name-hash.h: move declarations for name-hash.c from cache.h
[alt-git.git] / apply.c
blob3da33bb88fde0fc55695bb345fd29cb66898c6eb
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 #include "cache.h"
11 #include "abspath.h"
12 #include "alloc.h"
13 #include "base85.h"
14 #include "config.h"
15 #include "object-store.h"
16 #include "blob.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 "ll-merge.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 "quote.h"
31 #include "rerere.h"
32 #include "apply.h"
33 #include "entry.h"
34 #include "setup.h"
35 #include "symlinks.h"
36 #include "ws.h"
37 #include "wrapper.h"
39 struct gitdiff_data {
40 struct strbuf *root;
41 int linenr;
42 int p_value;
45 static void git_apply_config(void)
47 git_config_get_string("apply.whitespace", &apply_default_whitespace);
48 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
49 git_config(git_xmerge_config, NULL);
52 static int parse_whitespace_option(struct apply_state *state, const char *option)
54 if (!option) {
55 state->ws_error_action = warn_on_ws_error;
56 return 0;
58 if (!strcmp(option, "warn")) {
59 state->ws_error_action = warn_on_ws_error;
60 return 0;
62 if (!strcmp(option, "nowarn")) {
63 state->ws_error_action = nowarn_ws_error;
64 return 0;
66 if (!strcmp(option, "error")) {
67 state->ws_error_action = die_on_ws_error;
68 return 0;
70 if (!strcmp(option, "error-all")) {
71 state->ws_error_action = die_on_ws_error;
72 state->squelch_whitespace_errors = 0;
73 return 0;
75 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
76 state->ws_error_action = correct_ws_error;
77 return 0;
80 * Please update $__git_whitespacelist in git-completion.bash
81 * when you add new options.
83 return error(_("unrecognized whitespace option '%s'"), option);
86 static int parse_ignorewhitespace_option(struct apply_state *state,
87 const char *option)
89 if (!option || !strcmp(option, "no") ||
90 !strcmp(option, "false") || !strcmp(option, "never") ||
91 !strcmp(option, "none")) {
92 state->ws_ignore_action = ignore_ws_none;
93 return 0;
95 if (!strcmp(option, "change")) {
96 state->ws_ignore_action = ignore_ws_change;
97 return 0;
99 return error(_("unrecognized whitespace ignore option '%s'"), option);
102 int init_apply_state(struct apply_state *state,
103 struct repository *repo,
104 const char *prefix)
106 memset(state, 0, sizeof(*state));
107 state->prefix = prefix;
108 state->repo = repo;
109 state->apply = 1;
110 state->line_termination = '\n';
111 state->p_value = 1;
112 state->p_context = UINT_MAX;
113 state->squelch_whitespace_errors = 5;
114 state->ws_error_action = warn_on_ws_error;
115 state->ws_ignore_action = ignore_ws_none;
116 state->linenr = 1;
117 string_list_init_nodup(&state->fn_table);
118 string_list_init_nodup(&state->limit_by_name);
119 strset_init(&state->removed_symlinks);
120 strset_init(&state->kept_symlinks);
121 strbuf_init(&state->root, 0);
123 git_apply_config();
124 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
125 return -1;
126 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
127 return -1;
128 return 0;
131 void clear_apply_state(struct apply_state *state)
133 string_list_clear(&state->limit_by_name, 0);
134 strset_clear(&state->removed_symlinks);
135 strset_clear(&state->kept_symlinks);
136 strbuf_release(&state->root);
138 /* &state->fn_table is cleared at the end of apply_patch() */
141 static void mute_routine(const char *msg UNUSED, va_list params UNUSED)
143 /* do nothing */
146 int check_apply_state(struct apply_state *state, int force_apply)
148 int is_not_gitdir = !startup_info->have_repository;
150 if (state->apply_with_reject && state->threeway)
151 return error(_("options '%s' and '%s' cannot be used together"), "--reject", "--3way");
152 if (state->threeway) {
153 if (is_not_gitdir)
154 return error(_("'%s' outside a repository"), "--3way");
155 state->check_index = 1;
157 if (state->apply_with_reject) {
158 state->apply = 1;
159 if (state->apply_verbosity == verbosity_normal)
160 state->apply_verbosity = verbosity_verbose;
162 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
163 state->apply = 0;
164 if (state->check_index && is_not_gitdir)
165 return error(_("'%s' outside a repository"), "--index");
166 if (state->cached) {
167 if (is_not_gitdir)
168 return error(_("'%s' outside a repository"), "--cached");
169 state->check_index = 1;
171 if (state->ita_only && (state->check_index || is_not_gitdir))
172 state->ita_only = 0;
173 if (state->check_index)
174 state->unsafe_paths = 0;
176 if (state->apply_verbosity <= verbosity_silent) {
177 state->saved_error_routine = get_error_routine();
178 state->saved_warn_routine = get_warn_routine();
179 set_error_routine(mute_routine);
180 set_warn_routine(mute_routine);
183 return 0;
186 static void set_default_whitespace_mode(struct apply_state *state)
188 if (!state->whitespace_option && !apply_default_whitespace)
189 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
193 * This represents one "hunk" from a patch, starting with
194 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
195 * patch text is pointed at by patch, and its byte length
196 * is stored in size. leading and trailing are the number
197 * of context lines.
199 struct fragment {
200 unsigned long leading, trailing;
201 unsigned long oldpos, oldlines;
202 unsigned long newpos, newlines;
204 * 'patch' is usually borrowed from buf in apply_patch(),
205 * but some codepaths store an allocated buffer.
207 const char *patch;
208 unsigned free_patch:1,
209 rejected:1;
210 int size;
211 int linenr;
212 struct fragment *next;
216 * When dealing with a binary patch, we reuse "leading" field
217 * to store the type of the binary hunk, either deflated "delta"
218 * or deflated "literal".
220 #define binary_patch_method leading
221 #define BINARY_DELTA_DEFLATED 1
222 #define BINARY_LITERAL_DEFLATED 2
224 static void free_fragment_list(struct fragment *list)
226 while (list) {
227 struct fragment *next = list->next;
228 if (list->free_patch)
229 free((char *)list->patch);
230 free(list);
231 list = next;
235 void release_patch(struct patch *patch)
237 free_fragment_list(patch->fragments);
238 free(patch->def_name);
239 free(patch->old_name);
240 free(patch->new_name);
241 free(patch->result);
244 static void free_patch(struct patch *patch)
246 release_patch(patch);
247 free(patch);
250 static void free_patch_list(struct patch *list)
252 while (list) {
253 struct patch *next = list->next;
254 free_patch(list);
255 list = next;
260 * A line in a file, len-bytes long (includes the terminating LF,
261 * except for an incomplete line at the end if the file ends with
262 * one), and its contents hashes to 'hash'.
264 struct line {
265 size_t len;
266 unsigned hash : 24;
267 unsigned flag : 8;
268 #define LINE_COMMON 1
269 #define LINE_PATCHED 2
273 * This represents a "file", which is an array of "lines".
275 struct image {
276 char *buf;
277 size_t len;
278 size_t nr;
279 size_t alloc;
280 struct line *line_allocated;
281 struct line *line;
284 static uint32_t hash_line(const char *cp, size_t len)
286 size_t i;
287 uint32_t h;
288 for (i = 0, h = 0; i < len; i++) {
289 if (!isspace(cp[i])) {
290 h = h * 3 + (cp[i] & 0xff);
293 return h;
297 * Compare lines s1 of length n1 and s2 of length n2, ignoring
298 * whitespace difference. Returns 1 if they match, 0 otherwise
300 static int fuzzy_matchlines(const char *s1, size_t n1,
301 const char *s2, size_t n2)
303 const char *end1 = s1 + n1;
304 const char *end2 = s2 + n2;
306 /* ignore line endings */
307 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
308 end1--;
309 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
310 end2--;
312 while (s1 < end1 && s2 < end2) {
313 if (isspace(*s1)) {
315 * Skip whitespace. We check on both buffers
316 * because we don't want "a b" to match "ab".
318 if (!isspace(*s2))
319 return 0;
320 while (s1 < end1 && isspace(*s1))
321 s1++;
322 while (s2 < end2 && isspace(*s2))
323 s2++;
324 } else if (*s1++ != *s2++)
325 return 0;
328 /* If we reached the end on one side only, lines don't match. */
329 return s1 == end1 && s2 == end2;
332 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
334 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
335 img->line_allocated[img->nr].len = len;
336 img->line_allocated[img->nr].hash = hash_line(bol, len);
337 img->line_allocated[img->nr].flag = flag;
338 img->nr++;
342 * "buf" has the file contents to be patched (read from various sources).
343 * attach it to "image" and add line-based index to it.
344 * "image" now owns the "buf".
346 static void prepare_image(struct image *image, char *buf, size_t len,
347 int prepare_linetable)
349 const char *cp, *ep;
351 memset(image, 0, sizeof(*image));
352 image->buf = buf;
353 image->len = len;
355 if (!prepare_linetable)
356 return;
358 ep = image->buf + image->len;
359 cp = image->buf;
360 while (cp < ep) {
361 const char *next;
362 for (next = cp; next < ep && *next != '\n'; next++)
364 if (next < ep)
365 next++;
366 add_line_info(image, cp, next - cp, 0);
367 cp = next;
369 image->line = image->line_allocated;
372 static void clear_image(struct image *image)
374 free(image->buf);
375 free(image->line_allocated);
376 memset(image, 0, sizeof(*image));
379 /* fmt must contain _one_ %s and no other substitution */
380 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
382 struct strbuf sb = STRBUF_INIT;
384 if (patch->old_name && patch->new_name &&
385 strcmp(patch->old_name, patch->new_name)) {
386 quote_c_style(patch->old_name, &sb, NULL, 0);
387 strbuf_addstr(&sb, " => ");
388 quote_c_style(patch->new_name, &sb, NULL, 0);
389 } else {
390 const char *n = patch->new_name;
391 if (!n)
392 n = patch->old_name;
393 quote_c_style(n, &sb, NULL, 0);
395 fprintf(output, fmt, sb.buf);
396 fputc('\n', output);
397 strbuf_release(&sb);
400 #define SLOP (16)
403 * apply.c isn't equipped to handle arbitrarily large patches, because
404 * it intermingles `unsigned long` with `int` for the type used to store
405 * buffer lengths.
407 * Only process patches that are just shy of 1 GiB large in order to
408 * avoid any truncation or overflow issues.
410 #define MAX_APPLY_SIZE (1024UL * 1024 * 1023)
412 static int read_patch_file(struct strbuf *sb, int fd)
414 if (strbuf_read(sb, fd, 0) < 0 || sb->len >= MAX_APPLY_SIZE)
415 return error_errno("git apply: failed to read");
418 * Make sure that we have some slop in the buffer
419 * so that we can do speculative "memcmp" etc, and
420 * see to it that it is NUL-filled.
422 strbuf_grow(sb, SLOP);
423 memset(sb->buf + sb->len, 0, SLOP);
424 return 0;
427 static unsigned long linelen(const char *buffer, unsigned long size)
429 unsigned long len = 0;
430 while (size--) {
431 len++;
432 if (*buffer++ == '\n')
433 break;
435 return len;
438 static int is_dev_null(const char *str)
440 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
443 #define TERM_SPACE 1
444 #define TERM_TAB 2
446 static int name_terminate(int c, int terminate)
448 if (c == ' ' && !(terminate & TERM_SPACE))
449 return 0;
450 if (c == '\t' && !(terminate & TERM_TAB))
451 return 0;
453 return 1;
456 /* remove double slashes to make --index work with such filenames */
457 static char *squash_slash(char *name)
459 int i = 0, j = 0;
461 if (!name)
462 return NULL;
464 while (name[i]) {
465 if ((name[j++] = name[i++]) == '/')
466 while (name[i] == '/')
467 i++;
469 name[j] = '\0';
470 return name;
473 static char *find_name_gnu(struct strbuf *root,
474 const char *line,
475 int p_value)
477 struct strbuf name = STRBUF_INIT;
478 char *cp;
481 * Proposed "new-style" GNU patch/diff format; see
482 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
484 if (unquote_c_style(&name, line, NULL)) {
485 strbuf_release(&name);
486 return NULL;
489 for (cp = name.buf; p_value; p_value--) {
490 cp = strchr(cp, '/');
491 if (!cp) {
492 strbuf_release(&name);
493 return NULL;
495 cp++;
498 strbuf_remove(&name, 0, cp - name.buf);
499 if (root->len)
500 strbuf_insert(&name, 0, root->buf, root->len);
501 return squash_slash(strbuf_detach(&name, NULL));
504 static size_t sane_tz_len(const char *line, size_t len)
506 const char *tz, *p;
508 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
509 return 0;
510 tz = line + len - strlen(" +0500");
512 if (tz[1] != '+' && tz[1] != '-')
513 return 0;
515 for (p = tz + 2; p != line + len; p++)
516 if (!isdigit(*p))
517 return 0;
519 return line + len - tz;
522 static size_t tz_with_colon_len(const char *line, size_t len)
524 const char *tz, *p;
526 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
527 return 0;
528 tz = line + len - strlen(" +08:00");
530 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
531 return 0;
532 p = tz + 2;
533 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
534 !isdigit(*p++) || !isdigit(*p++))
535 return 0;
537 return line + len - tz;
540 static size_t date_len(const char *line, size_t len)
542 const char *date, *p;
544 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
545 return 0;
546 p = date = line + len - strlen("72-02-05");
548 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
549 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
550 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
551 return 0;
553 if (date - line >= strlen("19") &&
554 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
555 date -= strlen("19");
557 return line + len - date;
560 static size_t short_time_len(const char *line, size_t len)
562 const char *time, *p;
564 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
565 return 0;
566 p = time = line + len - strlen(" 07:01:32");
568 /* Permit 1-digit hours? */
569 if (*p++ != ' ' ||
570 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
571 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
572 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
573 return 0;
575 return line + len - time;
578 static size_t fractional_time_len(const char *line, size_t len)
580 const char *p;
581 size_t n;
583 /* Expected format: 19:41:17.620000023 */
584 if (!len || !isdigit(line[len - 1]))
585 return 0;
586 p = line + len - 1;
588 /* Fractional seconds. */
589 while (p > line && isdigit(*p))
590 p--;
591 if (*p != '.')
592 return 0;
594 /* Hours, minutes, and whole seconds. */
595 n = short_time_len(line, p - line);
596 if (!n)
597 return 0;
599 return line + len - p + n;
602 static size_t trailing_spaces_len(const char *line, size_t len)
604 const char *p;
606 /* Expected format: ' ' x (1 or more) */
607 if (!len || line[len - 1] != ' ')
608 return 0;
610 p = line + len;
611 while (p != line) {
612 p--;
613 if (*p != ' ')
614 return line + len - (p + 1);
617 /* All spaces! */
618 return len;
621 static size_t diff_timestamp_len(const char *line, size_t len)
623 const char *end = line + len;
624 size_t n;
627 * Posix: 2010-07-05 19:41:17
628 * GNU: 2010-07-05 19:41:17.620000023 -0500
631 if (!isdigit(end[-1]))
632 return 0;
634 n = sane_tz_len(line, end - line);
635 if (!n)
636 n = tz_with_colon_len(line, end - line);
637 end -= n;
639 n = short_time_len(line, end - line);
640 if (!n)
641 n = fractional_time_len(line, end - line);
642 end -= n;
644 n = date_len(line, end - line);
645 if (!n) /* No date. Too bad. */
646 return 0;
647 end -= n;
649 if (end == line) /* No space before date. */
650 return 0;
651 if (end[-1] == '\t') { /* Success! */
652 end--;
653 return line + len - end;
655 if (end[-1] != ' ') /* No space before date. */
656 return 0;
658 /* Whitespace damage. */
659 end -= trailing_spaces_len(line, end - line);
660 return line + len - end;
663 static char *find_name_common(struct strbuf *root,
664 const char *line,
665 const char *def,
666 int p_value,
667 const char *end,
668 int terminate)
670 int len;
671 const char *start = NULL;
673 if (p_value == 0)
674 start = line;
675 while (line != end) {
676 char c = *line;
678 if (!end && isspace(c)) {
679 if (c == '\n')
680 break;
681 if (name_terminate(c, terminate))
682 break;
684 line++;
685 if (c == '/' && !--p_value)
686 start = line;
688 if (!start)
689 return squash_slash(xstrdup_or_null(def));
690 len = line - start;
691 if (!len)
692 return squash_slash(xstrdup_or_null(def));
695 * Generally we prefer the shorter name, especially
696 * if the other one is just a variation of that with
697 * something else tacked on to the end (ie "file.orig"
698 * or "file~").
700 if (def) {
701 int deflen = strlen(def);
702 if (deflen < len && !strncmp(start, def, deflen))
703 return squash_slash(xstrdup(def));
706 if (root->len) {
707 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
708 return squash_slash(ret);
711 return squash_slash(xmemdupz(start, len));
714 static char *find_name(struct strbuf *root,
715 const char *line,
716 char *def,
717 int p_value,
718 int terminate)
720 if (*line == '"') {
721 char *name = find_name_gnu(root, line, p_value);
722 if (name)
723 return name;
726 return find_name_common(root, line, def, p_value, NULL, terminate);
729 static char *find_name_traditional(struct strbuf *root,
730 const char *line,
731 char *def,
732 int p_value)
734 size_t len;
735 size_t date_len;
737 if (*line == '"') {
738 char *name = find_name_gnu(root, line, p_value);
739 if (name)
740 return name;
743 len = strchrnul(line, '\n') - line;
744 date_len = diff_timestamp_len(line, len);
745 if (!date_len)
746 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
747 len -= date_len;
749 return find_name_common(root, line, def, p_value, line + len, 0);
753 * Given the string after "--- " or "+++ ", guess the appropriate
754 * p_value for the given patch.
