Merge branch 'jt/p4-spell-re-with-raw-string'
[alt-git.git] / apply.c
blob7608e3301ca0727dcfec0a579ed46390afa41ab0
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 "git-compat-util.h"
11 #include "abspath.h"
12 #include "base85.h"
13 #include "config.h"
14 #include "object-store-ll.h"
15 #include "delta.h"
16 #include "diff.h"
17 #include "dir.h"
18 #include "environment.h"
19 #include "gettext.h"
20 #include "hex.h"
21 #include "xdiff-interface.h"
22 #include "merge-ll.h"
23 #include "lockfile.h"
24 #include "name-hash.h"
25 #include "object-name.h"
26 #include "object-file.h"
27 #include "parse-options.h"
28 #include "path.h"
29 #include "quote.h"
30 #include "read-cache.h"
31 #include "rerere.h"
32 #include "apply.h"
33 #include "entry.h"
34 #include "setup.h"
35 #include "symlinks.h"
36 #include "wildmatch.h"
37 #include "ws.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)
415 return error_errno(_("failed to read patch"));
416 else if (sb->len >= MAX_APPLY_SIZE)
417 return error(_("patch too large"));
419 * Make sure that we have some slop in the buffer
420 * so that we can do speculative "memcmp" etc, and
421 * see to it that it is NUL-filled.
423 strbuf_grow(sb, SLOP);
424 memset(sb->buf + sb->len, 0, SLOP);
425 return 0;
428 static unsigned long linelen(const char *buffer, unsigned long size)
430 unsigned long len = 0;
431 while (size--) {
432 len++;
433 if (*buffer++ == '\n')
434 break;
436 return len;
439 static int is_dev_null(const char *str)
441 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
444 #define TERM_SPACE 1
445 #define TERM_TAB 2
447 static int name_terminate(int c, int terminate)
449 if (c == ' ' && !(terminate & TERM_SPACE))
450 return 0;
451 if (c == '\t' && !(terminate & TERM_TAB))
452 return 0;
454 return 1;
457 /* remove double slashes to make --index work with such filenames */
458 static char *squash_slash(char *name)
460 int i = 0, j = 0;
462 if (!name)
463 return NULL;
465 while (name[i]) {
466 if ((name[j++] = name[i++]) == '/')
467 while (name[i] == '/')
468 i++;
470 name[j] = '\0';
471 return name;
474 static char *find_name_gnu(struct strbuf *root,
475 const char *line,
476 int p_value)
478 struct strbuf name = STRBUF_INIT;
479 char *cp;
482 * Proposed "new-style" GNU patch/diff format; see
483 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
485 if (unquote_c_style(&name, line, NULL)) {
486 strbuf_release(&name);
487 return NULL;
490 for (cp = name.buf; p_value; p_value--) {
491 cp = strchr(cp, '/');
492 if (!cp) {
493 strbuf_release(&name);
494 return NULL;
496 cp++;
499 strbuf_remove(&name, 0, cp - name.buf);
500 if (root->len)
501 strbuf_insert(&name, 0, root->buf, root->len);
502 return squash_slash(strbuf_detach(&name, NULL));
505 static size_t sane_tz_len(const char *line, size_t len)
507 const char *tz, *p;
509 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
510 return 0;
511 tz = line + len - strlen(" +0500");
513 if (tz[1] != '+' && tz[1] != '-')
514 return 0;
516 for (p = tz + 2; p != line + len; p++)
517 if (!isdigit(*p))
518 return 0;
520 return line + len - tz;
523 static size_t tz_with_colon_len(const char *line, size_t len)
525 const char *tz, *p;
527 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
528 return 0;
529 tz = line + len - strlen(" +08:00");
531 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
532 return 0;
533 p = tz + 2;
534 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
535 !isdigit(*p++) || !isdigit(*p++))
536 return 0;
538 return line + len - tz;
541 static size_t date_len(const char *line, size_t len)
543 const char *date, *p;
545 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
546 return 0;
547 p = date = line + len - strlen("72-02-05");
549 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
550 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
551 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
552 return 0;
554 if (date - line >= strlen("19") &&
555 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
556 date -= strlen("19");
558 return line + len - date;
561 static size_t short_time_len(const char *line, size_t len)
563 const char *time, *p;
565 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
566 return 0;
567 p = time = line + len - strlen(" 07:01:32");
569 /* Permit 1-digit hours? */
570 if (*p++ != ' ' ||
571 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
572 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
573 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
574 return 0;
576 return line + len - time;
579 static size_t fractional_time_len(const char *line, size_t len)
581 const char *p;
582 size_t n;
584 /* Expected format: 19:41:17.620000023 */
585 if (!len || !isdigit(line[len - 1]))
586 return 0;
587 p = line + len - 1;
589 /* Fractional seconds. */
590 while (p > line && isdigit(*p))
591 p--;
592 if (*p != '.')
593 return 0;
595 /* Hours, minutes, and whole seconds. */
596 n = short_time_len(line, p - line);
597 if (!n)
598 return 0;
600 return line + len - p + n;
603 static size_t trailing_spaces_len(const char *line, size_t len)
605 const char *p;
607 /* Expected format: ' ' x (1 or more) */
608 if (!len || line[len - 1] != ' ')
609 return 0;
611 p = line + len;
612 while (p != line) {
613 p--;
614 if (*p != ' ')
615 return line + len - (p + 1);
618 /* All spaces! */
619 return len;
622 static size_t diff_timestamp_len(const char *line, size_t len)
624 const char *end = line + len;
625 size_t n;
628 * Posix: 2010-07-05 19:41:17
629 * GNU: 2010-07-05 19:41:17.620000023 -0500
632 if (!isdigit(end[-1]))
633 return 0;
635 n = sane_tz_len(line, end - line);
636 if (!n)
637 n = tz_with_colon_len(line, end - line);
638 end -= n;
640 n = short_time_len(line, end - line);
641 if (!n)
642 n = fractional_time_len(line, end - line);
643 end -= n;
645 n = date_len(line, end - line);
646 if (!n) /* No date. Too bad. */
647 return 0;
648 end -= n;
650 if (end == line) /* No space before date. */
651 return 0;
652 if (end[-1] == '\t') { /* Success! */
653 end--;
654 return line + len - end;
656 if (end[-1] != ' ') /* No space before date. */
657 return 0;
659 /* Whitespace damage. */
660 end -= trailing_spaces_len(line, end - line);
661 return line + len - end;
664 static char *find_name_common(struct strbuf *root,
665 const char *line,
666 const char *def,
667 int p_value,
668 const char *end,
669 int terminate)
671 int len;
672 const char *start = NULL;
674 if (p_value == 0)
675 start = line;
676 while (line != end) {
677 char c = *line;
679 if (!end && isspace(c)) {
680 if (c == '\n')
681 break;
682 if (name_terminate(c, terminate))
683 break;
685 line++;
686 if (c == '/' && !--p_value)
687 start = line;
689 if (!start)
690 return squash_slash(xstrdup_or_null(def));
691 len = line - start;
692 if (!len)
693 return squash_slash(xstrdup_or_null(def));
696 * Generally we prefer the shorter name, especially
697 * if the other one is just a variation of that with
698 * something else tacked on to the end (ie "file.orig"
699 * or "file~").
701 if (def) {
702 int deflen = strlen(def);
703 if (deflen < len && !strncmp(start, def, deflen))
704 return squash_slash(xstrdup(def));
707 if (root->len) {
708 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
709 return squash_slash(ret);
712 return squash_slash(xmemdupz(start, len));
715 static char *find_name(struct strbuf *root,
716 const char *line,
717 char *def,
718 int p_value,
719 int terminate)
721 if (*line == '"') {
722 char *name = find_name_gnu(root, line, p_value);
723 if (name)
724 return name;
727 return find_name_common(root, line, def, p_value, NULL, terminate);
730 static char *find_name_traditional(struct strbuf *root,
731 const char *line,
732 char *def,
733 int p_value)
735 size_t len;
736 size_t date_len;
738 if (*line == '"') {
739 char *name = find_name_gnu(root, line, p_value);
740 if (name)
741 return name;
744 len = strchrnul(line, '\n') - line;
745 date_len = diff_timestamp_len(line, len);
746 if (!date_len)
747 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
748 len -= date_len;
750 return find_name_common(root, line, def, p_value, line + len, 0);
754 * Given the string after "--- " or "+++ ", guess the appropriate
755 * p_value for the given patch.
757 static int guess_p_value(struct apply_state *state, const char *nameline)
759 char *name, *cp;
760 int val = -1;
762 if (is_dev_null(nameline))
763 return -1;
764 name = find_name_traditional(&state->root, nameline, NULL, 0);
765 if (!name)
766 return -1;
767 cp = strchr(name, '/');
768 if (!cp)
769 val = 0;
770 else if (state->prefix) {
772 * Does it begin with "a/$our-prefix" and such? Then this is
773 * very likely to apply to our directory.
775 if (starts_with(name, state->prefix))
776 val = count_slashes(state->prefix);
777 else {
778 cp++;
779 if (starts_with(cp, state->prefix))
780 val = count_slashes(state->prefix) + 1;
783 free(name);
784 return val;
788 * Does the ---/+++ line have the POSIX timestamp after the last HT?
789 * GNU diff puts epoch there to signal a creation/deletion event. Is
790 * this such a timestamp?
792 static int has_epoch_timestamp(const char *nameline)
795 * We are only interested in epoch timestamp; any non-zero
796 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
797 * For the same reason, the date must be either 1969-12-31 or
798 * 1970-01-01, and the seconds part must be "00".
800 const char stamp_regexp[] =
801 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
803 "([-+][0-2][0-9]:?[0-5][0-9])\n";
804 const char *timestamp = NULL, *cp, *colon;
805 static regex_t *stamp;
806 regmatch_t m[10];
807 int zoneoffset, epoch_hour, hour, minute;
808 int status;
810 for (cp = nameline; *cp != '\n'; cp++) {
811 if (*cp == '\t')
812 timestamp = cp + 1;
814 if (!timestamp)
815 return 0;
818 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
819 * (west of GMT) or 1970-01-01 (east of GMT)
821 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
822 epoch_hour = 24;
823 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
824 epoch_hour = 0;
825 else
826 return 0;
828 if (!stamp) {
829 stamp = xmalloc(sizeof(*stamp));
830 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
831 warning(_("Cannot prepare timestamp regexp %s"),
832 stamp_regexp);
833 return 0;
837 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
838 if (status) {
839 if (status != REG_NOMATCH)
840 warning(_("regexec returned %d for input: %s"),
841 status, timestamp);
842 return 0;
845 hour = strtol(timestamp, NULL, 10);
846 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
848 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
849 if (*colon == ':')
850 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
851 else
852 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
853 if (timestamp[m[3].rm_so] == '-')
854 zoneoffset = -zoneoffset;
856 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
860 * Get the name etc info from the ---/+++ lines of a traditional patch header
862 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
863 * files, we can happily check the index for a match, but for creating a
864 * new file we should try to match whatever "patch" does. I have no idea.
866 static int parse_traditional_patch(struct apply_state *state,
867 const char *first,
868 const char *second,
869 struct patch *patch)
871 char *name;
873 first += 4; /* skip "--- " */
874 second += 4; /* skip "+++ " */
875 if (!state->p_value_known) {
876 int p, q;
877 p = guess_p_value(state, first);
878 q = guess_p_value(state, second);
879 if (p < 0) p = q;
880 if (0 <= p && p == q) {
881 state->p_value = p;
882 state->p_value_known = 1;
885 if (is_dev_null(first)) {
886 patch->is_new = 1;
887 patch->is_delete = 0;
888 name = find_name_traditional(&state->root, second, NULL, state->p_value);
889 patch->new_name = name;
890 } else if (is_dev_null(second)) {
891 patch->is_new = 0;
892 patch->is_delete = 1;
893 name = find_name_traditional(&state->root, first, NULL, state->p_value);
894 patch->old_name = name;
895 } else {
896 char *first_name;
897 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
898 name = find_name_traditional(&state->root, second, first_name, state->p_value);
899 free(first_name);
900 if (has_epoch_timestamp(first)) {
901 patch->is_new = 1;
902 patch->is_delete = 0;
903 patch->new_name = name;
904 } else if (has_epoch_timestamp(second)) {
905 patch->is_new = 0;
906 patch->is_delete = 1;
907 patch->old_name = name;
908 } else {
909 patch->old_name = name;
910 patch->new_name = xstrdup_or_null(name);
913 if (!name)
914 return error(_("unable to find filename in patch at line %d"), state->linenr);
916 return 0;
919 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
920 const char *line UNUSED,
921 struct patch *patch UNUSED)
923 return 1;
927 * We're anal about diff header consistency, to make
928 * sure that we don't end up having strange ambiguous
929 * patches floating around.
931 * As a result, gitdiff_{old|new}name() will check
932 * their names against any previous information, just
933 * to make sure..
935 #define DIFF_OLD_NAME 0
936 #define DIFF_NEW_NAME 1
938 static int gitdiff_verify_name(struct gitdiff_data *state,
939 const char *line,
940 int isnull,
941 char **name,
942 int side)
944 if (!*name && !isnull) {
945 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
946 return 0;
949 if (*name) {
950 char *another;
951 if (isnull)
952 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
953 *name, state->linenr);
954 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
955 if (!another || strcmp(another, *name)) {
956 free(another);
957 return error((side == DIFF_NEW_NAME) ?
958 _("git apply: bad git-diff - inconsistent new filename on line %d") :
959 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
961 free(another);
962 } else {
963 if (!is_dev_null(line))
964 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
967 return 0;
970 static int gitdiff_oldname(struct gitdiff_data *state,
971 const char *line,
972 struct patch *patch)
974 return gitdiff_verify_name(state, line,
975 patch->is_new, &patch->old_name,
976 DIFF_OLD_NAME);
979 static int gitdiff_newname(struct gitdiff_data *state,
980 const char *line,
981 struct patch *patch)
983 return gitdiff_verify_name(state, line,
984 patch->is_delete, &patch->new_name,
985 DIFF_NEW_NAME);
988 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
990 char *end;
991 *mode = strtoul(line, &end, 8);
992 if (end == line || !isspace(*end))
993 return error(_("invalid mode on line %d: %s"), linenr, line);
994 return 0;
997 static int gitdiff_oldmode(struct gitdiff_data *state,
998 const char *line,
999 struct patch *patch)
1001 return parse_mode_line(line, state->linenr, &patch->old_mode);
1004 static int gitdiff_newmode(struct gitdiff_data *state,
1005 const char *line,
1006 struct patch *patch)
1008 return parse_mode_line(line, state->linenr, &patch->new_mode);
1011 static int gitdiff_delete(struct gitdiff_data *state,
1012 const char *line,
1013 struct patch *patch)
1015 patch->is_delete = 1;
1016 free(patch->old_name);
1017 patch->old_name = xstrdup_or_null(patch->def_name);
1018 return gitdiff_oldmode(state, line, patch);
1021 static int gitdiff_newfile(struct gitdiff_data *state,
1022 const char *line,
1023 struct patch *patch)
1025 patch->is_new = 1;
1026 free(patch->new_name);
1027 patch->new_name = xstrdup_or_null(patch->def_name);
1028 return gitdiff_newmode(state, line, patch);
1031 static int gitdiff_copysrc(struct gitdiff_data *state,
1032 const char *line,
1033 struct patch *patch)
1035 patch->is_copy = 1;
1036 free(patch->old_name);
1037 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1038 return 0;
1041 static int gitdiff_copydst(struct gitdiff_data *state,
1042 const char *line,
1043 struct patch *patch)
1045 patch->is_copy = 1;
1046 free(patch->new_name);
1047 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1048 return 0;
1051 static int gitdiff_renamesrc(struct gitdiff_data *state,
1052 const char *line,
1053 struct patch *patch)
1055 patch->is_rename = 1;
1056 free(patch->old_name);
1057 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1058 return 0;
1061 static int gitdiff_renamedst(struct gitdiff_data *state,
1062 const char *line,
1063 struct patch *patch)
1065 patch->is_rename = 1;
1066 free(patch->new_name);
1067 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1068 return 0;
1071 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1072 const char *line,
1073 struct patch *patch)
1075 unsigned long val = strtoul(line, NULL, 10);
1076 if (val <= 100)
1077 patch->score = val;
1078 return 0;
1081 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1082 const char *line,
1083 struct patch *patch)
1085 unsigned long val = strtoul(line, NULL, 10);
1086 if (val <= 100)
1087 patch->score = val;
1088 return 0;
1091 static int gitdiff_index(struct gitdiff_data *state,
1092 const char *line,
1093 struct patch *patch)
1096 * index line is N hexadecimal, "..", N hexadecimal,
1097 * and optional space with octal mode.
