gitfaq: add documentation on proxies
[git.git] / apply.c
blobff939f908add9e359e7aee268b7791920670f51f
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
10 #define USE_THE_REPOSITORY_VARIABLE
12 #include "git-compat-util.h"
13 #include "abspath.h"
14 #include "base85.h"
15 #include "config.h"
16 #include "object-store-ll.h"
17 #include "delta.h"
18 #include "diff.h"
19 #include "dir.h"
20 #include "environment.h"
21 #include "gettext.h"
22 #include "hex.h"
23 #include "xdiff-interface.h"
24 #include "merge-ll.h"
25 #include "lockfile.h"
26 #include "name-hash.h"
27 #include "object-name.h"
28 #include "object-file.h"
29 #include "parse-options.h"
30 #include "path.h"
31 #include "quote.h"
32 #include "read-cache.h"
33 #include "rerere.h"
34 #include "apply.h"
35 #include "entry.h"
36 #include "setup.h"
37 #include "symlinks.h"
38 #include "wildmatch.h"
39 #include "ws.h"
41 struct gitdiff_data {
42 struct strbuf *root;
43 int linenr;
44 int p_value;
47 static void git_apply_config(void)
49 git_config_get_string("apply.whitespace", &apply_default_whitespace);
50 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
51 git_config(git_xmerge_config, NULL);
54 static int parse_whitespace_option(struct apply_state *state, const char *option)
56 if (!option) {
57 state->ws_error_action = warn_on_ws_error;
58 return 0;
60 if (!strcmp(option, "warn")) {
61 state->ws_error_action = warn_on_ws_error;
62 return 0;
64 if (!strcmp(option, "nowarn")) {
65 state->ws_error_action = nowarn_ws_error;
66 return 0;
68 if (!strcmp(option, "error")) {
69 state->ws_error_action = die_on_ws_error;
70 return 0;
72 if (!strcmp(option, "error-all")) {
73 state->ws_error_action = die_on_ws_error;
74 state->squelch_whitespace_errors = 0;
75 return 0;
77 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
78 state->ws_error_action = correct_ws_error;
79 return 0;
82 * Please update $__git_whitespacelist in git-completion.bash,
83 * Documentation/git-apply.txt, and Documentation/git-am.txt
84 * when you add new options.
86 return error(_("unrecognized whitespace option '%s'"), option);
89 static int parse_ignorewhitespace_option(struct apply_state *state,
90 const char *option)
92 if (!option || !strcmp(option, "no") ||
93 !strcmp(option, "false") || !strcmp(option, "never") ||
94 !strcmp(option, "none")) {
95 state->ws_ignore_action = ignore_ws_none;
96 return 0;
98 if (!strcmp(option, "change")) {
99 state->ws_ignore_action = ignore_ws_change;
100 return 0;
102 return error(_("unrecognized whitespace ignore option '%s'"), option);
105 int init_apply_state(struct apply_state *state,
106 struct repository *repo,
107 const char *prefix)
109 memset(state, 0, sizeof(*state));
110 state->prefix = prefix;
111 state->repo = repo;
112 state->apply = 1;
113 state->line_termination = '\n';
114 state->p_value = 1;
115 state->p_context = UINT_MAX;
116 state->squelch_whitespace_errors = 5;
117 state->ws_error_action = warn_on_ws_error;
118 state->ws_ignore_action = ignore_ws_none;
119 state->linenr = 1;
120 string_list_init_nodup(&state->fn_table);
121 string_list_init_nodup(&state->limit_by_name);
122 strset_init(&state->removed_symlinks);
123 strset_init(&state->kept_symlinks);
124 strbuf_init(&state->root, 0);
126 git_apply_config();
127 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
128 return -1;
129 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
130 return -1;
131 return 0;
134 void clear_apply_state(struct apply_state *state)
136 string_list_clear(&state->limit_by_name, 0);
137 strset_clear(&state->removed_symlinks);
138 strset_clear(&state->kept_symlinks);
139 strbuf_release(&state->root);
141 /* &state->fn_table is cleared at the end of apply_patch() */
144 static void mute_routine(const char *msg UNUSED, va_list params UNUSED)
146 /* do nothing */
149 int check_apply_state(struct apply_state *state, int force_apply)
151 int is_not_gitdir = !startup_info->have_repository;
153 if (state->apply_with_reject && state->threeway)
154 return error(_("options '%s' and '%s' cannot be used together"), "--reject", "--3way");
155 if (state->threeway) {
156 if (is_not_gitdir)
157 return error(_("'%s' outside a repository"), "--3way");
158 state->check_index = 1;
160 if (state->apply_with_reject) {
161 state->apply = 1;
162 if (state->apply_verbosity == verbosity_normal)
163 state->apply_verbosity = verbosity_verbose;
165 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
166 state->apply = 0;
167 if (state->check_index && is_not_gitdir)
168 return error(_("'%s' outside a repository"), "--index");
169 if (state->cached) {
170 if (is_not_gitdir)
171 return error(_("'%s' outside a repository"), "--cached");
172 state->check_index = 1;
174 if (state->ita_only && (state->check_index || is_not_gitdir))
175 state->ita_only = 0;
176 if (state->check_index)
177 state->unsafe_paths = 0;
179 if (state->apply_verbosity <= verbosity_silent) {
180 state->saved_error_routine = get_error_routine();
181 state->saved_warn_routine = get_warn_routine();
182 set_error_routine(mute_routine);
183 set_warn_routine(mute_routine);
186 return 0;
189 static void set_default_whitespace_mode(struct apply_state *state)
191 if (!state->whitespace_option && !apply_default_whitespace)
192 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
196 * This represents one "hunk" from a patch, starting with
197 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
198 * patch text is pointed at by patch, and its byte length
199 * is stored in size. leading and trailing are the number
200 * of context lines.
202 struct fragment {
203 unsigned long leading, trailing;
204 unsigned long oldpos, oldlines;
205 unsigned long newpos, newlines;
207 * 'patch' is usually borrowed from buf in apply_patch(),
208 * but some codepaths store an allocated buffer.
210 const char *patch;
211 unsigned free_patch:1,
212 rejected:1;
213 int size;
214 int linenr;
215 struct fragment *next;
219 * When dealing with a binary patch, we reuse "leading" field
220 * to store the type of the binary hunk, either deflated "delta"
221 * or deflated "literal".
223 #define binary_patch_method leading
224 #define BINARY_DELTA_DEFLATED 1
225 #define BINARY_LITERAL_DEFLATED 2
227 static void free_fragment_list(struct fragment *list)
229 while (list) {
230 struct fragment *next = list->next;
231 if (list->free_patch)
232 free((char *)list->patch);
233 free(list);
234 list = next;
238 void release_patch(struct patch *patch)
240 free_fragment_list(patch->fragments);
241 free(patch->def_name);
242 free(patch->old_name);
243 free(patch->new_name);
244 free(patch->result);
247 static void free_patch(struct patch *patch)
249 release_patch(patch);
250 free(patch);
253 static void free_patch_list(struct patch *list)
255 while (list) {
256 struct patch *next = list->next;
257 free_patch(list);
258 list = next;
263 * A line in a file, len-bytes long (includes the terminating LF,
264 * except for an incomplete line at the end if the file ends with
265 * one), and its contents hashes to 'hash'.
267 struct line {
268 size_t len;
269 unsigned hash : 24;
270 unsigned flag : 8;
271 #define LINE_COMMON 1
272 #define LINE_PATCHED 2
276 * This represents a "file", which is an array of "lines".
278 struct image {
279 char *buf;
280 size_t len;
281 size_t nr;
282 size_t alloc;
283 struct line *line_allocated;
284 struct line *line;
287 static uint32_t hash_line(const char *cp, size_t len)
289 size_t i;
290 uint32_t h;
291 for (i = 0, h = 0; i < len; i++) {
292 if (!isspace(cp[i])) {
293 h = h * 3 + (cp[i] & 0xff);
296 return h;
300 * Compare lines s1 of length n1 and s2 of length n2, ignoring
301 * whitespace difference. Returns 1 if they match, 0 otherwise
303 static int fuzzy_matchlines(const char *s1, size_t n1,
304 const char *s2, size_t n2)
306 const char *end1 = s1 + n1;
307 const char *end2 = s2 + n2;
309 /* ignore line endings */
310 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
311 end1--;
312 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
313 end2--;
315 while (s1 < end1 && s2 < end2) {
316 if (isspace(*s1)) {
318 * Skip whitespace. We check on both buffers
319 * because we don't want "a b" to match "ab".
321 if (!isspace(*s2))
322 return 0;
323 while (s1 < end1 && isspace(*s1))
324 s1++;
325 while (s2 < end2 && isspace(*s2))
326 s2++;
327 } else if (*s1++ != *s2++)
328 return 0;
331 /* If we reached the end on one side only, lines don't match. */
332 return s1 == end1 && s2 == end2;
335 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
337 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
338 img->line_allocated[img->nr].len = len;
339 img->line_allocated[img->nr].hash = hash_line(bol, len);
340 img->line_allocated[img->nr].flag = flag;
341 img->nr++;
345 * "buf" has the file contents to be patched (read from various sources).
346 * attach it to "image" and add line-based index to it.
347 * "image" now owns the "buf".
349 static void prepare_image(struct image *image, char *buf, size_t len,
350 int prepare_linetable)
352 const char *cp, *ep;
354 memset(image, 0, sizeof(*image));
355 image->buf = buf;
356 image->len = len;
358 if (!prepare_linetable)
359 return;
361 ep = image->buf + image->len;
362 cp = image->buf;
363 while (cp < ep) {
364 const char *next;
365 for (next = cp; next < ep && *next != '\n'; next++)
367 if (next < ep)
368 next++;
369 add_line_info(image, cp, next - cp, 0);
370 cp = next;
372 image->line = image->line_allocated;
375 static void clear_image(struct image *image)
377 free(image->buf);
378 free(image->line_allocated);
379 memset(image, 0, sizeof(*image));
382 /* fmt must contain _one_ %s and no other substitution */
383 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
385 struct strbuf sb = STRBUF_INIT;
387 if (patch->old_name && patch->new_name &&
388 strcmp(patch->old_name, patch->new_name)) {
389 quote_c_style(patch->old_name, &sb, NULL, 0);
390 strbuf_addstr(&sb, " => ");
391 quote_c_style(patch->new_name, &sb, NULL, 0);
392 } else {
393 const char *n = patch->new_name;
394 if (!n)
395 n = patch->old_name;
396 quote_c_style(n, &sb, NULL, 0);
398 fprintf(output, fmt, sb.buf);
399 fputc('\n', output);
400 strbuf_release(&sb);
403 #define SLOP (16)
406 * apply.c isn't equipped to handle arbitrarily large patches, because
407 * it intermingles `unsigned long` with `int` for the type used to store
408 * buffer lengths.
410 * Only process patches that are just shy of 1 GiB large in order to
411 * avoid any truncation or overflow issues.
413 #define MAX_APPLY_SIZE (1024UL * 1024 * 1023)
415 static int read_patch_file(struct strbuf *sb, int fd)
417 if (strbuf_read(sb, fd, 0) < 0)
418 return error_errno(_("failed to read patch"));
419 else if (sb->len >= MAX_APPLY_SIZE)
420 return error(_("patch too large"));
422 * Make sure that we have some slop in the buffer
423 * so that we can do speculative "memcmp" etc, and
424 * see to it that it is NUL-filled.
426 strbuf_grow(sb, SLOP);
427 memset(sb->buf + sb->len, 0, SLOP);
428 return 0;
431 static unsigned long linelen(const char *buffer, unsigned long size)
433 unsigned long len = 0;
434 while (size--) {
435 len++;
436 if (*buffer++ == '\n')
437 break;
439 return len;
442 static int is_dev_null(const char *str)
444 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
447 #define TERM_SPACE 1
448 #define TERM_TAB 2
450 static int name_terminate(int c, int terminate)
452 if (c == ' ' && !(terminate & TERM_SPACE))
453 return 0;
454 if (c == '\t' && !(terminate & TERM_TAB))
455 return 0;
457 return 1;
460 /* remove double slashes to make --index work with such filenames */
461 static char *squash_slash(char *name)
463 int i = 0, j = 0;
465 if (!name)
466 return NULL;
468 while (name[i]) {
469 if ((name[j++] = name[i++]) == '/')
470 while (name[i] == '/')
471 i++;
473 name[j] = '\0';
474 return name;
477 static char *find_name_gnu(struct strbuf *root,
478 const char *line,
479 int p_value)
481 struct strbuf name = STRBUF_INIT;
482 char *cp;
485 * Proposed "new-style" GNU patch/diff format; see
486 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
488 if (unquote_c_style(&name, line, NULL)) {
489 strbuf_release(&name);
490 return NULL;
493 for (cp = name.buf; p_value; p_value--) {
494 cp = strchr(cp, '/');
495 if (!cp) {
496 strbuf_release(&name);
497 return NULL;
499 cp++;
502 strbuf_remove(&name, 0, cp - name.buf);
503 if (root->len)
504 strbuf_insert(&name, 0, root->buf, root->len);
505 return squash_slash(strbuf_detach(&name, NULL));
508 static size_t sane_tz_len(const char *line, size_t len)
510 const char *tz, *p;
512 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
513 return 0;
514 tz = line + len - strlen(" +0500");
516 if (tz[1] != '+' && tz[1] != '-')
517 return 0;
519 for (p = tz + 2; p != line + len; p++)
520 if (!isdigit(*p))
521 return 0;
523 return line + len - tz;
526 static size_t tz_with_colon_len(const char *line, size_t len)
528 const char *tz, *p;
530 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
531 return 0;
532 tz = line + len - strlen(" +08:00");
534 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
535 return 0;
536 p = tz + 2;
537 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
538 !isdigit(*p++) || !isdigit(*p++))
539 return 0;
541 return line + len - tz;
544 static size_t date_len(const char *line, size_t len)
546 const char *date, *p;
548 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
549 return 0;
550 p = date = line + len - strlen("72-02-05");
552 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
553 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
554 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
555 return 0;
557 if (date - line >= strlen("19") &&
558 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
559 date -= strlen("19");
561 return line + len - date;
564 static size_t short_time_len(const char *line, size_t len)
566 const char *time, *p;
568 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
569 return 0;
570 p = time = line + len - strlen(" 07:01:32");
572 /* Permit 1-digit hours? */
573 if (*p++ != ' ' ||
574 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
575 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
576 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
577 return 0;
579 return line + len - time;
582 static size_t fractional_time_len(const char *line, size_t len)
584 const char *p;
585 size_t n;
587 /* Expected format: 19:41:17.620000023 */
588 if (!len || !isdigit(line[len - 1]))
589 return 0;
590 p = line + len - 1;
592 /* Fractional seconds. */
593 while (p > line && isdigit(*p))
594 p--;
595 if (*p != '.')
596 return 0;
598 /* Hours, minutes, and whole seconds. */
599 n = short_time_len(line, p - line);
600 if (!n)
601 return 0;
603 return line + len - p + n;
606 static size_t trailing_spaces_len(const char *line, size_t len)
608 const char *p;
610 /* Expected format: ' ' x (1 or more) */
611 if (!len || line[len - 1] != ' ')
612 return 0;
614 p = line + len;
615 while (p != line) {
616 p--;
617 if (*p != ' ')
618 return line + len - (p + 1);
621 /* All spaces! */
622 return len;
625 static size_t diff_timestamp_len(const char *line, size_t len)
627 const char *end = line + len;
628 size_t n;
631 * Posix: 2010-07-05 19:41:17
632 * GNU: 2010-07-05 19:41:17.620000023 -0500
635 if (!isdigit(end[-1]))
636 return 0;
638 n = sane_tz_len(line, end - line);
639 if (!n)
640 n = tz_with_colon_len(line, end - line);
641 end -= n;
643 n = short_time_len(line, end - line);
644 if (!n)
645 n = fractional_time_len(line, end - line);
646 end -= n;
648 n = date_len(line, end - line);
649 if (!n) /* No date. Too bad. */
650 return 0;
651 end -= n;
653 if (end == line) /* No space before date. */
654 return 0;
655 if (end[-1] == '\t') { /* Success! */
656 end--;
657 return line + len - end;
659 if (end[-1] != ' ') /* No space before date. */
660 return 0;
662 /* Whitespace damage. */
663 end -= trailing_spaces_len(line, end - line);
664 return line + len - end;
667 static char *find_name_common(struct strbuf *root,
668 const char *line,
669 const char *def,
670 int p_value,
671 const char *end,
672 int terminate)
674 int len;
675 const char *start = NULL;
677 if (p_value == 0)
678 start = line;
679 while (line != end) {
680 char c = *line;
682 if (!end && isspace(c)) {
683 if (c == '\n')
684 break;
685 if (name_terminate(c, terminate))
686 break;
688 line++;
689 if (c == '/' && !--p_value)
690 start = line;
692 if (!start)
693 return squash_slash(xstrdup_or_null(def));
694 len = line - start;
695 if (!len)
696 return squash_slash(xstrdup_or_null(def));
699 * Generally we prefer the shorter name, especially
700 * if the other one is just a variation of that with
701 * something else tacked on to the end (ie "file.orig"
702 * or "file~").
704 if (def) {
705 int deflen = strlen(def);
706 if (deflen < len && !strncmp(start, def, deflen))
707 return squash_slash(xstrdup(def));
710 if (root->len) {
711 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
712 return squash_slash(ret);
715 return squash_slash(xmemdupz(start, len));
718 static char *find_name(struct strbuf *root,
719 const char *line,
720 char *def,
721 int p_value,
722 int terminate)
724 if (*line == '"') {
725 char *name = find_name_gnu(root, line, p_value);
726 if (name)
727 return name;
730 return find_name_common(root, line, def, p_value, NULL, terminate);
733 static char *find_name_traditional(struct strbuf *root,
734 const char *line,
735 char *def,
736 int p_value)
738 size_t len;
739 size_t date_len;
741 if (*line == '"') {
742 char *name = find_name_gnu(root, line, p_value);
743 if (name)
744 return name;
747 len = strchrnul(line, '\n') - line;
748 date_len = diff_timestamp_len(line, len);
749 if (!date_len)
750 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
751 len -= date_len;
753 return find_name_common(root, line, def, p_value, line + len, 0);
757 * Given the string after "--- " or "+++ ", guess the appropriate
758 * p_value for the given patch.