756 static int guess_p_value(struct apply_state *state, const char *nameline)
758 char *name, *cp;
759 int val = -1;
761 if (is_dev_null(nameline))
762 return -1;
763 name = find_name_traditional(&state->root, nameline, NULL, 0);
764 if (!name)
765 return -1;
766 cp = strchr(name, '/');
767 if (!cp)
768 val = 0;
769 else if (state->prefix) {
771 * Does it begin with "a/$our-prefix" and such? Then this is
772 * very likely to apply to our directory.
774 if (starts_with(name, state->prefix))
775 val = count_slashes(state->prefix);
776 else {
777 cp++;
778 if (starts_with(cp, state->prefix))
779 val = count_slashes(state->prefix) + 1;
782 free(name);
783 return val;
787 * Does the ---/+++ line have the POSIX timestamp after the last HT?
788 * GNU diff puts epoch there to signal a creation/deletion event. Is
789 * this such a timestamp?
791 static int has_epoch_timestamp(const char *nameline)
794 * We are only interested in epoch timestamp; any non-zero
795 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
796 * For the same reason, the date must be either 1969-12-31 or
797 * 1970-01-01, and the seconds part must be "00".
799 const char stamp_regexp[] =
800 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
802 "([-+][0-2][0-9]:?[0-5][0-9])\n";
803 const char *timestamp = NULL, *cp, *colon;
804 static regex_t *stamp;
805 regmatch_t m[10];
806 int zoneoffset, epoch_hour, hour, minute;
807 int status;
809 for (cp = nameline; *cp != '\n'; cp++) {
810 if (*cp == '\t')
811 timestamp = cp + 1;
813 if (!timestamp)
814 return 0;
817 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
818 * (west of GMT) or 1970-01-01 (east of GMT)
820 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
821 epoch_hour = 24;
822 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
823 epoch_hour = 0;
824 else
825 return 0;
827 if (!stamp) {
828 stamp = xmalloc(sizeof(*stamp));
829 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
830 warning(_("Cannot prepare timestamp regexp %s"),
831 stamp_regexp);
832 return 0;
836 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
837 if (status) {
838 if (status != REG_NOMATCH)
839 warning(_("regexec returned %d for input: %s"),
840 status, timestamp);
841 return 0;
844 hour = strtol(timestamp, NULL, 10);
845 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
847 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
848 if (*colon == ':')
849 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
850 else
851 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
852 if (timestamp[m[3].rm_so] == '-')
853 zoneoffset = -zoneoffset;
855 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
859 * Get the name etc info from the ---/+++ lines of a traditional patch header
861 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
862 * files, we can happily check the index for a match, but for creating a
863 * new file we should try to match whatever "patch" does. I have no idea.
865 static int parse_traditional_patch(struct apply_state *state,
866 const char *first,
867 const char *second,
868 struct patch *patch)
870 char *name;
872 first += 4; /* skip "--- " */
873 second += 4; /* skip "+++ " */
874 if (!state->p_value_known) {
875 int p, q;
876 p = guess_p_value(state, first);
877 q = guess_p_value(state, second);
878 if (p < 0) p = q;
879 if (0 <= p && p == q) {
880 state->p_value = p;
881 state->p_value_known = 1;
884 if (is_dev_null(first)) {
885 patch->is_new = 1;
886 patch->is_delete = 0;
887 name = find_name_traditional(&state->root, second, NULL, state->p_value);
888 patch->new_name = name;
889 } else if (is_dev_null(second)) {
890 patch->is_new = 0;
891 patch->is_delete = 1;
892 name = find_name_traditional(&state->root, first, NULL, state->p_value);
893 patch->old_name = name;
894 } else {
895 char *first_name;
896 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
897 name = find_name_traditional(&state->root, second, first_name, state->p_value);
898 free(first_name);
899 if (has_epoch_timestamp(first)) {
900 patch->is_new = 1;
901 patch->is_delete = 0;
902 patch->new_name = name;
903 } else if (has_epoch_timestamp(second)) {
904 patch->is_new = 0;
905 patch->is_delete = 1;
906 patch->old_name = name;
907 } else {
908 patch->old_name = name;
909 patch->new_name = xstrdup_or_null(name);
912 if (!name)
913 return error(_("unable to find filename in patch at line %d"), state->linenr);
915 return 0;
918 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
919 const char *line UNUSED,
920 struct patch *patch UNUSED)
922 return 1;
926 * We're anal about diff header consistency, to make
927 * sure that we don't end up having strange ambiguous
928 * patches floating around.
930 * As a result, gitdiff_{old|new}name() will check
931 * their names against any previous information, just
932 * to make sure..
934 #define DIFF_OLD_NAME 0
935 #define DIFF_NEW_NAME 1
937 static int gitdiff_verify_name(struct gitdiff_data *state,
938 const char *line,
939 int isnull,
940 char **name,
941 int side)
943 if (!*name && !isnull) {
944 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
945 return 0;
948 if (*name) {
949 char *another;
950 if (isnull)
951 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
952 *name, state->linenr);
953 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
954 if (!another || strcmp(another, *name)) {
955 free(another);
956 return error((side == DIFF_NEW_NAME) ?
957 _("git apply: bad git-diff - inconsistent new filename on line %d") :
958 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
960 free(another);
961 } else {
962 if (!is_dev_null(line))
963 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
966 return 0;
969 static int gitdiff_oldname(struct gitdiff_data *state,
970 const char *line,
971 struct patch *patch)
973 return gitdiff_verify_name(state, line,
974 patch->is_new, &patch->old_name,
975 DIFF_OLD_NAME);
978 static int gitdiff_newname(struct gitdiff_data *state,
979 const char *line,
980 struct patch *patch)
982 return gitdiff_verify_name(state, line,
983 patch->is_delete, &patch->new_name,
984 DIFF_NEW_NAME);
987 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
989 char *end;
990 *mode = strtoul(line, &end, 8);
991 if (end == line || !isspace(*end))
992 return error(_("invalid mode on line %d: %s"), linenr, line);
993 return 0;
996 static int gitdiff_oldmode(struct gitdiff_data *state,
997 const char *line,
998 struct patch *patch)
1000 return parse_mode_line(line, state->linenr, &patch->old_mode);
1003 static int gitdiff_newmode(struct gitdiff_data *state,
1004 const char *line,
1005 struct patch *patch)
1007 return parse_mode_line(line, state->linenr, &patch->new_mode);
1010 static int gitdiff_delete(struct gitdiff_data *state,
1011 const char *line,
1012 struct patch *patch)
1014 patch->is_delete = 1;
1015 free(patch->old_name);
1016 patch->old_name = xstrdup_or_null(patch->def_name);
1017 return gitdiff_oldmode(state, line, patch);
1020 static int gitdiff_newfile(struct gitdiff_data *state,
1021 const char *line,
1022 struct patch *patch)
1024 patch->is_new = 1;
1025 free(patch->new_name);
1026 patch->new_name = xstrdup_or_null(patch->def_name);
1027 return gitdiff_newmode(state, line, patch);
1030 static int gitdiff_copysrc(struct gitdiff_data *state,
1031 const char *line,
1032 struct patch *patch)
1034 patch->is_copy = 1;
1035 free(patch->old_name);
1036 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1037 return 0;
1040 static int gitdiff_copydst(struct gitdiff_data *state,
1041 const char *line,
1042 struct patch *patch)
1044 patch->is_copy = 1;
1045 free(patch->new_name);
1046 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1047 return 0;
1050 static int gitdiff_renamesrc(struct gitdiff_data *state,
1051 const char *line,
1052 struct patch *patch)
1054 patch->is_rename = 1;
1055 free(patch->old_name);
1056 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1057 return 0;
1060 static int gitdiff_renamedst(struct gitdiff_data *state,
1061 const char *line,
1062 struct patch *patch)
1064 patch->is_rename = 1;
1065 free(patch->new_name);
1066 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1067 return 0;
1070 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1071 const char *line,
1072 struct patch *patch)
1074 unsigned long val = strtoul(line, NULL, 10);
1075 if (val <= 100)
1076 patch->score = val;
1077 return 0;
1080 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1081 const char *line,
1082 struct patch *patch)
1084 unsigned long val = strtoul(line, NULL, 10);
1085 if (val <= 100)
1086 patch->score = val;
1087 return 0;
1090 static int gitdiff_index(struct gitdiff_data *state,
1091 const char *line,
1092 struct patch *patch)
1095 * index line is N hexadecimal, "..", N hexadecimal,
1096 * and optional space with octal mode.
1098 const char *ptr, *eol;
1099 int len;
1100 const unsigned hexsz = the_hash_algo->hexsz;
1102 ptr = strchr(line, '.');
1103 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1104 return 0;
1105 len = ptr - line;
1106 memcpy(patch->old_oid_prefix, line, len);
1107 patch->old_oid_prefix[len] = 0;
1109 line = ptr + 2;
1110 ptr = strchr(line, ' ');
1111 eol = strchrnul(line, '\n');
1113 if (!ptr || eol < ptr)
1114 ptr = eol;
1115 len = ptr - line;
1117 if (hexsz < len)
1118 return 0;
1119 memcpy(patch->new_oid_prefix, line, len);
1120 patch->new_oid_prefix[len] = 0;
1121 if (*ptr == ' ')
1122 return gitdiff_oldmode(state, ptr + 1, patch);
1123 return 0;
1127 * This is normal for a diff that doesn't change anything: we'll fall through
1128 * into the next diff. Tell the parser to break out.
1130 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1131 const char *line UNUSED,
1132 struct patch *patch UNUSED)
1134 return 1;
1138 * Skip p_value leading components from "line"; as we do not accept
1139 * absolute paths, return NULL in that case.
1141 static const char *skip_tree_prefix(int p_value,
1142 const char *line,
1143 int llen)
1145 int nslash;
1146 int i;
1148 if (!p_value)
1149 return (llen && line[0] == '/') ? NULL : line;
1151 nslash = p_value;
1152 for (i = 0; i < llen; i++) {
1153 int ch = line[i];
1154 if (ch == '/' && --nslash <= 0)
1155 return (i == 0) ? NULL : &line[i + 1];
1157 return NULL;
1161 * This is to extract the same name that appears on "diff --git"
1162 * line. We do not find and return anything if it is a rename
1163 * patch, and it is OK because we will find the name elsewhere.
1164 * We need to reliably find name only when it is mode-change only,
1165 * creation or deletion of an empty file. In any of these cases,
1166 * both sides are the same name under a/ and b/ respectively.
1168 static char *git_header_name(int p_value,
1169 const char *line,
1170 int llen)
1172 const char *name;
1173 const char *second = NULL;
1174 size_t len, line_len;
1176 line += strlen("diff --git ");
1177 llen -= strlen("diff --git ");
1179 if (*line == '"') {
1180 const char *cp;
1181 struct strbuf first = STRBUF_INIT;
1182 struct strbuf sp = STRBUF_INIT;
1184 if (unquote_c_style(&first, line, &second))
1185 goto free_and_fail1;
1187 /* strip the a/b prefix including trailing slash */
1188 cp = skip_tree_prefix(p_value, first.buf, first.len);
1189 if (!cp)
1190 goto free_and_fail1;
1191 strbuf_remove(&first, 0, cp - first.buf);
1194 * second points at one past closing dq of name.
1195 * find the second name.
1197 while ((second < line + llen) && isspace(*second))
1198 second++;
1200 if (line + llen <= second)
1201 goto free_and_fail1;
1202 if (*second == '"') {
1203 if (unquote_c_style(&sp, second, NULL))
1204 goto free_and_fail1;
1205 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1206 if (!cp)
1207 goto free_and_fail1;
1208 /* They must match, otherwise ignore */
1209 if (strcmp(cp, first.buf))
1210 goto free_and_fail1;
1211 strbuf_release(&sp);
1212 return strbuf_detach(&first, NULL);
1215 /* unquoted second */
1216 cp = skip_tree_prefix(p_value, second, line + llen - second);
1217 if (!cp)
1218 goto free_and_fail1;
1219 if (line + llen - cp != first.len ||
1220 memcmp(first.buf, cp, first.len))
1221 goto free_and_fail1;
1222 return strbuf_detach(&first, NULL);
1224 free_and_fail1:
1225 strbuf_release(&first);
1226 strbuf_release(&sp);
1227 return NULL;
1230 /* unquoted first name */
1231 name = skip_tree_prefix(p_value, line, llen);
1232 if (!name)
1233 return NULL;
1236 * since the first name is unquoted, a dq if exists must be
1237 * the beginning of the second name.
1239 for (second = name; second < line + llen; second++) {
1240 if (*second == '"') {
1241 struct strbuf sp = STRBUF_INIT;
1242 const char *np;
1244 if (unquote_c_style(&sp, second, NULL))
1245 goto free_and_fail2;
1247 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1248 if (!np)
1249 goto free_and_fail2;
1251 len = sp.buf + sp.len - np;
1252 if (len < second - name &&
1253 !strncmp(np, name, len) &&
1254 isspace(name[len])) {
1255 /* Good */
1256 strbuf_remove(&sp, 0, np - sp.buf);
1257 return strbuf_detach(&sp, NULL);
1260 free_and_fail2:
1261 strbuf_release(&sp);
1262 return NULL;
1267 * Accept a name only if it shows up twice, exactly the same
1268 * form.
1270 second = strchr(name, '\n');
1271 if (!second)
1272 return NULL;
1273 line_len = second - name;
1274 for (len = 0 ; ; len++) {
1275 switch (name[len]) {
1276 default:
1277 continue;
1278 case '\n':
1279 return NULL;
1280 case '\t': case ' ':
1282 * Is this the separator between the preimage
1283 * and the postimage pathname? Again, we are
1284 * only interested in the case where there is
1285 * no rename, as this is only to set def_name
1286 * and a rename patch has the names elsewhere
1287 * in an unambiguous form.
1289 if (!name[len + 1])
1290 return NULL; /* no postimage name */
1291 second = skip_tree_prefix(p_value, name + len + 1,
1292 line_len - (len + 1));
1293 if (!second)
1294 return NULL;
1296 * Does len bytes starting at "name" and "second"
1297 * (that are separated by one HT or SP we just
1298 * found) exactly match?
1300 if (second[len] == '\n' && !strncmp(name, second, len))
1301 return xmemdupz(name, len);
1306 static int check_header_line(int linenr, struct patch *patch)
1308 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1309 (patch->is_rename == 1) + (patch->is_copy == 1);
1310 if (extensions > 1)
1311 return error(_("inconsistent header lines %d and %d"),
1312 patch->extension_linenr, linenr);
1313 if (extensions && !patch->extension_linenr)
1314 patch->extension_linenr = linenr;
1315 return 0;
1318 int parse_git_diff_header(struct strbuf *root,
1319 int *linenr,
1320 int p_value,
1321 const char *line,
1322 int len,
1323 unsigned int size,
1324 struct patch *patch)
1326 unsigned long offset;
1327 struct gitdiff_data parse_hdr_state;
1329 /* A git diff has explicit new/delete information, so we don't guess */
1330 patch->is_new = 0;
1331 patch->is_delete = 0;
1334 * Some things may not have the old name in the
1335 * rest of the headers anywhere (pure mode changes,
1336 * or removing or adding empty files), so we get
1337 * the default name from the header.