1099 const char *ptr, *eol;
1100 int len;
1101 const unsigned hexsz = the_hash_algo->hexsz;
1103 ptr = strchr(line, '.');
1104 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1105 return 0;
1106 len = ptr - line;
1107 memcpy(patch->old_oid_prefix, line, len);
1108 patch->old_oid_prefix[len] = 0;
1110 line = ptr + 2;
1111 ptr = strchr(line, ' ');
1112 eol = strchrnul(line, '\n');
1114 if (!ptr || eol < ptr)
1115 ptr = eol;
1116 len = ptr - line;
1118 if (hexsz < len)
1119 return 0;
1120 memcpy(patch->new_oid_prefix, line, len);
1121 patch->new_oid_prefix[len] = 0;
1122 if (*ptr == ' ')
1123 return gitdiff_oldmode(state, ptr + 1, patch);
1124 return 0;
1128 * This is normal for a diff that doesn't change anything: we'll fall through
1129 * into the next diff. Tell the parser to break out.
1131 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1132 const char *line UNUSED,
1133 struct patch *patch UNUSED)
1135 return 1;
1139 * Skip p_value leading components from "line"; as we do not accept
1140 * absolute paths, return NULL in that case.
1142 static const char *skip_tree_prefix(int p_value,
1143 const char *line,
1144 int llen)
1146 int nslash;
1147 int i;
1149 if (!p_value)
1150 return (llen && line[0] == '/') ? NULL : line;
1152 nslash = p_value;
1153 for (i = 0; i < llen; i++) {
1154 int ch = line[i];
1155 if (ch == '/' && --nslash <= 0)
1156 return (i == 0) ? NULL : &line[i + 1];
1158 return NULL;
1162 * This is to extract the same name that appears on "diff --git"
1163 * line. We do not find and return anything if it is a rename
1164 * patch, and it is OK because we will find the name elsewhere.
1165 * We need to reliably find name only when it is mode-change only,
1166 * creation or deletion of an empty file. In any of these cases,
1167 * both sides are the same name under a/ and b/ respectively.
1169 static char *git_header_name(int p_value,
1170 const char *line,
1171 int llen)
1173 const char *name;
1174 const char *second = NULL;
1175 size_t len, line_len;
1177 line += strlen("diff --git ");
1178 llen -= strlen("diff --git ");
1180 if (*line == '"') {
1181 const char *cp;
1182 struct strbuf first = STRBUF_INIT;
1183 struct strbuf sp = STRBUF_INIT;
1185 if (unquote_c_style(&first, line, &second))
1186 goto free_and_fail1;
1188 /* strip the a/b prefix including trailing slash */
1189 cp = skip_tree_prefix(p_value, first.buf, first.len);
1190 if (!cp)
1191 goto free_and_fail1;
1192 strbuf_remove(&first, 0, cp - first.buf);
1195 * second points at one past closing dq of name.
1196 * find the second name.
1198 while ((second < line + llen) && isspace(*second))
1199 second++;
1201 if (line + llen <= second)
1202 goto free_and_fail1;
1203 if (*second == '"') {
1204 if (unquote_c_style(&sp, second, NULL))
1205 goto free_and_fail1;
1206 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1207 if (!cp)
1208 goto free_and_fail1;
1209 /* They must match, otherwise ignore */
1210 if (strcmp(cp, first.buf))
1211 goto free_and_fail1;
1212 strbuf_release(&sp);
1213 return strbuf_detach(&first, NULL);
1216 /* unquoted second */
1217 cp = skip_tree_prefix(p_value, second, line + llen - second);
1218 if (!cp)
1219 goto free_and_fail1;
1220 if (line + llen - cp != first.len ||
1221 memcmp(first.buf, cp, first.len))
1222 goto free_and_fail1;
1223 return strbuf_detach(&first, NULL);
1225 free_and_fail1:
1226 strbuf_release(&first);
1227 strbuf_release(&sp);
1228 return NULL;
1231 /* unquoted first name */
1232 name = skip_tree_prefix(p_value, line, llen);
1233 if (!name)
1234 return NULL;
1237 * since the first name is unquoted, a dq if exists must be
1238 * the beginning of the second name.
1240 for (second = name; second < line + llen; second++) {
1241 if (*second == '"') {
1242 struct strbuf sp = STRBUF_INIT;
1243 const char *np;
1245 if (unquote_c_style(&sp, second, NULL))
1246 goto free_and_fail2;
1248 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1249 if (!np)
1250 goto free_and_fail2;
1252 len = sp.buf + sp.len - np;
1253 if (len < second - name &&
1254 !strncmp(np, name, len) &&
1255 isspace(name[len])) {
1256 /* Good */
1257 strbuf_remove(&sp, 0, np - sp.buf);
1258 return strbuf_detach(&sp, NULL);
1261 free_and_fail2:
1262 strbuf_release(&sp);
1263 return NULL;
1268 * Accept a name only if it shows up twice, exactly the same
1269 * form.
1271 second = strchr(name, '\n');
1272 if (!second)
1273 return NULL;
1274 line_len = second - name;
1275 for (len = 0 ; ; len++) {
1276 switch (name[len]) {
1277 default:
1278 continue;
1279 case '\n':
1280 return NULL;
1281 case '\t': case ' ':
1283 * Is this the separator between the preimage
1284 * and the postimage pathname? Again, we are
1285 * only interested in the case where there is
1286 * no rename, as this is only to set def_name
1287 * and a rename patch has the names elsewhere
1288 * in an unambiguous form.
1290 if (!name[len + 1])
1291 return NULL; /* no postimage name */
1292 second = skip_tree_prefix(p_value, name + len + 1,
1293 line_len - (len + 1));
1294 if (!second)
1295 return NULL;
1297 * Does len bytes starting at "name" and "second"
1298 * (that are separated by one HT or SP we just
1299 * found) exactly match?
1301 if (second[len] == '\n' && !strncmp(name, second, len))
1302 return xmemdupz(name, len);
1307 static int check_header_line(int linenr, struct patch *patch)
1309 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1310 (patch->is_rename == 1) + (patch->is_copy == 1);
1311 if (extensions > 1)
1312 return error(_("inconsistent header lines %d and %d"),
1313 patch->extension_linenr, linenr);
1314 if (extensions && !patch->extension_linenr)
1315 patch->extension_linenr = linenr;
1316 return 0;
1319 int parse_git_diff_header(struct strbuf *root,
1320 int *linenr,
1321 int p_value,
1322 const char *line,
1323 int len,
1324 unsigned int size,
1325 struct patch *patch)
1327 unsigned long offset;
1328 struct gitdiff_data parse_hdr_state;
1330 /* A git diff has explicit new/delete information, so we don't guess */
1331 patch->is_new = 0;
1332 patch->is_delete = 0;
1335 * Some things may not have the old name in the
1336 * rest of the headers anywhere (pure mode changes,
1337 * or removing or adding empty files), so we get
1338 * the default name from the header.
1340 patch->def_name = git_header_name(p_value, line, len);
1341 if (patch->def_name && root->len) {
1342 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1343 free(patch->def_name);
1344 patch->def_name = s;
1347 line += len;
1348 size -= len;
1349 (*linenr)++;
1350 parse_hdr_state.root = root;
1351 parse_hdr_state.linenr = *linenr;
1352 parse_hdr_state.p_value = p_value;
1354 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1355 static const struct opentry {
1356 const char *str;
1357 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1358 } optable[] = {
1359 { "@@ -", gitdiff_hdrend },
1360 { "--- ", gitdiff_oldname },
1361 { "+++ ", gitdiff_newname },
1362 { "old mode ", gitdiff_oldmode },
1363 { "new mode ", gitdiff_newmode },
1364 { "deleted file mode ", gitdiff_delete },
1365 { "new file mode ", gitdiff_newfile },
1366 { "copy from ", gitdiff_copysrc },
1367 { "copy to ", gitdiff_copydst },
1368 { "rename old ", gitdiff_renamesrc },
1369 { "rename new ", gitdiff_renamedst },
1370 { "rename from ", gitdiff_renamesrc },
1371 { "rename to ", gitdiff_renamedst },
1372 { "similarity index ", gitdiff_similarity },
1373 { "dissimilarity index ", gitdiff_dissimilarity },
1374 { "index ", gitdiff_index },
1375 { "", gitdiff_unrecognized },
1377 int i;
1379 len = linelen(line, size);
1380 if (!len || line[len-1] != '\n')
1381 break;
1382 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1383 const struct opentry *p = optable + i;
1384 int oplen = strlen(p->str);
1385 int res;
1386 if (len < oplen || memcmp(p->str, line, oplen))
1387 continue;
1388 res = p->fn(&parse_hdr_state, line + oplen, patch);
1389 if (res < 0)
1390 return -1;
1391 if (check_header_line(*linenr, patch))
1392 return -1;
1393 if (res > 0)
1394 goto done;
1395 break;
1399 done:
1400 if (!patch->old_name && !patch->new_name) {
1401 if (!patch->def_name) {
1402 error(Q_("git diff header lacks filename information when removing "
1403 "%d leading pathname component (line %d)",
1404 "git diff header lacks filename information when removing "
1405 "%d leading pathname components (line %d)",
1406 parse_hdr_state.p_value),
1407 parse_hdr_state.p_value, *linenr);
1408 return -128;
1410 patch->old_name = xstrdup(patch->def_name);
1411 patch->new_name = xstrdup(patch->def_name);
1413 if ((!patch->new_name && !patch->is_delete) ||
1414 (!patch->old_name && !patch->is_new)) {
1415 error(_("git diff header lacks filename information "
1416 "(line %d)"), *linenr);
1417 return -128;
1419 patch->is_toplevel_relative = 1;
1420 return offset;
1423 static int parse_num(const char *line, unsigned long *p)
1425 char *ptr;
1427 if (!isdigit(*line))
1428 return 0;
1429 *p = strtoul(line, &ptr, 10);
1430 return ptr - line;
1433 static int parse_range(const char *line, int len, int offset, const char *expect,
1434 unsigned long *p1, unsigned long *p2)
1436 int digits, ex;
1438 if (offset < 0 || offset >= len)
1439 return -1;
1440 line += offset;
1441 len -= offset;
1443 digits = parse_num(line, p1);
1444 if (!digits)
1445 return -1;
1447 offset += digits;
1448 line += digits;
1449 len -= digits;
1451 *p2 = 1;
1452 if (*line == ',') {
1453 digits = parse_num(line+1, p2);
1454 if (!digits)
1455 return -1;
1457 offset += digits+1;
1458 line += digits+1;
1459 len -= digits+1;
1462 ex = strlen(expect);
1463 if (ex > len)
1464 return -1;
1465 if (memcmp(line, expect, ex))
1466 return -1;
1468 return offset + ex;
1471 static void recount_diff(const char *line, int size, struct fragment *fragment)
1473 int oldlines = 0, newlines = 0, ret = 0;
1475 if (size < 1) {
1476 warning("recount: ignore empty hunk");
1477 return;
1480 for (;;) {
1481 int len = linelen(line, size);
1482 size -= len;
1483 line += len;
1485 if (size < 1)
1486 break;
1488 switch (*line) {
1489 case ' ': case '\n':
1490 newlines++;
1491 /* fall through */
1492 case '-':
1493 oldlines++;
1494 continue;
1495 case '+':
1496 newlines++;
1497 continue;
1498 case '\\':
1499 continue;
1500 case '@':
1501 ret = size < 3 || !starts_with(line, "@@ ");
1502 break;
1503 case 'd':
1504 ret = size < 5 || !starts_with(line, "diff ");
1505 break;
1506 default:
1507 ret = -1;
1508 break;
1510 if (ret) {
1511 warning(_("recount: unexpected line: %.*s"),
1512 (int)linelen(line, size), line);
1513 return;
1515 break;
1517 fragment->oldlines = oldlines;
1518 fragment->newlines = newlines;
1522 * Parse a unified diff fragment header of the
1523 * form "@@ -a,b +c,d @@"
1525 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1527 int offset;
1529 if (!len || line[len-1] != '\n')
1530 return -1;
1532 /* Figure out the number of lines in a fragment */
1533 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1534 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1536 return offset;
1540 * Find file diff header
1542 * Returns:
1543 * -1 if no header was found
1544 * -128 in case of error
1545 * the size of the header in bytes (called "offset") otherwise
1547 static int find_header(struct apply_state *state,
1548 const char *line,
1549 unsigned long size,
1550 int *hdrsize,
1551 struct patch *patch)
1553 unsigned long offset, len;
1555 patch->is_toplevel_relative = 0;
1556 patch->is_rename = patch->is_copy = 0;
1557 patch->is_new = patch->is_delete = -1;
1558 patch->old_mode = patch->new_mode = 0;
1559 patch->old_name = patch->new_name = NULL;
1560 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1561 unsigned long nextlen;
1563 len = linelen(line, size);
1564 if (!len)
1565 break;
1567 /* Testing this early allows us to take a few shortcuts.. */
1568 if (len < 6)
1569 continue;
1572 * Make sure we don't find any unconnected patch fragments.
1573 * That's a sign that we didn't find a header, and that a
1574 * patch has become corrupted/broken up.
1576 if (!memcmp("@@ -", line, 4)) {
1577 struct fragment dummy;
1578 if (parse_fragment_header(line, len, &dummy) < 0)
1579 continue;
1580 error(_("patch fragment without header at line %d: %.*s"),
1581 state->linenr, (int)len-1, line);
1582 return -128;
1585 if (size < len + 6)
1586 break;
1589 * Git patch? It might not have a real patch, just a rename
1590 * or mode change, so we handle that specially
1592 if (!memcmp("diff --git ", line, 11)) {
1593 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1594 state->p_value, line, len,
1595 size, patch);
1596 if (git_hdr_len < 0)
1597 return -128;
1598 if (git_hdr_len <= len)
1599 continue;
1600 *hdrsize = git_hdr_len;
1601 return offset;
1604 /* --- followed by +++ ? */
1605 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1606 continue;
1609 * We only accept unified patches, so we want it to
1610 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1611 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1613 nextlen = linelen(line + len, size - len);
1614 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1615 continue;
1617 /* Ok, we'll consider it a patch */
1618 if (parse_traditional_patch(state, line, line+len, patch))
1619 return -128;
1620 *hdrsize = len + nextlen;
1621 state->linenr += 2;
1622 return offset;
1624 return -1;
1627 static void record_ws_error(struct apply_state *state,
1628 unsigned result,
1629 const char *line,
1630 int len,
1631 int linenr)
1633 char *err;
1635 if (!result)
1636 return;
1638 state->whitespace_error++;
1639 if (state->squelch_whitespace_errors &&
1640 state->squelch_whitespace_errors < state->whitespace_error)
1641 return;
1643 err = whitespace_error_string(result);
1644 if (state->apply_verbosity > verbosity_silent)
1645 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1646 state->patch_input_file, linenr, err, len, line);
1647 free(err);
1650 static void check_whitespace(struct apply_state *state,
1651 const char *line,
1652 int len,
1653 unsigned ws_rule)
1655 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1657 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1661 * Check if the patch has context lines with CRLF or
1662 * the patch wants to remove lines with CRLF.
1664 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1666 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1667 patch->ws_rule |= WS_CR_AT_EOL;
1668 patch->crlf_in_old = 1;
1674 * Parse a unified diff. Note that this really needs to parse each
1675 * fragment separately, since the only way to know the difference
1676 * between a "---" that is part of a patch, and a "---" that starts
1677 * the next patch is to look at the line counts..
