reftable/reader: inline `reader_close()`
[git/gitster.git] / apply.c
blob0f2f5dabe33aca01e7a52489a02971ce4027b782
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 return 0;
1001 static int gitdiff_oldmode(struct gitdiff_data *state,
1002 const char *line,
1003 struct patch *patch)
1005 return parse_mode_line(line, state->linenr, &patch->old_mode);
1008 static int gitdiff_newmode(struct gitdiff_data *state,
1009 const char *line,
1010 struct patch *patch)
1012 return parse_mode_line(line, state->linenr, &patch->new_mode);
1015 static int gitdiff_delete(struct gitdiff_data *state,
1016 const char *line,
1017 struct patch *patch)
1019 patch->is_delete = 1;
1020 free(patch->old_name);
1021 patch->old_name = xstrdup_or_null(patch->def_name);
1022 return gitdiff_oldmode(state, line, patch);
1025 static int gitdiff_newfile(struct gitdiff_data *state,
1026 const char *line,
1027 struct patch *patch)
1029 patch->is_new = 1;
1030 free(patch->new_name);
1031 patch->new_name = xstrdup_or_null(patch->def_name);
1032 return gitdiff_newmode(state, line, patch);
1035 static int gitdiff_copysrc(struct gitdiff_data *state,
1036 const char *line,
1037 struct patch *patch)
1039 patch->is_copy = 1;
1040 free(patch->old_name);
1041 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1042 return 0;
1045 static int gitdiff_copydst(struct gitdiff_data *state,
1046 const char *line,
1047 struct patch *patch)
1049 patch->is_copy = 1;
1050 free(patch->new_name);
1051 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1052 return 0;
1055 static int gitdiff_renamesrc(struct gitdiff_data *state,
1056 const char *line,
1057 struct patch *patch)
1059 patch->is_rename = 1;
1060 free(patch->old_name);
1061 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1062 return 0;
1065 static int gitdiff_renamedst(struct gitdiff_data *state,
1066 const char *line,
1067 struct patch *patch)
1069 patch->is_rename = 1;
1070 free(patch->new_name);
1071 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1072 return 0;
1075 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1076 const char *line,
1077 struct patch *patch)
1079 unsigned long val = strtoul(line, NULL, 10);
1080 if (val <= 100)
1081 patch->score = val;
1082 return 0;
1085 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1086 const char *line,
1087 struct patch *patch)
1089 unsigned long val = strtoul(line, NULL, 10);
1090 if (val <= 100)
1091 patch->score = val;
1092 return 0;
1095 static int gitdiff_index(struct gitdiff_data *state,
1096 const char *line,
1097 struct patch *patch)
1100 * index line is N hexadecimal, "..", N hexadecimal,
1101 * and optional space with octal mode.
1103 const char *ptr, *eol;
1104 int len;
1105 const unsigned hexsz = the_hash_algo->hexsz;
1107 ptr = strchr(line, '.');
1108 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1109 return 0;
1110 len = ptr - line;
1111 memcpy(patch->old_oid_prefix, line, len);
1112 patch->old_oid_prefix[len] = 0;
1114 line = ptr + 2;
1115 ptr = strchr(line, ' ');
1116 eol = strchrnul(line, '\n');
1118 if (!ptr || eol < ptr)
1119 ptr = eol;
1120 len = ptr - line;
1122 if (hexsz < len)
1123 return 0;
1124 memcpy(patch->new_oid_prefix, line, len);
1125 patch->new_oid_prefix[len] = 0;
1126 if (*ptr == ' ')
1127 return gitdiff_oldmode(state, ptr + 1, patch);
1128 return 0;
1132 * This is normal for a diff that doesn't change anything: we'll fall through
1133 * into the next diff. Tell the parser to break out.
1135 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1136 const char *line UNUSED,
1137 struct patch *patch UNUSED)
1139 return 1;
1143 * Skip p_value leading components from "line"; as we do not accept
1144 * absolute paths, return NULL in that case.
1146 static const char *skip_tree_prefix(int p_value,
1147 const char *line,
1148 int llen)
1150 int nslash;
1151 int i;
1153 if (!p_value)
1154 return (llen && line[0] == '/') ? NULL : line;
1156 nslash = p_value;
1157 for (i = 0; i < llen; i++) {
1158 int ch = line[i];
1159 if (ch == '/' && --nslash <= 0)
1160 return (i == 0) ? NULL : &line[i + 1];
1162 return NULL;
1166 * This is to extract the same name that appears on "diff --git"
1167 * line. We do not find and return anything if it is a rename
1168 * patch, and it is OK because we will find the name elsewhere.
1169 * We need to reliably find name only when it is mode-change only,
1170 * creation or deletion of an empty file. In any of these cases,
1171 * both sides are the same name under a/ and b/ respectively.
1173 static char *git_header_name(int p_value,
1174 const char *line,
1175 int llen)
1177 const char *name;
1178 const char *second = NULL;
1179 size_t len, line_len;
1181 line += strlen("diff --git ");
1182 llen -= strlen("diff --git ");
1184 if (*line == '"') {
1185 const char *cp;
1186 struct strbuf first = STRBUF_INIT;
1187 struct strbuf sp = STRBUF_INIT;
1189 if (unquote_c_style(&first, line, &second))
1190 goto free_and_fail1;
1192 /* strip the a/b prefix including trailing slash */
1193 cp = skip_tree_prefix(p_value, first.buf, first.len);
1194 if (!cp)
1195 goto free_and_fail1;
1196 strbuf_remove(&first, 0, cp - first.buf);
1199 * second points at one past closing dq of name.
1200 * find the second name.
1202 while ((second < line + llen) && isspace(*second))
1203 second++;
1205 if (line + llen <= second)
1206 goto free_and_fail1;
1207 if (*second == '"') {
1208 if (unquote_c_style(&sp, second, NULL))
1209 goto free_and_fail1;
1210 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1211 if (!cp)
1212 goto free_and_fail1;
1213 /* They must match, otherwise ignore */
1214 if (strcmp(cp, first.buf))
1215 goto free_and_fail1;
1216 strbuf_release(&sp);
1217 return strbuf_detach(&first, NULL);
1220 /* unquoted second */
1221 cp = skip_tree_prefix(p_value, second, line + llen - second);
1222 if (!cp)
1223 goto free_and_fail1;
1224 if (line + llen - cp != first.len ||
1225 memcmp(first.buf, cp, first.len))
1226 goto free_and_fail1;
1227 return strbuf_detach(&first, NULL);
1229 free_and_fail1:
1230 strbuf_release(&first);
1231 strbuf_release(&sp);
1232 return NULL;
1235 /* unquoted first name */
1236 name = skip_tree_prefix(p_value, line, llen);
1237 if (!name)
1238 return NULL;
1241 * since the first name is unquoted, a dq if exists must be
1242 * the beginning of the second name.
1244 for (second = name; second < line + llen; second++) {
1245 if (*second == '"') {
1246 struct strbuf sp = STRBUF_INIT;
1247 const char *np;
1249 if (unquote_c_style(&sp, second, NULL))
1250 goto free_and_fail2;
1252 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1253 if (!np)
1254 goto free_and_fail2;
1256 len = sp.buf + sp.len - np;
1257 if (len < second - name &&
1258 !strncmp(np, name, len) &&
1259 isspace(name[len])) {
1260 /* Good */
1261 strbuf_remove(&sp, 0, np - sp.buf);
1262 return strbuf_detach(&sp, NULL);
1265 free_and_fail2:
1266 strbuf_release(&sp);
1267 return NULL;
1272 * Accept a name only if it shows up twice, exactly the same
1273 * form.
1275 second = strchr(name, '\n');
1276 if (!second)
1277 return NULL;
1278 line_len = second - name;
1279 for (len = 0 ; ; len++) {
1280 switch (name[len]) {
1281 default:
1282 continue;
1283 case '\n':
1284 return NULL;
1285 case '\t': case ' ':
1287 * Is this the separator between the preimage
1288 * and the postimage pathname? Again, we are
1289 * only interested in the case where there is
1290 * no rename, as this is only to set def_name
1291 * and a rename patch has the names elsewhere
1292 * in an unambiguous form.
1294 if (!name[len + 1])
1295 return NULL; /* no postimage name */
1296 second = skip_tree_prefix(p_value, name + len + 1,
1297 line_len - (len + 1));
1299 * If we are at the SP at the end of a directory,
1300 * skip_tree_prefix() may return NULL as that makes
1301 * it appears as if we have an absolute path.
1302 * Keep going to find another SP.
1304 if (!second)
1305 continue;
1308 * Does len bytes starting at "name" and "second"
1309 * (that are separated by one HT or SP we just
1310 * found) exactly match?
1312 if (second[len] == '\n' && !strncmp(name, second, len))
1313 return xmemdupz(name, len);
1318 static int check_header_line(int linenr, struct patch *patch)
1320 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1321 (patch->is_rename == 1) + (patch->is_copy == 1);
1322 if (extensions > 1)
1323 return error(_("inconsistent header lines %d and %d"),
1324 patch->extension_linenr, linenr);
1325 if (extensions && !patch->extension_linenr)
1326 patch->extension_linenr = linenr;
1327 return 0;
1330 int parse_git_diff_header(struct strbuf *root,
1331 int *linenr,
1332 int p_value,
1333 const char *line,
1334 int len,
1335 unsigned int size,
1336 struct patch *patch)
1338 unsigned long offset;
1339 struct gitdiff_data parse_hdr_state;
1341 /* A git diff has explicit new/delete information, so we don't guess */
1342 patch->is_new = 0;
1343 patch->is_delete = 0;
1346 * Some things may not have the old name in the
1347 * rest of the headers anywhere (pure mode changes,
1348 * or removing or adding empty files), so we get
1349 * the default name from the header.
1351 patch->def_name = git_header_name(p_value, line, len);
1352 if (patch->def_name && root->len) {
1353 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1354 free(patch->def_name);
1355 patch->def_name = s;
1358 line += len;
1359 size -= len;
1360 (*linenr)++;
1361 parse_hdr_state.root = root;
1362 parse_hdr_state.linenr = *linenr;
1363 parse_hdr_state.p_value = p_value;
1365 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1366 static const struct opentry {
1367 const char *str;
1368 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1369 } optable[] = {
1370 { "@@ -", gitdiff_hdrend },
1371 { "--- ", gitdiff_oldname },
1372 { "+++ ", gitdiff_newname },
1373 { "old mode ", gitdiff_oldmode },
1374 { "new mode ", gitdiff_newmode },
1375 { "deleted file mode ", gitdiff_delete },
1376 { "new file mode ", gitdiff_newfile },
1377 { "copy from ", gitdiff_copysrc },
1378 { "copy to ", gitdiff_copydst },
1379 { "rename old ", gitdiff_renamesrc },
1380 { "rename new ", gitdiff_renamedst },
1381 { "rename from ", gitdiff_renamesrc },
1382 { "rename to ", gitdiff_renamedst },
1383 { "similarity index ", gitdiff_similarity },
1384 { "dissimilarity index ", gitdiff_dissimilarity },
1385 { "index ", gitdiff_index },
1386 { "", gitdiff_unrecognized },
1388 int i;
1390 len = linelen(line, size);
1391 if (!len || line[len-1] != '\n')
1392 break;
1393 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1394 const struct opentry *p = optable + i;
1395 int oplen = strlen(p->str);
1396 int res;
1397 if (len < oplen || memcmp(p->str, line, oplen))
1398 continue;
1399 res = p->fn(&parse_hdr_state, line + oplen, patch);
1400 if (res < 0)
1401 return -1;
1402 if (check_header_line(*linenr, patch))
1403 return -1;
1404 if (res > 0)
1405 goto done;
1406 break;
1410 done:
1411 if (!patch->old_name && !patch->new_name) {
1412 if (!patch->def_name) {
1413 error(Q_("git diff header lacks filename information when removing "
1414 "%d leading pathname component (line %d)",
1415 "git diff header lacks filename information when removing "
1416 "%d leading pathname components (line %d)",
1417 parse_hdr_state.p_value),
1418 parse_hdr_state.p_value, *linenr);
1419 return -128;
1421 patch->old_name = xstrdup(patch->def_name);
1422 patch->new_name = xstrdup(patch->def_name);
1424 if ((!patch->new_name && !patch->is_delete) ||
1425 (!patch->old_name && !patch->is_new)) {
1426 error(_("git diff header lacks filename information "
1427 "(line %d)"), *linenr);
1428 return -128;
1430 patch->is_toplevel_relative = 1;
1431 return offset;
1434 static int parse_num(const char *line, unsigned long *p)
1436 char *ptr;
1438 if (!isdigit(*line))
1439 return 0;
1440 *p = strtoul(line, &ptr, 10);
1441 return ptr - line;
1444 static int parse_range(const char *line, int len, int offset, const char *expect,
1445 unsigned long *p1, unsigned long *p2)
1447 int digits, ex;
1449 if (offset < 0 || offset >= len)
1450 return -1;
1451 line += offset;
1452 len -= offset;
1454 digits = parse_num(line, p1);
1455 if (!digits)
1456 return -1;
1458 offset += digits;
1459 line += digits;
1460 len -= digits;
1462 *p2 = 1;
1463 if (*line == ',') {
1464 digits = parse_num(line+1, p2);
1465 if (!digits)
1466 return -1;
1468 offset += digits+1;
1469 line += digits+1;
1470 len -= digits+1;
1473 ex = strlen(expect);
1474 if (ex > len)
1475 return -1;
1476 if (memcmp(line, expect, ex))
1477 return -1;
1479 return offset + ex;
1482 static void recount_diff(const char *line, int size, struct fragment *fragment)
1484 int oldlines = 0, newlines = 0, ret = 0;
1486 if (size < 1) {
1487 warning("recount: ignore empty hunk");
1488 return;
1491 for (;;) {
1492 int len = linelen(line, size);
1493 size -= len;
1494 line += len;
1496 if (size < 1)
1497 break;
1499 switch (*line) {
1500 case ' ': case '\n':
1501 newlines++;
1502 /* fall through */
1503 case '-':
1504 oldlines++;
1505 continue;
1506 case '+':
1507 newlines++;
1508 continue;
1509 case '\\':
1510 continue;
1511 case '@':
1512 ret = size < 3 || !starts_with(line, "@@ ");
1513 break;
1514 case 'd':
1515 ret = size < 5 || !starts_with(line, "diff ");
1516 break;
1517 default:
1518 ret = -1;
1519 break;
1521 if (ret) {
1522 warning(_("recount: unexpected line: %.*s"),
1523 (int)linelen(line, size), line);
1524 return;
1526 break;
1528 fragment->oldlines = oldlines;
1529 fragment->newlines = newlines;
1533 * Parse a unified diff fragment header of the
1534 * form "@@ -a,b +c,d @@"
1536 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1538 int offset;
1540 if (!len || line[len-1] != '\n')
1541 return -1;
1543 /* Figure out the number of lines in a fragment */
1544 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1545 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1547 return offset;
1551 * Find file diff header
1553 * Returns:
1554 * -1 if no header was found
1555 * -128 in case of error
1556 * the size of the header in bytes (called "offset") otherwise
1558 static int find_header(struct apply_state *state,
1559 const char *line,
1560 unsigned long size,
1561 int *hdrsize,
1562 struct patch *patch)
1564 unsigned long offset, len;
1566 patch->is_toplevel_relative = 0;
1567 patch->is_rename = patch->is_copy = 0;
1568 patch->is_new = patch->is_delete = -1;
1569 patch->old_mode = patch->new_mode = 0;
1570 patch->old_name = patch->new_name = NULL;
1571 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1572 unsigned long nextlen;
1574 len = linelen(line, size);
1575 if (!len)
1576 break;
1578 /* Testing this early allows us to take a few shortcuts.. */
1579 if (len < 6)
1580 continue;
1583 * Make sure we don't find any unconnected patch fragments.