1339 patch->def_name = git_header_name(p_value, line, len);
1340 if (patch->def_name && root->len) {
1341 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1342 free(patch->def_name);
1343 patch->def_name = s;
1346 line += len;
1347 size -= len;
1348 (*linenr)++;
1349 parse_hdr_state.root = root;
1350 parse_hdr_state.linenr = *linenr;
1351 parse_hdr_state.p_value = p_value;
1353 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1354 static const struct opentry {
1355 const char *str;
1356 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1357 } optable[] = {
1358 { "@@ -", gitdiff_hdrend },
1359 { "--- ", gitdiff_oldname },
1360 { "+++ ", gitdiff_newname },
1361 { "old mode ", gitdiff_oldmode },
1362 { "new mode ", gitdiff_newmode },
1363 { "deleted file mode ", gitdiff_delete },
1364 { "new file mode ", gitdiff_newfile },
1365 { "copy from ", gitdiff_copysrc },
1366 { "copy to ", gitdiff_copydst },
1367 { "rename old ", gitdiff_renamesrc },
1368 { "rename new ", gitdiff_renamedst },
1369 { "rename from ", gitdiff_renamesrc },
1370 { "rename to ", gitdiff_renamedst },
1371 { "similarity index ", gitdiff_similarity },
1372 { "dissimilarity index ", gitdiff_dissimilarity },
1373 { "index ", gitdiff_index },
1374 { "", gitdiff_unrecognized },
1376 int i;
1378 len = linelen(line, size);
1379 if (!len || line[len-1] != '\n')
1380 break;
1381 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1382 const struct opentry *p = optable + i;
1383 int oplen = strlen(p->str);
1384 int res;
1385 if (len < oplen || memcmp(p->str, line, oplen))
1386 continue;
1387 res = p->fn(&parse_hdr_state, line + oplen, patch);
1388 if (res < 0)
1389 return -1;
1390 if (check_header_line(*linenr, patch))
1391 return -1;
1392 if (res > 0)
1393 goto done;
1394 break;
1398 done:
1399 if (!patch->old_name && !patch->new_name) {
1400 if (!patch->def_name) {
1401 error(Q_("git diff header lacks filename information when removing "
1402 "%d leading pathname component (line %d)",
1403 "git diff header lacks filename information when removing "
1404 "%d leading pathname components (line %d)",
1405 parse_hdr_state.p_value),
1406 parse_hdr_state.p_value, *linenr);
1407 return -128;
1409 patch->old_name = xstrdup(patch->def_name);
1410 patch->new_name = xstrdup(patch->def_name);
1412 if ((!patch->new_name && !patch->is_delete) ||
1413 (!patch->old_name && !patch->is_new)) {
1414 error(_("git diff header lacks filename information "
1415 "(line %d)"), *linenr);
1416 return -128;
1418 patch->is_toplevel_relative = 1;
1419 return offset;
1422 static int parse_num(const char *line, unsigned long *p)
1424 char *ptr;
1426 if (!isdigit(*line))
1427 return 0;
1428 *p = strtoul(line, &ptr, 10);
1429 return ptr - line;
1432 static int parse_range(const char *line, int len, int offset, const char *expect,
1433 unsigned long *p1, unsigned long *p2)
1435 int digits, ex;
1437 if (offset < 0 || offset >= len)
1438 return -1;
1439 line += offset;
1440 len -= offset;
1442 digits = parse_num(line, p1);
1443 if (!digits)
1444 return -1;
1446 offset += digits;
1447 line += digits;
1448 len -= digits;
1450 *p2 = 1;
1451 if (*line == ',') {
1452 digits = parse_num(line+1, p2);
1453 if (!digits)
1454 return -1;
1456 offset += digits+1;
1457 line += digits+1;
1458 len -= digits+1;
1461 ex = strlen(expect);
1462 if (ex > len)
1463 return -1;
1464 if (memcmp(line, expect, ex))
1465 return -1;
1467 return offset + ex;
1470 static void recount_diff(const char *line, int size, struct fragment *fragment)
1472 int oldlines = 0, newlines = 0, ret = 0;
1474 if (size < 1) {
1475 warning("recount: ignore empty hunk");
1476 return;
1479 for (;;) {
1480 int len = linelen(line, size);
1481 size -= len;
1482 line += len;
1484 if (size < 1)
1485 break;
1487 switch (*line) {
1488 case ' ': case '\n':
1489 newlines++;
1490 /* fall through */
1491 case '-':
1492 oldlines++;
1493 continue;
1494 case '+':
1495 newlines++;
1496 continue;
1497 case '\\':
1498 continue;
1499 case '@':
1500 ret = size < 3 || !starts_with(line, "@@ ");
1501 break;
1502 case 'd':
1503 ret = size < 5 || !starts_with(line, "diff ");
1504 break;
1505 default:
1506 ret = -1;
1507 break;
1509 if (ret) {
1510 warning(_("recount: unexpected line: %.*s"),
1511 (int)linelen(line, size), line);
1512 return;
1514 break;
1516 fragment->oldlines = oldlines;
1517 fragment->newlines = newlines;
1521 * Parse a unified diff fragment header of the
1522 * form "@@ -a,b +c,d @@"
1524 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1526 int offset;
1528 if (!len || line[len-1] != '\n')
1529 return -1;
1531 /* Figure out the number of lines in a fragment */
1532 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1533 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1535 return offset;
1539 * Find file diff header
1541 * Returns:
1542 * -1 if no header was found
1543 * -128 in case of error
1544 * the size of the header in bytes (called "offset") otherwise
1546 static int find_header(struct apply_state *state,
1547 const char *line,
1548 unsigned long size,
1549 int *hdrsize,
1550 struct patch *patch)
1552 unsigned long offset, len;
1554 patch->is_toplevel_relative = 0;
1555 patch->is_rename = patch->is_copy = 0;
1556 patch->is_new = patch->is_delete = -1;
1557 patch->old_mode = patch->new_mode = 0;
1558 patch->old_name = patch->new_name = NULL;
1559 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1560 unsigned long nextlen;
1562 len = linelen(line, size);
1563 if (!len)
1564 break;
1566 /* Testing this early allows us to take a few shortcuts.. */
1567 if (len < 6)
1568 continue;
1571 * Make sure we don't find any unconnected patch fragments.
1572 * That's a sign that we didn't find a header, and that a
1573 * patch has become corrupted/broken up.
1575 if (!memcmp("@@ -", line, 4)) {
1576 struct fragment dummy;
1577 if (parse_fragment_header(line, len, &dummy) < 0)
1578 continue;
1579 error(_("patch fragment without header at line %d: %.*s"),
1580 state->linenr, (int)len-1, line);
1581 return -128;
1584 if (size < len + 6)
1585 break;
1588 * Git patch? It might not have a real patch, just a rename
1589 * or mode change, so we handle that specially
1591 if (!memcmp("diff --git ", line, 11)) {
1592 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1593 state->p_value, line, len,
1594 size, patch);
1595 if (git_hdr_len < 0)
1596 return -128;
1597 if (git_hdr_len <= len)
1598 continue;
1599 *hdrsize = git_hdr_len;
1600 return offset;
1603 /* --- followed by +++ ? */
1604 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1605 continue;
1608 * We only accept unified patches, so we want it to
1609 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1610 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1612 nextlen = linelen(line + len, size - len);
1613 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1614 continue;
1616 /* Ok, we'll consider it a patch */
1617 if (parse_traditional_patch(state, line, line+len, patch))
1618 return -128;
1619 *hdrsize = len + nextlen;
1620 state->linenr += 2;
1621 return offset;
1623 return -1;
1626 static void record_ws_error(struct apply_state *state,
1627 unsigned result,
1628 const char *line,
1629 int len,
1630 int linenr)
1632 char *err;
1634 if (!result)
1635 return;
1637 state->whitespace_error++;
1638 if (state->squelch_whitespace_errors &&
1639 state->squelch_whitespace_errors < state->whitespace_error)
1640 return;
1642 err = whitespace_error_string(result);
1643 if (state->apply_verbosity > verbosity_silent)
1644 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1645 state->patch_input_file, linenr, err, len, line);
1646 free(err);
1649 static void check_whitespace(struct apply_state *state,
1650 const char *line,
1651 int len,
1652 unsigned ws_rule)
1654 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1656 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1660 * Check if the patch has context lines with CRLF or
1661 * the patch wants to remove lines with CRLF.
1663 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1665 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1666 patch->ws_rule |= WS_CR_AT_EOL;
1667 patch->crlf_in_old = 1;
1673 * Parse a unified diff. Note that this really needs to parse each
1674 * fragment separately, since the only way to know the difference
1675 * between a "---" that is part of a patch, and a "---" that starts
1676 * the next patch is to look at the line counts..
1678 static int parse_fragment(struct apply_state *state,
1679 const char *line,
1680 unsigned long size,
1681 struct patch *patch,
1682 struct fragment *fragment)
1684 int added, deleted;
1685 int len = linelen(line, size), offset;
1686 unsigned long oldlines, newlines;
1687 unsigned long leading, trailing;
1689 offset = parse_fragment_header(line, len, fragment);
1690 if (offset < 0)
1691 return -1;
1692 if (offset > 0 && patch->recount)
1693 recount_diff(line + offset, size - offset, fragment);
1694 oldlines = fragment->oldlines;
1695 newlines = fragment->newlines;
1696 leading = 0;
1697 trailing = 0;
1699 /* Parse the thing.. */
1700 line += len;
1701 size -= len;
1702 state->linenr++;
1703 added = deleted = 0;
1704 for (offset = len;
1705 0 < size;
1706 offset += len, size -= len, line += len, state->linenr++) {
1707 if (!oldlines && !newlines)
1708 break;
1709 len = linelen(line, size);
1710 if (!len || line[len-1] != '\n')
1711 return -1;
1712 switch (*line) {
1713 default:
1714 return -1;
1715 case '\n': /* newer GNU diff, an empty context line */
1716 case ' ':
1717 oldlines--;
1718 newlines--;
1719 if (!deleted && !added)
1720 leading++;
1721 trailing++;
1722 check_old_for_crlf(patch, line, len);
1723 if (!state->apply_in_reverse &&
1724 state->ws_error_action == correct_ws_error)
1725 check_whitespace(state, line, len, patch->ws_rule);
1726 break;
1727 case '-':
1728 if (!state->apply_in_reverse)
1729 check_old_for_crlf(patch, line, len);
1730 if (state->apply_in_reverse &&
1731 state->ws_error_action != nowarn_ws_error)
1732 check_whitespace(state, line, len, patch->ws_rule);
1733 deleted++;
1734 oldlines--;
1735 trailing = 0;
1736 break;
1737 case '+':
1738 if (state->apply_in_reverse)
1739 check_old_for_crlf(patch, line, len);
1740 if (!state->apply_in_reverse &&
1741 state->ws_error_action != nowarn_ws_error)
1742 check_whitespace(state, line, len, patch->ws_rule);
1743 added++;
1744 newlines--;
1745 trailing = 0;
1746 break;
1749 * We allow "\ No newline at end of file". Depending
1750 * on locale settings when the patch was produced we
1751 * don't know what this line looks like. The only
1752 * thing we do know is that it begins with "\ ".
1753 * Checking for 12 is just for sanity check -- any
1754 * l10n of "\ No newline..." is at least that long.
1756 case '\\':
1757 if (len < 12 || memcmp(line, "\\ ", 2))
1758 return -1;
1759 break;
1762 if (oldlines || newlines)
1763 return -1;
1764 if (!patch->recount && !deleted && !added)
1765 return -1;
1767 fragment->leading = leading;
1768 fragment->trailing = trailing;
1771 * If a fragment ends with an incomplete line, we failed to include
1772 * it in the above loop because we hit oldlines == newlines == 0
1773 * before seeing it.
1775 if (12 < size && !memcmp(line, "\\ ", 2))
1776 offset += linelen(line, size);
1778 patch->lines_added += added;
1779 patch->lines_deleted += deleted;
1781 if (0 < patch->is_new && oldlines)
1782 return error(_("new file depends on old contents"));
1783 if (0 < patch->is_delete && newlines)
1784 return error(_("deleted file still has contents"));
1785 return offset;
1789 * We have seen "diff --git a/... b/..." header (or a traditional patch
1790 * header). Read hunks that belong to this patch into fragments and hang
1791 * them to the given patch structure.
1793 * The (fragment->patch, fragment->size) pair points into the memory given
1794 * by the caller, not a copy, when we return.
1796 * Returns:
1797 * -1 in case of error,
1798 * the number of bytes in the patch otherwise.
1800 static int parse_single_patch(struct apply_state *state,
1801 const char *line,
1802 unsigned long size,
1803 struct patch *patch)
1805 unsigned long offset = 0;
1806 unsigned long oldlines = 0, newlines = 0, context = 0;
1807 struct fragment **fragp = &patch->fragments;
1809 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1810 struct fragment *fragment;
1811 int len;
1813 CALLOC_ARRAY(fragment, 1);
1814 fragment->linenr = state->linenr;
1815 len = parse_fragment(state, line, size, patch, fragment);
1816 if (len <= 0) {
1817 free(fragment);
1818 return error(_("corrupt patch at line %d"), state->linenr);
1820 fragment->patch = line;
1821 fragment->size = len;
1822 oldlines += fragment->oldlines;
1823 newlines += fragment->newlines;
1824 context += fragment->leading + fragment->trailing;
1826 *fragp = fragment;
1827 fragp = &fragment->next;
1829 offset += len;
1830 line += len;
1831 size -= len;
1835 * If something was removed (i.e. we have old-lines) it cannot
1836 * be creation, and if something was added it cannot be
1837 * deletion. However, the reverse is not true; --unified=0
1838 * patches that only add are not necessarily creation even
1839 * though they do not have any old lines, and ones that only
1840 * delete are not necessarily deletion.
1842 * Unfortunately, a real creation/deletion patch do _not_ have
1843 * any context line by definition, so we cannot safely tell it
1844 * apart with --unified=0 insanity. At least if the patch has
1845 * more than one hunk it is not creation or deletion.
1847 if (patch->is_new < 0 &&
1848 (oldlines || (patch->fragments && patch->fragments->next)))
1849 patch->is_new = 0;
1850 if (patch->is_delete < 0 &&
1851 (newlines || (patch->fragments && patch->fragments->next)))
1852 patch->is_delete = 0;
1854 if (0 < patch->is_new && oldlines)
1855 return error(_("new file %s depends on old contents"), patch->new_name);
1856 if (0 < patch->is_delete && newlines)
1857 return error(_("deleted file %s still has contents"), patch->old_name);
1858 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1859 fprintf_ln(stderr,
1860 _("** warning: "
1861 "file %s becomes empty but is not deleted"),
1862 patch->new_name);
1864 return offset;
1867 static inline int metadata_changes(struct patch *patch)
1869 return patch->is_rename > 0 ||
1870 patch->is_copy > 0 ||
1871 patch->is_new > 0 ||
1872 patch->is_delete ||
1873 (patch->old_mode && patch->new_mode &&
1874 patch->old_mode != patch->new_mode);
1877 static char *inflate_it(const void *data, unsigned long size,
1878 unsigned long inflated_size)
1880 git_zstream stream;
1881 void *out;
1882 int st;
1884 memset(&stream, 0, sizeof(stream));
1886 stream.next_in = (unsigned char *)data;
1887 stream.avail_in = size;
1888 stream.next_out = out = xmalloc(inflated_size);
1889 stream.avail_out = inflated_size;
1890 git_inflate_init(&stream);
1891 st = git_inflate(&stream, Z_FINISH);
1892 git_inflate_end(&stream);
1893 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1894 free(out);
1895 return NULL;
1897 return out;
1901 * Read a binary hunk and return a new fragment; fragment->patch
1902 * points at an allocated memory that the caller must free, so
1903 * it is marked as "->free_patch = 1".
1905 static struct fragment *parse_binary_hunk(struct apply_state *state,
1906 char **buf_p,
1907 unsigned long *sz_p,
1908 int *status_p,
1909 int *used_p)
1912 * Expect a line that begins with binary patch method ("literal"
1913 * or "delta"), followed by the length of data before deflating.
1914 * a sequence of 'length-byte' followed by base-85 encoded data
1915 * should follow, terminated by a newline.
1917 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1918 * and we would limit the patch line to 66 characters,
1919 * so one line can fit up to 13 groups that would decode
1920 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1921 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1923 int llen, used;
1924 unsigned long size = *sz_p;
1925 char *buffer = *buf_p;
1926 int patch_method;
1927 unsigned long origlen;
1928 char *data = NULL;
1929 int hunk_size = 0;
1930 struct fragment *frag;
1932 llen = linelen(buffer, size);
1933 used = llen;
1935 *status_p = 0;
1937 if (starts_with(buffer, "delta ")) {
1938 patch_method = BINARY_DELTA_DEFLATED;
1939 origlen = strtoul(buffer + 6, NULL, 10);
1941 else if (starts_with(buffer, "literal ")) {
1942 patch_method = BINARY_LITERAL_DEFLATED;
1943 origlen = strtoul(buffer + 8, NULL, 10);
1945 else
1946 return NULL;
1948 state->linenr++;
1949 buffer += llen;
1950 size -= llen;
1951 while (1) {
1952 int byte_length, max_byte_length, newsize;
1953 llen = linelen(buffer, size);
1954 used += llen;
1955 state->linenr++;
1956 if (llen == 1) {
1957 /* consume the blank line */
1958 buffer++;
1959 size--;
1960 break;
1963 * Minimum line is "A00000\n" which is 7-byte long,
1964 * and the line length must be multiple of 5 plus 2.
1966 if ((llen < 7) || (llen-2) % 5)
1967 goto corrupt;
1968 max_byte_length = (llen - 2) / 5 * 4;
1969 byte_length = *buffer;
1970 if ('A' <= byte_length && byte_length <= 'Z')
1971 byte_length = byte_length - 'A' + 1;
1972 else if ('a' <= byte_length && byte_length <= 'z')
1973 byte_length = byte_length - 'a' + 27;
1974 else
1975 goto corrupt;
1976 /* if the input length was not multiple of 4, we would
1977 * have filler at the end but the filler should never
1978 * exceed 3 bytes
1980 if (max_byte_length < byte_length ||
1981 byte_length <= max_byte_length - 4)
1982 goto corrupt;
1983 newsize = hunk_size + byte_length;
1984 data = xrealloc(data, newsize);
1985 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1986 goto corrupt;
1987 hunk_size = newsize;
1988 buffer += llen;
1989 size -= llen;
1992 CALLOC_ARRAY(frag, 1);
1993 frag->patch = inflate_it(data, hunk_size, origlen);
1994 frag->free_patch = 1;
1995 if (!frag->patch)
1996 goto corrupt;
1997 free(data);
1998 frag->size = origlen;
1999 *buf_p = buffer;
2000 *sz_p = size;
2001 *used_p = used;
2002 frag->binary_patch_method = patch_method;
2003 return frag;
2005 corrupt:
2006 free(data);
2007 *status_p = -1;
2008 error(_("corrupt binary patch at line %d: %.*s"),
2009 state->linenr-1, llen-1, buffer);
2010 return NULL;
2014 * Returns:
2015 * -1 in case of error,
2016 * the length of the parsed binary patch otherwise
2018 static int parse_binary(struct apply_state *state,
2019 char *buffer,
2020 unsigned long size,
2021 struct patch *patch)
2024 * We have read "GIT binary patch\n"; what follows is a line
2025 * that says the patch method (currently, either "literal" or
2026 * "delta") and the length of data before deflating; a
2027 * sequence of 'length-byte' followed by base-85 encoded data
2028 * follows.
2030 * When a binary patch is reversible, there is another binary
2031 * hunk in the same format, starting with patch method (either
2032 * "literal" or "delta") with the length of data, and a sequence
2033 * of length-byte + base-85 encoded data, terminated with another
2034 * empty line. This data, when applied to the postimage, produces
2035 * the preimage.
2037 struct fragment *forward;
2038 struct fragment *reverse;
2039 int status;
2040 int used, used_1;
2042 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2043 if (!forward && !status)
2044 /* there has to be one hunk (forward hunk) */
2045 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2046 if (status)
2047 /* otherwise we already gave an error message */
2048 return status;
2050 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2051 if (reverse)
2052 used += used_1;
2053 else if (status) {
2055 * Not having reverse hunk is not an error, but having
2056 * a corrupt reverse hunk is.
