add-patch: handle "* Unmerged path" lines
[git/debian.git] / apply.c
blob8c7ae3f714b9c2c801ad083a90e8a309c98fe49e
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
10 #include "cache.h"
11 #include "config.h"
12 #include "object-store.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "diff.h"
16 #include "dir.h"
17 #include "xdiff-interface.h"
18 #include "ll-merge.h"
19 #include "lockfile.h"
20 #include "parse-options.h"
21 #include "quote.h"
22 #include "rerere.h"
23 #include "apply.h"
24 #include "entry.h"
26 struct gitdiff_data {
27 struct strbuf *root;
28 int linenr;
29 int p_value;
32 static void git_apply_config(void)
34 git_config_get_string("apply.whitespace", &apply_default_whitespace);
35 git_config_get_string("apply.ignorewhitespace", &apply_default_ignorewhitespace);
36 git_config(git_xmerge_config, NULL);
39 static int parse_whitespace_option(struct apply_state *state, const char *option)
41 if (!option) {
42 state->ws_error_action = warn_on_ws_error;
43 return 0;
45 if (!strcmp(option, "warn")) {
46 state->ws_error_action = warn_on_ws_error;
47 return 0;
49 if (!strcmp(option, "nowarn")) {
50 state->ws_error_action = nowarn_ws_error;
51 return 0;
53 if (!strcmp(option, "error")) {
54 state->ws_error_action = die_on_ws_error;
55 return 0;
57 if (!strcmp(option, "error-all")) {
58 state->ws_error_action = die_on_ws_error;
59 state->squelch_whitespace_errors = 0;
60 return 0;
62 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
63 state->ws_error_action = correct_ws_error;
64 return 0;
67 * Please update $__git_whitespacelist in git-completion.bash
68 * when you add new options.
70 return error(_("unrecognized whitespace option '%s'"), option);
73 static int parse_ignorewhitespace_option(struct apply_state *state,
74 const char *option)
76 if (!option || !strcmp(option, "no") ||
77 !strcmp(option, "false") || !strcmp(option, "never") ||
78 !strcmp(option, "none")) {
79 state->ws_ignore_action = ignore_ws_none;
80 return 0;
82 if (!strcmp(option, "change")) {
83 state->ws_ignore_action = ignore_ws_change;
84 return 0;
86 return error(_("unrecognized whitespace ignore option '%s'"), option);
89 int init_apply_state(struct apply_state *state,
90 struct repository *repo,
91 const char *prefix)
93 memset(state, 0, sizeof(*state));
94 state->prefix = prefix;
95 state->repo = repo;
96 state->apply = 1;
97 state->line_termination = '\n';
98 state->p_value = 1;
99 state->p_context = UINT_MAX;
100 state->squelch_whitespace_errors = 5;
101 state->ws_error_action = warn_on_ws_error;
102 state->ws_ignore_action = ignore_ws_none;
103 state->linenr = 1;
104 string_list_init_nodup(&state->fn_table);
105 string_list_init_nodup(&state->limit_by_name);
106 strset_init(&state->removed_symlinks);
107 strset_init(&state->kept_symlinks);
108 strbuf_init(&state->root, 0);
110 git_apply_config();
111 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
112 return -1;
113 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
114 return -1;
115 return 0;
118 void clear_apply_state(struct apply_state *state)
120 string_list_clear(&state->limit_by_name, 0);
121 strset_clear(&state->removed_symlinks);
122 strset_clear(&state->kept_symlinks);
123 strbuf_release(&state->root);
125 /* &state->fn_table is cleared at the end of apply_patch() */
128 static void mute_routine(const char *msg UNUSED, va_list params UNUSED)
130 /* do nothing */
133 int check_apply_state(struct apply_state *state, int force_apply)
135 int is_not_gitdir = !startup_info->have_repository;
137 if (state->apply_with_reject && state->threeway)
138 return error(_("options '%s' and '%s' cannot be used together"), "--reject", "--3way");
139 if (state->threeway) {
140 if (is_not_gitdir)
141 return error(_("'%s' outside a repository"), "--3way");
142 state->check_index = 1;
144 if (state->apply_with_reject) {
145 state->apply = 1;
146 if (state->apply_verbosity == verbosity_normal)
147 state->apply_verbosity = verbosity_verbose;
149 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
150 state->apply = 0;
151 if (state->check_index && is_not_gitdir)
152 return error(_("'%s' outside a repository"), "--index");
153 if (state->cached) {
154 if (is_not_gitdir)
155 return error(_("'%s' outside a repository"), "--cached");
156 state->check_index = 1;
158 if (state->ita_only && (state->check_index || is_not_gitdir))
159 state->ita_only = 0;
160 if (state->check_index)
161 state->unsafe_paths = 0;
163 if (state->apply_verbosity <= verbosity_silent) {
164 state->saved_error_routine = get_error_routine();
165 state->saved_warn_routine = get_warn_routine();
166 set_error_routine(mute_routine);
167 set_warn_routine(mute_routine);
170 return 0;
173 static void set_default_whitespace_mode(struct apply_state *state)
175 if (!state->whitespace_option && !apply_default_whitespace)
176 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
180 * This represents one "hunk" from a patch, starting with
181 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
182 * patch text is pointed at by patch, and its byte length
183 * is stored in size. leading and trailing are the number
184 * of context lines.