760 static int guess_p_value(struct apply_state *state, const char *nameline)
762 char *name, *cp;
763 int val = -1;
765 if (is_dev_null(nameline))
766 return -1;
767 name = find_name_traditional(&state->root, nameline, NULL, 0);
768 if (!name)
769 return -1;
770 cp = strchr(name, '/');
771 if (!cp)
772 val = 0;
773 else if (state->prefix) {
775 * Does it begin with "a/$our-prefix" and such? Then this is
776 * very likely to apply to our directory.
778 if (starts_with(name, state->prefix))
779 val = count_slashes(state->prefix);
780 else {
781 cp++;
782 if (starts_with(cp, state->prefix))
783 val = count_slashes(state->prefix) + 1;
786 free(name);
787 return val;
791 * Does the ---/+++ line have the POSIX timestamp after the last HT?
792 * GNU diff puts epoch there to signal a creation/deletion event. Is
793 * this such a timestamp?
795 static int has_epoch_timestamp(const char *nameline)
798 * We are only interested in epoch timestamp; any non-zero
799 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
800 * For the same reason, the date must be either 1969-12-31 or
801 * 1970-01-01, and the seconds part must be "00".
803 const char stamp_regexp[] =
804 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
806 "([-+][0-2][0-9]:?[0-5][0-9])\n";
807 const char *timestamp = NULL, *cp, *colon;
808 static regex_t *stamp;
809 regmatch_t m[10];
810 int zoneoffset, epoch_hour, hour, minute;
811 int status;
813 for (cp = nameline; *cp != '\n'; cp++) {
814 if (*cp == '\t')
815 timestamp = cp + 1;
817 if (!timestamp)
818 return 0;
821 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
822 * (west of GMT) or 1970-01-01 (east of GMT)
824 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
825 epoch_hour = 24;
826 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
827 epoch_hour = 0;
828 else
829 return 0;
831 if (!stamp) {
832 stamp = xmalloc(sizeof(*stamp));
833 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
834 warning(_("Cannot prepare timestamp regexp %s"),
835 stamp_regexp);
836 return 0;
840 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
841 if (status) {
842 if (status != REG_NOMATCH)
843 warning(_("regexec returned %d for input: %s"),
844 status, timestamp);
845 return 0;
848 hour = strtol(timestamp, NULL, 10);
849 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
851 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
852 if (*colon == ':')
853 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
854 else
855 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
856 if (timestamp[m[3].rm_so] == '-')
857 zoneoffset = -zoneoffset;
859 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
863 * Get the name etc info from the ---/+++ lines of a traditional patch header
865 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
866 * files, we can happily check the index for a match, but for creating a
867 * new file we should try to match whatever "patch" does. I have no idea.
869 static int parse_traditional_patch(struct apply_state *state,
870 const char *first,
871 const char *second,
872 struct patch *patch)
874 char *name;
876 first += 4; /* skip "--- " */
877 second += 4; /* skip "+++ " */
878 if (!state->p_value_known) {
879 int p, q;
880 p = guess_p_value(state, first);
881 q = guess_p_value(state, second);
882 if (p < 0) p = q;
883 if (0 <= p && p == q) {
884 state->p_value = p;
885 state->p_value_known = 1;
888 if (is_dev_null(first)) {
889 patch->is_new = 1;
890 patch->is_delete = 0;
891 name = find_name_traditional(&state->root, second, NULL, state->p_value);
892 patch->new_name = name;
893 } else if (is_dev_null(second)) {
894 patch->is_new = 0;
895 patch->is_delete = 1;
896 name = find_name_traditional(&state->root, first, NULL, state->p_value);
897 patch->old_name = name;
898 } else {
899 char *first_name;
900 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
901 name = find_name_traditional(&state->root, second, first_name, state->p_value);
902 free(first_name);
903 if (has_epoch_timestamp(first)) {
904 patch->is_new = 1;
905 patch->is_delete = 0;
906 patch->new_name = name;
907 } else if (has_epoch_timestamp(second)) {
908 patch->is_new = 0;
909 patch->is_delete = 1;
910 patch->old_name = name;
911 } else {
912 patch->old_name = name;
913 patch->new_name = xstrdup_or_null(name);
916 if (!name)
917 return error(_("unable to find filename in patch at line %d"), state->linenr);
919 return 0;
922 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
923 const char *line UNUSED,
924 struct patch *patch UNUSED)
926 return 1;
930 * We're anal about diff header consistency, to make
931 * sure that we don't end up having strange ambiguous
932 * patches floating around.
934 * As a result, gitdiff_{old|new}name() will check
935 * their names against any previous information, just
936 * to make sure..
938 #define DIFF_OLD_NAME 0
939 #define DIFF_NEW_NAME 1
941 static int gitdiff_verify_name(struct gitdiff_data *state,
942 const char *line,
943 int isnull,
944 char **name,
945 int side)
947 if (!*name && !isnull) {
948 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
949 return 0;
952 if (*name) {
953 char *another;
954 if (isnull)
955 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
956 *name, state->linenr);
957 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
958 if (!another || strcmp(another, *name)) {
959 free(another);
960 return error((side == DIFF_NEW_NAME) ?
961 _("git apply: bad git-diff - inconsistent new filename on line %d") :
962 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
964 free(another);
965 } else {
966 if (!is_dev_null(line))
967 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
970 return 0;
973 static int gitdiff_oldname(struct gitdiff_data *state,
974 const char *line,
975 struct patch *patch)
977 return gitdiff_verify_name(state, line,
978 patch->is_new, &patch->old_name,
979 DIFF_OLD_NAME);
982 static int gitdiff_newname(struct gitdiff_data *state,
983 const char *line,
984 struct patch *patch)
986 return gitdiff_verify_name(state, line,
987 patch->is_delete, &patch->new_name,
988 DIFF_NEW_NAME);
991 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
993 char *end;
994 *mode = strtoul(line, &end, 8);
995 if (end == line || !isspace(*end))
996 return error(_("invalid mode on line %d: %s"), linenr, line);
997 return 0;
1000 static int gitdiff_oldmode(struct gitdiff_data *state,
1001 const char *line,
1002 struct patch *patch)
1004 return parse_mode_line(line, state->linenr, &patch->old_mode);
1007 static int gitdiff_newmode(struct gitdiff_data *state,
1008 const char *line,
1009 struct patch *patch)
1011 return parse_mode_line(line, state->linenr, &patch->new_mode);
1014 static int gitdiff_delete(struct gitdiff_data *state,
1015 const char *line,
1016 struct patch *patch)
1018 patch->is_delete = 1;
1019 free(patch->old_name);
1020 patch->old_name = xstrdup_or_null(patch->def_name);
1021 return gitdiff_oldmode(state, line, patch);
1024 static int gitdiff_newfile(struct gitdiff_data *state,
1025 const char *line,
1026 struct patch *patch)
1028 patch->is_new = 1;
1029 free(patch->new_name);
1030 patch->new_name = xstrdup_or_null(patch->def_name);
1031 return gitdiff_newmode(state, line, patch);
1034 static int gitdiff_copysrc(struct gitdiff_data *state,
1035 const char *line,
1036 struct patch *patch)
1038 patch->is_copy = 1;
1039 free(patch->old_name);
1040 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1041 return 0;
1044 static int gitdiff_copydst(struct gitdiff_data *state,
1045 const char *line,
1046 struct patch *patch)
1048 patch->is_copy = 1;
1049 free(patch->new_name);
1050 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1051 return 0;
1054 static int gitdiff_renamesrc(struct gitdiff_data *state,
1055 const char *line,
1056 struct patch *patch)
1058 patch->is_rename = 1;
1059 free(patch->old_name);
1060 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1061 return 0;
1064 static int gitdiff_renamedst(struct gitdiff_data *state,
1065 const char *line,
1066 struct patch *patch)
1068 patch->is_rename = 1;
1069 free(patch->new_name);
1070 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1071 return 0;
1074 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1075 const char *line,
1076 struct patch *patch)
1078 unsigned long val = strtoul(line, NULL, 10);
1079 if (val <= 100)
1080 patch->score = val;
1081 return 0;
1084 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1085 const char *line,
1086 struct patch *patch)
1088 unsigned long val = strtoul(line, NULL, 10);
1089 if (val <= 100)
1090 patch->score = val;
1091 return 0;
1094 static int gitdiff_index(struct gitdiff_data *state,
1095 const char *line,
1096 struct patch *patch)
1099 * index line is N hexadecimal, "..", N hexadecimal,
1100 * and optional space with octal mode.
1102 const char *ptr, *eol;
1103 int len;
1104 const unsigned hexsz = the_hash_algo->hexsz;
1106 ptr = strchr(line, '.');
1107 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1108 return 0;
1109 len = ptr - line;
1110 memcpy(patch->old_oid_prefix, line, len);
1111 patch->old_oid_prefix[len] = 0;
1113 line = ptr + 2;
1114 ptr = strchr(line, ' ');
1115 eol = strchrnul(line, '\n');
1117 if (!ptr || eol < ptr)
1118 ptr = eol;
1119 len = ptr - line;
1121 if (hexsz < len)
1122 return 0;
1123 memcpy(patch->new_oid_prefix, line, len);
1124 patch->new_oid_prefix[len] = 0;
1125 if (*ptr == ' ')
1126 return gitdiff_oldmode(state, ptr + 1, patch);
1127 return 0;
1131 * This is normal for a diff that doesn't change anything: we'll fall through
1132 * into the next diff. Tell the parser to break out.
1134 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1135 const char *line UNUSED,
1136 struct patch *patch UNUSED)
1138 return 1;
1142 * Skip p_value leading components from "line"; as we do not accept
1143 * absolute paths, return NULL in that case.
1145 static const char *skip_tree_prefix(int p_value,
1146 const char *line,
1147 int llen)
1149 int nslash;
1150 int i;
1152 if (!p_value)
1153 return (llen && line[0] == '/') ? NULL : line;
1155 nslash = p_value;
1156 for (i = 0; i < llen; i++) {
1157 int ch = line[i];
1158 if (ch == '/' && --nslash <= 0)
1159 return (i == 0) ? NULL : &line[i + 1];
1161 return NULL;
1165 * This is to extract the same name that appears on "diff --git"
1166 * line. We do not find and return anything if it is a rename
1167 * patch, and it is OK because we will find the name elsewhere.
1168 * We need to reliably find name only when it is mode-change only,
1169 * creation or deletion of an empty file. In any of these cases,
1170 * both sides are the same name under a/ and b/ respectively.
1172 static char *git_header_name(int p_value,
1173 const char *line,
1174 int llen)
1176 const char *name;
1177 const char *second = NULL;
1178 size_t len, line_len;
1180 line += strlen("diff --git ");
1181 llen -= strlen("diff --git ");
1183 if (*line == '"') {
1184 const char *cp;
1185 struct strbuf first = STRBUF_INIT;
1186 struct strbuf sp = STRBUF_INIT;
1188 if (unquote_c_style(&first, line, &second))
1189 goto free_and_fail1;
1191 /* strip the a/b prefix including trailing slash */
1192 cp = skip_tree_prefix(p_value, first.buf, first.len);
1193 if (!cp)
1194 goto free_and_fail1;
1195 strbuf_remove(&first, 0, cp - first.buf);
1198 * second points at one past closing dq of name.
1199 * find the second name.
1201 while ((second < line + llen) && isspace(*second))
1202 second++;
1204 if (line + llen <= second)
1205 goto free_and_fail1;
1206 if (*second == '"') {
1207 if (unquote_c_style(&sp, second, NULL))
1208 goto free_and_fail1;
1209 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1210 if (!cp)
1211 goto free_and_fail1;
1212 /* They must match, otherwise ignore */
1213 if (strcmp(cp, first.buf))
1214 goto free_and_fail1;
1215 strbuf_release(&sp);
1216 return strbuf_detach(&first, NULL);
1219 /* unquoted second */
1220 cp = skip_tree_prefix(p_value, second, line + llen - second);
1221 if (!cp)
1222 goto free_and_fail1;
1223 if (line + llen - cp != first.len ||
1224 memcmp(first.buf, cp, first.len))
1225 goto free_and_fail1;
1226 return strbuf_detach(&first, NULL);
1228 free_and_fail1:
1229 strbuf_release(&first);
1230 strbuf_release(&sp);
1231 return NULL;
1234 /* unquoted first name */
1235 name = skip_tree_prefix(p_value, line, llen);
1236 if (!name)
1237 return NULL;
1240 * since the first name is unquoted, a dq if exists must be
1241 * the beginning of the second name.
1243 for (second = name; second < line + llen; second++) {
1244 if (*second == '"') {
1245 struct strbuf sp = STRBUF_INIT;
1246 const char *np;
1248 if (unquote_c_style(&sp, second, NULL))
1249 goto free_and_fail2;
1251 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1252 if (!np)
1253 goto free_and_fail2;
1255 len = sp.buf + sp.len - np;
1256 if (len < second - name &&
1257 !strncmp(np, name, len) &&
1258 isspace(name[len])) {
1259 /* Good */
1260 strbuf_remove(&sp, 0, np - sp.buf);
1261 return strbuf_detach(&sp, NULL);
1264 free_and_fail2:
1265 strbuf_release(&sp);
1266 return NULL;
1271 * Accept a name only if it shows up twice, exactly the same
1272 * form.
1274 second = strchr(name, '\n');
1275 if (!second)
1276 return NULL;
1277 line_len = second - name;
1278 for (len = 0 ; ; len++) {
1279 switch (name[len]) {
1280 default:
1281 continue;
1282 case '\n':
1283 return NULL;
1284 case '\t': case ' ':
1286 * Is this the separator between the preimage
1287 * and the postimage pathname? Again, we are
1288 * only interested in the case where there is
1289 * no rename, as this is only to set def_name
1290 * and a rename patch has the names elsewhere
1291 * in an unambiguous form.
1293 if (!name[len + 1])
1294 return NULL; /* no postimage name */
1295 second = skip_tree_prefix(p_value, name + len + 1,
1296 line_len - (len + 1));
1298 * If we are at the SP at the end of a directory,
1299 * skip_tree_prefix() may return NULL as that makes
1300 * it appears as if we have an absolute path.
1301 * Keep going to find another SP.
1303 if (!second)
1304 continue;
1307 * Does len bytes starting at "name" and "second"
1308 * (that are separated by one HT or SP we just
1309 * found) exactly match?
1311 if (second[len] == '\n' && !strncmp(name, second, len))
1312 return xmemdupz(name, len);
1317 static int check_header_line(int linenr, struct patch *patch)
1319 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1320 (patch->is_rename == 1) + (patch->is_copy == 1);
1321 if (extensions > 1)
1322 return error(_("inconsistent header lines %d and %d"),
1323 patch->extension_linenr, linenr);
1324 if (extensions && !patch->extension_linenr)
1325 patch->extension_linenr = linenr;
1326 return 0;
1329 int parse_git_diff_header(struct strbuf *root,
1330 int *linenr,
1331 int p_value,
1332 const char *line,
1333 int len,
1334 unsigned int size,
1335 struct patch *patch)
1337 unsigned long offset;
1338 struct gitdiff_data parse_hdr_state;
1340 /* A git diff has explicit new/delete information, so we don't guess */
1341 patch->is_new = 0;
1342 patch->is_delete = 0;
1345 * Some things may not have the old name in the
1346 * rest of the headers anywhere (pure mode changes,
1347 * or removing or adding empty files), so we get
1348 * the default name from the header.