1679 static int parse_fragment(struct apply_state *state,
1680 const char *line,
1681 unsigned long size,
1682 struct patch *patch,
1683 struct fragment *fragment)
1685 int added, deleted;
1686 int len = linelen(line, size), offset;
1687 unsigned long oldlines, newlines;
1688 unsigned long leading, trailing;
1690 offset = parse_fragment_header(line, len, fragment);
1691 if (offset < 0)
1692 return -1;
1693 if (offset > 0 && patch->recount)
1694 recount_diff(line + offset, size - offset, fragment);
1695 oldlines = fragment->oldlines;
1696 newlines = fragment->newlines;
1697 leading = 0;
1698 trailing = 0;
1700 /* Parse the thing.. */
1701 line += len;
1702 size -= len;
1703 state->linenr++;
1704 added = deleted = 0;
1705 for (offset = len;
1706 0 < size;
1707 offset += len, size -= len, line += len, state->linenr++) {
1708 if (!oldlines && !newlines)
1709 break;
1710 len = linelen(line, size);
1711 if (!len || line[len-1] != '\n')
1712 return -1;
1713 switch (*line) {
1714 default:
1715 return -1;
1716 case '\n': /* newer GNU diff, an empty context line */
1717 case ' ':
1718 oldlines--;
1719 newlines--;
1720 if (!deleted && !added)
1721 leading++;
1722 trailing++;
1723 check_old_for_crlf(patch, line, len);
1724 if (!state->apply_in_reverse &&
1725 state->ws_error_action == correct_ws_error)
1726 check_whitespace(state, line, len, patch->ws_rule);
1727 break;
1728 case '-':
1729 if (!state->apply_in_reverse)
1730 check_old_for_crlf(patch, line, len);
1731 if (state->apply_in_reverse &&
1732 state->ws_error_action != nowarn_ws_error)
1733 check_whitespace(state, line, len, patch->ws_rule);
1734 deleted++;
1735 oldlines--;
1736 trailing = 0;
1737 break;
1738 case '+':
1739 if (state->apply_in_reverse)
1740 check_old_for_crlf(patch, line, len);
1741 if (!state->apply_in_reverse &&
1742 state->ws_error_action != nowarn_ws_error)
1743 check_whitespace(state, line, len, patch->ws_rule);
1744 added++;
1745 newlines--;
1746 trailing = 0;
1747 break;
1750 * We allow "\ No newline at end of file". Depending
1751 * on locale settings when the patch was produced we
1752 * don't know what this line looks like. The only
1753 * thing we do know is that it begins with "\ ".
1754 * Checking for 12 is just for sanity check -- any
1755 * l10n of "\ No newline..." is at least that long.
1757 case '\\':
1758 if (len < 12 || memcmp(line, "\\ ", 2))
1759 return -1;
1760 break;
1763 if (oldlines || newlines)
1764 return -1;
1765 if (!patch->recount && !deleted && !added)
1766 return -1;
1768 fragment->leading = leading;
1769 fragment->trailing = trailing;
1772 * If a fragment ends with an incomplete line, we failed to include
1773 * it in the above loop because we hit oldlines == newlines == 0
1774 * before seeing it.
1776 if (12 < size && !memcmp(line, "\\ ", 2))
1777 offset += linelen(line, size);
1779 patch->lines_added += added;
1780 patch->lines_deleted += deleted;
1782 if (0 < patch->is_new && oldlines)
1783 return error(_("new file depends on old contents"));
1784 if (0 < patch->is_delete && newlines)
1785 return error(_("deleted file still has contents"));
1786 return offset;
1790 * We have seen "diff --git a/... b/..." header (or a traditional patch
1791 * header). Read hunks that belong to this patch into fragments and hang
1792 * them to the given patch structure.
1794 * The (fragment->patch, fragment->size) pair points into the memory given
1795 * by the caller, not a copy, when we return.
1797 * Returns:
1798 * -1 in case of error,
1799 * the number of bytes in the patch otherwise.
1801 static int parse_single_patch(struct apply_state *state,
1802 const char *line,
1803 unsigned long size,
1804 struct patch *patch)
1806 unsigned long offset = 0;
1807 unsigned long oldlines = 0, newlines = 0, context = 0;
1808 struct fragment **fragp = &patch->fragments;
1810 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1811 struct fragment *fragment;
1812 int len;
1814 CALLOC_ARRAY(fragment, 1);
1815 fragment->linenr = state->linenr;
1816 len = parse_fragment(state, line, size, patch, fragment);
1817 if (len <= 0) {
1818 free(fragment);
1819 return error(_("corrupt patch at line %d"), state->linenr);
1821 fragment->patch = line;
1822 fragment->size = len;
1823 oldlines += fragment->oldlines;
1824 newlines += fragment->newlines;
1825 context += fragment->leading + fragment->trailing;
1827 *fragp = fragment;
1828 fragp = &fragment->next;
1830 offset += len;
1831 line += len;
1832 size -= len;
1836 * If something was removed (i.e. we have old-lines) it cannot
1837 * be creation, and if something was added it cannot be
1838 * deletion. However, the reverse is not true; --unified=0
1839 * patches that only add are not necessarily creation even
1840 * though they do not have any old lines, and ones that only
1841 * delete are not necessarily deletion.
1843 * Unfortunately, a real creation/deletion patch do _not_ have
1844 * any context line by definition, so we cannot safely tell it
1845 * apart with --unified=0 insanity. At least if the patch has
1846 * more than one hunk it is not creation or deletion.
1848 if (patch->is_new < 0 &&
1849 (oldlines || (patch->fragments && patch->fragments->next)))
1850 patch->is_new = 0;
1851 if (patch->is_delete < 0 &&
1852 (newlines || (patch->fragments && patch->fragments->next)))
1853 patch->is_delete = 0;
1855 if (0 < patch->is_new && oldlines)
1856 return error(_("new file %s depends on old contents"), patch->new_name);
1857 if (0 < patch->is_delete && newlines)
1858 return error(_("deleted file %s still has contents"), patch->old_name);
1859 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1860 fprintf_ln(stderr,
1861 _("** warning: "
1862 "file %s becomes empty but is not deleted"),
1863 patch->new_name);
1865 return offset;
1868 static inline int metadata_changes(struct patch *patch)
1870 return patch->is_rename > 0 ||
1871 patch->is_copy > 0 ||
1872 patch->is_new > 0 ||
1873 patch->is_delete ||
1874 (patch->old_mode && patch->new_mode &&
1875 patch->old_mode != patch->new_mode);
1878 static char *inflate_it(const void *data, unsigned long size,
1879 unsigned long inflated_size)
1881 git_zstream stream;
1882 void *out;
1883 int st;
1885 memset(&stream, 0, sizeof(stream));
1887 stream.next_in = (unsigned char *)data;
1888 stream.avail_in = size;
1889 stream.next_out = out = xmalloc(inflated_size);
1890 stream.avail_out = inflated_size;
1891 git_inflate_init(&stream);
1892 st = git_inflate(&stream, Z_FINISH);
1893 git_inflate_end(&stream);
1894 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1895 free(out);
1896 return NULL;
1898 return out;
1902 * Read a binary hunk and return a new fragment; fragment->patch
1903 * points at an allocated memory that the caller must free, so
1904 * it is marked as "->free_patch = 1".
1906 static struct fragment *parse_binary_hunk(struct apply_state *state,
1907 char **buf_p,
1908 unsigned long *sz_p,
1909 int *status_p,
1910 int *used_p)
1913 * Expect a line that begins with binary patch method ("literal"
1914 * or "delta"), followed by the length of data before deflating.
1915 * a sequence of 'length-byte' followed by base-85 encoded data
1916 * should follow, terminated by a newline.
1918 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1919 * and we would limit the patch line to 66 characters,
1920 * so one line can fit up to 13 groups that would decode
1921 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1922 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1924 int llen, used;
1925 unsigned long size = *sz_p;
1926 char *buffer = *buf_p;
1927 int patch_method;
1928 unsigned long origlen;
1929 char *data = NULL;
1930 int hunk_size = 0;
1931 struct fragment *frag;
1933 llen = linelen(buffer, size);
1934 used = llen;
1936 *status_p = 0;
1938 if (starts_with(buffer, "delta ")) {
1939 patch_method = BINARY_DELTA_DEFLATED;
1940 origlen = strtoul(buffer + 6, NULL, 10);
1942 else if (starts_with(buffer, "literal ")) {
1943 patch_method = BINARY_LITERAL_DEFLATED;
1944 origlen = strtoul(buffer + 8, NULL, 10);
1946 else
1947 return NULL;
1949 state->linenr++;
1950 buffer += llen;
1951 size -= llen;
1952 while (1) {
1953 int byte_length, max_byte_length, newsize;
1954 llen = linelen(buffer, size);
1955 used += llen;
1956 state->linenr++;
1957 if (llen == 1) {
1958 /* consume the blank line */
1959 buffer++;
1960 size--;
1961 break;
1964 * Minimum line is "A00000\n" which is 7-byte long,
1965 * and the line length must be multiple of 5 plus 2.
1967 if ((llen < 7) || (llen-2) % 5)
1968 goto corrupt;
1969 max_byte_length = (llen - 2) / 5 * 4;
1970 byte_length = *buffer;
1971 if ('A' <= byte_length && byte_length <= 'Z')
1972 byte_length = byte_length - 'A' + 1;
1973 else if ('a' <= byte_length && byte_length <= 'z')
1974 byte_length = byte_length - 'a' + 27;
1975 else
1976 goto corrupt;
1977 /* if the input length was not multiple of 4, we would
1978 * have filler at the end but the filler should never
1979 * exceed 3 bytes
1981 if (max_byte_length < byte_length ||
1982 byte_length <= max_byte_length - 4)
1983 goto corrupt;
1984 newsize = hunk_size + byte_length;
1985 data = xrealloc(data, newsize);
1986 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1987 goto corrupt;
1988 hunk_size = newsize;
1989 buffer += llen;
1990 size -= llen;
1993 CALLOC_ARRAY(frag, 1);
1994 frag->patch = inflate_it(data, hunk_size, origlen);
1995 frag->free_patch = 1;
1996 if (!frag->patch)
1997 goto corrupt;
1998 free(data);
1999 frag->size = origlen;
2000 *buf_p = buffer;
2001 *sz_p = size;
2002 *used_p = used;
2003 frag->binary_patch_method = patch_method;
2004 return frag;
2006 corrupt:
2007 free(data);
2008 *status_p = -1;
2009 error(_("corrupt binary patch at line %d: %.*s"),
2010 state->linenr-1, llen-1, buffer);
2011 return NULL;
2015 * Returns:
2016 * -1 in case of error,
2017 * the length of the parsed binary patch otherwise
2019 static int parse_binary(struct apply_state *state,
2020 char *buffer,
2021 unsigned long size,
2022 struct patch *patch)
2025 * We have read "GIT binary patch\n"; what follows is a line
2026 * that says the patch method (currently, either "literal" or
2027 * "delta") and the length of data before deflating; a
2028 * sequence of 'length-byte' followed by base-85 encoded data
2029 * follows.
2031 * When a binary patch is reversible, there is another binary
2032 * hunk in the same format, starting with patch method (either
2033 * "literal" or "delta") with the length of data, and a sequence
2034 * of length-byte + base-85 encoded data, terminated with another
2035 * empty line. This data, when applied to the postimage, produces
2036 * the preimage.
2038 struct fragment *forward;
2039 struct fragment *reverse;
2040 int status;
2041 int used, used_1;
2043 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2044 if (!forward && !status)
2045 /* there has to be one hunk (forward hunk) */
2046 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2047 if (status)
2048 /* otherwise we already gave an error message */
2049 return status;
2051 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2052 if (reverse)
2053 used += used_1;
2054 else if (status) {
2056 * Not having reverse hunk is not an error, but having
2057 * a corrupt reverse hunk is.
2059 free((void*) forward->patch);
2060 free(forward);
2061 return status;
2063 forward->next = reverse;
2064 patch->fragments = forward;
2065 patch->is_binary = 1;
2066 return used;
2069 static void prefix_one(struct apply_state *state, char **name)
2071 char *old_name = *name;
2072 if (!old_name)
2073 return;
2074 *name = prefix_filename(state->prefix, *name);
2075 free(old_name);
2078 static void prefix_patch(struct apply_state *state, struct patch *p)
2080 if (!state->prefix || p->is_toplevel_relative)
2081 return;
2082 prefix_one(state, &p->new_name);
2083 prefix_one(state, &p->old_name);
2087 * include/exclude
2090 static void add_name_limit(struct apply_state *state,
2091 const char *name,
2092 int exclude)
2094 struct string_list_item *it;
2096 it = string_list_append(&state->limit_by_name, name);
2097 it->util = exclude ? NULL : (void *) 1;
2100 static int use_patch(struct apply_state *state, struct patch *p)
2102 const char *pathname = p->new_name ? p->new_name : p->old_name;
2103 int i;
2105 /* Paths outside are not touched regardless of "--include" */
2106 if (state->prefix && *state->prefix) {
2107 const char *rest;
2108 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2109 return 0;
2112 /* See if it matches any of exclude/include rule */
2113 for (i = 0; i < state->limit_by_name.nr; i++) {
2114 struct string_list_item *it = &state->limit_by_name.items[i];
2115 if (!wildmatch(it->string, pathname, 0))
2116 return (it->util != NULL);
2120 * If we had any include, a path that does not match any rule is
2121 * not used. Otherwise, we saw bunch of exclude rules (or none)
2122 * and such a path is used.
2124 return !state->has_include;
2128 * Read the patch text in "buffer" that extends for "size" bytes; stop
2129 * reading after seeing a single patch (i.e. changes to a single file).
2130 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2132 * Returns:
2133 * -1 if no header was found or parse_binary() failed,
2134 * -128 on another error,
2135 * the number of bytes consumed otherwise,
2136 * so that the caller can call us again for the next patch.
2138 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2140 int hdrsize, patchsize;
2141 int offset = find_header(state, buffer, size, &hdrsize, patch);
2143 if (offset < 0)
2144 return offset;
2146 prefix_patch(state, patch);
2148 if (!use_patch(state, patch))
2149 patch->ws_rule = 0;
2150 else if (patch->new_name)
2151 patch->ws_rule = whitespace_rule(state->repo->index,
2152 patch->new_name);
2153 else
2154 patch->ws_rule = whitespace_rule(state->repo->index,
2155 patch->old_name);
2157 patchsize = parse_single_patch(state,
2158 buffer + offset + hdrsize,
2159 size - offset - hdrsize,
2160 patch);
2162 if (patchsize < 0)
2163 return -128;
2165 if (!patchsize) {
2166 static const char git_binary[] = "GIT binary patch\n";
2167 int hd = hdrsize + offset;
2168 unsigned long llen = linelen(buffer + hd, size - hd);
2170 if (llen == sizeof(git_binary) - 1 &&
2171 !memcmp(git_binary, buffer + hd, llen)) {
2172 int used;
2173 state->linenr++;
2174 used = parse_binary(state, buffer + hd + llen,
2175 size - hd - llen, patch);
2176 if (used < 0)
2177 return -1;
2178 if (used)
2179 patchsize = used + llen;
2180 else
2181 patchsize = 0;
2183 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2184 static const char *binhdr[] = {
2185 "Binary files ",
2186 "Files ",
2187 NULL,
2189 int i;
2190 for (i = 0; binhdr[i]; i++) {
2191 int len = strlen(binhdr[i]);
2192 if (len < size - hd &&
2193 !memcmp(binhdr[i], buffer + hd, len)) {
2194 state->linenr++;
2195 patch->is_binary = 1;
2196 patchsize = llen;
2197 break;
2202 /* Empty patch cannot be applied if it is a text patch
2203 * without metadata change. A binary patch appears
2204 * empty to us here.
2206 if ((state->apply || state->check) &&
2207 (!patch->is_binary && !metadata_changes(patch))) {
2208 error(_("patch with only garbage at line %d"), state->linenr);
2209 return -128;
2213 return offset + hdrsize + patchsize;
2216 static void reverse_patches(struct patch *p)
2218 for (; p; p = p->next) {
2219 struct fragment *frag = p->fragments;
2221 SWAP(p->new_name, p->old_name);
2222 SWAP(p->new_mode, p->old_mode);
2223 SWAP(p->is_new, p->is_delete);
2224 SWAP(p->lines_added, p->lines_deleted);
2225 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2227 for (; frag; frag = frag->next) {
2228 SWAP(frag->newpos, frag->oldpos);
2229 SWAP(frag->newlines, frag->oldlines);
2234 static const char pluses[] =
2235 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2236 static const char minuses[]=
2237 "----------------------------------------------------------------------";
2239 static void show_stats(struct apply_state *state, struct patch *patch)
2241 struct strbuf qname = STRBUF_INIT;
2242 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2243 int max, add, del;
2245 quote_c_style(cp, &qname, NULL, 0);
2248 * "scale" the filename
2250 max = state->max_len;
2251 if (max > 50)
2252 max = 50;
2254 if (qname.len > max) {
2255 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2256 if (!cp)
2257 cp = qname.buf + qname.len + 3 - max;
2258 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2261 if (patch->is_binary) {
2262 printf(" %-*s | Bin\n", max, qname.buf);
2263 strbuf_release(&qname);
2264 return;
2267 printf(" %-*s |", max, qname.buf);
2268 strbuf_release(&qname);
2271 * scale the add/delete
2273 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2274 add = patch->lines_added;
2275 del = patch->lines_deleted;
2277 if (state->max_change > 0) {
2278 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2279 add = (add * max + state->max_change / 2) / state->max_change;
2280 del = total - add;
2282 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2283 add, pluses, del, minuses);
2286 static int read_old_data(struct stat *st, struct patch *patch,
2287 const char *path, struct strbuf *buf)
2289 int conv_flags = patch->crlf_in_old ?
