repository: drop `the_index` variable
[git/gitster.git] / apply.c
blob34f20326a7f0ad2328b83af969d2c5419c2349eb
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
10 #include "git-compat-util.h"
11 #include "abspath.h"
12 #include "base85.h"
13 #include "config.h"
14 #include "object-store-ll.h"
15 #include "delta.h"
16 #include "diff.h"
17 #include "dir.h"
18 #include "environment.h"
19 #include "gettext.h"
20 #include "hex.h"
21 #include "xdiff-interface.h"
22 #include "merge-ll.h"
23 #include "lockfile.h"
24 #include "name-hash.h"
25 #include "object-name.h"
26 #include "object-file.h"
27 #include "parse-options.h"
28 #include "path.h"
29 #include "quote.h"
30 #include "read-cache.h"
31 #include "rerere.h"
32 #include "apply.h"
33 #include "entry.h"
34 #include "setup.h"
35 #include "symlinks.h"
36 #include "wildmatch.h"
37 #include "ws.h"
39 struct gitdiff_data {
40 struct strbuf *root;
41 int linenr;
42 int p_value;
45 static void git_apply_config(void)
47 git_config_get_string("apply.whitespace", &apply_default_whitespace);
48 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
49 git_config(git_xmerge_config, NULL);
52 static int parse_whitespace_option(struct apply_state *state, const char *option)
54 if (!option) {
55 state->ws_error_action = warn_on_ws_error;
56 return 0;
58 if (!strcmp(option, "warn")) {
59 state->ws_error_action = warn_on_ws_error;
60 return 0;
62 if (!strcmp(option, "nowarn")) {
63 state->ws_error_action = nowarn_ws_error;
64 return 0;
66 if (!strcmp(option, "error")) {
67 state->ws_error_action = die_on_ws_error;
68 return 0;
70 if (!strcmp(option, "error-all")) {
71 state->ws_error_action = die_on_ws_error;
72 state->squelch_whitespace_errors = 0;
73 return 0;
75 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
76 state->ws_error_action = correct_ws_error;
77 return 0;
80 * Please update $__git_whitespacelist in git-completion.bash,
81 * Documentation/git-apply.txt, and Documentation/git-am.txt
82 * when you add new options.
84 return error(_("unrecognized whitespace option '%s'"), option);
87 static int parse_ignorewhitespace_option(struct apply_state *state,
88 const char *option)
90 if (!option || !strcmp(option, "no") ||
91 !strcmp(option, "false") || !strcmp(option, "never") ||
92 !strcmp(option, "none")) {
93 state->ws_ignore_action = ignore_ws_none;
94 return 0;
96 if (!strcmp(option, "change")) {
97 state->ws_ignore_action = ignore_ws_change;
98 return 0;
100 return error(_("unrecognized whitespace ignore option '%s'"), option);
103 int init_apply_state(struct apply_state *state,
104 struct repository *repo,
105 const char *prefix)
107 memset(state, 0, sizeof(*state));
108 state->prefix = prefix;
109 state->repo = repo;
110 state->apply = 1;
111 state->line_termination = '\n';
112 state->p_value = 1;
113 state->p_context = UINT_MAX;
114 state->squelch_whitespace_errors = 5;
115 state->ws_error_action = warn_on_ws_error;
116 state->ws_ignore_action = ignore_ws_none;
117 state->linenr = 1;
118 string_list_init_nodup(&state->fn_table);
119 string_list_init_nodup(&state->limit_by_name);
120 strset_init(&state->removed_symlinks);
121 strset_init(&state->kept_symlinks);
122 strbuf_init(&state->root, 0);
124 git_apply_config();
125 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
126 return -1;
127 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
128 return -1;
129 return 0;
132 void clear_apply_state(struct apply_state *state)
134 string_list_clear(&state->limit_by_name, 0);
135 strset_clear(&state->removed_symlinks);
136 strset_clear(&state->kept_symlinks);
137 strbuf_release(&state->root);
139 /* &state->fn_table is cleared at the end of apply_patch() */
142 static void mute_routine(const char *msg UNUSED, va_list params UNUSED)
144 /* do nothing */
147 int check_apply_state(struct apply_state *state, int force_apply)
149 int is_not_gitdir = !startup_info->have_repository;
151 if (state->apply_with_reject && state->threeway)
152 return error(_("options '%s' and '%s' cannot be used together"), "--reject", "--3way");
153 if (state->threeway) {
154 if (is_not_gitdir)
155 return error(_("'%s' outside a repository"), "--3way");
156 state->check_index = 1;
158 if (state->apply_with_reject) {
159 state->apply = 1;
160 if (state->apply_verbosity == verbosity_normal)
161 state->apply_verbosity = verbosity_verbose;
163 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
164 state->apply = 0;
165 if (state->check_index && is_not_gitdir)
166 return error(_("'%s' outside a repository"), "--index");
167 if (state->cached) {
168 if (is_not_gitdir)
169 return error(_("'%s' outside a repository"), "--cached");
170 state->check_index = 1;
172 if (state->ita_only && (state->check_index || is_not_gitdir))
173 state->ita_only = 0;
174 if (state->check_index)
175 state->unsafe_paths = 0;
177 if (state->apply_verbosity <= verbosity_silent) {
178 state->saved_error_routine = get_error_routine();
179 state->saved_warn_routine = get_warn_routine();
180 set_error_routine(mute_routine);
181 set_warn_routine(mute_routine);
184 return 0;
187 static void set_default_whitespace_mode(struct apply_state *state)
189 if (!state->whitespace_option && !apply_default_whitespace)
190 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
194 * This represents one "hunk" from a patch, starting with
195 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
196 * patch text is pointed at by patch, and its byte length
197 * is stored in size. leading and trailing are the number
198 * of context lines.
200 struct fragment {
201 unsigned long leading, trailing;
202 unsigned long oldpos, oldlines;
203 unsigned long newpos, newlines;
205 * 'patch' is usually borrowed from buf in apply_patch(),
206 * but some codepaths store an allocated buffer.
208 const char *patch;
209 unsigned free_patch:1,
210 rejected:1;
211 int size;
212 int linenr;
213 struct fragment *next;
217 * When dealing with a binary patch, we reuse "leading" field
218 * to store the type of the binary hunk, either deflated "delta"
219 * or deflated "literal".
221 #define binary_patch_method leading
222 #define BINARY_DELTA_DEFLATED 1
223 #define BINARY_LITERAL_DEFLATED 2
225 static void free_fragment_list(struct fragment *list)
227 while (list) {
228 struct fragment *next = list->next;
229 if (list->free_patch)
230 free((char *)list->patch);
231 free(list);
232 list = next;
236 void release_patch(struct patch *patch)
238 free_fragment_list(patch->fragments);
239 free(patch->def_name);
240 free(patch->old_name);
241 free(patch->new_name);
242 free(patch->result);
245 static void free_patch(struct patch *patch)
247 release_patch(patch);
248 free(patch);
251 static void free_patch_list(struct patch *list)
253 while (list) {
254 struct patch *next = list->next;
255 free_patch(list);
256 list = next;
261 * A line in a file, len-bytes long (includes the terminating LF,
262 * except for an incomplete line at the end if the file ends with
263 * one), and its contents hashes to 'hash'.
265 struct line {
266 size_t len;
267 unsigned hash : 24;
268 unsigned flag : 8;
269 #define LINE_COMMON 1
270 #define LINE_PATCHED 2
274 * This represents a "file", which is an array of "lines".
276 struct image {
277 char *buf;
278 size_t len;
279 size_t nr;
280 size_t alloc;
281 struct line *line_allocated;
282 struct line *line;
285 static uint32_t hash_line(const char *cp, size_t len)
287 size_t i;
288 uint32_t h;
289 for (i = 0, h = 0; i < len; i++) {
290 if (!isspace(cp[i])) {
291 h = h * 3 + (cp[i] & 0xff);
294 return h;
298 * Compare lines s1 of length n1 and s2 of length n2, ignoring
299 * whitespace difference. Returns 1 if they match, 0 otherwise
301 static int fuzzy_matchlines(const char *s1, size_t n1,
302 const char *s2, size_t n2)
304 const char *end1 = s1 + n1;
305 const char *end2 = s2 + n2;
307 /* ignore line endings */
308 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
309 end1--;
310 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
311 end2--;
313 while (s1 < end1 && s2 < end2) {
314 if (isspace(*s1)) {
316 * Skip whitespace. We check on both buffers
317 * because we don't want "a b" to match "ab".
319 if (!isspace(*s2))
320 return 0;
321 while (s1 < end1 && isspace(*s1))
322 s1++;
323 while (s2 < end2 && isspace(*s2))
324 s2++;
325 } else if (*s1++ != *s2++)
326 return 0;
329 /* If we reached the end on one side only, lines don't match. */
330 return s1 == end1 && s2 == end2;
333 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
335 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
336 img->line_allocated[img->nr].len = len;
337 img->line_allocated[img->nr].hash = hash_line(bol, len);
338 img->line_allocated[img->nr].flag = flag;
339 img->nr++;
343 * "buf" has the file contents to be patched (read from various sources).
344 * attach it to "image" and add line-based index to it.
345 * "image" now owns the "buf".
347 static void prepare_image(struct image *image, char *buf, size_t len,
348 int prepare_linetable)
350 const char *cp, *ep;
352 memset(image, 0, sizeof(*image));
353 image->buf = buf;
354 image->len = len;
356 if (!prepare_linetable)
357 return;
359 ep = image->buf + image->len;
360 cp = image->buf;
361 while (cp < ep) {
362 const char *next;
363 for (next = cp; next < ep && *next != '\n'; next++)
365 if (next < ep)
366 next++;
367 add_line_info(image, cp, next - cp, 0);
368 cp = next;
370 image->line = image->line_allocated;
373 static void clear_image(struct image *image)
375 free(image->buf);
376 free(image->line_allocated);
377 memset(image, 0, sizeof(*image));
380 /* fmt must contain _one_ %s and no other substitution */
381 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
383 struct strbuf sb = STRBUF_INIT;
385 if (patch->old_name && patch->new_name &&
386 strcmp(patch->old_name, patch->new_name)) {
387 quote_c_style(patch->old_name, &sb, NULL, 0);
388 strbuf_addstr(&sb, " => ");
389 quote_c_style(patch->new_name, &sb, NULL, 0);
390 } else {
391 const char *n = patch->new_name;
392 if (!n)
393 n = patch->old_name;
394 quote_c_style(n, &sb, NULL, 0);
396 fprintf(output, fmt, sb.buf);
397 fputc('\n', output);
398 strbuf_release(&sb);
401 #define SLOP (16)
404 * apply.c isn't equipped to handle arbitrarily large patches, because
405 * it intermingles `unsigned long` with `int` for the type used to store
406 * buffer lengths.
408 * Only process patches that are just shy of 1 GiB large in order to
409 * avoid any truncation or overflow issues.
411 #define MAX_APPLY_SIZE (1024UL * 1024 * 1023)
413 static int read_patch_file(struct strbuf *sb, int fd)
415 if (strbuf_read(sb, fd, 0) < 0)
416 return error_errno(_("failed to read patch"));
417 else if (sb->len >= MAX_APPLY_SIZE)
418 return error(_("patch too large"));
420 * Make sure that we have some slop in the buffer
421 * so that we can do speculative "memcmp" etc, and
422 * see to it that it is NUL-filled.
424 strbuf_grow(sb, SLOP);
425 memset(sb->buf + sb->len, 0, SLOP);
426 return 0;
429 static unsigned long linelen(const char *buffer, unsigned long size)
431 unsigned long len = 0;
432 while (size--) {
433 len++;
434 if (*buffer++ == '\n')
435 break;
437 return len;
440 static int is_dev_null(const char *str)
442 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
445 #define TERM_SPACE 1
446 #define TERM_TAB 2
448 static int name_terminate(int c, int terminate)
450 if (c == ' ' && !(terminate & TERM_SPACE))
451 return 0;
452 if (c == '\t' && !(terminate & TERM_TAB))
453 return 0;
455 return 1;
458 /* remove double slashes to make --index work with such filenames */
459 static char *squash_slash(char *name)
461 int i = 0, j = 0;
463 if (!name)
464 return NULL;
466 while (name[i]) {
467 if ((name[j++] = name[i++]) == '/')
468 while (name[i] == '/')
469 i++;
471 name[j] = '\0';
472 return name;
475 static char *find_name_gnu(struct strbuf *root,
476 const char *line,
477 int p_value)
479 struct strbuf name = STRBUF_INIT;
480 char *cp;
483 * Proposed "new-style" GNU patch/diff format; see
484 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
486 if (unquote_c_style(&name, line, NULL)) {
487 strbuf_release(&name);
488 return NULL;
491 for (cp = name.buf; p_value; p_value--) {
492 cp = strchr(cp, '/');
493 if (!cp) {
494 strbuf_release(&name);
495 return NULL;
497 cp++;
500 strbuf_remove(&name, 0, cp - name.buf);
501 if (root->len)
502 strbuf_insert(&name, 0, root->buf, root->len);
503 return squash_slash(strbuf_detach(&name, NULL));
506 static size_t sane_tz_len(const char *line, size_t len)
508 const char *tz, *p;
510 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
511 return 0;
512 tz = line + len - strlen(" +0500");
514 if (tz[1] != '+' && tz[1] != '-')
515 return 0;
517 for (p = tz + 2; p != line + len; p++)
518 if (!isdigit(*p))
519 return 0;
521 return line + len - tz;
524 static size_t tz_with_colon_len(const char *line, size_t len)
526 const char *tz, *p;
528 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
529 return 0;
530 tz = line + len - strlen(" +08:00");
532 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
533 return 0;
534 p = tz + 2;
535 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
536 !isdigit(*p++) || !isdigit(*p++))
537 return 0;
539 return line + len - tz;
542 static size_t date_len(const char *line, size_t len)
544 const char *date, *p;
546 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
547 return 0;
548 p = date = line + len - strlen("72-02-05");
550 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
551 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
552 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
553 return 0;
555 if (date - line >= strlen("19") &&
556 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
557 date -= strlen("19");
559 return line + len - date;
562 static size_t short_time_len(const char *line, size_t len)
564 const char *time, *p;
566 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
567 return 0;
568 p = time = line + len - strlen(" 07:01:32");
570 /* Permit 1-digit hours? */
571 if (*p++ != ' ' ||
572 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
573 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
574 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
575 return 0;
577 return line + len - time;
580 static size_t fractional_time_len(const char *line, size_t len)
582 const char *p;
583 size_t n;
585 /* Expected format: 19:41:17.620000023 */
586 if (!len || !isdigit(line[len - 1]))
587 return 0;
588 p = line + len - 1;
590 /* Fractional seconds. */
591 while (p > line && isdigit(*p))
592 p--;
593 if (*p != '.')
594 return 0;
596 /* Hours, minutes, and whole seconds. */
597 n = short_time_len(line, p - line);
598 if (!n)
599 return 0;
601 return line + len - p + n;
604 static size_t trailing_spaces_len(const char *line, size_t len)
606 const char *p;
608 /* Expected format: ' ' x (1 or more) */
609 if (!len || line[len - 1] != ' ')
610 return 0;
612 p = line + len;
613 while (p != line) {
614 p--;
615 if (*p != ' ')
616 return line + len - (p + 1);
619 /* All spaces! */
620 return len;
623 static size_t diff_timestamp_len(const char *line, size_t len)
625 const char *end = line + len;
626 size_t n;
629 * Posix: 2010-07-05 19:41:17
630 * GNU: 2010-07-05 19:41:17.620000023 -0500
633 if (!isdigit(end[-1]))
634 return 0;
636 n = sane_tz_len(line, end - line);
637 if (!n)
638 n = tz_with_colon_len(line, end - line);
639 end -= n;
641 n = short_time_len(line, end - line);
642 if (!n)
643 n = fractional_time_len(line, end - line);
644 end -= n;
646 n = date_len(line, end - line);
647 if (!n) /* No date. Too bad. */
648 return 0;
649 end -= n;
651 if (end == line) /* No space before date. */
652 return 0;
653 if (end[-1] == '\t') { /* Success! */
654 end--;
655 return line + len - end;
657 if (end[-1] != ' ') /* No space before date. */
658 return 0;
660 /* Whitespace damage. */
661 end -= trailing_spaces_len(line, end - line);
662 return line + len - end;
665 static char *find_name_common(struct strbuf *root,
666 const char *line,
667 const char *def,
668 int p_value,
669 const char *end,
670 int terminate)
672 int len;
673 const char *start = NULL;
675 if (p_value == 0)
676 start = line;
677 while (line != end) {
678 char c = *line;
680 if (!end && isspace(c)) {
681 if (c == '\n')
682 break;
683 if (name_terminate(c, terminate))
684 break;
686 line++;
687 if (c == '/' && !--p_value)
688 start = line;
690 if (!start)
691 return squash_slash(xstrdup_or_null(def));
692 len = line - start;
693 if (!len)
694 return squash_slash(xstrdup_or_null(def));
697 * Generally we prefer the shorter name, especially
698 * if the other one is just a variation of that with
699 * something else tacked on to the end (ie "file.orig"
700 * or "file~").
702 if (def) {
703 int deflen = strlen(def);
704 if (deflen < len && !strncmp(start, def, deflen))
705 return squash_slash(xstrdup(def));
708 if (root->len) {
709 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
710 return squash_slash(ret);
713 return squash_slash(xmemdupz(start, len));
716 static char *find_name(struct strbuf *root,
717 const char *line,
718 char *def,
719 int p_value,
720 int terminate)
722 if (*line == '"') {
723 char *name = find_name_gnu(root, line, p_value);
724 if (name)
725 return name;
728 return find_name_common(root, line, def, p_value, NULL, terminate);
731 static char *find_name_traditional(struct strbuf *root,
732 const char *line,
733 char *def,
734 int p_value)
736 size_t len;
737 size_t date_len;
739 if (*line == '"') {
740 char *name = find_name_gnu(root, line, p_value);
741 if (name)
742 return name;
745 len = strchrnul(line, '\n') - line;
746 date_len = diff_timestamp_len(line, len);
747 if (!date_len)
748 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
749 len -= date_len;
751 return find_name_common(root, line, def, p_value, line + len, 0);
755 * Given the string after "--- " or "+++ ", guess the appropriate
756 * p_value for the given patch.