2058 free((void*) forward->patch);
2059 free(forward);
2060 return status;
2062 forward->next = reverse;
2063 patch->fragments = forward;
2064 patch->is_binary = 1;
2065 return used;
2068 static void prefix_one(struct apply_state *state, char **name)
2070 char *old_name = *name;
2071 if (!old_name)
2072 return;
2073 *name = prefix_filename(state->prefix, *name);
2074 free(old_name);
2077 static void prefix_patch(struct apply_state *state, struct patch *p)
2079 if (!state->prefix || p->is_toplevel_relative)
2080 return;
2081 prefix_one(state, &p->new_name);
2082 prefix_one(state, &p->old_name);
2086 * include/exclude
2089 static void add_name_limit(struct apply_state *state,
2090 const char *name,
2091 int exclude)
2093 struct string_list_item *it;
2095 it = string_list_append(&state->limit_by_name, name);
2096 it->util = exclude ? NULL : (void *) 1;
2099 static int use_patch(struct apply_state *state, struct patch *p)
2101 const char *pathname = p->new_name ? p->new_name : p->old_name;
2102 int i;
2104 /* Paths outside are not touched regardless of "--include" */
2105 if (state->prefix && *state->prefix) {
2106 const char *rest;
2107 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2108 return 0;
2111 /* See if it matches any of exclude/include rule */
2112 for (i = 0; i < state->limit_by_name.nr; i++) {
2113 struct string_list_item *it = &state->limit_by_name.items[i];
2114 if (!wildmatch(it->string, pathname, 0))
2115 return (it->util != NULL);
2119 * If we had any include, a path that does not match any rule is
2120 * not used. Otherwise, we saw bunch of exclude rules (or none)
2121 * and such a path is used.
2123 return !state->has_include;
2127 * Read the patch text in "buffer" that extends for "size" bytes; stop
2128 * reading after seeing a single patch (i.e. changes to a single file).
2129 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2131 * Returns:
2132 * -1 if no header was found or parse_binary() failed,
2133 * -128 on another error,
2134 * the number of bytes consumed otherwise,
2135 * so that the caller can call us again for the next patch.
2137 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2139 int hdrsize, patchsize;
2140 int offset = find_header(state, buffer, size, &hdrsize, patch);
2142 if (offset < 0)
2143 return offset;
2145 prefix_patch(state, patch);
2147 if (!use_patch(state, patch))
2148 patch->ws_rule = 0;
2149 else if (patch->new_name)
2150 patch->ws_rule = whitespace_rule(state->repo->index,
2151 patch->new_name);
2152 else
2153 patch->ws_rule = whitespace_rule(state->repo->index,
2154 patch->old_name);
2156 patchsize = parse_single_patch(state,
2157 buffer + offset + hdrsize,
2158 size - offset - hdrsize,
2159 patch);
2161 if (patchsize < 0)
2162 return -128;
2164 if (!patchsize) {
2165 static const char git_binary[] = "GIT binary patch\n";
2166 int hd = hdrsize + offset;
2167 unsigned long llen = linelen(buffer + hd, size - hd);
2169 if (llen == sizeof(git_binary) - 1 &&
2170 !memcmp(git_binary, buffer + hd, llen)) {
2171 int used;
2172 state->linenr++;
2173 used = parse_binary(state, buffer + hd + llen,
2174 size - hd - llen, patch);
2175 if (used < 0)
2176 return -1;
2177 if (used)
2178 patchsize = used + llen;
2179 else
2180 patchsize = 0;
2182 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2183 static const char *binhdr[] = {
2184 "Binary files ",
2185 "Files ",
2186 NULL,
2188 int i;
2189 for (i = 0; binhdr[i]; i++) {
2190 int len = strlen(binhdr[i]);
2191 if (len < size - hd &&
2192 !memcmp(binhdr[i], buffer + hd, len)) {
2193 state->linenr++;
2194 patch->is_binary = 1;
2195 patchsize = llen;
2196 break;
2201 /* Empty patch cannot be applied if it is a text patch
2202 * without metadata change. A binary patch appears
2203 * empty to us here.
2205 if ((state->apply || state->check) &&
2206 (!patch->is_binary && !metadata_changes(patch))) {
2207 error(_("patch with only garbage at line %d"), state->linenr);
2208 return -128;
2212 return offset + hdrsize + patchsize;
2215 static void reverse_patches(struct patch *p)
2217 for (; p; p = p->next) {
2218 struct fragment *frag = p->fragments;
2220 SWAP(p->new_name, p->old_name);
2221 SWAP(p->new_mode, p->old_mode);
2222 SWAP(p->is_new, p->is_delete);
2223 SWAP(p->lines_added, p->lines_deleted);
2224 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2226 for (; frag; frag = frag->next) {
2227 SWAP(frag->newpos, frag->oldpos);
2228 SWAP(frag->newlines, frag->oldlines);
2233 static const char pluses[] =
2234 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2235 static const char minuses[]=
2236 "----------------------------------------------------------------------";
2238 static void show_stats(struct apply_state *state, struct patch *patch)
2240 struct strbuf qname = STRBUF_INIT;
2241 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2242 int max, add, del;
2244 quote_c_style(cp, &qname, NULL, 0);
2247 * "scale" the filename
2249 max = state->max_len;
2250 if (max > 50)
2251 max = 50;
2253 if (qname.len > max) {
2254 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2255 if (!cp)
2256 cp = qname.buf + qname.len + 3 - max;
2257 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2260 if (patch->is_binary) {
2261 printf(" %-*s | Bin\n", max, qname.buf);
2262 strbuf_release(&qname);
2263 return;
2266 printf(" %-*s |", max, qname.buf);
2267 strbuf_release(&qname);
2270 * scale the add/delete
2272 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2273 add = patch->lines_added;
2274 del = patch->lines_deleted;
2276 if (state->max_change > 0) {
2277 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2278 add = (add * max + state->max_change / 2) / state->max_change;
2279 del = total - add;
2281 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2282 add, pluses, del, minuses);
2285 static int read_old_data(struct stat *st, struct patch *patch,
2286 const char *path, struct strbuf *buf)
2288 int conv_flags = patch->crlf_in_old ?
2289 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2290 switch (st->st_mode & S_IFMT) {
2291 case S_IFLNK:
2292 if (strbuf_readlink(buf, path, st->st_size) < 0)
2293 return error(_("unable to read symlink %s"), path);
2294 return 0;
2295 case S_IFREG:
2296 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2297 return error(_("unable to open or read %s"), path);
2299 * "git apply" without "--index/--cached" should never look
2300 * at the index; the target file may not have been added to
2301 * the index yet, and we may not even be in any Git repository.
2302 * Pass NULL to convert_to_git() to stress this; the function
2303 * should never look at the index when explicit crlf option
2304 * is given.
2306 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2307 return 0;
2308 default:
2309 return -1;
2314 * Update the preimage, and the common lines in postimage,
2315 * from buffer buf of length len. If postlen is 0 the postimage
2316 * is updated in place, otherwise it's updated on a new buffer
2317 * of length postlen
2320 static void update_pre_post_images(struct image *preimage,
2321 struct image *postimage,
2322 char *buf,
2323 size_t len, size_t postlen)
2325 int i, ctx, reduced;
2326 char *new_buf, *old_buf, *fixed;
2327 struct image fixed_preimage;
2330 * Update the preimage with whitespace fixes. Note that we
2331 * are not losing preimage->buf -- apply_one_fragment() will
2332 * free "oldlines".
2334 prepare_image(&fixed_preimage, buf, len, 1);
2335 assert(postlen
2336 ? fixed_preimage.nr == preimage->nr
2337 : fixed_preimage.nr <= preimage->nr);
2338 for (i = 0; i < fixed_preimage.nr; i++)
2339 fixed_preimage.line[i].flag = preimage->line[i].flag;
2340 free(preimage->line_allocated);
2341 *preimage = fixed_preimage;
2344 * Adjust the common context lines in postimage. This can be
2345 * done in-place when we are shrinking it with whitespace
2346 * fixing, but needs a new buffer when ignoring whitespace or
2347 * expanding leading tabs to spaces.
2349 * We trust the caller to tell us if the update can be done
2350 * in place (postlen==0) or not.
2352 old_buf = postimage->buf;
2353 if (postlen)
2354 new_buf = postimage->buf = xmalloc(postlen);
2355 else
2356 new_buf = old_buf;
2357 fixed = preimage->buf;
2359 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2360 size_t l_len = postimage->line[i].len;
2361 if (!(postimage->line[i].flag & LINE_COMMON)) {
2362 /* an added line -- no counterparts in preimage */
2363 memmove(new_buf, old_buf, l_len);
2364 old_buf += l_len;
2365 new_buf += l_len;
2366 continue;
2369 /* a common context -- skip it in the original postimage */
2370 old_buf += l_len;
2372 /* and find the corresponding one in the fixed preimage */
2373 while (ctx < preimage->nr &&
2374 !(preimage->line[ctx].flag & LINE_COMMON)) {
2375 fixed += preimage->line[ctx].len;
2376 ctx++;
2380 * preimage is expected to run out, if the caller
2381 * fixed addition of trailing blank lines.
2383 if (preimage->nr <= ctx) {
2384 reduced++;
2385 continue;
2388 /* and copy it in, while fixing the line length */
2389 l_len = preimage->line[ctx].len;
2390 memcpy(new_buf, fixed, l_len);
2391 new_buf += l_len;
2392 fixed += l_len;
2393 postimage->line[i].len = l_len;
2394 ctx++;
2397 if (postlen
2398 ? postlen < new_buf - postimage->buf
2399 : postimage->len < new_buf - postimage->buf)
2400 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2401 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2403 /* Fix the length of the whole thing */
2404 postimage->len = new_buf - postimage->buf;
2405 postimage->nr -= reduced;
2408 static int line_by_line_fuzzy_match(struct image *img,
2409 struct image *preimage,
2410 struct image *postimage,
2411 unsigned long current,
2412 int current_lno,
2413 int preimage_limit)
2415 int i;
2416 size_t imgoff = 0;
2417 size_t preoff = 0;
2418 size_t postlen = postimage->len;
2419 size_t extra_chars;
2420 char *buf;
2421 char *preimage_eof;
2422 char *preimage_end;
2423 struct strbuf fixed;
2424 char *fixed_buf;
2425 size_t fixed_len;
2427 for (i = 0; i < preimage_limit; i++) {
2428 size_t prelen = preimage->line[i].len;
2429 size_t imglen = img->line[current_lno+i].len;
2431 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2432 preimage->buf + preoff, prelen))
2433 return 0;
2434 if (preimage->line[i].flag & LINE_COMMON)
2435 postlen += imglen - prelen;
2436 imgoff += imglen;
2437 preoff += prelen;
2441 * Ok, the preimage matches with whitespace fuzz.
2443 * imgoff now holds the true length of the target that
2444 * matches the preimage before the end of the file.
2446 * Count the number of characters in the preimage that fall
2447 * beyond the end of the file and make sure that all of them
2448 * are whitespace characters. (This can only happen if
2449 * we are removing blank lines at the end of the file.)
2451 buf = preimage_eof = preimage->buf + preoff;
2452 for ( ; i < preimage->nr; i++)
2453 preoff += preimage->line[i].len;
2454 preimage_end = preimage->buf + preoff;
2455 for ( ; buf < preimage_end; buf++)
2456 if (!isspace(*buf))
2457 return 0;
2460 * Update the preimage and the common postimage context
2461 * lines to use the same whitespace as the target.
2462 * If whitespace is missing in the target (i.e.
2463 * if the preimage extends beyond the end of the file),
2464 * use the whitespace from the preimage.
2466 extra_chars = preimage_end - preimage_eof;
2467 strbuf_init(&fixed, imgoff + extra_chars);
2468 strbuf_add(&fixed, img->buf + current, imgoff);
2469 strbuf_add(&fixed, preimage_eof, extra_chars);
2470 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2471 update_pre_post_images(preimage, postimage,
2472 fixed_buf, fixed_len, postlen);
2473 return 1;
2476 static int match_fragment(struct apply_state *state,
2477 struct image *img,
2478 struct image *preimage,
2479 struct image *postimage,
2480 unsigned long current,
2481 int current_lno,
2482 unsigned ws_rule,
2483 int match_beginning, int match_end)
2485 int i;
2486 char *fixed_buf, *buf, *orig, *target;
2487 struct strbuf fixed;
2488 size_t fixed_len, postlen;
2489 int preimage_limit;
2491 if (preimage->nr + current_lno <= img->nr) {
2493 * The hunk falls within the boundaries of img.
2495 preimage_limit = preimage->nr;
2496 if (match_end && (preimage->nr + current_lno != img->nr))
2497 return 0;
2498 } else if (state->ws_error_action == correct_ws_error &&
2499 (ws_rule & WS_BLANK_AT_EOF)) {
2501 * This hunk extends beyond the end of img, and we are
2502 * removing blank lines at the end of the file. This
2503 * many lines from the beginning of the preimage must
2504 * match with img, and the remainder of the preimage
2505 * must be blank.
2507 preimage_limit = img->nr - current_lno;
2508 } else {
2510 * The hunk extends beyond the end of the img and
2511 * we are not removing blanks at the end, so we
2512 * should reject the hunk at this position.
2514 return 0;
2517 if (match_beginning && current_lno)
2518 return 0;
2520 /* Quick hash check */
2521 for (i = 0; i < preimage_limit; i++)
2522 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2523 (preimage->line[i].hash != img->line[current_lno + i].hash))
2524 return 0;
2526 if (preimage_limit == preimage->nr) {
2528 * Do we have an exact match? If we were told to match
2529 * at the end, size must be exactly at current+fragsize,
2530 * otherwise current+fragsize must be still within the preimage,
2531 * and either case, the old piece should match the preimage
2532 * exactly.
2534 if ((match_end
2535 ? (current + preimage->len == img->len)
2536 : (current + preimage->len <= img->len)) &&
2537 !memcmp(img->buf + current, preimage->buf, preimage->len))
2538 return 1;
2539 } else {
2541 * The preimage extends beyond the end of img, so
2542 * there cannot be an exact match.
2544 * There must be one non-blank context line that match
2545 * a line before the end of img.
2547 char *buf_end;
2549 buf = preimage->buf;
2550 buf_end = buf;
2551 for (i = 0; i < preimage_limit; i++)
2552 buf_end += preimage->line[i].len;
2554 for ( ; buf < buf_end; buf++)
2555 if (!isspace(*buf))
2556 break;
2557 if (buf == buf_end)
2558 return 0;
2562 * No exact match. If we are ignoring whitespace, run a line-by-line
2563 * fuzzy matching. We collect all the line length information because
2564 * we need it to adjust whitespace if we match.
2566 if (state->ws_ignore_action == ignore_ws_change)
2567 return line_by_line_fuzzy_match(img, preimage, postimage,
2568 current, current_lno, preimage_limit);
2570 if (state->ws_error_action != correct_ws_error)
2571 return 0;
2574 * The hunk does not apply byte-by-byte, but the hash says
2575 * it might with whitespace fuzz. We weren't asked to
2576 * ignore whitespace, we were asked to correct whitespace
2577 * errors, so let's try matching after whitespace correction.
2579 * While checking the preimage against the target, whitespace
2580 * errors in both fixed, we count how large the corresponding
2581 * postimage needs to be. The postimage prepared by
2582 * apply_one_fragment() has whitespace errors fixed on added
2583 * lines already, but the common lines were propagated as-is,
2584 * which may become longer when their whitespace errors are
2585 * fixed.
2588 /* First count added lines in postimage */
2589 postlen = 0;
2590 for (i = 0; i < postimage->nr; i++) {
2591 if (!(postimage->line[i].flag & LINE_COMMON))
2592 postlen += postimage->line[i].len;
2596 * The preimage may extend beyond the end of the file,
2597 * but in this loop we will only handle the part of the
2598 * preimage that falls within the file.
2600 strbuf_init(&fixed, preimage->len + 1);
2601 orig = preimage->buf;
2602 target = img->buf + current;
2603 for (i = 0; i < preimage_limit; i++) {
2604 size_t oldlen = preimage->line[i].len;
2605 size_t tgtlen = img->line[current_lno + i].len;
2606 size_t fixstart = fixed.len;
2607 struct strbuf tgtfix;
2608 int match;
2610 /* Try fixing the line in the preimage */
2611 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2613 /* Try fixing the line in the target */
2614 strbuf_init(&tgtfix, tgtlen);
2615 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2618 * If they match, either the preimage was based on
2619 * a version before our tree fixed whitespace breakage,
2620 * or we are lacking a whitespace-fix patch the tree
2621 * the preimage was based on already had (i.e. target
2622 * has whitespace breakage, the preimage doesn't).
2623 * In either case, we are fixing the whitespace breakages
2624 * so we might as well take the fix together with their
2625 * real change.
2627 match = (tgtfix.len == fixed.len - fixstart &&
2628 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2629 fixed.len - fixstart));
2631 /* Add the length if this is common with the postimage */
2632 if (preimage->line[i].flag & LINE_COMMON)
2633 postlen += tgtfix.len;
2635 strbuf_release(&tgtfix);
2636 if (!match)
2637 goto unmatch_exit;
2639 orig += oldlen;
2640 target += tgtlen;
2645 * Now handle the lines in the preimage that falls beyond the
2646 * end of the file (if any). They will only match if they are
2647 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2648 * false).
2650 for ( ; i < preimage->nr; i++) {
2651 size_t fixstart = fixed.len; /* start of the fixed preimage */
2652 size_t oldlen = preimage->line[i].len;
2653 int j;
2655 /* Try fixing the line in the preimage */
2656 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2658 for (j = fixstart; j < fixed.len; j++)
2659 if (!isspace(fixed.buf[j]))
2660 goto unmatch_exit;
2662 orig += oldlen;
2666 * Yes, the preimage is based on an older version that still
2667 * has whitespace breakages unfixed, and fixing them makes the
2668 * hunk match. Update the context lines in the postimage.
