git-compat-util: convert skip_{prefix,suffix}{,_mem} to bool
[git/gitster.git] / apply.c
blob3d69fec836d41fbae10988606ac20ce96dad562d
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 "blob.h"
16 #include "delta.h"
17 #include "diff.h"
18 #include "dir.h"
19 #include "environment.h"
20 #include "gettext.h"
21 #include "hex.h"
22 #include "xdiff-interface.h"
23 #include "merge-ll.h"
24 #include "lockfile.h"
25 #include "name-hash.h"
26 #include "object-name.h"
27 #include "object-file.h"
28 #include "parse-options.h"
29 #include "path.h"
30 #include "quote.h"
31 #include "read-cache.h"
32 #include "rerere.h"
33 #include "apply.h"
34 #include "entry.h"
35 #include "setup.h"
36 #include "symlinks.h"
37 #include "wildmatch.h"
38 #include "ws.h"
40 struct gitdiff_data {
41 struct strbuf *root;
42 int linenr;
43 int p_value;
46 static void git_apply_config(void)
48 git_config_get_string("apply.whitespace", &apply_default_whitespace);
49 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
50 git_config(git_xmerge_config, NULL);
53 static int parse_whitespace_option(struct apply_state *state, const char *option)
55 if (!option) {
56 state->ws_error_action = warn_on_ws_error;
57 return 0;
59 if (!strcmp(option, "warn")) {
60 state->ws_error_action = warn_on_ws_error;
61 return 0;
63 if (!strcmp(option, "nowarn")) {
64 state->ws_error_action = nowarn_ws_error;
65 return 0;
67 if (!strcmp(option, "error")) {
68 state->ws_error_action = die_on_ws_error;
69 return 0;
71 if (!strcmp(option, "error-all")) {
72 state->ws_error_action = die_on_ws_error;
73 state->squelch_whitespace_errors = 0;
74 return 0;
76 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
77 state->ws_error_action = correct_ws_error;
78 return 0;
81 * Please update $__git_whitespacelist in git-completion.bash
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));
1295 if (!second)
1296 return NULL;
1298 * Does len bytes starting at "name" and "second"
1299 * (that are separated by one HT or SP we just
1300 * found) exactly match?
1302 if (second[len] == '\n' && !strncmp(name, second, len))
1303 return xmemdupz(name, len);
1308 static int check_header_line(int linenr, struct patch *patch)
1310 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1311 (patch->is_rename == 1) + (patch->is_copy == 1);
1312 if (extensions > 1)
1313 return error(_("inconsistent header lines %d and %d"),
1314 patch->extension_linenr, linenr);
1315 if (extensions && !patch->extension_linenr)
1316 patch->extension_linenr = linenr;
1317 return 0;
1320 int parse_git_diff_header(struct strbuf *root,
1321 int *linenr,
1322 int p_value,
1323 const char *line,
1324 int len,
1325 unsigned int size,
1326 struct patch *patch)
1328 unsigned long offset;
1329 struct gitdiff_data parse_hdr_state;
1331 /* A git diff has explicit new/delete information, so we don't guess */
1332 patch->is_new = 0;
1333 patch->is_delete = 0;
1336 * Some things may not have the old name in the
1337 * rest of the headers anywhere (pure mode changes,
1338 * or removing or adding empty files), so we get
1339 * the default name from the header.
1341 patch->def_name = git_header_name(p_value, line, len);
1342 if (patch->def_name && root->len) {
1343 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1344 free(patch->def_name);
1345 patch->def_name = s;
1348 line += len;
1349 size -= len;
1350 (*linenr)++;
1351 parse_hdr_state.root = root;
1352 parse_hdr_state.linenr = *linenr;
1353 parse_hdr_state.p_value = p_value;
1355 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1356 static const struct opentry {
1357 const char *str;
1358 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1359 } optable[] = {
1360 { "@@ -", gitdiff_hdrend },
1361 { "--- ", gitdiff_oldname },
1362 { "+++ ", gitdiff_newname },
1363 { "old mode ", gitdiff_oldmode },
1364 { "new mode ", gitdiff_newmode },
1365 { "deleted file mode ", gitdiff_delete },
1366 { "new file mode ", gitdiff_newfile },
1367 { "copy from ", gitdiff_copysrc },
1368 { "copy to ", gitdiff_copydst },
1369 { "rename old ", gitdiff_renamesrc },
1370 { "rename new ", gitdiff_renamedst },
1371 { "rename from ", gitdiff_renamesrc },
1372 { "rename to ", gitdiff_renamedst },
1373 { "similarity index ", gitdiff_similarity },
1374 { "dissimilarity index ", gitdiff_dissimilarity },
1375 { "index ", gitdiff_index },
1376 { "", gitdiff_unrecognized },
1378 int i;
1380 len = linelen(line, size);
1381 if (!len || line[len-1] != '\n')
1382 break;
1383 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1384 const struct opentry *p = optable + i;
1385 int oplen = strlen(p->str);
1386 int res;
1387 if (len < oplen || memcmp(p->str, line, oplen))
1388 continue;
1389 res = p->fn(&parse_hdr_state, line + oplen, patch);
1390 if (res < 0)
1391 return -1;
1392 if (check_header_line(*linenr, patch))
1393 return -1;
1394 if (res > 0)
1395 goto done;
1396 break;
1400 done:
1401 if (!patch->old_name && !patch->new_name) {
1402 if (!patch->def_name) {
1403 error(Q_("git diff header lacks filename information when removing "
1404 "%d leading pathname component (line %d)",
1405 "git diff header lacks filename information when removing "
1406 "%d leading pathname components (line %d)",
1407 parse_hdr_state.p_value),
1408 parse_hdr_state.p_value, *linenr);
1409 return -128;
1411 patch->old_name = xstrdup(patch->def_name);
1412 patch->new_name = xstrdup(patch->def_name);
1414 if ((!patch->new_name && !patch->is_delete) ||
1415 (!patch->old_name && !patch->is_new)) {
1416 error(_("git diff header lacks filename information "
1417 "(line %d)"), *linenr);
1418 return -128;
1420 patch->is_toplevel_relative = 1;
1421 return offset;
1424 static int parse_num(const char *line, unsigned long *p)
1426 char *ptr;
1428 if (!isdigit(*line))
1429 return 0;
1430 *p = strtoul(line, &ptr, 10);
1431 return ptr - line;
1434 static int parse_range(const char *line, int len, int offset, const char *expect,
1435 unsigned long *p1, unsigned long *p2)
1437 int digits, ex;
1439 if (offset < 0 || offset >= len)
1440 return -1;
1441 line += offset;
1442 len -= offset;
1444 digits = parse_num(line, p1);
1445 if (!digits)
1446 return -1;
1448 offset += digits;
1449 line += digits;
1450 len -= digits;
1452 *p2 = 1;
1453 if (*line == ',') {
1454 digits = parse_num(line+1, p2);
1455 if (!digits)
1456 return -1;
1458 offset += digits+1;
1459 line += digits+1;
1460 len -= digits+1;
1463 ex = strlen(expect);
1464 if (ex > len)
1465 return -1;
1466 if (memcmp(line, expect, ex))
1467 return -1;
1469 return offset + ex;
1472 static void recount_diff(const char *line, int size, struct fragment *fragment)
1474 int oldlines = 0, newlines = 0, ret = 0;
1476 if (size < 1) {
1477 warning("recount: ignore empty hunk");
1478 return;
1481 for (;;) {
1482 int len = linelen(line, size);
1483 size -= len;
1484 line += len;
1486 if (size < 1)
1487 break;
1489 switch (*line) {
1490 case ' ': case '\n':
1491 newlines++;
1492 /* fall through */
1493 case '-':
1494 oldlines++;
1495 continue;
1496 case '+':
1497 newlines++;
1498 continue;
1499 case '\\':
1500 continue;
1501 case '@':
1502 ret = size < 3 || !starts_with(line, "@@ ");
1503 break;
1504 case 'd':
1505 ret = size < 5 || !starts_with(line, "diff ");
1506 break;
1507 default:
1508 ret = -1;
1509 break;
1511 if (ret) {
1512 warning(_("recount: unexpected line: %.*s"),
1513 (int)linelen(line, size), line);
1514 return;
1516 break;
1518 fragment->oldlines = oldlines;
1519 fragment->newlines = newlines;
1523 * Parse a unified diff fragment header of the
1524 * form "@@ -a,b +c,d @@"
1526 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1528 int offset;
1530 if (!len || line[len-1] != '\n')
1531 return -1;
1533 /* Figure out the number of lines in a fragment */
1534 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1535 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1537 return offset;
1541 * Find file diff header
1543 * Returns:
1544 * -1 if no header was found
1545 * -128 in case of error
1546 * the size of the header in bytes (called "offset") otherwise
1548 static int find_header(struct apply_state *state,
1549 const char *line,
1550 unsigned long size,
1551 int *hdrsize,
1552 struct patch *patch)
1554 unsigned long offset, len;
1556 patch->is_toplevel_relative = 0;
1557 patch->is_rename = patch->is_copy = 0;
1558 patch->is_new = patch->is_delete = -1;
1559 patch->old_mode = patch->new_mode = 0;
1560 patch->old_name = patch->new_name = NULL;
1561 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1562 unsigned long nextlen;
1564 len = linelen(line, size);
1565 if (!len)
1566 break;
1568 /* Testing this early allows us to take a few shortcuts.. */
1569 if (len < 6)
1570 continue;
1573 * Make sure we don't find any unconnected patch fragments.
1574 * That's a sign that we didn't find a header, and that a
1575 * patch has become corrupted/broken up.
1577 if (!memcmp("@@ -", line, 4)) {
1578 struct fragment dummy;
1579 if (parse_fragment_header(line, len, &dummy) < 0)
1580 continue;
1581 error(_("patch fragment without header at line %d: %.*s"),
1582 state->linenr, (int)len-1, line);
1583 return -128;
1586 if (size < len + 6)
1587 break;
1590 * Git patch? It might not have a real patch, just a rename
1591 * or mode change, so we handle that specially
1593 if (!memcmp("diff --git ", line, 11)) {
1594 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1595 state->p_value, line, len,
1596 size, patch);
1597 if (git_hdr_len < 0)
1598 return -128;
1599 if (git_hdr_len <= len)
1600 continue;
1601 *hdrsize = git_hdr_len;
1602 return offset;
1605 /* --- followed by +++ ? */
1606 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1607 continue;
1610 * We only accept unified patches, so we want it to
1611 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1612 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1614 nextlen = linelen(line + len, size - len);
1615 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1616 continue;
1618 /* Ok, we'll consider it a patch */
1619 if (parse_traditional_patch(state, line, line+len, patch))
1620 return -128;
1621 *hdrsize = len + nextlen;
1622 state->linenr += 2;
1623 return offset;
1625 return -1;
1628 static void record_ws_error(struct apply_state *state,
1629 unsigned result,
1630 const char *line,
1631 int len,
1632 int linenr)
1634 char *err;
1636 if (!result)
1637 return;
1639 state->whitespace_error++;
1640 if (state->squelch_whitespace_errors &&
1641 state->squelch_whitespace_errors < state->whitespace_error)
1642 return;
1644 err = whitespace_error_string(result);
1645 if (state->apply_verbosity > verbosity_silent)
1646 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1647 state->patch_input_file, linenr, err, len, line);
1648 free(err);
1651 static void check_whitespace(struct apply_state *state,
1652 const char *line,
1653 int len,
1654 unsigned ws_rule)
1656 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1658 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1662 * Check if the patch has context lines with CRLF or
1663 * the patch wants to remove lines with CRLF.
1665 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1667 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1668 patch->ws_rule |= WS_CR_AT_EOL;
1669 patch->crlf_in_old = 1;
1675 * Parse a unified diff. Note that this really needs to parse each
1676 * fragment separately, since the only way to know the difference
1677 * between a "---" that is part of a patch, and a "---" that starts
1678 * the next patch is to look at the line counts..
1680 static int parse_fragment(struct apply_state *state,
1681 const char *line,
1682 unsigned long size,
1683 struct patch *patch,
1684 struct fragment *fragment)
1686 int added, deleted;
1687 int len = linelen(line, size), offset;
1688 unsigned long oldlines, newlines;
1689 unsigned long leading, trailing;
1691 offset = parse_fragment_header(line, len, fragment);
1692 if (offset < 0)
1693 return -1;
1694 if (offset > 0 && patch->recount)
1695 recount_diff(line + offset, size - offset, fragment);
1696 oldlines = fragment->oldlines;
1697 newlines = fragment->newlines;
1698 leading = 0;
1699 trailing = 0;
1701 /* Parse the thing.. */
1702 line += len;
1703 size -= len;
1704 state->linenr++;
1705 added = deleted = 0;
1706 for (offset = len;
1707 0 < size;
1708 offset += len, size -= len, line += len, state->linenr++) {
1709 if (!oldlines && !newlines)
1710 break;
1711 len = linelen(line, size);
1712 if (!len || line[len-1] != '\n')
1713 return -1;
1714 switch (*line) {
1715 default:
1716 return -1;
1717 case '\n': /* newer GNU diff, an empty context line */
1718 case ' ':
1719 oldlines--;
1720 newlines--;
1721 if (!deleted && !added)
1722 leading++;
1723 trailing++;
1724 check_old_for_crlf(patch, line, len);
1725 if (!state->apply_in_reverse &&
1726 state->ws_error_action == correct_ws_error)
1727 check_whitespace(state, line, len, patch->ws_rule);
1728 break;
1729 case '-':
1730 if (!state->apply_in_reverse)
1731 check_old_for_crlf(patch, line, len);
1732 if (state->apply_in_reverse &&
1733 state->ws_error_action != nowarn_ws_error)
1734 check_whitespace(state, line, len, patch->ws_rule);
1735 deleted++;
1736 oldlines--;
1737 trailing = 0;
1738 break;
1739 case '+':
1740 if (state->apply_in_reverse)
1741 check_old_for_crlf(patch, line, len);
1742 if (!state->apply_in_reverse &&
1743 state->ws_error_action != nowarn_ws_error)
1744 check_whitespace(state, line, len, patch->ws_rule);
1745 added++;
1746 newlines--;
1747 trailing = 0;
1748 break;
1751 * We allow "\ No newline at end of file". Depending
1752 * on locale settings when the patch was produced we
1753 * don't know what this line looks like. The only
1754 * thing we do know is that it begins with "\ ".