186 struct fragment {
187 unsigned long leading, trailing;
188 unsigned long oldpos, oldlines;
189 unsigned long newpos, newlines;
191 * 'patch' is usually borrowed from buf in apply_patch(),
192 * but some codepaths store an allocated buffer.
194 const char *patch;
195 unsigned free_patch:1,
196 rejected:1;
197 int size;
198 int linenr;
199 struct fragment *next;
203 * When dealing with a binary patch, we reuse "leading" field
204 * to store the type of the binary hunk, either deflated "delta"
205 * or deflated "literal".
207 #define binary_patch_method leading
208 #define BINARY_DELTA_DEFLATED 1
209 #define BINARY_LITERAL_DEFLATED 2
211 static void free_fragment_list(struct fragment *list)
213 while (list) {
214 struct fragment *next = list->next;
215 if (list->free_patch)
216 free((char *)list->patch);
217 free(list);
218 list = next;
222 void release_patch(struct patch *patch)
224 free_fragment_list(patch->fragments);
225 free(patch->def_name);
226 free(patch->old_name);
227 free(patch->new_name);
228 free(patch->result);
231 static void free_patch(struct patch *patch)
233 release_patch(patch);
234 free(patch);
237 static void free_patch_list(struct patch *list)
239 while (list) {
240 struct patch *next = list->next;
241 free_patch(list);
242 list = next;
247 * A line in a file, len-bytes long (includes the terminating LF,
248 * except for an incomplete line at the end if the file ends with
249 * one), and its contents hashes to 'hash'.
251 struct line {
252 size_t len;
253 unsigned hash : 24;
254 unsigned flag : 8;
255 #define LINE_COMMON 1
256 #define LINE_PATCHED 2
260 * This represents a "file", which is an array of "lines".
262 struct image {
263 char *buf;
264 size_t len;
265 size_t nr;
266 size_t alloc;
267 struct line *line_allocated;
268 struct line *line;
271 static uint32_t hash_line(const char *cp, size_t len)
273 size_t i;
274 uint32_t h;
275 for (i = 0, h = 0; i < len; i++) {
276 if (!isspace(cp[i])) {
277 h = h * 3 + (cp[i] & 0xff);
280 return h;
284 * Compare lines s1 of length n1 and s2 of length n2, ignoring
285 * whitespace difference. Returns 1 if they match, 0 otherwise
287 static int fuzzy_matchlines(const char *s1, size_t n1,
288 const char *s2, size_t n2)
290 const char *end1 = s1 + n1;
291 const char *end2 = s2 + n2;
293 /* ignore line endings */
294 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
295 end1--;
296 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
297 end2--;
299 while (s1 < end1 && s2 < end2) {
300 if (isspace(*s1)) {
302 * Skip whitespace. We check on both buffers
303 * because we don't want "a b" to match "ab".
305 if (!isspace(*s2))
306 return 0;
307 while (s1 < end1 && isspace(*s1))
308 s1++;
309 while (s2 < end2 && isspace(*s2))
310 s2++;
311 } else if (*s1++ != *s2++)
312 return 0;
315 /* If we reached the end on one side only, lines don't match. */
316 return s1 == end1 && s2 == end2;
319 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
321 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
322 img->line_allocated[img->nr].len = len;
323 img->line_allocated[img->nr].hash = hash_line(bol, len);
324 img->line_allocated[img->nr].flag = flag;
325 img->nr++;
329 * "buf" has the file contents to be patched (read from various sources).
330 * attach it to "image" and add line-based index to it.
331 * "image" now owns the "buf".