1350 patch->def_name = git_header_name(p_value, line, len);
1351 if (patch->def_name && root->len) {
1352 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1353 free(patch->def_name);
1354 patch->def_name = s;
1357 line += len;
1358 size -= len;
1359 (*linenr)++;
1360 parse_hdr_state.root = root;
1361 parse_hdr_state.linenr = *linenr;
1362 parse_hdr_state.p_value = p_value;
1364 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1365 static const struct opentry {
1366 const char *str;
1367 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1368 } optable[] = {
1369 { "@@ -", gitdiff_hdrend },
1370 { "--- ", gitdiff_oldname },
1371 { "+++ ", gitdiff_newname },
1372 { "old mode ", gitdiff_oldmode },
1373 { "new mode ", gitdiff_newmode },
1374 { "deleted file mode ", gitdiff_delete },
1375 { "new file mode ", gitdiff_newfile },
1376 { "copy from ", gitdiff_copysrc },
1377 { "copy to ", gitdiff_copydst },
1378 { "rename old ", gitdiff_renamesrc },
1379 { "rename new ", gitdiff_renamedst },
1380 { "rename from ", gitdiff_renamesrc },
1381 { "rename to ", gitdiff_renamedst },
1382 { "similarity index ", gitdiff_similarity },
1383 { "dissimilarity index ", gitdiff_dissimilarity },
1384 { "index ", gitdiff_index },
1385 { "", gitdiff_unrecognized },
1387 int i;
1389 len = linelen(line, size);
1390 if (!len || line[len-1] != '\n')
1391 break;
1392 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1393 const struct opentry *p = optable + i;
1394 int oplen = strlen(p->str);
1395 int res;
1396 if (len < oplen || memcmp(p->str, line, oplen))
1397 continue;
1398 res = p->fn(&parse_hdr_state, line + oplen, patch);
1399 if (res < 0)
1400 return -1;
1401 if (check_header_line(*linenr, patch))
1402 return -1;
1403 if (res > 0)
1404 goto done;
1405 break;
1409 done:
1410 if (!patch->old_name && !patch->new_name) {
1411 if (!patch->def_name) {
1412 error(Q_("git diff header lacks filename information when removing "
1413 "%d leading pathname component (line %d)",
1414 "git diff header lacks filename information when removing "
1415 "%d leading pathname components (line %d)",
1416 parse_hdr_state.p_value),
1417 parse_hdr_state.p_value, *linenr);
1418 return -128;
1420 patch->old_name = xstrdup(patch->def_name);
1421 patch->new_name = xstrdup(patch->def_name);
1423 if ((!patch->new_name && !patch->is_delete) ||
1424 (!patch->old_name && !patch->is_new)) {
1425 error(_("git diff header lacks filename information "
1426 "(line %d)"), *linenr);
1427 return -128;
1429 patch->is_toplevel_relative = 1;
1430 return offset;
1433 static int parse_num(const char *line, unsigned long *p)
1435 char *ptr;
1437 if (!isdigit(*line))
1438 return 0;
1439 *p = strtoul(line, &ptr, 10);
1440 return ptr - line;
1443 static int parse_range(const char *line, int len, int offset, const char *expect,
1444 unsigned long *p1, unsigned long *p2)
1446 int digits, ex;
1448 if (offset < 0 || offset >= len)
1449 return -1;
1450 line += offset;
1451 len -= offset;
1453 digits = parse_num(line, p1);
1454 if (!digits)
1455 return -1;
1457 offset += digits;
1458 line += digits;
1459 len -= digits;
1461 *p2 = 1;
1462 if (*line == ',') {
1463 digits = parse_num(line+1, p2);
1464 if (!digits)
1465 return -1;
1467 offset += digits+1;
1468 line += digits+1;
1469 len -= digits+1;
1472 ex = strlen(expect);
1473 if (ex > len)
1474 return -1;
1475 if (memcmp(line, expect, ex))
1476 return -1;
1478 return offset + ex;
1481 static void recount_diff(const char *line, int size, struct fragment *fragment)
1483 int oldlines = 0, newlines = 0, ret = 0;
1485 if (size < 1) {
1486 warning("recount: ignore empty hunk");
1487 return;
1490 for (;;) {
1491 int len = linelen(line, size);
1492 size -= len;
1493 line += len;
1495 if (size < 1)
1496 break;
1498 switch (*line) {
1499 case ' ': case '\n':
1500 newlines++;
1501 /* fall through */
1502 case '-':
1503 oldlines++;
1504 continue;
1505 case '+':
1506 newlines++;
1507 continue;
1508 case '\\':
1509 continue;
1510 case '@':
1511 ret = size < 3 || !starts_with(line, "@@ ");
1512 break;
1513 case 'd':
1514 ret = size < 5 || !starts_with(line, "diff ");
1515 break;
1516 default:
1517 ret = -1;
1518 break;
1520 if (ret) {
1521 warning(_("recount: unexpected line: %.*s"),
1522 (int)linelen(line, size), line);
1523 return;
1525 break;
1527 fragment->oldlines = oldlines;
1528 fragment->newlines = newlines;
1532 * Parse a unified diff fragment header of the
1533 * form "@@ -a,b +c,d @@"
1535 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1537 int offset;
1539 if (!len || line[len-1] != '\n')
1540 return -1;
1542 /* Figure out the number of lines in a fragment */
1543 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1544 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1546 return offset;
1550 * Find file diff header
1552 * Returns:
1553 * -1 if no header was found
1554 * -128 in case of error
1555 * the size of the header in bytes (called "offset") otherwise
1557 static int find_header(struct apply_state *state,
1558 const char *line,
1559 unsigned long size,
1560 int *hdrsize,
1561 struct patch *patch)
1563 unsigned long offset, len;
1565 patch->is_toplevel_relative = 0;
1566 patch->is_rename = patch->is_copy = 0;
1567 patch->is_new = patch->is_delete = -1;
1568 patch->old_mode = patch->new_mode = 0;
1569 patch->old_name = patch->new_name = NULL;
1570 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1571 unsigned long nextlen;
1573 len = linelen(line, size);
1574 if (!len)
1575 break;
1577 /* Testing this early allows us to take a few shortcuts.. */
1578 if (len < 6)
1579 continue;
1582 * Make sure we don't find any unconnected patch fragments.
1583 * That's a sign that we didn't find a header, and that a
1584 * patch has become corrupted/broken up.
1586 if (!memcmp("@@ -", line, 4)) {
1587 struct fragment dummy;
1588 if (parse_fragment_header(line, len, &dummy) < 0)
1589 continue;
1590 error(_("patch fragment without header at line %d: %.*s"),
1591 state->linenr, (int)len-1, line);
1592 return -128;
1595 if (size < len + 6)
1596 break;
1599 * Git patch? It might not have a real patch, just a rename
1600 * or mode change, so we handle that specially
1602 if (!memcmp("diff --git ", line, 11)) {
1603 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1604 state->p_value, line, len,
1605 size, patch);
1606 if (git_hdr_len < 0)
1607 return -128;
1608 if (git_hdr_len <= len)
1609 continue;
1610 *hdrsize = git_hdr_len;
1611 return offset;
1614 /* --- followed by +++ ? */
1615 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1616 continue;
1619 * We only accept unified patches, so we want it to
1620 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1621 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1623 nextlen = linelen(line + len, size - len);
1624 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1625 continue;
1627 /* Ok, we'll consider it a patch */
1628 if (parse_traditional_patch(state, line, line+len, patch))
1629 return -128;
1630 *hdrsize = len + nextlen;
1631 state->linenr += 2;
1632 return offset;
1634 return -1;
1637 static void record_ws_error(struct apply_state *state,
1638 unsigned result,
1639 const char *line,
1640 int len,
1641 int linenr)
1643 char *err;
1645 if (!result)
1646 return;
1648 state->whitespace_error++;
1649 if (state->squelch_whitespace_errors &&
1650 state->squelch_whitespace_errors < state->whitespace_error)
1651 return;
1653 err = whitespace_error_string(result);
1654 if (state->apply_verbosity > verbosity_silent)
1655 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1656 state->patch_input_file, linenr, err, len, line);
1657 free(err);
1660 static void check_whitespace(struct apply_state *state,
1661 const char *line,
1662 int len,
1663 unsigned ws_rule)
1665 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1667 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1671 * Check if the patch has context lines with CRLF or
1672 * the patch wants to remove lines with CRLF.
1674 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1676 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1677 patch->ws_rule |= WS_CR_AT_EOL;
1678 patch->crlf_in_old = 1;
1684 * Parse a unified diff. Note that this really needs to parse each
1685 * fragment separately, since the only way to know the difference
1686 * between a "---" that is part of a patch, and a "---" that starts
1687 * the next patch is to look at the line counts..
1689 static int parse_fragment(struct apply_state *state,
1690 const char *line,
1691 unsigned long size,
1692 struct patch *patch,
1693 struct fragment *fragment)
1695 int added, deleted;
1696 int len = linelen(line, size), offset;
1697 unsigned long oldlines, newlines;
1698 unsigned long leading, trailing;
1700 offset = parse_fragment_header(line, len, fragment);
1701 if (offset < 0)
1702 return -1;
1703 if (offset > 0 && patch->recount)
1704 recount_diff(line + offset, size - offset, fragment);
1705 oldlines = fragment->oldlines;
1706 newlines = fragment->newlines;
1707 leading = 0;
1708 trailing = 0;
1710 /* Parse the thing.. */
1711 line += len;
1712 size -= len;
1713 state->linenr++;
1714 added = deleted = 0;
1715 for (offset = len;
1716 0 < size;
1717 offset += len, size -= len, line += len, state->linenr++) {
1718 if (!oldlines && !newlines)
1719 break;
1720 len = linelen(line, size);
1721 if (!len || line[len-1] != '\n')
1722 return -1;
1723 switch (*line) {
1724 default:
1725 return -1;
1726 case '\n': /* newer GNU diff, an empty context line */
1727 case ' ':
1728 oldlines--;
1729 newlines--;
1730 if (!deleted && !added)
1731 leading++;
1732 trailing++;
1733 check_old_for_crlf(patch, line, len);
1734 if (!state->apply_in_reverse &&
1735 state->ws_error_action == correct_ws_error)
1736 check_whitespace(state, line, len, patch->ws_rule);
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 deleted++;
1745 oldlines--;
1746 trailing = 0;
1747 break;
1748 case '+':
1749 if (state->apply_in_reverse)
1750 check_old_for_crlf(patch, line, len);
1751 if (!state->apply_in_reverse &&
1752 state->ws_error_action != nowarn_ws_error)
1753 check_whitespace(state, line, len, patch->ws_rule);
1754 added++;
1755 newlines--;
1756 trailing = 0;
1757 break;
1760 * We allow "\ No newline at end of file". Depending
1761 * on locale settings when the patch was produced we
1762 * don't know what this line looks like. The only
1763 * thing we do know is that it begins with "\ ".
1764 * Checking for 12 is just for sanity check -- any
1765 * l10n of "\ No newline..." is at least that long.
1767 case '\\':
1768 if (len < 12 || memcmp(line, "\\ ", 2))
1769 return -1;
1770 break;
1773 if (oldlines || newlines)
1774 return -1;
1775 if (!patch->recount && !deleted && !added)
1776 return -1;
1778 fragment->leading = leading;
1779 fragment->trailing = trailing;
1782 * If a fragment ends with an incomplete line, we failed to include
1783 * it in the above loop because we hit oldlines == newlines == 0
1784 * before seeing it.
1786 if (12 < size && !memcmp(line, "\\ ", 2))
1787 offset += linelen(line, size);
1789 patch->lines_added += added;
1790 patch->lines_deleted += deleted;
1792 if (0 < patch->is_new && oldlines)
1793 return error(_("new file depends on old contents"));
1794 if (0 < patch->is_delete && newlines)
1795 return error(_("deleted file still has contents"));
1796 return offset;
1800 * We have seen "diff --git a/... b/..." header (or a traditional patch
1801 * header). Read hunks that belong to this patch into fragments and hang
1802 * them to the given patch structure.
1804 * The (fragment->patch, fragment->size) pair points into the memory given
1805 * by the caller, not a copy, when we return.
1807 * Returns:
1808 * -1 in case of error,
1809 * the number of bytes in the patch otherwise.
1811 static int parse_single_patch(struct apply_state *state,
1812 const char *line,
1813 unsigned long size,
1814 struct patch *patch)
1816 unsigned long offset = 0;
1817 unsigned long oldlines = 0, newlines = 0, context = 0;
1818 struct fragment **fragp = &patch->fragments;
1820 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1821 struct fragment *fragment;
1822 int len;
1824 CALLOC_ARRAY(fragment, 1);
1825 fragment->linenr = state->linenr;
1826 len = parse_fragment(state, line, size, patch, fragment);
1827 if (len <= 0) {
1828 free(fragment);
1829 return error(_("corrupt patch at line %d"), state->linenr);
1831 fragment->patch = line;
1832 fragment->size = len;
1833 oldlines += fragment->oldlines;
1834 newlines += fragment->newlines;
1835 context += fragment->leading + fragment->trailing;
1837 *fragp = fragment;
1838 fragp = &fragment->next;
1840 offset += len;
1841 line += len;
1842 size -= len;
1846 * If something was removed (i.e. we have old-lines) it cannot
1847 * be creation, and if something was added it cannot be
1848 * deletion. However, the reverse is not true; --unified=0
1849 * patches that only add are not necessarily creation even
1850 * though they do not have any old lines, and ones that only
1851 * delete are not necessarily deletion.
1853 * Unfortunately, a real creation/deletion patch do _not_ have
1854 * any context line by definition, so we cannot safely tell it
1855 * apart with --unified=0 insanity. At least if the patch has
1856 * more than one hunk it is not creation or deletion.
1858 if (patch->is_new < 0 &&
1859 (oldlines || (patch->fragments && patch->fragments->next)))
1860 patch->is_new = 0;
1861 if (patch->is_delete < 0 &&
1862 (newlines || (patch->fragments && patch->fragments->next)))
1863 patch->is_delete = 0;
1865 if (0 < patch->is_new && oldlines)
1866 return error(_("new file %s depends on old contents"), patch->new_name);
1867 if (0 < patch->is_delete && newlines)
1868 return error(_("deleted file %s still has contents"), patch->old_name);
1869 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1870 fprintf_ln(stderr,
1871 _("** warning: "
1872 "file %s becomes empty but is not deleted"),
1873 patch->new_name);
1875 return offset;
1878 static inline int metadata_changes(struct patch *patch)
1880 return patch->is_rename > 0 ||
1881 patch->is_copy > 0 ||
1882 patch->is_new > 0 ||
1883 patch->is_delete ||
1884 (patch->old_mode && patch->new_mode &&
1885 patch->old_mode != patch->new_mode);
1888 static char *inflate_it(const void *data, unsigned long size,
1889 unsigned long inflated_size)
1891 git_zstream stream;
1892 void *out;
1893 int st;
1895 memset(&stream, 0, sizeof(stream));
1897 stream.next_in = (unsigned char *)data;
1898 stream.avail_in = size;
1899 stream.next_out = out = xmalloc(inflated_size);
1900 stream.avail_out = inflated_size;
1901 git_inflate_init(&stream);
1902 st = git_inflate(&stream, Z_FINISH);
1903 git_inflate_end(&stream);
1904 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1905 free(out);
1906 return NULL;
1908 return out;
1912 * Read a binary hunk and return a new fragment; fragment->patch
1913 * points at an allocated memory that the caller must free, so
1914 * it is marked as "->free_patch = 1".
1916 static struct fragment *parse_binary_hunk(struct apply_state *state,
1917 char **buf_p,
1918 unsigned long *sz_p,
1919 int *status_p,
1920 int *used_p)
1923 * Expect a line that begins with binary patch method ("literal"
1924 * or "delta"), followed by the length of data before deflating.
1925 * a sequence of 'length-byte' followed by base-85 encoded data
1926 * should follow, terminated by a newline.
1928 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1929 * and we would limit the patch line to 66 characters,
1930 * so one line can fit up to 13 groups that would decode
1931 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1932 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1934 int llen, used;
1935 unsigned long size = *sz_p;
1936 char *buffer = *buf_p;
1937 int patch_method;
1938 unsigned long origlen;
1939 char *data = NULL;
1940 int hunk_size = 0;
1941 struct fragment *frag;
1943 llen = linelen(buffer, size);
1944 used = llen;
1946 *status_p = 0;
1948 if (starts_with(buffer, "delta ")) {
1949 patch_method = BINARY_DELTA_DEFLATED;
1950 origlen = strtoul(buffer + 6, NULL, 10);
1952 else if (starts_with(buffer, "literal ")) {
1953 patch_method = BINARY_LITERAL_DEFLATED;
1954 origlen = strtoul(buffer + 8, NULL, 10);
1956 else
1957 return NULL;
1959 state->linenr++;
1960 buffer += llen;
1961 size -= llen;
1962 while (1) {
1963 int byte_length, max_byte_length, newsize;
1964 llen = linelen(buffer, size);
1965 used += llen;
1966 state->linenr++;
1967 if (llen == 1) {
1968 /* consume the blank line */
1969 buffer++;
1970 size--;
1971 break;
1974 * Minimum line is "A00000\n" which is 7-byte long,
1975 * and the line length must be multiple of 5 plus 2.
1977 if ((llen < 7) || (llen-2) % 5)
1978 goto corrupt;
1979 max_byte_length = (llen - 2) / 5 * 4;
1980 byte_length = *buffer;
1981 if ('A' <= byte_length && byte_length <= 'Z')
1982 byte_length = byte_length - 'A' + 1;
1983 else if ('a' <= byte_length && byte_length <= 'z')
1984 byte_length = byte_length - 'a' + 27;
1985 else
1986 goto corrupt;
1987 /* if the input length was not multiple of 4, we would
1988 * have filler at the end but the filler should never
1989 * exceed 3 bytes
1991 if (max_byte_length < byte_length ||
1992 byte_length <= max_byte_length - 4)
1993 goto corrupt;
1994 newsize = hunk_size + byte_length;
1995 data = xrealloc(data, newsize);
1996 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1997 goto corrupt;
1998 hunk_size = newsize;
1999 buffer += llen;
2000 size -= llen;
2003 CALLOC_ARRAY(frag, 1);
2004 frag->patch = inflate_it(data, hunk_size, origlen);
2005 frag->free_patch = 1;
2006 if (!frag->patch)
2007 goto corrupt;
2008 free(data);
2009 frag->size = origlen;
2010 *buf_p = buffer;
2011 *sz_p = size;
2012 *used_p = used;
2013 frag->binary_patch_method = patch_method;
2014 return frag;
2016 corrupt:
2017 free(data);
2018 *status_p = -1;
2019 error(_("corrupt binary patch at line %d: %.*s"),
2020 state->linenr-1, llen-1, buffer);
2021 return NULL;
2025 * Returns:
2026 * -1 in case of error,
2027 * the length of the parsed binary patch otherwise
2029 static int parse_binary(struct apply_state *state,
2030 char *buffer,
2031 unsigned long size,
2032 struct patch *patch)
2035 * We have read "GIT binary patch\n"; what follows is a line
2036 * that says the patch method (currently, either "literal" or
2037 * "delta") and the length of data before deflating; a
2038 * sequence of 'length-byte' followed by base-85 encoded data
2039 * follows.
2041 * When a binary patch is reversible, there is another binary
2042 * hunk in the same format, starting with patch method (either
2043 * "literal" or "delta") with the length of data, and a sequence
2044 * of length-byte + base-85 encoded data, terminated with another
2045 * empty line. This data, when applied to the postimage, produces
2046 * the preimage.
2048 struct fragment *forward;
2049 struct fragment *reverse;
2050 int status;
2051 int used, used_1;
2053 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2054 if (!forward && !status)
2055 /* there has to be one hunk (forward hunk) */
2056 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2057 if (status)
2058 /* otherwise we already gave an error message */
2059 return status;
2061 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2062 if (reverse)
2063 used += used_1;
2064 else if (status) {
2066 * Not having reverse hunk is not an error, but having
2067 * a corrupt reverse hunk is.