1755 * Checking for 12 is just for sanity check -- any
1756 * l10n of "\ No newline..." is at least that long.
1758 case '\\':
1759 if (len < 12 || memcmp(line, "\\ ", 2))
1760 return -1;
1761 break;
1764 if (oldlines || newlines)
1765 return -1;
1766 if (!patch->recount && !deleted && !added)
1767 return -1;
1769 fragment->leading = leading;
1770 fragment->trailing = trailing;
1773 * If a fragment ends with an incomplete line, we failed to include
1774 * it in the above loop because we hit oldlines == newlines == 0
1775 * before seeing it.
1777 if (12 < size && !memcmp(line, "\\ ", 2))
1778 offset += linelen(line, size);
1780 patch->lines_added += added;
1781 patch->lines_deleted += deleted;
1783 if (0 < patch->is_new && oldlines)
1784 return error(_("new file depends on old contents"));
1785 if (0 < patch->is_delete && newlines)
1786 return error(_("deleted file still has contents"));
1787 return offset;
1791 * We have seen "diff --git a/... b/..." header (or a traditional patch
1792 * header). Read hunks that belong to this patch into fragments and hang
1793 * them to the given patch structure.
1795 * The (fragment->patch, fragment->size) pair points into the memory given
1796 * by the caller, not a copy, when we return.
1798 * Returns:
1799 * -1 in case of error,
1800 * the number of bytes in the patch otherwise.
1802 static int parse_single_patch(struct apply_state *state,
1803 const char *line,
1804 unsigned long size,
1805 struct patch *patch)
1807 unsigned long offset = 0;
1808 unsigned long oldlines = 0, newlines = 0, context = 0;
1809 struct fragment **fragp = &patch->fragments;
1811 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1812 struct fragment *fragment;
1813 int len;
1815 CALLOC_ARRAY(fragment, 1);
1816 fragment->linenr = state->linenr;
1817 len = parse_fragment(state, line, size, patch, fragment);
1818 if (len <= 0) {
1819 free(fragment);
1820 return error(_("corrupt patch at line %d"), state->linenr);
1822 fragment->patch = line;
1823 fragment->size = len;
1824 oldlines += fragment->oldlines;
1825 newlines += fragment->newlines;
1826 context += fragment->leading + fragment->trailing;
1828 *fragp = fragment;
1829 fragp = &fragment->next;
1831 offset += len;
1832 line += len;
1833 size -= len;
1837 * If something was removed (i.e. we have old-lines) it cannot
1838 * be creation, and if something was added it cannot be
1839 * deletion. However, the reverse is not true; --unified=0
1840 * patches that only add are not necessarily creation even
1841 * though they do not have any old lines, and ones that only
1842 * delete are not necessarily deletion.
1844 * Unfortunately, a real creation/deletion patch do _not_ have
1845 * any context line by definition, so we cannot safely tell it
1846 * apart with --unified=0 insanity. At least if the patch has
1847 * more than one hunk it is not creation or deletion.
1849 if (patch->is_new < 0 &&
1850 (oldlines || (patch->fragments && patch->fragments->next)))
1851 patch->is_new = 0;
1852 if (patch->is_delete < 0 &&
1853 (newlines || (patch->fragments && patch->fragments->next)))
1854 patch->is_delete = 0;
1856 if (0 < patch->is_new && oldlines)
1857 return error(_("new file %s depends on old contents"), patch->new_name);
1858 if (0 < patch->is_delete && newlines)
1859 return error(_("deleted file %s still has contents"), patch->old_name);
1860 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1861 fprintf_ln(stderr,
1862 _("** warning: "
1863 "file %s becomes empty but is not deleted"),
1864 patch->new_name);
1866 return offset;
1869 static inline int metadata_changes(struct patch *patch)
1871 return patch->is_rename > 0 ||
1872 patch->is_copy > 0 ||
1873 patch->is_new > 0 ||
1874 patch->is_delete ||
1875 (patch->old_mode && patch->new_mode &&
1876 patch->old_mode != patch->new_mode);
1879 static char *inflate_it(const void *data, unsigned long size,
1880 unsigned long inflated_size)
1882 git_zstream stream;
1883 void *out;
1884 int st;
1886 memset(&stream, 0, sizeof(stream));
1888 stream.next_in = (unsigned char *)data;
1889 stream.avail_in = size;
1890 stream.next_out = out = xmalloc(inflated_size);
1891 stream.avail_out = inflated_size;
1892 git_inflate_init(&stream);
1893 st = git_inflate(&stream, Z_FINISH);
1894 git_inflate_end(&stream);
1895 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1896 free(out);
1897 return NULL;
1899 return out;
1903 * Read a binary hunk and return a new fragment; fragment->patch
1904 * points at an allocated memory that the caller must free, so
1905 * it is marked as "->free_patch = 1".
1907 static struct fragment *parse_binary_hunk(struct apply_state *state,
1908 char **buf_p,
1909 unsigned long *sz_p,
1910 int *status_p,
1911 int *used_p)
1914 * Expect a line that begins with binary patch method ("literal"
1915 * or "delta"), followed by the length of data before deflating.
1916 * a sequence of 'length-byte' followed by base-85 encoded data
1917 * should follow, terminated by a newline.
1919 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1920 * and we would limit the patch line to 66 characters,
1921 * so one line can fit up to 13 groups that would decode
1922 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1923 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1925 int llen, used;
1926 unsigned long size = *sz_p;
1927 char *buffer = *buf_p;
1928 int patch_method;
1929 unsigned long origlen;
1930 char *data = NULL;
1931 int hunk_size = 0;
1932 struct fragment *frag;
1934 llen = linelen(buffer, size);
1935 used = llen;
1937 *status_p = 0;
1939 if (starts_with(buffer, "delta ")) {
1940 patch_method = BINARY_DELTA_DEFLATED;
1941 origlen = strtoul(buffer + 6, NULL, 10);
1943 else if (starts_with(buffer, "literal ")) {
1944 patch_method = BINARY_LITERAL_DEFLATED;
1945 origlen = strtoul(buffer + 8, NULL, 10);
1947 else
1948 return NULL;
1950 state->linenr++;
1951 buffer += llen;
1952 size -= llen;
1953 while (1) {
1954 int byte_length, max_byte_length, newsize;
1955 llen = linelen(buffer, size);
1956 used += llen;
1957 state->linenr++;
1958 if (llen == 1) {
1959 /* consume the blank line */
1960 buffer++;
1961 size--;
1962 break;
1965 * Minimum line is "A00000\n" which is 7-byte long,
1966 * and the line length must be multiple of 5 plus 2.
1968 if ((llen < 7) || (llen-2) % 5)
1969 goto corrupt;
1970 max_byte_length = (llen - 2) / 5 * 4;
1971 byte_length = *buffer;
1972 if ('A' <= byte_length && byte_length <= 'Z')
1973 byte_length = byte_length - 'A' + 1;
1974 else if ('a' <= byte_length && byte_length <= 'z')
1975 byte_length = byte_length - 'a' + 27;
1976 else
1977 goto corrupt;
1978 /* if the input length was not multiple of 4, we would
1979 * have filler at the end but the filler should never
1980 * exceed 3 bytes
1982 if (max_byte_length < byte_length ||
1983 byte_length <= max_byte_length - 4)
1984 goto corrupt;
1985 newsize = hunk_size + byte_length;
1986 data = xrealloc(data, newsize);
1987 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1988 goto corrupt;
1989 hunk_size = newsize;
1990 buffer += llen;
1991 size -= llen;
1994 CALLOC_ARRAY(frag, 1);
1995 frag->patch = inflate_it(data, hunk_size, origlen);
1996 frag->free_patch = 1;
1997 if (!frag->patch)
1998 goto corrupt;
1999 free(data);
2000 frag->size = origlen;
2001 *buf_p = buffer;
2002 *sz_p = size;
2003 *used_p = used;
2004 frag->binary_patch_method = patch_method;
2005 return frag;
2007 corrupt:
2008 free(data);
2009 *status_p = -1;
2010 error(_("corrupt binary patch at line %d: %.*s"),
2011 state->linenr-1, llen-1, buffer);
2012 return NULL;
2016 * Returns:
2017 * -1 in case of error,
2018 * the length of the parsed binary patch otherwise
2020 static int parse_binary(struct apply_state *state,
2021 char *buffer,
2022 unsigned long size,
2023 struct patch *patch)
2026 * We have read "GIT binary patch\n"; what follows is a line
2027 * that says the patch method (currently, either "literal" or
2028 * "delta") and the length of data before deflating; a
2029 * sequence of 'length-byte' followed by base-85 encoded data
2030 * follows.
2032 * When a binary patch is reversible, there is another binary
2033 * hunk in the same format, starting with patch method (either
2034 * "literal" or "delta") with the length of data, and a sequence
2035 * of length-byte + base-85 encoded data, terminated with another
2036 * empty line. This data, when applied to the postimage, produces
2037 * the preimage.
2039 struct fragment *forward;
2040 struct fragment *reverse;
2041 int status;
2042 int used, used_1;
2044 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2045 if (!forward && !status)
2046 /* there has to be one hunk (forward hunk) */
2047 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2048 if (status)
2049 /* otherwise we already gave an error message */
2050 return status;
2052 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2053 if (reverse)
2054 used += used_1;
2055 else if (status) {
2057 * Not having reverse hunk is not an error, but having
2058 * a corrupt reverse hunk is.
2060 free((void*) forward->patch);
2061 free(forward);
2062 return status;
2064 forward->next = reverse;
2065 patch->fragments = forward;
2066 patch->is_binary = 1;
2067 return used;
2070 static void prefix_one(struct apply_state *state, char **name)
2072 char *old_name = *name;
2073 if (!old_name)
2074 return;
2075 *name = prefix_filename(state->prefix, *name);
2076 free(old_name);
2079 static void prefix_patch(struct apply_state *state, struct patch *p)
2081 if (!state->prefix || p->is_toplevel_relative)
2082 return;
2083 prefix_one(state, &p->new_name);
2084 prefix_one(state, &p->old_name);
2088 * include/exclude
2091 static void add_name_limit(struct apply_state *state,
2092 const char *name,
2093 int exclude)
2095 struct string_list_item *it;
2097 it = string_list_append(&state->limit_by_name, name);
2098 it->util = exclude ? NULL : (void *) 1;
2101 static int use_patch(struct apply_state *state, struct patch *p)
2103 const char *pathname = p->new_name ? p->new_name : p->old_name;
2104 int i;
2106 /* Paths outside are not touched regardless of "--include" */
2107 if (state->prefix && *state->prefix) {
2108 const char *rest;
2109 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2110 return 0;
2113 /* See if it matches any of exclude/include rule */
2114 for (i = 0; i < state->limit_by_name.nr; i++) {
2115 struct string_list_item *it = &state->limit_by_name.items[i];
2116 if (!wildmatch(it->string, pathname, 0))
2117 return (it->util != NULL);
2121 * If we had any include, a path that does not match any rule is
2122 * not used. Otherwise, we saw bunch of exclude rules (or none)
2123 * and such a path is used.
2125 return !state->has_include;
2129 * Read the patch text in "buffer" that extends for "size" bytes; stop
2130 * reading after seeing a single patch (i.e. changes to a single file).
2131 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2133 * Returns:
2134 * -1 if no header was found or parse_binary() failed,
2135 * -128 on another error,
2136 * the number of bytes consumed otherwise,
2137 * so that the caller can call us again for the next patch.
2139 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2141 int hdrsize, patchsize;
2142 int offset = find_header(state, buffer, size, &hdrsize, patch);
2144 if (offset < 0)
2145 return offset;
2147 prefix_patch(state, patch);
2149 if (!use_patch(state, patch))
2150 patch->ws_rule = 0;
2151 else if (patch->new_name)
2152 patch->ws_rule = whitespace_rule(state->repo->index,
2153 patch->new_name);
2154 else
2155 patch->ws_rule = whitespace_rule(state->repo->index,
2156 patch->old_name);
2158 patchsize = parse_single_patch(state,
2159 buffer + offset + hdrsize,
2160 size - offset - hdrsize,
2161 patch);
2163 if (patchsize < 0)
2164 return -128;
2166 if (!patchsize) {
2167 static const char git_binary[] = "GIT binary patch\n";
2168 int hd = hdrsize + offset;
2169 unsigned long llen = linelen(buffer + hd, size - hd);
2171 if (llen == sizeof(git_binary) - 1 &&
2172 !memcmp(git_binary, buffer + hd, llen)) {
2173 int used;
2174 state->linenr++;
2175 used = parse_binary(state, buffer + hd + llen,
2176 size - hd - llen, patch);
2177 if (used < 0)
2178 return -1;
2179 if (used)
2180 patchsize = used + llen;
2181 else
2182 patchsize = 0;
2184 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2185 static const char *binhdr[] = {
2186 "Binary files ",
2187 "Files ",
2188 NULL,
2190 int i;
2191 for (i = 0; binhdr[i]; i++) {
2192 int len = strlen(binhdr[i]);
2193 if (len < size - hd &&
2194 !memcmp(binhdr[i], buffer + hd, len)) {
2195 state->linenr++;
2196 patch->is_binary = 1;
2197 patchsize = llen;
2198 break;
2203 /* Empty patch cannot be applied if it is a text patch
2204 * without metadata change. A binary patch appears
2205 * empty to us here.
