyond a
4073 * symbolic link will be prevented by load_patch_target() that
4074 * is called at the beginning of apply_data() so we do not
4075 * have to worry about a patch marked with "is_delete" bit
4076 * here. We however need to make sure that the patch result
4077 * is not deposited to a path that is beyond a symbolic link
4078 * here.
4080 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4081 return error(_("affected file '%s' is beyond a symbolic link"),
4082 patch->new_name);
4084 if (apply_data(state, patch, &st, ce) < 0)
4085 return error(_("%s: patch does not apply"), name);
4086 patch->rejected = 0;
4087 return 0;
4090 static int check_patch_list(struct apply_state *state, struct patch *patch)
4092 int err = 0;
4094 prepare_symlink_changes(state, patch);
4095 prepare_fn_table(state, patch);
4096 while (patch) {
4097 int res;
4098 if (state->apply_verbosity > verbosity_normal)
4099 say_patch_name(stderr,
4100 _("Checking patch %s..."), patch);
4101 res = check_patch(state, patch);
4102 if (res == -128)
4103 return -128;
4104 err |= res;
4105 patch = patch->next;
4107 return err;
4110 static int read_apply_cache(struct apply_state *state)
4112 if (state->index_file)
4113 return read_index_from(state->repo->index, state->index_file,
4114 get_git_dir());
4115 else
4116 return repo_read_index(state->repo);
4119 /* This function tries to read the object name from the current index */
4120 static int get_current_oid(struct apply_state *state, const char *path,
4121 struct object_id *oid)
4123 int pos;
4125 if (read_apply_cache(state) < 0)
4126 return -1;
4127 pos = index_name_pos(state->repo->index, path, strlen(path));
4128 if (pos < 0)
4129 return -1;
4130 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4131 return 0;
4134 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4137 * A usable gitlink patch has only one fragment (hunk) that looks like:
4138 * @@ -1 +1 @@
4139 * -Subproject commit <old sha1>
4140 * +Subproject commit <new sha1>
4141 * or
4142 * @@ -1 +0,0 @@
4143 * -Subproject commit <old sha1>
4144 * for a removal patch.
4146 struct fragment *hunk = p->fragments;
4147 static const char heading[] = "-Subproject commit ";
4148 char *preimage;
4150 if (/* does the patch have only one hunk? */
4151 hunk && !hunk->next &&
4152 /* is its preimage one line? */
4153 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4154 /* does preimage begin with the heading? */
4155 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4156 starts_with(++preimage, heading) &&
4157 /* does it record full SHA-1? */
4158 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4159 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4160 /* does the abbreviated name on the index line agree with it? */
4161 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4162 return 0; /* it all looks fine */
4164 /* we may have full object name on the index line */
4165 return get_oid_hex(p->old_oid_prefix, oid);
4168 /* Build an index that contains just the files needed for a 3way merge */
4169 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4171 struct patch *patch;
4172 struct index_state result = INDEX_STATE_INIT(state->repo);
4173 struct lock_file lock = LOCK_INIT;
4174 int res;
4176 /* Once we start supporting the reverse patch, it may be
4177 * worth showing the new sha1 prefix, but until then...