758 static int guess_p_value(struct apply_state *state, const char *nameline)
760 char *name, *cp;
761 int val = -1;
763 if (is_dev_null(nameline))
764 return -1;
765 name = find_name_traditional(&state->root, nameline, NULL, 0);
766 if (!name)
767 return -1;
768 cp = strchr(name, '/');
769 if (!cp)
770 val = 0;
771 else if (state->prefix) {
773 * Does it begin with "a/$our-prefix" and such? Then this is
774 * very likely to apply to our directory.
776 if (starts_with(name, state->prefix))
777 val = count_slashes(state->prefix);
778 else {
779 cp++;
780 if (starts_with(cp, state->prefix))
781 val = count_slashes(state->prefix) + 1;
784 free(name);
785 return val;
789 * Does the ---/+++ line have the POSIX timestamp after the last HT?
790 * GNU diff puts epoch there to signal a creation/deletion event. Is
791 * this such a timestamp?
793 static int has_epoch_timestamp(const char *nameline)
796 * We are only interested in epoch timestamp; any non-zero
797 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
798 * For the same reason, the date must be either 1969-12-31 or
799 * 1970-01-01, and the seconds part must be "00".
801 const char stamp_regexp[] =
802 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
804 "([-+][0-2][0-9]:?[0-5][0-9])\n";
805 const char *timestamp = NULL, *cp, *colon;
806 static regex_t *stamp;
807 regmatch_t m[10];
808 int zoneoffset, epoch_hour, hour, minute;
809 int status;
811 for (cp = nameline; *cp != '\n'; cp++) {
812 if (*cp == '\t')
813 timestamp = cp + 1;
815 if (!timestamp)
816 return 0;
819 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
820 * (west of GMT) or 1970-01-01 (east of GMT)
822 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
823 epoch_hour = 24;
824 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
825 epoch_hour = 0;
826 else
827 return 0;
829 if (!stamp) {
830 stamp = xmalloc(sizeof(*stamp));
831 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
832 warning(_("Cannot prepare timestamp regexp %s"),
833 stamp_regexp);
834 return 0;
838 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
839 if (status) {
840 if (status != REG_NOMATCH)
841 warning(_("regexec returned %d for input: %s"),
842 status, timestamp);
843 return 0;
846 hour = strtol(timestamp, NULL, 10);
847 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
849 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
850 if (*colon == ':')
851 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
852 else
853 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
854 if (timestamp[m[3].rm_so] == '-')
855 zoneoffset = -zoneoffset;
857 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
861 * Get the name etc info from the ---/+++ lines of a traditional patch header
863 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
864 * files, we can happily check the index for a match, but for creating a
865 * new file we should try to match whatever "patch" does. I have no idea.
867 static int parse_traditional_patch(struct apply_state *state,
868 const char *first,
869 const char *second,
870 struct patch *patch)
872 char *name;
874 first += 4; /* skip "--- " */
875 second += 4; /* skip "+++ " */
876 if (!state->p_value_known) {
877 int p, q;
878 p = guess_p_value(state, first);
879 q = guess_p_value(state, second);
880 if (p < 0) p = q;
881 if (0 <= p && p == q) {
882 state->p_value = p;
883 state->p_value_known = 1;
886 if (is_dev_null(first)) {
887 patch->is_new = 1;
888 patch->is_delete = 0;
889 name = find_name_traditional(&state->root, second, NULL, state->p_value);
890 patch->new_name = name;
891 } else if (is_dev_null(second)) {
892 patch->is_new = 0;
893 patch->is_delete = 1;
894 name = find_name_traditional(&state->root, first, NULL, state->p_value);
895 patch->old_name = name;
896 } else {
897 char *first_name;
898 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
899 name = find_name_traditional(&state->root, second, first_name, state->p_value);
900 free(first_name);
901 if (has_epoch_timestamp(first)) {
902 patch->is_new = 1;
903 patch->is_delete = 0;
904 patch->new_name = name;
905 } else if (has_epoch_timestamp(second)) {
906 patch->is_new = 0;
907 patch->is_delete = 1;
908 patch->old_name = name;
909 } else {
910 patch->old_name = name;
911 patch->new_name = xstrdup_or_null(name);
914 if (!name)
915 return error(_("unable to find filename in patch at line %d"), state->linenr);
917 return 0;
920 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
921 const char *line UNUSED,
922 struct patch *patch UNUSED)
924 return 1;
928 * We're anal about diff header consistency, to make
929 * sure that we don't end up having strange ambiguous
930 * patches floating around.
932 * As a result, gitdiff_{old|new}name() will check
933 * their names against any previous information, just
934 * to make sure..
936 #define DIFF_OLD_NAME 0
937 #define DIFF_NEW_NAME 1
939 static int gitdiff_verify_name(struct gitdiff_data *state,
940 const char *line,
941 int isnull,
942 char **name,
943 int side)
945 if (!*name && !isnull) {
946 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
947 return 0;
950 if (*name) {
951 char *another;
952 if (isnull)
953 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
954 *name, state->linenr);
955 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
956 if (!another || strcmp(another, *name)) {
957 free(another);
958 return error((side == DIFF_NEW_NAME) ?
959 _("git apply: bad git-diff - inconsistent new filename on line %d") :
960 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
962 free(another);
963 } else {
964 if (!is_dev_null(line))
965 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
968 return 0;
971 static int gitdiff_oldname(struct gitdiff_data *state,
972 const char *line,
973 struct patch *patch)
975 return gitdiff_verify_name(state, line,
976 patch->is_new, &patch->old_name,
977 DIFF_OLD_NAME);
980 static int gitdiff_newname(struct gitdiff_data *state,
981 const char *line,
982 struct patch *patch)
984 return gitdiff_verify_name(state, line,
985 patch->is_delete, &patch->new_name,
986 DIFF_NEW_NAME);
989 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
991 char *end;
992 *mode = strtoul(line, &end, 8);
993 if (end == line || !isspace(*end))
994 return error(_("invalid mode on line %d: %s"), linenr, line);
995 return 0;
998 static int gitdiff_oldmode(struct gitdiff_data *state,
999 const char *line,
1000 struct patch *patch)
1002 return parse_mode_line(line, state->linenr, &patch->old_mode);
1005 static int gitdiff_newmode(struct gitdiff_data *state,
1006 const char *line,
1007 struct patch *patch)
1009 return parse_mode_line(line, state->linenr, &patch->new_mode);
1012 static int gitdiff_delete(struct gitdiff_data *state,
1013 const char *line,
1014 struct patch *patch)
1016 patch->is_delete = 1;
1017 free(patch->old_name);
1018 patch->old_name = xstrdup_or_null(patch->def_name);
1019 return gitdiff_oldmode(state, line, patch);
1022 static int gitdiff_newfile(struct gitdiff_data *state,
1023 const char *line,
1024 struct patch *patch)
1026 patch->is_new = 1;
1027 free(patch->new_name);
1028 patch->new_name = xstrdup_or_null(patch->def_name);
1029 return gitdiff_newmode(state, line, patch);
1032 static int gitdiff_copysrc(struct gitdiff_data *state,
1033 const char *line,
1034 struct patch *patch)
1036 patch->is_copy = 1;
1037 free(patch->old_name);
1038 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1039 return 0;
1042 static int gitdiff_copydst(struct gitdiff_data *state,
1043 const char *line,
1044 struct patch *patch)
1046 patch->is_copy = 1;
1047 free(patch->new_name);
1048 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1049 return 0;
1052 static int gitdiff_renamesrc(struct gitdiff_data *state,
1053 const char *line,
1054 struct patch *patch)
1056 patch->is_rename = 1;
1057 free(patch->old_name);
1058 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1059 return 0;
1062 static int gitdiff_renamedst(struct gitdiff_data *state,
1063 const char *line,
1064 struct patch *patch)
1066 patch->is_rename = 1;
1067 free(patch->new_name);
1068 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1069 return 0;
1072 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1073 const char *line,
1074 struct patch *patch)
1076 unsigned long val = strtoul(line, NULL, 10);
1077 if (val <= 100)
1078 patch->score = val;
1079 return 0;
1082 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1083 const char *line,
1084 struct patch *patch)
1086 unsigned long val = strtoul(line, NULL, 10);
1087 if (val <= 100)
1088 patch->score = val;
1089 return 0;
1092 static int gitdiff_index(struct gitdiff_data *state,
1093 const char *line,
1094 struct patch *patch)
1097 * index line is N hexadecimal, "..", N hexadecimal,
1098 * and optional space with octal mode.
1100 const char *ptr, *eol;
1101 int len;
1102 const unsigned hexsz = the_hash_algo->hexsz;
1104 ptr = strchr(line, '.');
1105 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1106 return 0;
1107 len = ptr - line;
1108 memcpy(patch->old_oid_prefix, line, len);
1109 patch->old_oid_prefix[len] = 0;
1111 line = ptr + 2;
1112 ptr = strchr(line, ' ');
1113 eol = strchrnul(line, '\n');
1115 if (!ptr || eol < ptr)
1116 ptr = eol;
1117 len = ptr - line;
1119 if (hexsz < len)
1120 return 0;
1121 memcpy(patch->new_oid_prefix, line, len);
1122 patch->new_oid_prefix[len] = 0;
1123 if (*ptr == ' ')
1124 return gitdiff_oldmode(state, ptr + 1, patch);
1125 return 0;
1129 * This is normal for a diff that doesn't change anything: we'll fall through
1130 * into the next diff. Tell the parser to break out.
1132 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1133 const char *line UNUSED,
1134 struct patch *patch UNUSED)
1136 return 1;
1140 * Skip p_value leading components from "line"; as we do not accept
1141 * absolute paths, return NULL in that case.
1143 static const char *skip_tree_prefix(int p_value,
1144 const char *line,
1145 int llen)
1147 int nslash;
1148 int i;
1150 if (!p_value)
1151 return (llen && line[0] == '/') ? NULL : line;
1153 nslash = p_value;
1154 for (i = 0; i < llen; i++) {
1155 int ch = line[i];
1156 if (ch == '/' && --nslash <= 0)
1157 return (i == 0) ? NULL : &line[i + 1];
1159 return NULL;
1163 * This is to extract the same name that appears on "diff --git"
1164 * line. We do not find and return anything if it is a rename
1165 * patch, and it is OK because we will find the name elsewhere.
1166 * We need to reliably find name only when it is mode-change only,
1167 * creation or deletion of an empty file. In any of these cases,
1168 * both sides are the same name under a/ and b/ respectively.
1170 static char *git_header_name(int p_value,
1171 const char *line,
1172 int llen)
1174 const char *name;
1175 const char *second = NULL;
1176 size_t len, line_len;
1178 line += strlen("diff --git ");
1179 llen -= strlen("diff --git ");
1181 if (*line == '"') {
1182 const char *cp;
1183 struct strbuf first = STRBUF_INIT;
1184 struct strbuf sp = STRBUF_INIT;
1186 if (unquote_c_style(&first, line, &second))
1187 goto free_and_fail1;
1189 /* strip the a/b prefix including trailing slash */
1190 cp = skip_tree_prefix(p_value, first.buf, first.len);
1191 if (!cp)
1192 goto free_and_fail1;
1193 strbuf_remove(&first, 0, cp - first.buf);
1196 * second points at one past closing dq of name.
1197 * find the second name.
1199 while ((second < line + llen) && isspace(*second))
1200 second++;
1202 if (line + llen <= second)
1203 goto free_and_fail1;
1204 if (*second == '"') {
1205 if (unquote_c_style(&sp, second, NULL))
1206 goto free_and_fail1;
1207 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1208 if (!cp)
1209 goto free_and_fail1;
1210 /* They must match, otherwise ignore */
1211 if (strcmp(cp, first.buf))
1212 goto free_and_fail1;
1213 strbuf_release(&sp);
1214 return strbuf_detach(&first, NULL);
1217 /* unquoted second */
1218 cp = skip_tree_prefix(p_value, second, line + llen - second);
1219 if (!cp)
1220 goto free_and_fail1;
1221 if (line + llen - cp != first.len ||
1222 memcmp(first.buf, cp, first.len))
1223 goto free_and_fail1;
1224 return strbuf_detach(&first, NULL);
1226 free_and_fail1:
1227 strbuf_release(&first);
1228 strbuf_release(&sp);
1229 return NULL;
1232 /* unquoted first name */
1233 name = skip_tree_prefix(p_value, line, llen);
1234 if (!name)
1235 return NULL;
1238 * since the first name is unquoted, a dq if exists must be
1239 * the beginning of the second name.
1241 for (second = name; second < line + llen; second++) {
1242 if (*second == '"') {
1243 struct strbuf sp = STRBUF_INIT;
1244 const char *np;
1246 if (unquote_c_style(&sp, second, NULL))
1247 goto free_and_fail2;
1249 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1250 if (!np)
1251 goto free_and_fail2;
1253 len = sp.buf + sp.len - np;
1254 if (len < second - name &&
1255 !strncmp(np, name, len) &&
1256 isspace(name[len])) {
1257 /* Good */
1258 strbuf_remove(&sp, 0, np - sp.buf);
1259 return strbuf_detach(&sp, NULL);
1262 free_and_fail2:
1263 strbuf_release(&sp);
1264 return NULL;
1269 * Accept a name only if it shows up twice, exactly the same
1270 * form.
1272 second = strchr(name, '\n');
1273 if (!second)
1274 return NULL;
1275 line_len = second - name;
1276 for (len = 0 ; ; len++) {
1277 switch (name[len]) {
1278 default:
1279 continue;
1280 case '\n':
1281 return NULL;
1282 case '\t': case ' ':
1284 * Is this the separator between the preimage
1285 * and the postimage pathname? Again, we are
1286 * only interested in the case where there is
1287 * no rename, as this is only to set def_name
1288 * and a rename patch has the names elsewhere
1289 * in an unambiguous form.
1291 if (!name[len + 1])
1292 return NULL; /* no postimage name */
1293 second = skip_tree_prefix(p_value, name + len + 1,
1294 line_len - (len + 1));
1296 * If we are at the SP at the end of a directory,
1297 * skip_tree_prefix() may return NULL as that makes
1298 * it appears as if we have an absolute path.
1299 * Keep going to find another SP.
1301 if (!second)
1302 continue;
1305 * Does len bytes starting at "name" and "second"
1306 * (that are separated by one HT or SP we just
1307 * found) exactly match?
1309 if (second[len] == '\n' && !strncmp(name, second, len))
1310 return xmemdupz(name, len);
1315 static int check_header_line(int linenr, struct patch *patch)
1317 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1318 (patch->is_rename == 1) + (patch->is_copy == 1);
1319 if (extensions > 1)
1320 return error(_("inconsistent header lines %d and %d"),
1321 patch->extension_linenr, linenr);
1322 if (extensions && !patch->extension_linenr)
1323 patch->extension_linenr = linenr;
1324 return 0;
1327 int parse_git_diff_header(struct strbuf *root,
1328 int *linenr,
1329 int p_value,
1330 const char *line,
1331 int len,
1332 unsigned int size,
1333 struct patch *patch)
1335 unsigned long offset;
1336 struct gitdiff_data parse_hdr_state;
1338 /* A git diff has explicit new/delete information, so we don't guess */
1339 patch->is_new = 0;
1340 patch->is_delete = 0;
1343 * Some things may not have the old name in the
1344 * rest of the headers anywhere (pure mode changes,
1345 * or removing or adding empty files), so we get
1346 * the default name from the header.