2207 if ((state->apply || state->check) &&
2208 (!patch->is_binary && !metadata_changes(patch))) {
2209 error(_("patch with only garbage at line %d"), state->linenr);
2210 return -128;
2214 return offset + hdrsize + patchsize;
2217 static void reverse_patches(struct patch *p)
2219 for (; p; p = p->next) {
2220 struct fragment *frag = p->fragments;
2222 SWAP(p->new_name, p->old_name);
2223 SWAP(p->new_mode, p->old_mode);
2224 SWAP(p->is_new, p->is_delete);
2225 SWAP(p->lines_added, p->lines_deleted);
2226 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2228 for (; frag; frag = frag->next) {
2229 SWAP(frag->newpos, frag->oldpos);
2230 SWAP(frag->newlines, frag->oldlines);
2235 static const char pluses[] =
2236 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2237 static const char minuses[]=
2238 "----------------------------------------------------------------------";
2240 static void show_stats(struct apply_state *state, struct patch *patch)
2242 struct strbuf qname = STRBUF_INIT;
2243 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2244 int max, add, del;
2246 quote_c_style(cp, &qname, NULL, 0);
2249 * "scale" the filename
2251 max = state->max_len;
2252 if (max > 50)
2253 max = 50;
2255 if (qname.len > max) {
2256 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2257 if (!cp)
2258 cp = qname.buf + qname.len + 3 - max;
2259 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2262 if (patch->is_binary) {
2263 printf(" %-*s | Bin\n", max, qname.buf);
2264 strbuf_release(&qname);
2265 return;
2268 printf(" %-*s |", max, qname.buf);
2269 strbuf_release(&qname);
2272 * scale the add/delete
2274 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2275 add = patch->lines_added;
2276 del = patch->lines_deleted;
2278 if (state->max_change > 0) {
2279 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2280 add = (add * max + state->max_change / 2) / state->max_change;
2281 del = total - add;
2283 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2284 add, pluses, del, minuses);
2287 static int read_old_data(struct stat *st, struct patch *patch,
2288 const char *path, struct strbuf *buf)
2290 int conv_flags = patch->crlf_in_old ?
2291 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2292 switch (st->st_mode & S_IFMT) {
2293 case S_IFLNK:
2294 if (strbuf_readlink(buf, path, st->st_size) < 0)
2295 return error(_("unable to read symlink %s"), path);
2296 return 0;
2297 case S_IFREG:
2298 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2299 return error(_("unable to open or read %s"), path);
2301 * "git apply" without "--index/--cached" should never look
2302 * at the index; the target file may not have been added to
2303 * the index yet, and we may not even be in any Git repository.
2304 * Pass NULL to convert_to_git() to stress this; the function
2305 * should never look at the index when explicit crlf option
2306 * is given.
2308 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2309 return 0;
2310 default:
2311 return -1;
2316 * Update the preimage, and the common lines in postimage,
2317 * from buffer buf of length len. If postlen is 0 the postimage
2318 * is updated in place, otherwise it's updated on a new buffer
2319 * of length postlen
2322 static void update_pre_post_images(struct image *preimage,
2323 struct image *postimage,
2324 char *buf,
2325 size_t len, size_t postlen)
2327 int i, ctx, reduced;
2328 char *new_buf, *old_buf, *fixed;
2329 struct image fixed_preimage;
2332 * Update the preimage with whitespace fixes. Note that we
2333 * are not losing preimage->buf -- apply_one_fragment() will
2334 * free "oldlines".
2336 prepare_image(&fixed_preimage, buf, len, 1);
2337 assert(postlen
2338 ? fixed_preimage.nr == preimage->nr
2339 : fixed_preimage.nr <= preimage->nr);
2340 for (i = 0; i < fixed_preimage.nr; i++)
2341 fixed_preimage.line[i].flag = preimage->line[i].flag;
2342 free(preimage->line_allocated);
2343 *preimage = fixed_preimage;
2346 * Adjust the common context lines in postimage. This can be
2347 * done in-place when we are shrinking it with whitespace
2348 * fixing, but needs a new buffer when ignoring whitespace or
2349 * expanding leading tabs to spaces.
2351 * We trust the caller to tell us if the update can be done
2352 * in place (postlen==0) or not.
2354 old_buf = postimage->buf;
2355 if (postlen)
2356 new_buf = postimage->buf = xmalloc(postlen);
2357 else
2358 new_buf = old_buf;
2359 fixed = preimage->buf;
2361 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2362 size_t l_len = postimage->line[i].len;
2363 if (!(postimage->line[i].flag & LINE_COMMON)) {
2364 /* an added line -- no counterparts in preimage */
2365 memmove(new_buf, old_buf, l_len);
2366 old_buf += l_len;
2367 new_buf += l_len;
2368 continue;
2371 /* a common context -- skip it in the original postimage */
2372 old_buf += l_len;
2374 /* and find the corresponding one in the fixed preimage */
2375 while (ctx < preimage->nr &&
2376 !(preimage->line[ctx].flag & LINE_COMMON)) {
2377 fixed += preimage->line[ctx].len;
2378 ctx++;
2382 * preimage is expected to run out, if the caller
2383 * fixed addition of trailing blank lines.
2385 if (preimage->nr <= ctx) {
2386 reduced++;
2387 continue;
2390 /* and copy it in, while fixing the line length */
2391 l_len = preimage->line[ctx].len;
2392 memcpy(new_buf, fixed, l_len);
2393 new_buf += l_len;
2394 fixed += l_len;
2395 postimage->line[i].len = l_len;
2396 ctx++;
2399 if (postlen
2400 ? postlen < new_buf - postimage->buf
2401 : postimage->len < new_buf - postimage->buf)
2402 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2403 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2405 /* Fix the length of the whole thing */
2406 postimage->len = new_buf - postimage->buf;
2407 postimage->nr -= reduced;
2410 static int line_by_line_fuzzy_match(struct image *img,
2411 struct image *preimage,
2412 struct image *postimage,
2413 unsigned long current,
2414 int current_lno,
2415 int preimage_limit)
2417 int i;
2418 size_t imgoff = 0;
2419 size_t preoff = 0;
2420 size_t postlen = postimage->len;
2421 size_t extra_chars;
2422 char *buf;
2423 char *preimage_eof;
2424 char *preimage_end;
2425 struct strbuf fixed;
2426 char *fixed_buf;
2427 size_t fixed_len;
2429 for (i = 0; i < preimage_limit; i++) {
2430 size_t prelen = preimage->line[i].len;
2431 size_t imglen = img->line[current_lno+i].len;
2433 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2434 preimage->buf + preoff, prelen))
2435 return 0;
2436 if (preimage->line[i].flag & LINE_COMMON)
2437 postlen += imglen - prelen;
2438 imgoff += imglen;
2439 preoff += prelen;
2443 * Ok, the preimage matches with whitespace fuzz.
2445 * imgoff now holds the true length of the target that
2446 * matches the preimage before the end of the file.
2448 * Count the number of characters in the preimage that fall
2449 * beyond the end of the file and make sure that all of them
2450 * are whitespace characters. (This can only happen if
2451 * we are removing blank lines at the end of the file.)
2453 buf = preimage_eof = preimage->buf + preoff;
2454 for ( ; i < preimage->nr; i++)
2455 preoff += preimage->line[i].len;
2456 preimage_end = preimage->buf + preoff;
2457 for ( ; buf < preimage_end; buf++)
2458 if (!isspace(*buf))
2459 return 0;
2462 * Update the preimage and the common postimage context
2463 * lines to use the same whitespace as the target.
2464 * If whitespace is missing in the target (i.e.
2465 * if the preimage extends beyond the end of the file),
2466 * use the whitespace from the preimage.
2468 extra_chars = preimage_end - preimage_eof;
2469 strbuf_init(&fixed, imgoff + extra_chars);
2470 strbuf_add(&fixed, img->buf + current, imgoff);
2471 strbuf_add(&fixed, preimage_eof, extra_chars);
2472 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2473 update_pre_post_images(preimage, postimage,
2474 fixed_buf, fixed_len, postlen);
2475 return 1;
2478 static int match_fragment(struct apply_state *state,
2479 struct image *img,
2480 struct image *preimage,
2481 struct image *postimage,
2482 unsigned long current,
2483 int current_lno,
2484 unsigned ws_rule,
2485 int match_beginning, int match_end)
2487 int i;
2488 char *fixed_buf, *buf, *orig, *target;
2489 struct strbuf fixed;
2490 size_t fixed_len, postlen;
2491 int preimage_limit;
2493 if (preimage->nr + current_lno <= img->nr) {
2495 * The hunk falls within the boundaries of img.
2497 preimage_limit = preimage->nr;
2498 if (match_end && (preimage->nr + current_lno != img->nr))
2499 return 0;
2500 } else if (state->ws_error_action == correct_ws_error &&
2501 (ws_rule & WS_BLANK_AT_EOF)) {
2503 * This hunk extends beyond the end of img, and we are
2504 * removing blank lines at the end of the file. This
2505 * many lines from the beginning of the preimage must
2506 * match with img, and the remainder of the preimage
2507 * must be blank.
2509 preimage_limit = img->nr - current_lno;
2510 } else {
2512 * The hunk extends beyond the end of the img and
2513 * we are not removing blanks at the end, so we
2514 * should reject the hunk at this position.
2516 return 0;
2519 if (match_beginning && current_lno)
2520 return 0;
2522 /* Quick hash check */
2523 for (i = 0; i < preimage_limit; i++)
2524 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2525 (preimage->line[i].hash != img->line[current_lno + i].hash))
2526 return 0;
2528 if (preimage_limit == preimage->nr) {
2530 * Do we have an exact match? If we were told to match
2531 * at the end, size must be exactly at current+fragsize,
2532 * otherwise current+fragsize must be still within the preimage,
2533 * and either case, the old piece should match the preimage
2534 * exactly.
2536 if ((match_end
2537 ? (current + preimage->len == img->len)
2538 : (current + preimage->len <= img->len)) &&
2539 !memcmp(img->buf + current, preimage->buf, preimage->len))
2540 return 1;
2541 } else {
2543 * The preimage extends beyond the end of img, so
2544 * there cannot be an exact match.
2546 * There must be one non-blank context line that match
2547 * a line before the end of img.
2549 char *buf_end;
2551 buf = preimage->buf;
2552 buf_end = buf;
2553 for (i = 0; i < preimage_limit; i++)
2554 buf_end += preimage->line[i].len;
2556 for ( ; buf < buf_end; buf++)
2557 if (!isspace(*buf))
2558 break;
2559 if (buf == buf_end)
2560 return 0;
2564 * No exact match. If we are ignoring whitespace, run a line-by-line
2565 * fuzzy matching. We collect all the line length information because
2566 * we need it to adjust whitespace if we match.
2568 if (state->ws_ignore_action == ignore_ws_change)
2569 return line_by_line_fuzzy_match(img, preimage, postimage,
2570 current, current_lno, preimage_limit);
2572 if (state->ws_error_action != correct_ws_error)
2573 return 0;
2576 * The hunk does not apply byte-by-byte, but the hash says
2577 * it might with whitespace fuzz. We weren't asked to
2578 * ignore whitespace, we were asked to correct whitespace
2579 * errors, so let's try matching after whitespace correction.
2581 * While checking the preimage against the target, whitespace
2582 * errors in both fixed, we count how large the corresponding
2583 * postimage needs to be. The postimage prepared by
2584 * apply_one_fragment() has whitespace errors fixed on added
2585 * lines already, but the common lines were propagated as-is,
2586 * which may become longer when their whitespace errors are
2587 * fixed.
2590 /* First count added lines in postimage */
2591 postlen = 0;
2592 for (i = 0; i < postimage->nr; i++) {
2593 if (!(postimage->line[i].flag & LINE_COMMON))
2594 postlen += postimage->line[i].len;
2598 * The preimage may extend beyond the end of the file,
2599 * but in this loop we will only handle the part of the
2600 * preimage that falls within the file.
2602 strbuf_init(&fixed, preimage->len + 1);
2603 orig = preimage->buf;
2604 target = img->buf + current;
2605 for (i = 0; i < preimage_limit; i++) {
2606 size_t oldlen = preimage->line[i].len;
2607 size_t tgtlen = img->line[current_lno + i].len;
2608 size_t fixstart = fixed.len;
2609 struct strbuf tgtfix;
2610 int match;
2612 /* Try fixing the line in the preimage */
2613 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2615 /* Try fixing the line in the target */
2616 strbuf_init(&tgtfix, tgtlen);
2617 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2620 * If they match, either the preimage was based on
2621 * a version before our tree fixed whitespace breakage,
2622 * or we are lacking a whitespace-fix patch the tree
2623 * the preimage was based on already had (i.e. target
2624 * has whitespace breakage, the preimage doesn't).
2625 * In either case, we are fixing the whitespace breakages
2626 * so we might as well take the fix together with their
2627 * real change.
2629 match = (tgtfix.len == fixed.len - fixstart &&
2630 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2631 fixed.len - fixstart));
2633 /* Add the length if this is common with the postimage */
2634 if (preimage->line[i].flag & LINE_COMMON)
2635 postlen += tgtfix.len;
2637 strbuf_release(&tgtfix);
2638 if (!match)
2639 goto unmatch_exit;
2641 orig += oldlen;
2642 target += tgtlen;
2647 * Now handle the lines in the preimage that falls beyond the
2648 * end of the file (if any). They will only match if they are
2649 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2650 * false).
2652 for ( ; i < preimage->nr; i++) {
2653 size_t fixstart = fixed.len; /* start of the fixed preimage */
2654 size_t oldlen = preimage->line[i].len;
2655 int j;
2657 /* Try fixing the line in the preimage */
2658 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2660 for (j = fixstart; j < fixed.len; j++)
2661 if (!isspace(fixed.buf[j]))
2662 goto unmatch_exit;
2664 orig += oldlen;
2668 * Yes, the preimage is based on an older version that still
2669 * has whitespace breakages unfixed, and fixing them makes the
2670 * hunk match. Update the context lines in the postimage.
2672 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2673 if (postlen < postimage->len)
2674 postlen = 0;
2675 update_pre_post_images(preimage, postimage,
2676 fixed_buf, fixed_len, postlen);
2677 return 1;
2679 unmatch_exit:
2680 strbuf_release(&fixed);
2681 return 0;
2684 static int find_pos(struct apply_state *state,
2685 struct image *img,
2686 struct image *preimage,
2687 struct image *postimage,
2688 int line,
2689 unsigned ws_rule,
2690 int match_beginning, int match_end)
2692 int i;
2693 unsigned long backwards, forwards, current;
2694 int backwards_lno, forwards_lno, current_lno;
2697 * When running with --allow-overlap, it is possible that a hunk is
2698 * seen that pretends to start at the beginning (but no longer does),
2699 * and that *still* needs to match the end. So trust `match_end` more
2700 * than `match_beginning`.