4179 for (patch = list; patch; patch = patch->next) {
4180 struct object_id oid;
4181 struct cache_entry *ce;
4182 const char *name;
4184 name = patch->old_name ? patch->old_name : patch->new_name;
4185 if (0 < patch->is_new)
4186 continue;
4188 if (S_ISGITLINK(patch->old_mode)) {
4189 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4190 ; /* ok, the textual part looks sane */
4191 else
4192 return error(_("sha1 information is lacking or "
4193 "useless for submodule %s"), name);
4194 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4195 ; /* ok */
4196 } else if (!patch->lines_added && !patch->lines_deleted) {
4197 /* mode-only change: update the current */
4198 if (get_current_oid(state, patch->old_name, &oid))
4199 return error(_("mode change for %s, which is not "
4200 "in current HEAD"), name);
4201 } else
4202 return error(_("sha1 information is lacking or useless "
4203 "(%s)."), name);
4205 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4206 if (!ce)
4207 return error(_("make_cache_entry failed for path '%s'"),
4208 name);
4209 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4210 discard_cache_entry(ce);
4211 return error(_("could not add %s to temporary index"),
4212 name);
4216 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4217 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4218 discard_index(&result);
4220 if (res)
4221 return error(_("could not write temporary index to %s"),
4222 state->fake_ancestor);
4224 return 0;
4227 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4229 int files, adds, dels;
4231 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4232 files++;
4233 adds += patch->lines_added;
4234 dels += patch->lines_deleted;
4235 show_stats(state, patch);
4238 print_stat_summary(stdout, files, adds, dels);
4241 static void numstat_patch_list(struct apply_state *state,
4242 struct patch *patch)
4244 for ( ; patch; patch = patch->next) {
4245 const char *name;
4246 name = patch->new_name ? patch->new_name : patch->old_name;
4247 if (patch->is_binary)
4248 printf("-\t-\t");
4249 else
4250 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4251 write_name_quoted(name, stdout, state->line_termination);
4255 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4257 if (mode)
4258 printf(" %s mode %06o %s\n", newdelete, mode, name);
4259 else
4260 printf(" %s %s\n", newdelete, name);
4263 static void show_mode_change(struct patch *p, int show_name)
4265 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4266 if (show_name)
4267 printf(" mode change %06o => %06o %s\n",
4268 p->old_mode, p->new_mode, p->new_name);
4269 else
4270 printf(" mode change %06o => %06o\n",
4271 p->old_mode, p->new_mode);
4275 static void show_rename_copy(struct patch *p)
4277 const char *renamecopy = p->is_rename ? "rename" : "copy";
4278 const char *old_name, *new_name;
4280 /* Find common prefix */
4281 old_name = p->old_name;
4282 new_name = p->new_name;
4283 while (1) {
4284 const char *slash_old, *slash_new;
4285 slash_old = strchr(old_name, '/');
4286 slash_new = strchr(new_name, '/');
4287 if (!slash_old ||
4288 !slash_new ||
4289 slash_old - old_name != slash_new - new_name ||
4290 memcmp(old_name, new_name, slash_new - new_name))
4291 break;
4292 old_name = slash_old + 1;
4293 new_name = slash_new + 1;
4295 /* p->old_name through old_name is the common prefix, and old_name and
4296 * new_name through the end of names are renames
4298 if (old_name != p->old_name)
4299 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4300 (int)(old_name - p->old_name), p->old_name,
4301 old_name, new_name, p->score);
4302 else
4303 printf(" %s %s => %s (%d%%)\n", renamecopy,
4304 p->old_name, p->new_name, p->score);
4305 show_mode_change(p, 0);
4308 static void summary_patch_list(struct patch *patch)
4310 struct patch *p;
4312 for (p = patch; p; p = p->next) {
4313 if (p->is_new)
4314 show_file_mode_name("create", p->new_mode, p->new_name);
4315 else if (p->is_delete)
4316 show_file_mode_name("delete", p->old_mode, p->old_name);
4317 else {
4318 if (p->is_rename || p->is_copy)
4319 show_rename_copy(p);
4320 else {
4321 if (p->score) {
4322 printf(" rewrite %s (%d%%)\n",
4323 p->new_name, p->score);
4324 show_mode_change(p, 0);
4326 else
4327 show_mode_change(p, 1);
4333 static void patch_stats(struct apply_state *state, struct patch *patch)
4335 int lines = patch->lines_added + patch->lines_deleted;
4337 if (lines > state->max_change)
4338 state->max_change = lines;
4339 if (patch->old_name) {
4340 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4341 if (!len)