1348 patch->def_name = git_header_name(p_value, line, len);
1349 if (patch->def_name && root->len) {
1350 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1351 free(patch->def_name);
1352 patch->def_name = s;
1355 line += len;
1356 size -= len;
1357 (*linenr)++;
1358 parse_hdr_state.root = root;
1359 parse_hdr_state.linenr = *linenr;
1360 parse_hdr_state.p_value = p_value;
1362 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1363 static const struct opentry {
1364 const char *str;
1365 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1366 } optable[] = {
1367 { "@@ -", gitdiff_hdrend },
1368 { "--- ", gitdiff_oldname },
1369 { "+++ ", gitdiff_newname },
1370 { "old mode ", gitdiff_oldmode },
1371 { "new mode ", gitdiff_newmode },
1372 { "deleted file mode ", gitdiff_delete },
1373 { "new file mode ", gitdiff_newfile },
1374 { "copy from ", gitdiff_copysrc },
1375 { "copy to ", gitdiff_copydst },
1376 { "rename old ", gitdiff_renamesrc },
1377 { "rename new ", gitdiff_renamedst },
1378 { "rename from ", gitdiff_renamesrc },
1379 { "rename to ", gitdiff_renamedst },
1380 { "similarity index ", gitdiff_similarity },
1381 { "dissimilarity index ", gitdiff_dissimilarity },
1382 { "index ", gitdiff_index },
1383 { "", gitdiff_unrecognized },
1385 int i;
1387 len = linelen(line, size);
1388 if (!len || line[len-1] != '\n')
1389 break;
1390 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1391 const struct opentry *p = optable + i;
1392 int oplen = strlen(p->str);
1393 int res;
1394 if (len < oplen || memcmp(p->str, line, oplen))
1395 continue;
1396 res = p->fn(&parse_hdr_state, line + oplen, patch);
1397 if (res < 0)
1398 return -1;
1399 if (check_header_line(*linenr, patch))
1400 return -1;
1401 if (res > 0)
1402 goto done;
1403 break;
1407 done:
1408 if (!patch->old_name && !patch->new_name) {
1409 if (!patch->def_name) {
1410 error(Q_("git diff header lacks filename information when removing "
1411 "%d leading pathname component (line %d)",
1412 "git diff header lacks filename information when removing "
1413 "%d leading pathname components (line %d)",
1414 parse_hdr_state.p_value),
1415 parse_hdr_state.p_value, *linenr);
1416 return -128;
1418 patch->old_name = xstrdup(patch->def_name);
1419 patch->new_name = xstrdup(patch->def_name);
1421 if ((!patch->new_name && !patch->is_delete) ||
1422 (!patch->old_name && !patch->is_new)) {
1423 error(_("git diff header lacks filename information "
1424 "(line %d)"), *linenr);
1425 return -128;
1427 patch->is_toplevel_relative = 1;
1428 return offset;
1431 static int parse_num(const char *line, unsigned long *p)
1433 char *ptr;
1435 if (!isdigit(*line))
1436 return 0;
1437 *p = strtoul(line, &ptr, 10);
1438 return ptr - line;
1441 static int parse_range(const char *line, int len, int offset, const char *expect,
1442 unsigned long *p1, unsigned long *p2)
1444 int digits, ex;
1446 if (offset < 0 || offset >= len)
1447 return -1;
1448 line += offset;
1449 len -= offset;
1451 digits = parse_num(line, p1);
1452 if (!digits)
1453 return -1;
1455 offset += digits;
1456 line += digits;
1457 len -= digits;
1459 *p2 = 1;
1460 if (*line == ',') {
1461 digits = parse_num(line+1, p2);
1462 if (!digits)
1463 return -1;
1465 offset += digits+1;
1466 line += digits+1;
1467 len -= digits+1;
1470 ex = strlen(expect);
1471 if (ex > len)
1472 return -1;
1473 if (memcmp(line, expect, ex))
1474 return -1;
1476 return offset + ex;
1479 static void recount_diff(const char *line, int size, struct fragment *fragment)
1481 int oldlines = 0, newlines = 0, ret = 0;
1483 if (size < 1) {
1484 warning("recount: ignore empty hunk");
1485 return;
1488 for (;;) {
1489 int len = linelen(line, size);
1490 size -= len;
1491 line += len;
1493 if (size < 1)
1494 break;
1496 switch (*line) {
1497 case ' ': case '\n':
1498 newlines++;
1499 /* fall through */
1500 case '-':
1501 oldlines++;
1502 continue;
1503 case '+':
1504 newlines++;
1505 continue;
1506 case '\\':
1507 continue;
1508 case '@':
1509 ret = size < 3 || !starts_with(line, "@@ ");
1510 break;
1511 case 'd':
1512 ret = size < 5 || !starts_with(line, "diff ");
1513 break;
1514 default:
1515 ret = -1;
1516 break;
1518 if (ret) {
1519 warning(_("recount: unexpected line: %.*s"),
1520 (int)linelen(line, size), line);
1521 return;
1523 break;
1525 fragment->oldlines = oldlines;
1526 fragment->newlines = newlines;
1530 * Parse a unified diff fragment header of the
1531 * form "@@ -a,b +c,d @@"
1533 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1535 int offset;
1537 if (!len || line[len-1] != '\n')
1538 return -1;
1540 /* Figure out the number of lines in a fragment */
1541 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1542 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1544 return offset;
1548 * Find file diff header
1550 * Returns:
1551 * -1 if no header was found
1552 * -128 in case of error
1553 * the size of the header in bytes (called "offset") otherwise
1555 static int find_header(struct apply_state *state,
1556 const char *line,
1557 unsigned long size,
1558 int *hdrsize,
1559 struct patch *patch)
1561 unsigned long offset, len;
1563 patch->is_toplevel_relative = 0;
1564 patch->is_rename = patch->is_copy = 0;
1565 patch->is_new = patch->is_delete = -1;
1566 patch->old_mode = patch->new_mode = 0;
1567 patch->old_name = patch->new_name = NULL;
1568 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1569 unsigned long nextlen;
1571 len = linelen(line, size);
1572 if (!len)
1573 break;
1575 /* Testing this early allows us to take a few shortcuts.. */
1576 if (len < 6)
1577 continue;
1580 * Make sure we don't find any unconnected patch fragments.
1581 * That's a sign that we didn't find a header, and that a
1582 * patch has become corrupted/broken up.
1584 if (!memcmp("@@ -", line, 4)) {
1585 struct fragment dummy;
1586 if (parse_fragment_header(line, len, &dummy) < 0)
1587 continue;
1588 error(_("patch fragment without header at line %d: %.*s"),
1589 state->linenr, (int)len-1, line);
1590 return -128;
1593 if (size < len + 6)
1594 break;
1597 * Git patch? It might not have a real patch, just a rename
1598 * or mode change, so we handle that specially
1600 if (!memcmp("diff --git ", line, 11)) {
1601 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1602 state->p_value, line, len,
1603 size, patch);
1604 if (git_hdr_len < 0)
1605 return -128;
1606 if (git_hdr_len <= len)
1607 continue;
1608 *hdrsize = git_hdr_len;
1609 return offset;
1612 /* --- followed by +++ ? */
1613 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1614 continue;
1617 * We only accept unified patches, so we want it to
1618 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1619 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1621 nextlen = linelen(line + len, size - len);
1622 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1623 continue;
1625 /* Ok, we'll consider it a patch */
1626 if (parse_traditional_patch(state, line, line+len, patch))
1627 return -128;
1628 *hdrsize = len + nextlen;
1629 state->linenr += 2;
1630 return offset;
1632 return -1;
1635 static void record_ws_error(struct apply_state *state,
1636 unsigned result,
1637 const char *line,
1638 int len,
1639 int linenr)
1641 char *err;
1643 if (!result)
1644 return;
1646 state->whitespace_error++;
1647 if (state->squelch_whitespace_errors &&
1648 state->squelch_whitespace_errors < state->whitespace_error)
1649 return;
1651 err = whitespace_error_string(result);
1652 if (state->apply_verbosity > verbosity_silent)
1653 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1654 state->patch_input_file, linenr, err, len, line);
1655 free(err);
1658 static void check_whitespace(struct apply_state *state,
1659 const char *line,
1660 int len,
1661 unsigned ws_rule)
1663 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1665 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1669 * Check if the patch has context lines with CRLF or
1670 * the patch wants to remove lines with CRLF.
1672 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1674 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1675 patch->ws_rule |= WS_CR_AT_EOL;
1676 patch->crlf_in_old = 1;
1682 * Parse a unified diff. Note that this really needs to parse each
1683 * fragment separately, since the only way to know the difference
1684 * between a "---" that is part of a patch, and a "---" that starts
1685 * the next patch is to look at the line counts..
1687 static int parse_fragment(struct apply_state *state,
1688 const char *line,
1689 unsigned long size,
1690 struct patch *patch,
1691 struct fragment *fragment)
1693 int added, deleted;
1694 int len = linelen(line, size), offset;
1695 unsigned long oldlines, newlines;
1696 unsigned long leading, trailing;
1698 offset = parse_fragment_header(line, len, fragment);
1699 if (offset < 0)
1700 return -1;
1701 if (offset > 0 && patch->recount)
1702 recount_diff(line + offset, size - offset, fragment);
1703 oldlines = fragment->oldlines;
1704 newlines = fragment->newlines;
1705 leading = 0;
1706 trailing = 0;
1708 /* Parse the thing.. */
1709 line += len;
1710 size -= len;
1711 state->linenr++;
1712 added = deleted = 0;
1713 for (offset = len;
1714 0 < size;
1715 offset += len, size -= len, line += len, state->linenr++) {
1716 if (!oldlines && !newlines)
1717 break;
1718 len = linelen(line, size);
1719 if (!len || line[len-1] != '\n')
1720 return -1;
1721 switch (*line) {
1722 default:
1723 return -1;
1724 case '\n': /* newer GNU diff, an empty context line */
1725 case ' ':
1726 oldlines--;
1727 newlines--;
1728 if (!deleted && !added)
1729 leading++;
1730 trailing++;
1731 check_old_for_crlf(patch, line, len);
1732 if (!state->apply_in_reverse &&
1733 state->ws_error_action == correct_ws_error)
1734 check_whitespace(state, line, len, patch->ws_rule);
1735 break;
1736 case '-':
1737 if (!state->apply_in_reverse)
1738 check_old_for_crlf(patch, line, len);
1739 if (state->apply_in_reverse &&
1740 state->ws_error_action != nowarn_ws_error)
1741 check_whitespace(state, line, len, patch->ws_rule);
1742 deleted++;
1743 oldlines--;
1744 trailing = 0;
1745 break;
1746 case '+':
1747 if (state->apply_in_reverse)
1748 check_old_for_crlf(patch, line, len);
1749 if (!state->apply_in_reverse &&
1750 state->ws_error_action != nowarn_ws_error)
1751 check_whitespace(state, line, len, patch->ws_rule);
1752 added++;
1753 newlines--;
1754 trailing = 0;
1755 break;
1758 * We allow "\ No newline at end of file". Depending
1759 * on locale settings when the patch was produced we
1760 * don't know what this line looks like. The only
1761 * thing we do know is that it begins with "\ ".
1762 * Checking for 12 is just for sanity check -- any
1763 * l10n of "\ No newline..." is at least that long.
1765 case '\\':
1766 if (len < 12 || memcmp(line, "\\ ", 2))
1767 return -1;
1768 break;
1771 if (oldlines || newlines)
1772 return -1;
1773 if (!patch->recount && !deleted && !added)
1774 return -1;
1776 fragment->leading = leading;
1777 fragment->trailing = trailing;
1780 * If a fragment ends with an incomplete line, we failed to include
1781 * it in the above loop because we hit oldlines == newlines == 0
1782 * before seeing it.
1784 if (12 < size && !memcmp(line, "\\ ", 2))
1785 offset += linelen(line, size);
1787 patch->lines_added += added;
1788 patch->lines_deleted += deleted;
1790 if (0 < patch->is_new && oldlines)
1791 return error(_("new file depends on old contents"));
1792 if (0 < patch->is_delete && newlines)
1793 return error(_("deleted file still has contents"));
1794 return offset;
1798 * We have seen "diff --git a/... b/..." header (or a traditional patch
1799 * header). Read hunks that belong to this patch into fragments and hang
1800 * them to the given patch structure.
1802 * The (fragment->patch, fragment->size) pair points into the memory given
1803 * by the caller, not a copy, when we return.
1805 * Returns:
1806 * -1 in case of error,
1807 * the number of bytes in the patch otherwise.
1809 static int parse_single_patch(struct apply_state *state,
1810 const char *line,
1811 unsigned long size,
1812 struct patch *patch)
1814 unsigned long offset = 0;
1815 unsigned long oldlines = 0, newlines = 0, context = 0;
1816 struct fragment **fragp = &patch->fragments;
1818 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1819 struct fragment *fragment;
1820 int len;
1822 CALLOC_ARRAY(fragment, 1);
1823 fragment->linenr = state->linenr;
1824 len = parse_fragment(state, line, size, patch, fragment);
1825 if (len <= 0) {
1826 free(fragment);
1827 return error(_("corrupt patch at line %d"), state->linenr);
1829 fragment->patch = line;
1830 fragment->size = len;
1831 oldlines += fragment->oldlines;
1832 newlines += fragment->newlines;
1833 context += fragment->leading + fragment->trailing;
1835 *fragp = fragment;
1836 fragp = &fragment->next;
1838 offset += len;
1839 line += len;
1840 size -= len;
1844 * If something was removed (i.e. we have old-lines) it cannot
1845 * be creation, and if something was added it cannot be
1846 * deletion. However, the reverse is not true; --unified=0
1847 * patches that only add are not necessarily creation even
1848 * though they do not have any old lines, and ones that only
1849 * delete are not necessarily deletion.
1851 * Unfortunately, a real creation/deletion patch do _not_ have
1852 * any context line by definition, so we cannot safely tell it
1853 * apart with --unified=0 insanity. At least if the patch has
1854 * more than one hunk it is not creation or deletion.
1856 if (patch->is_new < 0 &&
1857 (oldlines || (patch->fragments && patch->fragments->next)))
1858 patch->is_new = 0;
1859 if (patch->is_delete < 0 &&
1860 (newlines || (patch->fragments && patch->fragments->next)))
1861 patch->is_delete = 0;
1863 if (0 < patch->is_new && oldlines)
1864 return error(_("new file %s depends on old contents"), patch->new_name);
1865 if (0 < patch->is_delete && newlines)
1866 return error(_("deleted file %s still has contents"), patch->old_name);
1867 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1868 fprintf_ln(stderr,
1869 _("** warning: "
1870 "file %s becomes empty but is not deleted"),
1871 patch->new_name);
1873 return offset;
1876 static inline int metadata_changes(struct patch *patch)
1878 return patch->is_rename > 0 ||
1879 patch->is_copy > 0 ||
1880 patch->is_new > 0 ||
1881 patch->is_delete ||
1882 (patch->old_mode && patch->new_mode &&
1883 patch->old_mode != patch->new_mode);
1886 static char *inflate_it(const void *data, unsigned long size,
1887 unsigned long inflated_size)
1889 git_zstream stream;
1890 void *out;
1891 int st;
1893 memset(&stream, 0, sizeof(stream));
1895 stream.next_in = (unsigned char *)data;
1896 stream.avail_in = size;
1897 stream.next_out = out = xmalloc(inflated_size);
1898 stream.avail_out = inflated_size;
1899 git_inflate_init(&stream);
1900 st = git_inflate(&stream, Z_FINISH);
1901 git_inflate_end(&stream);
1902 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1903 free(out);
1904 return NULL;
1906 return out;
1910 * Read a binary hunk and return a new fragment; fragment->patch
1911 * points at an allocated memory that the caller must free, so
1912 * it is marked as "->free_patch = 1".
1914 static struct fragment *parse_binary_hunk(struct apply_state *state,
1915 char **buf_p,
1916 unsigned long *sz_p,
1917 int *status_p,
1918 int *used_p)
1921 * Expect a line that begins with binary patch method ("literal"
1922 * or "delta"), followed by the length of data before deflating.
1923 * a sequence of 'length-byte' followed by base-85 encoded data
1924 * should follow, terminated by a newline.
1926 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1927 * and we would limit the patch line to 66 characters,
1928 * so one line can fit up to 13 groups that would decode
1929 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1930 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1932 int llen, used;
1933 unsigned long size = *sz_p;
1934 char *buffer = *buf_p;
1935 int patch_method;
1936 unsigned long origlen;
1937 char *data = NULL;
1938 int hunk_size = 0;
1939 struct fragment *frag;
1941 llen = linelen(buffer, size);
1942 used = llen;
1944 *status_p = 0;
1946 if (starts_with(buffer, "delta ")) {
1947 patch_method = BINARY_DELTA_DEFLATED;
1948 origlen = strtoul(buffer + 6, NULL, 10);
1950 else if (starts_with(buffer, "literal ")) {
1951 patch_method = BINARY_LITERAL_DEFLATED;
1952 origlen = strtoul(buffer + 8, NULL, 10);
1954 else
1955 return NULL;
1957 state->linenr++;
1958 buffer += llen;
1959 size -= llen;
1960 while (1) {
1961 int byte_length, max_byte_length, newsize;
1962 llen = linelen(buffer, size);
1963 used += llen;
1964 state->linenr++;
1965 if (llen == 1) {
1966 /* consume the blank line */
1967 buffer++;
1968 size--;
1969 break;
1972 * Minimum line is "A00000\n" which is 7-byte long,
1973 * and the line length must be multiple of 5 plus 2.
1975 if ((llen < 7) || (llen-2) % 5)
1976 goto corrupt;
1977 max_byte_length = (llen - 2) / 5 * 4;
1978 byte_length = *buffer;
1979 if ('A' <= byte_length && byte_length <= 'Z')
1980 byte_length = byte_length - 'A' + 1;
1981 else if ('a' <= byte_length && byte_length <= 'z')
1982 byte_length = byte_length - 'a' + 27;
1983 else
1984 goto corrupt;
1985 /* if the input length was not multiple of 4, we would
1986 * have filler at the end but the filler should never
1987 * exceed 3 bytes
1989 if (max_byte_length < byte_length ||
1990 byte_length <= max_byte_length - 4)
1991 goto corrupt;
1992 newsize = hunk_size + byte_length;
1993 data = xrealloc(data, newsize);
1994 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1995 goto corrupt;
1996 hunk_size = newsize;
1997 buffer += llen;
1998 size -= llen;
2001 CALLOC_ARRAY(frag, 1);
2002 frag->patch = inflate_it(data, hunk_size, origlen);
2003 frag->free_patch = 1;
2004 if (!frag->patch)
2005 goto corrupt;
2006 free(data);
2007 frag->size = origlen;
2008 *buf_p = buffer;
2009 *sz_p = size;
2010 *used_p = used;
2011 frag->binary_patch_method = patch_method;
2012 return frag;
2014 corrupt:
2015 free(data);
2016 *status_p = -1;
2017 error(_("corrupt binary patch at line %d: %.*s"),
2018 state->linenr-1, llen-1, buffer);
2019 return NULL;
2023 * Returns:
2024 * -1 in case of error,
2025 * the length of the parsed binary patch otherwise
2027 static int parse_binary(struct apply_state *state,
2028 char *buffer,
2029 unsigned long size,
2030 struct patch *patch)
2033 * We have read "GIT binary patch\n"; what follows is a line
2034 * that says the patch method (currently, either "literal" or
2035 * "delta") and the length of data before deflating; a
2036 * sequence of 'length-byte' followed by base-85 encoded data
2037 * follows.
2039 * When a binary patch is reversible, there is another binary
2040 * hunk in the same format, starting with patch method (either
2041 * "literal" or "delta") with the length of data, and a sequence
2042 * of length-byte + base-85 encoded data, terminated with another
2043 * empty line. This data, when applied to the postimage, produces
2044 * the preimage.
2046 struct fragment *forward;
2047 struct fragment *reverse;
2048 int status;
2049 int used, used_1;
2051 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2052 if (!forward && !status)
2053 /* there has to be one hunk (forward hunk) */
2054 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2055 if (status)
2056 /* otherwise we already gave an error message */
2057 return status;
2059 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2060 if (reverse)
2061 used += used_1;
2062 else if (status) {
2064 * Not having reverse hunk is not an error, but having
2065 * a corrupt reverse hunk is.