1584 * That's a sign that we didn't find a header, and that a
1585 * patch has become corrupted/broken up.
1587 if (!memcmp("@@ -", line, 4)) {
1588 struct fragment dummy;
1589 if (parse_fragment_header(line, len, &dummy) < 0)
1590 continue;
1591 error(_("patch fragment without header at line %d: %.*s"),
1592 state->linenr, (int)len-1, line);
1593 return -128;
1596 if (size < len + 6)
1597 break;
1600 * Git patch? It might not have a real patch, just a rename
1601 * or mode change, so we handle that specially
1603 if (!memcmp("diff --git ", line, 11)) {
1604 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1605 state->p_value, line, len,
1606 size, patch);
1607 if (git_hdr_len < 0)
1608 return -128;
1609 if (git_hdr_len <= len)
1610 continue;
1611 *hdrsize = git_hdr_len;
1612 return offset;
1615 /* --- followed by +++ ? */
1616 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1617 continue;
1620 * We only accept unified patches, so we want it to
1621 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1622 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1624 nextlen = linelen(line + len, size - len);
1625 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1626 continue;
1628 /* Ok, we'll consider it a patch */
1629 if (parse_traditional_patch(state, line, line+len, patch))
1630 return -128;
1631 *hdrsize = len + nextlen;
1632 state->linenr += 2;
1633 return offset;
1635 return -1;
1638 static void record_ws_error(struct apply_state *state,
1639 unsigned result,
1640 const char *line,
1641 int len,
1642 int linenr)
1644 char *err;
1646 if (!result)
1647 return;
1649 state->whitespace_error++;
1650 if (state->squelch_whitespace_errors &&
1651 state->squelch_whitespace_errors < state->whitespace_error)
1652 return;
1654 err = whitespace_error_string(result);
1655 if (state->apply_verbosity > verbosity_silent)
1656 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1657 state->patch_input_file, linenr, err, len, line);
1658 free(err);
1661 static void check_whitespace(struct apply_state *state,
1662 const char *line,
1663 int len,
1664 unsigned ws_rule)
1666 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1668 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1672 * Check if the patch has context lines with CRLF or
1673 * the patch wants to remove lines with CRLF.
1675 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1677 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1678 patch->ws_rule |= WS_CR_AT_EOL;
1679 patch->crlf_in_old = 1;
1685 * Parse a unified diff. Note that this really needs to parse each
1686 * fragment separately, since the only way to know the difference
1687 * between a "---" that is part of a patch, and a "---" that starts
1688 * the next patch is to look at the line counts..
1690 static int parse_fragment(struct apply_state *state,
1691 const char *line,
1692 unsigned long size,
1693 struct patch *patch,
1694 struct fragment *fragment)
1696 int added, deleted;
1697 int len = linelen(line, size), offset;
1698 unsigned long oldlines, newlines;
1699 unsigned long leading, trailing;
1701 offset = parse_fragment_header(line, len, fragment);
1702 if (offset < 0)
1703 return -1;
1704 if (offset > 0 && patch->recount)
1705 recount_diff(line + offset, size - offset, fragment);
1706 oldlines = fragment->oldlines;
1707 newlines = fragment->newlines;
1708 leading = 0;
1709 trailing = 0;
1711 /* Parse the thing.. */
1712 line += len;
1713 size -= len;
1714 state->linenr++;
1715 added = deleted = 0;
1716 for (offset = len;
1717 0 < size;
1718 offset += len, size -= len, line += len, state->linenr++) {
1719 if (!oldlines && !newlines)
1720 break;
1721 len = linelen(line, size);
1722 if (!len || line[len-1] != '\n')
1723 return -1;
1724 switch (*line) {
1725 default:
1726 return -1;
1727 case '\n': /* newer GNU diff, an empty context line */
1728 case ' ':
1729 oldlines--;
1730 newlines--;
1731 if (!deleted && !added)
1732 leading++;
1733 trailing++;
1734 check_old_for_crlf(patch, line, len);
1735 if (!state->apply_in_reverse &&
1736 state->ws_error_action == correct_ws_error)
1737 check_whitespace(state, line, len, patch->ws_rule);
1738 break;
1739 case '-':
1740 if (!state->apply_in_reverse)
1741 check_old_for_crlf(patch, line, len);
1742 if (state->apply_in_reverse &&
1743 state->ws_error_action != nowarn_ws_error)
1744 check_whitespace(state, line, len, patch->ws_rule);
1745 deleted++;
1746 oldlines--;
1747 trailing = 0;
1748 break;
1749 case '+':
1750 if (state->apply_in_reverse)
1751 check_old_for_crlf(patch, line, len);
1752 if (!state->apply_in_reverse &&
1753 state->ws_error_action != nowarn_ws_error)
1754 check_whitespace(state, line, len, patch->ws_rule);
1755 added++;
1756 newlines--;
1757 trailing = 0;
1758 break;
1761 * We allow "\ No newline at end of file". Depending
1762 * on locale settings when the patch was produced we
1763 * don't know what this line looks like. The only
1764 * thing we do know is that it begins with "\ ".
1765 * Checking for 12 is just for sanity check -- any
1766 * l10n of "\ No newline..." is at least that long.
1768 case '\\':
1769 if (len < 12 || memcmp(line, "\\ ", 2))
1770 return -1;
1771 break;
1774 if (oldlines || newlines)
1775 return -1;
1776 if (!patch->recount && !deleted && !added)
1777 return -1;
1779 fragment->leading = leading;
1780 fragment->trailing = trailing;
1783 * If a fragment ends with an incomplete line, we failed to include
1784 * it in the above loop because we hit oldlines == newlines == 0
1785 * before seeing it.
1787 if (12 < size && !memcmp(line, "\\ ", 2))
1788 offset += linelen(line, size);
1790 patch->lines_added += added;
1791 patch->lines_deleted += deleted;
1793 if (0 < patch->is_new && oldlines)
1794 return error(_("new file depends on old contents"));
1795 if (0 < patch->is_delete && newlines)
1796 return error(_("deleted file still has contents"));
1797 return offset;
1801 * We have seen "diff --git a/... b/..." header (or a traditional patch
1802 * header). Read hunks that belong to this patch into fragments and hang
1803 * them to the given patch structure.
1805 * The (fragment->patch, fragment->size) pair points into the memory given
1806 * by the caller, not a copy, when we return.
1808 * Returns:
1809 * -1 in case of error,
1810 * the number of bytes in the patch otherwise.
1812 static int parse_single_patch(struct apply_state *state,
1813 const char *line,
1814 unsigned long size,
1815 struct patch *patch)
1817 unsigned long offset = 0;
1818 unsigned long oldlines = 0, newlines = 0, context = 0;
1819 struct fragment **fragp = &patch->fragments;
1821 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1822 struct fragment *fragment;
1823 int len;
1825 CALLOC_ARRAY(fragment, 1);
1826 fragment->linenr = state->linenr;
1827 len = parse_fragment(state, line, size, patch, fragment);
1828 if (len <= 0) {
1829 free(fragment);
1830 return error(_("corrupt patch at line %d"), state->linenr);
1832 fragment->patch = line;
1833 fragment->size = len;
1834 oldlines += fragment->oldlines;
1835 newlines += fragment->newlines;
1836 context += fragment->leading + fragment->trailing;
1838 *fragp = fragment;
1839 fragp = &fragment->next;
1841 offset += len;
1842 line += len;
1843 size -= len;
1847 * If something was removed (i.e. we have old-lines) it cannot
1848 * be creation, and if something was added it cannot be
1849 * deletion. However, the reverse is not true; --unified=0
1850 * patches that only add are not necessarily creation even
1851 * though they do not have any old lines, and ones that only
1852 * delete are not necessarily deletion.
1854 * Unfortunately, a real creation/deletion patch do _not_ have
1855 * any context line by definition, so we cannot safely tell it
1856 * apart with --unified=0 insanity. At least if the patch has
1857 * more than one hunk it is not creation or deletion.
1859 if (patch->is_new < 0 &&
1860 (oldlines || (patch->fragments && patch->fragments->next)))
1861 patch->is_new = 0;
1862 if (patch->is_delete < 0 &&
1863 (newlines || (patch->fragments && patch->fragments->next)))
1864 patch->is_delete = 0;
1866 if (0 < patch->is_new && oldlines)
1867 return error(_("new file %s depends on old contents"), patch->new_name);
1868 if (0 < patch->is_delete && newlines)
1869 return error(_("deleted file %s still has contents"), patch->old_name);
1870 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1871 fprintf_ln(stderr,
1872 _("** warning: "
1873 "file %s becomes empty but is not deleted"),
1874 patch->new_name);
1876 return offset;
1879 static inline int metadata_changes(struct patch *patch)
1881 return patch->is_rename > 0 ||
1882 patch->is_copy > 0 ||
1883 patch->is_new > 0 ||
1884 patch->is_delete ||
1885 (patch->old_mode && patch->new_mode &&
1886 patch->old_mode != patch->new_mode);
1889 static char *inflate_it(const void *data, unsigned long size,
1890 unsigned long inflated_size)
1892 git_zstream stream;
1893 void *out;
1894 int st;
1896 memset(&stream, 0, sizeof(stream));
1898 stream.next_in = (unsigned char *)data;
1899 stream.avail_in = size;
1900 stream.next_out = out = xmalloc(inflated_size);
1901 stream.avail_out = inflated_size;
1902 git_inflate_init(&stream);
1903 st = git_inflate(&stream, Z_FINISH);
1904 git_inflate_end(&stream);
1905 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1906 free(out);
1907 return NULL;
1909 return out;
1913 * Read a binary hunk and return a new fragment; fragment->patch
1914 * points at an allocated memory that the caller must free, so
1915 * it is marked as "->free_patch = 1".
1917 static struct fragment *parse_binary_hunk(struct apply_state *state,
1918 char **buf_p,
1919 unsigned long *sz_p,
1920 int *status_p,
1921 int *used_p)
1924 * Expect a line that begins with binary patch method ("literal"
1925 * or "delta"), followed by the length of data before deflating.
1926 * a sequence of 'length-byte' followed by base-85 encoded data
1927 * should follow, terminated by a newline.
1929 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1930 * and we would limit the patch line to 66 characters,
1931 * so one line can fit up to 13 groups that would decode
1932 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1933 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1935 int llen, used;
1936 unsigned long size = *sz_p;
1937 char *buffer = *buf_p;
1938 int patch_method;
1939 unsigned long origlen;
1940 char *data = NULL;
1941 int hunk_size = 0;
1942 struct fragment *frag;
1944 llen = linelen(buffer, size);
1945 used = llen;
1947 *status_p = 0;
1949 if (starts_with(buffer, "delta ")) {
1950 patch_method = BINARY_DELTA_DEFLATED;
1951 origlen = strtoul(buffer + 6, NULL, 10);
1953 else if (starts_with(buffer, "literal ")) {
1954 patch_method = BINARY_LITERAL_DEFLATED;
1955 origlen = strtoul(buffer + 8, NULL, 10);
1957 else
1958 return NULL;
1960 state->linenr++;
1961 buffer += llen;
1962 size -= llen;
1963 while (1) {
1964 int byte_length, max_byte_length, newsize;
1965 llen = linelen(buffer, size);
1966 used += llen;
1967 state->linenr++;
1968 if (llen == 1) {
1969 /* consume the blank line */
1970 buffer++;
1971 size--;
1972 break;
1975 * Minimum line is "A00000\n" which is 7-byte long,
1976 * and the line length must be multiple of 5 plus 2.
1978 if ((llen < 7) || (llen-2) % 5)
1979 goto corrupt;
1980 max_byte_length = (llen - 2) / 5 * 4;
1981 byte_length = *buffer;
1982 if ('A' <= byte_length && byte_length <= 'Z')
1983 byte_length = byte_length - 'A' + 1;
1984 else if ('a' <= byte_length && byte_length <= 'z')
1985 byte_length = byte_length - 'a' + 27;
1986 else
1987 goto corrupt;
1988 /* if the input length was not multiple of 4, we would
1989 * have filler at the end but the filler should never
1990 * exceed 3 bytes
1992 if (max_byte_length < byte_length ||
1993 byte_length <= max_byte_length - 4)
1994 goto corrupt;
1995 newsize = hunk_size + byte_length;
1996 data = xrealloc(data, newsize);
1997 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1998 goto corrupt;
1999 hunk_size = newsize;
2000 buffer += llen;
2001 size -= llen;
2004 CALLOC_ARRAY(frag, 1);
2005 frag->patch = inflate_it(data, hunk_size, origlen);
2006 frag->free_patch = 1;
2007 if (!frag->patch)
2008 goto corrupt;
2009 free(data);
2010 frag->size = origlen;
2011 *buf_p = buffer;
2012 *sz_p = size;
2013 *used_p = used;
2014 frag->binary_patch_method = patch_method;
2015 return frag;
2017 corrupt:
2018 free(data);
2019 *status_p = -1;
2020 error(_("corrupt binary patch at line %d: %.*s"),
2021 state->linenr-1, llen-1, buffer);
2022 return NULL;
2026 * Returns:
2027 * -1 in case of error,
2028 * the length of the parsed binary patch otherwise
2030 static int parse_binary(struct apply_state *state,
2031 char *buffer,
2032 unsigned long size,
2033 struct patch *patch)
2036 * We have read "GIT binary patch\n"; what follows is a line
2037 * that says the patch method (currently, either "literal" or
2038 * "delta") and the length of data before deflating; a
2039 * sequence of 'length-byte' followed by base-85 encoded data
2040 * follows.
2042 * When a binary patch is reversible, there is another binary
2043 * hunk in the same format, starting with patch method (either
2044 * "literal" or "delta") with the length of data, and a sequence
2045 * of length-byte + base-85 encoded data, terminated with another
2046 * empty line. This data, when applied to the postimage, produces
2047 * the preimage.
2049 struct fragment *forward;
2050 struct fragment *reverse;
2051 int status;
2052 int used, used_1;
2054 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2055 if (!forward && !status)
2056 /* there has to be one hunk (forward hunk) */
2057 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2058 if (status)
2059 /* otherwise we already gave an error message */
2060 return status;
2062 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2063 if (reverse)
2064 used += used_1;
2065 else if (status) {
2067 * Not having reverse hunk is not an error, but having
2068 * a corrupt reverse hunk is.