2702 if (state->allow_overlap && match_beginning && match_end &&
2703 img->nr - preimage->nr != 0)
2704 match_beginning = 0;
2707 * If match_beginning or match_end is specified, there is no
2708 * point starting from a wrong line that will never match and
2709 * wander around and wait for a match at the specified end.
2711 if (match_beginning)
2712 line = 0;
2713 else if (match_end)
2714 line = img->nr - preimage->nr;
2717 * Because the comparison is unsigned, the following test
2718 * will also take care of a negative line number that can
2719 * result when match_end and preimage is larger than the target.
2721 if ((size_t) line > img->nr)
2722 line = img->nr;
2724 current = 0;
2725 for (i = 0; i < line; i++)
2726 current += img->line[i].len;
2729 * There's probably some smart way to do this, but I'll leave
2730 * that to the smart and beautiful people. I'm simple and stupid.
2732 backwards = current;
2733 backwards_lno = line;
2734 forwards = current;
2735 forwards_lno = line;
2736 current_lno = line;
2738 for (i = 0; ; i++) {
2739 if (match_fragment(state, img, preimage, postimage,
2740 current, current_lno, ws_rule,
2741 match_beginning, match_end))
2742 return current_lno;
2744 again:
2745 if (backwards_lno == 0 && forwards_lno == img->nr)
2746 break;
2748 if (i & 1) {
2749 if (backwards_lno == 0) {
2750 i++;
2751 goto again;
2753 backwards_lno--;
2754 backwards -= img->line[backwards_lno].len;
2755 current = backwards;
2756 current_lno = backwards_lno;
2757 } else {
2758 if (forwards_lno == img->nr) {
2759 i++;
2760 goto again;
2762 forwards += img->line[forwards_lno].len;
2763 forwards_lno++;
2764 current = forwards;
2765 current_lno = forwards_lno;
2769 return -1;
2772 static void remove_first_line(struct image *img)
2774 img->buf += img->line[0].len;
2775 img->len -= img->line[0].len;
2776 img->line++;
2777 img->nr--;
2780 static void remove_last_line(struct image *img)
2782 img->len -= img->line[--img->nr].len;
2786 * The change from "preimage" and "postimage" has been found to
2787 * apply at applied_pos (counts in line numbers) in "img".
2788 * Update "img" to remove "preimage" and replace it with "postimage".
2790 static void update_image(struct apply_state *state,
2791 struct image *img,
2792 int applied_pos,
2793 struct image *preimage,
2794 struct image *postimage)
2797 * remove the copy of preimage at offset in img
2798 * and replace it with postimage
2800 int i, nr;
2801 size_t remove_count, insert_count, applied_at = 0;
2802 char *result;
2803 int preimage_limit;
2806 * If we are removing blank lines at the end of img,
2807 * the preimage may extend beyond the end.
2808 * If that is the case, we must be careful only to
2809 * remove the part of the preimage that falls within
2810 * the boundaries of img. Initialize preimage_limit
2811 * to the number of lines in the preimage that falls
2812 * within the boundaries.
2814 preimage_limit = preimage->nr;
2815 if (preimage_limit > img->nr - applied_pos)
2816 preimage_limit = img->nr - applied_pos;
2818 for (i = 0; i < applied_pos; i++)
2819 applied_at += img->line[i].len;
2821 remove_count = 0;
2822 for (i = 0; i < preimage_limit; i++)
2823 remove_count += img->line[applied_pos + i].len;
2824 insert_count = postimage->len;
2826 /* Adjust the contents */
2827 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2828 memcpy(result, img->buf, applied_at);
2829 memcpy(result + applied_at, postimage->buf, postimage->len);
2830 memcpy(result + applied_at + postimage->len,
2831 img->buf + (applied_at + remove_count),
2832 img->len - (applied_at + remove_count));
2833 free(img->buf);
2834 img->buf = result;
2835 img->len += insert_count - remove_count;
2836 result[img->len] = '\0';
2838 /* Adjust the line table */
2839 nr = img->nr + postimage->nr - preimage_limit;
2840 if (preimage_limit < postimage->nr) {
2842 * NOTE: this knows that we never call remove_first_line()
2843 * on anything other than pre/post image.
2845 REALLOC_ARRAY(img->line, nr);
2846 img->line_allocated = img->line;
2848 if (preimage_limit != postimage->nr)
2849 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2850 img->line + applied_pos + preimage_limit,
2851 img->nr - (applied_pos + preimage_limit));
2852 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2853 if (!state->allow_overlap)
2854 for (i = 0; i < postimage->nr; i++)
2855 img->line[applied_pos + i].flag |= LINE_PATCHED;
2856 img->nr = nr;
2860 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2861 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2862 * replace the part of "img" with "postimage" text.
2864 static int apply_one_fragment(struct apply_state *state,
2865 struct image *img, struct fragment *frag,
2866 int inaccurate_eof, unsigned ws_rule,
2867 int nth_fragment)
2869 int match_beginning, match_end;
2870 const char *patch = frag->patch;
2871 int size = frag->size;
2872 char *old, *oldlines;
2873 struct strbuf newlines;
2874 int new_blank_lines_at_end = 0;
2875 int found_new_blank_lines_at_end = 0;
2876 int hunk_linenr = frag->linenr;
2877 unsigned long leading, trailing;
2878 int pos, applied_pos;
2879 struct image preimage;
2880 struct image postimage;
2882 memset(&preimage, 0, sizeof(preimage));
2883 memset(&postimage, 0, sizeof(postimage));
2884 oldlines = xmalloc(size);
2885 strbuf_init(&newlines, size);
2887 old = oldlines;
2888 while (size > 0) {
2889 char first;
2890 int len = linelen(patch, size);
2891 int plen;
2892 int added_blank_line = 0;
2893 int is_blank_context = 0;
2894 size_t start;
2896 if (!len)
2897 break;
2900 * "plen" is how much of the line we should use for
2901 * the actual patch data. Normally we just remove the
2902 * first character on the line, but if the line is
2903 * followed by "\ No newline", then we also remove the
2904 * last one (which is the newline, of course).
2906 plen = len - 1;
2907 if (len < size && patch[len] == '\\')
2908 plen--;
2909 first = *patch;
2910 if (state->apply_in_reverse) {
2911 if (first == '-')
2912 first = '+';
2913 else if (first == '+')
2914 first = '-';
2917 switch (first) {
2918 case '\n':
2919 /* Newer GNU diff, empty context line */
2920 if (plen < 0)
2921 /* ... followed by '\No newline'; nothing */
2922 break;
2923 *old++ = '\n';
2924 strbuf_addch(&newlines, '\n');
2925 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2926 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2927 is_blank_context = 1;
2928 break;
2929 case ' ':
2930 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2931 ws_blank_line(patch + 1, plen))
2932 is_blank_context = 1;
2933 /* fallthrough */
2934 case '-':
2935 memcpy(old, patch + 1, plen);
2936 add_line_info(&preimage, old, plen,
2937 (first == ' ' ? LINE_COMMON : 0));
2938 old += plen;
2939 if (first == '-')
2940 break;
2941 /* fallthrough */
2942 case '+':
2943 /* --no-add does not add new lines */
2944 if (first == '+' && state->no_add)
2945 break;
2947 start = newlines.len;
2948 if (first != '+' ||
2949 !state->whitespace_error ||
2950 state->ws_error_action != correct_ws_error) {
2951 strbuf_add(&newlines, patch + 1, plen);
2953 else {
2954 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2956 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2957 (first == '+' ? 0 : LINE_COMMON));
2958 if (first == '+' &&
2959 (ws_rule & WS_BLANK_AT_EOF) &&
2960 ws_blank_line(patch + 1, plen))
2961 added_blank_line = 1;
2962 break;
2963 case '@': case '\\':
2964 /* Ignore it, we already handled it */
2965 break;
2966 default:
2967 if (state->apply_verbosity > verbosity_normal)
2968 error(_("invalid start of line: '%c'"), first);
2969 applied_pos = -1;
2970 goto out;
2972 if (added_blank_line) {
2973 if (!new_blank_lines_at_end)
2974 found_new_blank_lines_at_end = hunk_linenr;
2975 new_blank_lines_at_end++;
2977 else if (is_blank_context)
2979 else
2980 new_blank_lines_at_end = 0;
2981 patch += len;
2982 size -= len;
2983 hunk_linenr++;
2985 if (inaccurate_eof &&
2986 old > oldlines && old[-1] == '\n' &&
2987 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2988 old--;
2989 strbuf_setlen(&newlines, newlines.len - 1);
2990 preimage.line_allocated[preimage.nr - 1].len--;
2991 postimage.line_allocated[postimage.nr - 1].len--;
2994 leading = frag->leading;
2995 trailing = frag->trailing;
2998 * A hunk to change lines at the beginning would begin with
2999 * @@ -1,L +N,M @@
3000 * but we need to be careful. -U0 that inserts before the second
3001 * line also has this pattern.
3003 * And a hunk to add to an empty file would begin with
3004 * @@ -0,0 +N,M @@
3006 * In other words, a hunk that is (frag->oldpos <= 1) with or
3007 * without leading context must match at the beginning.
3009 match_beginning = (!frag->oldpos ||
3010 (frag->oldpos == 1 && !state->unidiff_zero));
3013 * A hunk without trailing lines must match at the end.
3014 * However, we simply cannot tell if a hunk must match end
3015 * from the lack of trailing lines if the patch was generated
3016 * with unidiff without any context.
3018 match_end = !state->unidiff_zero && !trailing;
3020 pos = frag->newpos ? (frag->newpos - 1) : 0;
3021 preimage.buf = oldlines;
3022 preimage.len = old - oldlines;
3023 postimage.buf = newlines.buf;
3024 postimage.len = newlines.len;
3025 preimage.line = preimage.line_allocated;
3026 postimage.line = postimage.line_allocated;
3028 for (;;) {
3030 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3031 ws_rule, match_beginning, match_end);
3033 if (applied_pos >= 0)
3034 break;
3036 /* Am I at my context limits? */
3037 if ((leading <= state->p_context) && (trailing <= state->p_context))
3038 break;
3039 if (match_beginning || match_end) {
3040 match_beginning = match_end = 0;
3041 continue;
3045 * Reduce the number of context lines; reduce both
3046 * leading and trailing if they are equal otherwise
3047 * just reduce the larger context.
3049 if (leading >= trailing) {
3050 remove_first_line(&preimage);
3051 remove_first_line(&postimage);
3052 pos--;
3053 leading--;
3055 if (trailing > leading) {
3056 remove_last_line(&preimage);
3057 remove_last_line(&postimage);
3058 trailing--;
3062 if (applied_pos >= 0) {
3063 if (new_blank_lines_at_end &&
3064 preimage.nr + applied_pos >= img->nr &&
3065 (ws_rule & WS_BLANK_AT_EOF) &&
3066 state->ws_error_action != nowarn_ws_error) {
3067 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3068 found_new_blank_lines_at_end);
3069 if (state->ws_error_action == correct_ws_error) {
3070 while (new_blank_lines_at_end--)
3071 remove_last_line(&postimage);
3074 * We would want to prevent write_out_results()
3075 * from taking place in apply_patch() that follows
3076 * the callchain led us here, which is:
3077 * apply_patch->check_patch_list->check_patch->
3078 * apply_data->apply_fragments->apply_one_fragment
3080 if (state->ws_error_action == die_on_ws_error)
3081 state->apply = 0;
3084 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3085 int offset = applied_pos - pos;
3086 if (state->apply_in_reverse)
3087 offset = 0 - offset;
3088 fprintf_ln(stderr,
3089 Q_("Hunk #%d succeeded at %d (offset %d line).",
3090 "Hunk #%d succeeded at %d (offset %d lines).",
3091 offset),
3092 nth_fragment, applied_pos + 1, offset);
3096 * Warn if it was necessary to reduce the number
3097 * of context lines.
3099 if ((leading != frag->leading ||
3100 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3101 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3102 " to apply fragment at %d"),
3103 leading, trailing, applied_pos+1);
3104 update_image(state, img, applied_pos, &preimage, &postimage);
3105 } else {
3106 if (state->apply_verbosity > verbosity_normal)
3107 error(_("while searching for:\n%.*s"),
3108 (int)(old - oldlines), oldlines);
3111 out:
3112 free(oldlines);
3113 strbuf_release(&newlines);
3114 free(preimage.line_allocated);
3115 free(postimage.line_allocated);
3117 return (applied_pos < 0);
3120 static int apply_binary_fragment(struct apply_state *state,
3121 struct image *img,
3122 struct patch *patch)
3124 struct fragment *fragment = patch->fragments;
3125 unsigned long len;
3126 void *dst;
3128 if (!fragment)
3129 return error(_("missing binary patch data for '%s'"),
3130 patch->new_name ?
3131 patch->new_name :
3132 patch->old_name);
3134 /* Binary patch is irreversible without the optional second hunk */
3135 if (state->apply_in_reverse) {
3136 if (!fragment->next)
3137 return error(_("cannot reverse-apply a binary patch "
3138 "without the reverse hunk to '%s'"),
3139 patch->new_name
3140 ? patch->new_name : patch->old_name);
3141 fragment = fragment->next;
3143 switch (fragment->binary_patch_method) {
3144 case BINARY_DELTA_DEFLATED:
3145 dst = patch_delta(img->buf, img->len, fragment->patch,
3146 fragment->size, &len);
3147 if (!dst)
3148 return -1;
3149 clear_image(img);
3150 img->buf = dst;
3151 img->len = len;
3152 return 0;
3153 case BINARY_LITERAL_DEFLATED:
3154 clear_image(img);
3155 img->len = fragment->size;
3156 img->buf = xmemdupz(fragment->patch, img->len);
3157 return 0;
3159 return -1;
3163 * Replace "img" with the result of applying the binary patch.
3164 * The binary patch data itself in patch->fragment is still kept
3165 * but the preimage prepared by the caller in "img" is freed here
3166 * or in the helper function apply_binary_fragment() this calls.
3168 static int apply_binary(struct apply_state *state,
3169 struct image *img,
3170 struct patch *patch)
3172 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3173 struct object_id oid;
3174 const unsigned hexsz = the_hash_algo->hexsz;
3177 * For safety, we require patch index line to contain
3178 * full hex textual object ID for old and new, at least for now.