2067 free((void*) forward->patch);
2068 free(forward);
2069 return status;
2071 forward->next = reverse;
2072 patch->fragments = forward;
2073 patch->is_binary = 1;
2074 return used;
2077 static void prefix_one(struct apply_state *state, char **name)
2079 char *old_name = *name;
2080 if (!old_name)
2081 return;
2082 *name = prefix_filename(state->prefix, *name);
2083 free(old_name);
2086 static void prefix_patch(struct apply_state *state, struct patch *p)
2088 if (!state->prefix || p->is_toplevel_relative)
2089 return;
2090 prefix_one(state, &p->new_name);
2091 prefix_one(state, &p->old_name);
2095 * include/exclude
2098 static void add_name_limit(struct apply_state *state,
2099 const char *name,
2100 int exclude)
2102 struct string_list_item *it;
2104 it = string_list_append(&state->limit_by_name, name);
2105 it->util = exclude ? NULL : (void *) 1;
2108 static int use_patch(struct apply_state *state, struct patch *p)
2110 const char *pathname = p->new_name ? p->new_name : p->old_name;
2111 int i;
2113 /* Paths outside are not touched regardless of "--include" */
2114 if (state->prefix && *state->prefix) {
2115 const char *rest;
2116 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2117 return 0;
2120 /* See if it matches any of exclude/include rule */
2121 for (i = 0; i < state->limit_by_name.nr; i++) {
2122 struct string_list_item *it = &state->limit_by_name.items[i];
2123 if (!wildmatch(it->string, pathname, 0))
2124 return (it->util != NULL);
2128 * If we had any include, a path that does not match any rule is
2129 * not used. Otherwise, we saw bunch of exclude rules (or none)
2130 * and such a path is used.
2132 return !state->has_include;
2136 * Read the patch text in "buffer" that extends for "size" bytes; stop
2137 * reading after seeing a single patch (i.e. changes to a single file).
2138 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2140 * Returns:
2141 * -1 if no header was found or parse_binary() failed,
2142 * -128 on another error,
2143 * the number of bytes consumed otherwise,
2144 * so that the caller can call us again for the next patch.
2146 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2148 int hdrsize, patchsize;
2149 int offset = find_header(state, buffer, size, &hdrsize, patch);
2151 if (offset < 0)
2152 return offset;
2154 prefix_patch(state, patch);
2156 if (!use_patch(state, patch))
2157 patch->ws_rule = 0;
2158 else if (patch->new_name)
2159 patch->ws_rule = whitespace_rule(state->repo->index,
2160 patch->new_name);
2161 else
2162 patch->ws_rule = whitespace_rule(state->repo->index,
2163 patch->old_name);
2165 patchsize = parse_single_patch(state,
2166 buffer + offset + hdrsize,
2167 size - offset - hdrsize,
2168 patch);
2170 if (patchsize < 0)
2171 return -128;
2173 if (!patchsize) {
2174 static const char git_binary[] = "GIT binary patch\n";
2175 int hd = hdrsize + offset;
2176 unsigned long llen = linelen(buffer + hd, size - hd);
2178 if (llen == sizeof(git_binary) - 1 &&
2179 !memcmp(git_binary, buffer + hd, llen)) {
2180 int used;
2181 state->linenr++;
2182 used = parse_binary(state, buffer + hd + llen,
2183 size - hd - llen, patch);
2184 if (used < 0)
2185 return -1;
2186 if (used)
2187 patchsize = used + llen;
2188 else
2189 patchsize = 0;
2191 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2192 static const char *binhdr[] = {
2193 "Binary files ",
2194 "Files ",
2195 NULL,
2197 int i;
2198 for (i = 0; binhdr[i]; i++) {
2199 int len = strlen(binhdr[i]);
2200 if (len < size - hd &&
2201 !memcmp(binhdr[i], buffer + hd, len)) {
2202 state->linenr++;
2203 patch->is_binary = 1;
2204 patchsize = llen;
2205 break;
2210 /* Empty patch cannot be applied if it is a text patch
2211 * without metadata change. A binary patch appears
2212 * empty to us here.
2214 if ((state->apply || state->check) &&
2215 (!patch->is_binary && !metadata_changes(patch))) {
2216 error(_("patch with only garbage at line %d"), state->linenr);
2217 return -128;
2221 return offset + hdrsize + patchsize;
2224 static void reverse_patches(struct patch *p)
2226 for (; p; p = p->next) {
2227 struct fragment *frag = p->fragments;
2229 SWAP(p->new_name, p->old_name);
2230 if (p->new_mode)
2231 SWAP(p->new_mode, p->old_mode);
2232 SWAP(p->is_new, p->is_delete);
2233 SWAP(p->lines_added, p->lines_deleted);
2234 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2236 for (; frag; frag = frag->next) {
2237 SWAP(frag->newpos, frag->oldpos);
2238 SWAP(frag->newlines, frag->oldlines);
2243 static const char pluses[] =
2244 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2245 static const char minuses[]=
2246 "----------------------------------------------------------------------";
2248 static void show_stats(struct apply_state *state, struct patch *patch)
2250 struct strbuf qname = STRBUF_INIT;
2251 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2252 int max, add, del;
2254 quote_c_style(cp, &qname, NULL, 0);
2257 * "scale" the filename
2259 max = state->max_len;
2260 if (max > 50)
2261 max = 50;
2263 if (qname.len > max) {
2264 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2265 if (!cp)
2266 cp = qname.buf + qname.len + 3 - max;
2267 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2270 if (patch->is_binary) {
2271 printf(" %-*s | Bin\n", max, qname.buf);
2272 strbuf_release(&qname);
2273 return;
2276 printf(" %-*s |", max, qname.buf);
2277 strbuf_release(&qname);
2280 * scale the add/delete
2282 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2283 add = patch->lines_added;
2284 del = patch->lines_deleted;
2286 if (state->max_change > 0) {
2287 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2288 add = (add * max + state->max_change / 2) / state->max_change;
2289 del = total - add;
2291 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2292 add, pluses, del, minuses);
2295 static int read_old_data(struct stat *st, struct patch *patch,
2296 const char *path, struct strbuf *buf)
2298 int conv_flags = patch->crlf_in_old ?
2299 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2300 switch (st->st_mode & S_IFMT) {
2301 case S_IFLNK:
2302 if (strbuf_readlink(buf, path, st->st_size) < 0)
2303 return error(_("unable to read symlink %s"), path);
2304 return 0;
2305 case S_IFREG:
2306 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2307 return error(_("unable to open or read %s"), path);
2309 * "git apply" without "--index/--cached" should never look
2310 * at the index; the target file may not have been added to
2311 * the index yet, and we may not even be in any Git repository.
2312 * Pass NULL to convert_to_git() to stress this; the function
2313 * should never look at the index when explicit crlf option
2314 * is given.
2316 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2317 return 0;
2318 default:
2319 return -1;
2324 * Update the preimage, and the common lines in postimage,
2325 * from buffer buf of length len. If postlen is 0 the postimage
2326 * is updated in place, otherwise it's updated on a new buffer
2327 * of length postlen
2330 static void update_pre_post_images(struct image *preimage,
2331 struct image *postimage,
2332 char *buf,
2333 size_t len, size_t postlen)
2335 int i, ctx, reduced;
2336 char *new_buf, *old_buf, *fixed;
2337 struct image fixed_preimage;
2340 * Update the preimage with whitespace fixes. Note that we
2341 * are not losing preimage->buf -- apply_one_fragment() will
2342 * free "oldlines".
2344 prepare_image(&fixed_preimage, buf, len, 1);
2345 assert(postlen
2346 ? fixed_preimage.nr == preimage->nr
2347 : fixed_preimage.nr <= preimage->nr);
2348 for (i = 0; i < fixed_preimage.nr; i++)
2349 fixed_preimage.line[i].flag = preimage->line[i].flag;
2350 free(preimage->line_allocated);
2351 *preimage = fixed_preimage;
2354 * Adjust the common context lines in postimage. This can be
2355 * done in-place when we are shrinking it with whitespace
2356 * fixing, but needs a new buffer when ignoring whitespace or
2357 * expanding leading tabs to spaces.
2359 * We trust the caller to tell us if the update can be done
2360 * in place (postlen==0) or not.
2362 old_buf = postimage->buf;
2363 if (postlen)
2364 new_buf = postimage->buf = xmalloc(postlen);
2365 else
2366 new_buf = old_buf;
2367 fixed = preimage->buf;
2369 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2370 size_t l_len = postimage->line[i].len;
2371 if (!(postimage->line[i].flag & LINE_COMMON)) {
2372 /* an added line -- no counterparts in preimage */
2373 memmove(new_buf, old_buf, l_len);
2374 old_buf += l_len;
2375 new_buf += l_len;
2376 continue;
2379 /* a common context -- skip it in the original postimage */
2380 old_buf += l_len;
2382 /* and find the corresponding one in the fixed preimage */
2383 while (ctx < preimage->nr &&
2384 !(preimage->line[ctx].flag & LINE_COMMON)) {
2385 fixed += preimage->line[ctx].len;
2386 ctx++;
2390 * preimage is expected to run out, if the caller
2391 * fixed addition of trailing blank lines.
2393 if (preimage->nr <= ctx) {
2394 reduced++;
2395 continue;
2398 /* and copy it in, while fixing the line length */
2399 l_len = preimage->line[ctx].len;
2400 memcpy(new_buf, fixed, l_len);
2401 new_buf += l_len;
2402 fixed += l_len;
2403 postimage->line[i].len = l_len;
2404 ctx++;
2407 if (postlen
2408 ? postlen < new_buf - postimage->buf
2409 : postimage->len < new_buf - postimage->buf)
2410 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2411 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2413 /* Fix the length of the whole thing */
2414 postimage->len = new_buf - postimage->buf;
2415 postimage->nr -= reduced;
2418 static int line_by_line_fuzzy_match(struct image *img,
2419 struct image *preimage,
2420 struct image *postimage,
2421 unsigned long current,
2422 int current_lno,
2423 int preimage_limit)
2425 int i;
2426 size_t imgoff = 0;
2427 size_t preoff = 0;
2428 size_t postlen = postimage->len;
2429 size_t extra_chars;
2430 char *buf;
2431 char *preimage_eof;
2432 char *preimage_end;
2433 struct strbuf fixed;
2434 char *fixed_buf;
2435 size_t fixed_len;
2437 for (i = 0; i < preimage_limit; i++) {
2438 size_t prelen = preimage->line[i].len;
2439 size_t imglen = img->line[current_lno+i].len;
2441 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2442 preimage->buf + preoff, prelen))
2443 return 0;
2444 if (preimage->line[i].flag & LINE_COMMON)
2445 postlen += imglen - prelen;
2446 imgoff += imglen;
2447 preoff += prelen;
2451 * Ok, the preimage matches with whitespace fuzz.
2453 * imgoff now holds the true length of the target that
2454 * matches the preimage before the end of the file.
2456 * Count the number of characters in the preimage that fall
2457 * beyond the end of the file and make sure that all of them
2458 * are whitespace characters. (This can only happen if
2459 * we are removing blank lines at the end of the file.)
2461 buf = preimage_eof = preimage->buf + preoff;
2462 for ( ; i < preimage->nr; i++)
2463 preoff += preimage->line[i].len;
2464 preimage_end = preimage->buf + preoff;
2465 for ( ; buf < preimage_end; buf++)
2466 if (!isspace(*buf))
2467 return 0;
2470 * Update the preimage and the common postimage context
2471 * lines to use the same whitespace as the target.
2472 * If whitespace is missing in the target (i.e.
2473 * if the preimage extends beyond the end of the file),
2474 * use the whitespace from the preimage.
2476 extra_chars = preimage_end - preimage_eof;
2477 strbuf_init(&fixed, imgoff + extra_chars);
2478 strbuf_add(&fixed, img->buf + current, imgoff);
2479 strbuf_add(&fixed, preimage_eof, extra_chars);
2480 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2481 update_pre_post_images(preimage, postimage,
2482 fixed_buf, fixed_len, postlen);
2483 return 1;
2486 static int match_fragment(struct apply_state *state,
2487 struct image *img,
2488 struct image *preimage,
2489 struct image *postimage,
2490 unsigned long current,
2491 int current_lno,
2492 unsigned ws_rule,
2493 int match_beginning, int match_end)
2495 int i;
2496 char *fixed_buf, *buf, *orig, *target;
2497 struct strbuf fixed;
2498 size_t fixed_len, postlen;
2499 int preimage_limit;
2501 if (preimage->nr + current_lno <= img->nr) {
2503 * The hunk falls within the boundaries of img.
2505 preimage_limit = preimage->nr;
2506 if (match_end && (preimage->nr + current_lno != img->nr))
2507 return 0;
2508 } else if (state->ws_error_action == correct_ws_error &&
2509 (ws_rule & WS_BLANK_AT_EOF)) {
2511 * This hunk extends beyond the end of img, and we are
2512 * removing blank lines at the end of the file. This
2513 * many lines from the beginning of the preimage must
2514 * match with img, and the remainder of the preimage
2515 * must be blank.
2517 preimage_limit = img->nr - current_lno;
2518 } else {
2520 * The hunk extends beyond the end of the img and
2521 * we are not removing blanks at the end, so we
2522 * should reject the hunk at this position.
2524 return 0;
2527 if (match_beginning && current_lno)
2528 return 0;
2530 /* Quick hash check */
2531 for (i = 0; i < preimage_limit; i++)
2532 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2533 (preimage->line[i].hash != img->line[current_lno + i].hash))
2534 return 0;
2536 if (preimage_limit == preimage->nr) {
2538 * Do we have an exact match? If we were told to match
2539 * at the end, size must be exactly at current+fragsize,
2540 * otherwise current+fragsize must be still within the preimage,
2541 * and either case, the old piece should match the preimage
2542 * exactly.
2544 if ((match_end
2545 ? (current + preimage->len == img->len)
2546 : (current + preimage->len <= img->len)) &&
2547 !memcmp(img->buf + current, preimage->buf, preimage->len))
2548 return 1;
2549 } else {
2551 * The preimage extends beyond the end of img, so
2552 * there cannot be an exact match.
2554 * There must be one non-blank context line that match
2555 * a line before the end of img.
2557 char *buf_end;
2559 buf = preimage->buf;
2560 buf_end = buf;
2561 for (i = 0; i < preimage_limit; i++)
2562 buf_end += preimage->line[i].len;
2564 for ( ; buf < buf_end; buf++)
2565 if (!isspace(*buf))
2566 break;
2567 if (buf == buf_end)
2568 return 0;
2572 * No exact match. If we are ignoring whitespace, run a line-by-line
2573 * fuzzy matching. We collect all the line length information because
2574 * we need it to adjust whitespace if we match.
2576 if (state->ws_ignore_action == ignore_ws_change)
2577 return line_by_line_fuzzy_match(img, preimage, postimage,
2578 current, current_lno, preimage_limit);
2580 if (state->ws_error_action != correct_ws_error)
2581 return 0;
2584 * The hunk does not apply byte-by-byte, but the hash says
2585 * it might with whitespace fuzz. We weren't asked to
2586 * ignore whitespace, we were asked to correct whitespace
2587 * errors, so let's try matching after whitespace correction.
2589 * While checking the preimage against the target, whitespace
2590 * errors in both fixed, we count how large the corresponding
2591 * postimage needs to be. The postimage prepared by
2592 * apply_one_fragment() has whitespace errors fixed on added
2593 * lines already, but the common lines were propagated as-is,
2594 * which may become longer when their whitespace errors are
2595 * fixed.
2598 /* First count added lines in postimage */
2599 postlen = 0;
2600 for (i = 0; i < postimage->nr; i++) {
2601 if (!(postimage->line[i].flag & LINE_COMMON))
2602 postlen += postimage->line[i].len;
2606 * The preimage may extend beyond the end of the file,
2607 * but in this loop we will only handle the part of the
2608 * preimage that falls within the file.
2610 strbuf_init(&fixed, preimage->len + 1);
2611 orig = preimage->buf;
2612 target = img->buf + current;
2613 for (i = 0; i < preimage_limit; i++) {
2614 size_t oldlen = preimage->line[i].len;
2615 size_t tgtlen = img->line[current_lno + i].len;
2616 size_t fixstart = fixed.len;
2617 struct strbuf tgtfix;
2618 int match;
2620 /* Try fixing the line in the preimage */
2621 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2623 /* Try fixing the line in the target */
2624 strbuf_init(&tgtfix, tgtlen);
2625 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2628 * If they match, either the preimage was based on
2629 * a version before our tree fixed whitespace breakage,
2630 * or we are lacking a whitespace-fix patch the tree
2631 * the preimage was based on already had (i.e. target
2632 * has whitespace breakage, the preimage doesn't).
2633 * In either case, we are fixing the whitespace breakages
2634 * so we might as well take the fix together with their
2635 * real change.
2637 match = (tgtfix.len == fixed.len - fixstart &&
2638 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2639 fixed.len - fixstart));
2641 /* Add the length if this is common with the postimage */
2642 if (preimage->line[i].flag & LINE_COMMON)
2643 postlen += tgtfix.len;
2645 strbuf_release(&tgtfix);
2646 if (!match)
2647 goto unmatch_exit;
2649 orig += oldlen;
2650 target += tgtlen;
2655 * Now handle the lines in the preimage that falls beyond the
2656 * end of the file (if any). They will only match if they are
2657 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2658 * false).
2660 for ( ; i < preimage->nr; i++) {
2661 size_t fixstart = fixed.len; /* start of the fixed preimage */
2662 size_t oldlen = preimage->line[i].len;
2663 int j;
2665 /* Try fixing the line in the preimage */
2666 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2668 for (j = fixstart; j < fixed.len; j++)
2669 if (!isspace(fixed.buf[j]))
2670 goto unmatch_exit;
2672 orig += oldlen;
2676 * Yes, the preimage is based on an older version that still
2677 * has whitespace breakages unfixed, and fixing them makes the
2678 * hunk match. Update the context lines in the postimage.