2670 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2671 if (postlen < postimage->len)
2672 postlen = 0;
2673 update_pre_post_images(preimage, postimage,
2674 fixed_buf, fixed_len, postlen);
2675 return 1;
2677 unmatch_exit:
2678 strbuf_release(&fixed);
2679 return 0;
2682 static int find_pos(struct apply_state *state,
2683 struct image *img,
2684 struct image *preimage,
2685 struct image *postimage,
2686 int line,
2687 unsigned ws_rule,
2688 int match_beginning, int match_end)
2690 int i;
2691 unsigned long backwards, forwards, current;
2692 int backwards_lno, forwards_lno, current_lno;
2695 * When running with --allow-overlap, it is possible that a hunk is
2696 * seen that pretends to start at the beginning (but no longer does),
2697 * and that *still* needs to match the end. So trust `match_end` more
2698 * than `match_beginning`.
2700 if (state->allow_overlap && match_beginning && match_end &&
2701 img->nr - preimage->nr != 0)
2702 match_beginning = 0;
2705 * If match_beginning or match_end is specified, there is no
2706 * point starting from a wrong line that will never match and
2707 * wander around and wait for a match at the specified end.
2709 if (match_beginning)
2710 line = 0;
2711 else if (match_end)
2712 line = img->nr - preimage->nr;
2715 * Because the comparison is unsigned, the following test
2716 * will also take care of a negative line number that can
2717 * result when match_end and preimage is larger than the target.
2719 if ((size_t) line > img->nr)
2720 line = img->nr;
2722 current = 0;
2723 for (i = 0; i < line; i++)
2724 current += img->line[i].len;
2727 * There's probably some smart way to do this, but I'll leave
2728 * that to the smart and beautiful people. I'm simple and stupid.
2730 backwards = current;
2731 backwards_lno = line;
2732 forwards = current;
2733 forwards_lno = line;
2734 current_lno = line;
2736 for (i = 0; ; i++) {
2737 if (match_fragment(state, img, preimage, postimage,
2738 current, current_lno, ws_rule,
2739 match_beginning, match_end))
2740 return current_lno;
2742 again:
2743 if (backwards_lno == 0 && forwards_lno == img->nr)
2744 break;
2746 if (i & 1) {
2747 if (backwards_lno == 0) {
2748 i++;
2749 goto again;
2751 backwards_lno--;
2752 backwards -= img->line[backwards_lno].len;
2753 current = backwards;
2754 current_lno = backwards_lno;
2755 } else {
2756 if (forwards_lno == img->nr) {
2757 i++;
2758 goto again;
2760 forwards += img->line[forwards_lno].len;
2761 forwards_lno++;
2762 current = forwards;
2763 current_lno = forwards_lno;
2767 return -1;
2770 static void remove_first_line(struct image *img)
2772 img->buf += img->line[0].len;
2773 img->len -= img->line[0].len;
2774 img->line++;
2775 img->nr--;
2778 static void remove_last_line(struct image *img)
2780 img->len -= img->line[--img->nr].len;
2784 * The change from "preimage" and "postimage" has been found to
2785 * apply at applied_pos (counts in line numbers) in "img".
2786 * Update "img" to remove "preimage" and replace it with "postimage".
2788 static void update_image(struct apply_state *state,
2789 struct image *img,
2790 int applied_pos,
2791 struct image *preimage,
2792 struct image *postimage)
2795 * remove the copy of preimage at offset in img
2796 * and replace it with postimage
2798 int i, nr;
2799 size_t remove_count, insert_count, applied_at = 0;
2800 char *result;
2801 int preimage_limit;
2804 * If we are removing blank lines at the end of img,
2805 * the preimage may extend beyond the end.
2806 * If that is the case, we must be careful only to
2807 * remove the part of the preimage that falls within
2808 * the boundaries of img. Initialize preimage_limit
2809 * to the number of lines in the preimage that falls
2810 * within the boundaries.
2812 preimage_limit = preimage->nr;
2813 if (preimage_limit > img->nr - applied_pos)
2814 preimage_limit = img->nr - applied_pos;
2816 for (i = 0; i < applied_pos; i++)
2817 applied_at += img->line[i].len;
2819 remove_count = 0;
2820 for (i = 0; i < preimage_limit; i++)
2821 remove_count += img->line[applied_pos + i].len;
2822 insert_count = postimage->len;
2824 /* Adjust the contents */
2825 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2826 memcpy(result, img->buf, applied_at);
2827 memcpy(result + applied_at, postimage->buf, postimage->len);
2828 memcpy(result + applied_at + postimage->len,
2829 img->buf + (applied_at + remove_count),
2830 img->len - (applied_at + remove_count));
2831 free(img->buf);
2832 img->buf = result;
2833 img->len += insert_count - remove_count;
2834 result[img->len] = '\0';
2836 /* Adjust the line table */
2837 nr = img->nr + postimage->nr - preimage_limit;
2838 if (preimage_limit < postimage->nr) {
2840 * NOTE: this knows that we never call remove_first_line()
2841 * on anything other than pre/post image.
2843 REALLOC_ARRAY(img->line, nr);
2844 img->line_allocated = img->line;
2846 if (preimage_limit != postimage->nr)
2847 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2848 img->line + applied_pos + preimage_limit,
2849 img->nr - (applied_pos + preimage_limit));
2850 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2851 if (!state->allow_overlap)
2852 for (i = 0; i < postimage->nr; i++)
2853 img->line[applied_pos + i].flag |= LINE_PATCHED;
2854 img->nr = nr;
2858 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2859 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2860 * replace the part of "img" with "postimage" text.
2862 static int apply_one_fragment(struct apply_state *state,
2863 struct image *img, struct fragment *frag,
2864 int inaccurate_eof, unsigned ws_rule,
2865 int nth_fragment)
2867 int match_beginning, match_end;
2868 const char *patch = frag->patch;
2869 int size = frag->size;
2870 char *old, *oldlines;
2871 struct strbuf newlines;
2872 int new_blank_lines_at_end = 0;
2873 int found_new_blank_lines_at_end = 0;
2874 int hunk_linenr = frag->linenr;
2875 unsigned long leading, trailing;
2876 int pos, applied_pos;
2877 struct image preimage;
2878 struct image postimage;
2880 memset(&preimage, 0, sizeof(preimage));
2881 memset(&postimage, 0, sizeof(postimage));
2882 oldlines = xmalloc(size);
2883 strbuf_init(&newlines, size);
2885 old = oldlines;
2886 while (size > 0) {
2887 char first;
2888 int len = linelen(patch, size);
2889 int plen;
2890 int added_blank_line = 0;
2891 int is_blank_context = 0;
2892 size_t start;
2894 if (!len)
2895 break;
2898 * "plen" is how much of the line we should use for
2899 * the actual patch data. Normally we just remove the
2900 * first character on the line, but if the line is
2901 * followed by "\ No newline", then we also remove the
2902 * last one (which is the newline, of course).
2904 plen = len - 1;
2905 if (len < size && patch[len] == '\\')
2906 plen--;
2907 first = *patch;
2908 if (state->apply_in_reverse) {
2909 if (first == '-')
2910 first = '+';
2911 else if (first == '+')
2912 first = '-';
2915 switch (first) {
2916 case '\n':
2917 /* Newer GNU diff, empty context line */
2918 if (plen < 0)
2919 /* ... followed by '\No newline'; nothing */
2920 break;
2921 *old++ = '\n';
2922 strbuf_addch(&newlines, '\n');
2923 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2924 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2925 is_blank_context = 1;
2926 break;
2927 case ' ':
2928 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2929 ws_blank_line(patch + 1, plen))
2930 is_blank_context = 1;
2931 /* fallthrough */
2932 case '-':
2933 memcpy(old, patch + 1, plen);
2934 add_line_info(&preimage, old, plen,
2935 (first == ' ' ? LINE_COMMON : 0));
2936 old += plen;
2937 if (first == '-')
2938 break;
2939 /* fallthrough */
2940 case '+':
2941 /* --no-add does not add new lines */
2942 if (first == '+' && state->no_add)
2943 break;
2945 start = newlines.len;
2946 if (first != '+' ||
2947 !state->whitespace_error ||
2948 state->ws_error_action != correct_ws_error) {
2949 strbuf_add(&newlines, patch + 1, plen);
2951 else {
2952 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2954 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2955 (first == '+' ? 0 : LINE_COMMON));
2956 if (first == '+' &&
2957 (ws_rule & WS_BLANK_AT_EOF) &&
2958 ws_blank_line(patch + 1, plen))
2959 added_blank_line = 1;
2960 break;
2961 case '@': case '\\':
2962 /* Ignore it, we already handled it */
2963 break;
2964 default:
2965 if (state->apply_verbosity > verbosity_normal)
2966 error(_("invalid start of line: '%c'"), first);
2967 applied_pos = -1;
2968 goto out;
2970 if (added_blank_line) {
2971 if (!new_blank_lines_at_end)
2972 found_new_blank_lines_at_end = hunk_linenr;
2973 new_blank_lines_at_end++;
2975 else if (is_blank_context)
2977 else
2978 new_blank_lines_at_end = 0;
2979 patch += len;
2980 size -= len;
2981 hunk_linenr++;
2983 if (inaccurate_eof &&
2984 old > oldlines && old[-1] == '\n' &&
2985 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2986 old--;
2987 strbuf_setlen(&newlines, newlines.len - 1);
2988 preimage.line_allocated[preimage.nr - 1].len--;
2989 postimage.line_allocated[postimage.nr - 1].len--;
2992 leading = frag->leading;
2993 trailing = frag->trailing;
2996 * A hunk to change lines at the beginning would begin with
2997 * @@ -1,L +N,M @@
2998 * but we need to be careful. -U0 that inserts before the second
2999 * line also has this pattern.
3001 * And a hunk to add to an empty file would begin with
3002 * @@ -0,0 +N,M @@
3004 * In other words, a hunk that is (frag->oldpos <= 1) with or
3005 * without leading context must match at the beginning.
3007 match_beginning = (!frag->oldpos ||
3008 (frag->oldpos == 1 && !state->unidiff_zero));
3011 * A hunk without trailing lines must match at the end.
3012 * However, we simply cannot tell if a hunk must match end
3013 * from the lack of trailing lines if the patch was generated
3014 * with unidiff without any context.
3016 match_end = !state->unidiff_zero && !trailing;
3018 pos = frag->newpos ? (frag->newpos - 1) : 0;
3019 preimage.buf = oldlines;
3020 preimage.len = old - oldlines;
3021 postimage.buf = newlines.buf;
3022 postimage.len = newlines.len;
3023 preimage.line = preimage.line_allocated;
3024 postimage.line = postimage.line_allocated;
3026 for (;;) {
3028 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3029 ws_rule, match_beginning, match_end);
3031 if (applied_pos >= 0)
3032 break;
3034 /* Am I at my context limits? */
3035 if ((leading <= state->p_context) && (trailing <= state->p_context))
3036 break;
3037 if (match_beginning || match_end) {
3038 match_beginning = match_end = 0;
3039 continue;
3043 * Reduce the number of context lines; reduce both
3044 * leading and trailing if they are equal otherwise
3045 * just reduce the larger context.
3047 if (leading >= trailing) {
3048 remove_first_line(&preimage);
3049 remove_first_line(&postimage);
3050 pos--;
3051 leading--;
3053 if (trailing > leading) {
3054 remove_last_line(&preimage);
3055 remove_last_line(&postimage);
3056 trailing--;
3060 if (applied_pos >= 0) {
3061 if (new_blank_lines_at_end &&
3062 preimage.nr + applied_pos >= img->nr &&
3063 (ws_rule & WS_BLANK_AT_EOF) &&
3064 state->ws_error_action != nowarn_ws_error) {
3065 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3066 found_new_blank_lines_at_end);
3067 if (state->ws_error_action == correct_ws_error) {
3068 while (new_blank_lines_at_end--)
3069 remove_last_line(&postimage);
3072 * We would want to prevent write_out_results()
3073 * from taking place in apply_patch() that follows
3074 * the callchain led us here, which is:
3075 * apply_patch->check_patch_list->check_patch->
3076 * apply_data->apply_fragments->apply_one_fragment
3078 if (state->ws_error_action == die_on_ws_error)
3079 state->apply = 0;
3082 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3083 int offset = applied_pos - pos;
3084 if (state->apply_in_reverse)
3085 offset = 0 - offset;
3086 fprintf_ln(stderr,
3087 Q_("Hunk #%d succeeded at %d (offset %d line).",
3088 "Hunk #%d succeeded at %d (offset %d lines).",
3089 offset),
3090 nth_fragment, applied_pos + 1, offset);
3094 * Warn if it was necessary to reduce the number
3095 * of context lines.
3097 if ((leading != frag->leading ||
3098 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3099 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3100 " to apply fragment at %d"),
3101 leading, trailing, applied_pos+1);
3102 update_image(state, img, applied_pos, &preimage, &postimage);
3103 } else {
3104 if (state->apply_verbosity > verbosity_normal)
3105 error(_("while searching for:\n%.*s"),
3106 (int)(old - oldlines), oldlines);
3109 out:
3110 free(oldlines);
3111 strbuf_release(&newlines);
3112 free(preimage.line_allocated);
3113 free(postimage.line_allocated);
3115 return (applied_pos < 0);
3118 static int apply_binary_fragment(struct apply_state *state,
3119 struct image *img,
3120 struct patch *patch)
3122 struct fragment *fragment = patch->fragments;
3123 unsigned long len;
3124 void *dst;
3126 if (!fragment)
3127 return error(_("missing binary patch data for '%s'"),
3128 patch->new_name ?
3129 patch->new_name :
3130 patch->old_name);
3132 /* Binary patch is irreversible without the optional second hunk */
3133 if (state->apply_in_reverse) {
3134 if (!fragment->next)
3135 return error(_("cannot reverse-apply a binary patch "
3136 "without the reverse hunk to '%s'"),
3137 patch->new_name
3138 ? patch->new_name : patch->old_name);
3139 fragment = fragment->next;
3141 switch (fragment->binary_patch_method) {
3142 case BINARY_DELTA_DEFLATED:
3143 dst = patch_delta(img->buf, img->len, fragment->patch,
3144 fragment->size, &len);
3145 if (!dst)
3146 return -1;
3147 clear_image(img);
3148 img->buf = dst;
3149 img->len = len;
3150 return 0;
3151 case BINARY_LITERAL_DEFLATED:
3152 clear_image(img);
3153 img->len = fragment->size;
3154 img->buf = xmemdupz(fragment->patch, img->len);
3155 return 0;
3157 return -1;
3161 * Replace "img" with the result of applying the binary patch.
3162 * The binary patch data itself in patch->fragment is still kept
3163 * but the preimage prepared by the caller in "img" is freed here
3164 * or in the helper function apply_binary_fragment() this calls.
3166 static int apply_binary(struct apply_state *state,
3167 struct image *img,
3168 struct patch *patch)
3170 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3171 struct object_id oid;
3172 const unsigned hexsz = the_hash_algo->hexsz;
3175 * For safety, we require patch index line to contain
3176 * full hex textual object ID for old and new, at least for now.
3178 if (strlen(patch->old_oid_prefix) != hexsz ||
3179 strlen(patch->new_oid_prefix) != hexsz ||
3180 get_oid_hex(patch->old_oid_prefix, &oid) ||
3181 get_oid_hex(patch->new_oid_prefix, &oid))
3182 return error(_("cannot apply binary patch to '%s' "
3183 "without full index line"), name);
3185 if (patch->old_name) {
3187 * See if the old one matches what the patch
3188 * applies to.
3190 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3191 &oid);
3192 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3193 return error(_("the patch applies to '%s' (%s), "
3194 "which does not match the "
3195 "current contents."),
3196 name, oid_to_hex(&oid));
3198 else {
3199 /* Otherwise, the old one must be empty. */
3200 if (img->len)
3201 return error(_("the patch applies to an empty "
3202 "'%s' but it is not empty"), name);
3205 get_oid_hex(patch->new_oid_prefix, &oid);
3206 if (is_null_oid(&oid)) {
3207 clear_image(img);
3208 return 0; /* deletion patch */
3211 if (has_object(the_repository, &oid, 0)) {
3212 /* We already have the postimage */
3213 enum object_type type;
3214 unsigned long size;
3215 char *result;
3217 result = repo_read_object_file(the_repository, &oid, &type,
3218 &size);
3219 if (!result)
3220 return error(_("the necessary postimage %s for "
3221 "'%s' cannot be read"),
3222 patch->new_oid_prefix, name);
3223 clear_image(img);
3224 img->buf = result;
3225 img->len = size;
3226 } else {
3228 * We have verified buf matches the preimage;
3229 * apply the patch data to it, which is stored
3230 * in the patch->fragments->{patch,size}.
3232 if (apply_binary_fragment(state, img, patch))
3233 return error(_("binary patch does not apply to '%s'"),
3234 name);
3236 /* verify that the result matches */
3237 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3238 &oid);
3239 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3240 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3241 name, patch->new_oid_prefix, oid_to_hex(&oid));
3244 return 0;
3247 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3249 struct fragment *frag = patch->fragments;
3250 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3251 unsigned ws_rule = patch->ws_rule;
3252 unsigned inaccurate_eof = patch->inaccurate_eof;
3253 int nth = 0;
3255 if (patch->is_binary)
3256 return apply_binary(state, img, patch);
3258 while (frag) {
3259 nth++;
3260 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3261 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3262 if (!state->apply_with_reject)
3263 return -1;
3264 frag->rejected = 1;
3266 frag = frag->next;
3268 return 0;
3271 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3273 if (S_ISGITLINK(mode)) {
3274 strbuf_grow(buf, 100);
3275 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3276 } else {
3277 enum object_type type;
3278 unsigned long sz;
3279 char *result;
3281 result = repo_read_object_file(the_repository, oid, &type,
3282 &sz);
3283 if (!result)
3284 return -1;
3285 /* XXX read_sha1_file NUL-terminates */
3286 strbuf_attach(buf, result, sz, sz + 1);
3288 return 0;
3291 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3293 if (!ce)
3294 return 0;
3295 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3298 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3300 struct string_list_item *item;
3302 if (!name)
3303 return NULL;
3305 item = string_list_lookup(&state->fn_table, name);
3306 if (item)
3307 return (struct patch *)item->util;
3309 return NULL;
3313 * item->util in the filename table records the status of the path.