2069 free((void*) forward->patch);
2070 free(forward);
2071 return status;
2073 forward->next = reverse;
2074 patch->fragments = forward;
2075 patch->is_binary = 1;
2076 return used;
2079 static void prefix_one(struct apply_state *state, char **name)
2081 char *old_name = *name;
2082 if (!old_name)
2083 return;
2084 *name = prefix_filename(state->prefix, *name);
2085 free(old_name);
2088 static void prefix_patch(struct apply_state *state, struct patch *p)
2090 if (!state->prefix || p->is_toplevel_relative)
2091 return;
2092 prefix_one(state, &p->new_name);
2093 prefix_one(state, &p->old_name);
2097 * include/exclude
2100 static void add_name_limit(struct apply_state *state,
2101 const char *name,
2102 int exclude)
2104 struct string_list_item *it;
2106 it = string_list_append(&state->limit_by_name, name);
2107 it->util = exclude ? NULL : (void *) 1;
2110 static int use_patch(struct apply_state *state, struct patch *p)
2112 const char *pathname = p->new_name ? p->new_name : p->old_name;
2113 int i;
2115 /* Paths outside are not touched regardless of "--include" */
2116 if (state->prefix && *state->prefix) {
2117 const char *rest;
2118 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2119 return 0;
2122 /* See if it matches any of exclude/include rule */
2123 for (i = 0; i < state->limit_by_name.nr; i++) {
2124 struct string_list_item *it = &state->limit_by_name.items[i];
2125 if (!wildmatch(it->string, pathname, 0))
2126 return (it->util != NULL);
2130 * If we had any include, a path that does not match any rule is
2131 * not used. Otherwise, we saw bunch of exclude rules (or none)
2132 * and such a path is used.
2134 return !state->has_include;
2138 * Read the patch text in "buffer" that extends for "size" bytes; stop
2139 * reading after seeing a single patch (i.e. changes to a single file).
2140 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2142 * Returns:
2143 * -1 if no header was found or parse_binary() failed,
2144 * -128 on another error,
2145 * the number of bytes consumed otherwise,
2146 * so that the caller can call us again for the next patch.
2148 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2150 int hdrsize, patchsize;
2151 int offset = find_header(state, buffer, size, &hdrsize, patch);
2153 if (offset < 0)
2154 return offset;
2156 prefix_patch(state, patch);
2158 if (!use_patch(state, patch))
2159 patch->ws_rule = 0;
2160 else if (patch->new_name)
2161 patch->ws_rule = whitespace_rule(state->repo->index,
2162 patch->new_name);
2163 else
2164 patch->ws_rule = whitespace_rule(state->repo->index,
2165 patch->old_name);
2167 patchsize = parse_single_patch(state,
2168 buffer + offset + hdrsize,
2169 size - offset - hdrsize,
2170 patch);
2172 if (patchsize < 0)
2173 return -128;
2175 if (!patchsize) {
2176 static const char git_binary[] = "GIT binary patch\n";
2177 int hd = hdrsize + offset;
2178 unsigned long llen = linelen(buffer + hd, size - hd);
2180 if (llen == sizeof(git_binary) - 1 &&
2181 !memcmp(git_binary, buffer + hd, llen)) {
2182 int used;
2183 state->linenr++;
2184 used = parse_binary(state, buffer + hd + llen,
2185 size - hd - llen, patch);
2186 if (used < 0)
2187 return -1;
2188 if (used)
2189 patchsize = used + llen;
2190 else
2191 patchsize = 0;
2193 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2194 static const char *binhdr[] = {
2195 "Binary files ",
2196 "Files ",
2197 NULL,
2199 int i;
2200 for (i = 0; binhdr[i]; i++) {
2201 int len = strlen(binhdr[i]);
2202 if (len < size - hd &&
2203 !memcmp(binhdr[i], buffer + hd, len)) {
2204 state->linenr++;
2205 patch->is_binary = 1;
2206 patchsize = llen;
2207 break;
2212 /* Empty patch cannot be applied if it is a text patch
2213 * without metadata change. A binary patch appears
2214 * empty to us here.
2216 if ((state->apply || state->check) &&
2217 (!patch->is_binary && !metadata_changes(patch))) {
2218 error(_("patch with only garbage at line %d"), state->linenr);
2219 return -128;
2223 return offset + hdrsize + patchsize;
2226 static void reverse_patches(struct patch *p)
2228 for (; p; p = p->next) {
2229 struct fragment *frag = p->fragments;
2231 SWAP(p->new_name, p->old_name);
2232 if (p->new_mode)
2233 SWAP(p->new_mode, p->old_mode);
2234 SWAP(p->is_new, p->is_delete);
2235 SWAP(p->lines_added, p->lines_deleted);
2236 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2238 for (; frag; frag = frag->next) {
2239 SWAP(frag->newpos, frag->oldpos);
2240 SWAP(frag->newlines, frag->oldlines);
2245 static const char pluses[] =
2246 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2247 static const char minuses[]=
2248 "----------------------------------------------------------------------";
2250 static void show_stats(struct apply_state *state, struct patch *patch)
2252 struct strbuf qname = STRBUF_INIT;
2253 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2254 int max, add, del;
2256 quote_c_style(cp, &qname, NULL, 0);
2259 * "scale" the filename
2261 max = state->max_len;
2262 if (max > 50)
2263 max = 50;
2265 if (qname.len > max) {
2266 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2267 if (!cp)
2268 cp = qname.buf + qname.len + 3 - max;
2269 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2272 if (patch->is_binary) {
2273 printf(" %-*s | Bin\n", max, qname.buf);
2274 strbuf_release(&qname);
2275 return;
2278 printf(" %-*s |", max, qname.buf);
2279 strbuf_release(&qname);
2282 * scale the add/delete
2284 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2285 add = patch->lines_added;
2286 del = patch->lines_deleted;
2288 if (state->max_change > 0) {
2289 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2290 add = (add * max + state->max_change / 2) / state->max_change;
2291 del = total - add;
2293 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2294 add, pluses, del, minuses);
2297 static int read_old_data(struct stat *st, struct patch *patch,
2298 const char *path, struct strbuf *buf)
2300 int conv_flags = patch->crlf_in_old ?
2301 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2302 switch (st->st_mode & S_IFMT) {
2303 case S_IFLNK:
2304 if (strbuf_readlink(buf, path, st->st_size) < 0)
2305 return error(_("unable to read symlink %s"), path);
2306 return 0;
2307 case S_IFREG:
2308 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2309 return error(_("unable to open or read %s"), path);
2311 * "git apply" without "--index/--cached" should never look
2312 * at the index; the target file may not have been added to
2313 * the index yet, and we may not even be in any Git repository.
2314 * Pass NULL to convert_to_git() to stress this; the function
2315 * should never look at the index when explicit crlf option
2316 * is given.
2318 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2319 return 0;
2320 default:
2321 return -1;
2326 * Update the preimage, and the common lines in postimage,
2327 * from buffer buf of length len. If postlen is 0 the postimage
2328 * is updated in place, otherwise it's updated on a new buffer
2329 * of length postlen
2332 static void update_pre_post_images(struct image *preimage,
2333 struct image *postimage,
2334 char *buf,
2335 size_t len, size_t postlen)
2337 int i, ctx, reduced;
2338 char *new_buf, *old_buf, *fixed;
2339 struct image fixed_preimage;
2342 * Update the preimage with whitespace fixes. Note that we
2343 * are not losing preimage->buf -- apply_one_fragment() will
2344 * free "oldlines".
2346 prepare_image(&fixed_preimage, buf, len, 1);
2347 assert(postlen
2348 ? fixed_preimage.nr == preimage->nr
2349 : fixed_preimage.nr <= preimage->nr);
2350 for (i = 0; i < fixed_preimage.nr; i++)
2351 fixed_preimage.line[i].flag = preimage->line[i].flag;
2352 free(preimage->line_allocated);
2353 *preimage = fixed_preimage;
2356 * Adjust the common context lines in postimage. This can be
2357 * done in-place when we are shrinking it with whitespace
2358 * fixing, but needs a new buffer when ignoring whitespace or
2359 * expanding leading tabs to spaces.
2361 * We trust the caller to tell us if the update can be done
2362 * in place (postlen==0) or not.
2364 old_buf = postimage->buf;
2365 if (postlen)
2366 new_buf = postimage->buf = xmalloc(postlen);
2367 else
2368 new_buf = old_buf;
2369 fixed = preimage->buf;
2371 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2372 size_t l_len = postimage->line[i].len;
2373 if (!(postimage->line[i].flag & LINE_COMMON)) {
2374 /* an added line -- no counterparts in preimage */
2375 memmove(new_buf, old_buf, l_len);
2376 old_buf += l_len;
2377 new_buf += l_len;
2378 continue;
2381 /* a common context -- skip it in the original postimage */
2382 old_buf += l_len;
2384 /* and find the corresponding one in the fixed preimage */
2385 while (ctx < preimage->nr &&
2386 !(preimage->line[ctx].flag & LINE_COMMON)) {
2387 fixed += preimage->line[ctx].len;
2388 ctx++;
2392 * preimage is expected to run out, if the caller
2393 * fixed addition of trailing blank lines.
2395 if (preimage->nr <= ctx) {
2396 reduced++;
2397 continue;
2400 /* and copy it in, while fixing the line length */
2401 l_len = preimage->line[ctx].len;
2402 memcpy(new_buf, fixed, l_len);
2403 new_buf += l_len;
2404 fixed += l_len;
2405 postimage->line[i].len = l_len;
2406 ctx++;
2409 if (postlen
2410 ? postlen < new_buf - postimage->buf
2411 : postimage->len < new_buf - postimage->buf)
2412 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2413 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2415 /* Fix the length of the whole thing */
2416 postimage->len = new_buf - postimage->buf;
2417 postimage->nr -= reduced;
2420 static int line_by_line_fuzzy_match(struct image *img,
2421 struct image *preimage,
2422 struct image *postimage,
2423 unsigned long current,
2424 int current_lno,
2425 int preimage_limit)
2427 int i;
2428 size_t imgoff = 0;
2429 size_t preoff = 0;
2430 size_t postlen = postimage->len;
2431 size_t extra_chars;
2432 char *buf;
2433 char *preimage_eof;
2434 char *preimage_end;
2435 struct strbuf fixed;
2436 char *fixed_buf;
2437 size_t fixed_len;
2439 for (i = 0; i < preimage_limit; i++) {
2440 size_t prelen = preimage->line[i].len;
2441 size_t imglen = img->line[current_lno+i].len;
2443 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2444 preimage->buf + preoff, prelen))
2445 return 0;
2446 if (preimage->line[i].flag & LINE_COMMON)
2447 postlen += imglen - prelen;
2448 imgoff += imglen;
2449 preoff += prelen;
2453 * Ok, the preimage matches with whitespace fuzz.
2455 * imgoff now holds the true length of the target that
2456 * matches the preimage before the end of the file.
2458 * Count the number of characters in the preimage that fall
2459 * beyond the end of the file and make sure that all of them
2460 * are whitespace characters. (This can only happen if
2461 * we are removing blank lines at the end of the file.)
2463 buf = preimage_eof = preimage->buf + preoff;
2464 for ( ; i < preimage->nr; i++)
2465 preoff += preimage->line[i].len;
2466 preimage_end = preimage->buf + preoff;
2467 for ( ; buf < preimage_end; buf++)
2468 if (!isspace(*buf))
2469 return 0;
2472 * Update the preimage and the common postimage context
2473 * lines to use the same whitespace as the target.
2474 * If whitespace is missing in the target (i.e.
2475 * if the preimage extends beyond the end of the file),
2476 * use the whitespace from the preimage.
2478 extra_chars = preimage_end - preimage_eof;
2479 strbuf_init(&fixed, imgoff + extra_chars);
2480 strbuf_add(&fixed, img->buf + current, imgoff);
2481 strbuf_add(&fixed, preimage_eof, extra_chars);
2482 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2483 update_pre_post_images(preimage, postimage,
2484 fixed_buf, fixed_len, postlen);
2485 return 1;
2488 static int match_fragment(struct apply_state *state,
2489 struct image *img,
2490 struct image *preimage,
2491 struct image *postimage,
2492 unsigned long current,
2493 int current_lno,
2494 unsigned ws_rule,
2495 int match_beginning, int match_end)
2497 int i;
2498 char *fixed_buf, *buf, *orig, *target;
2499 struct strbuf fixed;
2500 size_t fixed_len, postlen;
2501 int preimage_limit;
2503 if (preimage->nr + current_lno <= img->nr) {
2505 * The hunk falls within the boundaries of img.
2507 preimage_limit = preimage->nr;
2508 if (match_end && (preimage->nr + current_lno != img->nr))
2509 return 0;
2510 } else if (state->ws_error_action == correct_ws_error &&
2511 (ws_rule & WS_BLANK_AT_EOF)) {
2513 * This hunk extends beyond the end of img, and we are
2514 * removing blank lines at the end of the file. This
2515 * many lines from the beginning of the preimage must
2516 * match with img, and the remainder of the preimage
2517 * must be blank.
2519 preimage_limit = img->nr - current_lno;
2520 } else {
2522 * The hunk extends beyond the end of the img and
2523 * we are not removing blanks at the end, so we
2524 * should reject the hunk at this position.
2526 return 0;
2529 if (match_beginning && current_lno)
2530 return 0;
2532 /* Quick hash check */
2533 for (i = 0; i < preimage_limit; i++)
2534 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2535 (preimage->line[i].hash != img->line[current_lno + i].hash))
2536 return 0;
2538 if (preimage_limit == preimage->nr) {
2540 * Do we have an exact match? If we were told to match
2541 * at the end, size must be exactly at current+fragsize,
2542 * otherwise current+fragsize must be still within the preimage,
2543 * and either case, the old piece should match the preimage
2544 * exactly.
2546 if ((match_end
2547 ? (current + preimage->len == img->len)
2548 : (current + preimage->len <= img->len)) &&
2549 !memcmp(img->buf + current, preimage->buf, preimage->len))
2550 return 1;
2551 } else {
2553 * The preimage extends beyond the end of img, so
2554 * there cannot be an exact match.
2556 * There must be one non-blank context line that match
2557 * a line before the end of img.
2559 char *buf_end;
2561 buf = preimage->buf;
2562 buf_end = buf;
2563 for (i = 0; i < preimage_limit; i++)
2564 buf_end += preimage->line[i].len;
2566 for ( ; buf < buf_end; buf++)
2567 if (!isspace(*buf))
2568 break;
2569 if (buf == buf_end)
2570 return 0;
2574 * No exact match. If we are ignoring whitespace, run a line-by-line
2575 * fuzzy matching. We collect all the line length information because
2576 * we need it to adjust whitespace if we match.
2578 if (state->ws_ignore_action == ignore_ws_change)
2579 return line_by_line_fuzzy_match(img, preimage, postimage,
2580 current, current_lno, preimage_limit);
2582 if (state->ws_error_action != correct_ws_error)
2583 return 0;
2586 * The hunk does not apply byte-by-byte, but the hash says
2587 * it might with whitespace fuzz. We weren't asked to
2588 * ignore whitespace, we were asked to correct whitespace
2589 * errors, so let's try matching after whitespace correction.
2591 * While checking the preimage against the target, whitespace
2592 * errors in both fixed, we count how large the corresponding
2593 * postimage needs to be. The postimage prepared by
2594 * apply_one_fragment() has whitespace errors fixed on added
2595 * lines already, but the common lines were propagated as-is,
2596 * which may become longer when their whitespace errors are
2597 * fixed.
2600 /* First count added lines in postimage */
2601 postlen = 0;
2602 for (i = 0; i < postimage->nr; i++) {
2603 if (!(postimage->line[i].flag & LINE_COMMON))
2604 postlen += postimage->line[i].len;
2608 * The preimage may extend beyond the end of the file,
2609 * but in this loop we will only handle the part of the
2610 * preimage that falls within the file.
2612 strbuf_init(&fixed, preimage->len + 1);
2613 orig = preimage->buf;
2614 target = img->buf + current;
2615 for (i = 0; i < preimage_limit; i++) {
2616 size_t oldlen = preimage->line[i].len;
2617 size_t tgtlen = img->line[current_lno + i].len;
2618 size_t fixstart = fixed.len;
2619 struct strbuf tgtfix;
2620 int match;
2622 /* Try fixing the line in the preimage */
2623 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2625 /* Try fixing the line in the target */
2626 strbuf_init(&tgtfix, tgtlen);
2627 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2630 * If they match, either the preimage was based on
2631 * a version before our tree fixed whitespace breakage,
2632 * or we are lacking a whitespace-fix patch the tree
2633 * the preimage was based on already had (i.e. target
2634 * has whitespace breakage, the preimage doesn't).
2635 * In either case, we are fixing the whitespace breakages
2636 * so we might as well take the fix together with their
2637 * real change.
2639 match = (tgtfix.len == fixed.len - fixstart &&
2640 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2641 fixed.len - fixstart));
2643 /* Add the length if this is common with the postimage */
2644 if (preimage->line[i].flag & LINE_COMMON)
2645 postlen += tgtfix.len;
2647 strbuf_release(&tgtfix);
2648 if (!match)
2649 goto unmatch_exit;
2651 orig += oldlen;
2652 target += tgtlen;
2657 * Now handle the lines in the preimage that falls beyond the
2658 * end of the file (if any). They will only match if they are
2659 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2660 * false).
2662 for ( ; i < preimage->nr; i++) {
2663 size_t fixstart = fixed.len; /* start of the fixed preimage */
2664 size_t oldlen = preimage->line[i].len;
2665 int j;
2667 /* Try fixing the line in the preimage */
2668 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2670 for (j = fixstart; j < fixed.len; j++)
2671 if (!isspace(fixed.buf[j]))
2672 goto unmatch_exit;
2674 orig += oldlen;
2678 * Yes, the preimage is based on an older version that still
2679 * has whitespace breakages unfixed, and fixing them makes the
2680 * hunk match. Update the context lines in the postimage.