2680 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2681 if (postlen < postimage->len)
2682 postlen = 0;
2683 update_pre_post_images(preimage, postimage,
2684 fixed_buf, fixed_len, postlen);
2685 return 1;
2687 unmatch_exit:
2688 strbuf_release(&fixed);
2689 return 0;
2692 static int find_pos(struct apply_state *state,
2693 struct image *img,
2694 struct image *preimage,
2695 struct image *postimage,
2696 int line,
2697 unsigned ws_rule,
2698 int match_beginning, int match_end)
2700 int i;
2701 unsigned long backwards, forwards, current;
2702 int backwards_lno, forwards_lno, current_lno;
2705 * When running with --allow-overlap, it is possible that a hunk is
2706 * seen that pretends to start at the beginning (but no longer does),
2707 * and that *still* needs to match the end. So trust `match_end` more
2708 * than `match_beginning`.
2710 if (state->allow_overlap && match_beginning && match_end &&
2711 img->nr - preimage->nr != 0)
2712 match_beginning = 0;
2715 * If match_beginning or match_end is specified, there is no
2716 * point starting from a wrong line that will never match and
2717 * wander around and wait for a match at the specified end.
2719 if (match_beginning)
2720 line = 0;
2721 else if (match_end)
2722 line = img->nr - preimage->nr;
2725 * Because the comparison is unsigned, the following test
2726 * will also take care of a negative line number that can
2727 * result when match_end and preimage is larger than the target.
2729 if ((size_t) line > img->nr)
2730 line = img->nr;
2732 current = 0;
2733 for (i = 0; i < line; i++)
2734 current += img->line[i].len;
2737 * There's probably some smart way to do this, but I'll leave
2738 * that to the smart and beautiful people. I'm simple and stupid.
2740 backwards = current;
2741 backwards_lno = line;
2742 forwards = current;
2743 forwards_lno = line;
2744 current_lno = line;
2746 for (i = 0; ; i++) {
2747 if (match_fragment(state, img, preimage, postimage,
2748 current, current_lno, ws_rule,
2749 match_beginning, match_end))
2750 return current_lno;
2752 again:
2753 if (backwards_lno == 0 && forwards_lno == img->nr)
2754 break;
2756 if (i & 1) {
2757 if (backwards_lno == 0) {
2758 i++;
2759 goto again;
2761 backwards_lno--;
2762 backwards -= img->line[backwards_lno].len;
2763 current = backwards;
2764 current_lno = backwards_lno;
2765 } else {
2766 if (forwards_lno == img->nr) {
2767 i++;
2768 goto again;
2770 forwards += img->line[forwards_lno].len;
2771 forwards_lno++;
2772 current = forwards;
2773 current_lno = forwards_lno;
2777 return -1;
2780 static void remove_first_line(struct image *img)
2782 img->buf += img->line[0].len;
2783 img->len -= img->line[0].len;
2784 img->line++;
2785 img->nr--;
2788 static void remove_last_line(struct image *img)
2790 img->len -= img->line[--img->nr].len;
2794 * The change from "preimage" and "postimage" has been found to
2795 * apply at applied_pos (counts in line numbers) in "img".
2796 * Update "img" to remove "preimage" and replace it with "postimage".
2798 static void update_image(struct apply_state *state,
2799 struct image *img,
2800 int applied_pos,
2801 struct image *preimage,
2802 struct image *postimage)
2805 * remove the copy of preimage at offset in img
2806 * and replace it with postimage
2808 int i, nr;
2809 size_t remove_count, insert_count, applied_at = 0;
2810 char *result;
2811 int preimage_limit;
2814 * If we are removing blank lines at the end of img,
2815 * the preimage may extend beyond the end.
2816 * If that is the case, we must be careful only to
2817 * remove the part of the preimage that falls within
2818 * the boundaries of img. Initialize preimage_limit
2819 * to the number of lines in the preimage that falls
2820 * within the boundaries.
2822 preimage_limit = preimage->nr;
2823 if (preimage_limit > img->nr - applied_pos)
2824 preimage_limit = img->nr - applied_pos;
2826 for (i = 0; i < applied_pos; i++)
2827 applied_at += img->line[i].len;
2829 remove_count = 0;
2830 for (i = 0; i < preimage_limit; i++)
2831 remove_count += img->line[applied_pos + i].len;
2832 insert_count = postimage->len;
2834 /* Adjust the contents */
2835 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2836 memcpy(result, img->buf, applied_at);
2837 memcpy(result + applied_at, postimage->buf, postimage->len);
2838 memcpy(result + applied_at + postimage->len,
2839 img->buf + (applied_at + remove_count),
2840 img->len - (applied_at + remove_count));
2841 free(img->buf);
2842 img->buf = result;
2843 img->len += insert_count - remove_count;
2844 result[img->len] = '\0';
2846 /* Adjust the line table */
2847 nr = img->nr + postimage->nr - preimage_limit;
2848 if (preimage_limit < postimage->nr) {
2850 * NOTE: this knows that we never call remove_first_line()
2851 * on anything other than pre/post image.
2853 REALLOC_ARRAY(img->line, nr);
2854 img->line_allocated = img->line;
2856 if (preimage_limit != postimage->nr)
2857 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2858 img->line + applied_pos + preimage_limit,
2859 img->nr - (applied_pos + preimage_limit));
2860 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2861 if (!state->allow_overlap)
2862 for (i = 0; i < postimage->nr; i++)
2863 img->line[applied_pos + i].flag |= LINE_PATCHED;
2864 img->nr = nr;
2868 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2869 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2870 * replace the part of "img" with "postimage" text.
2872 static int apply_one_fragment(struct apply_state *state,
2873 struct image *img, struct fragment *frag,
2874 int inaccurate_eof, unsigned ws_rule,
2875 int nth_fragment)
2877 int match_beginning, match_end;
2878 const char *patch = frag->patch;
2879 int size = frag->size;
2880 char *old, *oldlines;
2881 struct strbuf newlines;
2882 int new_blank_lines_at_end = 0;
2883 int found_new_blank_lines_at_end = 0;
2884 int hunk_linenr = frag->linenr;
2885 unsigned long leading, trailing;
2886 int pos, applied_pos;
2887 struct image preimage;
2888 struct image postimage;
2890 memset(&preimage, 0, sizeof(preimage));
2891 memset(&postimage, 0, sizeof(postimage));
2892 oldlines = xmalloc(size);
2893 strbuf_init(&newlines, size);
2895 old = oldlines;
2896 while (size > 0) {
2897 char first;
2898 int len = linelen(patch, size);
2899 int plen;
2900 int added_blank_line = 0;
2901 int is_blank_context = 0;
2902 size_t start;
2904 if (!len)
2905 break;
2908 * "plen" is how much of the line we should use for
2909 * the actual patch data. Normally we just remove the
2910 * first character on the line, but if the line is
2911 * followed by "\ No newline", then we also remove the
2912 * last one (which is the newline, of course).
2914 plen = len - 1;
2915 if (len < size && patch[len] == '\\')
2916 plen--;
2917 first = *patch;
2918 if (state->apply_in_reverse) {
2919 if (first == '-')
2920 first = '+';
2921 else if (first == '+')
2922 first = '-';
2925 switch (first) {
2926 case '\n':
2927 /* Newer GNU diff, empty context line */
2928 if (plen < 0)
2929 /* ... followed by '\No newline'; nothing */
2930 break;
2931 *old++ = '\n';
2932 strbuf_addch(&newlines, '\n');
2933 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2934 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2935 is_blank_context = 1;
2936 break;
2937 case ' ':
2938 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2939 ws_blank_line(patch + 1, plen))
2940 is_blank_context = 1;
2941 /* fallthrough */
2942 case '-':
2943 memcpy(old, patch + 1, plen);
2944 add_line_info(&preimage, old, plen,
2945 (first == ' ' ? LINE_COMMON : 0));
2946 old += plen;
2947 if (first == '-')
2948 break;
2949 /* fallthrough */
2950 case '+':
2951 /* --no-add does not add new lines */
2952 if (first == '+' && state->no_add)
2953 break;
2955 start = newlines.len;
2956 if (first != '+' ||
2957 !state->whitespace_error ||
2958 state->ws_error_action != correct_ws_error) {
2959 strbuf_add(&newlines, patch + 1, plen);
2961 else {
2962 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2964 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2965 (first == '+' ? 0 : LINE_COMMON));
2966 if (first == '+' &&
2967 (ws_rule & WS_BLANK_AT_EOF) &&
2968 ws_blank_line(patch + 1, plen))
2969 added_blank_line = 1;
2970 break;
2971 case '@': case '\\':
2972 /* Ignore it, we already handled it */
2973 break;
2974 default:
2975 if (state->apply_verbosity > verbosity_normal)
2976 error(_("invalid start of line: '%c'"), first);
2977 applied_pos = -1;
2978 goto out;
2980 if (added_blank_line) {
2981 if (!new_blank_lines_at_end)
2982 found_new_blank_lines_at_end = hunk_linenr;
2983 new_blank_lines_at_end++;
2985 else if (is_blank_context)
2987 else
2988 new_blank_lines_at_end = 0;
2989 patch += len;
2990 size -= len;
2991 hunk_linenr++;
2993 if (inaccurate_eof &&
2994 old > oldlines && old[-1] == '\n' &&
2995 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2996 old--;
2997 strbuf_setlen(&newlines, newlines.len - 1);
2998 preimage.line_allocated[preimage.nr - 1].len--;
2999 postimage.line_allocated[postimage.nr - 1].len--;
3002 leading = frag->leading;
3003 trailing = frag->trailing;
3006 * A hunk to change lines at the beginning would begin with
3007 * @@ -1,L +N,M @@
3008 * but we need to be careful. -U0 that inserts before the second
3009 * line also has this pattern.
3011 * And a hunk to add to an empty file would begin with
3012 * @@ -0,0 +N,M @@
3014 * In other words, a hunk that is (frag->oldpos <= 1) with or
3015 * without leading context must match at the beginning.
3017 match_beginning = (!frag->oldpos ||
3018 (frag->oldpos == 1 && !state->unidiff_zero));
3021 * A hunk without trailing lines must match at the end.
3022 * However, we simply cannot tell if a hunk must match end
3023 * from the lack of trailing lines if the patch was generated
3024 * with unidiff without any context.
3026 match_end = !state->unidiff_zero && !trailing;
3028 pos = frag->newpos ? (frag->newpos - 1) : 0;
3029 preimage.buf = oldlines;
3030 preimage.len = old - oldlines;
3031 postimage.buf = newlines.buf;
3032 postimage.len = newlines.len;
3033 preimage.line = preimage.line_allocated;
3034 postimage.line = postimage.line_allocated;
3036 for (;;) {
3038 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3039 ws_rule, match_beginning, match_end);
3041 if (applied_pos >= 0)
3042 break;
3044 /* Am I at my context limits? */
3045 if ((leading <= state->p_context) && (trailing <= state->p_context))
3046 break;
3047 if (match_beginning || match_end) {
3048 match_beginning = match_end = 0;
3049 continue;
3053 * Reduce the number of context lines; reduce both
3054 * leading and trailing if they are equal otherwise
3055 * just reduce the larger context.
3057 if (leading >= trailing) {
3058 remove_first_line(&preimage);
3059 remove_first_line(&postimage);
3060 pos--;
3061 leading--;
3063 if (trailing > leading) {
3064 remove_last_line(&preimage);
3065 remove_last_line(&postimage);
3066 trailing--;
3070 if (applied_pos >= 0) {
3071 if (new_blank_lines_at_end &&
3072 preimage.nr + applied_pos >= img->nr &&
3073 (ws_rule & WS_BLANK_AT_EOF) &&
3074 state->ws_error_action != nowarn_ws_error) {
3075 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3076 found_new_blank_lines_at_end);
3077 if (state->ws_error_action == correct_ws_error) {
3078 while (new_blank_lines_at_end--)
3079 remove_last_line(&postimage);
3082 * We would want to prevent write_out_results()
3083 * from taking place in apply_patch() that follows
3084 * the callchain led us here, which is:
3085 * apply_patch->check_patch_list->check_patch->
3086 * apply_data->apply_fragments->apply_one_fragment
3088 if (state->ws_error_action == die_on_ws_error)
3089 state->apply = 0;
3092 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3093 int offset = applied_pos - pos;
3094 if (state->apply_in_reverse)
3095 offset = 0 - offset;
3096 fprintf_ln(stderr,
3097 Q_("Hunk #%d succeeded at %d (offset %d line).",
3098 "Hunk #%d succeeded at %d (offset %d lines).",
3099 offset),
3100 nth_fragment, applied_pos + 1, offset);
3104 * Warn if it was necessary to reduce the number
3105 * of context lines.
3107 if ((leading != frag->leading ||
3108 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3109 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3110 " to apply fragment at %d"),
3111 leading, trailing, applied_pos+1);
3112 update_image(state, img, applied_pos, &preimage, &postimage);
3113 } else {
3114 if (state->apply_verbosity > verbosity_normal)
3115 error(_("while searching for:\n%.*s"),
3116 (int)(old - oldlines), oldlines);
3119 out:
3120 free(oldlines);
3121 strbuf_release(&newlines);
3122 free(preimage.line_allocated);
3123 free(postimage.line_allocated);
3125 return (applied_pos < 0);
3128 static int apply_binary_fragment(struct apply_state *state,
3129 struct image *img,
3130 struct patch *patch)
3132 struct fragment *fragment = patch->fragments;
3133 unsigned long len;
3134 void *dst;
3136 if (!fragment)
3137 return error(_("missing binary patch data for '%s'"),
3138 patch->new_name ?
3139 patch->new_name :
3140 patch->old_name);
3142 /* Binary patch is irreversible without the optional second hunk */
3143 if (state->apply_in_reverse) {
3144 if (!fragment->next)
3145 return error(_("cannot reverse-apply a binary patch "
3146 "without the reverse hunk to '%s'"),
3147 patch->new_name
3148 ? patch->new_name : patch->old_name);
3149 fragment = fragment->next;
3151 switch (fragment->binary_patch_method) {
3152 case BINARY_DELTA_DEFLATED:
3153 dst = patch_delta(img->buf, img->len, fragment->patch,
3154 fragment->size, &len);
3155 if (!dst)
3156 return -1;
3157 clear_image(img);
3158 img->buf = dst;
3159 img->len = len;
3160 return 0;
3161 case BINARY_LITERAL_DEFLATED:
3162 clear_image(img);
3163 img->len = fragment->size;
3164 img->buf = xmemdupz(fragment->patch, img->len);
3165 return 0;
3167 return -1;
3171 * Replace "img" with the result of applying the binary patch.
3172 * The binary patch data itself in patch->fragment is still kept
3173 * but the preimage prepared by the caller in "img" is freed here
3174 * or in the helper function apply_binary_fragment() this calls.
3176 static int apply_binary(struct apply_state *state,
3177 struct image *img,
3178 struct patch *patch)
3180 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3181 struct object_id oid;
3182 const unsigned hexsz = the_hash_algo->hexsz;
3185 * For safety, we require patch index line to contain
3186 * full hex textual object ID for old and new, at least for now.
3188 if (strlen(patch->old_oid_prefix) != hexsz ||
3189 strlen(patch->new_oid_prefix) != hexsz ||
3190 get_oid_hex(patch->old_oid_prefix, &oid) ||
3191 get_oid_hex(patch->new_oid_prefix, &oid))
3192 return error(_("cannot apply binary patch to '%s' "
3193 "without full index line"), name);
3195 if (patch->old_name) {
3197 * See if the old one matches what the patch
3198 * applies to.
3200 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3201 &oid);
3202 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3203 return error(_("the patch applies to '%s' (%s), "
3204 "which does not match the "
3205 "current contents."),
3206 name, oid_to_hex(&oid));
3208 else {
3209 /* Otherwise, the old one must be empty. */
3210 if (img->len)
3211 return error(_("the patch applies to an empty "
3212 "'%s' but it is not empty"), name);
3215 get_oid_hex(patch->new_oid_prefix, &oid);
3216 if (is_null_oid(&oid)) {
3217 clear_image(img);
3218 return 0; /* deletion patch */
3221 if (has_object(the_repository, &oid, 0)) {
3222 /* We already have the postimage */
3223 enum object_type type;
3224 unsigned long size;
3225 char *result;
3227 result = repo_read_object_file(the_repository, &oid, &type,
3228 &size);
3229 if (!result)
3230 return error(_("the necessary postimage %s for "
3231 "'%s' cannot be read"),
3232 patch->new_oid_prefix, name);
3233 clear_image(img);
3234 img->buf = result;
3235 img->len = size;
3236 } else {
3238 * We have verified buf matches the preimage;
3239 * apply the patch data to it, which is stored
3240 * in the patch->fragments->{patch,size}.
3242 if (apply_binary_fragment(state, img, patch))
3243 return error(_("binary patch does not apply to '%s'"),
3244 name);
3246 /* verify that the result matches */
3247 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3248 &oid);
3249 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3250 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3251 name, patch->new_oid_prefix, oid_to_hex(&oid));
3254 return 0;
3257 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3259 struct fragment *frag = patch->fragments;
3260 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3261 unsigned ws_rule = patch->ws_rule;
3262 unsigned inaccurate_eof = patch->inaccurate_eof;
3263 int nth = 0;
3265 if (patch->is_binary)
3266 return apply_binary(state, img, patch);
3268 while (frag) {
3269 nth++;
3270 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3271 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3272 if (!state->apply_with_reject)
3273 return -1;
3274 frag->rejected = 1;
3276 frag = frag->next;
3278 return 0;
3281 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3283 if (S_ISGITLINK(mode)) {
3284 strbuf_grow(buf, 100);
3285 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3286 } else {
3287 enum object_type type;
3288 unsigned long sz;
3289 char *result;
3291 result = repo_read_object_file(the_repository, oid, &type,
3292 &sz);
3293 if (!result)
3294 return -1;
3295 /* XXX read_sha1_file NUL-terminates */
3296 strbuf_attach(buf, result, sz, sz + 1);
3298 return 0;
3301 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3303 if (!ce)
3304 return 0;
3305 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3308 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3310 struct string_list_item *item;
3312 if (!name)
3313 return NULL;
3315 item = string_list_lookup(&state->fn_table, name);
3316 if (item)
3317 return (struct patch *)item->util;
3319 return NULL;
3323 * item->util in the filename table records the status of the path.