333 static void prepare_image(struct image *image, char *buf, size_t len,
334 int prepare_linetable)
336 const char *cp, *ep;
338 memset(image, 0, sizeof(*image));
339 image->buf = buf;
340 image->len = len;
342 if (!prepare_linetable)
343 return;
345 ep = image->buf + image->len;
346 cp = image->buf;
347 while (cp < ep) {
348 const char *next;
349 for (next = cp; next < ep && *next != '\n'; next++)
351 if (next < ep)
352 next++;
353 add_line_info(image, cp, next - cp, 0);
354 cp = next;
356 image->line = image->line_allocated;
359 static void clear_image(struct image *image)
361 free(image->buf);
362 free(image->line_allocated);
363 memset(image, 0, sizeof(*image));
366 /* fmt must contain _one_ %s and no other substitution */
367 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
369 struct strbuf sb = STRBUF_INIT;
371 if (patch->old_name && patch->new_name &&
372 strcmp(patch->old_name, patch->new_name)) {
373 quote_c_style(patch->old_name, &sb, NULL, 0);
374 strbuf_addstr(&sb, " => ");
375 quote_c_style(patch->new_name, &sb, NULL, 0);
376 } else {
377 const char *n = patch->new_name;
378 if (!n)
379 n = patch->old_name;
380 quote_c_style(n, &sb, NULL, 0);
382 fprintf(output, fmt, sb.buf);
383 fputc('\n', output);
384 strbuf_release(&sb);
387 #define SLOP (16)
390 * apply.c isn't equipped to handle arbitrarily large patches, because
391 * it intermingles `unsigned long` with `int` for the type used to store
392 * buffer lengths.
394 * Only process patches that are just shy of 1 GiB large in order to
395 * avoid any truncation or overflow issues.
397 #define MAX_APPLY_SIZE (1024UL * 1024 * 1023)
399 static int read_patch_file(struct strbuf *sb, int fd)
401 if (strbuf_read(sb, fd, 0) < 0 || sb->len >= MAX_APPLY_SIZE)
402 return error_errno("git apply: failed to read");
405 * Make sure that we have some slop in the buffer
406 * so that we can do speculative "memcmp" etc, and
407 * see to it that it is NUL-filled.
409 strbuf_grow(sb, SLOP);
410 memset(sb->buf + sb->len, 0, SLOP);
411 return 0;
414 static unsigned long linelen(const char *buffer, unsigned long size)
416 unsigned long len = 0;
417 while (size--) {
418 len++;
419 if (*buffer++ == '\n')
420 break;
422 return len;
425 static int is_dev_null(const char *str)
427 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
430 #define TERM_SPACE 1
431 #define TERM_TAB 2
433 static int name_terminate(int c, int terminate)
435 if (c == ' ' && !(terminate & TERM_SPACE))
436 return 0;
437 if (c == '\t' && !(terminate & TERM_TAB))
438 return 0;
440 return 1;
443 /* remove double slashes to make --index work with such filenames */
444 static char *squash_slash(char *name)
446 int i = 0, j = 0;
448 if (!name)
449 return NULL;
451 while (name[i]) {
452 if ((name[j++] = name[i++]) == '/')
453 while (name[i] == '/')
454 i++;
456 name[j] = '\0';
457 return name;
460 static char *find_name_gnu(struct strbuf *root,
461 const char *line,
462 int p_value)
464 struct strbuf name = STRBUF_INIT;
465 char *cp;
468 * Proposed "new-style" GNU patch/diff format; see
469 * https://lore.kernel.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
471 if (unquote_c_style(&name, line, NULL)) {
472 strbuf_release(&name);
473 return NULL;
476 for (cp = name.buf; p_value; p_value--) {
477 cp = strchr(cp, '/');
478 if (!cp) {
479 strbuf_release(&name);
480 return NULL;
482 cp++;
485 strbuf_remove(&name, 0, cp - name.buf);
486 if (root->len)
487 strbuf_insert(&name, 0, root->buf, root->len);
488 return squash_slash(strbuf_detach(&name, NULL));
491 static size_t sane_tz_len(const char *line, size_t len)
493 const char *tz, *p;
495 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
496 return 0;
497 tz = line + len - strlen(" +0500");
499 if (tz[1] != '+' && tz[1] != '-')
500 return 0;
502 for (p = tz + 2; p != line + len; p++)
503 if (!isdigit(*p))
504 return 0;
506 return line + len - tz;
509 static size_t tz_with_colon_len(const char *line, size_t len)
511 const char *tz, *p;
513 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
514 return 0;
515 tz = line + len - strlen(" +08:00");
517 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
518 return 0;
519 p = tz + 2;
520 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
521 !isdigit(*p++) || !isdigit(*p++))
522 return 0;
524 return line + len - tz;
527 static size_t date_len(const char *line, size_t len)
529 const char *date, *p;
531 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
532 return 0;
533 p = date = line + len - strlen("72-02-05");
535 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
536 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
537 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
538 return 0;
540 if (date - line >= strlen("19") &&
541 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
542 date -= strlen("19");
544 return line + len - date;
547 static size_t short_time_len(const char *line, size_t len)
549 const char *time, *p;
551 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
552 return 0;
553 p = time = line + len - strlen(" 07:01:32");
555 /* Permit 1-digit hours? */
556 if (*p++ != ' ' ||
557 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
558 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
559 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
560 return 0;
562 return line + len - time;
565 static size_t fractional_time_len(const char *line, size_t len)
567 const char *p;
568 size_t n;
570 /* Expected format: 19:41:17.620000023 */
571 if (!len || !isdigit(line[len - 1]))
572 return 0;
573 p = line + len - 1;
575 /* Fractional seconds. */
576 while (p > line && isdigit(*p))
577 p--;
578 if (*p != '.')