4342 len = strlen(patch->old_name);
4343 if (len > state->max_len)
4344 state->max_len = len;
4346 if (patch->new_name) {
4347 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4348 if (!len)
4349 len = strlen(patch->new_name);
4350 if (len > state->max_len)
4351 state->max_len = len;
4355 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4357 if (state->update_index && !state->ita_only) {
4358 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4359 return error(_("unable to remove %s from index"), patch->old_name);
4361 if (!state->cached) {
4362 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4363 remove_path(patch->old_name);
4366 return 0;
4369 static int add_index_file(struct apply_state *state,
4370 const char *path,
4371 unsigned mode,
4372 void *buf,
4373 unsigned long size)
4375 struct stat st;
4376 struct cache_entry *ce;
4377 int namelen = strlen(path);
4379 ce = make_empty_cache_entry(state->repo->index, namelen);
4380 memcpy(ce->name, path, namelen);
4381 ce->ce_mode = create_ce_mode(mode);
4382 ce->ce_flags = create_ce_flags(0);
4383 ce->ce_namelen = namelen;
4384 if (state->ita_only) {
4385 ce->ce_flags |= CE_INTENT_TO_ADD;
4386 set_object_name_for_intent_to_add_entry(ce);
4387 } else if (S_ISGITLINK(mode)) {
4388 const char *s;
4390 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4391 get_oid_hex(s, &ce->oid)) {
4392 discard_cache_entry(ce);
4393 return error(_("corrupt patch for submodule %s"), path);
4395 } else {
4396 if (!state->cached) {
4397 if (lstat(path, &st) < 0) {
4398 discard_cache_entry(ce);
4399 return error_errno(_("unable to stat newly "
4400 "created file '%s'"),
4401 path);
4403 fill_stat_cache_info(state->repo->index, ce, &st);
4405 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4406 discard_cache_entry(ce);
4407 return error(_("unable to create backing store "
4408 "for newly created file %s"), path);
4411 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4412 discard_cache_entry(ce);
4413 return error(_("unable to add cache entry for %s"), path);
4416 return 0;
4420 * Returns:
4421 * -1 if an unrecoverable error happened
4422 * 0 if everything went well
4423 * 1 if a recoverable error happened
4425 static int try_create_file(struct apply_state *state, const char *path,
4426 unsigned int mode, const char *buf,
4427 unsigned long size)
4429 int fd, res;
4430 struct strbuf nbuf = STRBUF_INIT;
4432 if (S_ISGITLINK(mode)) {
4433 struct stat st;
4434 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4435 return 0;
4436 return !!mkdir(path, 0777);
4439 if (has_symlinks && S_ISLNK(mode))
4440 /* Although buf:size is counted string, it also is NUL
4441 * terminated.
4443 return !!symlink(buf, path);
4445 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4446 if (fd < 0)
4447 return 1;
4449 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4450 size = nbuf.len;
4451 buf = nbuf.buf;
4454 res = write_in_full(fd, buf, size) < 0;
4455 if (res)
4456 error_errno(_("failed to write to '%s'"), path);
4457 strbuf_release(&nbuf);
4459 if (close(fd) < 0 && !res)
4460 return error_errno(_("closing file '%s'"), path);
4462 return res ? -1 : 0;
4466 * We optimistically assume that the directories exist,
4467 * which is true 99% of the time anyway. If they don't,
4468 * we create them and try again.
4470 * Returns:
4471 * -1 on error
4472 * 0 otherwise
4474 static int create_one_file(struct apply_state *state,
4475 char *path,
4476 unsigned mode,
4477 const char *buf,
4478 unsigned long size)
4480 char *newpath = NULL;
4481 int res;
4483 if (state->cached)
4484 return 0;
4487 * We already try to detect whether files are beyond a symlink in our
4488 * up-front checks. But in the case where symlinks are created by any
4489 * of the intermediate hunks it can happen that our up-front checks
4490 * didn't yet see the symlink, but at the point of arriving here there
4491 * in fact is one. We thus repeat the check for symlinks here.
4493 * Note that this does not make the up-front check obsolete as the
4494 * failure mode is different:
4496 * - The up-front checks cause us to abort before we have written
4497 * anything into the working directory. So when we exit this way the
4498 * working directory remains clean.
4500 * - The checks here happen in the middle of the action where we have
4501 * already started to apply the patch. The end result will be a dirty
4502 * working directory.
4504 * Ideally, we should update the up-front checks to catch what would
4505 * happen when we apply the patch before we damage the working tree.