3324 * Usually it points at a patch (whose result records the contents
3325 * of it after applying it), but it could be PATH_WAS_DELETED for a
3326 * path that a previously applied patch has already removed, or
3327 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3329 * The latter is needed to deal with a case where two paths A and B
3330 * are swapped by first renaming A to B and then renaming B to A;
3331 * moving A to B should not be prevented due to presence of B as we
3332 * will remove it in a later patch.
3334 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3335 #define PATH_WAS_DELETED ((struct patch *) -1)
3337 static int to_be_deleted(struct patch *patch)
3339 return patch == PATH_TO_BE_DELETED;
3342 static int was_deleted(struct patch *patch)
3344 return patch == PATH_WAS_DELETED;
3347 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3349 struct string_list_item *item;
3352 * Always add new_name unless patch is a deletion
3353 * This should cover the cases for normal diffs,
3354 * file creations and copies
3356 if (patch->new_name) {
3357 item = string_list_insert(&state->fn_table, patch->new_name);
3358 item->util = patch;
3362 * store a failure on rename/deletion cases because
3363 * later chunks shouldn't patch old names
3365 if ((patch->new_name == NULL) || (patch->is_rename)) {
3366 item = string_list_insert(&state->fn_table, patch->old_name);
3367 item->util = PATH_WAS_DELETED;
3371 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3374 * store information about incoming file deletion
3376 while (patch) {
3377 if ((patch->new_name == NULL) || (patch->is_rename)) {
3378 struct string_list_item *item;
3379 item = string_list_insert(&state->fn_table, patch->old_name);
3380 item->util = PATH_TO_BE_DELETED;
3382 patch = patch->next;
3386 static int checkout_target(struct index_state *istate,
3387 struct cache_entry *ce, struct stat *st)
3389 struct checkout costate = CHECKOUT_INIT;
3391 costate.refresh_cache = 1;
3392 costate.istate = istate;
3393 if (checkout_entry(ce, &costate, NULL, NULL) ||
3394 lstat(ce->name, st))
3395 return error(_("cannot checkout %s"), ce->name);
3396 return 0;
3399 static struct patch *previous_patch(struct apply_state *state,
3400 struct patch *patch,
3401 int *gone)
3403 struct patch *previous;
3405 *gone = 0;
3406 if (patch->is_copy || patch->is_rename)
3407 return NULL; /* "git" patches do not depend on the order */
3409 previous = in_fn_table(state, patch->old_name);
3410 if (!previous)
3411 return NULL;
3413 if (to_be_deleted(previous))
3414 return NULL; /* the deletion hasn't happened yet */
3416 if (was_deleted(previous))
3417 *gone = 1;
3419 return previous;
3422 static int verify_index_match(struct apply_state *state,
3423 const struct cache_entry *ce,
3424 struct stat *st)
3426 if (S_ISGITLINK(ce->ce_mode)) {
3427 if (!S_ISDIR(st->st_mode))
3428 return -1;
3429 return 0;
3431 return ie_match_stat(state->repo->index, ce, st,
3432 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3435 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3437 static int load_patch_target(struct apply_state *state,
3438 struct strbuf *buf,
3439 const struct cache_entry *ce,
3440 struct stat *st,
3441 struct patch *patch,
3442 const char *name,
3443 unsigned expected_mode)
3445 if (state->cached || state->check_index) {
3446 if (read_file_or_gitlink(ce, buf))
3447 return error(_("failed to read %s"), name);
3448 } else if (name) {
3449 if (S_ISGITLINK(expected_mode)) {
3450 if (ce)
3451 return read_file_or_gitlink(ce, buf);
3452 else
3453 return SUBMODULE_PATCH_WITHOUT_INDEX;
3454 } else if (has_symlink_leading_path(name, strlen(name))) {
3455 return error(_("reading from '%s' beyond a symbolic link"), name);
3456 } else {
3457 if (read_old_data(st, patch, name, buf))
3458 return error(_("failed to read %s"), name);
3461 return 0;
3465 * We are about to apply "patch"; populate the "image" with the
3466 * current version we have, from the working tree or from the index,
3467 * depending on the situation e.g. --cached/--index. If we are
3468 * applying a non-git patch that incrementally updates the tree,
3469 * we read from the result of a previous diff.
3471 static int load_preimage(struct apply_state *state,
3472 struct image *image,
3473 struct patch *patch, struct stat *st,
3474 const struct cache_entry *ce)
3476 struct strbuf buf = STRBUF_INIT;
3477 size_t len;
3478 char *img;
3479 struct patch *previous;
3480 int status;
3482 previous = previous_patch(state, patch, &status);
3483 if (status)
3484 return error(_("path %s has been renamed/deleted"),
3485 patch->old_name);
3486 if (previous) {
3487 /* We have a patched copy in memory; use that. */
3488 strbuf_add(&buf, previous->result, previous->resultsize);
3489 } else {
3490 status = load_patch_target(state, &buf, ce, st, patch,
3491 patch->old_name, patch->old_mode);
3492 if (status < 0)
3493 return status;
3494 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3496 * There is no way to apply subproject
3497 * patch without looking at the index.
3498 * NEEDSWORK: shouldn't this be flagged
3499 * as an error???
3501 free_fragment_list(patch->fragments);
3502 patch->fragments = NULL;
3503 } else if (status) {
3504 return error(_("failed to read %s"), patch->old_name);
3508 img = strbuf_detach(&buf, &len);
3509 prepare_image(image, img, len, !patch->is_binary);
3510 return 0;
3513 static int resolve_to(struct image *image, const struct object_id *result_id)
3515 unsigned long size;
3516 enum object_type type;
3518 clear_image(image);
3520 image->buf = repo_read_object_file(the_repository, result_id, &type,
3521 &size);
3522 if (!image->buf || type != OBJ_BLOB)
3523 die("unable to read blob object %s", oid_to_hex(result_id));
3524 image->len = size;
3526 return 0;
3529 static int three_way_merge(struct apply_state *state,
3530 struct image *image,
3531 char *path,
3532 const struct object_id *base,
3533 const struct object_id *ours,
3534 const struct object_id *theirs)
3536 mmfile_t base_file, our_file, their_file;
3537 mmbuffer_t result = { NULL };
3538 enum ll_merge_result status;
3540 /* resolve trivial cases first */
3541 if (oideq(base, ours))
3542 return resolve_to(image, theirs);
3543 else if (oideq(base, theirs) || oideq(ours, theirs))
3544 return resolve_to(image, ours);
3546 read_mmblob(&base_file, base);
3547 read_mmblob(&our_file, ours);
3548 read_mmblob(&their_file, theirs);
3549 status = ll_merge(&result, path,
3550 &base_file, "base",
3551 &our_file, "ours",
3552 &their_file, "theirs",
3553 state->repo->index,
3554 NULL);
3555 if (status == LL_MERGE_BINARY_CONFLICT)
3556 warning("Cannot merge binary files: %s (%s vs. %s)",
3557 path, "ours", "theirs");
3558 free(base_file.ptr);
3559 free(our_file.ptr);
3560 free(their_file.ptr);
3561 if (status < 0 || !result.ptr) {
3562 free(result.ptr);
3563 return -1;
3565 clear_image(image);
3566 image->buf = result.ptr;
3567 image->len = result.size;
3569 return status;
3573 * When directly falling back to add/add three-way merge, we read from
3574 * the current contents of the new_name. In no cases other than that
3575 * this function will be called.
3577 static int load_current(struct apply_state *state,
3578 struct image *image,
3579 struct patch *patch)
3581 struct strbuf buf = STRBUF_INIT;
3582 int status, pos;
3583 size_t len;
3584 char *img;
3585 struct stat st;
3586 struct cache_entry *ce;
3587 char *name = patch->new_name;
3588 unsigned mode = patch->new_mode;
3590 if (!patch->is_new)
3591 BUG("patch to %s is not a creation", patch->old_name);
3593 pos = index_name_pos(state->repo->index, name, strlen(name));
3594 if (pos < 0)
3595 return error(_("%s: does not exist in index"), name);
3596 ce = state->repo->index->cache[pos];
3597 if (lstat(name, &st)) {
3598 if (errno != ENOENT)
3599 return error_errno("%s", name);
3600 if (checkout_target(state->repo->index, ce, &st))
3601 return -1;
3603 if (verify_index_match(state, ce, &st))
3604 return error(_("%s: does not match index"), name);
3606 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3607 if (status < 0)
3608 return status;
3609 else if (status)
3610 return -1;
3611 img = strbuf_detach(&buf, &len);
3612 prepare_image(image, img, len, !patch->is_binary);
3613 return 0;
3616 static int try_threeway(struct apply_state *state,
3617 struct image *image,
3618 struct patch *patch,
3619 struct stat *st,
3620 const struct cache_entry *ce)
3622 struct object_id pre_oid, post_oid, our_oid;
3623 struct strbuf buf = STRBUF_INIT;
3624 size_t len;
3625 int status;
3626 char *img;
3627 struct image tmp_image;
3629 /* No point falling back to 3-way merge in these cases */
3630 if (patch->is_delete ||
3631 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3632 (patch->is_new && !patch->direct_to_threeway) ||
3633 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3634 return -1;
3636 /* Preimage the patch was prepared for */
3637 if (patch->is_new)
3638 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3639 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3640 read_blob_object(&buf, &pre_oid, patch->old_mode))
3641 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3643 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3644 fprintf(stderr, _("Performing three-way merge...\n"));
3646 img = strbuf_detach(&buf, &len);
3647 prepare_image(&tmp_image, img, len, 1);
3648 /* Apply the patch to get the post image */
3649 if (apply_fragments(state, &tmp_image, patch) < 0) {
3650 clear_image(&tmp_image);
3651 return -1;
3653 /* post_oid is theirs */
3654 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3655 clear_image(&tmp_image);
3657 /* our_oid is ours */
3658 if (patch->is_new) {
3659 if (load_current(state, &tmp_image, patch))
3660 return error(_("cannot read the current contents of '%s'"),
3661 patch->new_name);
3662 } else {
3663 if (load_preimage(state, &tmp_image, patch, st, ce))
3664 return error(_("cannot read the current contents of '%s'"),
3665 patch->old_name);
3667 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3668 clear_image(&tmp_image);
3670 /* in-core three-way merge between post and our using pre as base */
3671 status = three_way_merge(state, image, patch->new_name,
3672 &pre_oid, &our_oid, &post_oid);
3673 if (status < 0) {
3674 if (state->apply_verbosity > verbosity_silent)
3675 fprintf(stderr,
3676 _("Failed to perform three-way merge...\n"));
3677 return status;
3680 if (status) {
3681 patch->conflicted_threeway = 1;
3682 if (patch->is_new)
3683 oidclr(&patch->threeway_stage[0]);
3684 else
3685 oidcpy(&patch->threeway_stage[0], &pre_oid);
3686 oidcpy(&patch->threeway_stage[1], &our_oid);
3687 oidcpy(&patch->threeway_stage[2], &post_oid);
3688 if (state->apply_verbosity > verbosity_silent)
3689 fprintf(stderr,
3690 _("Applied patch to '%s' with conflicts.\n"),
3691 patch->new_name);
3692 } else {
3693 if (state->apply_verbosity > verbosity_silent)
3694 fprintf(stderr,
3695 _("Applied patch to '%s' cleanly.\n"),
3696 patch->new_name);
3698 return 0;
3701 static int apply_data(struct apply_state *state, struct patch *patch,
3702 struct stat *st, const struct cache_entry *ce)
3704 struct image image;
3706 if (load_preimage(state, &image, patch, st, ce) < 0)
3707 return -1;
3709 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3710 if (state->apply_verbosity > verbosity_silent &&
3711 state->threeway && !patch->direct_to_threeway)
3712 fprintf(stderr, _("Falling back to direct application...\n"));
3714 /* Note: with --reject, apply_fragments() returns 0 */
3715 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0)
3716 return -1;
3718 patch->result = image.buf;
3719 patch->resultsize = image.len;
3720 add_to_fn_table(state, patch);
3721 free(image.line_allocated);
3723 if (0 < patch->is_delete && patch->resultsize)
3724 return error(_("removal patch leaves file contents"));
3726 return 0;
3730 * If "patch" that we are looking at modifies or deletes what we have,
3731 * we would want it not to lose any local modification we have, either
3732 * in the working tree or in the index.
3734 * This also decides if a non-git patch is a creation patch or a
3735 * modification to an existing empty file. We do not check the state
3736 * of the current tree for a creation patch in this function; the caller
3737 * check_patch() separately makes sure (and errors out otherwise) that
3738 * the path the patch creates does not exist in the current tree.
3740 static int check_preimage(struct apply_state *state,
3741 struct patch *patch,
3742 struct cache_entry **ce,
3743 struct stat *st)
3745 const char *old_name = patch->old_name;
3746 struct patch *previous = NULL;
3747 int stat_ret = 0, status;
3748 unsigned st_mode = 0;
3750 if (!old_name)
3751 return 0;
3753 assert(patch->is_new <= 0);
3754 previous = previous_patch(state, patch, &status);
3756 if (status)
3757 return error(_("path %s has been renamed/deleted"), old_name);
3758 if (previous) {
3759 st_mode = previous->new_mode;
3760 } else if (!state->cached) {
3761 stat_ret = lstat(old_name, st);
3762 if (stat_ret && errno != ENOENT)
3763 return error_errno("%s", old_name);
3766 if (state->check_index && !previous) {
3767 int pos = index_name_pos(state->repo->index, old_name,
3768 strlen(old_name));
3769 if (pos < 0) {
3770 if (patch->is_new < 0)
3771 goto is_new;
3772 return error(_("%s: does not exist in index"), old_name);
3774 *ce = state->repo->index->cache[pos];
3775 if (stat_ret < 0) {
3776 if (checkout_target(state->repo->index, *ce, st))
3777 return -1;
3779 if (!state->cached && verify_index_match(state, *ce, st))
3780 return error(_("%s: does not match index"), old_name);
3781 if (state->cached)
3782 st_mode = (*ce)->ce_mode;
3783 } else if (stat_ret < 0) {
3784 if (patch->is_new < 0)
3785 goto is_new;
3786 return error_errno("%s", old_name);
3789 if (!state->cached && !previous) {
3790 if (*ce && !(*ce)->ce_mode)
3791 BUG("ce_mode == 0 for path '%s'", old_name);
3793 if (trust_executable_bit)
3794 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3795 else if (*ce)
3796 st_mode = (*ce)->ce_mode;
3797 else
3798 st_mode = patch->old_mode;
3801 if (patch->is_new < 0)
3802 patch->is_new = 0;
3803 if (!patch->old_mode)
3804 patch->old_mode = st_mode;
3805 if ((st_mode ^ patch->old_mode) & S_IFMT)
3806 return error(_("%s: wrong type"), old_name);
3807 if (st_mode != patch->old_mode)
3808 warning(_("%s has type %o, expected %o"),
3809 old_name, st_mode, patch->old_mode);
3810 if (!patch->new_mode && !patch->is_delete)
3811 patch->new_mode = st_mode;
3812 return 0;
3814 is_new:
3815 patch->is_new = 1;
3816 patch->is_delete = 0;
3817 FREE_AND_NULL(patch->old_name);
3818 return 0;
3822 #define EXISTS_IN_INDEX 1
3823 #define EXISTS_IN_WORKTREE 2
3824 #define EXISTS_IN_INDEX_AS_ITA 3
3826 static int check_to_create(struct apply_state *state,
3827 const char *new_name,
3828 int ok_if_exists)
3830 struct stat nst;
3832 if (state->check_index && (!ok_if_exists || !state->cached)) {
3833 int pos;
3835 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3836 if (pos >= 0) {
3837 struct cache_entry *ce = state->repo->index->cache[pos];
3839 /* allow ITA, as they do not yet exist in the index */
3840 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3841 return EXISTS_IN_INDEX;
3843 /* ITA entries can never match working tree files */
3844 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3845 return EXISTS_IN_INDEX_AS_ITA;
3849 if (state->cached)
3850 return 0;
3852 if (!lstat(new_name, &nst)) {
3853 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3854 return 0;
3856 * A leading component of new_name might be a symlink
3857 * that is going to be removed with this patch, but
3858 * still pointing at somewhere that has the path.
3859 * In such a case, path "new_name" does not exist as
3860 * far as git is concerned.
3862 if (has_symlink_leading_path(new_name, strlen(new_name)))
3863 return 0;
3865 return EXISTS_IN_WORKTREE;
3866 } else if (!is_missing_file_error(errno)) {
3867 return error_errno("%s", new_name);
3869 return 0;
3872 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3874 for ( ; patch; patch = patch->next) {
3875 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3876 (patch->is_rename || patch->is_delete))
3877 /* the symlink at patch->old_name is removed */
3878 strset_add(&state->removed_symlinks, patch->old_name);
3880 if (patch->new_name && S_ISLNK(patch->new_mode))
3881 /* the symlink at patch->new_name is created or remains */
3882 strset_add(&state->kept_symlinks, patch->new_name);
3886 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3888 do {
3889 while (--name->len && name->buf[name->len] != '/')
3890 ; /* scan backwards */
3891 if (!name->len)
3892 break;
3893 name->buf[name->len] = '\0';
3894 if (strset_contains(&state->kept_symlinks, name->buf))
3895 return 1;
3896 if (strset_contains(&state->removed_symlinks, name->buf))
3898 * This cannot be "return 0", because we may
3899 * see a new one created at a higher level.