3314 * Usually it points at a patch (whose result records the contents
3315 * of it after applying it), but it could be PATH_WAS_DELETED for a
3316 * path that a previously applied patch has already removed, or
3317 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3319 * The latter is needed to deal with a case where two paths A and B
3320 * are swapped by first renaming A to B and then renaming B to A;
3321 * moving A to B should not be prevented due to presence of B as we
3322 * will remove it in a later patch.
3324 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3325 #define PATH_WAS_DELETED ((struct patch *) -1)
3327 static int to_be_deleted(struct patch *patch)
3329 return patch == PATH_TO_BE_DELETED;
3332 static int was_deleted(struct patch *patch)
3334 return patch == PATH_WAS_DELETED;
3337 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3339 struct string_list_item *item;
3342 * Always add new_name unless patch is a deletion
3343 * This should cover the cases for normal diffs,
3344 * file creations and copies
3346 if (patch->new_name) {
3347 item = string_list_insert(&state->fn_table, patch->new_name);
3348 item->util = patch;
3352 * store a failure on rename/deletion cases because
3353 * later chunks shouldn't patch old names
3355 if ((patch->new_name == NULL) || (patch->is_rename)) {
3356 item = string_list_insert(&state->fn_table, patch->old_name);
3357 item->util = PATH_WAS_DELETED;
3361 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3364 * store information about incoming file deletion
3366 while (patch) {
3367 if ((patch->new_name == NULL) || (patch->is_rename)) {
3368 struct string_list_item *item;
3369 item = string_list_insert(&state->fn_table, patch->old_name);
3370 item->util = PATH_TO_BE_DELETED;
3372 patch = patch->next;
3376 static int checkout_target(struct index_state *istate,
3377 struct cache_entry *ce, struct stat *st)
3379 struct checkout costate = CHECKOUT_INIT;
3381 costate.refresh_cache = 1;
3382 costate.istate = istate;
3383 if (checkout_entry(ce, &costate, NULL, NULL) ||
3384 lstat(ce->name, st))
3385 return error(_("cannot checkout %s"), ce->name);
3386 return 0;
3389 static struct patch *previous_patch(struct apply_state *state,
3390 struct patch *patch,
3391 int *gone)
3393 struct patch *previous;
3395 *gone = 0;
3396 if (patch->is_copy || patch->is_rename)
3397 return NULL; /* "git" patches do not depend on the order */
3399 previous = in_fn_table(state, patch->old_name);
3400 if (!previous)
3401 return NULL;
3403 if (to_be_deleted(previous))
3404 return NULL; /* the deletion hasn't happened yet */
3406 if (was_deleted(previous))
3407 *gone = 1;
3409 return previous;
3412 static int verify_index_match(struct apply_state *state,
3413 const struct cache_entry *ce,
3414 struct stat *st)
3416 if (S_ISGITLINK(ce->ce_mode)) {
3417 if (!S_ISDIR(st->st_mode))
3418 return -1;
3419 return 0;
3421 return ie_match_stat(state->repo->index, ce, st,
3422 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3425 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3427 static int load_patch_target(struct apply_state *state,
3428 struct strbuf *buf,
3429 const struct cache_entry *ce,
3430 struct stat *st,
3431 struct patch *patch,
3432 const char *name,
3433 unsigned expected_mode)
3435 if (state->cached || state->check_index) {
3436 if (read_file_or_gitlink(ce, buf))
3437 return error(_("failed to read %s"), name);
3438 } else if (name) {
3439 if (S_ISGITLINK(expected_mode)) {
3440 if (ce)
3441 return read_file_or_gitlink(ce, buf);
3442 else
3443 return SUBMODULE_PATCH_WITHOUT_INDEX;
3444 } else if (has_symlink_leading_path(name, strlen(name))) {
3445 return error(_("reading from '%s' beyond a symbolic link"), name);
3446 } else {
3447 if (read_old_data(st, patch, name, buf))
3448 return error(_("failed to read %s"), name);
3451 return 0;
3455 * We are about to apply "patch"; populate the "image" with the
3456 * current version we have, from the working tree or from the index,
3457 * depending on the situation e.g. --cached/--index. If we are
3458 * applying a non-git patch that incrementally updates the tree,
3459 * we read from the result of a previous diff.
3461 static int load_preimage(struct apply_state *state,
3462 struct image *image,
3463 struct patch *patch, struct stat *st,
3464 const struct cache_entry *ce)
3466 struct strbuf buf = STRBUF_INIT;
3467 size_t len;
3468 char *img;
3469 struct patch *previous;
3470 int status;
3472 previous = previous_patch(state, patch, &status);
3473 if (status)
3474 return error(_("path %s has been renamed/deleted"),
3475 patch->old_name);
3476 if (previous) {
3477 /* We have a patched copy in memory; use that. */
3478 strbuf_add(&buf, previous->result, previous->resultsize);
3479 } else {
3480 status = load_patch_target(state, &buf, ce, st, patch,
3481 patch->old_name, patch->old_mode);
3482 if (status < 0)
3483 return status;
3484 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3486 * There is no way to apply subproject
3487 * patch without looking at the index.
3488 * NEEDSWORK: shouldn't this be flagged
3489 * as an error???
3491 free_fragment_list(patch->fragments);
3492 patch->fragments = NULL;
3493 } else if (status) {
3494 return error(_("failed to read %s"), patch->old_name);
3498 img = strbuf_detach(&buf, &len);
3499 prepare_image(image, img, len, !patch->is_binary);
3500 return 0;
3503 static int resolve_to(struct image *image, const struct object_id *result_id)
3505 unsigned long size;
3506 enum object_type type;
3508 clear_image(image);
3510 image->buf = repo_read_object_file(the_repository, result_id, &type,
3511 &size);
3512 if (!image->buf || type != OBJ_BLOB)
3513 die("unable to read blob object %s", oid_to_hex(result_id));
3514 image->len = size;
3516 return 0;
3519 static int three_way_merge(struct apply_state *state,
3520 struct image *image,
3521 char *path,
3522 const struct object_id *base,
3523 const struct object_id *ours,
3524 const struct object_id *theirs)
3526 mmfile_t base_file, our_file, their_file;
3527 mmbuffer_t result = { NULL };
3528 enum ll_merge_result status;
3530 /* resolve trivial cases first */
3531 if (oideq(base, ours))
3532 return resolve_to(image, theirs);
3533 else if (oideq(base, theirs) || oideq(ours, theirs))
3534 return resolve_to(image, ours);
3536 read_mmblob(&base_file, base);
3537 read_mmblob(&our_file, ours);
3538 read_mmblob(&their_file, theirs);
3539 status = ll_merge(&result, path,
3540 &base_file, "base",
3541 &our_file, "ours",
3542 &their_file, "theirs",
3543 state->repo->index,
3544 NULL);
3545 if (status == LL_MERGE_BINARY_CONFLICT)
3546 warning("Cannot merge binary files: %s (%s vs. %s)",
3547 path, "ours", "theirs");
3548 free(base_file.ptr);
3549 free(our_file.ptr);
3550 free(their_file.ptr);
3551 if (status < 0 || !result.ptr) {
3552 free(result.ptr);
3553 return -1;
3555 clear_image(image);
3556 image->buf = result.ptr;
3557 image->len = result.size;
3559 return status;
3563 * When directly falling back to add/add three-way merge, we read from
3564 * the current contents of the new_name. In no cases other than that
3565 * this function will be called.
3567 static int load_current(struct apply_state *state,
3568 struct image *image,
3569 struct patch *patch)
3571 struct strbuf buf = STRBUF_INIT;
3572 int status, pos;
3573 size_t len;
3574 char *img;
3575 struct stat st;
3576 struct cache_entry *ce;
3577 char *name = patch->new_name;
3578 unsigned mode = patch->new_mode;
3580 if (!patch->is_new)
3581 BUG("patch to %s is not a creation", patch->old_name);
3583 pos = index_name_pos(state->repo->index, name, strlen(name));
3584 if (pos < 0)
3585 return error(_("%s: does not exist in index"), name);
3586 ce = state->repo->index->cache[pos];
3587 if (lstat(name, &st)) {
3588 if (errno != ENOENT)
3589 return error_errno("%s", name);
3590 if (checkout_target(state->repo->index, ce, &st))
3591 return -1;
3593 if (verify_index_match(state, ce, &st))
3594 return error(_("%s: does not match index"), name);
3596 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3597 if (status < 0)
3598 return status;
3599 else if (status)
3600 return -1;
3601 img = strbuf_detach(&buf, &len);
3602 prepare_image(image, img, len, !patch->is_binary);
3603 return 0;
3606 static int try_threeway(struct apply_state *state,
3607 struct image *image,
3608 struct patch *patch,
3609 struct stat *st,
3610 const struct cache_entry *ce)
3612 struct object_id pre_oid, post_oid, our_oid;
3613 struct strbuf buf = STRBUF_INIT;
3614 size_t len;
3615 int status;
3616 char *img;
3617 struct image tmp_image;
3619 /* No point falling back to 3-way merge in these cases */
3620 if (patch->is_delete ||
3621 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3622 (patch->is_new && !patch->direct_to_threeway) ||
3623 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3624 return -1;
3626 /* Preimage the patch was prepared for */
3627 if (patch->is_new)
3628 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3629 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3630 read_blob_object(&buf, &pre_oid, patch->old_mode))
3631 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3633 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3634 fprintf(stderr, _("Performing three-way merge...\n"));
3636 img = strbuf_detach(&buf, &len);
3637 prepare_image(&tmp_image, img, len, 1);
3638 /* Apply the patch to get the post image */
3639 if (apply_fragments(state, &tmp_image, patch) < 0) {
3640 clear_image(&tmp_image);
3641 return -1;
3643 /* post_oid is theirs */
3644 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3645 clear_image(&tmp_image);
3647 /* our_oid is ours */
3648 if (patch->is_new) {
3649 if (load_current(state, &tmp_image, patch))
3650 return error(_("cannot read the current contents of '%s'"),
3651 patch->new_name);
3652 } else {
3653 if (load_preimage(state, &tmp_image, patch, st, ce))
3654 return error(_("cannot read the current contents of '%s'"),
3655 patch->old_name);
3657 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3658 clear_image(&tmp_image);
3660 /* in-core three-way merge between post and our using pre as base */
3661 status = three_way_merge(state, image, patch->new_name,
3662 &pre_oid, &our_oid, &post_oid);
3663 if (status < 0) {
3664 if (state->apply_verbosity > verbosity_silent)
3665 fprintf(stderr,
3666 _("Failed to perform three-way merge...\n"));
3667 return status;
3670 if (status) {
3671 patch->conflicted_threeway = 1;
3672 if (patch->is_new)
3673 oidclr(&patch->threeway_stage[0]);
3674 else
3675 oidcpy(&patch->threeway_stage[0], &pre_oid);
3676 oidcpy(&patch->threeway_stage[1], &our_oid);
3677 oidcpy(&patch->threeway_stage[2], &post_oid);
3678 if (state->apply_verbosity > verbosity_silent)
3679 fprintf(stderr,
3680 _("Applied patch to '%s' with conflicts.\n"),
3681 patch->new_name);
3682 } else {
3683 if (state->apply_verbosity > verbosity_silent)
3684 fprintf(stderr,
3685 _("Applied patch to '%s' cleanly.\n"),
3686 patch->new_name);
3688 return 0;
3691 static int apply_data(struct apply_state *state, struct patch *patch,
3692 struct stat *st, const struct cache_entry *ce)
3694 struct image image;
3696 if (load_preimage(state, &image, patch, st, ce) < 0)
3697 return -1;
3699 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3700 if (state->apply_verbosity > verbosity_silent &&
3701 state->threeway && !patch->direct_to_threeway)
3702 fprintf(stderr, _("Falling back to direct application...\n"));
3704 /* Note: with --reject, apply_fragments() returns 0 */
3705 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0)
3706 return -1;
3708 patch->result = image.buf;
3709 patch->resultsize = image.len;
3710 add_to_fn_table(state, patch);
3711 free(image.line_allocated);
3713 if (0 < patch->is_delete && patch->resultsize)
3714 return error(_("removal patch leaves file contents"));
3716 return 0;
3720 * If "patch" that we are looking at modifies or deletes what we have,
3721 * we would want it not to lose any local modification we have, either
3722 * in the working tree or in the index.
3724 * This also decides if a non-git patch is a creation patch or a
3725 * modification to an existing empty file. We do not check the state
3726 * of the current tree for a creation patch in this function; the caller
3727 * check_patch() separately makes sure (and errors out otherwise) that
3728 * the path the patch creates does not exist in the current tree.
3730 static int check_preimage(struct apply_state *state,
3731 struct patch *patch,
3732 struct cache_entry **ce,
3733 struct stat *st)
3735 const char *old_name = patch->old_name;
3736 struct patch *previous = NULL;
3737 int stat_ret = 0, status;
3738 unsigned st_mode = 0;
3740 if (!old_name)
3741 return 0;
3743 assert(patch->is_new <= 0);
3744 previous = previous_patch(state, patch, &status);
3746 if (status)
3747 return error(_("path %s has been renamed/deleted"), old_name);
3748 if (previous) {
3749 st_mode = previous->new_mode;
3750 } else if (!state->cached) {
3751 stat_ret = lstat(old_name, st);
3752 if (stat_ret && errno != ENOENT)
3753 return error_errno("%s", old_name);
3756 if (state->check_index && !previous) {
3757 int pos = index_name_pos(state->repo->index, old_name,
3758 strlen(old_name));
3759 if (pos < 0) {
3760 if (patch->is_new < 0)
3761 goto is_new;
3762 return error(_("%s: does not exist in index"), old_name);
3764 *ce = state->repo->index->cache[pos];
3765 if (stat_ret < 0) {
3766 if (checkout_target(state->repo->index, *ce, st))
3767 return -1;
3769 if (!state->cached && verify_index_match(state, *ce, st))
3770 return error(_("%s: does not match index"), old_name);
3771 if (state->cached)
3772 st_mode = (*ce)->ce_mode;
3773 } else if (stat_ret < 0) {
3774 if (patch->is_new < 0)
3775 goto is_new;
3776 return error_errno("%s", old_name);
3779 if (!state->cached && !previous)
3780 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3782 if (patch->is_new < 0)
3783 patch->is_new = 0;
3784 if (!patch->old_mode)
3785 patch->old_mode = st_mode;
3786 if ((st_mode ^ patch->old_mode) & S_IFMT)
3787 return error(_("%s: wrong type"), old_name);
3788 if (st_mode != patch->old_mode)
3789 warning(_("%s has type %o, expected %o"),
3790 old_name, st_mode, patch->old_mode);
3791 if (!patch->new_mode && !patch->is_delete)
3792 patch->new_mode = st_mode;
3793 return 0;
3795 is_new:
3796 patch->is_new = 1;
3797 patch->is_delete = 0;
3798 FREE_AND_NULL(patch->old_name);
3799 return 0;
3803 #define EXISTS_IN_INDEX 1
3804 #define EXISTS_IN_WORKTREE 2
3805 #define EXISTS_IN_INDEX_AS_ITA 3
3807 static int check_to_create(struct apply_state *state,
3808 const char *new_name,
3809 int ok_if_exists)
3811 struct stat nst;
3813 if (state->check_index && (!ok_if_exists || !state->cached)) {
3814 int pos;
3816 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3817 if (pos >= 0) {
3818 struct cache_entry *ce = state->repo->index->cache[pos];
3820 /* allow ITA, as they do not yet exist in the index */
3821 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3822 return EXISTS_IN_INDEX;
3824 /* ITA entries can never match working tree files */
3825 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3826 return EXISTS_IN_INDEX_AS_ITA;
3830 if (state->cached)
3831 return 0;
3833 if (!lstat(new_name, &nst)) {
3834 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3835 return 0;
3837 * A leading component of new_name might be a symlink
3838 * that is going to be removed with this patch, but
3839 * still pointing at somewhere that has the path.
3840 * In such a case, path "new_name" does not exist as
3841 * far as git is concerned.
3843 if (has_symlink_leading_path(new_name, strlen(new_name)))
3844 return 0;
3846 return EXISTS_IN_WORKTREE;
3847 } else if (!is_missing_file_error(errno)) {
3848 return error_errno("%s", new_name);
3850 return 0;
3853 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3855 for ( ; patch; patch = patch->next) {
3856 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3857 (patch->is_rename || patch->is_delete))
3858 /* the symlink at patch->old_name is removed */
3859 strset_add(&state->removed_symlinks, patch->old_name);
3861 if (patch->new_name && S_ISLNK(patch->new_mode))
3862 /* the symlink at patch->new_name is created or remains */
3863 strset_add(&state->kept_symlinks, patch->new_name);
3867 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3869 do {
3870 while (--name->len && name->buf[name->len] != '/')
3871 ; /* scan backwards */
3872 if (!name->len)
3873 break;
3874 name->buf[name->len] = '\0';
3875 if (strset_contains(&state->kept_symlinks, name->buf))
3876 return 1;
3877 if (strset_contains(&state->removed_symlinks, name->buf))
3879 * This cannot be "return 0", because we may
3880 * see a new one created at a higher level.