4506 * We have all the information necessary to do so. But for now, as a
4507 * part of embargoed security work, having this check would serve as a
4508 * reasonable first step.
4510 if (path_is_beyond_symlink(state, path))
4511 return error(_("affected file '%s' is beyond a symbolic link"), path);
4513 res = try_create_file(state, path, mode, buf, size);
4514 if (res < 0)
4515 return -1;
4516 if (!res)
4517 return 0;
4519 if (errno == ENOENT) {
4520 if (safe_create_leading_directories_no_share(path))
4521 return 0;
4522 res = try_create_file(state, path, mode, buf, size);
4523 if (res < 0)
4524 return -1;
4525 if (!res)
4526 return 0;
4529 if (errno == EEXIST || errno == EACCES) {
4530 /* We may be trying to create a file where a directory
4531 * used to be.
4533 struct stat st;
4534 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4535 errno = EEXIST;
4538 if (errno == EEXIST) {
4539 unsigned int nr = getpid();
4541 for (;;) {
4542 newpath = mkpathdup("%s~%u", path, nr);
4543 res = try_create_file(state, newpath, mode, buf, size);
4544 if (res < 0)
4545 goto out;
4546 if (!res) {
4547 if (!rename(newpath, path))
4548 goto out;
4549 unlink_or_warn(newpath);
4550 break;
4552 if (errno != EEXIST)
4553 break;
4554 ++nr;
4555 FREE_AND_NULL(newpath);
4558 res = error_errno(_("unable to write file '%s' mode %o"), path, mode);
4559 out:
4560 free(newpath);
4561 return res;
4564 static int add_conflicted_stages_file(struct apply_state *state,
4565 struct patch *patch)
4567 int stage, namelen;
4568 unsigned mode;
4569 struct cache_entry *ce;
4571 if (!state->update_index)
4572 return 0;
4573 namelen = strlen(patch->new_name);
4574 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4576 remove_file_from_index(state->repo->index, patch->new_name);
4577 for (stage = 1; stage < 4; stage++) {
4578 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4579 continue;
4580 ce = make_empty_cache_entry(state->repo->index, namelen);
4581 memcpy(ce->name, patch->new_name, namelen);
4582 ce->ce_mode = create_ce_mode(mode);
4583 ce->ce_flags = create_ce_flags(stage);
4584 ce->ce_namelen = namelen;
4585 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4586 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4587 discard_cache_entry(ce);
4588 return error(_("unable to add cache entry for %s"),
4589 patch->new_name);
4593 return 0;
4596 static int create_file(struct apply_state *state, struct patch *patch)
4598 char *path = patch->new_name;
4599 unsigned mode = patch->new_mode;
4600 unsigned long size = patch->resultsize;
4601 char *buf = patch->result;
4603 if (!mode)
4604 mode = S_IFREG | 0644;
4605 if (create_one_file(state, path, mode, buf, size))
4606 return -1;
4608 if (patch->conflicted_threeway)
4609 return add_conflicted_stages_file(state, patch);
4610 else if (state->update_index)
4611 return add_index_file(state, path, mode, buf, size);
4612 return 0;
4615 /* phase zero is to remove, phase one is to create */
4616 static int write_out_one_result(struct apply_state *state,
4617 struct patch *patch,
4618 int phase)
4620 if (patch->is_delete > 0) {
4621 if (phase == 0)
4622 return remove_file(state, patch, 1);
4623 return 0;
4625 if (patch->is_new > 0 || patch->is_copy) {
4626 if (phase == 1)
4627 return create_file(state, patch);
4628 return 0;
4631 * Rename or modification boils down to the same
4632 * thing: remove the old, write the new
4634 if (phase == 0)
4635 return remove_file(state, patch, patch->is_rename);
4636 if (phase == 1)
4637 return create_file(state, patch);
4638 return 0;
4641 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4643 FILE *rej;
4644 char *namebuf;
4645 struct fragment *frag;
4646 int fd, cnt = 0;
4647 struct strbuf sb = STRBUF_INIT;
4649 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4650 if (!frag->rejected)
4651 continue;
4652 cnt++;
4655 if (!cnt) {
4656 if (state->apply_verbosity > verbosity_normal)
4657 say_patch_name(stderr,
4658 _("Applied patch %s cleanly."), patch);
4659 return 0;
4662 /* This should not happen, because a removal patch that leaves
4663 * contents are marked "rejected" at the patch level.