2070 free((void*) forward->patch);
2071 free(forward);
2072 return status;
2074 forward->next = reverse;
2075 patch->fragments = forward;
2076 patch->is_binary = 1;
2077 return used;
2080 static void prefix_one(struct apply_state *state, char **name)
2082 char *old_name = *name;
2083 if (!old_name)
2084 return;
2085 *name = prefix_filename(state->prefix, *name);
2086 free(old_name);
2089 static void prefix_patch(struct apply_state *state, struct patch *p)
2091 if (!state->prefix || p->is_toplevel_relative)
2092 return;
2093 prefix_one(state, &p->new_name);
2094 prefix_one(state, &p->old_name);
2098 * include/exclude
2101 static void add_name_limit(struct apply_state *state,
2102 const char *name,
2103 int exclude)
2105 struct string_list_item *it;
2107 it = string_list_append(&state->limit_by_name, name);
2108 it->util = exclude ? NULL : (void *) 1;
2111 static int use_patch(struct apply_state *state, struct patch *p)
2113 const char *pathname = p->new_name ? p->new_name : p->old_name;
2114 int i;
2116 /* Paths outside are not touched regardless of "--include" */
2117 if (state->prefix && *state->prefix) {
2118 const char *rest;
2119 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2120 return 0;
2123 /* See if it matches any of exclude/include rule */
2124 for (i = 0; i < state->limit_by_name.nr; i++) {
2125 struct string_list_item *it = &state->limit_by_name.items[i];
2126 if (!wildmatch(it->string, pathname, 0))
2127 return (it->util != NULL);
2131 * If we had any include, a path that does not match any rule is
2132 * not used. Otherwise, we saw bunch of exclude rules (or none)
2133 * and such a path is used.
2135 return !state->has_include;
2139 * Read the patch text in "buffer" that extends for "size" bytes; stop
2140 * reading after seeing a single patch (i.e. changes to a single file).
2141 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2143 * Returns:
2144 * -1 if no header was found or parse_binary() failed,
2145 * -128 on another error,
2146 * the number of bytes consumed otherwise,
2147 * so that the caller can call us again for the next patch.
2149 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2151 int hdrsize, patchsize;
2152 int offset = find_header(state, buffer, size, &hdrsize, patch);
2154 if (offset < 0)
2155 return offset;
2157 prefix_patch(state, patch);
2159 if (!use_patch(state, patch))
2160 patch->ws_rule = 0;
2161 else if (patch->new_name)
2162 patch->ws_rule = whitespace_rule(state->repo->index,
2163 patch->new_name);
2164 else
2165 patch->ws_rule = whitespace_rule(state->repo->index,
2166 patch->old_name);
2168 patchsize = parse_single_patch(state,
2169 buffer + offset + hdrsize,
2170 size - offset - hdrsize,
2171 patch);
2173 if (patchsize < 0)
2174 return -128;
2176 if (!patchsize) {
2177 static const char git_binary[] = "GIT binary patch\n";
2178 int hd = hdrsize + offset;
2179 unsigned long llen = linelen(buffer + hd, size - hd);
2181 if (llen == sizeof(git_binary) - 1 &&
2182 !memcmp(git_binary, buffer + hd, llen)) {
2183 int used;
2184 state->linenr++;
2185 used = parse_binary(state, buffer + hd + llen,
2186 size - hd - llen, patch);
2187 if (used < 0)
2188 return -1;
2189 if (used)
2190 patchsize = used + llen;
2191 else
2192 patchsize = 0;
2194 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2195 static const char *binhdr[] = {
2196 "Binary files ",
2197 "Files ",
2198 NULL,
2200 int i;
2201 for (i = 0; binhdr[i]; i++) {
2202 int len = strlen(binhdr[i]);
2203 if (len < size - hd &&
2204 !memcmp(binhdr[i], buffer + hd, len)) {
2205 state->linenr++;
2206 patch->is_binary = 1;
2207 patchsize = llen;
2208 break;
2213 /* Empty patch cannot be applied if it is a text patch
2214 * without metadata change. A binary patch appears
2215 * empty to us here.
2217 if ((state->apply || state->check) &&
2218 (!patch->is_binary && !metadata_changes(patch))) {
2219 error(_("patch with only garbage at line %d"), state->linenr);
2220 return -128;
2224 return offset + hdrsize + patchsize;
2227 static void reverse_patches(struct patch *p)
2229 for (; p; p = p->next) {
2230 struct fragment *frag = p->fragments;
2232 SWAP(p->new_name, p->old_name);
2233 if (p->new_mode)
2234 SWAP(p->new_mode, p->old_mode);
2235 SWAP(p->is_new, p->is_delete);
2236 SWAP(p->lines_added, p->lines_deleted);
2237 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2239 for (; frag; frag = frag->next) {
2240 SWAP(frag->newpos, frag->oldpos);
2241 SWAP(frag->newlines, frag->oldlines);
2246 static const char pluses[] =
2247 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2248 static const char minuses[]=
2249 "----------------------------------------------------------------------";
2251 static void show_stats(struct apply_state *state, struct patch *patch)
2253 struct strbuf qname = STRBUF_INIT;
2254 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2255 int max, add, del;
2257 quote_c_style(cp, &qname, NULL, 0);
2260 * "scale" the filename
2262 max = state->max_len;
2263 if (max > 50)
2264 max = 50;
2266 if (qname.len > max) {
2267 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2268 if (!cp)
2269 cp = qname.buf + qname.len + 3 - max;
2270 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2273 if (patch->is_binary) {
2274 printf(" %-*s | Bin\n", max, qname.buf);
2275 strbuf_release(&qname);
2276 return;
2279 printf(" %-*s |", max, qname.buf);
2280 strbuf_release(&qname);
2283 * scale the add/delete
2285 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2286 add = patch->lines_added;
2287 del = patch->lines_deleted;
2289 if (state->max_change > 0) {
2290 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2291 add = (add * max + state->max_change / 2) / state->max_change;
2292 del = total - add;
2294 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2295 add, pluses, del, minuses);
2298 static int read_old_data(struct stat *st, struct patch *patch,
2299 const char *path, struct strbuf *buf)
2301 int conv_flags = patch->crlf_in_old ?
2302 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2303 switch (st->st_mode & S_IFMT) {
2304 case S_IFLNK:
2305 if (strbuf_readlink(buf, path, st->st_size) < 0)
2306 return error(_("unable to read symlink %s"), path);
2307 return 0;
2308 case S_IFREG:
2309 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2310 return error(_("unable to open or read %s"), path);
2312 * "git apply" without "--index/--cached" should never look
2313 * at the index; the target file may not have been added to
2314 * the index yet, and we may not even be in any Git repository.
2315 * Pass NULL to convert_to_git() to stress this; the function
2316 * should never look at the index when explicit crlf option
2317 * is given.
2319 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2320 return 0;
2321 default:
2322 return -1;
2327 * Update the preimage, and the common lines in postimage,
2328 * from buffer buf of length len. If postlen is 0 the postimage
2329 * is updated in place, otherwise it's updated on a new buffer
2330 * of length postlen
2333 static void update_pre_post_images(struct image *preimage,
2334 struct image *postimage,
2335 char *buf,
2336 size_t len, size_t postlen)
2338 int i, ctx, reduced;
2339 char *new_buf, *old_buf, *fixed;
2340 struct image fixed_preimage;
2343 * Update the preimage with whitespace fixes. Note that we
2344 * are not losing preimage->buf -- apply_one_fragment() will
2345 * free "oldlines".
2347 prepare_image(&fixed_preimage, buf, len, 1);
2348 assert(postlen
2349 ? fixed_preimage.nr == preimage->nr
2350 : fixed_preimage.nr <= preimage->nr);
2351 for (i = 0; i < fixed_preimage.nr; i++)
2352 fixed_preimage.line[i].flag = preimage->line[i].flag;
2353 free(preimage->line_allocated);
2354 *preimage = fixed_preimage;
2357 * Adjust the common context lines in postimage. This can be
2358 * done in-place when we are shrinking it with whitespace
2359 * fixing, but needs a new buffer when ignoring whitespace or
2360 * expanding leading tabs to spaces.
2362 * We trust the caller to tell us if the update can be done
2363 * in place (postlen==0) or not.
2365 old_buf = postimage->buf;
2366 if (postlen)
2367 new_buf = postimage->buf = xmalloc(postlen);
2368 else
2369 new_buf = old_buf;
2370 fixed = preimage->buf;
2372 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2373 size_t l_len = postimage->line[i].len;
2374 if (!(postimage->line[i].flag & LINE_COMMON)) {
2375 /* an added line -- no counterparts in preimage */
2376 memmove(new_buf, old_buf, l_len);
2377 old_buf += l_len;
2378 new_buf += l_len;
2379 continue;
2382 /* a common context -- skip it in the original postimage */
2383 old_buf += l_len;
2385 /* and find the corresponding one in the fixed preimage */
2386 while (ctx < preimage->nr &&
2387 !(preimage->line[ctx].flag & LINE_COMMON)) {
2388 fixed += preimage->line[ctx].len;
2389 ctx++;
2393 * preimage is expected to run out, if the caller
2394 * fixed addition of trailing blank lines.
2396 if (preimage->nr <= ctx) {
2397 reduced++;
2398 continue;
2401 /* and copy it in, while fixing the line length */
2402 l_len = preimage->line[ctx].len;
2403 memcpy(new_buf, fixed, l_len);
2404 new_buf += l_len;
2405 fixed += l_len;
2406 postimage->line[i].len = l_len;
2407 ctx++;
2410 if (postlen
2411 ? postlen < new_buf - postimage->buf
2412 : postimage->len < new_buf - postimage->buf)
2413 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2414 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2416 /* Fix the length of the whole thing */
2417 postimage->len = new_buf - postimage->buf;
2418 postimage->nr -= reduced;
2421 static int line_by_line_fuzzy_match(struct image *img,
2422 struct image *preimage,
2423 struct image *postimage,
2424 unsigned long current,
2425 int current_lno,
2426 int preimage_limit)
2428 int i;
2429 size_t imgoff = 0;
2430 size_t preoff = 0;
2431 size_t postlen = postimage->len;
2432 size_t extra_chars;
2433 char *buf;
2434 char *preimage_eof;
2435 char *preimage_end;
2436 struct strbuf fixed;
2437 char *fixed_buf;
2438 size_t fixed_len;
2440 for (i = 0; i < preimage_limit; i++) {
2441 size_t prelen = preimage->line[i].len;
2442 size_t imglen = img->line[current_lno+i].len;
2444 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2445 preimage->buf + preoff, prelen))
2446 return 0;
2447 if (preimage->line[i].flag & LINE_COMMON)
2448 postlen += imglen - prelen;
2449 imgoff += imglen;
2450 preoff += prelen;
2454 * Ok, the preimage matches with whitespace fuzz.
2456 * imgoff now holds the true length of the target that
2457 * matches the preimage before the end of the file.
2459 * Count the number of characters in the preimage that fall
2460 * beyond the end of the file and make sure that all of them
2461 * are whitespace characters. (This can only happen if
2462 * we are removing blank lines at the end of the file.)
2464 buf = preimage_eof = preimage->buf + preoff;
2465 for ( ; i < preimage->nr; i++)
2466 preoff += preimage->line[i].len;
2467 preimage_end = preimage->buf + preoff;
2468 for ( ; buf < preimage_end; buf++)
2469 if (!isspace(*buf))
2470 return 0;
2473 * Update the preimage and the common postimage context
2474 * lines to use the same whitespace as the target.
2475 * If whitespace is missing in the target (i.e.
2476 * if the preimage extends beyond the end of the file),
2477 * use the whitespace from the preimage.
2479 extra_chars = preimage_end - preimage_eof;
2480 strbuf_init(&fixed, imgoff + extra_chars);
2481 strbuf_add(&fixed, img->buf + current, imgoff);
2482 strbuf_add(&fixed, preimage_eof, extra_chars);
2483 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2484 update_pre_post_images(preimage, postimage,
2485 fixed_buf, fixed_len, postlen);
2486 return 1;
2489 static int match_fragment(struct apply_state *state,
2490 struct image *img,
2491 struct image *preimage,
2492 struct image *postimage,
2493 unsigned long current,
2494 int current_lno,
2495 unsigned ws_rule,
2496 int match_beginning, int match_end)
2498 int i;
2499 const char *orig, *target;
2500 struct strbuf fixed = STRBUF_INIT;
2501 size_t postlen;
2502 int preimage_limit;
2503 int ret;
2505 if (preimage->nr + current_lno <= img->nr) {
2507 * The hunk falls within the boundaries of img.
2509 preimage_limit = preimage->nr;
2510 if (match_end && (preimage->nr + current_lno != img->nr)) {
2511 ret = 0;
2512 goto out;
2514 } else if (state->ws_error_action == correct_ws_error &&
2515 (ws_rule & WS_BLANK_AT_EOF)) {
2517 * This hunk extends beyond the end of img, and we are
2518 * removing blank lines at the end of the file. This
2519 * many lines from the beginning of the preimage must
2520 * match with img, and the remainder of the preimage
2521 * must be blank.
2523 preimage_limit = img->nr - current_lno;
2524 } else {
2526 * The hunk extends beyond the end of the img and
2527 * we are not removing blanks at the end, so we
2528 * should reject the hunk at this position.
2530 ret = 0;
2531 goto out;
2534 if (match_beginning && current_lno) {
2535 ret = 0;
2536 goto out;
2539 /* Quick hash check */
2540 for (i = 0; i < preimage_limit; i++) {
2541 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2542 (preimage->line[i].hash != img->line[current_lno + i].hash)) {
2543 ret = 0;
2544 goto out;
2548 if (preimage_limit == preimage->nr) {
2550 * Do we have an exact match? If we were told to match
2551 * at the end, size must be exactly at current+fragsize,
2552 * otherwise current+fragsize must be still within the preimage,
2553 * and either case, the old piece should match the preimage
2554 * exactly.
2556 if ((match_end
2557 ? (current + preimage->len == img->len)
2558 : (current + preimage->len <= img->len)) &&
2559 !memcmp(img->buf + current, preimage->buf, preimage->len)) {
2560 ret = 1;
2561 goto out;
2563 } else {
2565 * The preimage extends beyond the end of img, so
2566 * there cannot be an exact match.
2568 * There must be one non-blank context line that match
2569 * a line before the end of img.
2571 const char *buf, *buf_end;
2573 buf = preimage->buf;
2574 buf_end = buf;
2575 for (i = 0; i < preimage_limit; i++)
2576 buf_end += preimage->line[i].len;
2578 for ( ; buf < buf_end; buf++)
2579 if (!isspace(*buf))
2580 break;
2581 if (buf == buf_end) {
2582 ret = 0;
2583 goto out;
2588 * No exact match. If we are ignoring whitespace, run a line-by-line
2589 * fuzzy matching. We collect all the line length information because
2590 * we need it to adjust whitespace if we match.