3180 if (strlen(patch->old_oid_prefix) != hexsz ||
3181 strlen(patch->new_oid_prefix) != hexsz ||
3182 get_oid_hex(patch->old_oid_prefix, &oid) ||
3183 get_oid_hex(patch->new_oid_prefix, &oid))
3184 return error(_("cannot apply binary patch to '%s' "
3185 "without full index line"), name);
3187 if (patch->old_name) {
3189 * See if the old one matches what the patch
3190 * applies to.
3192 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3193 &oid);
3194 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3195 return error(_("the patch applies to '%s' (%s), "
3196 "which does not match the "
3197 "current contents."),
3198 name, oid_to_hex(&oid));
3200 else {
3201 /* Otherwise, the old one must be empty. */
3202 if (img->len)
3203 return error(_("the patch applies to an empty "
3204 "'%s' but it is not empty"), name);
3207 get_oid_hex(patch->new_oid_prefix, &oid);
3208 if (is_null_oid(&oid)) {
3209 clear_image(img);
3210 return 0; /* deletion patch */
3213 if (has_object(the_repository, &oid, 0)) {
3214 /* We already have the postimage */
3215 enum object_type type;
3216 unsigned long size;
3217 char *result;
3219 result = repo_read_object_file(the_repository, &oid, &type,
3220 &size);
3221 if (!result)
3222 return error(_("the necessary postimage %s for "
3223 "'%s' cannot be read"),
3224 patch->new_oid_prefix, name);
3225 clear_image(img);
3226 img->buf = result;
3227 img->len = size;
3228 } else {
3230 * We have verified buf matches the preimage;
3231 * apply the patch data to it, which is stored
3232 * in the patch->fragments->{patch,size}.
3234 if (apply_binary_fragment(state, img, patch))
3235 return error(_("binary patch does not apply to '%s'"),
3236 name);
3238 /* verify that the result matches */
3239 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3240 &oid);
3241 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3242 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3243 name, patch->new_oid_prefix, oid_to_hex(&oid));
3246 return 0;
3249 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3251 struct fragment *frag = patch->fragments;
3252 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3253 unsigned ws_rule = patch->ws_rule;
3254 unsigned inaccurate_eof = patch->inaccurate_eof;
3255 int nth = 0;
3257 if (patch->is_binary)
3258 return apply_binary(state, img, patch);
3260 while (frag) {
3261 nth++;
3262 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3263 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3264 if (!state->apply_with_reject)
3265 return -1;
3266 frag->rejected = 1;
3268 frag = frag->next;
3270 return 0;
3273 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3275 if (S_ISGITLINK(mode)) {
3276 strbuf_grow(buf, 100);
3277 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3278 } else {
3279 enum object_type type;
3280 unsigned long sz;
3281 char *result;
3283 result = repo_read_object_file(the_repository, oid, &type,
3284 &sz);
3285 if (!result)
3286 return -1;
3287 /* XXX read_sha1_file NUL-terminates */
3288 strbuf_attach(buf, result, sz, sz + 1);
3290 return 0;
3293 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3295 if (!ce)
3296 return 0;
3297 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3300 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3302 struct string_list_item *item;
3304 if (!name)
3305 return NULL;
3307 item = string_list_lookup(&state->fn_table, name);
3308 if (item)
3309 return (struct patch *)item->util;
3311 return NULL;
3315 * item->util in the filename table records the status of the path.
3316 * Usually it points at a patch (whose result records the contents
3317 * of it after applying it), but it could be PATH_WAS_DELETED for a
3318 * path that a previously applied patch has already removed, or
3319 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3321 * The latter is needed to deal with a case where two paths A and B
3322 * are swapped by first renaming A to B and then renaming B to A;
3323 * moving A to B should not be prevented due to presence of B as we
3324 * will remove it in a later patch.
3326 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3327 #define PATH_WAS_DELETED ((struct patch *) -1)
3329 static int to_be_deleted(struct patch *patch)
3331 return patch == PATH_TO_BE_DELETED;
3334 static int was_deleted(struct patch *patch)
3336 return patch == PATH_WAS_DELETED;
3339 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3341 struct string_list_item *item;
3344 * Always add new_name unless patch is a deletion
3345 * This should cover the cases for normal diffs,
3346 * file creations and copies
3348 if (patch->new_name) {
3349 item = string_list_insert(&state->fn_table, patch->new_name);
3350 item->util = patch;
3354 * store a failure on rename/deletion cases because
3355 * later chunks shouldn't patch old names
3357 if ((patch->new_name == NULL) || (patch->is_rename)) {
3358 item = string_list_insert(&state->fn_table, patch->old_name);
3359 item->util = PATH_WAS_DELETED;
3363 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3366 * store information about incoming file deletion
3368 while (patch) {
3369 if ((patch->new_name == NULL) || (patch->is_rename)) {
3370 struct string_list_item *item;
3371 item = string_list_insert(&state->fn_table, patch->old_name);
3372 item->util = PATH_TO_BE_DELETED;
3374 patch = patch->next;
3378 static int checkout_target(struct index_state *istate,
3379 struct cache_entry *ce, struct stat *st)
3381 struct checkout costate = CHECKOUT_INIT;
3383 costate.refresh_cache = 1;
3384 costate.istate = istate;
3385 if (checkout_entry(ce, &costate, NULL, NULL) ||
3386 lstat(ce->name, st))
3387 return error(_("cannot checkout %s"), ce->name);
3388 return 0;
3391 static struct patch *previous_patch(struct apply_state *state,
3392 struct patch *patch,
3393 int *gone)
3395 struct patch *previous;
3397 *gone = 0;
3398 if (patch->is_copy || patch->is_rename)
3399 return NULL; /* "git" patches do not depend on the order */
3401 previous = in_fn_table(state, patch->old_name);
3402 if (!previous)
3403 return NULL;
3405 if (to_be_deleted(previous))
3406 return NULL; /* the deletion hasn't happened yet */
3408 if (was_deleted(previous))
3409 *gone = 1;
3411 return previous;
3414 static int verify_index_match(struct apply_state *state,
3415 const struct cache_entry *ce,
3416 struct stat *st)
3418 if (S_ISGITLINK(ce->ce_mode)) {
3419 if (!S_ISDIR(st->st_mode))
3420 return -1;
3421 return 0;
3423 return ie_match_stat(state->repo->index, ce, st,
3424 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3427 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3429 static int load_patch_target(struct apply_state *state,
3430 struct strbuf *buf,
3431 const struct cache_entry *ce,
3432 struct stat *st,
3433 struct patch *patch,
3434 const char *name,
3435 unsigned expected_mode)
3437 if (state->cached || state->check_index) {
3438 if (read_file_or_gitlink(ce, buf))
3439 return error(_("failed to read %s"), name);
3440 } else if (name) {
3441 if (S_ISGITLINK(expected_mode)) {
3442 if (ce)
3443 return read_file_or_gitlink(ce, buf);
3444 else
3445 return SUBMODULE_PATCH_WITHOUT_INDEX;
3446 } else if (has_symlink_leading_path(name, strlen(name))) {
3447 return error(_("reading from '%s' beyond a symbolic link"), name);
3448 } else {
3449 if (read_old_data(st, patch, name, buf))
3450 return error(_("failed to read %s"), name);
3453 return 0;
3457 * We are about to apply "patch"; populate the "image" with the
3458 * current version we have, from the working tree or from the index,
3459 * depending on the situation e.g. --cached/--index. If we are
3460 * applying a non-git patch that incrementally updates the tree,
3461 * we read from the result of a previous diff.
3463 static int load_preimage(struct apply_state *state,
3464 struct image *image,
3465 struct patch *patch, struct stat *st,
3466 const struct cache_entry *ce)
3468 struct strbuf buf = STRBUF_INIT;
3469 size_t len;
3470 char *img;
3471 struct patch *previous;
3472 int status;
3474 previous = previous_patch(state, patch, &status);
3475 if (status)
3476 return error(_("path %s has been renamed/deleted"),
3477 patch->old_name);
3478 if (previous) {
3479 /* We have a patched copy in memory; use that. */
3480 strbuf_add(&buf, previous->result, previous->resultsize);
3481 } else {
3482 status = load_patch_target(state, &buf, ce, st, patch,
3483 patch->old_name, patch->old_mode);
3484 if (status < 0)
3485 return status;
3486 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3488 * There is no way to apply subproject
3489 * patch without looking at the index.
3490 * NEEDSWORK: shouldn't this be flagged
3491 * as an error???
3493 free_fragment_list(patch->fragments);
3494 patch->fragments = NULL;
3495 } else if (status) {
3496 return error(_("failed to read %s"), patch->old_name);
3500 img = strbuf_detach(&buf, &len);
3501 prepare_image(image, img, len, !patch->is_binary);
3502 return 0;
3505 static int resolve_to(struct image *image, const struct object_id *result_id)
3507 unsigned long size;
3508 enum object_type type;
3510 clear_image(image);
3512 image->buf = repo_read_object_file(the_repository, result_id, &type,
3513 &size);
3514 if (!image->buf || type != OBJ_BLOB)
3515 die("unable to read blob object %s", oid_to_hex(result_id));
3516 image->len = size;
3518 return 0;
3521 static int three_way_merge(struct apply_state *state,
3522 struct image *image,
3523 char *path,
3524 const struct object_id *base,
3525 const struct object_id *ours,
3526 const struct object_id *theirs)
3528 mmfile_t base_file, our_file, their_file;
3529 mmbuffer_t result = { NULL };
3530 enum ll_merge_result status;
3532 /* resolve trivial cases first */
3533 if (oideq(base, ours))
3534 return resolve_to(image, theirs);
3535 else if (oideq(base, theirs) || oideq(ours, theirs))
3536 return resolve_to(image, ours);
3538 read_mmblob(&base_file, base);
3539 read_mmblob(&our_file, ours);
3540 read_mmblob(&their_file, theirs);
3541 status = ll_merge(&result, path,
3542 &base_file, "base",
3543 &our_file, "ours",
3544 &their_file, "theirs",
3545 state->repo->index,
3546 NULL);
3547 if (status == LL_MERGE_BINARY_CONFLICT)
3548 warning("Cannot merge binary files: %s (%s vs. %s)",
3549 path, "ours", "theirs");
3550 free(base_file.ptr);
3551 free(our_file.ptr);
3552 free(their_file.ptr);
3553 if (status < 0 || !result.ptr) {
3554 free(result.ptr);
3555 return -1;
3557 clear_image(image);
3558 image->buf = result.ptr;
3559 image->len = result.size;
3561 return status;
3565 * When directly falling back to add/add three-way merge, we read from
3566 * the current contents of the new_name. In no cases other than that
3567 * this function will be called.
3569 static int load_current(struct apply_state *state,
3570 struct image *image,
3571 struct patch *patch)
3573 struct strbuf buf = STRBUF_INIT;
3574 int status, pos;
3575 size_t len;
3576 char *img;
3577 struct stat st;
3578 struct cache_entry *ce;
3579 char *name = patch->new_name;
3580 unsigned mode = patch->new_mode;
3582 if (!patch->is_new)
3583 BUG("patch to %s is not a creation", patch->old_name);
3585 pos = index_name_pos(state->repo->index, name, strlen(name));
3586 if (pos < 0)
3587 return error(_("%s: does not exist in index"), name);
3588 ce = state->repo->index->cache[pos];
3589 if (lstat(name, &st)) {
3590 if (errno != ENOENT)
3591 return error_errno("%s", name);
3592 if (checkout_target(state->repo->index, ce, &st))
3593 return -1;
3595 if (verify_index_match(state, ce, &st))
3596 return error(_("%s: does not match index"), name);
3598 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3599 if (status < 0)
3600 return status;
3601 else if (status)
3602 return -1;
3603 img = strbuf_detach(&buf, &len);
3604 prepare_image(image, img, len, !patch->is_binary);
3605 return 0;
3608 static int try_threeway(struct apply_state *state,
3609 struct image *image,
3610 struct patch *patch,
3611 struct stat *st,
3612 const struct cache_entry *ce)
3614 struct object_id pre_oid, post_oid, our_oid;
3615 struct strbuf buf = STRBUF_INIT;
3616 size_t len;
3617 int status;
3618 char *img;
3619 struct image tmp_image;
3621 /* No point falling back to 3-way merge in these cases */
3622 if (patch->is_delete ||
3623 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3624 (patch->is_new && !patch->direct_to_threeway) ||
3625 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3626 return -1;
3628 /* Preimage the patch was prepared for */
3629 if (patch->is_new)
3630 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3631 else if (repo_get_oid(the_repository, patch->old_oid_prefix, &pre_oid) ||
3632 read_blob_object(&buf, &pre_oid, patch->old_mode))
3633 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3635 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3636 fprintf(stderr, _("Performing three-way merge...\n"));
3638 img = strbuf_detach(&buf, &len);
3639 prepare_image(&tmp_image, img, len, 1);
3640 /* Apply the patch to get the post image */
3641 if (apply_fragments(state, &tmp_image, patch) < 0) {
3642 clear_image(&tmp_image);
3643 return -1;
3645 /* post_oid is theirs */
3646 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3647 clear_image(&tmp_image);
3649 /* our_oid is ours */
3650 if (patch->is_new) {
3651 if (load_current(state, &tmp_image, patch))
3652 return error(_("cannot read the current contents of '%s'"),
3653 patch->new_name);
3654 } else {
3655 if (load_preimage(state, &tmp_image, patch, st, ce))
3656 return error(_("cannot read the current contents of '%s'"),
3657 patch->old_name);
3659 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3660 clear_image(&tmp_image);
3662 /* in-core three-way merge between post and our using pre as base */
3663 status = three_way_merge(state, image, patch->new_name,
3664 &pre_oid, &our_oid, &post_oid);
3665 if (status < 0) {
3666 if (state->apply_verbosity > verbosity_silent)
3667 fprintf(stderr,
3668 _("Failed to perform three-way merge...\n"));
3669 return status;
3672 if (status) {
3673 patch->conflicted_threeway = 1;
3674 if (patch->is_new)
3675 oidclr(&patch->threeway_stage[0]);
3676 else
3677 oidcpy(&patch->threeway_stage[0], &pre_oid);
3678 oidcpy(&patch->threeway_stage[1], &our_oid);
3679 oidcpy(&patch->threeway_stage[2], &post_oid);
3680 if (state->apply_verbosity > verbosity_silent)
3681 fprintf(stderr,
3682 _("Applied patch to '%s' with conflicts.\n"),
3683 patch->new_name);
3684 } else {
3685 if (state->apply_verbosity > verbosity_silent)
3686 fprintf(stderr,
3687 _("Applied patch to '%s' cleanly.\n"),
3688 patch->new_name);
3690 return 0;
3693 static int apply_data(struct apply_state *state, struct patch *patch,
3694 struct stat *st, const struct cache_entry *ce)
3696 struct image image;
3698 if (load_preimage(state, &image, patch, st, ce) < 0)
3699 return -1;
3701 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3702 if (state->apply_verbosity > verbosity_silent &&
3703 state->threeway && !patch->direct_to_threeway)
3704 fprintf(stderr, _("Falling back to direct application...\n"));
3706 /* Note: with --reject, apply_fragments() returns 0 */
3707 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0)
3708 return -1;
3710 patch->result = image.buf;
3711 patch->resultsize = image.len;
3712 add_to_fn_table(state, patch);
3713 free(image.line_allocated);
3715 if (0 < patch->is_delete && patch->resultsize)
3716 return error(_("removal patch leaves file contents"));
3718 return 0;
3722 * If "patch" that we are looking at modifies or deletes what we have,
3723 * we would want it not to lose any local modification we have, either
3724 * in the working tree or in the index.