2290 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2291 switch (st->st_mode & S_IFMT) {
2292 case S_IFLNK:
2293 if (strbuf_readlink(buf, path, st->st_size) < 0)
2294 return error(_("unable to read symlink %s"), path);
2295 return 0;
2296 case S_IFREG:
2297 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2298 return error(_("unable to open or read %s"), path);
2300 * "git apply" without "--index/--cached" should never look
2301 * at the index; the target file may not have been added to
2302 * the index yet, and we may not even be in any Git repository.
2303 * Pass NULL to convert_to_git() to stress this; the function
2304 * should never look at the index when explicit crlf option
2305 * is given.
2307 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2308 return 0;
2309 default:
2310 return -1;
2315 * Update the preimage, and the common lines in postimage,
2316 * from buffer buf of length len. If postlen is 0 the postimage
2317 * is updated in place, otherwise it's updated on a new buffer
2318 * of length postlen
2321 static void update_pre_post_images(struct image *preimage,
2322 struct image *postimage,
2323 char *buf,
2324 size_t len, size_t postlen)
2326 int i, ctx, reduced;
2327 char *new_buf, *old_buf, *fixed;
2328 struct image fixed_preimage;
2331 * Update the preimage with whitespace fixes. Note that we
2332 * are not losing preimage->buf -- apply_one_fragment() will
2333 * free "oldlines".
2335 prepare_image(&fixed_preimage, buf, len, 1);
2336 assert(postlen
2337 ? fixed_preimage.nr == preimage->nr
2338 : fixed_preimage.nr <= preimage->nr);
2339 for (i = 0; i < fixed_preimage.nr; i++)
2340 fixed_preimage.line[i].flag = preimage->line[i].flag;
2341 free(preimage->line_allocated);
2342 *preimage = fixed_preimage;
2345 * Adjust the common context lines in postimage. This can be
2346 * done in-place when we are shrinking it with whitespace
2347 * fixing, but needs a new buffer when ignoring whitespace or
2348 * expanding leading tabs to spaces.
2350 * We trust the caller to tell us if the update can be done
2351 * in place (postlen==0) or not.
2353 old_buf = postimage->buf;
2354 if (postlen)
2355 new_buf = postimage->buf = xmalloc(postlen);
2356 else
2357 new_buf = old_buf;
2358 fixed = preimage->buf;
2360 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2361 size_t l_len = postimage->line[i].len;
2362 if (!(postimage->line[i].flag & LINE_COMMON)) {
2363 /* an added line -- no counterparts in preimage */
2364 memmove(new_buf, old_buf, l_len);
2365 old_buf += l_len;
2366 new_buf += l_len;
2367 continue;
2370 /* a common context -- skip it in the original postimage */
2371 old_buf += l_len;
2373 /* and find the corresponding one in the fixed preimage */
2374 while (ctx < preimage->nr &&
2375 !(preimage->line[ctx].flag & LINE_COMMON)) {
2376 fixed += preimage->line[ctx].len;
2377 ctx++;
2381 * preimage is expected to run out, if the caller
2382 * fixed addition of trailing blank lines.
2384 if (preimage->nr <= ctx) {
2385 reduced++;
2386 continue;
2389 /* and copy it in, while fixing the line length */
2390 l_len = preimage->line[ctx].len;
2391 memcpy(new_buf, fixed, l_len);
2392 new_buf += l_len;
2393 fixed += l_len;
2394 postimage->line[i].len = l_len;
2395 ctx++;
2398 if (postlen
2399 ? postlen < new_buf - postimage->buf
2400 : postimage->len < new_buf - postimage->buf)
2401 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2402 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2404 /* Fix the length of the whole thing */
2405 postimage->len = new_buf - postimage->buf;
2406 postimage->nr -= reduced;
2409 static int line_by_line_fuzzy_match(struct image *img,
2410 struct image *preimage,
2411 struct image *postimage,
2412 unsigned long current,
2413 int current_lno,
2414 int preimage_limit)
2416 int i;
2417 size_t imgoff = 0;
2418 size_t preoff = 0;
2419 size_t postlen = postimage->len;
2420 size_t extra_chars;
2421 char *buf;
2422 char *preimage_eof;
2423 char *preimage_end;
2424 struct strbuf fixed;
2425 char *fixed_buf;
2426 size_t fixed_len;
2428 for (i = 0; i < preimage_limit; i++) {
2429 size_t prelen = preimage->line[i].len;
2430 size_t imglen = img->line[current_lno+i].len;
2432 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2433 preimage->buf + preoff, prelen))
2434 return 0;
2435 if (preimage->line[i].flag & LINE_COMMON)
2436 postlen += imglen - prelen;
2437 imgoff += imglen;
2438 preoff += prelen;
2442 * Ok, the preimage matches with whitespace fuzz.
2444 * imgoff now holds the true length of the target that
2445 * matches the preimage before the end of the file.
2447 * Count the number of characters in the preimage that fall
2448 * beyond the end of the file and make sure that all of them
2449 * are whitespace characters. (This can only happen if
2450 * we are removing blank lines at the end of the file.)
2452 buf = preimage_eof = preimage->buf + preoff;
2453 for ( ; i < preimage->nr; i++)
2454 preoff += preimage->line[i].len;
2455 preimage_end = preimage->buf + preoff;
2456 for ( ; buf < preimage_end; buf++)
2457 if (!isspace(*buf))
2458 return 0;
2461 * Update the preimage and the common postimage context
2462 * lines to use the same whitespace as the target.
2463 * If whitespace is missing in the target (i.e.
2464 * if the preimage extends beyond the end of the file),
2465 * use the whitespace from the preimage.
2467 extra_chars = preimage_end - preimage_eof;
2468 strbuf_init(&fixed, imgoff + extra_chars);
2469 strbuf_add(&fixed, img->buf + current, imgoff);
2470 strbuf_add(&fixed, preimage_eof, extra_chars);
2471 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2472 update_pre_post_images(preimage, postimage,
2473 fixed_buf, fixed_len, postlen);
2474 return 1;
2477 static int match_fragment(struct apply_state *state,
2478 struct image *img,
2479 struct image *preimage,
2480 struct image *postimage,
2481 unsigned long current,
2482 int current_lno,
2483 unsigned ws_rule,
2484 int match_beginning, int match_end)
2486 int i;
2487 char *fixed_buf, *buf, *orig, *target;
2488 struct strbuf fixed;
2489 size_t fixed_len, postlen;
2490 int preimage_limit;
2492 if (preimage->nr + current_lno <= img->nr) {
2494 * The hunk falls within the boundaries of img.
2496 preimage_limit = preimage->nr;
2497 if (match_end && (preimage->nr + current_lno != img->nr))
2498 return 0;
2499 } else if (state->ws_error_action == correct_ws_error &&
2500 (ws_rule & WS_BLANK_AT_EOF)) {
2502 * This hunk extends beyond the end of img, and we are
2503 * removing blank lines at the end of the file. This
2504 * many lines from the beginning of the preimage must
2505 * match with img, and the remainder of the preimage
2506 * must be blank.
2508 preimage_limit = img->nr - current_lno;
2509 } else {
2511 * The hunk extends beyond the end of the img and
2512 * we are not removing blanks at the end, so we
2513 * should reject the hunk at this position.
2515 return 0;
2518 if (match_beginning && current_lno)
2519 return 0;
2521 /* Quick hash check */
2522 for (i = 0; i < preimage_limit; i++)
2523 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2524 (preimage->line[i].hash != img->line[current_lno + i].hash))
2525 return 0;
2527 if (preimage_limit == preimage->nr) {
2529 * Do we have an exact match? If we were told to match
2530 * at the end, size must be exactly at current+fragsize,
2531 * otherwise current+fragsize must be still within the preimage,
2532 * and either case, the old piece should match the preimage
2533 * exactly.
2535 if ((match_end
2536 ? (current + preimage->len == img->len)
2537 : (current + preimage->len <= img->len)) &&
2538 !memcmp(img->buf + current, preimage->buf, preimage->len))
2539 return 1;
2540 } else {
2542 * The preimage extends beyond the end of img, so
2543 * there cannot be an exact match.
2545 * There must be one non-blank context line that match
2546 * a line before the end of img.
2548 char *buf_end;
2550 buf = preimage->buf;
2551 buf_end = buf;
2552 for (i = 0; i < preimage_limit; i++)
2553 buf_end += preimage->line[i].len;
2555 for ( ; buf < buf_end; buf++)
2556 if (!isspace(*buf))
2557 break;
2558 if (buf == buf_end)
2559 return 0;
2563 * No exact match. If we are ignoring whitespace, run a line-by-line
2564 * fuzzy matching. We collect all the line length information because
2565 * we need it to adjust whitespace if we match.
2567 if (state->ws_ignore_action == ignore_ws_change)
2568 return line_by_line_fuzzy_match(img, preimage, postimage,
2569 current, current_lno, preimage_limit);
2571 if (state->ws_error_action != correct_ws_error)
2572 return 0;
2575 * The hunk does not apply byte-by-byte, but the hash says
2576 * it might with whitespace fuzz. We weren't asked to
2577 * ignore whitespace, we were asked to correct whitespace
2578 * errors, so let's try matching after whitespace correction.
2580 * While checking the preimage against the target, whitespace
2581 * errors in both fixed, we count how large the corresponding
2582 * postimage needs to be. The postimage prepared by
2583 * apply_one_fragment() has whitespace errors fixed on added
2584 * lines already, but the common lines were propagated as-is,
2585 * which may become longer when their whitespace errors are
2586 * fixed.
2589 /* First count added lines in postimage */
2590 postlen = 0;
2591 for (i = 0; i < postimage->nr; i++) {
2592 if (!(postimage->line[i].flag & LINE_COMMON))
2593 postlen += postimage->line[i].len;
2597 * The preimage may extend beyond the end of the file,
2598 * but in this loop we will only handle the part of the
2599 * preimage that falls within the file.
2601 strbuf_init(&fixed, preimage->len + 1);
2602 orig = preimage->buf;
2603 target = img->buf + current;
2604 for (i = 0; i < preimage_limit; i++) {
2605 size_t oldlen = preimage->line[i].len;
2606 size_t tgtlen = img->line[current_lno + i].len;
2607 size_t fixstart = fixed.len;
2608 struct strbuf tgtfix;
2609 int match;
2611 /* Try fixing the line in the preimage */
2612 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2614 /* Try fixing the line in the target */
2615 strbuf_init(&tgtfix, tgtlen);
2616 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2619 * If they match, either the preimage was based on
2620 * a version before our tree fixed whitespace breakage,
2621 * or we are lacking a whitespace-fix patch the tree
2622 * the preimage was based on already had (i.e. target
2623 * has whitespace breakage, the preimage doesn't).
2624 * In either case, we are fixing the whitespace breakages
2625 * so we might as well take the fix together with their
2626 * real change.
2628 match = (tgtfix.len == fixed.len - fixstart &&
2629 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2630 fixed.len - fixstart));
2632 /* Add the length if this is common with the postimage */
2633 if (preimage->line[i].flag & LINE_COMMON)
2634 postlen += tgtfix.len;
2636 strbuf_release(&tgtfix);
2637 if (!match)
2638 goto unmatch_exit;
2640 orig += oldlen;
2641 target += tgtlen;
2646 * Now handle the lines in the preimage that falls beyond the
2647 * end of the file (if any). They will only match if they are
2648 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2649 * false).
2651 for ( ; i < preimage->nr; i++) {
2652 size_t fixstart = fixed.len; /* start of the fixed preimage */
2653 size_t oldlen = preimage->line[i].len;
2654 int j;
2656 /* Try fixing the line in the preimage */
2657 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2659 for (j = fixstart; j < fixed.len; j++)
2660 if (!isspace(fixed.buf[j]))
2661 goto unmatch_exit;
2663 orig += oldlen;
2667 * Yes, the preimage is based on an older version that still
2668 * has whitespace breakages unfixed, and fixing them makes the
2669 * hunk match. Update the context lines in the postimage.
2671 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2672 if (postlen < postimage->len)
2673 postlen = 0;
2674 update_pre_post_images(preimage, postimage,
2675 fixed_buf, fixed_len, postlen);
2676 return 1;
2678 unmatch_exit:
2679 strbuf_release(&fixed);
2680 return 0;
2683 static int find_pos(struct apply_state *state,
2684 struct image *img,
2685 struct image *preimage,
2686 struct image *postimage,
2687 int line,
2688 unsigned ws_rule,
2689 int match_beginning, int match_end)
2691 int i;
2692 unsigned long backwards, forwards, current;
2693 int backwards_lno, forwards_lno, current_lno;
2696 * When running with --allow-overlap, it is possible that a hunk is
2697 * seen that pretends to start at the beginning (but no longer does),
2698 * and that *still* needs to match the end. So trust `match_end` more
2699 * than `match_beginning`.
2701 if (state->allow_overlap && match_beginning && match_end &&
2702 img->nr - preimage->nr != 0)
2703 match_beginning = 0;
2706 * If match_beginning or match_end is specified, there is no
2707 * point starting from a wrong line that will never match and
2708 * wander around and wait for a match at the specified end.
2710 if (match_beginning)
2711 line = 0;
2712 else if (match_end)
2713 line = img->nr - preimage->nr;
2716 * Because the comparison is unsigned, the following test
2717 * will also take care of a negative line number that can
2718 * result when match_end and preimage is larger than the target.
2720 if ((size_t) line > img->nr)
2721 line = img->nr;
2723 current = 0;
2724 for (i = 0; i < line; i++)
2725 current += img->line[i].len;
2728 * There's probably some smart way to do this, but I'll leave
2729 * that to the smart and beautiful people. I'm simple and stupid.
2731 backwards = current;
2732 backwards_lno = line;
2733 forwards = current;
2734 forwards_lno = line;
2735 current_lno = line;
2737 for (i = 0; ; i++) {
2738 if (match_fragment(state, img, preimage, postimage,
2739 current, current_lno, ws_rule,
2740 match_beginning, match_end))
2741 return current_lno;
2743 again:
2744 if (backwards_lno == 0 && forwards_lno == img->nr)
2745 break;
2747 if (i & 1) {
2748 if (backwards_lno == 0) {
2749 i++;
2750 goto again;
2752 backwards_lno--;
2753 backwards -= img->line[backwards_lno].len;
2754 current = backwards;
2755 current_lno = backwards_lno;
2756 } else {
2757 if (forwards_lno == img->nr) {
2758 i++;
2759 goto again;
2761 forwards += img->line[forwards_lno].len;
2762 forwards_lno++;
2763 current = forwards;
2764 current_lno = forwards_lno;
2768 return -1;
2771 static void remove_first_line(struct image *img)
2773 img->buf += img->line[0].len;
2774 img->len -= img->line[0].len;
2775 img->line++;
2776 img->nr--;
2779 static void remove_last_line(struct image *img)
2781 img->len -= img->line[--img->nr].len;
2785 * The change from "preimage" and "postimage" has been found to
2786 * apply at applied_pos (counts in line numbers) in "img".
2787 * Update "img" to remove "preimage" and replace it with "postimage".
2789 static void update_image(struct apply_state *state,
2790 struct image *img,
2791 int applied_pos,
2792 struct image *preimage,
2793 struct image *postimage)
2796 * remove the copy of preimage at offset in img
2797 * and replace it with postimage
2799 int i, nr;
2800 size_t remove_count, insert_count, applied_at = 0;
2801 char *result;
2802 int preimage_limit;
2805 * If we are removing blank lines at the end of img,
2806 * the preimage may extend beyond the end.
2807 * If that is the case, we must be careful only to
2808 * remove the part of the preimage that falls within
2809 * the boundaries of img. Initialize preimage_limit
2810 * to the number of lines in the preimage that falls
2811 * within the boundaries.
2813 preimage_limit = preimage->nr;
2814 if (preimage_limit > img->nr - applied_pos)
2815 preimage_limit = img->nr - applied_pos;
2817 for (i = 0; i < applied_pos; i++)
2818 applied_at += img->line[i].len;
2820 remove_count = 0;
2821 for (i = 0; i < preimage_limit; i++)
2822 remove_count += img->line[applied_pos + i].len;
2823 insert_count = postimage->len;
2825 /* Adjust the contents */
2826 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2827 memcpy(result, img->buf, applied_at);
2828 memcpy(result + applied_at, postimage->buf, postimage->len);
2829 memcpy(result + applied_at + postimage->len,
2830 img->buf + (applied_at + remove_count),
2831 img->len - (applied_at + remove_count));
2832 free(img->buf);
2833 img->buf = result;
2834 img->len += insert_count - remove_count;
2835 result[img->len] = '\0';
2837 /* Adjust the line table */
2838 nr = img->nr + postimage->nr - preimage_limit;
2839 if (preimage_limit < postimage->nr) {
2841 * NOTE: this knows that we never call remove_first_line()
2842 * on anything other than pre/post image.