2592 if (state->ws_ignore_action == ignore_ws_change) {
2593 ret = line_by_line_fuzzy_match(img, preimage, postimage,
2594 current, current_lno, preimage_limit);
2595 goto out;
2598 if (state->ws_error_action != correct_ws_error) {
2599 ret = 0;
2600 goto out;
2604 * The hunk does not apply byte-by-byte, but the hash says
2605 * it might with whitespace fuzz. We weren't asked to
2606 * ignore whitespace, we were asked to correct whitespace
2607 * errors, so let's try matching after whitespace correction.
2609 * While checking the preimage against the target, whitespace
2610 * errors in both fixed, we count how large the corresponding
2611 * postimage needs to be. The postimage prepared by
2612 * apply_one_fragment() has whitespace errors fixed on added
2613 * lines already, but the common lines were propagated as-is,
2614 * which may become longer when their whitespace errors are
2615 * fixed.
2618 /* First count added lines in postimage */
2619 postlen = 0;
2620 for (i = 0; i < postimage->nr; i++) {
2621 if (!(postimage->line[i].flag & LINE_COMMON))
2622 postlen += postimage->line[i].len;
2626 * The preimage may extend beyond the end of the file,
2627 * but in this loop we will only handle the part of the
2628 * preimage that falls within the file.
2630 strbuf_grow(&fixed, preimage->len + 1);
2631 orig = preimage->buf;
2632 target = img->buf + current;
2633 for (i = 0; i < preimage_limit; i++) {
2634 size_t oldlen = preimage->line[i].len;
2635 size_t tgtlen = img->line[current_lno + i].len;
2636 size_t fixstart = fixed.len;
2637 struct strbuf tgtfix;
2638 int match;
2640 /* Try fixing the line in the preimage */
2641 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2643 /* Try fixing the line in the target */
2644 strbuf_init(&tgtfix, tgtlen);
2645 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2648 * If they match, either the preimage was based on
2649 * a version before our tree fixed whitespace breakage,
2650 * or we are lacking a whitespace-fix patch the tree
2651 * the preimage was based on already had (i.e. target
2652 * has whitespace breakage, the preimage doesn't).
2653 * In either case, we are fixing the whitespace breakages
2654 * so we might as well take the fix together with their
2655 * real change.
2657 match = (tgtfix.len == fixed.len - fixstart &&
2658 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2659 fixed.len - fixstart));
2661 /* Add the length if this is common with the postimage */
2662 if (preimage->line[i].flag & LINE_COMMON)
2663 postlen += tgtfix.len;
2665 strbuf_release(&tgtfix);
2666 if (!match) {
2667 ret = 0;
2668 goto out;
2671 orig += oldlen;
2672 target += tgtlen;
2677 * Now handle the lines in the preimage that falls beyond the
2678 * end of the file (if any). They will only match if they are
2679 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2680 * false).
2682 for ( ; i < preimage->nr; i++) {
2683 size_t fixstart = fixed.len; /* start of the fixed preimage */
2684 size_t oldlen = preimage->line[i].len;
2685 int j;
2687 /* Try fixing the line in the preimage */
2688 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2690 for (j = fixstart; j < fixed.len; j++) {
2691 if (!isspace(fixed.buf[j])) {
2692 ret = 0;
2693 goto out;
2698 orig += oldlen;
2702 * Yes, the preimage is based on an older version that still
2703 * has whitespace breakages unfixed, and fixing them makes the
2704 * hunk match. Update the context lines in the postimage.
2706 if (postlen < postimage->len)
2707 postlen = 0;
2708 update_pre_post_images(preimage, postimage,
2709 fixed.buf, fixed.len, postlen);
2711 ret = 1;
2713 out:
2714 strbuf_release(&fixed);
2715 return ret;
2718 static int find_pos(struct apply_state *state,
2719 struct image *img,
2720 struct image *preimage,
2721 struct image *postimage,
2722 int line,
2723 unsigned ws_rule,
2724 int match_beginning, int match_end)
2726 int i;
2727 unsigned long backwards, forwards, current;
2728 int backwards_lno, forwards_lno, current_lno;
2731 * When running with --allow-overlap, it is possible that a hunk is
2732 * seen that pretends to start at the beginning (but no longer does),
2733 * and that *still* needs to match the end. So trust `match_end` more
2734 * than `match_beginning`.
2736 if (state->allow_overlap && match_beginning && match_end &&
2737 img->nr - preimage->nr != 0)
2738 match_beginning = 0;
2741 * If match_beginning or match_end is specified, there is no
2742 * point starting from a wrong line that will never match and
2743 * wander around and wait for a match at the specified end.
2745 if (match_beginning)
2746 line = 0;
2747 else if (match_end)
2748 line = img->nr - preimage->nr;
2751 * Because the comparison is unsigned, the following test
2752 * will also take care of a negative line number that can
2753 * result when match_end and preimage is larger than the target.
2755 if ((size_t) line > img->nr)
2756 line = img->nr;
2758 current = 0;
2759 for (i = 0; i < line; i++)
2760 current += img->line[i].len;
2763 * There's probably some smart way to do this, but I'll leave
2764 * that to the smart and beautiful people. I'm simple and stupid.
2766 backwards = current;
2767 backwards_lno = line;
2768 forwards = current;
2769 forwards_lno = line;
2770 current_lno = line;
2772 for (i = 0; ; i++) {
2773 if (match_fragment(state, img, preimage, postimage,
2774 current, current_lno, ws_rule,
2775 match_beginning, match_end))
2776 return current_lno;
2778 again:
2779 if (backwards_lno == 0 && forwards_lno == img->nr)
2780 break;
2782 if (i & 1) {
2783 if (backwards_lno == 0) {
2784 i++;
2785 goto again;
2787 backwards_lno--;
2788 backwards -= img->line[backwards_lno].len;
2789 current = backwards;
2790 current_lno = backwards_lno;
2791 } else {
2792 if (forwards_lno == img->nr) {
2793 i++;
2794 goto again;
2796 forwards += img->line[forwards_lno].len;
2797 forwards_lno++;
2798 current = forwards;
2799 current_lno = forwards_lno;
2803 return -1;
2806 static void remove_first_line(struct image *img)
2808 img->buf += img->line[0].len;
2809 img->len -= img->line[0].len;
2810 img->line++;
2811 img->nr--;
2814 static void remove_last_line(struct image *img)
2816 img->len -= img->line[--img->nr].len;
2820 * The change from "preimage" and "postimage" has been found to
2821 * apply at applied_pos (counts in line numbers) in "img".
2822 * Update "img" to remove "preimage" and replace it with "postimage".
2824 static void update_image(struct apply_state *state,
2825 struct image *img,
2826 int applied_pos,
2827 struct image *preimage,
2828 struct image *postimage)
2831 * remove the copy of preimage at offset in img
2832 * and replace it with postimage
2834 int i, nr;
2835 size_t remove_count, insert_count, applied_at = 0;
2836 char *result;
2837 int preimage_limit;
2840 * If we are removing blank lines at the end of img,
2841 * the preimage may extend beyond the end.
2842 * If that is the case, we must be careful only to
2843 * remove the part of the preimage that falls within
2844 * the boundaries of img. Initialize preimage_limit
2845 * to the number of lines in the preimage that falls
2846 * within the boundaries.
2848 preimage_limit = preimage->nr;
2849 if (preimage_limit > img->nr - applied_pos)
2850 preimage_limit = img->nr - applied_pos;
2852 for (i = 0; i < applied_pos; i++)
2853 applied_at += img->line[i].len;
2855 remove_count = 0;
2856 for (i = 0; i < preimage_limit; i++)
2857 remove_count += img->line[applied_pos + i].len;
2858 insert_count = postimage->len;
2860 /* Adjust the contents */
2861 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2862 memcpy(result, img->buf, applied_at);
2863 memcpy(result + applied_at, postimage->buf, postimage->len);
2864 memcpy(result + applied_at + postimage->len,
2865 img->buf + (applied_at + remove_count),
2866 img->len - (applied_at + remove_count));
2867 free(img->buf);
2868 img->buf = result;
2869 img->len += insert_count - remove_count;
2870 result[img->len] = '\0';
2872 /* Adjust the line table */
2873 nr = img->nr + postimage->nr - preimage_limit;
2874 if (preimage_limit < postimage->nr) {
2876 * NOTE: this knows that we never call remove_first_line()
2877 * on anything other than pre/post image.
2879 REALLOC_ARRAY(img->line, nr);
2880 img->line_allocated = img->line;
2882 if (preimage_limit != postimage->nr)
2883 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2884 img->line + applied_pos + preimage_limit,
2885 img->nr - (applied_pos + preimage_limit));
2886 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2887 if (!state->allow_overlap)
2888 for (i = 0; i < postimage->nr; i++)
2889 img->line[applied_pos + i].flag |= LINE_PATCHED;
2890 img->nr = nr;
2894 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2895 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2896 * replace the part of "img" with "postimage" text.
2898 static int apply_one_fragment(struct apply_state *state,
2899 struct image *img, struct fragment *frag,
2900 int inaccurate_eof, unsigned ws_rule,
2901 int nth_fragment)
2903 int match_beginning, match_end;
2904 const char *patch = frag->patch;
2905 int size = frag->size;
2906 char *old, *oldlines;
2907 struct strbuf newlines;
2908 int new_blank_lines_at_end = 0;
2909 int found_new_blank_lines_at_end = 0;
2910 int hunk_linenr = frag->linenr;
2911 unsigned long leading, trailing;
2912 int pos, applied_pos;
2913 struct image preimage;
2914 struct image postimage;
2916 memset(&preimage, 0, sizeof(preimage));
2917 memset(&postimage, 0, sizeof(postimage));
2918 oldlines = xmalloc(size);
2919 strbuf_init(&newlines, size);
2921 old = oldlines;
2922 while (size > 0) {
2923 char first;
2924 int len = linelen(patch, size);
2925 int plen;
2926 int added_blank_line = 0;
2927 int is_blank_context = 0;
2928 size_t start;
2930 if (!len)
2931 break;
2934 * "plen" is how much of the line we should use for
2935 * the actual patch data. Normally we just remove the
2936 * first character on the line, but if the line is
2937 * followed by "\ No newline", then we also remove the
2938 * last one (which is the newline, of course).
2940 plen = len - 1;
2941 if (len < size && patch[len] == '\\')
2942 plen--;
2943 first = *patch;
2944 if (state->apply_in_reverse) {
2945 if (first == '-')
2946 first = '+';
2947 else if (first == '+')
2948 first = '-';
2951 switch (first) {
2952 case '\n':
2953 /* Newer GNU diff, empty context line */
2954 if (plen < 0)
2955 /* ... followed by '\No newline'; nothing */
2956 break;
2957 *old++ = '\n';
2958 strbuf_addch(&newlines, '\n');
2959 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2960 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2961 is_blank_context = 1;
2962 break;
2963 case ' ':
2964 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2965 ws_blank_line(patch + 1, plen))
2966 is_blank_context = 1;
2967 /* fallthrough */
2968 case '-':
2969 memcpy(old, patch + 1, plen);
2970 add_line_info(&preimage, old, plen,
2971 (first == ' ' ? LINE_COMMON : 0));
2972 old += plen;
2973 if (first == '-')
2974 break;
2975 /* fallthrough */
2976 case '+':
2977 /* --no-add does not add new lines */
2978 if (first == '+' && state->no_add)
2979 break;
2981 start = newlines.len;
2982 if (first != '+' ||
2983 !state->whitespace_error ||
2984 state->ws_error_action != correct_ws_error) {
2985 strbuf_add(&newlines, patch + 1, plen);
2987 else {
2988 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2990 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2991 (first == '+' ? 0 : LINE_COMMON));
2992 if (first == '+' &&
2993 (ws_rule & WS_BLANK_AT_EOF) &&
2994 ws_blank_line(patch + 1, plen))
2995 added_blank_line = 1;
2996 break;
2997 case '@': case '\\':
2998 /* Ignore it, we already handled it */
2999 break;
3000 default:
3001 if (state->apply_verbosity > verbosity_normal)
3002 error(_("invalid start of line: '%c'"), first);
3003 applied_pos = -1;
3004 goto out;
3006 if (added_blank_line) {
3007 if (!new_blank_lines_at_end)
3008 found_new_blank_lines_at_end = hunk_linenr;
3009 new_blank_lines_at_end++;
3011 else if (is_blank_context)
3013 else
3014 new_blank_lines_at_end = 0;
3015 patch += len;
3016 size -= len;
3017 hunk_linenr++;
3019 if (inaccurate_eof &&
3020 old > oldlines && old[-1] == '\n' &&
3021 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
3022 old--;
3023 strbuf_setlen(&newlines, newlines.len - 1);
3024 preimage.line_allocated[preimage.nr - 1].len--;
3025 postimage.line_allocated[postimage.nr - 1].len--;
3028 leading = frag->leading;
3029 trailing = frag->trailing;
3032 * A hunk to change lines at the beginning would begin with
3033 * @@ -1,L +N,M @@
3034 * but we need to be careful. -U0 that inserts before the second
3035 * line also has this pattern.
3037 * And a hunk to add to an empty file would begin with
3038 * @@ -0,0 +N,M @@
3040 * In other words, a hunk that is (frag->oldpos <= 1) with or
3041 * without leading context must match at the beginning.
3043 match_beginning = (!frag->oldpos ||
3044 (frag->oldpos == 1 && !state->unidiff_zero));
3047 * A hunk without trailing lines must match at the end.
3048 * However, we simply cannot tell if a hunk must match end
3049 * from the lack of trailing lines if the patch was generated
3050 * with unidiff without any context.
3052 match_end = !state->unidiff_zero && !trailing;
3054 pos = frag->newpos ? (frag->newpos - 1) : 0;
3055 preimage.buf = oldlines;
3056 preimage.len = old - oldlines;
3057 postimage.buf = newlines.buf;
3058 postimage.len = newlines.len;
3059 preimage.line = preimage.line_allocated;
3060 postimage.line = postimage.line_allocated;
3062 for (;;) {
3064 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3065 ws_rule, match_beginning, match_end);
3067 if (applied_pos >= 0)
3068 break;
3070 /* Am I at my context limits? */
3071 if ((leading <= state->p_context) && (trailing <= state->p_context))
3072 break;
3073 if (match_beginning || match_end) {
3074 match_beginning = match_end = 0;
3075 continue;
3079 * Reduce the number of context lines; reduce both
3080 * leading and trailing if they are equal otherwise
3081 * just reduce the larger context.