3901 continue;
3903 /* otherwise, check the preimage */
3904 if (state->check_index) {
3905 struct cache_entry *ce;
3907 ce = index_file_exists(state->repo->index, name->buf,
3908 name->len, ignore_case);
3909 if (ce && S_ISLNK(ce->ce_mode))
3910 return 1;
3911 } else {
3912 struct stat st;
3913 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3914 return 1;
3916 } while (1);
3917 return 0;
3920 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3922 int ret;
3923 struct strbuf name = STRBUF_INIT;
3925 assert(*name_ != '\0');
3926 strbuf_addstr(&name, name_);
3927 ret = path_is_beyond_symlink_1(state, &name);
3928 strbuf_release(&name);
3930 return ret;
3933 static int check_unsafe_path(struct patch *patch)
3935 const char *old_name = NULL;
3936 const char *new_name = NULL;
3937 if (patch->is_delete)
3938 old_name = patch->old_name;
3939 else if (!patch->is_new && !patch->is_copy)
3940 old_name = patch->old_name;
3941 if (!patch->is_delete)
3942 new_name = patch->new_name;
3944 if (old_name && !verify_path(old_name, patch->old_mode))
3945 return error(_("invalid path '%s'"), old_name);
3946 if (new_name && !verify_path(new_name, patch->new_mode))
3947 return error(_("invalid path '%s'"), new_name);
3948 return 0;
3952 * Check and apply the patch in-core; leave the result in patch->result
3953 * for the caller to write it out to the final destination.
3955 static int check_patch(struct apply_state *state, struct patch *patch)
3957 struct stat st;
3958 const char *old_name = patch->old_name;
3959 const char *new_name = patch->new_name;
3960 const char *name = old_name ? old_name : new_name;
3961 struct cache_entry *ce = NULL;
3962 struct patch *tpatch;
3963 int ok_if_exists;
3964 int status;
3966 patch->rejected = 1; /* we will drop this after we succeed */
3968 status = check_preimage(state, patch, &ce, &st);
3969 if (status)
3970 return status;
3971 old_name = patch->old_name;
3974 * A type-change diff is always split into a patch to delete
3975 * old, immediately followed by a patch to create new (see
3976 * diff.c::run_diff()); in such a case it is Ok that the entry
3977 * to be deleted by the previous patch is still in the working
3978 * tree and in the index.
3980 * A patch to swap-rename between A and B would first rename A
3981 * to B and then rename B to A. While applying the first one,
3982 * the presence of B should not stop A from getting renamed to
3983 * B; ask to_be_deleted() about the later rename. Removal of
3984 * B and rename from A to B is handled the same way by asking
3985 * was_deleted().
3987 if ((tpatch = in_fn_table(state, new_name)) &&
3988 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3989 ok_if_exists = 1;
3990 else
3991 ok_if_exists = 0;
3993 if (new_name &&
3994 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3995 int err = check_to_create(state, new_name, ok_if_exists);
3997 if (err && state->threeway) {
3998 patch->direct_to_threeway = 1;
3999 } else switch (err) {
4000 case 0:
4001 break; /* happy */
4002 case EXISTS_IN_INDEX:
4003 return error(_("%s: already exists in index"), new_name);
4004 case EXISTS_IN_INDEX_AS_ITA:
4005 return error(_("%s: does not match index"), new_name);
4006 case EXISTS_IN_WORKTREE:
4007 return error(_("%s: already exists in working directory"),
4008 new_name);
4009 default:
4010 return err;
4013 if (!patch->new_mode) {
4014 if (0 < patch->is_new)
4015 patch->new_mode = S_IFREG | 0644;
4016 else
4017 patch->new_mode = patch->old_mode;
4021 if (new_name && old_name) {
4022 int same = !strcmp(old_name, new_name);
4023 if (!patch->new_mode)
4024 patch->new_mode = patch->old_mode;
4025 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4026 if (same)
4027 return error(_("new mode (%o) of %s does not "
4028 "match old mode (%o)"),
4029 patch->new_mode, new_name,
4030 patch->old_mode);
4031 else
4032 return error(_("new mode (%o) of %s does not "
4033 "match old mode (%o) of %s"),
4034 patch->new_mode, new_name,
4035 patch->old_mode, old_name);
4039 if (!state->unsafe_paths && check_unsafe_path(patch))
4040 return -128;
4043 * An attempt to read from or delete a path that is beyond a
4044 * symbolic link will be prevented by load_patch_target() that
4045 * is called at the beginning of apply_data() so we do not
4046 * have to worry about a patch marked with "is_delete" bit
4047 * here. We however need to make sure that the patch result
4048 * is not deposited to a path that is beyond a symbolic link
4049 * here.
4051 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4052 return error(_("affected file '%s' is beyond a symbolic link"),
4053 patch->new_name);
4055 if (apply_data(state, patch, &st, ce) < 0)
4056 return error(_("%s: patch does not apply"), name);
4057 patch->rejected = 0;
4058 return 0;
4061 static int check_patch_list(struct apply_state *state, struct patch *patch)
4063 int err = 0;
4065 prepare_symlink_changes(state, patch);
4066 prepare_fn_table(state, patch);
4067 while (patch) {
4068 int res;
4069 if (state->apply_verbosity > verbosity_normal)
4070 say_patch_name(stderr,
4071 _("Checking patch %s..."), patch);
4072 res = check_patch(state, patch);
4073 if (res == -128)
4074 return -128;
4075 err |= res;
4076 patch = patch->next;
4078 return err;
4081 static int read_apply_cache(struct apply_state *state)
4083 if (state->index_file)
4084 return read_index_from(state->repo->index, state->index_file,
4085 get_git_dir());
4086 else
4087 return repo_read_index(state->repo);
4090 /* This function tries to read the object name from the current index */
4091 static int get_current_oid(struct apply_state *state, const char *path,
4092 struct object_id *oid)
4094 int pos;
4096 if (read_apply_cache(state) < 0)
4097 return -1;
4098 pos = index_name_pos(state->repo->index, path, strlen(path));
4099 if (pos < 0)
4100 return -1;
4101 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4102 return 0;
4105 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4108 * A usable gitlink patch has only one fragment (hunk) that looks like:
4109 * @@ -1 +1 @@
4110 * -Subproject commit <old sha1>
4111 * +Subproject commit <new sha1>
4112 * or
4113 * @@ -1 +0,0 @@
4114 * -Subproject commit <old sha1>
4115 * for a removal patch.
4117 struct fragment *hunk = p->fragments;
4118 static const char heading[] = "-Subproject commit ";
4119 char *preimage;
4121 if (/* does the patch have only one hunk? */
4122 hunk && !hunk->next &&
4123 /* is its preimage one line? */
4124 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4125 /* does preimage begin with the heading? */
4126 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4127 starts_with(++preimage, heading) &&
4128 /* does it record full SHA-1? */
4129 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4130 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4131 /* does the abbreviated name on the index line agree with it? */
4132 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4133 return 0; /* it all looks fine */
4135 /* we may have full object name on the index line */
4136 return get_oid_hex(p->old_oid_prefix, oid);
4139 /* Build an index that contains just the files needed for a 3way merge */
4140 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4142 struct patch *patch;
4143 struct index_state result = INDEX_STATE_INIT(state->repo);
4144 struct lock_file lock = LOCK_INIT;
4145 int res;
4147 /* Once we start supporting the reverse patch, it may be
4148 * worth showing the new sha1 prefix, but until then...
4150 for (patch = list; patch; patch = patch->next) {
4151 struct object_id oid;
4152 struct cache_entry *ce;
4153 const char *name;
4155 name = patch->old_name ? patch->old_name : patch->new_name;
4156 if (0 < patch->is_new)
4157 continue;
4159 if (S_ISGITLINK(patch->old_mode)) {
4160 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4161 ; /* ok, the textual part looks sane */
4162 else
4163 return error(_("sha1 information is lacking or "
4164 "useless for submodule %s"), name);
4165 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4166 ; /* ok */
4167 } else if (!patch->lines_added && !patch->lines_deleted) {
4168 /* mode-only change: update the current */
4169 if (get_current_oid(state, patch->old_name, &oid))
4170 return error(_("mode change for %s, which is not "
4171 "in current HEAD"), name);
4172 } else
4173 return error(_("sha1 information is lacking or useless "
4174 "(%s)."), name);
4176 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4177 if (!ce)
4178 return error(_("make_cache_entry failed for path '%s'"),
4179 name);
4180 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4181 discard_cache_entry(ce);
4182 return error(_("could not add %s to temporary index"),
4183 name);
4187 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4188 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4189 discard_index(&result);
4191 if (res)
4192 return error(_("could not write temporary index to %s"),
4193 state->fake_ancestor);
4195 return 0;
4198 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4200 int files, adds, dels;
4202 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4203 files++;
4204 adds += patch->lines_added;
4205 dels += patch->lines_deleted;
4206 show_stats(state, patch);
4209 print_stat_summary(stdout, files, adds, dels);
4212 static void numstat_patch_list(struct apply_state *state,
4213 struct patch *patch)
4215 for ( ; patch; patch = patch->next) {
4216 const char *name;
4217 name = patch->new_name ? patch->new_name : patch->old_name;
4218 if (patch->is_binary)
4219 printf("-\t-\t");
4220 else
4221 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4222 write_name_quoted(name, stdout, state->line_termination);
4226 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4228 if (mode)
4229 printf(" %s mode %06o %s\n", newdelete, mode, name);
4230 else
4231 printf(" %s %s\n", newdelete, name);
4234 static void show_mode_change(struct patch *p, int show_name)
4236 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4237 if (show_name)
4238 printf(" mode change %06o => %06o %s\n",
4239 p->old_mode, p->new_mode, p->new_name);
4240 else
4241 printf(" mode change %06o => %06o\n",
4242 p->old_mode, p->new_mode);
4246 static void show_rename_copy(struct patch *p)
4248 const char *renamecopy = p->is_rename ? "rename" : "copy";
4249 const char *old_name, *new_name;
4251 /* Find common prefix */
4252 old_name = p->old_name;
4253 new_name = p->new_name;
4254 while (1) {
4255 const char *slash_old, *slash_new;
4256 slash_old = strchr(old_name, '/');
4257 slash_new = strchr(new_name, '/');
4258 if (!slash_old ||
4259 !slash_new ||
4260 slash_old - old_name != slash_new - new_name ||
4261 memcmp(old_name, new_name, slash_new - new_name))
4262 break;
4263 old_name = slash_old + 1;
4264 new_name = slash_new + 1;
4266 /* p->old_name through old_name is the common prefix, and old_name and
4267 * new_name through the end of names are renames
4269 if (old_name != p->old_name)
4270 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4271 (int)(old_name - p->old_name), p->old_name,
4272 old_name, new_name, p->score);
4273 else
4274 printf(" %s %s => %s (%d%%)\n", renamecopy,
4275 p->old_name, p->new_name, p->score);
4276 show_mode_change(p, 0);
4279 static void summary_patch_list(struct patch *patch)
4281 struct patch *p;
4283 for (p = patch; p; p = p->next) {
4284 if (p->is_new)
4285 show_file_mode_name("create", p->new_mode, p->new_name);
4286 else if (p->is_delete)
4287 show_file_mode_name("delete", p->old_mode, p->old_name);
4288 else {
4289 if (p->is_rename || p->is_copy)
4290 show_rename_copy(p);
4291 else {
4292 if (p->score) {
4293 printf(" rewrite %s (%d%%)\n",
4294 p->new_name, p->score);
4295 show_mode_change(p, 0);
4297 else
4298 show_mode_change(p, 1);
4304 static void patch_stats(struct apply_state *state, struct patch *patch)
4306 int lines = patch->lines_added + patch->lines_deleted;
4308 if (lines > state->max_change)
4309 state->max_change = lines;
4310 if (patch->old_name) {
4311 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4312 if (!len)
4313 len = strlen(patch->old_name);
4314 if (len > state->max_len)
4315 state->max_len = len;
4317 if (patch->new_name) {
4318 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4319 if (!len)
4320 len = strlen(patch->new_name);
4321 if (len > state->max_len)
4322 state->max_len = len;
4326 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4328 if (state->update_index && !state->ita_only) {
4329 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4330 return error(_("unable to remove %s from index"), patch->old_name);
4332 if (!state->cached) {
4333 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4334 remove_path(patch->old_name);
4337 return 0;
4340 static int add_index_file(struct apply_state *state,
4341 const char *path,
4342 unsigned mode,
4343 void *buf,
4344 unsigned long size)
4346 struct stat st;
4347 struct cache_entry *ce;
4348 int namelen = strlen(path);
4350 ce = make_empty_cache_entry(state->repo->index, namelen);
4351 memcpy(ce->name, path, namelen);
4352 ce->ce_mode = create_ce_mode(mode);
4353 ce->ce_flags = create_ce_flags(0);
4354 ce->ce_namelen = namelen;
4355 if (state->ita_only) {
4356 ce->ce_flags |= CE_INTENT_TO_ADD;
4357 set_object_name_for_intent_to_add_entry(ce);
4358 } else if (S_ISGITLINK(mode)) {
4359 const char *s;
4361 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4362 get_oid_hex(s, &ce->oid)) {
4363 discard_cache_entry(ce);
4364 return error(_("corrupt patch for submodule %s"), path);
4366 } else {
4367 if (!state->cached) {
4368 if (lstat(path, &st) < 0) {
4369 discard_cache_entry(ce);
4370 return error_errno(_("unable to stat newly "
4371 "created file '%s'"),
4372 path);
4374 fill_stat_cache_info(state->repo->index, ce, &st);
4376 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4377 discard_cache_entry(ce);
4378 return error(_("unable to create backing store "
4379 "for newly created file %s"), path);
4382 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4383 discard_cache_entry(ce);
4384 return error(_("unable to add cache entry for %s"), path);
4387 return 0;
4391 * Returns:
4392 * -1 if an unrecoverable error happened
4393 * 0 if everything went well
4394 * 1 if a recoverable error happened
4396 static int try_create_file(struct apply_state *state, const char *path,
4397 unsigned int mode, const char *buf,
4398 unsigned long size)
4400 int fd, res;
4401 struct strbuf nbuf = STRBUF_INIT;
4403 if (S_ISGITLINK(mode)) {
4404 struct stat st;
4405 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4406 return 0;
4407 return !!mkdir(path, 0777);
4410 if (has_symlinks && S_ISLNK(mode))
4411 /* Although buf:size is counted string, it also is NUL
4412 * terminated.
4414 return !!symlink(buf, path);
4416 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4417 if (fd < 0)
4418 return 1;
4420 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4421 size = nbuf.len;
4422 buf = nbuf.buf;
4425 res = write_in_full(fd, buf, size) < 0;
4426 if (res)
4427 error_errno(_("failed to write to '%s'"), path);
4428 strbuf_release(&nbuf);
4430 if (close(fd) < 0 && !res)
4431 return error_errno(_("closing file '%s'"), path);
4433 return res ? -1 : 0;
4437 * We optimistically assume that the directories exist,
4438 * which is true 99% of the time anyway. If they don't,
4439 * we create them and try again.
4441 * Returns:
4442 * -1 on error
4443 * 0 otherwise
4445 static int create_one_file(struct apply_state *state,
4446 char *path,
4447 unsigned mode,
4448 const char *buf,
4449 unsigned long size)
4451 char *newpath = NULL;
4452 int res;
4454 if (state->cached)
4455 return 0;
4458 * We already try to detect whether files are beyond a symlink in our
4459 * up-front checks. But in the case where symlinks are created by any
4460 * of the intermediate hunks it can happen that our up-front checks
4461 * didn't yet see the symlink, but at the point of arriving here there
4462 * in fact is one. We thus repeat the check for symlinks here.
4464 * Note that this does not make the up-front check obsolete as the
4465 * failure mode is different:
4467 * - The up-front checks cause us to abort before we have written
4468 * anything into the working directory. So when we exit this way the
4469 * working directory remains clean.
4471 * - The checks here happen in the middle of the action where we have
4472 * already started to apply the patch. The end result will be a dirty
4473 * working directory.
4475 * Ideally, we should update the up-front checks to catch what would
4476 * happen when we apply the patch before we damage the working tree.
4477 * We have all the information necessary to do so. But for now, as a
4478 * part of embargoed security work, having this check would serve as a
4479 * reasonable first step.
4481 if (path_is_beyond_symlink(state, path))
4482 return error(_("affected file '%s' is beyond a symbolic link"), path);
4484 res = try_create_file(state, path, mode, buf, size);
4485 if (res < 0)
4486 return -1;
4487 if (!res)
4488 return 0;
4490 if (errno == ENOENT) {
4491 if (safe_create_leading_directories_no_share(path))
4492 return 0;
4493 res = try_create_file(state, path, mode, buf, size);
4494 if (res < 0)
4495 return -1;
4496 if (!res)
4497 return 0;
4500 if (errno == EEXIST || errno == EACCES) {
4501 /* We may be trying to create a file where a directory
4502 * used to be.