4665 if (!patch->new_name)
4666 die(_("internal error"));
4668 /* Say this even without --verbose */
4669 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4670 "Applying patch %%s with %d rejects...",
4671 cnt),
4672 cnt);
4673 if (state->apply_verbosity > verbosity_silent)
4674 say_patch_name(stderr, sb.buf, patch);
4675 strbuf_release(&sb);
4677 namebuf = xstrfmt("%s.rej", patch->new_name);
4679 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4680 if (fd < 0) {
4681 if (errno != EEXIST) {
4682 error_errno(_("cannot open %s"), namebuf);
4683 goto error;
4685 if (unlink(namebuf)) {
4686 error_errno(_("cannot unlink '%s'"), namebuf);
4687 goto error;
4689 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4690 if (fd < 0) {
4691 error_errno(_("cannot open %s"), namebuf);
4692 goto error;
4695 rej = fdopen(fd, "w");
4696 if (!rej) {
4697 error_errno(_("cannot open %s"), namebuf);
4698 close(fd);
4699 goto error;
4702 /* Normal git tools never deal with .rej, so do not pretend
4703 * this is a git patch by saying --git or giving extended
4704 * headers. While at it, maybe please "kompare" that wants
4705 * the trailing TAB and some garbage at the end of line ;-).
4707 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4708 patch->new_name, patch->new_name);
4709 for (cnt = 1, frag = patch->fragments;
4710 frag;
4711 cnt++, frag = frag->next) {
4712 if (!frag->rejected) {
4713 if (state->apply_verbosity > verbosity_silent)
4714 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4715 continue;
4717 if (state->apply_verbosity > verbosity_silent)
4718 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4719 fprintf(rej, "%.*s", frag->size, frag->patch);
4720 if (frag->patch[frag->size-1] != '\n')
4721 fputc('\n', rej);
4723 fclose(rej);
4724 error:
4725 free(namebuf);
4726 return -1;
4730 * Returns:
4731 * -1 if an error happened
4732 * 0 if the patch applied cleanly
4733 * 1 if the patch did not apply cleanly
4735 static int write_out_results(struct apply_state *state, struct patch *list)
4737 int phase;
4738 int errs = 0;
4739 struct patch *l;
4740 struct string_list cpath = STRING_LIST_INIT_DUP;
4742 for (phase = 0; phase < 2; phase++) {
4743 l = list;
4744 while (l) {
4745 if (l->rejected)
4746 errs = 1;
4747 else {
4748 if (write_out_one_result(state, l, phase)) {
4749 string_list_clear(&cpath, 0);
4750 return -1;
4752 if (phase == 1) {
4753 if (write_out_one_reject(state, l))
4754 errs = 1;
4755 if (l->conflicted_threeway) {
4756 string_list_append(&cpath, l->new_name);
4757 errs = 1;
4761 l = l->next;
4765 if (cpath.nr) {
4766 struct string_list_item *item;
4768 string_list_sort(&cpath);
4769 if (state->apply_verbosity > verbosity_silent) {
4770 for_each_string_list_item(item, &cpath)
4771 fprintf(stderr, "U %s\n", item->string);
4773 string_list_clear(&cpath, 0);
4776 * rerere relies on the partially merged result being in the working
4777 * tree with conflict markers, but that isn't written with --cached.
4779 if (!state->cached)
4780 repo_rerere(state->repo, 0);
4783 return errs;
4787 * Try to apply a patch.