2844 REALLOC_ARRAY(img->line, nr);
2845 img->line_allocated = img->line;
2847 if (preimage_limit != postimage->nr)
2848 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2849 img->line + applied_pos + preimage_limit,
2850 img->nr - (applied_pos + preimage_limit));
2851 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2852 if (!state->allow_overlap)
2853 for (i = 0; i < postimage->nr; i++)
2854 img->line[applied_pos + i].flag |= LINE_PATCHED;
2855 img->nr = nr;
2859 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2860 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2861 * replace the part of "img" with "postimage" text.
2863 static int apply_one_fragment(struct apply_state *state,
2864 struct image *img, struct fragment *frag,
2865 int inaccurate_eof, unsigned ws_rule,
2866 int nth_fragment)
2868 int match_beginning, match_end;
2869 const char *patch = frag->patch;
2870 int size = frag->size;
2871 char *old, *oldlines;
2872 struct strbuf newlines;
2873 int new_blank_lines_at_end = 0;
2874 int found_new_blank_lines_at_end = 0;
2875 int hunk_linenr = frag->linenr;
2876 unsigned long leading, trailing;
2877 int pos, applied_pos;
2878 struct image preimage;
2879 struct image postimage;
2881 memset(&preimage, 0, sizeof(preimage));
2882 memset(&postimage, 0, sizeof(postimage));
2883 oldlines = xmalloc(size);
2884 strbuf_init(&newlines, size);
2886 old = oldlines;
2887 while (size > 0) {
2888 char first;
2889 int len = linelen(patch, size);
2890 int plen;
2891 int added_blank_line = 0;
2892 int is_blank_context = 0;
2893 size_t start;
2895 if (!len)
2896 break;
2899 * "plen" is how much of the line we should use for
2900 * the actual patch data. Normally we just remove the
2901 * first character on the line, but if the line is
2902 * followed by "\ No newline", then we also remove the
2903 * last one (which is the newline, of course).
2905 plen = len - 1;
2906 if (len < size && patch[len] == '\\')
2907 plen--;
2908 first = *patch;
2909 if (state->apply_in_reverse) {
2910 if (first == '-')
2911 first = '+';
2912 else if (first == '+')
2913 first = '-';
2916 switch (first) {
2917 case '\n':
2918 /* Newer GNU diff, empty context line */
2919 if (plen < 0)
2920 /* ... followed by '\No newline'; nothing */
2921 break;
2922 *old++ = '\n';
2923 strbuf_addch(&newlines, '\n');
2924 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2925 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2926 is_blank_context = 1;
2927 break;
2928 case ' ':
2929 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2930 ws_blank_line(patch + 1, plen))
2931 is_blank_context = 1;
2932 /* fallthrough */
2933 case '-':
2934 memcpy(old, patch + 1, plen);
2935 add_line_info(&preimage, old, plen,
2936 (first == ' ' ? LINE_COMMON : 0));
2937 old += plen;
2938 if (first == '-')
2939 break;
2940 /* fallthrough */
2941 case '+':
2942 /* --no-add does not add new lines */
2943 if (first == '+' && state->no_add)
2944 break;
2946 start = newlines.len;
2947 if (first != '+' ||
2948 !state->whitespace_error ||
2949 state->ws_error_action != correct_ws_error) {
2950 strbuf_add(&newlines, patch + 1, plen);
2952 else {
2953 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2955 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2956 (first == '+' ? 0 : LINE_COMMON));
2957 if (first == '+' &&
2958 (ws_rule & WS_BLANK_AT_EOF) &&
2959 ws_blank_line(patch + 1, plen))
2960 added_blank_line = 1;
2961 break;
2962 case '@': case '\\':
2963 /* Ignore it, we already handled it */
2964 break;
2965 default:
2966 if (state->apply_verbosity > verbosity_normal)
2967 error(_("invalid start of line: '%c'"), first);
2968 applied_pos = -1;
2969 goto out;
2971 if (added_blank_line) {
2972 if (!new_blank_lines_at_end)
2973 found_new_blank_lines_at_end = hunk_linenr;
2974 new_blank_lines_at_end++;
2976 else if (is_blank_context)
2978 else
2979 new_blank_lines_at_end = 0;
2980 patch += len;
2981 size -= len;
2982 hunk_linenr++;
2984 if (inaccurate_eof &&
2985 old > oldlines && old[-1] == '\n' &&
2986 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2987 old--;
2988 strbuf_setlen(&newlines, newlines.len - 1);
2989 preimage.line_allocated[preimage.nr - 1].len--;
2990 postimage.line_allocated[postimage.nr - 1].len--;
2993 leading = frag->leading;
2994 trailing = frag->trailing;
2997 * A hunk to change lines at the beginning would begin with
2998 * @@ -1,L +N,M @@
2999 * but we need to be careful. -U0 that inserts before the second
3000 * line also has this pattern.
3002 * And a hunk to add to an empty file would begin with
3003 * @@ -0,0 +N,M @@
3005 * In other words, a hunk that is (frag->oldpos <= 1) with or
3006 * without leading context must match at the beginning.
3008 match_beginning = (!frag->oldpos ||
3009 (frag->oldpos == 1 && !state->unidiff_zero));
3012 * A hunk without trailing lines must match at the end.
3013 * However, we simply cannot tell if a hunk must match end
3014 * from the lack of trailing lines if the patch was generated
3015 * with unidiff without any context.
3017 match_end = !state->unidiff_zero && !trailing;
3019 pos = frag->newpos ? (frag->newpos - 1) : 0;
3020 preimage.buf = oldlines;
3021 preimage.len = old - oldlines;
3022 postimage.buf = newlines.buf;
3023 postimage.len = newlines.len;
3024 preimage.line = preimage.line_allocated;
3025 postimage.line = postimage.line_allocated;
3027 for (;;) {
3029 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3030 ws_rule, match_beginning, match_end);
3032 if (applied_pos >= 0)
3033 break;
3035 /* Am I at my context limits? */
3036 if ((leading <= state->p_context) && (trailing <= state->p_context))
3037 break;
3038 if (match_beginning || match_end) {
3039 match_beginning = match_end = 0;
3040 continue;
3044 * Reduce the number of context lines; reduce both
3045 * leading and trailing if they are equal otherwise
3046 * just reduce the larger context.
3048 if (leading >= trailing) {
3049 remove_first_line(&preimage);
3050 remove_first_line(&postimage);
3051 pos--;
3052 leading--;
3054 if (trailing > leading) {
3055 remove_last_line(&preimage);
3056 remove_last_line(&postimage);
3057 trailing--;
3061 if (applied_pos >= 0) {
3062 if (new_blank_lines_at_end &&
3063 preimage.nr + applied_pos >= img->nr &&
3064 (ws_rule & WS_BLANK_AT_EOF) &&
3065 state->ws_error_action != nowarn_ws_error) {
3066 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3067 found_new_blank_lines_at_end);
3068 if (state->ws_error_action == correct_ws_error) {
3069 while (new_blank_lines_at_end--)
3070 remove_last_line(&postimage);
3073 * We would want to prevent write_out_results()
3074 * from taking place in apply_patch() that follows
3075 * the callchain led us here, which is:
3076 * apply_patch->check_patch_list->check_patch->
3077 * apply_data->apply_fragments->apply_one_fragment
3079 if (state->ws_error_action == die_on_ws_error)
3080 state->apply = 0;
3083 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3084 int offset = applied_pos - pos;
3085 if (state->apply_in_reverse)
3086 offset = 0 - offset;
3087 fprintf_ln(stderr,
3088 Q_("Hunk #%d succeeded at %d (offset %d line).",
3089 "Hunk #%d succeeded at %d (offset %d lines).",
3090 offset),
3091 nth_fragment, applied_pos + 1, offset);
3095 * Warn if it was necessary to reduce the number
3096 * of context lines.
3098 if ((leading != frag->leading ||
3099 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3100 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3101 " to apply fragment at %d"),
3102 leading, trailing, applied_pos+1);
3103 update_image(state, img, applied_pos, &preimage, &postimage);
3104 } else {
3105 if (state->apply_verbosity > verbosity_normal)
3106 error(_("while searching for:\n%.*s"),
3107 (int)(old - oldlines), oldlines);
3110 out:
3111 free(oldlines);
3112 strbuf_release(&newlines);
3113 free(preimage.line_allocated);
3114 free(postimage.line_allocated);
3116 return (applied_pos < 0);
3119 static int apply_binary_fragment(struct apply_state *state,
3120 struct image *img,
3121 struct patch *patch)
3123 struct fragment *fragment = patch->fragments;
3124 unsigned long len;
3125 void *dst;
3127 if (!fragment)
3128 return error(_("missing binary patch data for '%s'"),
3129 patch->new_name ?
3130 patch->new_name :
3131 patch->old_name);
3133 /* Binary patch is irreversible without the optional second hunk */
3134 if (state->apply_in_reverse) {
3135 if (!fragment->next)
3136 return error(_("cannot reverse-apply a binary patch "
3137 "without the reverse hunk to '%s'"),
3138 patch->new_name
3139 ? patch->new_name : patch->old_name);
3140 fragment = fragment->next;
3142 switch (fragment->binary_patch_method) {
3143 case BINARY_DELTA_DEFLATED:
3144 dst = patch_delta(img->buf, img->len, fragment->patch,
3145 fragment->size, &len);
3146 if (!dst)
3147 return -1;
3148 clear_image(img);
3149 img->buf = dst;
3150 img->len = len;
3151 return 0;
3152 case BINARY_LITERAL_DEFLATED:
3153 clear_image(img);
3154 img->len = fragment->size;
3155 img->buf = xmemdupz(fragment->patch, img->len);
3156 return 0;
3158 return -1;
3162 * Replace "img" with the result of applying the binary patch.
3163 * The binary patch data itself in patch->fragment is still kept
3164 * but the preimage prepared by the caller in "img" is freed here
3165 * or in the helper function apply_binary_fragment() this calls.
3167 static int apply_binary(struct apply_state *state,
3168 struct image *img,
3169 struct patch *patch)
3171 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3172 struct object_id oid;
3173 const unsigned hexsz = the_hash_algo->hexsz;
3176 * For safety, we require patch index line to contain
3177 * full hex textual object ID for old and new, at least for now.
3179 if (strlen(patch->old_oid_prefix) != hexsz ||
3180 strlen(patch->new_oid_prefix) != hexsz ||
3181 get_oid_hex(patch->old_oid_prefix, &oid) ||
3182 get_oid_hex(patch->new_oid_prefix, &oid))
3183 return error(_("cannot apply binary patch to '%s' "
3184 "without full index line"), name);
3186 if (patch->old_name) {
3188 * See if the old one matches what the patch
3189 * applies to.
3191 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3192 &oid);
3193 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3194 return error(_("the patch applies to '%s' (%s), "
3195 "which does not match the "
3196 "current contents."),
3197 name, oid_to_hex(&oid));
3199 else {
3200 /* Otherwise, the old one must be empty. */
3201 if (img->len)
3202 return error(_("the patch applies to an empty "
3203 "'%s' but it is not empty"), name);
3206 get_oid_hex(patch->new_oid_prefix, &oid);
3207 if (is_null_oid(&oid)) {
3208 clear_image(img);
3209 return 0; /* deletion patch */
3212 if (has_object(the_repository, &oid, 0)) {
3213 /* We already have the postimage */
3214 enum object_type type;
3215 unsigned long size;
3216 char *result;
3218 result = repo_read_object_file(the_repository, &oid, &type,
3219 &size);
3220 if (!result)
3221 return error(_("the necessary postimage %s for "
3222 "'%s' cannot be read"),
3223 patch->new_oid_prefix, name);
3224 clear_image(img);
3225 img->buf = result;
3226 img->len = size;
3227 } else {
3229 * We have verified buf matches the preimage;
3230 * apply the patch data to it, which is stored
3231 * in the patch->fragments->{patch,size}.
3233 if (apply_binary_fragment(state, img, patch))
3234 return error(_("binary patch does not apply to '%s'"),
3235 name);
3237 /* verify that the result matches */
3238 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3239 &oid);
3240 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3241 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3242 name, patch->new_oid_prefix, oid_to_hex(&oid));
3245 return 0;
3248 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3250 struct fragment *frag = patch->fragments;
3251 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3252 unsigned ws_rule = patch->ws_rule;
3253 unsigned inaccurate_eof = patch->inaccurate_eof;
3254 int nth = 0;
3256 if (patch->is_binary)
3257 return apply_binary(state, img, patch);
3259 while (frag) {
3260 nth++;
3261 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3262 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3263 if (!state->apply_with_reject)
3264 return -1;
3265 frag->rejected = 1;
3267 frag = frag->next;
3269 return 0;
3272 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3274 if (S_ISGITLINK(mode)) {
3275 strbuf_grow(buf, 100);
3276 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3277 } else {
3278 enum object_type type;
3279 unsigned long sz;
3280 char *result;
3282 result = repo_read_object_file(the_repository, oid, &type,
3283 &sz);
3284 if (!result)
3285 return -1;
3286 /* XXX read_sha1_file NUL-terminates */
3287 strbuf_attach(buf, result, sz, sz + 1);
3289 return 0;
3292 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3294 if (!ce)
3295 return 0;
3296 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3299 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3301 struct string_list_item *item;
3303 if (!name)
3304 return NULL;
3306 item = string_list_lookup(&state->fn_table, name);
3307 if (item)
3308 return (struct patch *)item->util;
3310 return NULL;
3314 * item->util in the filename table records the status of the path.
3315 * Usually it points at a patch (whose result records the contents
3316 * of it after applying it), but it could be PATH_WAS_DELETED for a
3317 * path that a previously applied patch has already removed, or
3318 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3320 * The latter is needed to deal with a case where two paths A and B
3321 * are swapped by first renaming A to B and then renaming B to A;
3322 * moving A to B should not be prevented due to presence of B as we
3323 * will remove it in a later patch.
3325 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3326 #define PATH_WAS_DELETED ((struct patch *) -1)
3328 static int to_be_deleted(struct patch *patch)
3330 return patch == PATH_TO_BE_DELETED;
3333 static int was_deleted(struct patch *patch)
3335 return patch == PATH_WAS_DELETED;
3338 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3340 struct string_list_item *item;
3343 * Always add new_name unless patch is a deletion
3344 * This should cover the cases for normal diffs,
3345 * file creations and copies
3347 if (patch->new_name) {
3348 item = string_list_insert(&state->fn_table, patch->new_name);
3349 item->util = patch;
3353 * store a failure on rename/deletion cases because
3354 * later chunks shouldn't patch old names
3356 if ((patch->new_name == NULL) || (patch->is_rename)) {
3357 item = string_list_insert(&state->fn_table, patch->old_name);
3358 item->util = PATH_WAS_DELETED;
3362 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3365 * store information about incoming file deletion
3367 while (patch) {
3368 if ((patch->new_name == NULL) || (patch->is_rename)) {
3369 struct string_list_item *item;
3370 item = string_list_insert(&state->fn_table, patch->old_name);
3371 item->util = PATH_TO_BE_DELETED;
3373 patch = patch->next;
3377 static int checkout_target(struct index_state *istate,
3378 struct cache_entry *ce, struct stat *st)
3380 struct checkout costate = CHECKOUT_INIT;
3382 costate.refresh_cache = 1;
3383 costate.istate = istate;
3384 if (checkout_entry(ce, &costate, NULL, NULL) ||
3385 lstat(ce->name, st))
3386 return error(_("cannot checkout %s"), ce->name);
3387 return 0;
3390 static struct patch *previous_patch(struct apply_state *state,
3391 struct patch *patch,
3392 int *gone)
3394 struct patch *previous;
3396 *gone = 0;
3397 if (patch->is_copy || patch->is_rename)
3398 return NULL; /* "git" patches do not depend on the order */
3400 previous = in_fn_table(state, patch->old_name);
3401 if (!previous)
3402 return NULL;
3404 if (to_be_deleted(previous))
3405 return NULL; /* the deletion hasn't happened yet */
3407 if (was_deleted(previous))
3408 *gone = 1;
3410 return previous;
3413 static int verify_index_match(struct apply_state *state,
3414 const struct cache_entry *ce,
3415 struct stat *st)
3417 if (S_ISGITLINK(ce->ce_mode)) {
3418 if (!S_ISDIR(st->st_mode))
3419 return -1;
3420 return 0;
3422 return ie_match_stat(state->repo->index, ce, st,
3423 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3426 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3428 static int load_patch_target(struct apply_state *state,
3429 struct strbuf *buf,
3430 const struct cache_entry *ce,
3431 struct stat *st,
3432 struct patch *patch,
3433 const char *name,
3434 unsigned expected_mode)
3436 if (state->cached || state->check_index) {
3437 if (read_file_or_gitlink(ce, buf))
3438 return error(_("failed to read %s"), name);
3439 } else if (name) {
3440 if (S_ISGITLINK(expected_mode)) {
3441 if (ce)
3442 return read_file_or_gitlink(ce, buf);
3443 else
3444 return SUBMODULE_PATCH_WITHOUT_INDEX;
3445 } else if (has_symlink_leading_path(name, strlen(name))) {
3446 return error(_("reading from '%s' beyond a symbolic link"), name);
3447 } else {
3448 if (read_old_data(st, patch, name, buf))
3449 return error(_("failed to read %s"), name);
3452 return 0;
3456 * We are about to apply "patch"; populate the "image" with the
3457 * current version we have, from the working tree or from the index,
3458 * depending on the situation e.g. --cached/--index. If we are
3459 * applying a non-git patch that incrementally updates the tree,
3460 * we read from the result of a previous diff.