3726 * This also decides if a non-git patch is a creation patch or a
3727 * modification to an existing empty file. We do not check the state
3728 * of the current tree for a creation patch in this function; the caller
3729 * check_patch() separately makes sure (and errors out otherwise) that
3730 * the path the patch creates does not exist in the current tree.
3732 static int check_preimage(struct apply_state *state,
3733 struct patch *patch,
3734 struct cache_entry **ce,
3735 struct stat *st)
3737 const char *old_name = patch->old_name;
3738 struct patch *previous = NULL;
3739 int stat_ret = 0, status;
3740 unsigned st_mode = 0;
3742 if (!old_name)
3743 return 0;
3745 assert(patch->is_new <= 0);
3746 previous = previous_patch(state, patch, &status);
3748 if (status)
3749 return error(_("path %s has been renamed/deleted"), old_name);
3750 if (previous) {
3751 st_mode = previous->new_mode;
3752 } else if (!state->cached) {
3753 stat_ret = lstat(old_name, st);
3754 if (stat_ret && errno != ENOENT)
3755 return error_errno("%s", old_name);
3758 if (state->check_index && !previous) {
3759 int pos = index_name_pos(state->repo->index, old_name,
3760 strlen(old_name));
3761 if (pos < 0) {
3762 if (patch->is_new < 0)
3763 goto is_new;
3764 return error(_("%s: does not exist in index"), old_name);
3766 *ce = state->repo->index->cache[pos];
3767 if (stat_ret < 0) {
3768 if (checkout_target(state->repo->index, *ce, st))
3769 return -1;
3771 if (!state->cached && verify_index_match(state, *ce, st))
3772 return error(_("%s: does not match index"), old_name);
3773 if (state->cached)
3774 st_mode = (*ce)->ce_mode;
3775 } else if (stat_ret < 0) {
3776 if (patch->is_new < 0)
3777 goto is_new;
3778 return error_errno("%s", old_name);
3781 if (!state->cached && !previous)
3782 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3784 if (patch->is_new < 0)
3785 patch->is_new = 0;
3786 if (!patch->old_mode)
3787 patch->old_mode = st_mode;
3788 if ((st_mode ^ patch->old_mode) & S_IFMT)
3789 return error(_("%s: wrong type"), old_name);
3790 if (st_mode != patch->old_mode)
3791 warning(_("%s has type %o, expected %o"),
3792 old_name, st_mode, patch->old_mode);
3793 if (!patch->new_mode && !patch->is_delete)
3794 patch->new_mode = st_mode;
3795 return 0;
3797 is_new:
3798 patch->is_new = 1;
3799 patch->is_delete = 0;
3800 FREE_AND_NULL(patch->old_name);
3801 return 0;
3805 #define EXISTS_IN_INDEX 1
3806 #define EXISTS_IN_WORKTREE 2
3807 #define EXISTS_IN_INDEX_AS_ITA 3
3809 static int check_to_create(struct apply_state *state,
3810 const char *new_name,
3811 int ok_if_exists)
3813 struct stat nst;
3815 if (state->check_index && (!ok_if_exists || !state->cached)) {
3816 int pos;
3818 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3819 if (pos >= 0) {
3820 struct cache_entry *ce = state->repo->index->cache[pos];
3822 /* allow ITA, as they do not yet exist in the index */
3823 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3824 return EXISTS_IN_INDEX;
3826 /* ITA entries can never match working tree files */
3827 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3828 return EXISTS_IN_INDEX_AS_ITA;
3832 if (state->cached)
3833 return 0;
3835 if (!lstat(new_name, &nst)) {
3836 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3837 return 0;
3839 * A leading component of new_name might be a symlink
3840 * that is going to be removed with this patch, but
3841 * still pointing at somewhere that has the path.
3842 * In such a case, path "new_name" does not exist as
3843 * far as git is concerned.
3845 if (has_symlink_leading_path(new_name, strlen(new_name)))
3846 return 0;
3848 return EXISTS_IN_WORKTREE;
3849 } else if (!is_missing_file_error(errno)) {
3850 return error_errno("%s", new_name);
3852 return 0;
3855 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3857 for ( ; patch; patch = patch->next) {
3858 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3859 (patch->is_rename || patch->is_delete))
3860 /* the symlink at patch->old_name is removed */
3861 strset_add(&state->removed_symlinks, patch->old_name);
3863 if (patch->new_name && S_ISLNK(patch->new_mode))
3864 /* the symlink at patch->new_name is created or remains */
3865 strset_add(&state->kept_symlinks, patch->new_name);
3869 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3871 do {
3872 while (--name->len && name->buf[name->len] != '/')
3873 ; /* scan backwards */
3874 if (!name->len)
3875 break;
3876 name->buf[name->len] = '\0';
3877 if (strset_contains(&state->kept_symlinks, name->buf))
3878 return 1;
3879 if (strset_contains(&state->removed_symlinks, name->buf))
3881 * This cannot be "return 0", because we may
3882 * see a new one created at a higher level.
3884 continue;
3886 /* otherwise, check the preimage */
3887 if (state->check_index) {
3888 struct cache_entry *ce;
3890 ce = index_file_exists(state->repo->index, name->buf,
3891 name->len, ignore_case);
3892 if (ce && S_ISLNK(ce->ce_mode))
3893 return 1;
3894 } else {
3895 struct stat st;
3896 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3897 return 1;
3899 } while (1);
3900 return 0;
3903 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3905 int ret;
3906 struct strbuf name = STRBUF_INIT;
3908 assert(*name_ != '\0');
3909 strbuf_addstr(&name, name_);
3910 ret = path_is_beyond_symlink_1(state, &name);
3911 strbuf_release(&name);
3913 return ret;
3916 static int check_unsafe_path(struct patch *patch)
3918 const char *old_name = NULL;
3919 const char *new_name = NULL;
3920 if (patch->is_delete)
3921 old_name = patch->old_name;
3922 else if (!patch->is_new && !patch->is_copy)
3923 old_name = patch->old_name;
3924 if (!patch->is_delete)
3925 new_name = patch->new_name;
3927 if (old_name && !verify_path(old_name, patch->old_mode))
3928 return error(_("invalid path '%s'"), old_name);
3929 if (new_name && !verify_path(new_name, patch->new_mode))
3930 return error(_("invalid path '%s'"), new_name);
3931 return 0;
3935 * Check and apply the patch in-core; leave the result in patch->result
3936 * for the caller to write it out to the final destination.
3938 static int check_patch(struct apply_state *state, struct patch *patch)
3940 struct stat st;
3941 const char *old_name = patch->old_name;
3942 const char *new_name = patch->new_name;
3943 const char *name = old_name ? old_name : new_name;
3944 struct cache_entry *ce = NULL;
3945 struct patch *tpatch;
3946 int ok_if_exists;
3947 int status;
3949 patch->rejected = 1; /* we will drop this after we succeed */
3951 status = check_preimage(state, patch, &ce, &st);
3952 if (status)
3953 return status;
3954 old_name = patch->old_name;
3957 * A type-change diff is always split into a patch to delete
3958 * old, immediately followed by a patch to create new (see
3959 * diff.c::run_diff()); in such a case it is Ok that the entry
3960 * to be deleted by the previous patch is still in the working
3961 * tree and in the index.
3963 * A patch to swap-rename between A and B would first rename A
3964 * to B and then rename B to A. While applying the first one,
3965 * the presence of B should not stop A from getting renamed to
3966 * B; ask to_be_deleted() about the later rename. Removal of
3967 * B and rename from A to B is handled the same way by asking
3968 * was_deleted().
3970 if ((tpatch = in_fn_table(state, new_name)) &&
3971 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3972 ok_if_exists = 1;
3973 else
3974 ok_if_exists = 0;
3976 if (new_name &&
3977 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3978 int err = check_to_create(state, new_name, ok_if_exists);
3980 if (err && state->threeway) {
3981 patch->direct_to_threeway = 1;
3982 } else switch (err) {
3983 case 0:
3984 break; /* happy */
3985 case EXISTS_IN_INDEX:
3986 return error(_("%s: already exists in index"), new_name);
3987 case EXISTS_IN_INDEX_AS_ITA:
3988 return error(_("%s: does not match index"), new_name);
3989 case EXISTS_IN_WORKTREE:
3990 return error(_("%s: already exists in working directory"),
3991 new_name);
3992 default:
3993 return err;
3996 if (!patch->new_mode) {
3997 if (0 < patch->is_new)
3998 patch->new_mode = S_IFREG | 0644;
3999 else
4000 patch->new_mode = patch->old_mode;
4004 if (new_name && old_name) {
4005 int same = !strcmp(old_name, new_name);
4006 if (!patch->new_mode)
4007 patch->new_mode = patch->old_mode;
4008 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
4009 if (same)
4010 return error(_("new mode (%o) of %s does not "
4011 "match old mode (%o)"),
4012 patch->new_mode, new_name,
4013 patch->old_mode);
4014 else
4015 return error(_("new mode (%o) of %s does not "
4016 "match old mode (%o) of %s"),
4017 patch->new_mode, new_name,
4018 patch->old_mode, old_name);
4022 if (!state->unsafe_paths && check_unsafe_path(patch))
4023 return -128;
4026 * An attempt to read from or delete a path that is beyond a
4027 * symbolic link will be prevented by load_patch_target() that
4028 * is called at the beginning of apply_data() so we do not
4029 * have to worry about a patch marked with "is_delete" bit
4030 * here. We however need to make sure that the patch result
4031 * is not deposited to a path that is beyond a symbolic link
4032 * here.
4034 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4035 return error(_("affected file '%s' is beyond a symbolic link"),
4036 patch->new_name);
4038 if (apply_data(state, patch, &st, ce) < 0)
4039 return error(_("%s: patch does not apply"), name);
4040 patch->rejected = 0;
4041 return 0;
4044 static int check_patch_list(struct apply_state *state, struct patch *patch)
4046 int err = 0;
4048 prepare_symlink_changes(state, patch);
4049 prepare_fn_table(state, patch);
4050 while (patch) {
4051 int res;
4052 if (state->apply_verbosity > verbosity_normal)
4053 say_patch_name(stderr,
4054 _("Checking patch %s..."), patch);
4055 res = check_patch(state, patch);
4056 if (res == -128)
4057 return -128;
4058 err |= res;
4059 patch = patch->next;
4061 return err;
4064 static int read_apply_cache(struct apply_state *state)
4066 if (state->index_file)
4067 return read_index_from(state->repo->index, state->index_file,
4068 get_git_dir());
4069 else
4070 return repo_read_index(state->repo);
4073 /* This function tries to read the object name from the current index */
4074 static int get_current_oid(struct apply_state *state, const char *path,
4075 struct object_id *oid)
4077 int pos;
4079 if (read_apply_cache(state) < 0)
4080 return -1;
4081 pos = index_name_pos(state->repo->index, path, strlen(path));
4082 if (pos < 0)
4083 return -1;
4084 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4085 return 0;
4088 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4091 * A usable gitlink patch has only one fragment (hunk) that looks like:
4092 * @@ -1 +1 @@
4093 * -Subproject commit <old sha1>
4094 * +Subproject commit <new sha1>
4095 * or
4096 * @@ -1 +0,0 @@
4097 * -Subproject commit <old sha1>
4098 * for a removal patch.
4100 struct fragment *hunk = p->fragments;
4101 static const char heading[] = "-Subproject commit ";
4102 char *preimage;
4104 if (/* does the patch have only one hunk? */
4105 hunk && !hunk->next &&
4106 /* is its preimage one line? */
4107 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4108 /* does preimage begin with the heading? */
4109 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4110 starts_with(++preimage, heading) &&
4111 /* does it record full SHA-1? */
4112 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4113 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4114 /* does the abbreviated name on the index line agree with it? */
4115 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4116 return 0; /* it all looks fine */
4118 /* we may have full object name on the index line */
4119 return get_oid_hex(p->old_oid_prefix, oid);
4122 /* Build an index that contains just the files needed for a 3way merge */
4123 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4125 struct patch *patch;
4126 struct index_state result = INDEX_STATE_INIT(state->repo);
4127 struct lock_file lock = LOCK_INIT;
4128 int res;
4130 /* Once we start supporting the reverse patch, it may be
4131 * worth showing the new sha1 prefix, but until then...