4789 * Returns:
4790 * -128 if a bad error happened (like patch unreadable)
4791 * -1 if patch did not apply and user cannot deal with it
4792 * 0 if the patch applied
4793 * 1 if the patch did not apply but user might fix it
4795 static int apply_patch(struct apply_state *state,
4796 int fd,
4797 const char *filename,
4798 int options)
4800 size_t offset;
4801 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4802 struct patch *list = NULL, **listp = &list;
4803 int skipped_patch = 0;
4804 int res = 0;
4805 int flush_attributes = 0;
4807 state->patch_input_file = filename;
4808 if (read_patch_file(&buf, fd) < 0)
4809 return -128;
4810 offset = 0;
4811 while (offset < buf.len) {
4812 struct patch *patch;
4813 int nr;
4815 CALLOC_ARRAY(patch, 1);
4816 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4817 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4818 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4819 if (nr < 0) {
4820 free_patch(patch);
4821 if (nr == -128) {
4822 res = -128;
4823 goto end;
4825 break;
4827 if (state->apply_in_reverse)
4828 reverse_patches(patch);
4829 if (use_patch(state, patch)) {
4830 patch_stats(state, patch);
4831 if (!list || !state->apply_in_reverse) {
4832 *listp = patch;
4833 listp = &patch->next;
4834 } else {
4835 patch->next = list;
4836 list = patch;
4839 if ((patch->new_name &&
4840 ends_with_path_components(patch->new_name,
4841 GITATTRIBUTES_FILE)) ||
4842 (patch->old_name &&
4843 ends_with_path_components(patch->old_name,
4844 GITATTRIBUTES_FILE)))
4845 flush_attributes = 1;
4847 else {
4848 if (state->apply_verbosity > verbosity_normal)
4849 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4850 free_patch(patch);
4851 skipped_patch++;
4853 offset += nr;
4856 if (!list && !skipped_patch) {
4857 if (!state->allow_empty) {
4858 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4859 res = -128;
4861 goto end;
4864 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4865 state->apply = 0;
4867 state->update_index = (state->check_index || state->ita_only) && state->apply;
4868 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4869 if (state->index_file)
4870 hold_lock_file_for_update(&state->lock_file,
4871 state->index_file,
4872 LOCK_DIE_ON_ERROR);
4873 else
4874 repo_hold_locked_index(state->repo, &state->lock_file,
4875 LOCK_DIE_ON_ERROR);
4878 if (state->check_index && read_apply_cache(state) < 0) {
4879 error(_("unable to read index file"));
4880 res = -128;
4881 goto end;
4884 if (state->check || state->apply) {
4885 int r = check_patch_list(state, list);
4886 if (r == -128) {
4887 res = -128;
4888 goto end;
4890 if (r < 0 && !state->apply_with_reject) {
4891 res = -1;
4892 goto end;
4896 if (state->apply) {
4897 int write_res = write_out_results(state, list);
4898 if (write_res < 0) {
4899 res = -128;
4900 goto end;
4902 if (write_res > 0) {
4903 /* with --3way, we still need to write the index out */
4904 res = state->apply_with_reject ? -1 : 1;
4905 goto end;
4909 if (state->fake_ancestor &&
4910 build_fake_ancestor(state, list)) {
4911 res = -128;
4912 goto end;
4915 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4916 stat_patch_list(state, list);
4918 if (state->numstat && state->apply_verbosity > verbosity_silent)
4919 numstat_patch_list(state, list);
4921 if (state->summary && state->apply_verbosity > verbosity_silent)
4922 summary_patch_list(list);
4924 if (flush_attributes)
4925 reset_parsed_attributes();
4926 end:
4927 free_patch_list(list);
4928 strbuf_release(&buf);
4929 string_list_clear(&state->fn_table, 0);
4930 return res;
4933 static int apply_option_parse_exclude(const struct option *opt,
4934 const char *arg, int unset)
4936 struct apply_state *state = opt->value;
4938 BUG_ON_OPT_NEG(unset);
4940 add_name_limit(state, arg, 1);
4941 return 0;
4944 static int apply_option_parse_include(const struct option *opt,
4945 const char *arg, int unset)
4947 struct apply_state *state = opt->value;