3882 continue;
3884 /* otherwise, check the preimage */
3885 if (state->check_index) {
3886 struct cache_entry *ce;
3888 ce = index_file_exists(state->repo->index, name->buf,
3889 name->len, ignore_case);
3890 if (ce && S_ISLNK(ce->ce_mode))
3891 return 1;
3892 } else {
3893 struct stat st;
3894 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3895 return 1;
3897 } while (1);
3898 return 0;
3901 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3903 int ret;
3904 struct strbuf name = STRBUF_INIT;
3906 assert(*name_ != '\0');
3907 strbuf_addstr(&name, name_);
3908 ret = path_is_beyond_symlink_1(state, &name);
3909 strbuf_release(&name);
3911 return ret;
3914 static int check_unsafe_path(struct patch *patch)
3916 const char *old_name = NULL;
3917 const char *new_name = NULL;
3918 if (patch->is_delete)
3919 old_name = patch->old_name;
3920 else if (!patch->is_new && !patch->is_copy)
3921 old_name = patch->old_name;
3922 if (!patch->is_delete)
3923 new_name = patch->new_name;
3925 if (old_name && !verify_path(old_name, patch->old_mode))
3926 return error(_("invalid path '%s'"), old_name);
3927 if (new_name && !verify_path(new_name, patch->new_mode))
3928 return error(_("invalid path '%s'"), new_name);
3929 return 0;
3933 * Check and apply the patch in-core; leave the result in patch->result
3934 * for the caller to write it out to the final destination.
3936 static int check_patch(struct apply_state *state, struct patch *patch)
3938 struct stat st;
3939 const char *old_name = patch->old_name;
3940 const char *new_name = patch->new_name;
3941 const char *name = old_name ? old_name : new_name;
3942 struct cache_entry *ce = NULL;
3943 struct patch *tpatch;
3944 int ok_if_exists;
3945 int status;
3947 patch->rejected = 1; /* we will drop this after we succeed */
3949 status = check_preimage(state, patch, &ce, &st);
3950 if (status)
3951 return status;
3952 old_name = patch->old_name;
3955 * A type-change diff is always split into a patch to delete
3956 * old, immediately followed by a patch to create new (see
3957 * diff.c::run_diff()); in such a case it is Ok that the entry
3958 * to be deleted by the previous patch is still in the working
3959 * tree and in the index.
3961 * A patch to swap-rename between A and B would first rename A
3962 * to B and then rename B to A. While applying the first one,
3963 * the presence of B should not stop A from getting renamed to
3964 * B; ask to_be_deleted() about the later rename. Removal of
3965 * B and rename from A to B is handled the same way by asking
3966 * was_deleted().
3968 if ((tpatch = in_fn_table(state, new_name)) &&
3969 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3970 ok_if_exists = 1;
3971 else
3972 ok_if_exists = 0;
3974 if (new_name &&
3975 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3976 int err = check_to_create(state, new_name, ok_if_exists);
3978 if (err && state->threeway) {
3979 patch->direct_to_threeway = 1;
3980 } else switch (err) {
3981 case 0:
3982 break; /* happy */
3983 case EXISTS_IN_INDEX:
3984 return error(_("%s: already exists in index"), new_name);
3985 case EXISTS_IN_INDEX_AS_ITA:
3986 return error(_("%s: does not match index"), new_name);
3987 case EXISTS_IN_WORKTREE:
3988 return error(_("%s: already exists in working directory"),
3989 new_name);
3990 default:
3991 return err;
3994 if (!patch->new_mode) {
3995 if (0 < patch->is_new)
3996 patch->new_mode = S_IFREG | 0644;
3997 else
3998 patch->new_mode = patch->old_mode;
4002 if (new_name && old_name) {
4003 int same = !strcmp(old_name, new_name);
4004 if (!patch->new_mode)
4005 patch->new_mode = patch->old_mode;
4006 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4007 if (same)
4008 return error(_("new mode (%o) of %s does not "
4009 "match old mode (%o)"),
4010 patch->new_mode, new_name,
4011 patch->old_mode);
4012 else
4013 return error(_("new mode (%o) of %s does not "
4014 "match old mode (%o) of %s"),
4015 patch->new_mode, new_name,
4016 patch->old_mode, old_name);
4020 if (!state->unsafe_paths && check_unsafe_path(patch))
4021 return -128;
4024 * An attempt to read from or delete a path that is beyond a
4025 * symbolic link will be prevented by load_patch_target() that
4026 * is called at the beginning of apply_data() so we do not
4027 * have to worry about a patch marked with "is_delete" bit
4028 * here. We however need to make sure that the patch result
4029 * is not deposited to a path that is beyond a symbolic link
4030 * here.
4032 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4033 return error(_("affected file '%s' is beyond a symbolic link"),
4034 patch->new_name);
4036 if (apply_data(state, patch, &st, ce) < 0)
4037 return error(_("%s: patch does not apply"), name);
4038 patch->rejected = 0;
4039 return 0;
4042 static int check_patch_list(struct apply_state *state, struct patch *patch)
4044 int err = 0;
4046 prepare_symlink_changes(state, patch);
4047 prepare_fn_table(state, patch);
4048 while (patch) {
4049 int res;
4050 if (state->apply_verbosity > verbosity_normal)
4051 say_patch_name(stderr,
4052 _("Checking patch %s..."), patch);
4053 res = check_patch(state, patch);
4054 if (res == -128)
4055 return -128;
4056 err |= res;
4057 patch = patch->next;
4059 return err;
4062 static int read_apply_cache(struct apply_state *state)
4064 if (state->index_file)
4065 return read_index_from(state->repo->index, state->index_file,
4066 get_git_dir());
4067 else
4068 return repo_read_index(state->repo);
4071 /* This function tries to read the object name from the current index */
4072 static int get_current_oid(struct apply_state *state, const char *path,
4073 struct object_id *oid)
4075 int pos;
4077 if (read_apply_cache(state) < 0)
4078 return -1;
4079 pos = index_name_pos(state->repo->index, path, strlen(path));
4080 if (pos < 0)
4081 return -1;
4082 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4083 return 0;
4086 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4089 * A usable gitlink patch has only one fragment (hunk) that looks like:
4090 * @@ -1 +1 @@
4091 * -Subproject commit <old sha1>
4092 * +Subproject commit <new sha1>
4093 * or
4094 * @@ -1 +0,0 @@
4095 * -Subproject commit <old sha1>
4096 * for a removal patch.
4098 struct fragment *hunk = p->fragments;
4099 static const char heading[] = "-Subproject commit ";
4100 char *preimage;
4102 if (/* does the patch have only one hunk? */
4103 hunk && !hunk->next &&
4104 /* is its preimage one line? */
4105 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4106 /* does preimage begin with the heading? */
4107 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4108 starts_with(++preimage, heading) &&
4109 /* does it record full SHA-1? */
4110 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4111 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4112 /* does the abbreviated name on the index line agree with it? */
4113 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4114 return 0; /* it all looks fine */
4116 /* we may have full object name on the index line */
4117 return get_oid_hex(p->old_oid_prefix, oid);
4120 /* Build an index that contains just the files needed for a 3way merge */
4121 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4123 struct patch *patch;
4124 struct index_state result = INDEX_STATE_INIT(state->repo);
4125 struct lock_file lock = LOCK_INIT;
4126 int res;
4128 /* Once we start supporting the reverse patch, it may be
4129 * worth showing the new sha1 prefix, but until then...
4131 for (patch = list; patch; patch = patch->next) {
4132 struct object_id oid;
4133 struct cache_entry *ce;
4134 const char *name;
4136 name = patch->old_name ? patch->old_name : patch->new_name;
4137 if (0 < patch->is_new)
4138 continue;
4140 if (S_ISGITLINK(patch->old_mode)) {
4141 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4142 ; /* ok, the textual part looks sane */
4143 else
4144 return error(_("sha1 information is lacking or "
4145 "useless for submodule %s"), name);
4146 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4147 ; /* ok */
4148 } else if (!patch->lines_added && !patch->lines_deleted) {
4149 /* mode-only change: update the current */
4150 if (get_current_oid(state, patch->old_name, &oid))
4151 return error(_("mode change for %s, which is not "
4152 "in current HEAD"), name);
4153 } else
4154 return error(_("sha1 information is lacking or useless "
4155 "(%s)."), name);
4157 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4158 if (!ce)
4159 return error(_("make_cache_entry failed for path '%s'"),
4160 name);
4161 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4162 discard_cache_entry(ce);
4163 return error(_("could not add %s to temporary index"),
4164 name);
4168 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4169 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4170 discard_index(&result);
4172 if (res)
4173 return error(_("could not write temporary index to %s"),
4174 state->fake_ancestor);
4176 return 0;
4179 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4181 int files, adds, dels;
4183 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4184 files++;
4185 adds += patch->lines_added;
4186 dels += patch->lines_deleted;
4187 show_stats(state, patch);
4190 print_stat_summary(stdout, files, adds, dels);
4193 static void numstat_patch_list(struct apply_state *state,
4194 struct patch *patch)
4196 for ( ; patch; patch = patch->next) {
4197 const char *name;
4198 name = patch->new_name ? patch->new_name : patch->old_name;
4199 if (patch->is_binary)
4200 printf("-\t-\t");
4201 else
4202 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4203 write_name_quoted(name, stdout, state->line_termination);
4207 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4209 if (mode)
4210 printf(" %s mode %06o %s\n", newdelete, mode, name);
4211 else
4212 printf(" %s %s\n", newdelete, name);
4215 static void show_mode_change(struct patch *p, int show_name)
4217 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4218 if (show_name)
4219 printf(" mode change %06o => %06o %s\n",
4220 p->old_mode, p->new_mode, p->new_name);
4221 else
4222 printf(" mode change %06o => %06o\n",
4223 p->old_mode, p->new_mode);
4227 static void show_rename_copy(struct patch *p)
4229 const char *renamecopy = p->is_rename ? "rename" : "copy";
4230 const char *old_name, *new_name;
4232 /* Find common prefix */
4233 old_name = p->old_name;
4234 new_name = p->new_name;
4235 while (1) {
4236 const char *slash_old, *slash_new;
4237 slash_old = strchr(old_name, '/');
4238 slash_new = strchr(new_name, '/');
4239 if (!slash_old ||
4240 !slash_new ||
4241 slash_old - old_name != slash_new - new_name ||
4242 memcmp(old_name, new_name, slash_new - new_name))
4243 break;
4244 old_name = slash_old + 1;
4245 new_name = slash_new + 1;
4247 /* p->old_name through old_name is the common prefix, and old_name and
4248 * new_name through the end of names are renames
4250 if (old_name != p->old_name)
4251 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4252 (int)(old_name - p->old_name), p->old_name,
4253 old_name, new_name, p->score);
4254 else
4255 printf(" %s %s => %s (%d%%)\n", renamecopy,
4256 p->old_name, p->new_name, p->score);
4257 show_mode_change(p, 0);
4260 static void summary_patch_list(struct patch *patch)
4262 struct patch *p;
4264 for (p = patch; p; p = p->next) {
4265 if (p->is_new)
4266 show_file_mode_name("create", p->new_mode, p->new_name);
4267 else if (p->is_delete)
4268 show_file_mode_name("delete", p->old_mode, p->old_name);
4269 else {
4270 if (p->is_rename || p->is_copy)
4271 show_rename_copy(p);
4272 else {
4273 if (p->score) {
4274 printf(" rewrite %s (%d%%)\n",
4275 p->new_name, p->score);
4276 show_mode_change(p, 0);
4278 else
4279 show_mode_change(p, 1);
4285 static void patch_stats(struct apply_state *state, struct patch *patch)
4287 int lines = patch->lines_added + patch->lines_deleted;
4289 if (lines > state->max_change)
4290 state->max_change = lines;
4291 if (patch->old_name) {
4292 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4293 if (!len)
4294 len = strlen(patch->old_name);
4295 if (len > state->max_len)
4296 state->max_len = len;
4298 if (patch->new_name) {
4299 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4300 if (!len)
4301 len = strlen(patch->new_name);
4302 if (len > state->max_len)
4303 state->max_len = len;
4307 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4309 if (state->update_index && !state->ita_only) {
4310 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4311 return error(_("unable to remove %s from index"), patch->old_name);
4313 if (!state->cached) {
4314 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4315 remove_path(patch->old_name);
4318 return 0;
4321 static int add_index_file(struct apply_state *state,
4322 const char *path,
4323 unsigned mode,
4324 void *buf,
4325 unsigned long size)
4327 struct stat st;
4328 struct cache_entry *ce;
4329 int namelen = strlen(path);
4331 ce = make_empty_cache_entry(state->repo->index, namelen);
4332 memcpy(ce->name, path, namelen);
4333 ce->ce_mode = create_ce_mode(mode);
4334 ce->ce_flags = create_ce_flags(0);
4335 ce->ce_namelen = namelen;
4336 if (state->ita_only) {
4337 ce->ce_flags |= CE_INTENT_TO_ADD;
4338 set_object_name_for_intent_to_add_entry(ce);
4339 } else if (S_ISGITLINK(mode)) {
4340 const char *s;
4342 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4343 get_oid_hex(s, &ce->oid)) {
4344 discard_cache_entry(ce);
4345 return error(_("corrupt patch for submodule %s"), path);
4347 } else {
4348 if (!state->cached) {
4349 if (lstat(path, &st) < 0) {
4350 discard_cache_entry(ce);
4351 return error_errno(_("unable to stat newly "
4352 "created file '%s'"),
4353 path);
4355 fill_stat_cache_info(state->repo->index, ce, &st);
4357 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4358 discard_cache_entry(ce);
4359 return error(_("unable to create backing store "
4360 "for newly created file %s"), path);
4363 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4364 discard_cache_entry(ce);
4365 return error(_("unable to add cache entry for %s"), path);
4368 return 0;
4372 * Returns:
4373 * -1 if an unrecoverable error happened
4374 * 0 if everything went well
4375 * 1 if a recoverable error happened
4377 static int try_create_file(struct apply_state *state, const char *path,
4378 unsigned int mode, const char *buf,
4379 unsigned long size)
4381 int fd, res;
4382 struct strbuf nbuf = STRBUF_INIT;
4384 if (S_ISGITLINK(mode)) {
4385 struct stat st;
4386 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4387 return 0;
4388 return !!mkdir(path, 0777);
4391 if (has_symlinks && S_ISLNK(mode))
4392 /* Although buf:size is counted string, it also is NUL
4393 * terminated.
4395 return !!symlink(buf, path);
4397 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4398 if (fd < 0)
4399 return 1;
4401 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4402 size = nbuf.len;
4403 buf = nbuf.buf;
4406 res = write_in_full(fd, buf, size) < 0;
4407 if (res)
4408 error_errno(_("failed to write to '%s'"), path);
4409 strbuf_release(&nbuf);
4411 if (close(fd) < 0 && !res)
4412 return error_errno(_("closing file '%s'"), path);
4414 return res ? -1 : 0;
4418 * We optimistically assume that the directories exist,
4419 * which is true 99% of the time anyway. If they don't,
4420 * we create them and try again.
4422 * Returns:
4423 * -1 on error
4424 * 0 otherwise
4426 static int create_one_file(struct apply_state *state,
4427 char *path,
4428 unsigned mode,
4429 const char *buf,
4430 unsigned long size)
4432 int res;
4434 if (state->cached)
4435 return 0;
4438 * We already try to detect whether files are beyond a symlink in our
4439 * up-front checks. But in the case where symlinks are created by any
4440 * of the intermediate hunks it can happen that our up-front checks
4441 * didn't yet see the symlink, but at the point of arriving here there
4442 * in fact is one. We thus repeat the check for symlinks here.
4444 * Note that this does not make the up-front check obsolete as the
4445 * failure mode is different:
4447 * - The up-front checks cause us to abort before we have written
4448 * anything into the working directory. So when we exit this way the
4449 * working directory remains clean.
4451 * - The checks here happen in the middle of the action where we have
4452 * already started to apply the patch. The end result will be a dirty
4453 * working directory.
4455 * Ideally, we should update the up-front checks to catch what would
4456 * happen when we apply the patch before we damage the working tree.
4457 * We have all the information necessary to do so. But for now, as a
4458 * part of embargoed security work, having this check would serve as a
4459 * reasonable first step.
4461 if (path_is_beyond_symlink(state, path))
4462 return error(_("affected file '%s' is beyond a symbolic link"), path);
4464 res = try_create_file(state, path, mode, buf, size);
4465 if (res < 0)
4466 return -1;
4467 if (!res)
4468 return 0;
4470 if (errno == ENOENT) {
4471 if (safe_create_leading_directories_no_share(path))
4472 return 0;
4473 res = try_create_file(state, path, mode, buf, size);
4474 if (res < 0)
4475 return -1;
4476 if (!res)
4477 return 0;
4480 if (errno == EEXIST || errno == EACCES) {
4481 /* We may be trying to create a file where a directory
4482 * used to be.