2682 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2683 if (postlen < postimage->len)
2684 postlen = 0;
2685 update_pre_post_images(preimage, postimage,
2686 fixed_buf, fixed_len, postlen);
2687 return 1;
2689 unmatch_exit:
2690 strbuf_release(&fixed);
2691 return 0;
2694 static int find_pos(struct apply_state *state,
2695 struct image *img,
2696 struct image *preimage,
2697 struct image *postimage,
2698 int line,
2699 unsigned ws_rule,
2700 int match_beginning, int match_end)
2702 int i;
2703 unsigned long backwards, forwards, current;
2704 int backwards_lno, forwards_lno, current_lno;
2707 * When running with --allow-overlap, it is possible that a hunk is
2708 * seen that pretends to start at the beginning (but no longer does),
2709 * and that *still* needs to match the end. So trust `match_end` more
2710 * than `match_beginning`.
2712 if (state->allow_overlap && match_beginning && match_end &&
2713 img->nr - preimage->nr != 0)
2714 match_beginning = 0;
2717 * If match_beginning or match_end is specified, there is no
2718 * point starting from a wrong line that will never match and
2719 * wander around and wait for a match at the specified end.
2721 if (match_beginning)
2722 line = 0;
2723 else if (match_end)
2724 line = img->nr - preimage->nr;
2727 * Because the comparison is unsigned, the following test
2728 * will also take care of a negative line number that can
2729 * result when match_end and preimage is larger than the target.
2731 if ((size_t) line > img->nr)
2732 line = img->nr;
2734 current = 0;
2735 for (i = 0; i < line; i++)
2736 current += img->line[i].len;
2739 * There's probably some smart way to do this, but I'll leave
2740 * that to the smart and beautiful people. I'm simple and stupid.
2742 backwards = current;
2743 backwards_lno = line;
2744 forwards = current;
2745 forwards_lno = line;
2746 current_lno = line;
2748 for (i = 0; ; i++) {
2749 if (match_fragment(state, img, preimage, postimage,
2750 current, current_lno, ws_rule,
2751 match_beginning, match_end))
2752 return current_lno;
2754 again:
2755 if (backwards_lno == 0 && forwards_lno == img->nr)
2756 break;
2758 if (i & 1) {
2759 if (backwards_lno == 0) {
2760 i++;
2761 goto again;
2763 backwards_lno--;
2764 backwards -= img->line[backwards_lno].len;
2765 current = backwards;
2766 current_lno = backwards_lno;
2767 } else {
2768 if (forwards_lno == img->nr) {
2769 i++;
2770 goto again;
2772 forwards += img->line[forwards_lno].len;
2773 forwards_lno++;
2774 current = forwards;
2775 current_lno = forwards_lno;
2779 return -1;
2782 static void remove_first_line(struct image *img)
2784 img->buf += img->line[0].len;
2785 img->len -= img->line[0].len;
2786 img->line++;
2787 img->nr--;
2790 static void remove_last_line(struct image *img)
2792 img->len -= img->line[--img->nr].len;
2796 * The change from "preimage" and "postimage" has been found to
2797 * apply at applied_pos (counts in line numbers) in "img".
2798 * Update "img" to remove "preimage" and replace it with "postimage".
2800 static void update_image(struct apply_state *state,
2801 struct image *img,
2802 int applied_pos,
2803 struct image *preimage,
2804 struct image *postimage)
2807 * remove the copy of preimage at offset in img
2808 * and replace it with postimage
2810 int i, nr;
2811 size_t remove_count, insert_count, applied_at = 0;
2812 char *result;
2813 int preimage_limit;
2816 * If we are removing blank lines at the end of img,
2817 * the preimage may extend beyond the end.
2818 * If that is the case, we must be careful only to
2819 * remove the part of the preimage that falls within
2820 * the boundaries of img. Initialize preimage_limit
2821 * to the number of lines in the preimage that falls
2822 * within the boundaries.
2824 preimage_limit = preimage->nr;
2825 if (preimage_limit > img->nr - applied_pos)
2826 preimage_limit = img->nr - applied_pos;
2828 for (i = 0; i < applied_pos; i++)
2829 applied_at += img->line[i].len;
2831 remove_count = 0;
2832 for (i = 0; i < preimage_limit; i++)
2833 remove_count += img->line[applied_pos + i].len;
2834 insert_count = postimage->len;
2836 /* Adjust the contents */
2837 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2838 memcpy(result, img->buf, applied_at);
2839 memcpy(result + applied_at, postimage->buf, postimage->len);
2840 memcpy(result + applied_at + postimage->len,
2841 img->buf + (applied_at + remove_count),
2842 img->len - (applied_at + remove_count));
2843 free(img->buf);
2844 img->buf = result;
2845 img->len += insert_count - remove_count;
2846 result[img->len] = '\0';
2848 /* Adjust the line table */
2849 nr = img->nr + postimage->nr - preimage_limit;
2850 if (preimage_limit < postimage->nr) {
2852 * NOTE: this knows that we never call remove_first_line()
2853 * on anything other than pre/post image.
2855 REALLOC_ARRAY(img->line, nr);
2856 img->line_allocated = img->line;
2858 if (preimage_limit != postimage->nr)
2859 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2860 img->line + applied_pos + preimage_limit,
2861 img->nr - (applied_pos + preimage_limit));
2862 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2863 if (!state->allow_overlap)
2864 for (i = 0; i < postimage->nr; i++)
2865 img->line[applied_pos + i].flag |= LINE_PATCHED;
2866 img->nr = nr;
2870 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2871 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2872 * replace the part of "img" with "postimage" text.
2874 static int apply_one_fragment(struct apply_state *state,
2875 struct image *img, struct fragment *frag,
2876 int inaccurate_eof, unsigned ws_rule,
2877 int nth_fragment)
2879 int match_beginning, match_end;
2880 const char *patch = frag->patch;
2881 int size = frag->size;
2882 char *old, *oldlines;
2883 struct strbuf newlines;
2884 int new_blank_lines_at_end = 0;
2885 int found_new_blank_lines_at_end = 0;
2886 int hunk_linenr = frag->linenr;
2887 unsigned long leading, trailing;
2888 int pos, applied_pos;
2889 struct image preimage;
2890 struct image postimage;
2892 memset(&preimage, 0, sizeof(preimage));
2893 memset(&postimage, 0, sizeof(postimage));
2894 oldlines = xmalloc(size);
2895 strbuf_init(&newlines, size);
2897 old = oldlines;
2898 while (size > 0) {
2899 char first;
2900 int len = linelen(patch, size);
2901 int plen;
2902 int added_blank_line = 0;
2903 int is_blank_context = 0;
2904 size_t start;
2906 if (!len)
2907 break;
2910 * "plen" is how much of the line we should use for
2911 * the actual patch data. Normally we just remove the
2912 * first character on the line, but if the line is
2913 * followed by "\ No newline", then we also remove the
2914 * last one (which is the newline, of course).
2916 plen = len - 1;
2917 if (len < size && patch[len] == '\\')
2918 plen--;
2919 first = *patch;
2920 if (state->apply_in_reverse) {
2921 if (first == '-')
2922 first = '+';
2923 else if (first == '+')
2924 first = '-';
2927 switch (first) {
2928 case '\n':
2929 /* Newer GNU diff, empty context line */
2930 if (plen < 0)
2931 /* ... followed by '\No newline'; nothing */
2932 break;
2933 *old++ = '\n';
2934 strbuf_addch(&newlines, '\n');
2935 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2936 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2937 is_blank_context = 1;
2938 break;
2939 case ' ':
2940 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2941 ws_blank_line(patch + 1, plen))
2942 is_blank_context = 1;
2943 /* fallthrough */
2944 case '-':
2945 memcpy(old, patch + 1, plen);
2946 add_line_info(&preimage, old, plen,
2947 (first == ' ' ? LINE_COMMON : 0));
2948 old += plen;
2949 if (first == '-')
2950 break;
2951 /* fallthrough */
2952 case '+':
2953 /* --no-add does not add new lines */
2954 if (first == '+' && state->no_add)
2955 break;
2957 start = newlines.len;
2958 if (first != '+' ||
2959 !state->whitespace_error ||
2960 state->ws_error_action != correct_ws_error) {
2961 strbuf_add(&newlines, patch + 1, plen);
2963 else {
2964 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2966 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2967 (first == '+' ? 0 : LINE_COMMON));
2968 if (first == '+' &&
2969 (ws_rule & WS_BLANK_AT_EOF) &&
2970 ws_blank_line(patch + 1, plen))
2971 added_blank_line = 1;
2972 break;
2973 case '@': case '\\':
2974 /* Ignore it, we already handled it */
2975 break;
2976 default:
2977 if (state->apply_verbosity > verbosity_normal)
2978 error(_("invalid start of line: '%c'"), first);
2979 applied_pos = -1;
2980 goto out;
2982 if (added_blank_line) {
2983 if (!new_blank_lines_at_end)
2984 found_new_blank_lines_at_end = hunk_linenr;
2985 new_blank_lines_at_end++;
2987 else if (is_blank_context)
2989 else
2990 new_blank_lines_at_end = 0;
2991 patch += len;
2992 size -= len;
2993 hunk_linenr++;
2995 if (inaccurate_eof &&
2996 old > oldlines && old[-1] == '\n' &&
2997 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2998 old--;
2999 strbuf_setlen(&newlines, newlines.len - 1);
3000 preimage.line_allocated[preimage.nr - 1].len--;
3001 postimage.line_allocated[postimage.nr - 1].len--;
3004 leading = frag->leading;
3005 trailing = frag->trailing;
3008 * A hunk to change lines at the beginning would begin with
3009 * @@ -1,L +N,M @@
3010 * but we need to be careful. -U0 that inserts before the second
3011 * line also has this pattern.
3013 * And a hunk to add to an empty file would begin with
3014 * @@ -0,0 +N,M @@
3016 * In other words, a hunk that is (frag->oldpos <= 1) with or
3017 * without leading context must match at the beginning.
3019 match_beginning = (!frag->oldpos ||
3020 (frag->oldpos == 1 && !state->unidiff_zero));
3023 * A hunk without trailing lines must match at the end.
3024 * However, we simply cannot tell if a hunk must match end
3025 * from the lack of trailing lines if the patch was generated
3026 * with unidiff without any context.
3028 match_end = !state->unidiff_zero && !trailing;
3030 pos = frag->newpos ? (frag->newpos - 1) : 0;
3031 preimage.buf = oldlines;
3032 preimage.len = old - oldlines;
3033 postimage.buf = newlines.buf;
3034 postimage.len = newlines.len;
3035 preimage.line = preimage.line_allocated;
3036 postimage.line = postimage.line_allocated;
3038 for (;;) {
3040 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3041 ws_rule, match_beginning, match_end);
3043 if (applied_pos >= 0)
3044 break;
3046 /* Am I at my context limits? */
3047 if ((leading <= state->p_context) && (trailing <= state->p_context))
3048 break;
3049 if (match_beginning || match_end) {
3050 match_beginning = match_end = 0;
3051 continue;
3055 * Reduce the number of context lines; reduce both
3056 * leading and trailing if they are equal otherwise
3057 * just reduce the larger context.
3059 if (leading >= trailing) {
3060 remove_first_line(&preimage);
3061 remove_first_line(&postimage);
3062 pos--;
3063 leading--;
3065 if (trailing > leading) {
3066 remove_last_line(&preimage);
3067 remove_last_line(&postimage);
3068 trailing--;
3072 if (applied_pos >= 0) {
3073 if (new_blank_lines_at_end &&
3074 preimage.nr + applied_pos >= img->nr &&
3075 (ws_rule & WS_BLANK_AT_EOF) &&
3076 state->ws_error_action != nowarn_ws_error) {
3077 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3078 found_new_blank_lines_at_end);
3079 if (state->ws_error_action == correct_ws_error) {
3080 while (new_blank_lines_at_end--)
3081 remove_last_line(&postimage);
3084 * We would want to prevent write_out_results()
3085 * from taking place in apply_patch() that follows
3086 * the callchain led us here, which is:
3087 * apply_patch->check_patch_list->check_patch->
3088 * apply_data->apply_fragments->apply_one_fragment
3090 if (state->ws_error_action == die_on_ws_error)
3091 state->apply = 0;
3094 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3095 int offset = applied_pos - pos;
3096 if (state->apply_in_reverse)
3097 offset = 0 - offset;
3098 fprintf_ln(stderr,
3099 Q_("Hunk #%d succeeded at %d (offset %d line).",
3100 "Hunk #%d succeeded at %d (offset %d lines).",
3101 offset),
3102 nth_fragment, applied_pos + 1, offset);
3106 * Warn if it was necessary to reduce the number
3107 * of context lines.
3109 if ((leading != frag->leading ||
3110 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3111 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3112 " to apply fragment at %d"),
3113 leading, trailing, applied_pos+1);
3114 update_image(state, img, applied_pos, &preimage, &postimage);
3115 } else {
3116 if (state->apply_verbosity > verbosity_normal)
3117 error(_("while searching for:\n%.*s"),
3118 (int)(old - oldlines), oldlines);
3121 out:
3122 free(oldlines);
3123 strbuf_release(&newlines);
3124 free(preimage.line_allocated);
3125 free(postimage.line_allocated);
3127 return (applied_pos < 0);
3130 static int apply_binary_fragment(struct apply_state *state,
3131 struct image *img,
3132 struct patch *patch)
3134 struct fragment *fragment = patch->fragments;
3135 unsigned long len;
3136 void *dst;
3138 if (!fragment)
3139 return error(_("missing binary patch data for '%s'"),
3140 patch->new_name ?
3141 patch->new_name :
3142 patch->old_name);
3144 /* Binary patch is irreversible without the optional second hunk */
3145 if (state->apply_in_reverse) {
3146 if (!fragment->next)
3147 return error(_("cannot reverse-apply a binary patch "
3148 "without the reverse hunk to '%s'"),
3149 patch->new_name
3150 ? patch->new_name : patch->old_name);
3151 fragment = fragment->next;
3153 switch (fragment->binary_patch_method) {
3154 case BINARY_DELTA_DEFLATED:
3155 dst = patch_delta(img->buf, img->len, fragment->patch,
3156 fragment->size, &len);
3157 if (!dst)
3158 return -1;
3159 clear_image(img);
3160 img->buf = dst;
3161 img->len = len;
3162 return 0;
3163 case BINARY_LITERAL_DEFLATED:
3164 clear_image(img);
3165 img->len = fragment->size;
3166 img->buf = xmemdupz(fragment->patch, img->len);
3167 return 0;
3169 return -1;
3173 * Replace "img" with the result of applying the binary patch.
3174 * The binary patch data itself in patch->fragment is still kept
3175 * but the preimage prepared by the caller in "img" is freed here
3176 * or in the helper function apply_binary_fragment() this calls.
3178 static int apply_binary(struct apply_state *state,
3179 struct image *img,
3180 struct patch *patch)
3182 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3183 struct object_id oid;
3184 const unsigned hexsz = the_hash_algo->hexsz;
3187 * For safety, we require patch index line to contain
3188 * full hex textual object ID for old and new, at least for now.
3190 if (strlen(patch->old_oid_prefix) != hexsz ||
3191 strlen(patch->new_oid_prefix) != hexsz ||
3192 get_oid_hex(patch->old_oid_prefix, &oid) ||
3193 get_oid_hex(patch->new_oid_prefix, &oid))
3194 return error(_("cannot apply binary patch to '%s' "
3195 "without full index line"), name);
3197 if (patch->old_name) {
3199 * See if the old one matches what the patch
3200 * applies to.
3202 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3203 &oid);
3204 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3205 return error(_("the patch applies to '%s' (%s), "
3206 "which does not match the "
3207 "current contents."),
3208 name, oid_to_hex(&oid));
3210 else {
3211 /* Otherwise, the old one must be empty. */
3212 if (img->len)
3213 return error(_("the patch applies to an empty "
3214 "'%s' but it is not empty"), name);
3217 get_oid_hex(patch->new_oid_prefix, &oid);
3218 if (is_null_oid(&oid)) {
3219 clear_image(img);
3220 return 0; /* deletion patch */
3223 if (has_object(the_repository, &oid, 0)) {
3224 /* We already have the postimage */
3225 enum object_type type;
3226 unsigned long size;
3227 char *result;
3229 result = repo_read_object_file(the_repository, &oid, &type,
3230 &size);
3231 if (!result)
3232 return error(_("the necessary postimage %s for "
3233 "'%s' cannot be read"),
3234 patch->new_oid_prefix, name);
3235 clear_image(img);
3236 img->buf = result;
3237 img->len = size;
3238 } else {
3240 * We have verified buf matches the preimage;
3241 * apply the patch data to it, which is stored
3242 * in the patch->fragments->{patch,size}.
3244 if (apply_binary_fragment(state, img, patch))
3245 return error(_("binary patch does not apply to '%s'"),
3246 name);
3248 /* verify that the result matches */
3249 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3250 &oid);
3251 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3252 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3253 name, patch->new_oid_prefix, oid_to_hex(&oid));
3256 return 0;
3259 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3261 struct fragment *frag = patch->fragments;
3262 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3263 unsigned ws_rule = patch->ws_rule;
3264 unsigned inaccurate_eof = patch->inaccurate_eof;
3265 int nth = 0;
3267 if (patch->is_binary)
3268 return apply_binary(state, img, patch);
3270 while (frag) {
3271 nth++;
3272 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3273 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3274 if (!state->apply_with_reject)
3275 return -1;
3276 frag->rejected = 1;
3278 frag = frag->next;
3280 return 0;
3283 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3285 if (S_ISGITLINK(mode)) {
3286 strbuf_grow(buf, 100);
3287 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3288 } else {
3289 enum object_type type;
3290 unsigned long sz;
3291 char *result;
3293 result = repo_read_object_file(the_repository, oid, &type,
3294 &sz);
3295 if (!result)
3296 return -1;
3297 /* XXX read_sha1_file NUL-terminates */
3298 strbuf_attach(buf, result, sz, sz + 1);
3300 return 0;
3303 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3305 if (!ce)
3306 return 0;
3307 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3310 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3312 struct string_list_item *item;
3314 if (!name)
3315 return NULL;
3317 item = string_list_lookup(&state->fn_table, name);
3318 if (item)
3319 return (struct patch *)item->util;
3321 return NULL;
3325 * item->util in the filename table records the status of the path.