4133 for (patch = list; patch; patch = patch->next) {
4134 struct object_id oid;
4135 struct cache_entry *ce;
4136 const char *name;
4138 name = patch->old_name ? patch->old_name : patch->new_name;
4139 if (0 < patch->is_new)
4140 continue;
4142 if (S_ISGITLINK(patch->old_mode)) {
4143 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4144 ; /* ok, the textual part looks sane */
4145 else
4146 return error(_("sha1 information is lacking or "
4147 "useless for submodule %s"), name);
4148 } else if (!repo_get_oid_blob(the_repository, patch->old_oid_prefix, &oid)) {
4149 ; /* ok */
4150 } else if (!patch->lines_added && !patch->lines_deleted) {
4151 /* mode-only change: update the current */
4152 if (get_current_oid(state, patch->old_name, &oid))
4153 return error(_("mode change for %s, which is not "
4154 "in current HEAD"), name);
4155 } else
4156 return error(_("sha1 information is lacking or useless "
4157 "(%s)."), name);
4159 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4160 if (!ce)
4161 return error(_("make_cache_entry failed for path '%s'"),
4162 name);
4163 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4164 discard_cache_entry(ce);
4165 return error(_("could not add %s to temporary index"),
4166 name);
4170 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4171 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4172 discard_index(&result);
4174 if (res)
4175 return error(_("could not write temporary index to %s"),
4176 state->fake_ancestor);
4178 return 0;
4181 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4183 int files, adds, dels;
4185 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4186 files++;
4187 adds += patch->lines_added;
4188 dels += patch->lines_deleted;
4189 show_stats(state, patch);
4192 print_stat_summary(stdout, files, adds, dels);
4195 static void numstat_patch_list(struct apply_state *state,
4196 struct patch *patch)
4198 for ( ; patch; patch = patch->next) {
4199 const char *name;
4200 name = patch->new_name ? patch->new_name : patch->old_name;
4201 if (patch->is_binary)
4202 printf("-\t-\t");
4203 else
4204 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4205 write_name_quoted(name, stdout, state->line_termination);
4209 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4211 if (mode)
4212 printf(" %s mode %06o %s\n", newdelete, mode, name);
4213 else
4214 printf(" %s %s\n", newdelete, name);
4217 static void show_mode_change(struct patch *p, int show_name)
4219 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4220 if (show_name)
4221 printf(" mode change %06o => %06o %s\n",
4222 p->old_mode, p->new_mode, p->new_name);
4223 else
4224 printf(" mode change %06o => %06o\n",
4225 p->old_mode, p->new_mode);
4229 static void show_rename_copy(struct patch *p)
4231 const char *renamecopy = p->is_rename ? "rename" : "copy";
4232 const char *old_name, *new_name;
4234 /* Find common prefix */
4235 old_name = p->old_name;
4236 new_name = p->new_name;
4237 while (1) {
4238 const char *slash_old, *slash_new;
4239 slash_old = strchr(old_name, '/');
4240 slash_new = strchr(new_name, '/');
4241 if (!slash_old ||
4242 !slash_new ||
4243 slash_old - old_name != slash_new - new_name ||
4244 memcmp(old_name, new_name, slash_new - new_name))
4245 break;
4246 old_name = slash_old + 1;
4247 new_name = slash_new + 1;
4249 /* p->old_name through old_name is the common prefix, and old_name and
4250 * new_name through the end of names are renames
4252 if (old_name != p->old_name)
4253 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4254 (int)(old_name - p->old_name), p->old_name,
4255 old_name, new_name, p->score);
4256 else
4257 printf(" %s %s => %s (%d%%)\n", renamecopy,
4258 p->old_name, p->new_name, p->score);
4259 show_mode_change(p, 0);
4262 static void summary_patch_list(struct patch *patch)
4264 struct patch *p;
4266 for (p = patch; p; p = p->next) {
4267 if (p->is_new)
4268 show_file_mode_name("create", p->new_mode, p->new_name);
4269 else if (p->is_delete)
4270 show_file_mode_name("delete", p->old_mode, p->old_name);
4271 else {
4272 if (p->is_rename || p->is_copy)
4273 show_rename_copy(p);
4274 else {
4275 if (p->score) {
4276 printf(" rewrite %s (%d%%)\n",
4277 p->new_name, p->score);
4278 show_mode_change(p, 0);
4280 else
4281 show_mode_change(p, 1);
4287 static void patch_stats(struct apply_state *state, struct patch *patch)
4289 int lines = patch->lines_added + patch->lines_deleted;
4291 if (lines > state->max_change)
4292 state->max_change = lines;
4293 if (patch->old_name) {
4294 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4295 if (!len)
4296 len = strlen(patch->old_name);
4297 if (len > state->max_len)
4298 state->max_len = len;
4300 if (patch->new_name) {
4301 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4302 if (!len)
4303 len = strlen(patch->new_name);
4304 if (len > state->max_len)
4305 state->max_len = len;
4309 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4311 if (state->update_index && !state->ita_only) {
4312 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4313 return error(_("unable to remove %s from index"), patch->old_name);
4315 if (!state->cached) {
4316 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4317 remove_path(patch->old_name);
4320 return 0;
4323 static int add_index_file(struct apply_state *state,
4324 const char *path,
4325 unsigned mode,
4326 void *buf,
4327 unsigned long size)
4329 struct stat st;
4330 struct cache_entry *ce;
4331 int namelen = strlen(path);
4333 ce = make_empty_cache_entry(state->repo->index, namelen);
4334 memcpy(ce->name, path, namelen);
4335 ce->ce_mode = create_ce_mode(mode);
4336 ce->ce_flags = create_ce_flags(0);
4337 ce->ce_namelen = namelen;
4338 if (state->ita_only) {
4339 ce->ce_flags |= CE_INTENT_TO_ADD;
4340 set_object_name_for_intent_to_add_entry(ce);
4341 } else if (S_ISGITLINK(mode)) {
4342 const char *s;
4344 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4345 get_oid_hex(s, &ce->oid)) {
4346 discard_cache_entry(ce);
4347 return error(_("corrupt patch for submodule %s"), path);
4349 } else {
4350 if (!state->cached) {
4351 if (lstat(path, &st) < 0) {
4352 discard_cache_entry(ce);
4353 return error_errno(_("unable to stat newly "
4354 "created file '%s'"),
4355 path);
4357 fill_stat_cache_info(state->repo->index, ce, &st);
4359 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4360 discard_cache_entry(ce);
4361 return error(_("unable to create backing store "
4362 "for newly created file %s"), path);
4365 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4366 discard_cache_entry(ce);
4367 return error(_("unable to add cache entry for %s"), path);
4370 return 0;
4374 * Returns:
4375 * -1 if an unrecoverable error happened
4376 * 0 if everything went well
4377 * 1 if a recoverable error happened
4379 static int try_create_file(struct apply_state *state, const char *path,
4380 unsigned int mode, const char *buf,
4381 unsigned long size)
4383 int fd, res;
4384 struct strbuf nbuf = STRBUF_INIT;
4386 if (S_ISGITLINK(mode)) {
4387 struct stat st;
4388 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4389 return 0;
4390 return !!mkdir(path, 0777);
4393 if (has_symlinks && S_ISLNK(mode))
4394 /* Although buf:size is counted string, it also is NUL
4395 * terminated.
4397 return !!symlink(buf, path);
4399 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4400 if (fd < 0)
4401 return 1;
4403 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4404 size = nbuf.len;
4405 buf = nbuf.buf;
4408 res = write_in_full(fd, buf, size) < 0;
4409 if (res)
4410 error_errno(_("failed to write to '%s'"), path);
4411 strbuf_release(&nbuf);
4413 if (close(fd) < 0 && !res)
4414 return error_errno(_("closing file '%s'"), path);
4416 return res ? -1 : 0;
4420 * We optimistically assume that the directories exist,
4421 * which is true 99% of the time anyway. If they don't,
4422 * we create them and try again.
4424 * Returns:
4425 * -1 on error
4426 * 0 otherwise
4428 static int create_one_file(struct apply_state *state,
4429 char *path,
4430 unsigned mode,
4431 const char *buf,
4432 unsigned long size)
4434 int res;
4436 if (state->cached)
4437 return 0;
4440 * We already try to detect whether files are beyond a symlink in our
4441 * up-front checks. But in the case where symlinks are created by any
4442 * of the intermediate hunks it can happen that our up-front checks
4443 * didn't yet see the symlink, but at the point of arriving here there
4444 * in fact is one. We thus repeat the check for symlinks here.
4446 * Note that this does not make the up-front check obsolete as the
4447 * failure mode is different:
4449 * - The up-front checks cause us to abort before we have written
4450 * anything into the working directory. So when we exit this way the
4451 * working directory remains clean.
4453 * - The checks here happen in the middle of the action where we have
4454 * already started to apply the patch. The end result will be a dirty
4455 * working directory.
4457 * Ideally, we should update the up-front checks to catch what would
4458 * happen when we apply the patch before we damage the working tree.
4459 * We have all the information necessary to do so. But for now, as a
4460 * part of embargoed security work, having this check would serve as a
4461 * reasonable first step.
4463 if (path_is_beyond_symlink(state, path))
4464 return error(_("affected file '%s' is beyond a symbolic link"), path);
4466 res = try_create_file(state, path, mode, buf, size);
4467 if (res < 0)
4468 return -1;
4469 if (!res)
4470 return 0;
4472 if (errno == ENOENT) {
4473 if (safe_create_leading_directories_no_share(path))
4474 return 0;
4475 res = try_create_file(state, path, mode, buf, size);
4476 if (res < 0)
4477 return -1;
4478 if (!res)
4479 return 0;
4482 if (errno == EEXIST || errno == EACCES) {
4483 /* We may be trying to create a file where a directory
4484 * used to be.
4486 struct stat st;
4487 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4488 errno = EEXIST;
4491 if (errno == EEXIST) {
4492 unsigned int nr = getpid();
4494 for (;;) {
4495 char newpath[PATH_MAX];
4496 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4497 res = try_create_file(state, newpath, mode, buf, size);
4498 if (res < 0)
4499 return -1;
4500 if (!res) {
4501 if (!rename(newpath, path))
4502 return 0;
4503 unlink_or_warn(newpath);
4504 break;
4506 if (errno != EEXIST)
4507 break;
4508 ++nr;
4511 return error_errno(_("unable to write file '%s' mode %o"),
4512 path, mode);
4515 static int add_conflicted_stages_file(struct apply_state *state,
4516 struct patch *patch)
4518 int stage, namelen;
4519 unsigned mode;
4520 struct cache_entry *ce;
4522 if (!state->update_index)
4523 return 0;
4524 namelen = strlen(patch->new_name);
4525 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4527 remove_file_from_index(state->repo->index, patch->new_name);
4528 for (stage = 1; stage < 4; stage++) {
4529 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4530 continue;
4531 ce = make_empty_cache_entry(state->repo->index, namelen);
4532 memcpy(ce->name, patch->new_name, namelen);
4533 ce->ce_mode = create_ce_mode(mode);
4534 ce->ce_flags = create_ce_flags(stage);
4535 ce->ce_namelen = namelen;
4536 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4537 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4538 discard_cache_entry(ce);
4539 return error(_("unable to add cache entry for %s"),
4540 patch->new_name);
4544 return 0;
4547 static int create_file(struct apply_state *state, struct patch *patch)
4549 char *path = patch->new_name;
4550 unsigned mode = patch->new_mode;
4551 unsigned long size = patch->resultsize;
4552 char *buf = patch->result;
4554 if (!mode)
4555 mode = S_IFREG | 0644;
4556 if (create_one_file(state, path, mode, buf, size))
4557 return -1;
4559 if (patch->conflicted_threeway)
4560 return add_conflicted_stages_file(state, patch);
4561 else if (state->update_index)
4562 return add_index_file(state, path, mode, buf, size);
4563 return 0;
4566 /* phase zero is to remove, phase one is to create */
4567 static int write_out_one_result(struct apply_state *state,
4568 struct patch *patch,
4569 int phase)
4571 if (patch->is_delete > 0) {
4572 if (phase == 0)
4573 return remove_file(state, patch, 1);
4574 return 0;
4576 if (patch->is_new > 0 || patch->is_copy) {
4577 if (phase == 1)
4578 return create_file(state, patch);
4579 return 0;
4582 * Rename or modification boils down to the same
4583 * thing: remove the old, write the new
4585 if (phase == 0)
4586 return remove_file(state, patch, patch->is_rename);
4587 if (phase == 1)
4588 return create_file(state, patch);
4589 return 0;
4592 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4594 FILE *rej;
4595 char namebuf[PATH_MAX];
4596 struct fragment *frag;
4597 int fd, cnt = 0;
4598 struct strbuf sb = STRBUF_INIT;
4600 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4601 if (!frag->rejected)
4602 continue;
4603 cnt++;
4606 if (!cnt) {
4607 if (state->apply_verbosity > verbosity_normal)
4608 say_patch_name(stderr,
4609 _("Applied patch %s cleanly."), patch);
4610 return 0;
4613 /* This should not happen, because a removal patch that leaves
4614 * contents are marked "rejected" at the patch level.
4616 if (!patch->new_name)
4617 die(_("internal error"));
4619 /* Say this even without --verbose */
4620 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4621 "Applying patch %%s with %d rejects...",
4622 cnt),
4623 cnt);
4624 if (state->apply_verbosity > verbosity_silent)
4625 say_patch_name(stderr, sb.buf, patch);
4626 strbuf_release(&sb);
4628 cnt = strlen(patch->new_name);
4629 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4630 cnt = ARRAY_SIZE(namebuf) - 5;
4631 warning(_("truncating .rej filename to %.*s.rej"),
4632 cnt - 1, patch->new_name);
4634 memcpy(namebuf, patch->new_name, cnt);
4635 memcpy(namebuf + cnt, ".rej", 5);
4637 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4638 if (fd < 0) {
4639 if (errno != EEXIST)
4640 return error_errno(_("cannot open %s"), namebuf);
4641 if (unlink(namebuf))
4642 return error_errno(_("cannot unlink '%s'"), namebuf);
4643 fd = open(namebuf, O_CREAT | O_EXCL | O_WRONLY, 0666);
4644 if (fd < 0)
4645 return error_errno(_("cannot open %s"), namebuf);
4647 rej = fdopen(fd, "w");
4648 if (!rej)
4649 return error_errno(_("cannot open %s"), namebuf);
4651 /* Normal git tools never deal with .rej, so do not pretend
4652 * this is a git patch by saying --git or giving extended
4653 * headers. While at it, maybe please "kompare" that wants
4654 * the trailing TAB and some garbage at the end of line ;-).