4484 struct stat st;
4485 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4486 errno = EEXIST;
4489 if (errno == EEXIST) {
4490 unsigned int nr = getpid();
4492 for (;;) {
4493 char newpath[PATH_MAX];
4494 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4495 res = try_create_file(state, newpath, mode, buf, size);
4496 if (res < 0)
4497 return -1;
4498 if (!res) {
4499 if (!rename(newpath, path))
4500 return 0;
4501 unlink_or_warn(newpath);
4502 break;
4504 if (errno != EEXIST)
4505 break;
4506 ++nr;
4509 return error_errno(_("unable to write file '%s' mode %o"),
4510 path, mode);
4513 static int add_conflicted_stages_file(struct apply_state *state,
4514 struct patch *patch)
4516 int stage, namelen;
4517 unsigned mode;
4518 struct cache_entry *ce;
4520 if (!state->update_index)
4521 return 0;
4522 namelen = strlen(patch->new_name);
4523 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4525 remove_file_from_index(state->repo->index, patch->new_name);
4526 for (stage = 1; stage < 4; stage++) {
4527 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4528 continue;
4529 ce = make_empty_cache_entry(state->repo->index, namelen);
4530 memcpy(ce->name, patch->new_name, namelen);
4531 ce->ce_mode = create_ce_mode(mode);
4532 ce->ce_flags = create_ce_flags(stage);
4533 ce->ce_namelen = namelen;
4534 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4535 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4536 discard_cache_entry(ce);
4537 return error(_("unable to add cache entry for %s"),
4538 patch->new_name);
4542 return 0;
4545 static int create_file(struct apply_state *state, struct patch *patch)
4547 char *path = patch->new_name;
4548 unsigned mode = patch->new_mode;
4549 unsigned long size = patch->resultsize;
4550 char *buf = patch->result;
4552 if (!mode)
4553 mode = S_IFREG | 0644;
4554 if (create_one_file(state, path, mode, buf, size))
4555 return -1;
4557 if (patch->conflicted_threeway)
4558 return add_conflicted_stages_file(state, patch);
4559 else if (state->update_index)
4560 return add_index_file(state, path, mode, buf, size);
4561 return 0;
4564 /* phase zero is to remove, phase one is to create */
4565 static int write_out_one_result(struct apply_state *state,
4566 struct patch *patch,
4567 int phase)
4569 if (patch->is_delete > 0) {
4570 if (phase == 0)
4571 return remove_file(state, patch, 1);
4572 return 0;
4574 if (patch->is_new > 0 || patch->is_copy) {
4575 if (phase == 1)
4576 return create_file(state, patch);
4577 return 0;
4580 * Rename or modification boils down to the same
4581 * thing: remove the old, write the new
4583 if (phase == 0)
4584 return remove_file(state, patch, patch->is_rename);
4585 if (phase == 1)
4586 return create_file(state, patch);
4587 return 0;
4590 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4592 FILE *rej;
4593 char namebuf[PATH_MAX];
4594 struct fragment *frag;
4595 int fd, cnt = 0;
4596 struct strbuf sb = STRBUF_INIT;
4598 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4599 if (!frag->rejected)
4600 continue;
4601 cnt++;
4604 if (!cnt) {
4605 if (state->apply_verbosity > verbosity_normal)
4606 say_patch_name(stderr,
4607 _("Applied patch %s cleanly."), patch);
4608 return 0;
4611 /* This should not happen, because a removal patch that leaves
4612 * contents are marked "rejected" at the patch level.
4614 if (!patch->new_name)
4615 die(_("internal error"));
4617 /* Say this even without --verbose */
4618 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4619 "Applying patch %%s with %d rejects...",
4620 cnt),
4621 cnt);
4622 if (state->apply_verbosity > verbosity_silent)
4623 say_patch_name(stderr, sb.buf, patch);
4624 strbuf_release(&sb);
4626 cnt = strlen(patch->new_name);
4627 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4628 cnt = ARRAY_SIZE(namebuf) - 5;
4629 warning(_("truncating .rej filename to %.*s.rej"),
4630 cnt - 1, patch->new_name);
4632 memcpy(namebuf, patch->new_name, cnt);
4633 memcpy(namebuf + cnt, ".rej", 5);
4635 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4636 if (fd < 0) {
4637 if (errno != EEXIST)
4638 return error_errno(_("cannot open %s"), namebuf);
4639 if (unlink(namebuf))
4640 return error_errno(_("cannot unlink '%s'"), namebuf);
4641 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4642 if (fd < 0)
4643 return error_errno(_("cannot open %s"), namebuf);
4645 rej = fdopen(fd, "w");
4646 if (!rej)
4647 return error_errno(_("cannot open %s"), namebuf);
4649 /* Normal git tools never deal with .rej, so do not pretend
4650 * this is a git patch by saying --git or giving extended
4651 * headers. While at it, maybe please "kompare" that wants
4652 * the trailing TAB and some garbage at the end of line ;-).
4654 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4655 patch->new_name, patch->new_name);
4656 for (cnt = 1, frag = patch->fragments;
4657 frag;
4658 cnt++, frag = frag->next) {
4659 if (!frag->rejected) {
4660 if (state->apply_verbosity > verbosity_silent)
4661 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4662 continue;
4664 if (state->apply_verbosity > verbosity_silent)
4665 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4666 fprintf(rej, "%.*s", frag->size, frag->patch);
4667 if (frag->patch[frag->size-1] != '\n')
4668 fputc('\n', rej);
4670 fclose(rej);
4671 return -1;
4675 * Returns:
4676 * -1 if an error happened
4677 * 0 if the patch applied cleanly
4678 * 1 if the patch did not apply cleanly
4680 static int write_out_results(struct apply_state *state, struct patch *list)
4682 int phase;
4683 int errs = 0;
4684 struct patch *l;
4685 struct string_list cpath = STRING_LIST_INIT_DUP;
4687 for (phase = 0; phase < 2; phase++) {
4688 l = list;
4689 while (l) {
4690 if (l->rejected)
4691 errs = 1;
4692 else {
4693 if (write_out_one_result(state, l, phase)) {
4694 string_list_clear(&cpath, 0);
4695 return -1;
4697 if (phase == 1) {
4698 if (write_out_one_reject(state, l))
4699 errs = 1;
4700 if (l->conflicted_threeway) {
4701 string_list_append(&cpath, l->new_name);
4702 errs = 1;
4706 l = l->next;
4710 if (cpath.nr) {
4711 struct string_list_item *item;
4713 string_list_sort(&cpath);
4714 if (state->apply_verbosity > verbosity_silent) {
4715 for_each_string_list_item(item, &cpath)
4716 fprintf(stderr, "U %s\n", item->string);
4718 string_list_clear(&cpath, 0);
4721 * rerere relies on the partially merged result being in the working
4722 * tree with conflict markers, but that isn't written with --cached.
4724 if (!state->cached)
4725 repo_rerere(state->repo, 0);
4728 return errs;
4732 * Try to apply a patch.
4734 * Returns:
4735 * -128 if a bad error happened (like patch unreadable)
4736 * -1 if patch did not apply and user cannot deal with it
4737 * 0 if the patch applied
4738 * 1 if the patch did not apply but user might fix it
4740 static int apply_patch(struct apply_state *state,
4741 int fd,
4742 const char *filename,
4743 int options)
4745 size_t offset;
4746 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4747 struct patch *list = NULL, **listp = &list;
4748 int skipped_patch = 0;
4749 int res = 0;
4750 int flush_attributes = 0;
4752 state->patch_input_file = filename;
4753 if (read_patch_file(&buf, fd) < 0)
4754 return -128;
4755 offset = 0;
4756 while (offset < buf.len) {
4757 struct patch *patch;
4758 int nr;
4760 CALLOC_ARRAY(patch, 1);
4761 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4762 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4763 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4764 if (nr < 0) {
4765 free_patch(patch);
4766 if (nr == -128) {
4767 res = -128;
4768 goto end;
4770 break;
4772 if (state->apply_in_reverse)
4773 reverse_patches(patch);
4774 if (use_patch(state, patch)) {
4775 patch_stats(state, patch);
4776 if (!list || !state->apply_in_reverse) {
4777 *listp = patch;
4778 listp = &patch->next;
4779 } else {
4780 patch->next = list;
4781 list = patch;
4784 if ((patch->new_name &&
4785 ends_with_path_components(patch->new_name,
4786 GITATTRIBUTES_FILE)) ||
4787 (patch->old_name &&
4788 ends_with_path_components(patch->old_name,
4789 GITATTRIBUTES_FILE)))
4790 flush_attributes = 1;
4792 else {
4793 if (state->apply_verbosity > verbosity_normal)
4794 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4795 free_patch(patch);
4796 skipped_patch++;
4798 offset += nr;
4801 if (!list && !skipped_patch) {
4802 if (!state->allow_empty) {
4803 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4804 res = -128;
4806 goto end;
4809 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4810 state->apply = 0;
4812 state->update_index = (state->check_index || state->ita_only) && state->apply;
4813 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4814 if (state->index_file)
4815 hold_lock_file_for_update(&state->lock_file,
4816 state->index_file,
4817 LOCK_DIE_ON_ERROR);
4818 else
4819 repo_hold_locked_index(state->repo, &state->lock_file,
4820 LOCK_DIE_ON_ERROR);
4823 if (state->check_index && read_apply_cache(state) < 0) {
4824 error(_("unable to read index file"));
4825 res = -128;
4826 goto end;
4829 if (state->check || state->apply) {
4830 int r = check_patch_list(state, list);
4831 if (r == -128) {
4832 res = -128;
4833 goto end;
4835 if (r < 0 && !state->apply_with_reject) {
4836 res = -1;
4837 goto end;
4841 if (state->apply) {
4842 int write_res = write_out_results(state, list);
4843 if (write_res < 0) {
4844 res = -128;
4845 goto end;
4847 if (write_res > 0) {
4848 /* with --3way, we still need to write the index out */
4849 res = state->apply_with_reject ? -1 : 1;
4850 goto end;
4854 if (state->fake_ancestor &&
4855 build_fake_ancestor(state, list)) {
4856 res = -128;
4857 goto end;
4860 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4861 stat_patch_list(state, list);
4863 if (state->numstat && state->apply_verbosity > verbosity_silent)
4864 numstat_patch_list(state, list);
4866 if (state->summary && state->apply_verbosity > verbosity_silent)
4867 summary_patch_list(list);
4869 if (flush_attributes)
4870 reset_parsed_attributes();
4871 end:
4872 free_patch_list(list);
4873 strbuf_release(&buf);
4874 string_list_clear(&state->fn_table, 0);
4875 return res;
4878 static int apply_option_parse_exclude(const struct option *opt,
4879 const char *arg, int unset)
4881 struct apply_state *state = opt->value;
4883 BUG_ON_OPT_NEG(unset);
4885 add_name_limit(state, arg, 1);
4886 return 0;
4889 static int apply_option_parse_include(const struct option *opt,
4890 const char *arg, int unset)
4892 struct apply_state *state = opt->value;
4894 BUG_ON_OPT_NEG(unset);
4896 add_name_limit(state, arg, 0);
4897 state->has_include = 1;
4898 return 0;
4901 static int apply_option_parse_p(const struct option *opt,
4902 const char *arg,
4903 int unset)
4905 struct apply_state *state = opt->value;
4907 BUG_ON_OPT_NEG(unset);
4909 state->p_value = atoi(arg);
4910 state->p_value_known = 1;
4911 return 0;
4914 static int apply_option_parse_space_change(const struct option *opt,
4915 const char *arg, int unset)
4917 struct apply_state *state = opt->value;
4919 BUG_ON_OPT_ARG(arg);
4921 if (unset)
4922 state->ws_ignore_action = ignore_ws_none;
4923 else
4924 state->ws_ignore_action = ignore_ws_change;
4925 return 0;
4928 static int apply_option_parse_whitespace(const struct option *opt,
4929 const char *arg, int unset)
4931 struct apply_state *state = opt->value;
4933 BUG_ON_OPT_NEG(unset);
4935 state->whitespace_option = arg;
4936 if (parse_whitespace_option(state, arg))
4937 return -1;
4938 return 0;
4941 static int apply_option_parse_directory(const struct option *opt,
4942 const char *arg, int unset)
4944 struct apply_state *state = opt->value;
4946 BUG_ON_OPT_NEG(unset);
4948 strbuf_reset(&state->root);
4949 strbuf_addstr(&state->root, arg);
4950 strbuf_complete(&state->root, '/');
4951 return 0;
4954 int apply_all_patches(struct apply_state *state,
4955 int argc,
4956 const char **argv,
4957 int options)
4959 int i;
4960 int res;
4961 int errs = 0;
4962 int read_stdin = 1;
4964 for (i = 0; i < argc; i++) {
4965 const char *arg = argv[i];
4966 char *to_free = NULL;
4967 int fd;
4969 if (!strcmp(arg, "-")) {
4970 res = apply_patch(state, 0, "<stdin>", options);
4971 if (res < 0)
4972 goto end;
4973 errs |= res;
4974 read_stdin = 0;
4975 continue;
4976 } else
4977 arg = to_free = prefix_filename(state->prefix, arg);
4979 fd = open(arg, O_RDONLY);
4980 if (fd < 0) {
4981 error(_("can't open patch '%s': %s"), arg, strerror(errno));
4982 res = -128;
4983 free(to_free);
4984 goto end;
4986 read_stdin = 0;
4987 set_default_whitespace_mode(state);
4988 res = apply_patch(state, fd, arg, options);
4989 close(fd);
4990 free(to_free);
4991 if (res < 0)
4992 goto end;
4993 errs |= res;
4995 set_default_whitespace_mode(state);
4996 if (read_stdin) {
4997 res = apply_patch(state, 0, "<stdin>", options);
4998 if (res < 0)
4999 goto end;
5000 errs |= res;
5003 if (state->whitespace_error) {
5004 if (state->squelch_whitespace_errors &&
5005 state->squelch_whitespace_errors < state->whitespace_error) {
5006 int squelched =
5007 state->whitespace_error - state->squelch_whitespace_errors;
5008 warning(Q_("squelched %d whitespace error",
5009 "squelched %d whitespace errors",
5010 squelched),
5011 squelched);
5013 if (state->ws_error_action == die_on_ws_error) {
5014 error(Q_("%d line adds whitespace errors.",
5015 "%d lines add whitespace errors.",
5016 state->whitespace_error),
5017 state->whitespace_error);
5018 res = -128;
5019 goto end;
5021 if (state->applied_after_fixing_ws && state->apply)
5022 warning(Q_("%d line applied after"
5023 " fixing whitespace errors.",
5024 "%d lines applied after"
5025 " fixing whitespace errors.",
5026 state->applied_after_fixing_ws),
5027 state->applied_after_fixing_ws);
5028 else if (state->whitespace_error)
5029 warning(Q_("%d line adds whitespace errors.",
5030 "%d lines add whitespace errors.",
5031 state->whitespace_error),
5032 state->whitespace_error);
5035 if (state->update_index) {
5036 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5037 if (res) {
5038 error(_("Unable to write new index file"));
5039 res = -128;
5040 goto end;
5044 res = !!errs;
5046 end:
5047 rollback_lock_file(&state->lock_file);
5049 if (state->apply_verbosity <= verbosity_silent) {
5050 set_error_routine(state->saved_error_routine);
5051 set_warn_routine(state->saved_warn_routine);
5054 if (res > -1)
5055 return res;
5056 return (res == -1 ? 1 : 128);
5059 int apply_parse_options(int argc, const char **argv,
5060 struct apply_state *state,
5061 int *force_apply, int *options,
5062 const char * const *apply_usage)
5064 struct option builtin_apply_options[] = {
5065 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5066 N_("don't apply changes matching the given path"),
5067 PARSE_OPT_NONEG, apply_option_parse_exclude),
5068 OPT_CALLBACK_F(0, "include", state, N_("path"),
5069 N_("apply changes matching the given path"),
5070 PARSE_OPT_NONEG, apply_option_parse_include),
5071 OPT_CALLBACK('p', NULL, state, N_("num"),
5072 N_("remove <num> leading slashes from traditional diff paths"),
5073 apply_option_parse_p),
5074 OPT_BOOL(0, "no-add", &state->no_add,
5075 N_("ignore additions made by the patch")),
5076 OPT_BOOL(0, "stat", &state->diffstat,
5077 N_("instead of applying the patch, output diffstat for the input")),
5078 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5079 OPT_NOOP_NOARG(0, "binary"),
5080 OPT_BOOL(0, "numstat", &state->numstat,
5081 N_("show number of added and deleted lines in decimal notation")),
5082 OPT_BOOL(0, "summary", &state->summary,
5083 N_("instead of applying the patch, output a summary for the input")),
5084 OPT_BOOL(0, "check", &state->check,
5085 N_("instead of applying the patch, see if the patch is applicable")),
5086 OPT_BOOL(0, "index", &state->check_index,
5087 N_("make sure the patch is applicable to the current index")),
5088 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5089 N_("mark new files with `git add --intent-to-add`")),
5090 OPT_BOOL(0, "cached", &state->cached,
5091 N_("apply a patch without touching the working tree")),
5092 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5093 N_("accept a patch that touches outside the working area"),
5094 PARSE_OPT_NOCOMPLETE),
5095 OPT_BOOL(0, "apply", force_apply,
5096 N_("also apply the patch (use with --stat/--summary/--check)")),
5097 OPT_BOOL('3', "3way", &state->threeway,
5098 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5099 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5100 N_("build a temporary index based on embedded index information")),
5101 /* Think twice before adding "--nul" synonym to this */
5102 OPT_SET_INT('z', NULL, &state->line_termination,
5103 N_("paths are separated with NUL character"), '\0'),
5104 OPT_INTEGER('C', NULL, &state->p_context,
5105 N_("ensure at least <n> lines of context match")),
5106 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5107 N_("detect new or modified lines that have whitespace errors"),
5108 apply_option_parse_whitespace),
5109 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5110 N_("ignore changes in whitespace when finding context"),
5111 PARSE_OPT_NOARG, apply_option_parse_space_change),
5112 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5113 N_("ignore changes in whitespace when finding context"),
5114 PARSE_OPT_NOARG, apply_option_parse_space_change),
5115 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5116 N_("apply the patch in reverse")),
5117 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5118 N_("don't expect at least one line of context")),
5119 OPT_BOOL(0, "reject", &state->apply_with_reject,
5120 N_("leave the rejected hunks in corresponding *.rej files")),
5121 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5122 N_("allow overlapping hunks")),
5123 OPT__VERBOSITY(&state->apply_verbosity),
5124 OPT_BIT(0, "inaccurate-eof", options,
5125 N_("tolerate incorrectly detected missing new-line at the end of file"),
5126 APPLY_OPT_INACCURATE_EOF),
5127 OPT_BIT(0, "recount", options,
5128 N_("do not trust the line counts in the hunk headers"),
5129 APPLY_OPT_RECOUNT),
5130 OPT_CALLBACK(0, "directory", state, N_("root"),
5131 N_("prepend <root> to all filenames"),
5132 apply_option_parse_directory),
5133 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5134 N_("don't return error for empty patches")),
5135 OPT_END()
5138 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);