3462 static int load_preimage(struct apply_state *state,
3463 struct image *image,
3464 struct patch *patch, struct stat *st,
3465 const struct cache_entry *ce)
3467 struct strbuf buf = STRBUF_INIT;
3468 size_t len;
3469 char *img;
3470 struct patch *previous;
3471 int status;
3473 previous = previous_patch(state, patch, &status);
3474 if (status)
3475 return error(_("path %s has been renamed/deleted"),
3476 patch->old_name);
3477 if (previous) {
3478 /* We have a patched copy in memory; use that. */
3479 strbuf_add(&buf, previous->result, previous->resultsize);
3480 } else {
3481 status = load_patch_target(state, &buf, ce, st, patch,
3482 patch->old_name, patch->old_mode);
3483 if (status < 0)
3484 return status;
3485 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3487 * There is no way to apply subproject
3488 * patch without looking at the index.
3489 * NEEDSWORK: shouldn't this be flagged
3490 * as an error???
3492 free_fragment_list(patch->fragments);
3493 patch->fragments = NULL;
3494 } else if (status) {
3495 return error(_("failed to read %s"), patch->old_name);
3499 img = strbuf_detach(&buf, &len);
3500 prepare_image(image, img, len, !patch->is_binary);
3501 return 0;
3504 static int resolve_to(struct image *image, const struct object_id *result_id)
3506 unsigned long size;
3507 enum object_type type;
3509 clear_image(image);
3511 image->buf = repo_read_object_file(the_repository, result_id, &type,
3512 &size);
3513 if (!image->buf || type != OBJ_BLOB)
3514 die("unable to read blob object %s", oid_to_hex(result_id));
3515 image->len = size;
3517 return 0;
3520 static int three_way_merge(struct apply_state *state,
3521 struct image *image,
3522 char *path,
3523 const struct object_id *base,
3524 const struct object_id *ours,
3525 const struct object_id *theirs)
3527 mmfile_t base_file, our_file, their_file;
3528 mmbuffer_t result = { NULL };
3529 enum ll_merge_result status;
3531 /* resolve trivial cases first */
3532 if (oideq(base, ours))
3533 return resolve_to(image, theirs);
3534 else if (oideq(base, theirs) || oideq(ours, theirs))
3535 return resolve_to(image, ours);
3537 read_mmblob(&base_file, base);
3538 read_mmblob(&our_file, ours);
3539 read_mmblob(&their_file, theirs);
3540 status = ll_merge(&result, path,
3541 &base_file, "base",
3542 &our_file, "ours",
3543 &their_file, "theirs",
3544 state->repo->index,
3545 NULL);
3546 if (status == LL_MERGE_BINARY_CONFLICT)
3547 warning("Cannot merge binary files: %s (%s vs. %s)",
3548 path, "ours", "theirs");
3549 free(base_file.ptr);
3550 free(our_file.ptr);
3551 free(their_file.ptr);
3552 if (status < 0 || !result.ptr) {
3553 free(result.ptr);
3554 return -1;
3556 clear_image(image);
3557 image->buf = result.ptr;
3558 image->len = result.size;
3560 return status;
3564 * When directly falling back to add/add three-way merge, we read from
3565 * the current contents of the new_name. In no cases other than that
3566 * this function will be called.
3568 static int load_current(struct apply_state *state,
3569 struct image *image,
3570 struct patch *patch)
3572 struct strbuf buf = STRBUF_INIT;
3573 int status, pos;
3574 size_t len;
3575 char *img;
3576 struct stat st;
3577 struct cache_entry *ce;
3578 char *name = patch->new_name;
3579 unsigned mode = patch->new_mode;
3581 if (!patch->is_new)
3582 BUG("patch to %s is not a creation", patch->old_name);
3584 pos = index_name_pos(state->repo->index, name, strlen(name));
3585 if (pos < 0)
3586 return error(_("%s: does not exist in index"), name);
3587 ce = state->repo->index->cache[pos];
3588 if (lstat(name, &st)) {
3589 if (errno != ENOENT)
3590 return error_errno("%s", name);
3591 if (checkout_target(state->repo->index, ce, &st))
3592 return -1;
3594 if (verify_index_match(state, ce, &st))
3595 return error(_("%s: does not match index"), name);
3597 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3598 if (status < 0)
3599 return status;
3600 else if (status)
3601 return -1;
3602 img = strbuf_detach(&buf, &len);
3603 prepare_image(image, img, len, !patch->is_binary);
3604 return 0;
3607 static int try_threeway(struct apply_state *state,
3608 struct image *image,
3609 struct patch *patch,
3610 struct stat *st,
3611 const struct cache_entry *ce)
3613 struct object_id pre_oid, post_oid, our_oid;
3614 struct strbuf buf = STRBUF_INIT;
3615 size_t len;
3616 int status;
3617 char *img;
3618 struct image tmp_image;
3620 /* No point falling back to 3-way merge in these cases */
3621 if (patch->is_delete ||
3622 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3623 (patch->is_new && !patch->direct_to_threeway) ||
3624 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3625 return -1;
3627 /* Preimage the patch was prepared for */
3628 if (patch->is_new)
3629 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3630 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3631 read_blob_object(&buf, &pre_oid, patch->old_mode))
3632 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3634 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3635 fprintf(stderr, _("Performing three-way merge...\n"));
3637 img = strbuf_detach(&buf, &len);
3638 prepare_image(&tmp_image, img, len, 1);
3639 /* Apply the patch to get the post image */
3640 if (apply_fragments(state, &tmp_image, patch) < 0) {
3641 clear_image(&tmp_image);
3642 return -1;
3644 /* post_oid is theirs */
3645 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3646 clear_image(&tmp_image);
3648 /* our_oid is ours */
3649 if (patch->is_new) {
3650 if (load_current(state, &tmp_image, patch))
3651 return error(_("cannot read the current contents of '%s'"),
3652 patch->new_name);
3653 } else {
3654 if (load_preimage(state, &tmp_image, patch, st, ce))
3655 return error(_("cannot read the current contents of '%s'"),
3656 patch->old_name);
3658 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3659 clear_image(&tmp_image);
3661 /* in-core three-way merge between post and our using pre as base */
3662 status = three_way_merge(state, image, patch->new_name,
3663 &pre_oid, &our_oid, &post_oid);
3664 if (status < 0) {
3665 if (state->apply_verbosity > verbosity_silent)
3666 fprintf(stderr,
3667 _("Failed to perform three-way merge...\n"));
3668 return status;
3671 if (status) {
3672 patch->conflicted_threeway = 1;
3673 if (patch->is_new)
3674 oidclr(&patch->threeway_stage[0]);
3675 else
3676 oidcpy(&patch->threeway_stage[0], &pre_oid);
3677 oidcpy(&patch->threeway_stage[1], &our_oid);
3678 oidcpy(&patch->threeway_stage[2], &post_oid);
3679 if (state->apply_verbosity > verbosity_silent)
3680 fprintf(stderr,
3681 _("Applied patch to '%s' with conflicts.\n"),
3682 patch->new_name);
3683 } else {
3684 if (state->apply_verbosity > verbosity_silent)
3685 fprintf(stderr,
3686 _("Applied patch to '%s' cleanly.\n"),
3687 patch->new_name);
3689 return 0;
3692 static int apply_data(struct apply_state *state, struct patch *patch,
3693 struct stat *st, const struct cache_entry *ce)
3695 struct image image;
3697 if (load_preimage(state, &image, patch, st, ce) < 0)
3698 return -1;
3700 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3701 if (state->apply_verbosity > verbosity_silent &&
3702 state->threeway && !patch->direct_to_threeway)
3703 fprintf(stderr, _("Falling back to direct application...\n"));
3705 /* Note: with --reject, apply_fragments() returns 0 */
3706 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0)
3707 return -1;
3709 patch->result = image.buf;
3710 patch->resultsize = image.len;
3711 add_to_fn_table(state, patch);
3712 free(image.line_allocated);
3714 if (0 < patch->is_delete && patch->resultsize)
3715 return error(_("removal patch leaves file contents"));
3717 return 0;
3721 * If "patch" that we are looking at modifies or deletes what we have,
3722 * we would want it not to lose any local modification we have, either
3723 * in the working tree or in the index.
3725 * This also decides if a non-git patch is a creation patch or a
3726 * modification to an existing empty file. We do not check the state
3727 * of the current tree for a creation patch in this function; the caller
3728 * check_patch() separately makes sure (and errors out otherwise) that
3729 * the path the patch creates does not exist in the current tree.
3731 static int check_preimage(struct apply_state *state,
3732 struct patch *patch,
3733 struct cache_entry **ce,
3734 struct stat *st)
3736 const char *old_name = patch->old_name;
3737 struct patch *previous = NULL;
3738 int stat_ret = 0, status;
3739 unsigned st_mode = 0;
3741 if (!old_name)
3742 return 0;
3744 assert(patch->is_new <= 0);
3745 previous = previous_patch(state, patch, &status);
3747 if (status)
3748 return error(_("path %s has been renamed/deleted"), old_name);
3749 if (previous) {
3750 st_mode = previous->new_mode;
3751 } else if (!state->cached) {
3752 stat_ret = lstat(old_name, st);
3753 if (stat_ret && errno != ENOENT)
3754 return error_errno("%s", old_name);
3757 if (state->check_index && !previous) {
3758 int pos = index_name_pos(state->repo->index, old_name,
3759 strlen(old_name));
3760 if (pos < 0) {
3761 if (patch->is_new < 0)
3762 goto is_new;
3763 return error(_("%s: does not exist in index"), old_name);
3765 *ce = state->repo->index->cache[pos];
3766 if (stat_ret < 0) {
3767 if (checkout_target(state->repo->index, *ce, st))
3768 return -1;
3770 if (!state->cached && verify_index_match(state, *ce, st))
3771 return error(_("%s: does not match index"), old_name);
3772 if (state->cached)
3773 st_mode = (*ce)->ce_mode;
3774 } else if (stat_ret < 0) {
3775 if (patch->is_new < 0)
3776 goto is_new;
3777 return error_errno("%s", old_name);
3780 if (!state->cached && !previous)
3781 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3783 if (patch->is_new < 0)
3784 patch->is_new = 0;
3785 if (!patch->old_mode)
3786 patch->old_mode = st_mode;
3787 if ((st_mode ^ patch->old_mode) & S_IFMT)
3788 return error(_("%s: wrong type"), old_name);
3789 if (st_mode != patch->old_mode)
3790 warning(_("%s has type %o, expected %o"),
3791 old_name, st_mode, patch->old_mode);
3792 if (!patch->new_mode && !patch->is_delete)
3793 patch->new_mode = st_mode;
3794 return 0;
3796 is_new:
3797 patch->is_new = 1;
3798 patch->is_delete = 0;
3799 FREE_AND_NULL(patch->old_name);
3800 return 0;
3804 #define EXISTS_IN_INDEX 1
3805 #define EXISTS_IN_WORKTREE 2
3806 #define EXISTS_IN_INDEX_AS_ITA 3
3808 static int check_to_create(struct apply_state *state,
3809 const char *new_name,
3810 int ok_if_exists)
3812 struct stat nst;
3814 if (state->check_index && (!ok_if_exists || !state->cached)) {
3815 int pos;
3817 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3818 if (pos >= 0) {
3819 struct cache_entry *ce = state->repo->index->cache[pos];
3821 /* allow ITA, as they do not yet exist in the index */
3822 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3823 return EXISTS_IN_INDEX;
3825 /* ITA entries can never match working tree files */
3826 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3827 return EXISTS_IN_INDEX_AS_ITA;
3831 if (state->cached)
3832 return 0;
3834 if (!lstat(new_name, &nst)) {
3835 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3836 return 0;
3838 * A leading component of new_name might be a symlink
3839 * that is going to be removed with this patch, but
3840 * still pointing at somewhere that has the path.
3841 * In such a case, path "new_name" does not exist as
3842 * far as git is concerned.
3844 if (has_symlink_leading_path(new_name, strlen(new_name)))
3845 return 0;
3847 return EXISTS_IN_WORKTREE;
3848 } else if (!is_missing_file_error(errno)) {
3849 return error_errno("%s", new_name);
3851 return 0;
3854 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3856 for ( ; patch; patch = patch->next) {
3857 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3858 (patch->is_rename || patch->is_delete))
3859 /* the symlink at patch->old_name is removed */
3860 strset_add(&state->removed_symlinks, patch->old_name);
3862 if (patch->new_name && S_ISLNK(patch->new_mode))
3863 /* the symlink at patch->new_name is created or remains */
3864 strset_add(&state->kept_symlinks, patch->new_name);
3868 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3870 do {
3871 while (--name->len && name->buf[name->len] != '/')
3872 ; /* scan backwards */
3873 if (!name->len)
3874 break;
3875 name->buf[name->len] = '\0';
3876 if (strset_contains(&state->kept_symlinks, name->buf))
3877 return 1;
3878 if (strset_contains(&state->removed_symlinks, name->buf))
3880 * This cannot be "return 0", because we may
3881 * see a new one created at a higher level.
3883 continue;
3885 /* otherwise, check the preimage */
3886 if (state->check_index) {
3887 struct cache_entry *ce;
3889 ce = index_file_exists(state->repo->index, name->buf,
3890 name->len, ignore_case);
3891 if (ce && S_ISLNK(ce->ce_mode))
3892 return 1;
3893 } else {
3894 struct stat st;
3895 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3896 return 1;
3898 } while (1);
3899 return 0;
3902 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3904 int ret;
3905 struct strbuf name = STRBUF_INIT;
3907 assert(*name_ != '\0');
3908 strbuf_addstr(&name, name_);
3909 ret = path_is_beyond_symlink_1(state, &name);
3910 strbuf_release(&name);
3912 return ret;
3915 static int check_unsafe_path(struct patch *patch)
3917 const char *old_name = NULL;
3918 const char *new_name = NULL;
3919 if (patch->is_delete)
3920 old_name = patch->old_name;
3921 else if (!patch->is_new && !patch->is_copy)
3922 old_name = patch->old_name;
3923 if (!patch->is_delete)
3924 new_name = patch->new_name;
3926 if (old_name && !verify_path(old_name, patch->old_mode))
3927 return error(_("invalid path '%s'"), old_name);
3928 if (new_name && !verify_path(new_name, patch->new_mode))
3929 return error(_("invalid path '%s'"), new_name);
3930 return 0;
3934 * Check and apply the patch in-core; leave the result in patch->result
3935 * for the caller to write it out to the final destination.
3937 static int check_patch(struct apply_state *state, struct patch *patch)
3939 struct stat st;
3940 const char *old_name = patch->old_name;
3941 const char *new_name = patch->new_name;
3942 const char *name = old_name ? old_name : new_name;
3943 struct cache_entry *ce = NULL;
3944 struct patch *tpatch;
3945 int ok_if_exists;
3946 int status;
3948 patch->rejected = 1; /* we will drop this after we succeed */
3950 status = check_preimage(state, patch, &ce, &st);
3951 if (status)
3952 return status;
3953 old_name = patch->old_name;
3956 * A type-change diff is always split into a patch to delete
3957 * old, immediately followed by a patch to create new (see
3958 * diff.c::run_diff()); in such a case it is Ok that the entry
3959 * to be deleted by the previous patch is still in the working
3960 * tree and in the index.
3962 * A patch to swap-rename between A and B would first rename A
3963 * to B and then rename B to A. While applying the first one,
3964 * the presence of B should not stop A from getting renamed to
3965 * B; ask to_be_deleted() about the later rename. Removal of
3966 * B and rename from A to B is handled the same way by asking
3967 * was_deleted().