4504 struct stat st;
4505 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4506 errno = EEXIST;
4509 if (errno == EEXIST) {
4510 unsigned int nr = getpid();
4512 for (;;) {
4513 newpath = mkpathdup("%s~%u", path, nr);
4514 res = try_create_file(state, newpath, mode, buf, size);
4515 if (res < 0)
4516 goto out;
4517 if (!res) {
4518 if (!rename(newpath, path))
4519 goto out;
4520 unlink_or_warn(newpath);
4521 break;
4523 if (errno != EEXIST)
4524 break;
4525 ++nr;
4526 FREE_AND_NULL(newpath);
4529 res = error_errno(_("unable to write file '%s' mode %o"), path, mode);
4530 out:
4531 free(newpath);
4532 return res;
4535 static int add_conflicted_stages_file(struct apply_state *state,
4536 struct patch *patch)
4538 int stage, namelen;
4539 unsigned mode;
4540 struct cache_entry *ce;
4542 if (!state->update_index)
4543 return 0;
4544 namelen = strlen(patch->new_name);
4545 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4547 remove_file_from_index(state->repo->index, patch->new_name);
4548 for (stage = 1; stage < 4; stage++) {
4549 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4550 continue;
4551 ce = make_empty_cache_entry(state->repo->index, namelen);
4552 memcpy(ce->name, patch->new_name, namelen);
4553 ce->ce_mode = create_ce_mode(mode);
4554 ce->ce_flags = create_ce_flags(stage);
4555 ce->ce_namelen = namelen;
4556 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4557 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4558 discard_cache_entry(ce);
4559 return error(_("unable to add cache entry for %s"),
4560 patch->new_name);
4564 return 0;
4567 static int create_file(struct apply_state *state, struct patch *patch)
4569 char *path = patch->new_name;
4570 unsigned mode = patch->new_mode;
4571 unsigned long size = patch->resultsize;
4572 char *buf = patch->result;
4574 if (!mode)
4575 mode = S_IFREG | 0644;
4576 if (create_one_file(state, path, mode, buf, size))
4577 return -1;
4579 if (patch->conflicted_threeway)
4580 return add_conflicted_stages_file(state, patch);
4581 else if (state->update_index)
4582 return add_index_file(state, path, mode, buf, size);
4583 return 0;
4586 /* phase zero is to remove, phase one is to create */
4587 static int write_out_one_result(struct apply_state *state,
4588 struct patch *patch,
4589 int phase)
4591 if (patch->is_delete > 0) {
4592 if (phase == 0)
4593 return remove_file(state, patch, 1);
4594 return 0;
4596 if (patch->is_new > 0 || patch->is_copy) {
4597 if (phase == 1)
4598 return create_file(state, patch);
4599 return 0;
4602 * Rename or modification boils down to the same
4603 * thing: remove the old, write the new
4605 if (phase == 0)
4606 return remove_file(state, patch, patch->is_rename);
4607 if (phase == 1)
4608 return create_file(state, patch);
4609 return 0;
4612 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4614 FILE *rej;
4615 char namebuf[PATH_MAX];
4616 struct fragment *frag;
4617 int fd, cnt = 0;
4618 struct strbuf sb = STRBUF_INIT;
4620 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4621 if (!frag->rejected)
4622 continue;
4623 cnt++;
4626 if (!cnt) {
4627 if (state->apply_verbosity > verbosity_normal)
4628 say_patch_name(stderr,
4629 _("Applied patch %s cleanly."), patch);
4630 return 0;
4633 /* This should not happen, because a removal patch that leaves
4634 * contents are marked "rejected" at the patch level.
4636 if (!patch->new_name)
4637 die(_("internal error"));
4639 /* Say this even without --verbose */
4640 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4641 "Applying patch %%s with %d rejects...",
4642 cnt),
4643 cnt);
4644 if (state->apply_verbosity > verbosity_silent)
4645 say_patch_name(stderr, sb.buf, patch);
4646 strbuf_release(&sb);
4648 cnt = strlen(patch->new_name);
4649 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4650 cnt = ARRAY_SIZE(namebuf) - 5;
4651 warning(_("truncating .rej filename to %.*s.rej"),
4652 cnt - 1, patch->new_name);
4654 memcpy(namebuf, patch->new_name, cnt);
4655 memcpy(namebuf + cnt, ".rej", 5);
4657 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4658 if (fd < 0) {
4659 if (errno != EEXIST)
4660 return error_errno(_("cannot open %s"), namebuf);
4661 if (unlink(namebuf))
4662 return error_errno(_("cannot unlink '%s'"), namebuf);
4663 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4664 if (fd < 0)
4665 return error_errno(_("cannot open %s"), namebuf);
4667 rej = fdopen(fd, "w");
4668 if (!rej) {
4669 error_errno(_("cannot open %s"), namebuf);
4670 close(fd);
4671 return -1;
4674 /* Normal git tools never deal with .rej, so do not pretend
4675 * this is a git patch by saying --git or giving extended
4676 * headers. While at it, maybe please "kompare" that wants
4677 * the trailing TAB and some garbage at the end of line ;-).
4679 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4680 patch->new_name, patch->new_name);
4681 for (cnt = 1, frag = patch->fragments;
4682 frag;
4683 cnt++, frag = frag->next) {
4684 if (!frag->rejected) {
4685 if (state->apply_verbosity > verbosity_silent)
4686 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4687 continue;
4689 if (state->apply_verbosity > verbosity_silent)
4690 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4691 fprintf(rej, "%.*s", frag->size, frag->patch);
4692 if (frag->patch[frag->size-1] != '\n')
4693 fputc('\n', rej);
4695 fclose(rej);
4696 return -1;
4700 * Returns:
4701 * -1 if an error happened
4702 * 0 if the patch applied cleanly
4703 * 1 if the patch did not apply cleanly
4705 static int write_out_results(struct apply_state *state, struct patch *list)
4707 int phase;
4708 int errs = 0;
4709 struct patch *l;
4710 struct string_list cpath = STRING_LIST_INIT_DUP;
4712 for (phase = 0; phase < 2; phase++) {
4713 l = list;
4714 while (l) {
4715 if (l->rejected)
4716 errs = 1;
4717 else {
4718 if (write_out_one_result(state, l, phase)) {
4719 string_list_clear(&cpath, 0);
4720 return -1;
4722 if (phase == 1) {
4723 if (write_out_one_reject(state, l))
4724 errs = 1;
4725 if (l->conflicted_threeway) {
4726 string_list_append(&cpath, l->new_name);
4727 errs = 1;
4731 l = l->next;
4735 if (cpath.nr) {
4736 struct string_list_item *item;
4738 string_list_sort(&cpath);
4739 if (state->apply_verbosity > verbosity_silent) {
4740 for_each_string_list_item(item, &cpath)
4741 fprintf(stderr, "U %s\n", item->string);
4743 string_list_clear(&cpath, 0);
4746 * rerere relies on the partially merged result being in the working
4747 * tree with conflict markers, but that isn't written with --cached.
4749 if (!state->cached)
4750 repo_rerere(state->repo, 0);
4753 return errs;
4757 * Try to apply a patch.
4759 * Returns:
4760 * -128 if a bad error happened (like patch unreadable)
4761 * -1 if patch did not apply and user cannot deal with it
4762 * 0 if the patch applied
4763 * 1 if the patch did not apply but user might fix it
4765 static int apply_patch(struct apply_state *state,
4766 int fd,
4767 const char *filename,
4768 int options)
4770 size_t offset;
4771 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4772 struct patch *list = NULL, **listp = &list;
4773 int skipped_patch = 0;
4774 int res = 0;
4775 int flush_attributes = 0;
4777 state->patch_input_file = filename;
4778 if (read_patch_file(&buf, fd) < 0)
4779 return -128;
4780 offset = 0;
4781 while (offset < buf.len) {
4782 struct patch *patch;
4783 int nr;
4785 CALLOC_ARRAY(patch, 1);
4786 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4787 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4788 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4789 if (nr < 0) {
4790 free_patch(patch);
4791 if (nr == -128) {
4792 res = -128;
4793 goto end;
4795 break;
4797 if (state->apply_in_reverse)
4798 reverse_patches(patch);
4799 if (use_patch(state, patch)) {
4800 patch_stats(state, patch);
4801 if (!list || !state->apply_in_reverse) {
4802 *listp = patch;
4803 listp = &patch->next;
4804 } else {
4805 patch->next = list;
4806 list = patch;
4809 if ((patch->new_name &&
4810 ends_with_path_components(patch->new_name,
4811 GITATTRIBUTES_FILE)) ||
4812 (patch->old_name &&
4813 ends_with_path_components(patch->old_name,
4814 GITATTRIBUTES_FILE)))
4815 flush_attributes = 1;
4817 else {
4818 if (state->apply_verbosity > verbosity_normal)
4819 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4820 free_patch(patch);
4821 skipped_patch++;
4823 offset += nr;
4826 if (!list && !skipped_patch) {
4827 if (!state->allow_empty) {
4828 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4829 res = -128;
4831 goto end;
4834 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4835 state->apply = 0;
4837 state->update_index = (state->check_index || state->ita_only) && state->apply;
4838 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4839 if (state->index_file)
4840 hold_lock_file_for_update(&state->lock_file,
4841 state->index_file,
4842 LOCK_DIE_ON_ERROR);
4843 else
4844 repo_hold_locked_index(state->repo, &state->lock_file,
4845 LOCK_DIE_ON_ERROR);
4848 if (state->check_index && read_apply_cache(state) < 0) {
4849 error(_("unable to read index file"));
4850 res = -128;
4851 goto end;
4854 if (state->check || state->apply) {
4855 int r = check_patch_list(state, list);
4856 if (r == -128) {
4857 res = -128;
4858 goto end;
4860 if (r < 0 && !state->apply_with_reject) {
4861 res = -1;
4862 goto end;
4866 if (state->apply) {
4867 int write_res = write_out_results(state, list);
4868 if (write_res < 0) {
4869 res = -128;
4870 goto end;
4872 if (write_res > 0) {
4873 /* with --3way, we still need to write the index out */
4874 res = state->apply_with_reject ? -1 : 1;
4875 goto end;
4879 if (state->fake_ancestor &&
4880 build_fake_ancestor(state, list)) {
4881 res = -128;
4882 goto end;
4885 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4886 stat_patch_list(state, list);
4888 if (state->numstat && state->apply_verbosity > verbosity_silent)
4889 numstat_patch_list(state, list);
4891 if (state->summary && state->apply_verbosity > verbosity_silent)
4892 summary_patch_list(list);
4894 if (flush_attributes)
4895 reset_parsed_attributes();
4896 end:
4897 free_patch_list(list);
4898 strbuf_release(&buf);
4899 string_list_clear(&state->fn_table, 0);
4900 return res;
4903 static int apply_option_parse_exclude(const struct option *opt,
4904 const char *arg, int unset)
4906 struct apply_state *state = opt->value;
4908 BUG_ON_OPT_NEG(unset);
4910 add_name_limit(state, arg, 1);
4911 return 0;
4914 static int apply_option_parse_include(const struct option *opt,
4915 const char *arg, int unset)
4917 struct apply_state *state = opt->value;
4919 BUG_ON_OPT_NEG(unset);
4921 add_name_limit(state, arg, 0);
4922 state->has_include = 1;
4923 return 0;
4926 static int apply_option_parse_p(const struct option *opt,
4927 const char *arg,
4928 int unset)
4930 struct apply_state *state = opt->value;
4932 BUG_ON_OPT_NEG(unset);
4934 state->p_value = atoi(arg);
4935 state->p_value_known = 1;
4936 return 0;
4939 static int apply_option_parse_space_change(const struct option *opt,
4940 const char *arg, int unset)
4942 struct apply_state *state = opt->value;
4944 BUG_ON_OPT_ARG(arg);
4946 if (unset)
4947 state->ws_ignore_action = ignore_ws_none;
4948 else
4949 state->ws_ignore_action = ignore_ws_change;
4950 return 0;
4953 static int apply_option_parse_whitespace(const struct option *opt,
4954 const char *arg, int unset)
4956 struct apply_state *state = opt->value;
4958 BUG_ON_OPT_NEG(unset);
4960 state->whitespace_option = arg;
4961 if (parse_whitespace_option(state, arg))
4962 return -1;
4963 return 0;
4966 static int apply_option_parse_directory(const struct option *opt,
4967 const char *arg, int unset)
4969 struct apply_state *state = opt->value;
4971 BUG_ON_OPT_NEG(unset);
4973 strbuf_reset(&state->root);
4974 strbuf_addstr(&state->root, arg);
4975 strbuf_complete(&state->root, '/');
4976 return 0;
4979 int apply_all_patches(struct apply_state *state,
4980 int argc,
4981 const char **argv,
4982 int options)
4984 int i;
4985 int res;
4986 int errs = 0;
4987 int read_stdin = 1;
4989 for (i = 0; i < argc; i++) {
4990 const char *arg = argv[i];
4991 char *to_free = NULL;
4992 int fd;
4994 if (!strcmp(arg, "-")) {
4995 res = apply_patch(state, 0, "<stdin>", options);
4996 if (res < 0)
4997 goto end;
4998 errs |= res;
4999 read_stdin = 0;
5000 continue;
5001 } else
5002 arg = to_free = prefix_filename(state->prefix, arg);
5004 fd = open(arg, O_RDONLY);
5005 if (fd < 0) {
5006 error(_("can't open patch '%s': %s"), arg, strerror(errno));
5007 res = -128;
5008 free(to_free);
5009 goto end;
5011 read_stdin = 0;
5012 set_default_whitespace_mode(state);
5013 res = apply_patch(state, fd, arg, options);
5014 close(fd);
5015 free(to_free);
5016 if (res < 0)
5017 goto end;
5018 errs |= res;
5020 set_default_whitespace_mode(state);
5021 if (read_stdin) {
5022 res = apply_patch(state, 0, "<stdin>", options);
5023 if (res < 0)
5024 goto end;
5025 errs |= res;
5028 if (state->whitespace_error) {
5029 if (state->squelch_whitespace_errors &&
5030 state->squelch_whitespace_errors < state->whitespace_error) {
5031 int squelched =
5032 state->whitespace_error - state->squelch_whitespace_errors;
5033 warning(Q_("squelched %d whitespace error",
5034 "squelched %d whitespace errors",
5035 squelched),
5036 squelched);
5038 if (state->ws_error_action == die_on_ws_error) {
5039 error(Q_("%d line adds whitespace errors.",
5040 "%d lines add whitespace errors.",
5041 state->whitespace_error),
5042 state->whitespace_error);
5043 res = -128;
5044 goto end;
5046 if (state->applied_after_fixing_ws && state->apply)
5047 warning(Q_("%d line applied after"
5048 " fixing whitespace errors.",
5049 "%d lines applied after"
5050 " fixing whitespace errors.",
5051 state->applied_after_fixing_ws),
5052 state->applied_after_fixing_ws);
5053 else if (state->whitespace_error)
5054 warning(Q_("%d line adds whitespace errors.",
5055 "%d lines add whitespace errors.",
5056 state->whitespace_error),
5057 state->whitespace_error);
5060 if (state->update_index) {
5061 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5062 if (res) {
5063 error(_("Unable to write new index file"));
5064 res = -128;
5065 goto end;
5069 res = !!errs;
5071 end:
5072 rollback_lock_file(&state->lock_file);
5074 if (state->apply_verbosity <= verbosity_silent) {
5075 set_error_routine(state->saved_error_routine);
5076 set_warn_routine(state->saved_warn_routine);
5079 if (res > -1)
5080 return res;
5081 return (res == -1 ? 1 : 128);
5084 int apply_parse_options(int argc, const char **argv,
5085 struct apply_state *state,
5086 int *force_apply, int *options,
5087 const char * const *apply_usage)
5089 struct option builtin_apply_options[] = {
5090 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5091 N_("don't apply changes matching the given path"),
5092 PARSE_OPT_NONEG, apply_option_parse_exclude),
5093 OPT_CALLBACK_F(0, "include", state, N_("path"),
5094 N_("apply changes matching the given path"),
5095 PARSE_OPT_NONEG, apply_option_parse_include),
5096 OPT_CALLBACK('p', NULL, state, N_("num"),
5097 N_("remove <num> leading slashes from traditional diff paths"),
5098 apply_option_parse_p),
5099 OPT_BOOL(0, "no-add", &state->no_add,
5100 N_("ignore additions made by the patch")),
5101 OPT_BOOL(0, "stat", &state->diffstat,
5102 N_("instead of applying the patch, output diffstat for the input")),
5103 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5104 OPT_NOOP_NOARG(0, "binary"),
5105 OPT_BOOL(0, "numstat", &state->numstat,
5106 N_("show number of added and deleted lines in decimal notation")),
5107 OPT_BOOL(0, "summary", &state->summary,
5108 N_("instead of applying the patch, output a summary for the input")),
5109 OPT_BOOL(0, "check", &state->check,
5110 N_("instead of applying the patch, see if the patch is applicable")),
5111 OPT_BOOL(0, "index", &state->check_index,
5112 N_("make sure the patch is applicable to the current index")),
5113 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5114 N_("mark new files with `git add --intent-to-add`")),
5115 OPT_BOOL(0, "cached", &state->cached,
5116 N_("apply a patch without touching the working tree")),
5117 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5118 N_("accept a patch that touches outside the working area"),
5119 PARSE_OPT_NOCOMPLETE),
5120 OPT_BOOL(0, "apply", force_apply,
5121 N_("also apply the patch (use with --stat/--summary/--check)")),
5122 OPT_BOOL('3', "3way", &state->threeway,
5123 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5124 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5125 N_("build a temporary index based on embedded index information")),
5126 /* Think twice before adding "--nul" synonym to this */
5127 OPT_SET_INT('z', NULL, &state->line_termination,
5128 N_("paths are separated with NUL character"), '\0'),
5129 OPT_INTEGER('C', NULL, &state->p_context,
5130 N_("ensure at least <n> lines of context match")),
5131 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5132 N_("detect new or modified lines that have whitespace errors"),
5133 apply_option_parse_whitespace),
5134 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5135 N_("ignore changes in whitespace when finding context"),
5136 PARSE_OPT_NOARG, apply_option_parse_space_change),
5137 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5138 N_("ignore changes in whitespace when finding context"),
5139 PARSE_OPT_NOARG, apply_option_parse_space_change),
5140 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5141 N_("apply the patch in reverse")),
5142 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5143 N_("don't expect at least one line of context")),
5144 OPT_BOOL(0, "reject", &state->apply_with_reject,
5145 N_("leave the rejected hunks in corresponding *.rej files")),
5146 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5147 N_("allow overlapping hunks")),
5148 OPT__VERBOSITY(&state->apply_verbosity),
5149 OPT_BIT(0, "inaccurate-eof", options,
5150 N_("tolerate incorrectly detected missing new-line at the end of file"),
5151 APPLY_OPT_INACCURATE_EOF),
5152 OPT_BIT(0, "recount", options,
5153 N_("do not trust the line counts in the hunk headers"),
5154 APPLY_OPT_RECOUNT),
5155 OPT_CALLBACK(0, "directory", state, N_("root"),
5156 N_("prepend <root> to all filenames"),
5157 apply_option_parse_directory),
5158 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5159 N_("don't return error for empty patches")),
5160 OPT_END()
5163 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);