4949 BUG_ON_OPT_NEG(unset);
4951 add_name_limit(state, arg, 0);
4952 state->has_include = 1;
4953 return 0;
4956 static int apply_option_parse_p(const struct option *opt,
4957 const char *arg,
4958 int unset)
4960 struct apply_state *state = opt->value;
4962 BUG_ON_OPT_NEG(unset);
4964 state->p_value = atoi(arg);
4965 state->p_value_known = 1;
4966 return 0;
4969 static int apply_option_parse_space_change(const struct option *opt,
4970 const char *arg, int unset)
4972 struct apply_state *state = opt->value;
4974 BUG_ON_OPT_ARG(arg);
4976 if (unset)
4977 state->ws_ignore_action = ignore_ws_none;
4978 else
4979 state->ws_ignore_action = ignore_ws_change;
4980 return 0;
4983 static int apply_option_parse_whitespace(const struct option *opt,
4984 const char *arg, int unset)
4986 struct apply_state *state = opt->value;
4988 BUG_ON_OPT_NEG(unset);
4990 state->whitespace_option = arg;
4991 if (parse_whitespace_option(state, arg))
4992 return -1;
4993 return 0;
4996 static int apply_option_parse_directory(const struct option *opt,
4997 const char *arg, int unset)
4999 struct apply_state *state = opt->value;
5001 BUG_ON_OPT_NEG(unset);
5003 strbuf_reset(&state->root);
5004 strbuf_addstr(&state->root, arg);
5005 strbuf_complete(&state->root, '/');
5006 return 0;
5009 int apply_all_patches(struct apply_state *state,
5010 int argc,
5011 const char **argv,
5012 int options)
5014 int i;
5015 int res;
5016 int errs = 0;
5017 int read_stdin = 1;
5019 for (i = 0; i < argc; i++) {
5020 const char *arg = argv[i];
5021 char *to_free = NULL;
5022 int fd;
5024 if (!strcmp(arg, "-")) {
5025 res = apply_patch(state, 0, "<stdin>", options);
5026 if (res < 0)
5027 goto end;
5028 errs |= res;
5029 read_stdin = 0;
5030 continue;
5031 } else
5032 arg = to_free = prefix_filename(state->prefix, arg);
5034 fd = open(arg, O_RDONLY);
5035 if (fd < 0) {
5036 error(_("can't open patch '%s': %s"), arg, strerror(errno));
5037 res = -128;
5038 free(to_free);
5039 goto end;
5041 read_stdin = 0;
5042 set_default_whitespace_mode(state);
5043 res = apply_patch(state, fd, arg, options);
5044 close(fd);
5045 free(to_free);
5046 if (res < 0)
5047 goto end;
5048 errs |= res;
5050 set_default_whitespace_mode(state);
5051 if (read_stdin) {
5052 res = apply_patch(state, 0, "<stdin>", options);
5053 if (res < 0)
5054 goto end;
5055 errs |= res;
5058 if (state->whitespace_error) {
5059 if (state->squelch_whitespace_errors &&
5060 state->squelch_whitespace_errors < state->whitespace_error) {
5061 int squelched =
5062 state->whitespace_error - state->squelch_whitespace_errors;
5063 warning(Q_("squelched %d whitespace error",
5064 "squelched %d whitespace errors",
5065 squelched),
5066 squelched);
5068 if (state->ws_error_action == die_on_ws_error) {
5069 error(Q_("%d line adds whitespace errors.",
5070 "%d lines add whitespace errors.",
5071 state->whitespace_error),
5072 state->whitespace_error);
5073 res = -128;
5074 goto end;
5076 if (state->applied_after_fixing_ws && state->apply)
5077 warning(Q_("%d line applied after"
5078 " fixing whitespace errors.",
5079 "%d lines applied after"
5080 " fixing whitespace errors.",
5081 state->applied_after_fixing_ws),
5082 state->applied_after_fixing_ws);
5083 else if (state->whitespace_error)
5084 warning(Q_("%d line adds whitespace errors.",
5085 "%d lines add whitespace errors.",
5086 state->whitespace_error),
5087 state->whitespace_error);
5090 if (state->update_index) {
5091 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5092 if (res) {
5093 error(_("Unable to write new index file"));
5094 res = -128;
5095 goto end;
5099 res = !!errs;
5101 end:
5102 rollback_lock_file(&state->lock_file);
5104 if (state->apply_verbosity <= verbosity_silent) {
5105 set_error_routine(state->saved_error_routine);
5106 set_warn_routine(state->saved_warn_routine);
5109 if (res > -1)
5110 return res;
5111 return (res == -1 ? 1 : 128);
5114 int apply_parse_options(int argc, const char **argv,