3969 if ((tpatch = in_fn_table(state, new_name)) &&
3970 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3971 ok_if_exists = 1;
3972 else
3973 ok_if_exists = 0;
3975 if (new_name &&
3976 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3977 int err = check_to_create(state, new_name, ok_if_exists);
3979 if (err && state->threeway) {
3980 patch->direct_to_threeway = 1;
3981 } else switch (err) {
3982 case 0:
3983 break; /* happy */
3984 case EXISTS_IN_INDEX:
3985 return error(_("%s: already exists in index"), new_name);
3986 case EXISTS_IN_INDEX_AS_ITA:
3987 return error(_("%s: does not match index"), new_name);
3988 case EXISTS_IN_WORKTREE:
3989 return error(_("%s: already exists in working directory"),
3990 new_name);
3991 default:
3992 return err;
3995 if (!patch->new_mode) {
3996 if (0 < patch->is_new)
3997 patch->new_mode = S_IFREG | 0644;
3998 else
3999 patch->new_mode = patch->old_mode;
4003 if (new_name && old_name) {
4004 int same = !strcmp(old_name, new_name);
4005 if (!patch->new_mode)
4006 patch->new_mode = patch->old_mode;
4007 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4008 if (same)
4009 return error(_("new mode (%o) of %s does not "
4010 "match old mode (%o)"),
4011 patch->new_mode, new_name,
4012 patch->old_mode);
4013 else
4014 return error(_("new mode (%o) of %s does not "
4015 "match old mode (%o) of %s"),
4016 patch->new_mode, new_name,
4017 patch->old_mode, old_name);
4021 if (!state->unsafe_paths && check_unsafe_path(patch))
4022 return -128;
4025 * An attempt to read from or delete a path that is beyond a
4026 * symbolic link will be prevented by load_patch_target() that
4027 * is called at the beginning of apply_data() so we do not
4028 * have to worry about a patch marked with "is_delete" bit
4029 * here. We however need to make sure that the patch result
4030 * is not deposited to a path that is beyond a symbolic link
4031 * here.
4033 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4034 return error(_("affected file '%s' is beyond a symbolic link"),
4035 patch->new_name);
4037 if (apply_data(state, patch, &st, ce) < 0)
4038 return error(_("%s: patch does not apply"), name);
4039 patch->rejected = 0;
4040 return 0;
4043 static int check_patch_list(struct apply_state *state, struct patch *patch)
4045 int err = 0;
4047 prepare_symlink_changes(state, patch);
4048 prepare_fn_table(state, patch);
4049 while (patch) {
4050 int res;
4051 if (state->apply_verbosity > verbosity_normal)
4052 say_patch_name(stderr,
4053 _("Checking patch %s..."), patch);
4054 res = check_patch(state, patch);
4055 if (res == -128)
4056 return -128;
4057 err |= res;
4058 patch = patch->next;
4060 return err;
4063 static int read_apply_cache(struct apply_state *state)
4065 if (state->index_file)
4066 return read_index_from(state->repo->index, state->index_file,
4067 get_git_dir());
4068 else
4069 return repo_read_index(state->repo);
4072 /* This function tries to read the object name from the current index */
4073 static int get_current_oid(struct apply_state *state, const char *path,
4074 struct object_id *oid)
4076 int pos;
4078 if (read_apply_cache(state) < 0)
4079 return -1;
4080 pos = index_name_pos(state->repo->index, path, strlen(path));
4081 if (pos < 0)
4082 return -1;
4083 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4084 return 0;
4087 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4090 * A usable gitlink patch has only one fragment (hunk) that looks like:
4091 * @@ -1 +1 @@
4092 * -Subproject commit <old sha1>
4093 * +Subproject commit <new sha1>
4094 * or
4095 * @@ -1 +0,0 @@
4096 * -Subproject commit <old sha1>
4097 * for a removal patch.
4099 struct fragment *hunk = p->fragments;
4100 static const char heading[] = "-Subproject commit ";
4101 char *preimage;
4103 if (/* does the patch have only one hunk? */
4104 hunk && !hunk->next &&
4105 /* is its preimage one line? */
4106 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4107 /* does preimage begin with the heading? */
4108 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4109 starts_with(++preimage, heading) &&
4110 /* does it record full SHA-1? */
4111 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4112 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4113 /* does the abbreviated name on the index line agree with it? */
4114 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4115 return 0; /* it all looks fine */
4117 /* we may have full object name on the index line */
4118 return get_oid_hex(p->old_oid_prefix, oid);
4121 /* Build an index that contains just the files needed for a 3way merge */
4122 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4124 struct patch *patch;
4125 struct index_state result = INDEX_STATE_INIT(state->repo);
4126 struct lock_file lock = LOCK_INIT;
4127 int res;
4129 /* Once we start supporting the reverse patch, it may be
4130 * worth showing the new sha1 prefix, but until then...
4132 for (patch = list; patch; patch = patch->next) {
4133 struct object_id oid;
4134 struct cache_entry *ce;
4135 const char *name;
4137 name = patch->old_name ? patch->old_name : patch->new_name;
4138 if (0 < patch->is_new)
4139 continue;
4141 if (S_ISGITLINK(patch->old_mode)) {
4142 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4143 ; /* ok, the textual part looks sane */
4144 else
4145 return error(_("sha1 information is lacking or "
4146 "useless for submodule %s"), name);
4147 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4148 ; /* ok */
4149 } else if (!patch->lines_added && !patch->lines_deleted) {
4150 /* mode-only change: update the current */
4151 if (get_current_oid(state, patch->old_name, &oid))
4152 return error(_("mode change for %s, which is not "
4153 "in current HEAD"), name);
4154 } else
4155 return error(_("sha1 information is lacking or useless "
4156 "(%s)."), name);
4158 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4159 if (!ce)
4160 return error(_("make_cache_entry failed for path '%s'"),
4161 name);
4162 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4163 discard_cache_entry(ce);
4164 return error(_("could not add %s to temporary index"),
4165 name);
4169 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4170 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4171 discard_index(&result);
4173 if (res)
4174 return error(_("could not write temporary index to %s"),
4175 state->fake_ancestor);
4177 return 0;
4180 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4182 int files, adds, dels;
4184 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4185 files++;
4186 adds += patch->lines_added;
4187 dels += patch->lines_deleted;
4188 show_stats(state, patch);
4191 print_stat_summary(stdout, files, adds, dels);
4194 static void numstat_patch_list(struct apply_state *state,
4195 struct patch *patch)
4197 for ( ; patch; patch = patch->next) {
4198 const char *name;
4199 name = patch->new_name ? patch->new_name : patch->old_name;
4200 if (patch->is_binary)
4201 printf("-\t-\t");
4202 else
4203 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4204 write_name_quoted(name, stdout, state->line_termination);
4208 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4210 if (mode)
4211 printf(" %s mode %06o %s\n", newdelete, mode, name);
4212 else
4213 printf(" %s %s\n", newdelete, name);
4216 static void show_mode_change(struct patch *p, int show_name)
4218 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4219 if (show_name)
4220 printf(" mode change %06o => %06o %s\n",
4221 p->old_mode, p->new_mode, p->new_name);
4222 else
4223 printf(" mode change %06o => %06o\n",
4224 p->old_mode, p->new_mode);
4228 static void show_rename_copy(struct patch *p)
4230 const char *renamecopy = p->is_rename ? "rename" : "copy";
4231 const char *old_name, *new_name;
4233 /* Find common prefix */
4234 old_name = p->old_name;
4235 new_name = p->new_name;
4236 while (1) {
4237 const char *slash_old, *slash_new;
4238 slash_old = strchr(old_name, '/');
4239 slash_new = strchr(new_name, '/');
4240 if (!slash_old ||
4241 !slash_new ||
4242 slash_old - old_name != slash_new - new_name ||
4243 memcmp(old_name, new_name, slash_new - new_name))
4244 break;
4245 old_name = slash_old + 1;
4246 new_name = slash_new + 1;
4248 /* p->old_name through old_name is the common prefix, and old_name and
4249 * new_name through the end of names are renames
4251 if (old_name != p->old_name)
4252 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4253 (int)(old_name - p->old_name), p->old_name,
4254 old_name, new_name, p->score);
4255 else
4256 printf(" %s %s => %s (%d%%)\n", renamecopy,
4257 p->old_name, p->new_name, p->score);
4258 show_mode_change(p, 0);
4261 static void summary_patch_list(struct patch *patch)
4263 struct patch *p;
4265 for (p = patch; p; p = p->next) {
4266 if (p->is_new)
4267 show_file_mode_name("create", p->new_mode, p->new_name);
4268 else if (p->is_delete)
4269 show_file_mode_name("delete", p->old_mode, p->old_name);
4270 else {
4271 if (p->is_rename || p->is_copy)
4272 show_rename_copy(p);
4273 else {
4274 if (p->score) {
4275 printf(" rewrite %s (%d%%)\n",
4276 p->new_name, p->score);
4277 show_mode_change(p, 0);
4279 else
4280 show_mode_change(p, 1);
4286 static void patch_stats(struct apply_state *state, struct patch *patch)
4288 int lines = patch->lines_added + patch->lines_deleted;
4290 if (lines > state->max_change)
4291 state->max_change = lines;
4292 if (patch->old_name) {
4293 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4294 if (!len)
4295 len = strlen(patch->old_name);
4296 if (len > state->max_len)
4297 state->max_len = len;
4299 if (patch->new_name) {
4300 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4301 if (!len)
4302 len = strlen(patch->new_name);
4303 if (len > state->max_len)
4304 state->max_len = len;
4308 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4310 if (state->update_index && !state->ita_only) {
4311 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4312 return error(_("unable to remove %s from index"), patch->old_name);
4314 if (!state->cached) {
4315 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4316 remove_path(patch->old_name);
4319 return 0;
4322 static int add_index_file(struct apply_state *state,
4323 const char *path,
4324 unsigned mode,
4325 void *buf,
4326 unsigned long size)
4328 struct stat st;
4329 struct cache_entry *ce;
4330 int namelen = strlen(path);
4332 ce = make_empty_cache_entry(state->repo->index, namelen);
4333 memcpy(ce->name, path, namelen);
4334 ce->ce_mode = create_ce_mode(mode);
4335 ce->ce_flags = create_ce_flags(0);
4336 ce->ce_namelen = namelen;
4337 if (state->ita_only) {
4338 ce->ce_flags |= CE_INTENT_TO_ADD;
4339 set_object_name_for_intent_to_add_entry(ce);
4340 } else if (S_ISGITLINK(mode)) {
4341 const char *s;
4343 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4344 get_oid_hex(s, &ce->oid)) {
4345 discard_cache_entry(ce);
4346 return error(_("corrupt patch for submodule %s"), path);
4348 } else {
4349 if (!state->cached) {
4350 if (lstat(path, &st) < 0) {
4351 discard_cache_entry(ce);
4352 return error_errno(_("unable to stat newly "
4353 "created file '%s'"),
4354 path);
4356 fill_stat_cache_info(state->repo->index, ce, &st);
4358 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4359 discard_cache_entry(ce);
4360 return error(_("unable to create backing store "
4361 "for newly created file %s"), path);
4364 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4365 discard_cache_entry(ce);
4366 return error(_("unable to add cache entry for %s"), path);
4369 return 0;
4373 * Returns:
4374 * -1 if an unrecoverable error happened
4375 * 0 if everything went well
4376 * 1 if a recoverable error happened
4378 static int try_create_file(struct apply_state *state, const char *path,
4379 unsigned int mode, const char *buf,
4380 unsigned long size)
4382 int fd, res;
4383 struct strbuf nbuf = STRBUF_INIT;
4385 if (S_ISGITLINK(mode)) {
4386 struct stat st;
4387 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4388 return 0;
4389 return !!mkdir(path, 0777);
4392 if (has_symlinks && S_ISLNK(mode))
4393 /* Although buf:size is counted string, it also is NUL
4394 * terminated.
4396 return !!symlink(buf, path);
4398 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4399 if (fd < 0)
4400 return 1;
4402 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4403 size = nbuf.len;
4404 buf = nbuf.buf;
4407 res = write_in_full(fd, buf, size) < 0;
4408 if (res)
4409 error_errno(_("failed to write to '%s'"), path);
4410 strbuf_release(&nbuf);
4412 if (close(fd) < 0 && !res)
4413 return error_errno(_("closing file '%s'"), path);
4415 return res ? -1 : 0;
4419 * We optimistically assume that the directories exist,
4420 * which is true 99% of the time anyway. If they don't,
4421 * we create them and try again.
4423 * Returns:
4424 * -1 on error
4425 * 0 otherwise
4427 static int create_one_file(struct apply_state *state,
4428 char *path,
4429 unsigned mode,
4430 const char *buf,
4431 unsigned long size)
4433 int res;
4435 if (state->cached)
4436 return 0;
4439 * We already try to detect whether files are beyond a symlink in our
4440 * up-front checks. But in the case where symlinks are created by any
4441 * of the intermediate hunks it can happen that our up-front checks
4442 * didn't yet see the symlink, but at the point of arriving here there
4443 * in fact is one. We thus repeat the check for symlinks here.
4445 * Note that this does not make the up-front check obsolete as the
4446 * failure mode is different:
4448 * - The up-front checks cause us to abort before we have written
4449 * anything into the working directory. So when we exit this way the
4450 * working directory remains clean.
4452 * - The checks here happen in the middle of the action where we have
4453 * already started to apply the patch. The end result will be a dirty
4454 * working directory.
4456 * Ideally, we should update the up-front checks to catch what would
4457 * happen when we apply the patch before we damage the working tree.
4458 * We have all the information necessary to do so. But for now, as a
4459 * part of embargoed security work, having this check would serve as a
4460 * reasonable first step.
4462 if (path_is_beyond_symlink(state, path))
4463 return error(_("affected file '%s' is beyond a symbolic link"), path);
4465 res = try_create_file(state, path, mode, buf, size);
4466 if (res < 0)
4467 return -1;
4468 if (!res)
4469 return 0;
4471 if (errno == ENOENT) {
4472 if (safe_create_leading_directories_no_share(path))
4473 return 0;
4474 res = try_create_file(state, path, mode, buf, size);
4475 if (res < 0)
4476 return -1;
4477 if (!res)
4478 return 0;
4481 if (errno == EEXIST || errno == EACCES) {
4482 /* We may be trying to create a file where a directory
4483 * used to be.
4485 struct stat st;
4486 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4487 errno = EEXIST;
4490 if (errno == EEXIST) {
4491 unsigned int nr = getpid();
4493 for (;;) {
4494 char newpath[PATH_MAX];
4495 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4496 res = try_create_file(state, newpath, mode, buf, size);
4497 if (res < 0)
4498 return -1;
4499 if (!res) {
4500 if (!rename(newpath, path))
4501 return 0;
4502 unlink_or_warn(newpath);
4503 break;
4505 if (errno != EEXIST)
4506 break;
4507 ++nr;
4510 return error_errno(_("unable to write file '%s' mode %o"),
4511 path, mode);
4514 static int add_conflicted_stages_file(struct apply_state *state,
4515 struct patch *patch)
4517 int stage, namelen;
4518 unsigned mode;
4519 struct cache_entry *ce;
4521 if (!state->update_index)
4522 return 0;
4523 namelen = strlen(patch->new_name);
4524 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4526 remove_file_from_index(state->repo->index, patch->new_name);
4527 for (stage = 1; stage < 4; stage++) {
4528 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4529 continue;
4530 ce = make_empty_cache_entry(state->repo->index, namelen);
4531 memcpy(ce->name, patch->new_name, namelen);
4532 ce->ce_mode = create_ce_mode(mode);
4533 ce->ce_flags = create_ce_flags(stage);
4534 ce->ce_namelen = namelen;
4535 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4536 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4537 discard_cache_entry(ce);
4538 return error(_("unable to add cache entry for %s"),
4539 patch->new_name);
4543 return 0;
4546 static int create_file(struct apply_state *state, struct patch *patch)
4548 char *path = patch->new_name;
4549 unsigned mode = patch->new_mode;
4550 unsigned long size = patch->resultsize;
4551 char *buf = patch->result;
4553 if (!mode)
4554 mode = S_IFREG | 0644;
4555 if (create_one_file(state, path, mode, buf, size))
4556 return -1;
4558 if (patch->conflicted_threeway)
4559 return add_conflicted_stages_file(state, patch);
4560 else if (state->update_index)
4561 return add_index_file(state, path, mode, buf, size);
4562 return 0;
4565 /* phase zero is to remove, phase one is to create */
4566 static int write_out_one_result(struct apply_state *state,
4567 struct patch *patch,
4568 int phase)
4570 if (patch->is_delete > 0) {
4571 if (phase == 0)
4572 return remove_file(state, patch, 1);
4573 return 0;
4575 if (patch->is_new > 0 || patch->is_copy) {
4576 if (phase == 1)
4577 return create_file(state, patch);
4578 return 0;
4581 * Rename or modification boils down to the same
4582 * thing: remove the old, write the new
4584 if (phase == 0)
4585 return remove_file(state, patch, patch->is_rename);
4586 if (phase == 1)
4587 return create_file(state, patch);
4588 return 0;
4591 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4593 FILE *rej;
4594 char namebuf[PATH_MAX];
4595 struct fragment *frag;
4596 int fd, cnt = 0;
4597 struct strbuf sb = STRBUF_INIT;
4599 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4600 if (!frag->rejected)
4601 continue;
4602 cnt++;
4605 if (!cnt) {
4606 if (state->apply_verbosity > verbosity_normal)
4607 say_patch_name(stderr,
4608 _("Applied patch %s cleanly."), patch);
4609 return 0;
4612 /* This should not happen, because a removal patch that leaves
4613 * contents are marked "rejected" at the patch level.