3083 if (leading >= trailing) {
3084 remove_first_line(&preimage);
3085 remove_first_line(&postimage);
3086 pos--;
3087 leading--;
3089 if (trailing > leading) {
3090 remove_last_line(&preimage);
3091 remove_last_line(&postimage);
3092 trailing--;
3096 if (applied_pos >= 0) {
3097 if (new_blank_lines_at_end &&
3098 preimage.nr + applied_pos >= img->nr &&
3099 (ws_rule & WS_BLANK_AT_EOF) &&
3100 state->ws_error_action != nowarn_ws_error) {
3101 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3102 found_new_blank_lines_at_end);
3103 if (state->ws_error_action == correct_ws_error) {
3104 while (new_blank_lines_at_end--)
3105 remove_last_line(&postimage);
3108 * We would want to prevent write_out_results()
3109 * from taking place in apply_patch() that follows
3110 * the callchain led us here, which is:
3111 * apply_patch->check_patch_list->check_patch->
3112 * apply_data->apply_fragments->apply_one_fragment
3114 if (state->ws_error_action == die_on_ws_error)
3115 state->apply = 0;
3118 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3119 int offset = applied_pos - pos;
3120 if (state->apply_in_reverse)
3121 offset = 0 - offset;
3122 fprintf_ln(stderr,
3123 Q_("Hunk #%d succeeded at %d (offset %d line).",
3124 "Hunk #%d succeeded at %d (offset %d lines).",
3125 offset),
3126 nth_fragment, applied_pos + 1, offset);
3130 * Warn if it was necessary to reduce the number
3131 * of context lines.
3133 if ((leading != frag->leading ||
3134 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3135 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3136 " to apply fragment at %d"),
3137 leading, trailing, applied_pos+1);
3138 update_image(state, img, applied_pos, &preimage, &postimage);
3139 } else {
3140 if (state->apply_verbosity > verbosity_normal)
3141 error(_("while searching for:\n%.*s"),
3142 (int)(old - oldlines), oldlines);
3145 out:
3146 free(oldlines);
3147 strbuf_release(&newlines);
3148 free(preimage.line_allocated);
3149 free(postimage.line_allocated);
3151 return (applied_pos < 0);
3154 static int apply_binary_fragment(struct apply_state *state,
3155 struct image *img,
3156 struct patch *patch)
3158 struct fragment *fragment = patch->fragments;
3159 unsigned long len;
3160 void *dst;
3162 if (!fragment)
3163 return error(_("missing binary patch data for '%s'"),
3164 patch->new_name ?
3165 patch->new_name :
3166 patch->old_name);
3168 /* Binary patch is irreversible without the optional second hunk */
3169 if (state->apply_in_reverse) {
3170 if (!fragment->next)
3171 return error(_("cannot reverse-apply a binary patch "
3172 "without the reverse hunk to '%s'"),
3173 patch->new_name
3174 ? patch->new_name : patch->old_name);
3175 fragment = fragment->next;
3177 switch (fragment->binary_patch_method) {
3178 case BINARY_DELTA_DEFLATED:
3179 dst = patch_delta(img->buf, img->len, fragment->patch,
3180 fragment->size, &len);
3181 if (!dst)
3182 return -1;
3183 clear_image(img);
3184 img->buf = dst;
3185 img->len = len;
3186 return 0;
3187 case BINARY_LITERAL_DEFLATED:
3188 clear_image(img);
3189 img->len = fragment->size;
3190 img->buf = xmemdupz(fragment->patch, img->len);
3191 return 0;
3193 return -1;
3197 * Replace "img" with the result of applying the binary patch.
3198 * The binary patch data itself in patch->fragment is still kept
3199 * but the preimage prepared by the caller in "img" is freed here
3200 * or in the helper function apply_binary_fragment() this calls.
3202 static int apply_binary(struct apply_state *state,
3203 struct image *img,
3204 struct patch *patch)
3206 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3207 struct object_id oid;
3208 const unsigned hexsz = the_hash_algo->hexsz;
3211 * For safety, we require patch index line to contain
3212 * full hex textual object ID for old and new, at least for now.
3214 if (strlen(patch->old_oid_prefix) != hexsz ||
3215 strlen(patch->new_oid_prefix) != hexsz ||
3216 get_oid_hex(patch->old_oid_prefix, &oid) ||
3217 get_oid_hex(patch->new_oid_prefix, &oid))
3218 return error(_("cannot apply binary patch to '%s' "
3219 "without full index line"), name);
3221 if (patch->old_name) {
3223 * See if the old one matches what the patch
3224 * applies to.
3226 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3227 &oid);
3228 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3229 return error(_("the patch applies to '%s' (%s), "
3230 "which does not match the "
3231 "current contents."),
3232 name, oid_to_hex(&oid));
3234 else {
3235 /* Otherwise, the old one must be empty. */
3236 if (img->len)
3237 return error(_("the patch applies to an empty "
3238 "'%s' but it is not empty"), name);
3241 get_oid_hex(patch->new_oid_prefix, &oid);
3242 if (is_null_oid(&oid)) {
3243 clear_image(img);
3244 return 0; /* deletion patch */
3247 if (has_object(the_repository, &oid, 0)) {
3248 /* We already have the postimage */
3249 enum object_type type;
3250 unsigned long size;
3251 char *result;
3253 result = repo_read_object_file(the_repository, &oid, &type,
3254 &size);
3255 if (!result)
3256 return error(_("the necessary postimage %s for "
3257 "'%s' cannot be read"),
3258 patch->new_oid_prefix, name);
3259 clear_image(img);
3260 img->buf = result;
3261 img->len = size;
3262 } else {
3264 * We have verified buf matches the preimage;
3265 * apply the patch data to it, which is stored
3266 * in the patch->fragments->{patch,size}.
3268 if (apply_binary_fragment(state, img, patch))
3269 return error(_("binary patch does not apply to '%s'"),
3270 name);
3272 /* verify that the result matches */
3273 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3274 &oid);
3275 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3276 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3277 name, patch->new_oid_prefix, oid_to_hex(&oid));
3280 return 0;
3283 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3285 struct fragment *frag = patch->fragments;
3286 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3287 unsigned ws_rule = patch->ws_rule;
3288 unsigned inaccurate_eof = patch->inaccurate_eof;
3289 int nth = 0;
3291 if (patch->is_binary)
3292 return apply_binary(state, img, patch);
3294 while (frag) {
3295 nth++;
3296 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3297 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3298 if (!state->apply_with_reject)
3299 return -1;
3300 frag->rejected = 1;
3302 frag = frag->next;
3304 return 0;
3307 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3309 if (S_ISGITLINK(mode)) {
3310 strbuf_grow(buf, 100);
3311 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3312 } else {
3313 enum object_type type;
3314 unsigned long sz;
3315 char *result;
3317 result = repo_read_object_file(the_repository, oid, &type,
3318 &sz);
3319 if (!result)
3320 return -1;
3321 /* XXX read_sha1_file NUL-terminates */
3322 strbuf_attach(buf, result, sz, sz + 1);
3324 return 0;
3327 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3329 if (!ce)
3330 return 0;
3331 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3334 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3336 struct string_list_item *item;
3338 if (!name)
3339 return NULL;
3341 item = string_list_lookup(&state->fn_table, name);
3342 if (item)
3343 return (struct patch *)item->util;
3345 return NULL;
3349 * item->util in the filename table records the status of the path.
3350 * Usually it points at a patch (whose result records the contents
3351 * of it after applying it), but it could be PATH_WAS_DELETED for a
3352 * path that a previously applied patch has already removed, or
3353 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3355 * The latter is needed to deal with a case where two paths A and B
3356 * are swapped by first renaming A to B and then renaming B to A;
3357 * moving A to B should not be prevented due to presence of B as we
3358 * will remove it in a later patch.
3360 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3361 #define PATH_WAS_DELETED ((struct patch *) -1)
3363 static int to_be_deleted(struct patch *patch)
3365 return patch == PATH_TO_BE_DELETED;
3368 static int was_deleted(struct patch *patch)
3370 return patch == PATH_WAS_DELETED;
3373 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3375 struct string_list_item *item;
3378 * Always add new_name unless patch is a deletion
3379 * This should cover the cases for normal diffs,
3380 * file creations and copies
3382 if (patch->new_name) {
3383 item = string_list_insert(&state->fn_table, patch->new_name);
3384 item->util = patch;
3388 * store a failure on rename/deletion cases because
3389 * later chunks shouldn't patch old names
3391 if ((patch->new_name == NULL) || (patch->is_rename)) {
3392 item = string_list_insert(&state->fn_table, patch->old_name);
3393 item->util = PATH_WAS_DELETED;
3397 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3400 * store information about incoming file deletion
3402 while (patch) {
3403 if ((patch->new_name == NULL) || (patch->is_rename)) {
3404 struct string_list_item *item;
3405 item = string_list_insert(&state->fn_table, patch->old_name);
3406 item->util = PATH_TO_BE_DELETED;
3408 patch = patch->next;
3412 static int checkout_target(struct index_state *istate,
3413 struct cache_entry *ce, struct stat *st)
3415 struct checkout costate = CHECKOUT_INIT;
3417 costate.refresh_cache = 1;
3418 costate.istate = istate;
3419 if (checkout_entry(ce, &costate, NULL, NULL) ||
3420 lstat(ce->name, st))
3421 return error(_("cannot checkout %s"), ce->name);
3422 return 0;
3425 static struct patch *previous_patch(struct apply_state *state,
3426 struct patch *patch,
3427 int *gone)
3429 struct patch *previous;
3431 *gone = 0;
3432 if (patch->is_copy || patch->is_rename)
3433 return NULL; /* "git" patches do not depend on the order */
3435 previous = in_fn_table(state, patch->old_name);
3436 if (!previous)
3437 return NULL;
3439 if (to_be_deleted(previous))
3440 return NULL; /* the deletion hasn't happened yet */
3442 if (was_deleted(previous))
3443 *gone = 1;
3445 return previous;
3448 static int verify_index_match(struct apply_state *state,
3449 const struct cache_entry *ce,
3450 struct stat *st)
3452 if (S_ISGITLINK(ce->ce_mode)) {
3453 if (!S_ISDIR(st->st_mode))
3454 return -1;
3455 return 0;
3457 return ie_match_stat(state->repo->index, ce, st,
3458 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3461 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3463 static int load_patch_target(struct apply_state *state,
3464 struct strbuf *buf,
3465 const struct cache_entry *ce,
3466 struct stat *st,
3467 struct patch *patch,
3468 const char *name,
3469 unsigned expected_mode)
3471 if (state->cached || state->check_index) {
3472 if (read_file_or_gitlink(ce, buf))
3473 return error(_("failed to read %s"), name);
3474 } else if (name) {
3475 if (S_ISGITLINK(expected_mode)) {
3476 if (ce)
3477 return read_file_or_gitlink(ce, buf);
3478 else
3479 return SUBMODULE_PATCH_WITHOUT_INDEX;
3480 } else if (has_symlink_leading_path(name, strlen(name))) {
3481 return error(_("reading from '%s' beyond a symbolic link"), name);
3482 } else {
3483 if (read_old_data(st, patch, name, buf))
3484 return error(_("failed to read %s"), name);
3487 return 0;
3491 * We are about to apply "patch"; populate the "image" with the
3492 * current version we have, from the working tree or from the index,
3493 * depending on the situation e.g. --cached/--index. If we are
3494 * applying a non-git patch that incrementally updates the tree,
3495 * we read from the result of a previous diff.
3497 static int load_preimage(struct apply_state *state,
3498 struct image *image,
3499 struct patch *patch, struct stat *st,
3500 const struct cache_entry *ce)
3502 struct strbuf buf = STRBUF_INIT;
3503 size_t len;
3504 char *img;
3505 struct patch *previous;
3506 int status;
3508 previous = previous_patch(state, patch, &status);
3509 if (status)
3510 return error(_("path %s has been renamed/deleted"),
3511 patch->old_name);
3512 if (previous) {
3513 /* We have a patched copy in memory; use that. */
3514 strbuf_add(&buf, previous->result, previous->resultsize);
3515 } else {
3516 status = load_patch_target(state, &buf, ce, st, patch,
3517 patch->old_name, patch->old_mode);
3518 if (status < 0)
3519 return status;
3520 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3522 * There is no way to apply subproject
3523 * patch without looking at the index.
3524 * NEEDSWORK: shouldn't this be flagged
3525 * as an error???
3527 free_fragment_list(patch->fragments);
3528 patch->fragments = NULL;
3529 } else if (status) {
3530 return error(_("failed to read %s"), patch->old_name);
3534 img = strbuf_detach(&buf, &len);
3535 prepare_image(image, img, len, !patch->is_binary);
3536 return 0;
3539 static int resolve_to(struct image *image, const struct object_id *result_id)
3541 unsigned long size;
3542 enum object_type type;
3544 clear_image(image);
3546 image->buf = repo_read_object_file(the_repository, result_id, &type,
3547 &size);
3548 if (!image->buf || type != OBJ_BLOB)
3549 die("unable to read blob object %s", oid_to_hex(result_id));
3550 image->len = size;
3552 return 0;
3555 static int three_way_merge(struct apply_state *state,
3556 struct image *image,
3557 char *path,
3558 const struct object_id *base,
3559 const struct object_id *ours,
3560 const struct object_id *theirs)
3562 mmfile_t base_file, our_file, their_file;
3563 mmbuffer_t result = { NULL };
3564 enum ll_merge_result status;
3566 /* resolve trivial cases first */
3567 if (oideq(base, ours))
3568 return resolve_to(image, theirs);
3569 else if (oideq(base, theirs) || oideq(ours, theirs))
3570 return resolve_to(image, ours);
3572 read_mmblob(&base_file, base);
3573 read_mmblob(&our_file, ours);
3574 read_mmblob(&their_file, theirs);
3575 status = ll_merge(&result, path,
3576 &base_file, "base",
3577 &our_file, "ours",
3578 &their_file, "theirs",
3579 state->repo->index,
3580 NULL);
3581 if (status == LL_MERGE_BINARY_CONFLICT)
3582 warning("Cannot merge binary files: %s (%s vs. %s)",
3583 path, "ours", "theirs");
3584 free(base_file.ptr);
3585 free(our_file.ptr);
3586 free(their_file.ptr);
3587 if (status < 0 || !result.ptr) {
3588 free(result.ptr);
3589 return -1;
3591 clear_image(image);
3592 image->buf = result.ptr;
3593 image->len = result.size;
3595 return status;
3599 * When directly falling back to add/add three-way merge, we read from
3600 * the current contents of the new_name. In no cases other than that
3601 * this function will be called.