5115 struct apply_state *state,
5116 int *force_apply, int *options,
5117 const char * const *apply_usage)
5119 struct option builtin_apply_options[] = {
5120 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5121 N_("don't apply changes matching the given path"),
5122 PARSE_OPT_NONEG, apply_option_parse_exclude),
5123 OPT_CALLBACK_F(0, "include", state, N_("path"),
5124 N_("apply changes matching the given path"),
5125 PARSE_OPT_NONEG, apply_option_parse_include),
5126 OPT_CALLBACK('p', NULL, state, N_("num"),
5127 N_("remove <num> leading slashes from traditional diff paths"),
5128 apply_option_parse_p),
5129 OPT_BOOL(0, "no-add", &state->no_add,
5130 N_("ignore additions made by the patch")),
5131 OPT_BOOL(0, "stat", &state->diffstat,
5132 N_("instead of applying the patch, output diffstat for the input")),
5133 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5134 OPT_NOOP_NOARG(0, "binary"),
5135 OPT_BOOL(0, "numstat", &state->numstat,
5136 N_("show number of added and deleted lines in decimal notation")),
5137 OPT_BOOL(0, "summary", &state->summary,
5138 N_("instead of applying the patch, output a summary for the input")),
5139 OPT_BOOL(0, "check", &state->check,
5140 N_("instead of applying the patch, see if the patch is applicable")),
5141 OPT_BOOL(0, "index", &state->check_index,
5142 N_("make sure the patch is applicable to the current index")),
5143 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5144 N_("mark new files with `git add --intent-to-add`")),
5145 OPT_BOOL(0, "cached", &state->cached,
5146 N_("apply a patch without touching the working tree")),
5147 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5148 N_("accept a patch that touches outside the working area"),
5149 PARSE_OPT_NOCOMPLETE),
5150 OPT_BOOL(0, "apply", force_apply,
5151 N_("also apply the patch (use with --stat/--summary/--check)")),
5152 OPT_BOOL('3', "3way", &state->threeway,
5153 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5154 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5155 N_("build a temporary index based on embedded index information")),
5156 /* Think twice before adding "--nul" synonym to this */
5157 OPT_SET_INT('z', NULL, &state->line_termination,
5158 N_("paths are separated with NUL character"), '\0'),
5159 OPT_INTEGER('C', NULL, &state->p_context,
5160 N_("ensure at least <n> lines of context match")),
5161 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5162 N_("detect new or modified lines that have whitespace errors"),
5163 apply_option_parse_whitespace),
5164 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5165 N_("ignore changes in whitespace when finding context"),
5166 PARSE_OPT_NOARG, apply_option_parse_space_change),
5167 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5168 N_("ignore changes in whitespace when finding context"),
5169 PARSE_OPT_NOARG, apply_option_parse_space_change),
5170 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5171 N_("apply the patch in reverse")),
5172 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5173 N_("don't expect at least one line of context")),
5174 OPT_BOOL(0, "reject", &state->apply_with_reject,
5175 N_("leave the rejected hunks in corresponding *.rej files")),
5176 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5177 N_("allow overlapping hunks")),
5178 OPT__VERBOSITY(&state->apply_verbosity),
5179 OPT_BIT(0, "inaccurate-eof", options,
5180 N_("tolerate incorrectly detected missing new-line at the end of file"),
5181 APPLY_OPT_INACCURATE_EOF),
5182 OPT_BIT(0, "recount", options,
5183 N_("do not trust the line counts in the hunk headers"),
5184 APPLY_OPT_RECOUNT),
5185 OPT_CALLBACK(0, "directory", state, N_("root"),
5186 N_("prepend <root> to all filenames"),
5187 apply_option_parse_directory),
5188 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5189 N_("don't return error for empty patches")),
5190 OPT_END()
5193 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);