3326 * Usually it points at a patch (whose result records the contents
3327 * of it after applying it), but it could be PATH_WAS_DELETED for a
3328 * path that a previously applied patch has already removed, or
3329 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3331 * The latter is needed to deal with a case where two paths A and B
3332 * are swapped by first renaming A to B and then renaming B to A;
3333 * moving A to B should not be prevented due to presence of B as we
3334 * will remove it in a later patch.
3336 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3337 #define PATH_WAS_DELETED ((struct patch *) -1)
3339 static int to_be_deleted(struct patch *patch)
3341 return patch == PATH_TO_BE_DELETED;
3344 static int was_deleted(struct patch *patch)
3346 return patch == PATH_WAS_DELETED;
3349 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3351 struct string_list_item *item;
3354 * Always add new_name unless patch is a deletion
3355 * This should cover the cases for normal diffs,
3356 * file creations and copies
3358 if (patch->new_name) {
3359 item = string_list_insert(&state->fn_table, patch->new_name);
3360 item->util = patch;
3364 * store a failure on rename/deletion cases because
3365 * later chunks shouldn't patch old names
3367 if ((patch->new_name == NULL) || (patch->is_rename)) {
3368 item = string_list_insert(&state->fn_table, patch->old_name);
3369 item->util = PATH_WAS_DELETED;
3373 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3376 * store information about incoming file deletion
3378 while (patch) {
3379 if ((patch->new_name == NULL) || (patch->is_rename)) {
3380 struct string_list_item *item;
3381 item = string_list_insert(&state->fn_table, patch->old_name);
3382 item->util = PATH_TO_BE_DELETED;
3384 patch = patch->next;
3388 static int checkout_target(struct index_state *istate,
3389 struct cache_entry *ce, struct stat *st)
3391 struct checkout costate = CHECKOUT_INIT;
3393 costate.refresh_cache = 1;
3394 costate.istate = istate;
3395 if (checkout_entry(ce, &costate, NULL, NULL) ||
3396 lstat(ce->name, st))
3397 return error(_("cannot checkout %s"), ce->name);
3398 return 0;
3401 static struct patch *previous_patch(struct apply_state *state,
3402 struct patch *patch,
3403 int *gone)
3405 struct patch *previous;
3407 *gone = 0;
3408 if (patch->is_copy || patch->is_rename)
3409 return NULL; /* "git" patches do not depend on the order */
3411 previous = in_fn_table(state, patch->old_name);
3412 if (!previous)
3413 return NULL;
3415 if (to_be_deleted(previous))
3416 return NULL; /* the deletion hasn't happened yet */
3418 if (was_deleted(previous))
3419 *gone = 1;
3421 return previous;
3424 static int verify_index_match(struct apply_state *state,
3425 const struct cache_entry *ce,
3426 struct stat *st)
3428 if (S_ISGITLINK(ce->ce_mode)) {
3429 if (!S_ISDIR(st->st_mode))
3430 return -1;
3431 return 0;
3433 return ie_match_stat(state->repo->index, ce, st,
3434 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3437 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3439 static int load_patch_target(struct apply_state *state,
3440 struct strbuf *buf,
3441 const struct cache_entry *ce,
3442 struct stat *st,
3443 struct patch *patch,
3444 const char *name,
3445 unsigned expected_mode)
3447 if (state->cached || state->check_index) {
3448 if (read_file_or_gitlink(ce, buf))
3449 return error(_("failed to read %s"), name);
3450 } else if (name) {
3451 if (S_ISGITLINK(expected_mode)) {
3452 if (ce)
3453 return read_file_or_gitlink(ce, buf);
3454 else
3455 return SUBMODULE_PATCH_WITHOUT_INDEX;
3456 } else if (has_symlink_leading_path(name, strlen(name))) {
3457 return error(_("reading from '%s' beyond a symbolic link"), name);
3458 } else {
3459 if (read_old_data(st, patch, name, buf))
3460 return error(_("failed to read %s"), name);
3463 return 0;
3467 * We are about to apply "patch"; populate the "image" with the
3468 * current version we have, from the working tree or from the index,
3469 * depending on the situation e.g. --cached/--index. If we are
3470 * applying a non-git patch that incrementally updates the tree,
3471 * we read from the result of a previous diff.
3473 static int load_preimage(struct apply_state *state,
3474 struct image *image,
3475 struct patch *patch, struct stat *st,
3476 const struct cache_entry *ce)
3478 struct strbuf buf = STRBUF_INIT;
3479 size_t len;
3480 char *img;
3481 struct patch *previous;
3482 int status;
3484 previous = previous_patch(state, patch, &status);
3485 if (status)
3486 return error(_("path %s has been renamed/deleted"),
3487 patch->old_name);
3488 if (previous) {
3489 /* We have a patched copy in memory; use that. */
3490 strbuf_add(&buf, previous->result, previous->resultsize);
3491 } else {
3492 status = load_patch_target(state, &buf, ce, st, patch,
3493 patch->old_name, patch->old_mode);
3494 if (status < 0)
3495 return status;
3496 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3498 * There is no way to apply subproject
3499 * patch without looking at the index.
3500 * NEEDSWORK: shouldn't this be flagged
3501 * as an error???
3503 free_fragment_list(patch->fragments);
3504 patch->fragments = NULL;
3505 } else if (status) {
3506 return error(_("failed to read %s"), patch->old_name);
3510 img = strbuf_detach(&buf, &len);
3511 prepare_image(image, img, len, !patch->is_binary);
3512 return 0;
3515 static int resolve_to(struct image *image, const struct object_id *result_id)
3517 unsigned long size;
3518 enum object_type type;
3520 clear_image(image);
3522 image->buf = repo_read_object_file(the_repository, result_id, &type,
3523 &size);
3524 if (!image->buf || type != OBJ_BLOB)
3525 die("unable to read blob object %s", oid_to_hex(result_id));
3526 image->len = size;
3528 return 0;
3531 static int three_way_merge(struct apply_state *state,
3532 struct image *image,
3533 char *path,
3534 const struct object_id *base,
3535 const struct object_id *ours,
3536 const struct object_id *theirs)
3538 mmfile_t base_file, our_file, their_file;
3539 mmbuffer_t result = { NULL };
3540 enum ll_merge_result status;
3542 /* resolve trivial cases first */
3543 if (oideq(base, ours))
3544 return resolve_to(image, theirs);
3545 else if (oideq(base, theirs) || oideq(ours, theirs))
3546 return resolve_to(image, ours);
3548 read_mmblob(&base_file, base);
3549 read_mmblob(&our_file, ours);
3550 read_mmblob(&their_file, theirs);
3551 status = ll_merge(&result, path,
3552 &base_file, "base",
3553 &our_file, "ours",
3554 &their_file, "theirs",
3555 state->repo->index,
3556 NULL);
3557 if (status == LL_MERGE_BINARY_CONFLICT)
3558 warning("Cannot merge binary files: %s (%s vs. %s)",
3559 path, "ours", "theirs");
3560 free(base_file.ptr);
3561 free(our_file.ptr);
3562 free(their_file.ptr);
3563 if (status < 0 || !result.ptr) {
3564 free(result.ptr);
3565 return -1;
3567 clear_image(image);
3568 image->buf = result.ptr;
3569 image->len = result.size;
3571 return status;
3575 * When directly falling back to add/add three-way merge, we read from
3576 * the current contents of the new_name. In no cases other than that
3577 * this function will be called.
3579 static int load_current(struct apply_state *state,
3580 struct image *image,
3581 struct patch *patch)
3583 struct strbuf buf = STRBUF_INIT;
3584 int status, pos;
3585 size_t len;
3586 char *img;
3587 struct stat st;
3588 struct cache_entry *ce;
3589 char *name = patch->new_name;
3590 unsigned mode = patch->new_mode;
3592 if (!patch->is_new)
3593 BUG("patch to %s is not a creation", patch->old_name);
3595 pos = index_name_pos(state->repo->index, name, strlen(name));
3596 if (pos < 0)
3597 return error(_("%s: does not exist in index"), name);
3598 ce = state->repo->index->cache[pos];
3599 if (lstat(name, &st)) {
3600 if (errno != ENOENT)
3601 return error_errno("%s", name);
3602 if (checkout_target(state->repo->index, ce, &st))
3603 return -1;
3605 if (verify_index_match(state, ce, &st))
3606 return error(_("%s: does not match index"), name);
3608 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3609 if (status < 0)
3610 return status;
3611 else if (status)
3612 return -1;
3613 img = strbuf_detach(&buf, &len);
3614 prepare_image(image, img, len, !patch->is_binary);
3615 return 0;
3618 static int try_threeway(struct apply_state *state,
3619 struct image *image,
3620 struct patch *patch,
3621 struct stat *st,
3622 const struct cache_entry *ce)
3624 struct object_id pre_oid, post_oid, our_oid;
3625 struct strbuf buf = STRBUF_INIT;
3626 size_t len;
3627 int status;
3628 char *img;
3629 struct image tmp_image;
3631 /* No point falling back to 3-way merge in these cases */
3632 if (patch->is_delete ||
3633 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3634 (patch->is_new && !patch->direct_to_threeway) ||
3635 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3636 return -1;
3638 /* Preimage the patch was prepared for */
3639 if (patch->is_new)
3640 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3641 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3642 read_blob_object(&buf, &pre_oid, patch->old_mode))
3643 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3645 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3646 fprintf(stderr, _("Performing three-way merge...\n"));
3648 img = strbuf_detach(&buf, &len);
3649 prepare_image(&tmp_image, img, len, 1);
3650 /* Apply the patch to get the post image */
3651 if (apply_fragments(state, &tmp_image, patch) < 0) {
3652 clear_image(&tmp_image);
3653 return -1;
3655 /* post_oid is theirs */
3656 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3657 clear_image(&tmp_image);
3659 /* our_oid is ours */
3660 if (patch->is_new) {
3661 if (load_current(state, &tmp_image, patch))
3662 return error(_("cannot read the current contents of '%s'"),
3663 patch->new_name);
3664 } else {
3665 if (load_preimage(state, &tmp_image, patch, st, ce))
3666 return error(_("cannot read the current contents of '%s'"),
3667 patch->old_name);
3669 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3670 clear_image(&tmp_image);
3672 /* in-core three-way merge between post and our using pre as base */
3673 status = three_way_merge(state, image, patch->new_name,
3674 &pre_oid, &our_oid, &post_oid);
3675 if (status < 0) {
3676 if (state->apply_verbosity > verbosity_silent)
3677 fprintf(stderr,
3678 _("Failed to perform three-way merge...\n"));
3679 return status;
3682 if (status) {
3683 patch->conflicted_threeway = 1;
3684 if (patch->is_new)
3685 oidclr(&patch->threeway_stage[0], the_repository->hash_algo);
3686 else
3687 oidcpy(&patch->threeway_stage[0], &pre_oid);
3688 oidcpy(&patch->threeway_stage[1], &our_oid);
3689 oidcpy(&patch->threeway_stage[2], &post_oid);
3690 if (state->apply_verbosity > verbosity_silent)
3691 fprintf(stderr,
3692 _("Applied patch to '%s' with conflicts.\n"),
3693 patch->new_name);
3694 } else {
3695 if (state->apply_verbosity > verbosity_silent)
3696 fprintf(stderr,
3697 _("Applied patch to '%s' cleanly.\n"),
3698 patch->new_name);
3700 return 0;
3703 static int apply_data(struct apply_state *state, struct patch *patch,
3704 struct stat *st, const struct cache_entry *ce)
3706 struct image image;
3708 if (load_preimage(state, &image, patch, st, ce) < 0)
3709 return -1;
3711 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3712 if (state->apply_verbosity > verbosity_silent &&
3713 state->threeway && !patch->direct_to_threeway)
3714 fprintf(stderr, _("Falling back to direct application...\n"));
3716 /* Note: with --reject, apply_fragments() returns 0 */
3717 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0) {
3718 clear_image(&image);
3719 return -1;
3722 patch->result = image.buf;
3723 patch->resultsize = image.len;
3724 add_to_fn_table(state, patch);
3725 free(image.line_allocated);
3727 if (0 < patch->is_delete && patch->resultsize)
3728 return error(_("removal patch leaves file contents"));
3730 return 0;
3734 * If "patch" that we are looking at modifies or deletes what we have,
3735 * we would want it not to lose any local modification we have, either
3736 * in the working tree or in the index.
3738 * This also decides if a non-git patch is a creation patch or a
3739 * modification to an existing empty file. We do not check the state
3740 * of the current tree for a creation patch in this function; the caller
3741 * check_patch() separately makes sure (and errors out otherwise) that
3742 * the path the patch creates does not exist in the current tree.
3744 static int check_preimage(struct apply_state *state,
3745 struct patch *patch,
3746 struct cache_entry **ce,
3747 struct stat *st)
3749 const char *old_name = patch->old_name;
3750 struct patch *previous = NULL;
3751 int stat_ret = 0, status;
3752 unsigned st_mode = 0;
3754 if (!old_name)
3755 return 0;
3757 assert(patch->is_new <= 0);
3758 previous = previous_patch(state, patch, &status);
3760 if (status)
3761 return error(_("path %s has been renamed/deleted"), old_name);
3762 if (previous) {
3763 st_mode = previous->new_mode;
3764 } else if (!state->cached) {
3765 stat_ret = lstat(old_name, st);
3766 if (stat_ret && errno != ENOENT)
3767 return error_errno("%s", old_name);
3770 if (state->check_index && !previous) {
3771 int pos = index_name_pos(state->repo->index, old_name,
3772 strlen(old_name));
3773 if (pos < 0) {
3774 if (patch->is_new < 0)
3775 goto is_new;
3776 return error(_("%s: does not exist in index"), old_name);
3778 *ce = state->repo->index->cache[pos];
3779 if (stat_ret < 0) {
3780 if (checkout_target(state->repo->index, *ce, st))
3781 return -1;
3783 if (!state->cached && verify_index_match(state, *ce, st))
3784 return error(_("%s: does not match index"), old_name);
3785 if (state->cached)
3786 st_mode = (*ce)->ce_mode;
3787 } else if (stat_ret < 0) {
3788 if (patch->is_new < 0)
3789 goto is_new;
3790 return error_errno("%s", old_name);
3793 if (!state->cached && !previous) {
3794 if (*ce && !(*ce)->ce_mode)
3795 BUG("ce_mode == 0 for path '%s'", old_name);
3797 if (trust_executable_bit)
3798 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3799 else if (*ce)
3800 st_mode = (*ce)->ce_mode;
3801 else
3802 st_mode = patch->old_mode;
3805 if (patch->is_new < 0)
3806 patch->is_new = 0;
3807 if (!patch->old_mode)
3808 patch->old_mode = st_mode;
3809 if ((st_mode ^ patch->old_mode) & S_IFMT)
3810 return error(_("%s: wrong type"), old_name);
3811 if (st_mode != patch->old_mode)
3812 warning(_("%s has type %o, expected %o"),
3813 old_name, st_mode, patch->old_mode);
3814 if (!patch->new_mode && !patch->is_delete)
3815 patch->new_mode = st_mode;
3816 return 0;
3818 is_new:
3819 patch->is_new = 1;
3820 patch->is_delete = 0;
3821 FREE_AND_NULL(patch->old_name);
3822 return 0;
3826 #define EXISTS_IN_INDEX 1
3827 #define EXISTS_IN_WORKTREE 2
3828 #define EXISTS_IN_INDEX_AS_ITA 3
3830 static int check_to_create(struct apply_state *state,
3831 const char *new_name,
3832 int ok_if_exists)
3834 struct stat nst;
3836 if (state->check_index && (!ok_if_exists || !state->cached)) {
3837 int pos;
3839 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3840 if (pos >= 0) {
3841 struct cache_entry *ce = state->repo->index->cache[pos];
3843 /* allow ITA, as they do not yet exist in the index */
3844 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3845 return EXISTS_IN_INDEX;
3847 /* ITA entries can never match working tree files */
3848 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3849 return EXISTS_IN_INDEX_AS_ITA;
3853 if (state->cached)
3854 return 0;
3856 if (!lstat(new_name, &nst)) {
3857 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3858 return 0;
3860 * A leading component of new_name might be a symlink
3861 * that is going to be removed with this patch, but
3862 * still pointing at somewhere that has the path.
3863 * In such a case, path "new_name" does not exist as
3864 * far as git is concerned.
3866 if (has_symlink_leading_path(new_name, strlen(new_name)))
3867 return 0;
3869 return EXISTS_IN_WORKTREE;
3870 } else if (!is_missing_file_error(errno)) {
3871 return error_errno("%s", new_name);
3873 return 0;
3876 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3878 for ( ; patch; patch = patch->next) {
3879 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3880 (patch->is_rename || patch->is_delete))
3881 /* the symlink at patch->old_name is removed */
3882 strset_add(&state->removed_symlinks, patch->old_name);
3884 if (patch->new_name && S_ISLNK(patch->new_mode))
3885 /* the symlink at patch->new_name is created or remains */
3886 strset_add(&state->kept_symlinks, patch->new_name);
3890 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3892 do {
3893 while (--name->len && name->buf[name->len] != '/')
3894 ; /* scan backwards */
3895 if (!name->len)
3896 break;
3897 name->buf[name->len] = '\0';
3898 if (strset_contains(&state->kept_symlinks, name->buf))
3899 return 1;
3900 if (strset_contains(&state->removed_symlinks, name->buf))
3902 * This cannot be "return 0", because we may
3903 * see a new one created at a higher level.