4615 if (!patch->new_name)
4616 die(_("internal error"));
4618 /* Say this even without --verbose */
4619 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4620 "Applying patch %%s with %d rejects...",
4621 cnt),
4622 cnt);
4623 if (state->apply_verbosity > verbosity_silent)
4624 say_patch_name(stderr, sb.buf, patch);
4625 strbuf_release(&sb);
4627 cnt = strlen(patch->new_name);
4628 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4629 cnt = ARRAY_SIZE(namebuf) - 5;
4630 warning(_("truncating .rej filename to %.*s.rej"),
4631 cnt - 1, patch->new_name);
4633 memcpy(namebuf, patch->new_name, cnt);
4634 memcpy(namebuf + cnt, ".rej", 5);
4636 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4637 if (fd < 0) {
4638 if (errno != EEXIST)
4639 return error_errno(_("cannot open %s"), namebuf);
4640 if (unlink(namebuf))
4641 return error_errno(_("cannot unlink '%s'"), namebuf);
4642 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4643 if (fd < 0)
4644 return error_errno(_("cannot open %s"), namebuf);
4646 rej = fdopen(fd, "w");
4647 if (!rej)
4648 return error_errno(_("cannot open %s"), namebuf);
4650 /* Normal git tools never deal with .rej, so do not pretend
4651 * this is a git patch by saying --git or giving extended
4652 * headers. While at it, maybe please "kompare" that wants
4653 * the trailing TAB and some garbage at the end of line ;-).
4655 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4656 patch->new_name, patch->new_name);
4657 for (cnt = 1, frag = patch->fragments;
4658 frag;
4659 cnt++, frag = frag->next) {
4660 if (!frag->rejected) {
4661 if (state->apply_verbosity > verbosity_silent)
4662 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4663 continue;
4665 if (state->apply_verbosity > verbosity_silent)
4666 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4667 fprintf(rej, "%.*s", frag->size, frag->patch);
4668 if (frag->patch[frag->size-1] != '\n')
4669 fputc('\n', rej);
4671 fclose(rej);
4672 return -1;
4676 * Returns:
4677 * -1 if an error happened
4678 * 0 if the patch applied cleanly
4679 * 1 if the patch did not apply cleanly
4681 static int write_out_results(struct apply_state *state, struct patch *list)
4683 int phase;
4684 int errs = 0;
4685 struct patch *l;
4686 struct string_list cpath = STRING_LIST_INIT_DUP;
4688 for (phase = 0; phase < 2; phase++) {
4689 l = list;
4690 while (l) {
4691 if (l->rejected)
4692 errs = 1;
4693 else {
4694 if (write_out_one_result(state, l, phase)) {
4695 string_list_clear(&cpath, 0);
4696 return -1;
4698 if (phase == 1) {
4699 if (write_out_one_reject(state, l))
4700 errs = 1;
4701 if (l->conflicted_threeway) {
4702 string_list_append(&cpath, l->new_name);
4703 errs = 1;
4707 l = l->next;
4711 if (cpath.nr) {
4712 struct string_list_item *item;
4714 string_list_sort(&cpath);
4715 if (state->apply_verbosity > verbosity_silent) {
4716 for_each_string_list_item(item, &cpath)
4717 fprintf(stderr, "U %s\n", item->string);
4719 string_list_clear(&cpath, 0);
4722 * rerere relies on the partially merged result being in the working
4723 * tree with conflict markers, but that isn't written with --cached.
4725 if (!state->cached)
4726 repo_rerere(state->repo, 0);
4729 return errs;
4733 * Try to apply a patch.
4735 * Returns:
4736 * -128 if a bad error happened (like patch unreadable)
4737 * -1 if patch did not apply and user cannot deal with it
4738 * 0 if the patch applied
4739 * 1 if the patch did not apply but user might fix it
4741 static int apply_patch(struct apply_state *state,
4742 int fd,
4743 const char *filename,
4744 int options)
4746 size_t offset;
4747 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4748 struct patch *list = NULL, **listp = &list;
4749 int skipped_patch = 0;
4750 int res = 0;
4751 int flush_attributes = 0;
4753 state->patch_input_file = filename;
4754 if (read_patch_file(&buf, fd) < 0)
4755 return -128;
4756 offset = 0;
4757 while (offset < buf.len) {
4758 struct patch *patch;
4759 int nr;
4761 CALLOC_ARRAY(patch, 1);
4762 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4763 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4764 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4765 if (nr < 0) {
4766 free_patch(patch);
4767 if (nr == -128) {
4768 res = -128;
4769 goto end;
4771 break;
4773 if (state->apply_in_reverse)
4774 reverse_patches(patch);
4775 if (use_patch(state, patch)) {
4776 patch_stats(state, patch);
4777 if (!list || !state->apply_in_reverse) {
4778 *listp = patch;
4779 listp = &patch->next;
4780 } else {
4781 patch->next = list;
4782 list = patch;
4785 if ((patch->new_name &&
4786 ends_with_path_components(patch->new_name,
4787 GITATTRIBUTES_FILE)) ||
4788 (patch->old_name &&
4789 ends_with_path_components(patch->old_name,
4790 GITATTRIBUTES_FILE)))
4791 flush_attributes = 1;
4793 else {
4794 if (state->apply_verbosity > verbosity_normal)
4795 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4796 free_patch(patch);
4797 skipped_patch++;
4799 offset += nr;
4802 if (!list && !skipped_patch) {
4803 if (!state->allow_empty) {
4804 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4805 res = -128;
4807 goto end;
4810 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4811 state->apply = 0;
4813 state->update_index = (state->check_index || state->ita_only) && state->apply;
4814 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4815 if (state->index_file)
4816 hold_lock_file_for_update(&state->lock_file,
4817 state->index_file,
4818 LOCK_DIE_ON_ERROR);
4819 else
4820 repo_hold_locked_index(state->repo, &state->lock_file,
4821 LOCK_DIE_ON_ERROR);
4824 if (state->check_index && read_apply_cache(state) < 0) {
4825 error(_("unable to read index file"));
4826 res = -128;
4827 goto end;
4830 if (state->check || state->apply) {
4831 int r = check_patch_list(state, list);
4832 if (r == -128) {
4833 res = -128;
4834 goto end;
4836 if (r < 0 && !state->apply_with_reject) {
4837 res = -1;
4838 goto end;
4842 if (state->apply) {
4843 int write_res = write_out_results(state, list);
4844 if (write_res < 0) {
4845 res = -128;
4846 goto end;
4848 if (write_res > 0) {
4849 /* with --3way, we still need to write the index out */
4850 res = state->apply_with_reject ? -1 : 1;
4851 goto end;
4855 if (state->fake_ancestor &&
4856 build_fake_ancestor(state, list)) {
4857 res = -128;
4858 goto end;
4861 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4862 stat_patch_list(state, list);
4864 if (state->numstat && state->apply_verbosity > verbosity_silent)
4865 numstat_patch_list(state, list);
4867 if (state->summary && state->apply_verbosity > verbosity_silent)
4868 summary_patch_list(list);
4870 if (flush_attributes)
4871 reset_parsed_attributes();
4872 end:
4873 free_patch_list(list);
4874 strbuf_release(&buf);
4875 string_list_clear(&state->fn_table, 0);
4876 return res;
4879 static int apply_option_parse_exclude(const struct option *opt,
4880 const char *arg, int unset)
4882 struct apply_state *state = opt->value;
4884 BUG_ON_OPT_NEG(unset);
4886 add_name_limit(state, arg, 1);
4887 return 0;
4890 static int apply_option_parse_include(const struct option *opt,
4891 const char *arg, int unset)
4893 struct apply_state *state = opt->value;
4895 BUG_ON_OPT_NEG(unset);
4897 add_name_limit(state, arg, 0);
4898 state->has_include = 1;
4899 return 0;
4902 static int apply_option_parse_p(const struct option *opt,
4903 const char *arg,
4904 int unset)
4906 struct apply_state *state = opt->value;
4908 BUG_ON_OPT_NEG(unset);
4910 state->p_value = atoi(arg);
4911 state->p_value_known = 1;
4912 return 0;
4915 static int apply_option_parse_space_change(const struct option *opt,
4916 const char *arg, int unset)
4918 struct apply_state *state = opt->value;
4920 BUG_ON_OPT_ARG(arg);
4922 if (unset)
4923 state->ws_ignore_action = ignore_ws_none;
4924 else
4925 state->ws_ignore_action = ignore_ws_change;
4926 return 0;
4929 static int apply_option_parse_whitespace(const struct option *opt,
4930 const char *arg, int unset)
4932 struct apply_state *state = opt->value;
4934 BUG_ON_OPT_NEG(unset);
4936 state->whitespace_option = arg;
4937 if (parse_whitespace_option(state, arg))
4938 return -1;
4939 return 0;
4942 static int apply_option_parse_directory(const struct option *opt,
4943 const char *arg, int unset)
4945 struct apply_state *state = opt->value;
4947 BUG_ON_OPT_NEG(unset);
4949 strbuf_reset(&state->root);
4950 strbuf_addstr(&state->root, arg);
4951 strbuf_complete(&state->root, '/');
4952 return 0;
4955 int apply_all_patches(struct apply_state *state,
4956 int argc,
4957 const char **argv,
4958 int options)
4960 int i;
4961 int res;
4962 int errs = 0;
4963 int read_stdin = 1;
4965 for (i = 0; i < argc; i++) {
4966 const char *arg = argv[i];
4967 char *to_free = NULL;
4968 int fd;
4970 if (!strcmp(arg, "-")) {
4971 res = apply_patch(state, 0, "<stdin>", options);
4972 if (res < 0)
4973 goto end;
4974 errs |= res;
4975 read_stdin = 0;
4976 continue;
4977 } else
4978 arg = to_free = prefix_filename(state->prefix, arg);
4980 fd = open(arg, O_RDONLY);
4981 if (fd < 0) {
4982 error(_("can't open patch '%s': %s"), arg, strerror(errno));
4983 res = -128;
4984 free(to_free);
4985 goto end;
4987 read_stdin = 0;
4988 set_default_whitespace_mode(state);
4989 res = apply_patch(state, fd, arg, options);
4990 close(fd);
4991 free(to_free);
4992 if (res < 0)
4993 goto end;
4994 errs |= res;
4996 set_default_whitespace_mode(state);
4997 if (read_stdin) {
4998 res = apply_patch(state, 0, "<stdin>", options);
4999 if (res < 0)
5000 goto end;
5001 errs |= res;
5004 if (state->whitespace_error) {
5005 if (state->squelch_whitespace_errors &&
5006 state->squelch_whitespace_errors < state->whitespace_error) {
5007 int squelched =
5008 state->whitespace_error - state->squelch_whitespace_errors;
5009 warning(Q_("squelched %d whitespace error",
5010 "squelched %d whitespace errors",
5011 squelched),
5012 squelched);
5014 if (state->ws_error_action == die_on_ws_error) {
5015 error(Q_("%d line adds whitespace errors.",
5016 "%d lines add whitespace errors.",
5017 state->whitespace_error),
5018 state->whitespace_error);
5019 res = -128;
5020 goto end;
5022 if (state->applied_after_fixing_ws && state->apply)
5023 warning(Q_("%d line applied after"
5024 " fixing whitespace errors.",
5025 "%d lines applied after"
5026 " fixing whitespace errors.",
5027 state->applied_after_fixing_ws),
5028 state->applied_after_fixing_ws);
5029 else if (state->whitespace_error)
5030 warning(Q_("%d line adds whitespace errors.",
5031 "%d lines add whitespace errors.",
5032 state->whitespace_error),
5033 state->whitespace_error);
5036 if (state->update_index) {
5037 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5038 if (res) {
5039 error(_("Unable to write new index file"));
5040 res = -128;
5041 goto end;
5045 res = !!errs;
5047 end:
5048 rollback_lock_file(&state->lock_file);
5050 if (state->apply_verbosity <= verbosity_silent) {
5051 set_error_routine(state->saved_error_routine);
5052 set_warn_routine(state->saved_warn_routine);
5055 if (res > -1)
5056 return res;
5057 return (res == -1 ? 1 : 128);
5060 int apply_parse_options(int argc, const char **argv,
5061 struct apply_state *state,
5062 int *force_apply, int *options,
5063 const char * const *apply_usage)
5065 struct option builtin_apply_options[] = {
5066 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5067 N_("don't apply changes matching the given path"),
5068 PARSE_OPT_NONEG, apply_option_parse_exclude),
5069 OPT_CALLBACK_F(0, "include", state, N_("path"),
5070 N_("apply changes matching the given path"),
5071 PARSE_OPT_NONEG, apply_option_parse_include),
5072 OPT_CALLBACK('p', NULL, state, N_("num"),
5073 N_("remove <num> leading slashes from traditional diff paths"),
5074 apply_option_parse_p),
5075 OPT_BOOL(0, "no-add", &state->no_add,
5076 N_("ignore additions made by the patch")),
5077 OPT_BOOL(0, "stat", &state->diffstat,
5078 N_("instead of applying the patch, output diffstat for the input")),
5079 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5080 OPT_NOOP_NOARG(0, "binary"),
5081 OPT_BOOL(0, "numstat", &state->numstat,
5082 N_("show number of added and deleted lines in decimal notation")),
5083 OPT_BOOL(0, "summary", &state->summary,
5084 N_("instead of applying the patch, output a summary for the input")),
5085 OPT_BOOL(0, "check", &state->check,
5086 N_("instead of applying the patch, see if the patch is applicable")),
5087 OPT_BOOL(0, "index", &state->check_index,
5088 N_("make sure the patch is applicable to the current index")),
5089 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5090 N_("mark new files with `git add --intent-to-add`")),
5091 OPT_BOOL(0, "cached", &state->cached,
5092 N_("apply a patch without touching the working tree")),
5093 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5094 N_("accept a patch that touches outside the working area"),
5095 PARSE_OPT_NOCOMPLETE),
5096 OPT_BOOL(0, "apply", force_apply,
5097 N_("also apply the patch (use with --stat/--summary/--check)")),
5098 OPT_BOOL('3', "3way", &state->threeway,
5099 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5100 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5101 N_("build a temporary index based on embedded index information")),
5102 /* Think twice before adding "--nul" synonym to this */
5103 OPT_SET_INT('z', NULL, &state->line_termination,
5104 N_("paths are separated with NUL character"), '\0'),
5105 OPT_INTEGER('C', NULL, &state->p_context,
5106 N_("ensure at least <n> lines of context match")),
5107 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5108 N_("detect new or modified lines that have whitespace errors"),
5109 apply_option_parse_whitespace),
5110 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5111 N_("ignore changes in whitespace when finding context"),
5112 PARSE_OPT_NOARG, apply_option_parse_space_change),
5113 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5114 N_("ignore changes in whitespace when finding context"),
5115 PARSE_OPT_NOARG, apply_option_parse_space_change),
5116 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5117 N_("apply the patch in reverse")),
5118 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5119 N_("don't expect at least one line of context")),
5120 OPT_BOOL(0, "reject", &state->apply_with_reject,
5121 N_("leave the rejected hunks in corresponding *.rej files")),
5122 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5123 N_("allow overlapping hunks")),
5124 OPT__VERBOSITY(&state->apply_verbosity),
5125 OPT_BIT(0, "inaccurate-eof", options,
5126 N_("tolerate incorrectly detected missing new-line at the end of file"),
5127 APPLY_OPT_INACCURATE_EOF),
5128 OPT_BIT(0, "recount", options,
5129 N_("do not trust the line counts in the hunk headers"),
5130 APPLY_OPT_RECOUNT),
5131 OPT_CALLBACK(0, "directory", state, N_("root"),
5132 N_("prepend <root> to all filenames"),
5133 apply_option_parse_directory),
5134 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5135 N_("don't return error for empty patches")),
5136 OPT_END()
5139 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);