3603 static int load_current(struct apply_state *state,
3604 struct image *image,
3605 struct patch *patch)
3607 struct strbuf buf = STRBUF_INIT;
3608 int status, pos;
3609 size_t len;
3610 char *img;
3611 struct stat st;
3612 struct cache_entry *ce;
3613 char *name = patch->new_name;
3614 unsigned mode = patch->new_mode;
3616 if (!patch->is_new)
3617 BUG("patch to %s is not a creation", patch->old_name);
3619 pos = index_name_pos(state->repo->index, name, strlen(name));
3620 if (pos < 0)
3621 return error(_("%s: does not exist in index"), name);
3622 ce = state->repo->index->cache[pos];
3623 if (lstat(name, &st)) {
3624 if (errno != ENOENT)
3625 return error_errno("%s", name);
3626 if (checkout_target(state->repo->index, ce, &st))
3627 return -1;
3629 if (verify_index_match(state, ce, &st))
3630 return error(_("%s: does not match index"), name);
3632 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3633 if (status < 0)
3634 return status;
3635 else if (status)
3636 return -1;
3637 img = strbuf_detach(&buf, &len);
3638 prepare_image(image, img, len, !patch->is_binary);
3639 return 0;
3642 static int try_threeway(struct apply_state *state,
3643 struct image *image,
3644 struct patch *patch,
3645 struct stat *st,
3646 const struct cache_entry *ce)
3648 struct object_id pre_oid, post_oid, our_oid;
3649 struct strbuf buf = STRBUF_INIT;
3650 size_t len;
3651 int status;
3652 char *img;
3653 struct image tmp_image;
3655 /* No point falling back to 3-way merge in these cases */
3656 if (patch->is_delete ||
3657 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3658 (patch->is_new && !patch->direct_to_threeway) ||
3659 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3660 return -1;
3662 /* Preimage the patch was prepared for */
3663 if (patch->is_new)
3664 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3665 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3666 read_blob_object(&buf, &pre_oid, patch->old_mode))
3667 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3669 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3670 fprintf(stderr, _("Performing three-way merge...\n"));
3672 img = strbuf_detach(&buf, &len);
3673 prepare_image(&tmp_image, img, len, 1);
3674 /* Apply the patch to get the post image */
3675 if (apply_fragments(state, &tmp_image, patch) < 0) {
3676 clear_image(&tmp_image);
3677 return -1;
3679 /* post_oid is theirs */
3680 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3681 clear_image(&tmp_image);
3683 /* our_oid is ours */
3684 if (patch->is_new) {
3685 if (load_current(state, &tmp_image, patch))
3686 return error(_("cannot read the current contents of '%s'"),
3687 patch->new_name);
3688 } else {
3689 if (load_preimage(state, &tmp_image, patch, st, ce))
3690 return error(_("cannot read the current contents of '%s'"),
3691 patch->old_name);
3693 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3694 clear_image(&tmp_image);
3696 /* in-core three-way merge between post and our using pre as base */
3697 status = three_way_merge(state, image, patch->new_name,
3698 &pre_oid, &our_oid, &post_oid);
3699 if (status < 0) {
3700 if (state->apply_verbosity > verbosity_silent)
3701 fprintf(stderr,
3702 _("Failed to perform three-way merge...\n"));
3703 return status;
3706 if (status) {
3707 patch->conflicted_threeway = 1;
3708 if (patch->is_new)
3709 oidclr(&patch->threeway_stage[0], the_repository->hash_algo);
3710 else
3711 oidcpy(&patch->threeway_stage[0], &pre_oid);
3712 oidcpy(&patch->threeway_stage[1], &our_oid);
3713 oidcpy(&patch->threeway_stage[2], &post_oid);
3714 if (state->apply_verbosity > verbosity_silent)
3715 fprintf(stderr,
3716 _("Applied patch to '%s' with conflicts.\n"),
3717 patch->new_name);
3718 } else {
3719 if (state->apply_verbosity > verbosity_silent)
3720 fprintf(stderr,
3721 _("Applied patch to '%s' cleanly.\n"),
3722 patch->new_name);
3724 return 0;
3727 static int apply_data(struct apply_state *state, struct patch *patch,
3728 struct stat *st, const struct cache_entry *ce)
3730 struct image image;
3732 if (load_preimage(state, &image, patch, st, ce) < 0)
3733 return -1;
3735 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3736 if (state->apply_verbosity > verbosity_silent &&
3737 state->threeway && !patch->direct_to_threeway)
3738 fprintf(stderr, _("Falling back to direct application...\n"));
3740 /* Note: with --reject, apply_fragments() returns 0 */
3741 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0) {
3742 clear_image(&image);
3743 return -1;
3746 patch->result = image.buf;
3747 patch->resultsize = image.len;
3748 add_to_fn_table(state, patch);
3749 free(image.line_allocated);
3751 if (0 < patch->is_delete && patch->resultsize)
3752 return error(_("removal patch leaves file contents"));
3754 return 0;
3758 * If "patch" that we are looking at modifies or deletes what we have,
3759 * we would want it not to lose any local modification we have, either
3760 * in the working tree or in the index.
3762 * This also decides if a non-git patch is a creation patch or a
3763 * modification to an existing empty file. We do not check the state
3764 * of the current tree for a creation patch in this function; the caller
3765 * check_patch() separately makes sure (and errors out otherwise) that
3766 * the path the patch creates does not exist in the current tree.
3768 static int check_preimage(struct apply_state *state,
3769 struct patch *patch,
3770 struct cache_entry **ce,
3771 struct stat *st)
3773 const char *old_name = patch->old_name;
3774 struct patch *previous = NULL;
3775 int stat_ret = 0, status;
3776 unsigned st_mode = 0;
3778 if (!old_name)
3779 return 0;
3781 assert(patch->is_new <= 0);
3782 previous = previous_patch(state, patch, &status);
3784 if (status)
3785 return error(_("path %s has been renamed/deleted"), old_name);
3786 if (previous) {
3787 st_mode = previous->new_mode;
3788 } else if (!state->cached) {
3789 stat_ret = lstat(old_name, st);
3790 if (stat_ret && errno != ENOENT)
3791 return error_errno("%s", old_name);
3794 if (state->check_index && !previous) {
3795 int pos = index_name_pos(state->repo->index, old_name,
3796 strlen(old_name));
3797 if (pos < 0) {
3798 if (patch->is_new < 0)
3799 goto is_new;
3800 return error(_("%s: does not exist in index"), old_name);
3802 *ce = state->repo->index->cache[pos];
3803 if (stat_ret < 0) {
3804 if (checkout_target(state->repo->index, *ce, st))
3805 return -1;
3807 if (!state->cached && verify_index_match(state, *ce, st))
3808 return error(_("%s: does not match index"), old_name);
3809 if (state->cached)
3810 st_mode = (*ce)->ce_mode;
3811 } else if (stat_ret < 0) {
3812 if (patch->is_new < 0)
3813 goto is_new;
3814 return error_errno("%s", old_name);
3817 if (!state->cached && !previous) {
3818 if (*ce && !(*ce)->ce_mode)
3819 BUG("ce_mode == 0 for path '%s'", old_name);
3821 if (trust_executable_bit)
3822 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3823 else if (*ce)
3824 st_mode = (*ce)->ce_mode;
3825 else
3826 st_mode = patch->old_mode;
3829 if (patch->is_new < 0)
3830 patch->is_new = 0;
3831 if (!patch->old_mode)
3832 patch->old_mode = st_mode;
3833 if ((st_mode ^ patch->old_mode) & S_IFMT)
3834 return error(_("%s: wrong type"), old_name);
3835 if (st_mode != patch->old_mode)
3836 warning(_("%s has type %o, expected %o"),
3837 old_name, st_mode, patch->old_mode);
3838 if (!patch->new_mode && !patch->is_delete)
3839 patch->new_mode = st_mode;
3840 return 0;
3842 is_new:
3843 patch->is_new = 1;
3844 patch->is_delete = 0;
3845 FREE_AND_NULL(patch->old_name);
3846 return 0;
3850 #define EXISTS_IN_INDEX 1
3851 #define EXISTS_IN_WORKTREE 2
3852 #define EXISTS_IN_INDEX_AS_ITA 3
3854 static int check_to_create(struct apply_state *state,
3855 const char *new_name,
3856 int ok_if_exists)
3858 struct stat nst;
3860 if (state->check_index && (!ok_if_exists || !state->cached)) {
3861 int pos;
3863 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3864 if (pos >= 0) {
3865 struct cache_entry *ce = state->repo->index->cache[pos];
3867 /* allow ITA, as they do not yet exist in the index */
3868 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3869 return EXISTS_IN_INDEX;
3871 /* ITA entries can never match working tree files */
3872 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3873 return EXISTS_IN_INDEX_AS_ITA;
3877 if (state->cached)
3878 return 0;
3880 if (!lstat(new_name, &nst)) {
3881 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3882 return 0;
3884 * A leading component of new_name might be a symlink
3885 * that is going to be removed with this patch, but
3886 * still pointing at somewhere that has the path.
3887 * In such a case, path "new_name" does not exist as
3888 * far as git is concerned.
3890 if (has_symlink_leading_path(new_name, strlen(new_name)))
3891 return 0;
3893 return EXISTS_IN_WORKTREE;
3894 } else if (!is_missing_file_error(errno)) {
3895 return error_errno("%s", new_name);
3897 return 0;
3900 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3902 for ( ; patch; patch = patch->next) {
3903 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3904 (patch->is_rename || patch->is_delete))
3905 /* the symlink at patch->old_name is removed */
3906 strset_add(&state->removed_symlinks, patch->old_name);
3908 if (patch->new_name && S_ISLNK(patch->new_mode))
3909 /* the symlink at patch->new_name is created or remains */
3910 strset_add(&state->kept_symlinks, patch->new_name);
3914 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3916 do {
3917 while (--name->len && name->buf[name->len] != '/')
3918 ; /* scan backwards */
3919 if (!name->len)
3920 break;
3921 name->buf[name->len] = '\0';
3922 if (strset_contains(&state->kept_symlinks, name->buf))
3923 return 1;
3924 if (strset_contains(&state->removed_symlinks, name->buf))
3926 * This cannot be "return 0", because we may
3927 * see a new one created at a higher level.
3929 continue;
3931 /* otherwise, check the preimage */
3932 if (state->check_index) {
3933 struct cache_entry *ce;
3935 ce = index_file_exists(state->repo->index, name->buf,
3936 name->len, ignore_case);
3937 if (ce && S_ISLNK(ce->ce_mode))
3938 return 1;
3939 } else {
3940 struct stat st;
3941 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3942 return 1;
3944 } while (1);
3945 return 0;
3948 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3950 int ret;
3951 struct strbuf name = STRBUF_INIT;
3953 assert(*name_ != '\0');
3954 strbuf_addstr(&name, name_);
3955 ret = path_is_beyond_symlink_1(state, &name);
3956 strbuf_release(&name);
3958 return ret;
3961 static int check_unsafe_path(struct patch *patch)
3963 const char *old_name = NULL;
3964 const char *new_name = NULL;
3965 if (patch->is_delete)
3966 old_name = patch->old_name;
3967 else if (!patch->is_new && !patch->is_copy)
3968 old_name = patch->old_name;
3969 if (!patch->is_delete)
3970 new_name = patch->new_name;
3972 if (old_name && !verify_path(old_name, patch->old_mode))
3973 return error(_("invalid path '%s'"), old_name);
3974 if (new_name && !verify_path(new_name, patch->new_mode))
3975 return error(_("invalid path '%s'"), new_name);
3976 return 0;
3980 * Check and apply the patch in-core; leave the result in patch->result
3981 * for the caller to write it out to the final destination.
3983 static int check_patch(struct apply_state *state, struct patch *patch)
3985 struct stat st;
3986 const char *old_name = patch->old_name;
3987 const char *new_name = patch->new_name;
3988 const char *name = old_name ? old_name : new_name;
3989 struct cache_entry *ce = NULL;
3990 struct patch *tpatch;
3991 int ok_if_exists;
3992 int status;
3994 patch->rejected = 1; /* we will drop this after we succeed */
3996 status = check_preimage(state, patch, &ce, &st);
3997 if (status)
3998 return status;
3999 old_name = patch->old_name;
4002 * A type-change diff is always split into a patch to delete
4003 * old, immediately followed by a patch to create new (see
4004 * diff.c::run_diff()); in such a case it is Ok that the entry
4005 * to be deleted by the previous patch is still in the working
4006 * tree and in the index.
4008 * A patch to swap-rename between A and B would first rename A
4009 * to B and then rename B to A. While applying the first one,
4010 * the presence of B should not stop A from getting renamed to
4011 * B; ask to_be_deleted() about the later rename. Removal of
4012 * B and rename from A to B is handled the same way by asking
4013 * was_deleted().
4015 if ((tpatch = in_fn_table(state, new_name)) &&
4016 (was_deleted(tpatch) || to_be_deleted(tpatch)))
4017 ok_if_exists = 1;
4018 else
4019 ok_if_exists = 0;
4021 if (new_name &&
4022 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
4023 int err = check_to_create(state, new_name, ok_if_exists);
4025 if (err && state->threeway) {
4026 patch->direct_to_threeway = 1;
4027 } else switch (err) {
4028 case 0:
4029 break; /* happy */
4030 case EXISTS_IN_INDEX:
4031 return error(_("%s: already exists in index"), new_name);
4032 case EXISTS_IN_INDEX_AS_ITA:
4033 return error(_("%s: does not match index"), new_name);
4034 case EXISTS_IN_WORKTREE:
4035 return error(_("%s: already exists in working directory"),
4036 new_name);
4037 default:
4038 return err;
4041 if (!patch->new_mode) {
4042 if (0 < patch->is_new)
4043 patch->new_mode = S_IFREG | 0644;
4044 else
4045 patch->new_mode = patch->old_mode;
4049 if (new_name && old_name) {
4050 int same = !strcmp(old_name, new_name);
4051 if (!patch->new_mode)
4052 patch->new_mode = patch->old_mode;
4053 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4054 if (same)
4055 return error(_("new mode (%o) of %s does not "
4056 "match old mode (%o)"),
4057 patch->new_mode, new_name,
4058 patch->old_mode);
4059 else
4060 return error(_("new mode (%o) of %s does not "
4061 "match old mode (%o) of %s"),
4062 patch->new_mode, new_name,
4063 patch->old_mode, old_name);
4067 if (!state->unsafe_paths && check_unsafe_path(patch))
4068 return -128;
4071 * An attempt to read from or delete a path that is beyond a
4072 * symbolic link will be prevented by load_patch_target() that
4073 * is called at the beginning of apply_data() so we do not
4074 * have to worry about a patch marked with "is_delete" bit
4075 * here. We however need to make sure that the patch result
4076 * is not deposited to a path that is beyond a symbolic link
4077 * here.
4079 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4080 return error(_("affected file '%s' is beyond a symbolic link"),
4081 patch->new_name);
4083 if (apply_data(state, patch, &st, ce) < 0)
4084 return error(_("%s: patch does not apply"), name);
4085 patch->rejected = 0;
4086 return 0;
4089 static int check_patch_list(struct apply_state *state, struct patch *patch)
4091 int err = 0;
4093 prepare_symlink_changes(state, patch);
4094 prepare_fn_table(state, patch);
4095 while (patch) {
4096 int res;
4097 if (state->apply_verbosity > verbosity_normal)
4098 say_patch_name(stderr,
4099 _("Checking patch %s..."), patch);
4100 res = check_patch(state, patch);
4101 if (res == -128)
4102 return -128;
4103 err |= res;
4104 patch = patch->next;
4106 return err;
4109 static int read_apply_cache(struct apply_state *state)
4111 if (state->index_file)
4112 return read_index_from(state->repo->index, state->index_file,
4113 get_git_dir());
4114 else
4115 return repo_read_index(state->repo);
4118 /* This function tries to read the object name from the current index */
4119 static int get_current_oid(struct apply_state *state, const char *path,
4120 struct object_id *oid)
4122 int pos;
4124 if (read_apply_cache(state) < 0)
4125 return -1;
4126 pos = index_name_pos(state->repo->index, path, strlen(path));
4127 if (pos < 0)
4128 return -1;
4129 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4130 return 0;
4133 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4136 * A usable gitlink patch has only one fragment (hunk) that looks like:
4137 * @@ -1 +1 @@
4138 * -Subproject commit <old sha1>
4139 * +Subproject commit <new sha1>
4140 * or
4141 * @@ -1 +0,0 @@
4142 * -Subproject commit <old sha1>
4143 * for a removal patch.