3905 continue;
3907 /* otherwise, check the preimage */
3908 if (state->check_index) {
3909 struct cache_entry *ce;
3911 ce = index_file_exists(state->repo->index, name->buf,
3912 name->len, ignore_case);
3913 if (ce && S_ISLNK(ce->ce_mode))
3914 return 1;
3915 } else {
3916 struct stat st;
3917 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3918 return 1;
3920 } while (1);
3921 return 0;
3924 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3926 int ret;
3927 struct strbuf name = STRBUF_INIT;
3929 assert(*name_ != '\0');
3930 strbuf_addstr(&name, name_);
3931 ret = path_is_beyond_symlink_1(state, &name);
3932 strbuf_release(&name);
3934 return ret;
3937 static int check_unsafe_path(struct patch *patch)
3939 const char *old_name = NULL;
3940 const char *new_name = NULL;
3941 if (patch->is_delete)
3942 old_name = patch->old_name;
3943 else if (!patch->is_new && !patch->is_copy)
3944 old_name = patch->old_name;
3945 if (!patch->is_delete)
3946 new_name = patch->new_name;
3948 if (old_name && !verify_path(old_name, patch->old_mode))
3949 return error(_("invalid path '%s'"), old_name);
3950 if (new_name && !verify_path(new_name, patch->new_mode))
3951 return error(_("invalid path '%s'"), new_name);
3952 return 0;
3956 * Check and apply the patch in-core; leave the result in patch->result
3957 * for the caller to write it out to the final destination.
3959 static int check_patch(struct apply_state *state, struct patch *patch)
3961 struct stat st;
3962 const char *old_name = patch->old_name;
3963 const char *new_name = patch->new_name;
3964 const char *name = old_name ? old_name : new_name;
3965 struct cache_entry *ce = NULL;
3966 struct patch *tpatch;
3967 int ok_if_exists;
3968 int status;
3970 patch->rejected = 1; /* we will drop this after we succeed */
3972 status = check_preimage(state, patch, &ce, &st);
3973 if (status)
3974 return status;
3975 old_name = patch->old_name;
3978 * A type-change diff is always split into a patch to delete
3979 * old, immediately followed by a patch to create new (see
3980 * diff.c::run_diff()); in such a case it is Ok that the entry
3981 * to be deleted by the previous patch is still in the working
3982 * tree and in the index.
3984 * A patch to swap-rename between A and B would first rename A
3985 * to B and then rename B to A. While applying the first one,
3986 * the presence of B should not stop A from getting renamed to
3987 * B; ask to_be_deleted() about the later rename. Removal of
3988 * B and rename from A to B is handled the same way by asking
3989 * was_deleted().
3991 if ((tpatch = in_fn_table(state, new_name)) &&
3992 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3993 ok_if_exists = 1;
3994 else
3995 ok_if_exists = 0;
3997 if (new_name &&
3998 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3999 int err = check_to_create(state, new_name, ok_if_exists);
4001 if (err && state->threeway) {
4002 patch->direct_to_threeway = 1;
4003 } else switch (err) {
4004 case 0:
4005 break; /* happy */
4006 case EXISTS_IN_INDEX:
4007 return error(_("%s: already exists in index"), new_name);
4008 case EXISTS_IN_INDEX_AS_ITA:
4009 return error(_("%s: does not match index"), new_name);
4010 case EXISTS_IN_WORKTREE:
4011 return error(_("%s: already exists in working directory"),
4012 new_name);
4013 default:
4014 return err;
4017 if (!patch->new_mode) {
4018 if (0 < patch->is_new)
4019 patch->new_mode = S_IFREG | 0644;
4020 else
4021 patch->new_mode = patch->old_mode;
4025 if (new_name && old_name) {
4026 int same = !strcmp(old_name, new_name);
4027 if (!patch->new_mode)
4028 patch->new_mode = patch->old_mode;
4029 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4030 if (same)
4031 return error(_("new mode (%o) of %s does not "
4032 "match old mode (%o)"),
4033 patch->new_mode, new_name,
4034 patch->old_mode);
4035 else
4036 return error(_("new mode (%o) of %s does not "
4037 "match old mode (%o) of %s"),
4038 patch->new_mode, new_name,
4039 patch->old_mode, old_name);
4043 if (!state->unsafe_paths && check_unsafe_path(patch))
4044 return -128;
4047 * An attempt to read from or delete a path that is beyond a
4048 * symbolic link will be prevented by load_patch_target() that
4049 * is called at the beginning of apply_data() so we do not
4050 * have to worry about a patch marked with "is_delete" bit
4051 * here. We however need to make sure that the patch result
4052 * is not deposited to a path that is beyond a symbolic link
4053 * here.
4055 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4056 return error(_("affected file '%s' is beyond a symbolic link"),
4057 patch->new_name);
4059 if (apply_data(state, patch, &st, ce) < 0)
4060 return error(_("%s: patch does not apply"), name);
4061 patch->rejected = 0;
4062 return 0;
4065 static int check_patch_list(struct apply_state *state, struct patch *patch)
4067 int err = 0;
4069 prepare_symlink_changes(state, patch);
4070 prepare_fn_table(state, patch);
4071 while (patch) {
4072 int res;
4073 if (state->apply_verbosity > verbosity_normal)
4074 say_patch_name(stderr,
4075 _("Checking patch %s..."), patch);
4076 res = check_patch(state, patch);
4077 if (res == -128)
4078 return -128;
4079 err |= res;
4080 patch = patch->next;
4082 return err;
4085 static int read_apply_cache(struct apply_state *state)
4087 if (state->index_file)
4088 return read_index_from(state->repo->index, state->index_file,
4089 get_git_dir());
4090 else
4091 return repo_read_index(state->repo);
4094 /* This function tries to read the object name from the current index */
4095 static int get_current_oid(struct apply_state *state, const char *path,
4096 struct object_id *oid)
4098 int pos;
4100 if (read_apply_cache(state) < 0)
4101 return -1;
4102 pos = index_name_pos(state->repo->index, path, strlen(path));
4103 if (pos < 0)
4104 return -1;
4105 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4106 return 0;
4109 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4112 * A usable gitlink patch has only one fragment (hunk) that looks like:
4113 * @@ -1 +1 @@
4114 * -Subproject commit <old sha1>
4115 * +Subproject commit <new sha1>
4116 * or
4117 * @@ -1 +0,0 @@
4118 * -Subproject commit <old sha1>
4119 * for a removal patch.
4121 struct fragment *hunk = p->fragments;
4122 static const char heading[] = "-Subproject commit ";
4123 char *preimage;
4125 if (/* does the patch have only one hunk? */
4126 hunk && !hunk->next &&
4127 /* is its preimage one line? */
4128 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4129 /* does preimage begin with the heading? */
4130 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4131 starts_with(++preimage, heading) &&
4132 /* does it record full SHA-1? */
4133 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4134 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4135 /* does the abbreviated name on the index line agree with it? */
4136 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4137 return 0; /* it all looks fine */
4139 /* we may have full object name on the index line */
4140 return get_oid_hex(p->old_oid_prefix, oid);
4143 /* Build an index that contains just the files needed for a 3way merge */
4144 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4146 struct patch *patch;
4147 struct index_state result = INDEX_STATE_INIT(state->repo);
4148 struct lock_file lock = LOCK_INIT;
4149 int res;
4151 /* Once we start supporting the reverse patch, it may be
4152 * worth showing the new sha1 prefix, but until then...
4154 for (patch = list; patch; patch = patch->next) {
4155 struct object_id oid;
4156 struct cache_entry *ce;
4157 const char *name;
4159 name = patch->old_name ? patch->old_name : patch->new_name;
4160 if (0 < patch->is_new)
4161 continue;
4163 if (S_ISGITLINK(patch->old_mode)) {
4164 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4165 ; /* ok, the textual part looks sane */
4166 else
4167 return error(_("sha1 information is lacking or "
4168 "useless for submodule %s"), name);
4169 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4170 ; /* ok */
4171 } else if (!patch->lines_added && !patch->lines_deleted) {
4172 /* mode-only change: update the current */
4173 if (get_current_oid(state, patch->old_name, &oid))
4174 return error(_("mode change for %s, which is not "
4175 "in current HEAD"), name);
4176 } else
4177 return error(_("sha1 information is lacking or useless "
4178 "(%s)."), name);
4180 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4181 if (!ce)
4182 return error(_("make_cache_entry failed for path '%s'"),
4183 name);
4184 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4185 discard_cache_entry(ce);
4186 return error(_("could not add %s to temporary index"),
4187 name);
4191 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4192 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4193 discard_index(&result);
4195 if (res)
4196 return error(_("could not write temporary index to %s"),
4197 state->fake_ancestor);
4199 return 0;
4202 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4204 int files, adds, dels;
4206 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4207 files++;
4208 adds += patch->lines_added;
4209 dels += patch->lines_deleted;
4210 show_stats(state, patch);
4213 print_stat_summary(stdout, files, adds, dels);
4216 static void numstat_patch_list(struct apply_state *state,
4217 struct patch *patch)
4219 for ( ; patch; patch = patch->next) {
4220 const char *name;
4221 name = patch->new_name ? patch->new_name : patch->old_name;
4222 if (patch->is_binary)
4223 printf("-\t-\t");
4224 else
4225 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4226 write_name_quoted(name, stdout, state->line_termination);
4230 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4232 if (mode)
4233 printf(" %s mode %06o %s\n", newdelete, mode, name);
4234 else
4235 printf(" %s %s\n", newdelete, name);
4238 static void show_mode_change(struct patch *p, int show_name)
4240 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4241 if (show_name)
4242 printf(" mode change %06o => %06o %s\n",
4243 p->old_mode, p->new_mode, p->new_name);
4244 else
4245 printf(" mode change %06o => %06o\n",
4246 p->old_mode, p->new_mode);
4250 static void show_rename_copy(struct patch *p)
4252 const char *renamecopy = p->is_rename ? "rename" : "copy";
4253 const char *old_name, *new_name;
4255 /* Find common prefix */
4256 old_name = p->old_name;
4257 new_name = p->new_name;
4258 while (1) {
4259 const char *slash_old, *slash_new;
4260 slash_old = strchr(old_name, '/');
4261 slash_new = strchr(new_name, '/');
4262 if (!slash_old ||
4263 !slash_new ||
4264 slash_old - old_name != slash_new - new_name ||
4265 memcmp(old_name, new_name, slash_new - new_name))
4266 break;
4267 old_name = slash_old + 1;
4268 new_name = slash_new + 1;
4270 /* p->old_name through old_name is the common prefix, and old_name and
4271 * new_name through the end of names are renames
4273 if (old_name != p->old_name)
4274 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4275 (int)(old_name - p->old_name), p->old_name,
4276 old_name, new_name, p->score);
4277 else
4278 printf(" %s %s => %s (%d%%)\n", renamecopy,
4279 p->old_name, p->new_name, p->score);
4280 show_mode_change(p, 0);
4283 static void summary_patch_list(struct patch *patch)
4285 struct patch *p;
4287 for (p = patch; p; p = p->next) {
4288 if (p->is_new)
4289 show_file_mode_name("create", p->new_mode, p->new_name);
4290 else if (p->is_delete)
4291 show_file_mode_name("delete", p->old_mode, p->old_name);
4292 else {
4293 if (p->is_rename || p->is_copy)
4294 show_rename_copy(p);
4295 else {
4296 if (p->score) {
4297 printf(" rewrite %s (%d%%)\n",
4298 p->new_name, p->score);
4299 show_mode_change(p, 0);
4301 else
4302 show_mode_change(p, 1);
4308 static void patch_stats(struct apply_state *state, struct patch *patch)
4310 int lines = patch->lines_added + patch->lines_deleted;
4312 if (lines > state->max_change)
4313 state->max_change = lines;
4314 if (patch->old_name) {
4315 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4316 if (!len)
4317 len = strlen(patch->old_name);
4318 if (len > state->max_len)
4319 state->max_len = len;
4321 if (patch->new_name) {
4322 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4323 if (!len)
4324 len = strlen(patch->new_name);
4325 if (len > state->max_len)
4326 state->max_len = len;
4330 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4332 if (state->update_index && !state->ita_only) {
4333 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4334 return error(_("unable to remove %s from index"), patch->old_name);
4336 if (!state->cached) {
4337 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4338 remove_path(patch->old_name);
4341 return 0;
4344 static int add_index_file(struct apply_state *state,
4345 const char *path,
4346 unsigned mode,
4347 void *buf,
4348 unsigned long size)
4350 struct stat st;
4351 struct cache_entry *ce;
4352 int namelen = strlen(path);
4354 ce = make_empty_cache_entry(state->repo->index, namelen);
4355 memcpy(ce->name, path, namelen);
4356 ce->ce_mode = create_ce_mode(mode);
4357 ce->ce_flags = create_ce_flags(0);
4358 ce->ce_namelen = namelen;
4359 if (state->ita_only) {
4360 ce->ce_flags |= CE_INTENT_TO_ADD;
4361 set_object_name_for_intent_to_add_entry(ce);
4362 } else if (S_ISGITLINK(mode)) {
4363 const char *s;
4365 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4366 get_oid_hex(s, &ce->oid)) {
4367 discard_cache_entry(ce);
4368 return error(_("corrupt patch for submodule %s"), path);
4370 } else {
4371 if (!state->cached) {
4372 if (lstat(path, &st) < 0) {
4373 discard_cache_entry(ce);
4374 return error_errno(_("unable to stat newly "
4375 "created file '%s'"),
4376 path);
4378 fill_stat_cache_info(state->repo->index, ce, &st);
4380 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4381 discard_cache_entry(ce);
4382 return error(_("unable to create backing store "
4383 "for newly created file %s"), path);
4386 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4387 discard_cache_entry(ce);
4388 return error(_("unable to add cache entry for %s"), path);
4391 return 0;
4395 * Returns:
4396 * -1 if an unrecoverable error happened
4397 * 0 if everything went well
4398 * 1 if a recoverable error happened
4400 static int try_create_file(struct apply_state *state, const char *path,
4401 unsigned int mode, const char *buf,
4402 unsigned long size)
4404 int fd, res;
4405 struct strbuf nbuf = STRBUF_INIT;
4407 if (S_ISGITLINK(mode)) {
4408 struct stat st;
4409 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4410 return 0;
4411 return !!mkdir(path, 0777);
4414 if (has_symlinks && S_ISLNK(mode))
4415 /* Although buf:size is counted string, it also is NUL
4416 * terminated.
4418 return !!symlink(buf, path);
4420 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4421 if (fd < 0)
4422 return 1;
4424 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4425 size = nbuf.len;
4426 buf = nbuf.buf;
4429 res = write_in_full(fd, buf, size) < 0;
4430 if (res)
4431 error_errno(_("failed to write to '%s'"), path);
4432 strbuf_release(&nbuf);
4434 if (close(fd) < 0 && !res)
4435 return error_errno(_("closing file '%s'"), path);
4437 return res ? -1 : 0;
4441 * We optimistically assume that the directories exist,
4442 * which is true 99% of the time anyway. If they don't,
4443 * we create them and try again.
4445 * Returns:
4446 * -1 on error
4447 * 0 otherwise
4449 static int create_one_file(struct apply_state *state,
4450 char *path,
4451 unsigned mode,
4452 const char *buf,
4453 unsigned long size)
4455 char *newpath = NULL;
4456 int res;
4458 if (state->cached)
4459 return 0;
4462 * We already try to detect whether files are beyond a symlink in our
4463 * up-front checks. But in the case where symlinks are created by any
4464 * of the intermediate hunks it can happen that our up-front checks
4465 * didn't yet see the symlink, but at the point of arriving here there
4466 * in fact is one. We thus repeat the check for symlinks here.
4468 * Note that this does not make the up-front check obsolete as the
4469 * failure mode is different:
4471 * - The up-front checks cause us to abort before we have written
4472 * anything into the working directory. So when we exit this way the
4473 * working directory remains clean.
4475 * - The checks here happen in the middle of the action where we have
4476 * already started to apply the patch. The end result will be a dirty
4477 * working directory.
4479 * Ideally, we should update the up-front checks to catch what would
4480 * happen when we apply the patch before we damage the working tree.
4481 * We have all the information necessary to do so. But for now, as a
4482 * part of embargoed security work, having this check would serve as a
4483 * reasonable first step.
4485 if (path_is_beyond_symlink(state, path))
4486 return error(_("affected file '%s' is beyond a symbolic link"), path);
4488 res = try_create_file(state, path, mode, buf, size);
4489 if (res < 0)
4490 return -1;
4491 if (!res)
4492 return 0;
4494 if (errno == ENOENT) {
4495 if (safe_create_leading_directories_no_share(path))
4496 return 0;
4497 res = try_create_file(state, path, mode, buf, size);
4498 if (res < 0)
4499 return -1;
4500 if (!res)
4501 return 0;
4504 if (errno == EEXIST || errno == EACCES) {
4505 /* We may be trying to create a file where a directory
4506 * used to be.