4656 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4657 patch->new_name, patch->new_name);
4658 for (cnt = 1, frag = patch->fragments;
4659 frag;
4660 cnt++, frag = frag->next) {
4661 if (!frag->rejected) {
4662 if (state->apply_verbosity > verbosity_silent)
4663 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4664 continue;
4666 if (state->apply_verbosity > verbosity_silent)
4667 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4668 fprintf(rej, "%.*s", frag->size, frag->patch);
4669 if (frag->patch[frag->size-1] != '\n')
4670 fputc('\n', rej);
4672 fclose(rej);
4673 return -1;
4677 * Returns:
4678 * -1 if an error happened
4679 * 0 if the patch applied cleanly
4680 * 1 if the patch did not apply cleanly
4682 static int write_out_results(struct apply_state *state, struct patch *list)
4684 int phase;
4685 int errs = 0;
4686 struct patch *l;
4687 struct string_list cpath = STRING_LIST_INIT_DUP;
4689 for (phase = 0; phase < 2; phase++) {
4690 l = list;
4691 while (l) {
4692 if (l->rejected)
4693 errs = 1;
4694 else {
4695 if (write_out_one_result(state, l, phase)) {
4696 string_list_clear(&cpath, 0);
4697 return -1;
4699 if (phase == 1) {
4700 if (write_out_one_reject(state, l))
4701 errs = 1;
4702 if (l->conflicted_threeway) {
4703 string_list_append(&cpath, l->new_name);
4704 errs = 1;
4708 l = l->next;
4712 if (cpath.nr) {
4713 struct string_list_item *item;
4715 string_list_sort(&cpath);
4716 if (state->apply_verbosity > verbosity_silent) {
4717 for_each_string_list_item(item, &cpath)
4718 fprintf(stderr, "U %s\n", item->string);
4720 string_list_clear(&cpath, 0);
4723 * rerere relies on the partially merged result being in the working
4724 * tree with conflict markers, but that isn't written with --cached.
4726 if (!state->cached)
4727 repo_rerere(state->repo, 0);
4730 return errs;
4734 * Try to apply a patch.
4736 * Returns:
4737 * -128 if a bad error happened (like patch unreadable)
4738 * -1 if patch did not apply and user cannot deal with it
4739 * 0 if the patch applied
4740 * 1 if the patch did not apply but user might fix it
4742 static int apply_patch(struct apply_state *state,
4743 int fd,
4744 const char *filename,
4745 int options)
4747 size_t offset;
4748 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4749 struct patch *list = NULL, **listp = &list;
4750 int skipped_patch = 0;
4751 int res = 0;
4752 int flush_attributes = 0;
4754 state->patch_input_file = filename;
4755 if (read_patch_file(&buf, fd) < 0)
4756 return -128;
4757 offset = 0;
4758 while (offset < buf.len) {
4759 struct patch *patch;
4760 int nr;
4762 CALLOC_ARRAY(patch, 1);
4763 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4764 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4765 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4766 if (nr < 0) {
4767 free_patch(patch);
4768 if (nr == -128) {
4769 res = -128;
4770 goto end;
4772 break;
4774 if (state->apply_in_reverse)
4775 reverse_patches(patch);
4776 if (use_patch(state, patch)) {
4777 patch_stats(state, patch);
4778 if (!list || !state->apply_in_reverse) {
4779 *listp = patch;
4780 listp = &patch->next;
4781 } else {
4782 patch->next = list;
4783 list = patch;
4786 if ((patch->new_name &&
4787 ends_with_path_components(patch->new_name,
4788 GITATTRIBUTES_FILE)) ||
4789 (patch->old_name &&
4790 ends_with_path_components(patch->old_name,
4791 GITATTRIBUTES_FILE)))
4792 flush_attributes = 1;
4794 else {
4795 if (state->apply_verbosity > verbosity_normal)
4796 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4797 free_patch(patch);
4798 skipped_patch++;
4800 offset += nr;
4803 if (!list && !skipped_patch) {
4804 if (!state->allow_empty) {
4805 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4806 res = -128;
4808 goto end;
4811 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4812 state->apply = 0;
4814 state->update_index = (state->check_index || state->ita_only) && state->apply;
4815 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4816 if (state->index_file)
4817 hold_lock_file_for_update(&state->lock_file,
4818 state->index_file,
4819 LOCK_DIE_ON_ERROR);
4820 else
4821 repo_hold_locked_index(state->repo, &state->lock_file,
4822 LOCK_DIE_ON_ERROR);
4825 if (state->check_index && read_apply_cache(state) < 0) {
4826 error(_("unable to read index file"));
4827 res = -128;
4828 goto end;
4831 if (state->check || state->apply) {
4832 int r = check_patch_list(state, list);
4833 if (r == -128) {
4834 res = -128;
4835 goto end;
4837 if (r < 0 && !state->apply_with_reject) {
4838 res = -1;
4839 goto end;
4843 if (state->apply) {
4844 int write_res = write_out_results(state, list);
4845 if (write_res < 0) {
4846 res = -128;
4847 goto end;
4849 if (write_res > 0) {
4850 /* with --3way, we still need to write the index out */
4851 res = state->apply_with_reject ? -1 : 1;
4852 goto end;
4856 if (state->fake_ancestor &&
4857 build_fake_ancestor(state, list)) {
4858 res = -128;
4859 goto end;
4862 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4863 stat_patch_list(state, list);
4865 if (state->numstat && state->apply_verbosity > verbosity_silent)
4866 numstat_patch_list(state, list);
4868 if (state->summary && state->apply_verbosity > verbosity_silent)
4869 summary_patch_list(list);
4871 if (flush_attributes)
4872 reset_parsed_attributes();
4873 end:
4874 free_patch_list(list);
4875 strbuf_release(&buf);
4876 string_list_clear(&state->fn_table, 0);
4877 return res;
4880 static int apply_option_parse_exclude(const struct option *opt,
4881 const char *arg, int unset)
4883 struct apply_state *state = opt->value;
4885 BUG_ON_OPT_NEG(unset);
4887 add_name_limit(state, arg, 1);
4888 return 0;
4891 static int apply_option_parse_include(const struct option *opt,
4892 const char *arg, int unset)
4894 struct apply_state *state = opt->value;
4896 BUG_ON_OPT_NEG(unset);
4898 add_name_limit(state, arg, 0);
4899 state->has_include = 1;
4900 return 0;
4903 static int apply_option_parse_p(const struct option *opt,
4904 const char *arg,
4905 int unset)
4907 struct apply_state *state = opt->value;
4909 BUG_ON_OPT_NEG(unset);
4911 state->p_value = atoi(arg);
4912 state->p_value_known = 1;
4913 return 0;
4916 static int apply_option_parse_space_change(const struct option *opt,
4917 const char *arg, int unset)
4919 struct apply_state *state = opt->value;
4921 BUG_ON_OPT_ARG(arg);
4923 if (unset)
4924 state->ws_ignore_action = ignore_ws_none;
4925 else
4926 state->ws_ignore_action = ignore_ws_change;
4927 return 0;
4930 static int apply_option_parse_whitespace(const struct option *opt,
4931 const char *arg, int unset)
4933 struct apply_state *state = opt->value;
4935 BUG_ON_OPT_NEG(unset);
4937 state->whitespace_option = arg;
4938 if (parse_whitespace_option(state, arg))
4939 return -1;
4940 return 0;
4943 static int apply_option_parse_directory(const struct option *opt,
4944 const char *arg, int unset)
4946 struct apply_state *state = opt->value;
4948 BUG_ON_OPT_NEG(unset);
4950 strbuf_reset(&state->root);
4951 strbuf_addstr(&state->root, arg);
4952 strbuf_complete(&state->root, '/');
4953 return 0;
4956 int apply_all_patches(struct apply_state *state,
4957 int argc,
4958 const char **argv,
4959 int options)
4961 int i;
4962 int res;
4963 int errs = 0;
4964 int read_stdin = 1;
4966 for (i = 0; i < argc; i++) {
4967 const char *arg = argv[i];
4968 char *to_free = NULL;
4969 int fd;
4971 if (!strcmp(arg, "-")) {
4972 res = apply_patch(state, 0, "<stdin>", options);
4973 if (res < 0)
4974 goto end;
4975 errs |= res;
4976 read_stdin = 0;
4977 continue;
4978 } else
4979 arg = to_free = prefix_filename(state->prefix, arg);
4981 fd = open(arg, O_RDONLY);
4982 if (fd < 0) {
4983 error(_("can't open patch '%s': %s"), arg, strerror(errno));
4984 res = -128;
4985 free(to_free);
4986 goto end;
4988 read_stdin = 0;
4989 set_default_whitespace_mode(state);
4990 res = apply_patch(state, fd, arg, options);
4991 close(fd);
4992 free(to_free);
4993 if (res < 0)
4994 goto end;
4995 errs |= res;
4997 set_default_whitespace_mode(state);
4998 if (read_stdin) {
4999 res = apply_patch(state, 0, "<stdin>", options);
5000 if (res < 0)
5001 goto end;
5002 errs |= res;
5005 if (state->whitespace_error) {
5006 if (state->squelch_whitespace_errors &&
5007 state->squelch_whitespace_errors < state->whitespace_error) {
5008 int squelched =
5009 state->whitespace_error - state->squelch_whitespace_errors;
5010 warning(Q_("squelched %d whitespace error",
5011 "squelched %d whitespace errors",
5012 squelched),
5013 squelched);
5015 if (state->ws_error_action == die_on_ws_error) {
5016 error(Q_("%d line adds whitespace errors.",
5017 "%d lines add whitespace errors.",
5018 state->whitespace_error),
5019 state->whitespace_error);
5020 res = -128;
5021 goto end;
5023 if (state->applied_after_fixing_ws && state->apply)
5024 warning(Q_("%d line applied after"
5025 " fixing whitespace errors.",
5026 "%d lines applied after"
5027 " fixing whitespace errors.",
5028 state->applied_after_fixing_ws),
5029 state->applied_after_fixing_ws);
5030 else if (state->whitespace_error)
5031 warning(Q_("%d line adds whitespace errors.",
5032 "%d lines add whitespace errors.",
5033 state->whitespace_error),
5034 state->whitespace_error);
5037 if (state->update_index) {
5038 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5039 if (res) {
5040 error(_("Unable to write new index file"));
5041 res = -128;
5042 goto end;
5046 res = !!errs;
5048 end:
5049 rollback_lock_file(&state->lock_file);
5051 if (state->apply_verbosity <= verbosity_silent) {
5052 set_error_routine(state->saved_error_routine);
5053 set_warn_routine(state->saved_warn_routine);
5056 if (res > -1)
5057 return res;
5058 return (res == -1 ? 1 : 128);
5061 int apply_parse_options(int argc, const char **argv,
5062 struct apply_state *state,
5063 int *force_apply, int *options,
5064 const char * const *apply_usage)
5066 struct option builtin_apply_options[] = {
5067 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5068 N_("don't apply changes matching the given path"),
5069 PARSE_OPT_NONEG, apply_option_parse_exclude),
5070 OPT_CALLBACK_F(0, "include", state, N_("path"),
5071 N_("apply changes matching the given path"),
5072 PARSE_OPT_NONEG, apply_option_parse_include),
5073 OPT_CALLBACK('p', NULL, state, N_("num"),
5074 N_("remove <num> leading slashes from traditional diff paths"),
5075 apply_option_parse_p),
5076 OPT_BOOL(0, "no-add", &state->no_add,
5077 N_("ignore additions made by the patch")),
5078 OPT_BOOL(0, "stat", &state->diffstat,
5079 N_("instead of applying the patch, output diffstat for the input")),
5080 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5081 OPT_NOOP_NOARG(0, "binary"),
5082 OPT_BOOL(0, "numstat", &state->numstat,
5083 N_("show number of added and deleted lines in decimal notation")),
5084 OPT_BOOL(0, "summary", &state->summary,
5085 N_("instead of applying the patch, output a summary for the input")),
5086 OPT_BOOL(0, "check", &state->check,
5087 N_("instead of applying the patch, see if the patch is applicable")),
5088 OPT_BOOL(0, "index", &state->check_index,
5089 N_("make sure the patch is applicable to the current index")),
5090 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5091 N_("mark new files with `git add --intent-to-add`")),
5092 OPT_BOOL(0, "cached", &state->cached,
5093 N_("apply a patch without touching the working tree")),
5094 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5095 N_("accept a patch that touches outside the working area"),
5096 PARSE_OPT_NOCOMPLETE),
5097 OPT_BOOL(0, "apply", force_apply,
5098 N_("also apply the patch (use with --stat/--summary/--check)")),
5099 OPT_BOOL('3', "3way", &state->threeway,
5100 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5101 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5102 N_("build a temporary index based on embedded index information")),
5103 /* Think twice before adding "--nul" synonym to this */
5104 OPT_SET_INT('z', NULL, &state->line_termination,
5105 N_("paths are separated with NUL character"), '\0'),
5106 OPT_INTEGER('C', NULL, &state->p_context,
5107 N_("ensure at least <n> lines of context match")),
5108 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5109 N_("detect new or modified lines that have whitespace errors"),
5110 apply_option_parse_whitespace),
5111 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5112 N_("ignore changes in whitespace when finding context"),
5113 PARSE_OPT_NOARG, apply_option_parse_space_change),
5114 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5115 N_("ignore changes in whitespace when finding context"),
5116 PARSE_OPT_NOARG, apply_option_parse_space_change),
5117 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5118 N_("apply the patch in reverse")),
5119 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5120 N_("don't expect at least one line of context")),
5121 OPT_BOOL(0, "reject", &state->apply_with_reject,
5122 N_("leave the rejected hunks in corresponding *.rej files")),
5123 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5124 N_("allow overlapping hunks")),
5125 OPT__VERBOSITY(&state->apply_verbosity),
5126 OPT_BIT(0, "inaccurate-eof", options,
5127 N_("tolerate incorrectly detected missing new-line at the end of file"),
5128 APPLY_OPT_INACCURATE_EOF),
5129 OPT_BIT(0, "recount", options,
5130 N_("do not trust the line counts in the hunk headers"),
5131 APPLY_OPT_RECOUNT),
5132 OPT_CALLBACK(0, "directory", state, N_("root"),
5133 N_("prepend <root> to all filenames"),
5134 apply_option_parse_directory),
5135 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5136 N_("don't return error for empty patches")),
5137 OPT_END()
5140 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);