4145 struct fragment *hunk = p->fragments;
4146 static const char heading[] = "-Subproject commit ";
4147 char *preimage;
4149 if (/* does the patch have only one hunk? */
4150 hunk && !hunk->next &&
4151 /* is its preimage one line? */
4152 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4153 /* does preimage begin with the heading? */
4154 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4155 starts_with(++preimage, heading) &&
4156 /* does it record full SHA-1? */
4157 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4158 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4159 /* does the abbreviated name on the index line agree with it? */
4160 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4161 return 0; /* it all looks fine */
4163 /* we may have full object name on the index line */
4164 return get_oid_hex(p->old_oid_prefix, oid);
4167 /* Build an index that contains just the files needed for a 3way merge */
4168 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4170 struct patch *patch;
4171 struct index_state result = INDEX_STATE_INIT(state->repo);
4172 struct lock_file lock = LOCK_INIT;
4173 int res;
4175 /* Once we start supporting the reverse patch, it may be
4176 * worth showing the new sha1 prefix, but until then...
4178 for (patch = list; patch; patch = patch->next) {
4179 struct object_id oid;
4180 struct cache_entry *ce;
4181 const char *name;
4183 name = patch->old_name ? patch->old_name : patch->new_name;
4184 if (0 < patch->is_new)
4185 continue;
4187 if (S_ISGITLINK(patch->old_mode)) {
4188 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4189 ; /* ok, the textual part looks sane */
4190 else
4191 return error(_("sha1 information is lacking or "
4192 "useless for submodule %s"), name);
4193 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4194 ; /* ok */
4195 } else if (!patch->lines_added && !patch->lines_deleted) {
4196 /* mode-only change: update the current */
4197 if (get_current_oid(state, patch->old_name, &oid))
4198 return error(_("mode change for %s, which is not "
4199 "in current HEAD"), name);
4200 } else
4201 return error(_("sha1 information is lacking or useless "
4202 "(%s)."), name);
4204 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4205 if (!ce)
4206 return error(_("make_cache_entry failed for path '%s'"),
4207 name);
4208 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4209 discard_cache_entry(ce);
4210 return error(_("could not add %s to temporary index"),
4211 name);
4215 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4216 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4217 discard_index(&result);
4219 if (res)
4220 return error(_("could not write temporary index to %s"),
4221 state->fake_ancestor);
4223 return 0;
4226 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4228 int files, adds, dels;
4230 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4231 files++;
4232 adds += patch->lines_added;
4233 dels += patch->lines_deleted;
4234 show_stats(state, patch);
4237 print_stat_summary(stdout, files, adds, dels);
4240 static void numstat_patch_list(struct apply_state *state,
4241 struct patch *patch)
4243 for ( ; patch; patch = patch->next) {
4244 const char *name;
4245 name = patch->new_name ? patch->new_name : patch->old_name;
4246 if (patch->is_binary)
4247 printf("-\t-\t");
4248 else
4249 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4250 write_name_quoted(name, stdout, state->line_termination);
4254 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4256 if (mode)
4257 printf(" %s mode %06o %s\n", newdelete, mode, name);
4258 else
4259 printf(" %s %s\n", newdelete, name);
4262 static void show_mode_change(struct patch *p, int show_name)
4264 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4265 if (show_name)
4266 printf(" mode change %06o => %06o %s\n",
4267 p->old_mode, p->new_mode, p->new_name);
4268 else
4269 printf(" mode change %06o => %06o\n",
4270 p->old_mode, p->new_mode);
4274 static void show_rename_copy(struct patch *p)
4276 const char *renamecopy = p->is_rename ? "rename" : "copy";
4277 const char *old_name, *new_name;
4279 /* Find common prefix */
4280 old_name = p->old_name;
4281 new_name = p->new_name;
4282 while (1) {
4283 const char *slash_old, *slash_new;
4284 slash_old = strchr(old_name, '/');
4285 slash_new = strchr(new_name, '/');
4286 if (!slash_old ||
4287 !slash_new ||
4288 slash_old - old_name != slash_new - new_name ||
4289 memcmp(old_name, new_name, slash_new - new_name))
4290 break;
4291 old_name = slash_old + 1;
4292 new_name = slash_new + 1;
4294 /* p->old_name through old_name is the common prefix, and old_name and
4295 * new_name through the end of names are renames
4297 if (old_name != p->old_name)
4298 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4299 (int)(old_name - p->old_name), p->old_name,
4300 old_name, new_name, p->score);
4301 else
4302 printf(" %s %s => %s (%d%%)\n", renamecopy,
4303 p->old_name, p->new_name, p->score);
4304 show_mode_change(p, 0);
4307 static void summary_patch_list(struct patch *patch)
4309 struct patch *p;
4311 for (p = patch; p; p = p->next) {
4312 if (p->is_new)
4313 show_file_mode_name("create", p->new_mode, p->new_name);
4314 else if (p->is_delete)
4315 show_file_mode_name("delete", p->old_mode, p->old_name);
4316 else {
4317 if (p->is_rename || p->is_copy)
4318 show_rename_copy(p);
4319 else {
4320 if (p->score) {
4321 printf(" rewrite %s (%d%%)\n",
4322 p->new_name, p->score);
4323 show_mode_change(p, 0);
4325 else
4326 show_mode_change(p, 1);
4332 static void patch_stats(struct apply_state *state, struct patch *patch)
4334 int lines = patch->lines_added + patch->lines_deleted;
4336 if (lines > state->max_change)
4337 state->max_change = lines;
4338 if (patch->old_name) {
4339 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4340 if (!len)
4341 len = strlen(patch->old_name);
4342 if (len > state->max_len)
4343 state->max_len = len;
4345 if (patch->new_name) {
4346 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4347 if (!len)
4348 len = strlen(patch->new_name);
4349 if (len > state->max_len)
4350 state->max_len = len;
4354 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4356 if (state->update_index && !state->ita_only) {
4357 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4358 return error(_("unable to remove %s from index"), patch->old_name);
4360 if (!state->cached) {
4361 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4362 remove_path(patch->old_name);
4365 return 0;
4368 static int add_index_file(struct apply_state *state,
4369 const char *path,
4370 unsigned mode,
4371 void *buf,
4372 unsigned long size)
4374 struct stat st;
4375 struct cache_entry *ce;
4376 int namelen = strlen(path);
4378 ce = make_empty_cache_entry(state->repo->index, namelen);
4379 memcpy(ce->name, path, namelen);
4380 ce->ce_mode = create_ce_mode(mode);
4381 ce->ce_flags = create_ce_flags(0);
4382 ce->ce_namelen = namelen;
4383 if (state->ita_only) {
4384 ce->ce_flags |= CE_INTENT_TO_ADD;
4385 set_object_name_for_intent_to_add_entry(ce);
4386 } else if (S_ISGITLINK(mode)) {
4387 const char *s;
4389 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4390 get_oid_hex(s, &ce->oid)) {
4391 discard_cache_entry(ce);
4392 return error(_("corrupt patch for submodule %s"), path);
4394 } else {
4395 if (!state->cached) {
4396 if (lstat(path, &st) < 0) {
4397 discard_cache_entry(ce);
4398 return error_errno(_("unable to stat newly "
4399 "created file '%s'"),
4400 path);
4402 fill_stat_cache_info(state->repo->index, ce, &st);
4404 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4405 discard_cache_entry(ce);
4406 return error(_("unable to create backing store "
4407 "for newly created file %s"), path);
4410 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4411 discard_cache_entry(ce);
4412 return error(_("unable to add cache entry for %s"), path);
4415 return 0;
4419 * Returns:
4420 * -1 if an unrecoverable error happened
4421 * 0 if everything went well
4422 * 1 if a recoverable error happened
4424 static int try_create_file(struct apply_state *state, const char *path,
4425 unsigned int mode, const char *buf,
4426 unsigned long size)
4428 int fd, res;
4429 struct strbuf nbuf = STRBUF_INIT;
4431 if (S_ISGITLINK(mode)) {
4432 struct stat st;
4433 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4434 return 0;
4435 return !!mkdir(path, 0777);
4438 if (has_symlinks && S_ISLNK(mode))
4439 /* Although buf:size is counted string, it also is NUL
4440 * terminated.
4442 return !!symlink(buf, path);
4444 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4445 if (fd < 0)
4446 return 1;
4448 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4449 size = nbuf.len;
4450 buf = nbuf.buf;
4453 res = write_in_full(fd, buf, size) < 0;
4454 if (res)
4455 error_errno(_("failed to write to '%s'"), path);
4456 strbuf_release(&nbuf);
4458 if (close(fd) < 0 && !res)
4459 return error_errno(_("closing file '%s'"), path);
4461 return res ? -1 : 0;
4465 * We optimistically assume that the directories exist,
4466 * which is true 99% of the time anyway. If they don't,
4467 * we create them and try again.
4469 * Returns:
4470 * -1 on error
4471 * 0 otherwise
4473 static int create_one_file(struct apply_state *state,
4474 char *path,
4475 unsigned mode,
4476 const char *buf,
4477 unsigned long size)
4479 char *newpath = NULL;
4480 int res;
4482 if (state->cached)
4483 return 0;
4486 * We already try to detect whether files are beyond a symlink in our
4487 * up-front checks. But in the case where symlinks are created by any
4488 * of the intermediate hunks it can happen that our up-front checks
4489 * didn't yet see the symlink, but at the point of arriving here there
4490 * in fact is one. We thus repeat the check for symlinks here.
4492 * Note that this does not make the up-front check obsolete as the
4493 * failure mode is different:
4495 * - The up-front checks cause us to abort before we have written
4496 * anything into the working directory. So when we exit this way the
4497 * working directory remains clean.
4499 * - The checks here happen in the middle of the action where we have
4500 * already started to apply the patch. The end result will be a dirty
4501 * working directory.
4503 * Ideally, we should update the up-front checks to catch what would
4504 * happen when we apply the patch before we damage the working tree.
4505 * We have all the information necessary to do so. But for now, as a
4506 * part of embargoed security work, having this check would serve as a
4507 * reasonable first step.
4509 if (path_is_beyond_symlink(state, path))
4510 return error(_("affected file '%s' is beyond a symbolic link"), path);
4512 res = try_create_file(state, path, mode, buf, size);
4513 if (res < 0)
4514 return -1;
4515 if (!res)
4516 return 0;
4518 if (errno == ENOENT) {
4519 if (safe_create_leading_directories_no_share(path))
4520 return 0;
4521 res = try_create_file(state, path, mode, buf, size);
4522 if (res < 0)
4523 return -1;
4524 if (!res)
4525 return 0;
4528 if (errno == EEXIST || errno == EACCES) {
4529 /* We may be trying to create a file where a directory
4530 * used to be.
4532 struct stat st;
4533 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4534 errno = EEXIST;
4537 if (errno == EEXIST) {
4538 unsigned int nr = getpid();
4540 for (;;) {
4541 newpath = mkpathdup("%s~%u", path, nr);
4542 res = try_create_file(state, newpath, mode, buf, size);
4543 if (res < 0)
4544 goto out;
4545 if (!res) {
4546 if (!rename(newpath, path))
4547 goto out;
4548 unlink_or_warn(newpath);
4549 break;
4551 if (errno != EEXIST)
4552 break;
4553 ++nr;
4554 FREE_AND_NULL(newpath);
4557 res = error_errno(_("unable to write file '%s' mode %o"), path, mode);
4558 out:
4559 free(newpath);
4560 return res;
4563 static int add_conflicted_stages_file(struct apply_state *state,
4564 struct patch *patch)
4566 int stage, namelen;
4567 unsigned mode;
4568 struct cache_entry *ce;
4570 if (!state->update_index)
4571 return 0;
4572 namelen = strlen(patch->new_name);
4573 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4575 remove_file_from_index(state->repo->index, patch->new_name);
4576 for (stage = 1; stage < 4; stage++) {
4577 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4578 continue;
4579 ce = make_empty_cache_entry(state->repo->index, namelen);
4580 memcpy(ce->name, patch->new_name, namelen);
4581 ce->ce_mode = create_ce_mode(mode);
4582 ce->ce_flags = create_ce_flags(stage);
4583 ce->ce_namelen = namelen;
4584 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4585 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4586 discard_cache_entry(ce);
4587 return error(_("unable to add cache entry for %s"),
4588 patch->new_name);
4592 return 0;
4595 static int create_file(struct apply_state *state, struct patch *patch)
4597 char *path = patch->new_name;
4598 unsigned mode = patch->new_mode;
4599 unsigned long size = patch->resultsize;
4600 char *buf = patch->result;
4602 if (!mode)
4603 mode = S_IFREG | 0644;
4604 if (create_one_file(state, path, mode, buf, size))
4605 return -1;
4607 if (patch->conflicted_threeway)
4608 return add_conflicted_stages_file(state, patch);
4609 else if (state->update_index)
4610 return add_index_file(state, path, mode, buf, size);
4611 return 0;
4614 /* phase zero is to remove, phase one is to create */
4615 static int write_out_one_result(struct apply_state *state,
4616 struct patch *patch,
4617 int phase)
4619 if (patch->is_delete > 0) {
4620 if (phase == 0)
4621 return remove_file(state, patch, 1);
4622 return 0;
4624 if (patch->is_new > 0 || patch->is_copy) {
4625 if (phase == 1)
4626 return create_file(state, patch);
4627 return 0;
4630 * Rename or modification boils down to the same
4631 * thing: remove the old, write the new
4633 if (phase == 0)
4634 return remove_file(state, patch, patch->is_rename);
4635 if (phase == 1)
4636 return create_file(state, patch);
4637 return 0;
4640 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4642 FILE *rej;
4643 char *namebuf;
4644 struct fragment *frag;
4645 int fd, cnt = 0;
4646 struct strbuf sb = STRBUF_INIT;
4648 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4649 if (!frag->rejected)
4650 continue;
4651 cnt++;
4654 if (!cnt) {
4655 if (state->apply_verbosity > verbosity_normal)
4656 say_patch_name(stderr,
4657 _("Applied patch %s cleanly."), patch);
4658 return 0;
4661 /* This should not happen, because a removal patch that leaves
4662 * contents are marked "rejected" at the patch level.