4508 struct stat st;
4509 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4510 errno = EEXIST;
4513 if (errno == EEXIST) {
4514 unsigned int nr = getpid();
4516 for (;;) {
4517 newpath = mkpathdup("%s~%u", path, nr);
4518 res = try_create_file(state, newpath, mode, buf, size);
4519 if (res < 0)
4520 goto out;
4521 if (!res) {
4522 if (!rename(newpath, path))
4523 goto out;
4524 unlink_or_warn(newpath);
4525 break;
4527 if (errno != EEXIST)
4528 break;
4529 ++nr;
4530 FREE_AND_NULL(newpath);
4533 res = error_errno(_("unable to write file '%s' mode %o"), path, mode);
4534 out:
4535 free(newpath);
4536 return res;
4539 static int add_conflicted_stages_file(struct apply_state *state,
4540 struct patch *patch)
4542 int stage, namelen;
4543 unsigned mode;
4544 struct cache_entry *ce;
4546 if (!state->update_index)
4547 return 0;
4548 namelen = strlen(patch->new_name);
4549 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4551 remove_file_from_index(state->repo->index, patch->new_name);
4552 for (stage = 1; stage < 4; stage++) {
4553 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4554 continue;
4555 ce = make_empty_cache_entry(state->repo->index, namelen);
4556 memcpy(ce->name, patch->new_name, namelen);
4557 ce->ce_mode = create_ce_mode(mode);
4558 ce->ce_flags = create_ce_flags(stage);
4559 ce->ce_namelen = namelen;
4560 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4561 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4562 discard_cache_entry(ce);
4563 return error(_("unable to add cache entry for %s"),
4564 patch->new_name);
4568 return 0;
4571 static int create_file(struct apply_state *state, struct patch *patch)
4573 char *path = patch->new_name;
4574 unsigned mode = patch->new_mode;
4575 unsigned long size = patch->resultsize;
4576 char *buf = patch->result;
4578 if (!mode)
4579 mode = S_IFREG | 0644;
4580 if (create_one_file(state, path, mode, buf, size))
4581 return -1;
4583 if (patch->conflicted_threeway)
4584 return add_conflicted_stages_file(state, patch);
4585 else if (state->update_index)
4586 return add_index_file(state, path, mode, buf, size);
4587 return 0;
4590 /* phase zero is to remove, phase one is to create */
4591 static int write_out_one_result(struct apply_state *state,
4592 struct patch *patch,
4593 int phase)
4595 if (patch->is_delete > 0) {
4596 if (phase == 0)
4597 return remove_file(state, patch, 1);
4598 return 0;
4600 if (patch->is_new > 0 || patch->is_copy) {
4601 if (phase == 1)
4602 return create_file(state, patch);
4603 return 0;
4606 * Rename or modification boils down to the same
4607 * thing: remove the old, write the new
4609 if (phase == 0)
4610 return remove_file(state, patch, patch->is_rename);
4611 if (phase == 1)
4612 return create_file(state, patch);
4613 return 0;
4616 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4618 FILE *rej;
4619 char *namebuf;
4620 struct fragment *frag;
4621 int fd, cnt = 0;
4622 struct strbuf sb = STRBUF_INIT;
4624 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4625 if (!frag->rejected)
4626 continue;
4627 cnt++;
4630 if (!cnt) {
4631 if (state->apply_verbosity > verbosity_normal)
4632 say_patch_name(stderr,
4633 _("Applied patch %s cleanly."), patch);
4634 return 0;
4637 /* This should not happen, because a removal patch that leaves
4638 * contents are marked "rejected" at the patch level.
4640 if (!patch->new_name)
4641 die(_("internal error"));
4643 /* Say this even without --verbose */
4644 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4645 "Applying patch %%s with %d rejects...",
4646 cnt),
4647 cnt);
4648 if (state->apply_verbosity > verbosity_silent)
4649 say_patch_name(stderr, sb.buf, patch);
4650 strbuf_release(&sb);
4652 namebuf = xstrfmt("%s.rej", patch->new_name);
4654 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4655 if (fd < 0) {
4656 if (errno != EEXIST) {
4657 error_errno(_("cannot open %s"), namebuf);
4658 goto error;
4660 if (unlink(namebuf)) {
4661 error_errno(_("cannot unlink '%s'"), namebuf);
4662 goto error;
4664 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4665 if (fd < 0) {
4666 error_errno(_("cannot open %s"), namebuf);
4667 goto error;
4670 rej = fdopen(fd, "w");
4671 if (!rej) {
4672 error_errno(_("cannot open %s"), namebuf);
4673 close(fd);
4674 goto error;
4677 /* Normal git tools never deal with .rej, so do not pretend
4678 * this is a git patch by saying --git or giving extended
4679 * headers. While at it, maybe please "kompare" that wants
4680 * the trailing TAB and some garbage at the end of line ;-).
4682 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4683 patch->new_name, patch->new_name);
4684 for (cnt = 1, frag = patch->fragments;
4685 frag;
4686 cnt++, frag = frag->next) {
4687 if (!frag->rejected) {
4688 if (state->apply_verbosity > verbosity_silent)
4689 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4690 continue;
4692 if (state->apply_verbosity > verbosity_silent)
4693 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4694 fprintf(rej, "%.*s", frag->size, frag->patch);
4695 if (frag->patch[frag->size-1] != '\n')
4696 fputc('\n', rej);
4698 fclose(rej);
4699 error:
4700 free(namebuf);
4701 return -1;
4705 * Returns:
4706 * -1 if an error happened
4707 * 0 if the patch applied cleanly
4708 * 1 if the patch did not apply cleanly
4710 static int write_out_results(struct apply_state *state, struct patch *list)
4712 int phase;
4713 int errs = 0;
4714 struct patch *l;
4715 struct string_list cpath = STRING_LIST_INIT_DUP;
4717 for (phase = 0; phase < 2; phase++) {
4718 l = list;
4719 while (l) {
4720 if (l->rejected)
4721 errs = 1;
4722 else {
4723 if (write_out_one_result(state, l, phase)) {
4724 string_list_clear(&cpath, 0);
4725 return -1;
4727 if (phase == 1) {
4728 if (write_out_one_reject(state, l))
4729 errs = 1;
4730 if (l->conflicted_threeway) {
4731 string_list_append(&cpath, l->new_name);
4732 errs = 1;
4736 l = l->next;
4740 if (cpath.nr) {
4741 struct string_list_item *item;
4743 string_list_sort(&cpath);
4744 if (state->apply_verbosity > verbosity_silent) {
4745 for_each_string_list_item(item, &cpath)
4746 fprintf(stderr, "U %s\n", item->string);
4748 string_list_clear(&cpath, 0);
4751 * rerere relies on the partially merged result being in the working
4752 * tree with conflict markers, but that isn't written with --cached.
4754 if (!state->cached)
4755 repo_rerere(state->repo, 0);
4758 return errs;
4762 * Try to apply a patch.
4764 * Returns:
4765 * -128 if a bad error happened (like patch unreadable)
4766 * -1 if patch did not apply and user cannot deal with it
4767 * 0 if the patch applied
4768 * 1 if the patch did not apply but user might fix it
4770 static int apply_patch(struct apply_state *state,
4771 int fd,
4772 const char *filename,
4773 int options)
4775 size_t offset;
4776 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4777 struct patch *list = NULL, **listp = &list;
4778 int skipped_patch = 0;
4779 int res = 0;
4780 int flush_attributes = 0;
4782 state->patch_input_file = filename;
4783 if (read_patch_file(&buf, fd) < 0)
4784 return -128;
4785 offset = 0;
4786 while (offset < buf.len) {
4787 struct patch *patch;
4788 int nr;
4790 CALLOC_ARRAY(patch, 1);
4791 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4792 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4793 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4794 if (nr < 0) {
4795 free_patch(patch);
4796 if (nr == -128) {
4797 res = -128;
4798 goto end;
4800 break;
4802 if (state->apply_in_reverse)
4803 reverse_patches(patch);
4804 if (use_patch(state, patch)) {
4805 patch_stats(state, patch);
4806 if (!list || !state->apply_in_reverse) {
4807 *listp = patch;
4808 listp = &patch->next;
4809 } else {
4810 patch->next = list;
4811 list = patch;
4814 if ((patch->new_name &&
4815 ends_with_path_components(patch->new_name,
4816 GITATTRIBUTES_FILE)) ||
4817 (patch->old_name &&
4818 ends_with_path_components(patch->old_name,
4819 GITATTRIBUTES_FILE)))
4820 flush_attributes = 1;
4822 else {
4823 if (state->apply_verbosity > verbosity_normal)
4824 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4825 free_patch(patch);
4826 skipped_patch++;
4828 offset += nr;
4831 if (!list && !skipped_patch) {
4832 if (!state->allow_empty) {
4833 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4834 res = -128;
4836 goto end;
4839 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4840 state->apply = 0;
4842 state->update_index = (state->check_index || state->ita_only) && state->apply;
4843 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4844 if (state->index_file)
4845 hold_lock_file_for_update(&state->lock_file,
4846 state->index_file,
4847 LOCK_DIE_ON_ERROR);
4848 else
4849 repo_hold_locked_index(state->repo, &state->lock_file,
4850 LOCK_DIE_ON_ERROR);
4853 if (state->check_index && read_apply_cache(state) < 0) {
4854 error(_("unable to read index file"));
4855 res = -128;
4856 goto end;
4859 if (state->check || state->apply) {
4860 int r = check_patch_list(state, list);
4861 if (r == -128) {
4862 res = -128;
4863 goto end;
4865 if (r < 0 && !state->apply_with_reject) {
4866 res = -1;
4867 goto end;
4871 if (state->apply) {
4872 int write_res = write_out_results(state, list);
4873 if (write_res < 0) {
4874 res = -128;
4875 goto end;
4877 if (write_res > 0) {
4878 /* with --3way, we still need to write the index out */
4879 res = state->apply_with_reject ? -1 : 1;
4880 goto end;
4884 if (state->fake_ancestor &&
4885 build_fake_ancestor(state, list)) {
4886 res = -128;
4887 goto end;
4890 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4891 stat_patch_list(state, list);
4893 if (state->numstat && state->apply_verbosity > verbosity_silent)
4894 numstat_patch_list(state, list);
4896 if (state->summary && state->apply_verbosity > verbosity_silent)
4897 summary_patch_list(list);
4899 if (flush_attributes)
4900 reset_parsed_attributes();
4901 end:
4902 free_patch_list(list);
4903 strbuf_release(&buf);
4904 string_list_clear(&state->fn_table, 0);
4905 return res;
4908 static int apply_option_parse_exclude(const struct option *opt,
4909 const char *arg, int unset)
4911 struct apply_state *state = opt->value;
4913 BUG_ON_OPT_NEG(unset);
4915 add_name_limit(state, arg, 1);
4916 return 0;
4919 static int apply_option_parse_include(const struct option *opt,
4920 const char *arg, int unset)
4922 struct apply_state *state = opt->value;
4924 BUG_ON_OPT_NEG(unset);
4926 add_name_limit(state, arg, 0);
4927 state->has_include = 1;
4928 return 0;
4931 static int apply_option_parse_p(const struct option *opt,
4932 const char *arg,
4933 int unset)
4935 struct apply_state *state = opt->value;
4937 BUG_ON_OPT_NEG(unset);
4939 state->p_value = atoi(arg);
4940 state->p_value_known = 1;
4941 return 0;
4944 static int apply_option_parse_space_change(const struct option *opt,
4945 const char *arg, int unset)
4947 struct apply_state *state = opt->value;
4949 BUG_ON_OPT_ARG(arg);
4951 if (unset)
4952 state->ws_ignore_action = ignore_ws_none;
4953 else
4954 state->ws_ignore_action = ignore_ws_change;
4955 return 0;
4958 static int apply_option_parse_whitespace(const struct option *opt,
4959 const char *arg, int unset)
4961 struct apply_state *state = opt->value;
4963 BUG_ON_OPT_NEG(unset);
4965 state->whitespace_option = arg;
4966 if (parse_whitespace_option(state, arg))
4967 return -1;
4968 return 0;
4971 static int apply_option_parse_directory(const struct option *opt,
4972 const char *arg, int unset)
4974 struct apply_state *state = opt->value;
4976 BUG_ON_OPT_NEG(unset);
4978 strbuf_reset(&state->root);
4979 strbuf_addstr(&state->root, arg);
4980 strbuf_complete(&state->root, '/');
4981 return 0;
4984 int apply_all_patches(struct apply_state *state,
4985 int argc,
4986 const char **argv,
4987 int options)
4989 int i;
4990 int res;
4991 int errs = 0;
4992 int read_stdin = 1;
4994 for (i = 0; i < argc; i++) {
4995 const char *arg = argv[i];
4996 char *to_free = NULL;
4997 int fd;
4999 if (!strcmp(arg, "-")) {
5000 res = apply_patch(state, 0, "<stdin>", options);
5001 if (res < 0)
5002 goto end;
5003 errs |= res;
5004 read_stdin = 0;
5005 continue;
5006 } else
5007 arg = to_free = prefix_filename(state->prefix, arg);
5009 fd = open(arg, O_RDONLY);
5010 if (fd < 0) {
5011 error(_("can't open patch '%s': %s"), arg, strerror(errno));
5012 res = -128;
5013 free(to_free);
5014 goto end;
5016 read_stdin = 0;
5017 set_default_whitespace_mode(state);
5018 res = apply_patch(state, fd, arg, options);
5019 close(fd);
5020 free(to_free);
5021 if (res < 0)
5022 goto end;
5023 errs |= res;
5025 set_default_whitespace_mode(state);
5026 if (read_stdin) {
5027 res = apply_patch(state, 0, "<stdin>", options);
5028 if (res < 0)
5029 goto end;
5030 errs |= res;
5033 if (state->whitespace_error) {
5034 if (state->squelch_whitespace_errors &&
5035 state->squelch_whitespace_errors < state->whitespace_error) {
5036 int squelched =
5037 state->whitespace_error - state->squelch_whitespace_errors;
5038 warning(Q_("squelched %d whitespace error",
5039 "squelched %d whitespace errors",
5040 squelched),
5041 squelched);
5043 if (state->ws_error_action == die_on_ws_error) {
5044 error(Q_("%d line adds whitespace errors.",
5045 "%d lines add whitespace errors.",
5046 state->whitespace_error),
5047 state->whitespace_error);
5048 res = -128;
5049 goto end;
5051 if (state->applied_after_fixing_ws && state->apply)
5052 warning(Q_("%d line applied after"
5053 " fixing whitespace errors.",
5054 "%d lines applied after"
5055 " fixing whitespace errors.",
5056 state->applied_after_fixing_ws),
5057 state->applied_after_fixing_ws);
5058 else if (state->whitespace_error)
5059 warning(Q_("%d line adds whitespace errors.",
5060 "%d lines add whitespace errors.",
5061 state->whitespace_error),
5062 state->whitespace_error);
5065 if (state->update_index) {
5066 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5067 if (res) {
5068 error(_("Unable to write new index file"));
5069 res = -128;
5070 goto end;
5074 res = !!errs;
5076 end:
5077 rollback_lock_file(&state->lock_file);
5079 if (state->apply_verbosity <= verbosity_silent) {
5080 set_error_routine(state->saved_error_routine);
5081 set_warn_routine(state->saved_warn_routine);
5084 if (res > -1)
5085 return res;
5086 return (res == -1 ? 1 : 128);
5089 int apply_parse_options(int argc, const char **argv,
5090 struct apply_state *state,
5091 int *force_apply, int *options,
5092 const char * const *apply_usage)
5094 struct option builtin_apply_options[] = {
5095 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5096 N_("don't apply changes matching the given path"),
5097 PARSE_OPT_NONEG, apply_option_parse_exclude),
5098 OPT_CALLBACK_F(0, "include", state, N_("path"),
5099 N_("apply changes matching the given path"),
5100 PARSE_OPT_NONEG, apply_option_parse_include),
5101 OPT_CALLBACK('p', NULL, state, N_("num"),
5102 N_("remove <num> leading slashes from traditional diff paths"),
5103 apply_option_parse_p),
5104 OPT_BOOL(0, "no-add", &state->no_add,
5105 N_("ignore additions made by the patch")),
5106 OPT_BOOL(0, "stat", &state->diffstat,
5107 N_("instead of applying the patch, output diffstat for the input")),
5108 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5109 OPT_NOOP_NOARG(0, "binary"),
5110 OPT_BOOL(0, "numstat", &state->numstat,
5111 N_("show number of added and deleted lines in decimal notation")),
5112 OPT_BOOL(0, "summary", &state->summary,
5113 N_("instead of applying the patch, output a summary for the input")),
5114 OPT_BOOL(0, "check", &state->check,
5115 N_("instead of applying the patch, see if the patch is applicable")),
5116 OPT_BOOL(0, "index", &state->check_index,
5117 N_("make sure the patch is applicable to the current index")),
5118 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5119 N_("mark new files with `git add --intent-to-add`")),
5120 OPT_BOOL(0, "cached", &state->cached,
5121 N_("apply a patch without touching the working tree")),
5122 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5123 N_("accept a patch that touches outside the working area"),
5124 PARSE_OPT_NOCOMPLETE),
5125 OPT_BOOL(0, "apply", force_apply,
5126 N_("also apply the patch (use with --stat/--summary/--check)")),
5127 OPT_BOOL('3', "3way", &state->threeway,
5128 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5129 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5130 N_("build a temporary index based on embedded index information")),
5131 /* Think twice before adding "--nul" synonym to this */
5132 OPT_SET_INT('z', NULL, &state->line_termination,
5133 N_("paths are separated with NUL character"), '\0'),
5134 OPT_INTEGER('C', NULL, &state->p_context,
5135 N_("ensure at least <n> lines of context match")),
5136 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5137 N_("detect new or modified lines that have whitespace errors"),
5138 apply_option_parse_whitespace),
5139 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5140 N_("ignore changes in whitespace when finding context"),
5141 PARSE_OPT_NOARG, apply_option_parse_space_change),
5142 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5143 N_("ignore changes in whitespace when finding context"),
5144 PARSE_OPT_NOARG, apply_option_parse_space_change),
5145 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5146 N_("apply the patch in reverse")),
5147 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5148 N_("don't expect at least one line of context")),
5149 OPT_BOOL(0, "reject", &state->apply_with_reject,
5150 N_("leave the rejected hunks in corresponding *.rej files")),
5151 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5152 N_("allow overlapping hunks")),
5153 OPT__VERBOSITY(&state->apply_verbosity),
5154 OPT_BIT(0, "inaccurate-eof", options,
5155 N_("tolerate incorrectly detected missing new-line at the end of file"),
5156 APPLY_OPT_INACCURATE_EOF),
5157 OPT_BIT(0, "recount", options,
5158 N_("do not trust the line counts in the hunk headers"),
5159 APPLY_OPT_RECOUNT),
5160 OPT_CALLBACK(0, "directory", state, N_("root"),
5161 N_("prepend <root> to all filenames"),
5162 apply_option_parse_directory),
5163 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5164 N_("don't return error for empty patches")),
5165 OPT_END()
5168 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);