4664 if (!patch->new_name)
4665 die(_("internal error"));
4667 /* Say this even without --verbose */
4668 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4669 "Applying patch %%s with %d rejects...",
4670 cnt),
4671 cnt);
4672 if (state->apply_verbosity > verbosity_silent)
4673 say_patch_name(stderr, sb.buf, patch);
4674 strbuf_release(&sb);
4676 namebuf = xstrfmt("%s.rej", patch->new_name);
4678 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4679 if (fd < 0) {
4680 if (errno != EEXIST) {
4681 error_errno(_("cannot open %s"), namebuf);
4682 goto error;
4684 if (unlink(namebuf)) {
4685 error_errno(_("cannot unlink '%s'"), namebuf);
4686 goto error;
4688 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4689 if (fd < 0) {
4690 error_errno(_("cannot open %s"), namebuf);
4691 goto error;
4694 rej = fdopen(fd, "w");
4695 if (!rej) {
4696 error_errno(_("cannot open %s"), namebuf);
4697 close(fd);
4698 goto error;
4701 /* Normal git tools never deal with .rej, so do not pretend
4702 * this is a git patch by saying --git or giving extended
4703 * headers. While at it, maybe please "kompare" that wants
4704 * the trailing TAB and some garbage at the end of line ;-).
4706 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4707 patch->new_name, patch->new_name);
4708 for (cnt = 1, frag = patch->fragments;
4709 frag;
4710 cnt++, frag = frag->next) {
4711 if (!frag->rejected) {
4712 if (state->apply_verbosity > verbosity_silent)
4713 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4714 continue;
4716 if (state->apply_verbosity > verbosity_silent)
4717 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4718 fprintf(rej, "%.*s", frag->size, frag->patch);
4719 if (frag->patch[frag->size-1] != '\n')
4720 fputc('\n', rej);
4722 fclose(rej);
4723 error:
4724 free(namebuf);
4725 return -1;
4729 * Returns:
4730 * -1 if an error happened
4731 * 0 if the patch applied cleanly
4732 * 1 if the patch did not apply cleanly
4734 static int write_out_results(struct apply_state *state, struct patch *list)
4736 int phase;
4737 int errs = 0;
4738 struct patch *l;
4739 struct string_list cpath = STRING_LIST_INIT_DUP;
4741 for (phase = 0; phase < 2; phase++) {
4742 l = list;
4743 while (l) {
4744 if (l->rejected)
4745 errs = 1;
4746 else {
4747 if (write_out_one_result(state, l, phase)) {
4748 string_list_clear(&cpath, 0);
4749 return -1;
4751 if (phase == 1) {
4752 if (write_out_one_reject(state, l))
4753 errs = 1;
4754 if (l->conflicted_threeway) {
4755 string_list_append(&cpath, l->new_name);
4756 errs = 1;
4760 l = l->next;
4764 if (cpath.nr) {
4765 struct string_list_item *item;
4767 string_list_sort(&cpath);
4768 if (state->apply_verbosity > verbosity_silent) {
4769 for_each_string_list_item(item, &cpath)
4770 fprintf(stderr, "U %s\n", item->string);
4772 string_list_clear(&cpath, 0);
4775 * rerere relies on the partially merged result being in the working
4776 * tree with conflict markers, but that isn't written with --cached.
4778 if (!state->cached)
4779 repo_rerere(state->repo, 0);
4782 return errs;
4786 * Try to apply a patch.
4788 * Returns:
4789 * -128 if a bad error happened (like patch unreadable)
4790 * -1 if patch did not apply and user cannot deal with it
4791 * 0 if the patch applied
4792 * 1 if the patch did not apply but user might fix it
4794 static int apply_patch(struct apply_state *state,
4795 int fd,
4796 const char *filename,
4797 int options)
4799 size_t offset;
4800 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4801 struct patch *list = NULL, **listp = &list;
4802 int skipped_patch = 0;
4803 int res = 0;
4804 int flush_attributes = 0;
4806 state->patch_input_file = filename;
4807 if (read_patch_file(&buf, fd) < 0)
4808 return -128;
4809 offset = 0;
4810 while (offset < buf.len) {
4811 struct patch *patch;
4812 int nr;
4814 CALLOC_ARRAY(patch, 1);
4815 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4816 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4817 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4818 if (nr < 0) {
4819 free_patch(patch);
4820 if (nr == -128) {
4821 res = -128;
4822 goto end;
4824 break;
4826 if (state->apply_in_reverse)
4827 reverse_patches(patch);
4828 if (use_patch(state, patch)) {
4829 patch_stats(state, patch);
4830 if (!list || !state->apply_in_reverse) {
4831 *listp = patch;
4832 listp = &patch->next;
4833 } else {
4834 patch->next = list;
4835 list = patch;
4838 if ((patch->new_name &&
4839 ends_with_path_components(patch->new_name,
4840 GITATTRIBUTES_FILE)) ||
4841 (patch->old_name &&
4842 ends_with_path_components(patch->old_name,
4843 GITATTRIBUTES_FILE)))
4844 flush_attributes = 1;
4846 else {
4847 if (state->apply_verbosity > verbosity_normal)
4848 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4849 free_patch(patch);
4850 skipped_patch++;
4852 offset += nr;
4855 if (!list && !skipped_patch) {
4856 if (!state->allow_empty) {
4857 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4858 res = -128;
4860 goto end;
4863 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4864 state->apply = 0;
4866 state->update_index = (state->check_index || state->ita_only) && state->apply;
4867 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4868 if (state->index_file)
4869 hold_lock_file_for_update(&state->lock_file,
4870 state->index_file,
4871 LOCK_DIE_ON_ERROR);
4872 else
4873 repo_hold_locked_index(state->repo, &state->lock_file,
4874 LOCK_DIE_ON_ERROR);
4877 if (state->check_index && read_apply_cache(state) < 0) {
4878 error(_("unable to read index file"));
4879 res = -128;
4880 goto end;
4883 if (state->check || state->apply) {
4884 int r = check_patch_list(state, list);
4885 if (r == -128) {
4886 res = -128;
4887 goto end;
4889 if (r < 0 && !state->apply_with_reject) {
4890 res = -1;
4891 goto end;
4895 if (state->apply) {
4896 int write_res = write_out_results(state, list);
4897 if (write_res < 0) {
4898 res = -128;
4899 goto end;
4901 if (write_res > 0) {
4902 /* with --3way, we still need to write the index out */
4903 res = state->apply_with_reject ? -1 : 1;
4904 goto end;
4908 if (state->fake_ancestor &&
4909 build_fake_ancestor(state, list)) {
4910 res = -128;
4911 goto end;
4914 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4915 stat_patch_list(state, list);
4917 if (state->numstat && state->apply_verbosity > verbosity_silent)
4918 numstat_patch_list(state, list);
4920 if (state->summary && state->apply_verbosity > verbosity_silent)
4921 summary_patch_list(list);
4923 if (flush_attributes)
4924 reset_parsed_attributes();
4925 end:
4926 free_patch_list(list);
4927 strbuf_release(&buf);
4928 string_list_clear(&state->fn_table, 0);
4929 return res;
4932 static int apply_option_parse_exclude(const struct option *opt,
4933 const char *arg, int unset)
4935 struct apply_state *state = opt->value;
4937 BUG_ON_OPT_NEG(unset);
4939 add_name_limit(state, arg, 1);
4940 return 0;
4943 static int apply_option_parse_include(const struct option *opt,
4944 const char *arg, int unset)
4946 struct apply_state *state = opt->value;
4948 BUG_ON_OPT_NEG(unset);
4950 add_name_limit(state, arg, 0);
4951 state->has_include = 1;
4952 return 0;
4955 static int apply_option_parse_p(const struct option *opt,
4956 const char *arg,
4957 int unset)
4959 struct apply_state *state = opt->value;
4961 BUG_ON_OPT_NEG(unset);
4963 state->p_value = atoi(arg);
4964 state->p_value_known = 1;
4965 return 0;
4968 static int apply_option_parse_space_change(const struct option *opt,
4969 const char *arg, int unset)
4971 struct apply_state *state = opt->value;
4973 BUG_ON_OPT_ARG(arg);
4975 if (unset)
4976 state->ws_ignore_action = ignore_ws_none;
4977 else
4978 state->ws_ignore_action = ignore_ws_change;
4979 return 0;
4982 static int apply_option_parse_whitespace(const struct option *opt,
4983 const char *arg, int unset)
4985 struct apply_state *state = opt->value;
4987 BUG_ON_OPT_NEG(unset);
4989 state->whitespace_option = arg;
4990 if (parse_whitespace_option(state, arg))
4991 return -1;
4992 return 0;
4995 static int apply_option_parse_directory(const struct option *opt,
4996 const char *arg, int unset)
4998 struct apply_state *state = opt->value;
5000 BUG_ON_OPT_NEG(unset);
5002 strbuf_reset(&state->root);
5003 strbuf_addstr(&state->root, arg);
5004 strbuf_complete(&state->root, '/');
5005 return 0;
5008 int apply_all_patches(struct apply_state *state,
5009 int argc,
5010 const char **argv,
5011 int options)
5013 int i;
5014 int res;
5015 int errs = 0;
5016 int read_stdin = 1;
5018 for (i = 0; i < argc; i++) {
5019 const char *arg = argv[i];
5020 char *to_free = NULL;
5021 int fd;
5023 if (!strcmp(arg, "-")) {
5024 res = apply_patch(state, 0, "<stdin>", options);
5025 if (res < 0)
5026 goto end;
5027 errs |= res;
5028 read_stdin = 0;
5029 continue;
5030 } else
5031 arg = to_free = prefix_filename(state->prefix, arg);
5033 fd = open(arg, O_RDONLY);
5034 if (fd < 0) {
5035 error(_("can't open patch '%s': %s"), arg, strerror(errno));
5036 res = -128;
5037 free(to_free);
5038 goto end;
5040 read_stdin = 0;
5041 set_default_whitespace_mode(state);
5042 res = apply_patch(state, fd, arg, options);
5043 close(fd);
5044 free(to_free);
5045 if (res < 0)
5046 goto end;
5047 errs |= res;
5049 set_default_whitespace_mode(state);
5050 if (read_stdin) {
5051 res = apply_patch(state, 0, "<stdin>", options);
5052 if (res < 0)
5053 goto end;
5054 errs |= res;
5057 if (state->whitespace_error) {
5058 if (state->squelch_whitespace_errors &&
5059 state->squelch_whitespace_errors < state->whitespace_error) {
5060 int squelched =
5061 state->whitespace_error - state->squelch_whitespace_errors;
5062 warning(Q_("squelched %d whitespace error",
5063 "squelched %d whitespace errors",
5064 squelched),
5065 squelched);
5067 if (state->ws_error_action == die_on_ws_error) {
5068 error(Q_("%d line adds whitespace errors.",
5069 "%d lines add whitespace errors.",
5070 state->whitespace_error),
5071 state->whitespace_error);
5072 res = -128;
5073 goto end;
5075 if (state->applied_after_fixing_ws && state->apply)
5076 warning(Q_("%d line applied after"
5077 " fixing whitespace errors.",
5078 "%d lines applied after"
5079 " fixing whitespace errors.",
5080 state->applied_after_fixing_ws),
5081 state->applied_after_fixing_ws);
5082 else if (state->whitespace_error)
5083 warning(Q_("%d line adds whitespace errors.",
5084 "%d lines add whitespace errors.",
5085 state->whitespace_error),
5086 state->whitespace_error);
5089 if (state->update_index) {
5090 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5091 if (res) {
5092 error(_("Unable to write new index file"));
5093 res = -128;
5094 goto end;
5098 res = !!errs;
5100 end:
5101 rollback_lock_file(&state->lock_file);
5103 if (state->apply_verbosity <= verbosity_silent) {
5104 set_error_routine(state->saved_error_routine);
5105 set_warn_routine(state->saved_warn_routine);
5108 if (res > -1)
5109 return res;
5110 return (res == -1 ? 1 : 128);
5113 int apply_parse_options(int argc, const char **argv,
5114 struct apply_state *state,
5115 int *force_apply, int *options,
5116 const char * const *apply_usage)
5118 struct option builtin_apply_options[] = {
5119 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5120 N_("don't apply changes matching the given path"),
5121 PARSE_OPT_NONEG, apply_option_parse_exclude),
5122 OPT_CALLBACK_F(0, "include", state, N_("path"),
5123 N_("apply changes matching the given path"),
5124 PARSE_OPT_NONEG, apply_option_parse_include),
5125 OPT_CALLBACK('p', NULL, state, N_("num"),
5126 N_("remove <num> leading slashes from traditional diff paths"),
5127 apply_option_parse_p),
5128 OPT_BOOL(0, "no-add", &state->no_add,
5129 N_("ignore additions made by the patch")),
5130 OPT_BOOL(0, "stat", &state->diffstat,
5131 N_("instead of applying the patch, output diffstat for the input")),
5132 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5133 OPT_NOOP_NOARG(0, "binary"),
5134 OPT_BOOL(0, "numstat", &state->numstat,
5135 N_("show number of added and deleted lines in decimal notation")),
5136 OPT_BOOL(0, "summary", &state->summary,
5137 N_("instead of applying the patch, output a summary for the input")),
5138 OPT_BOOL(0, "check", &state->check,
5139 N_("instead of applying the patch, see if the patch is applicable")),
5140 OPT_BOOL(0, "index", &state->check_index,
5141 N_("make sure the patch is applicable to the current index")),
5142 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5143 N_("mark new files with `git add --intent-to-add`")),
5144 OPT_BOOL(0, "cached", &state->cached,
5145 N_("apply a patch without touching the working tree")),
5146 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5147 N_("accept a patch that touches outside the working area"),
5148 PARSE_OPT_NOCOMPLETE),
5149 OPT_BOOL(0, "apply", force_apply,
5150 N_("also apply the patch (use with --stat/--summary/--check)")),
5151 OPT_BOOL('3', "3way", &state->threeway,
5152 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5153 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5154 N_("build a temporary index based on embedded index information")),
5155 /* Think twice before adding "--nul" synonym to this */
5156 OPT_SET_INT('z', NULL, &state->line_termination,
5157 N_("paths are separated with NUL character"), '\0'),
5158 OPT_INTEGER('C', NULL, &state->p_context,
5159 N_("ensure at least <n> lines of context match")),
5160 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5161 N_("detect new or modified lines that have whitespace errors"),
5162 apply_option_parse_whitespace),
5163 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5164 N_("ignore changes in whitespace when finding context"),
5165 PARSE_OPT_NOARG, apply_option_parse_space_change),
5166 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5167 N_("ignore changes in whitespace when finding context"),
5168 PARSE_OPT_NOARG, apply_option_parse_space_change),
5169 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5170 N_("apply the patch in reverse")),
5171 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5172 N_("don't expect at least one line of context")),
5173 OPT_BOOL(0, "reject", &state->apply_with_reject,
5174 N_("leave the rejected hunks in corresponding *.rej files")),
5175 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5176 N_("allow overlapping hunks")),
5177 OPT__VERBOSITY(&state->apply_verbosity),
5178 OPT_BIT(0, "inaccurate-eof", options,
5179 N_("tolerate incorrectly detected missing new-line at the end of file"),
5180 APPLY_OPT_INACCURATE_EOF),
5181 OPT_BIT(0, "recount", options,
5182 N_("do not trust the line counts in the hunk headers"),
5183 APPLY_OPT_RECOUNT),
5184 OPT_CALLBACK(0, "directory", state, N_("root"),
5185 N_("prepend <root> to all filenames"),
5186 apply_option_parse_directory),
5187 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5188 N_("don't return error for empty patches")),
5189 OPT_END()
5192 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);