579 return 0;
581 /* Hours, minutes, and whole seconds. */
582 n = short_time_len(line, p - line);
583 if (!n)
584 return 0;
586 return line + len - p + n;
589 static size_t trailing_spaces_len(const char *line, size_t len)
591 const char *p;
593 /* Expected format: ' ' x (1 or more) */
594 if (!len || line[len - 1] != ' ')
595 return 0;
597 p = line + len;
598 while (p != line) {
599 p--;
600 if (*p != ' ')
601 return line + len - (p + 1);
604 /* All spaces! */
605 return len;
608 static size_t diff_timestamp_len(const char *line, size_t len)
610 const char *end = line + len;
611 size_t n;
614 * Posix: 2010-07-05 19:41:17
615 * GNU: 2010-07-05 19:41:17.620000023 -0500
618 if (!isdigit(end[-1]))
619 return 0;
621 n = sane_tz_len(line, end - line);
622 if (!n)
623 n = tz_with_colon_len(line, end - line);
624 end -= n;
626 n = short_time_len(line, end - line);
627 if (!n)
628 n = fractional_time_len(line, end - line);
629 end -= n;
631 n = date_len(line, end - line);
632 if (!n) /* No date. Too bad. */
633 return 0;
634 end -= n;
636 if (end == line) /* No space before date. */
637 return 0;
638 if (end[-1] == '\t') { /* Success! */
639 end--;
640 return line + len - end;
642 if (end[-1] != ' ') /* No space before date. */
643 return 0;
645 /* Whitespace damage. */
646 end -= trailing_spaces_len(line, end - line);
647 return line + len - end;
650 static char *find_name_common(struct strbuf *root,
651 const char *line,
652 const char *def,
653 int p_value,
654 const char *end,
655 int terminate)
657 int len;
658 const char *start = NULL;
660 if (p_value == 0)
661 start = line;
662 while (line != end) {
663 char c = *line;
665 if (!end && isspace(c)) {
666 if (c == '\n')
667 break;
668 if (name_terminate(c, terminate))
669 break;
671 line++;
672 if (c == '/' && !--p_value)
673 start = line;
675 if (!start)
676 return squash_slash(xstrdup_or_null(def));
677 len = line - start;
678 if (!len)
679 return squash_slash(xstrdup_or_null(def));
682 * Generally we prefer the shorter name, especially
683 * if the other one is just a variation of that with
684 * something else tacked on to the end (ie "file.orig"
685 * or "file~").
687 if (def) {
688 int deflen = strlen(def);
689 if (deflen < len && !strncmp(start, def, deflen))
690 return squash_slash(xstrdup(def));
693 if (root->len) {
694 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
695 return squash_slash(ret);
698 return squash_slash(xmemdupz(start, len));
701 static char *find_name(struct strbuf *root,
702 const char *line,
703 char *def,
704 int p_value,
705 int terminate)
707 if (*line == '"') {
708 char *name = find_name_gnu(root, line, p_value);
709 if (name)
710 return name;
713 return find_name_common(root, line, def, p_value, NULL, terminate);
716 static char *find_name_traditional(struct strbuf *root,
717 const char *line,
718 char *def,
719 int p_value)
721 size_t len;
722 size_t date_len;
724 if (*line == '"') {
725 char *name = find_name_gnu(root, line, p_value);
726 if (name)
727 return name;
730 len = strchrnul(line, '\n') - line;
731 date_len = diff_timestamp_len(line, len);
732 if (!date_len)
733 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
734 len -= date_len;
736 return find_name_common(root, line, def, p_value, line + len, 0);
740 * Given the string after "--- " or "+++ ", guess the appropriate
741 * p_value for the given patch.
743 static int guess_p_value(struct apply_state *state, const char *nameline)
745 char *name, *cp;
746 int val = -1;
748 if (is_dev_null(nameline))
749 return -1;
750 name = find_name_traditional(&state->root, nameline, NULL, 0);
751 if (!name)
752 return -1;
753 cp = strchr(name, '/');
754 if (!cp)
755 val = 0;
756 else if (state->prefix) {
758 * Does it begin with "a/$our-prefix" and such? Then this is
759 * very likely to apply to our directory.
761 if (starts_with(name, state->prefix))
762 val = count_slashes(state->prefix);
763 else {
764 cp++;
765 if (starts_with(cp, state->prefix))
766 val = count_slashes(state->prefix) + 1;
769 free(name);
770 return val;
774 * Does the ---/+++ line have the POSIX timestamp after the last HT?
775 * GNU diff puts epoch there to signal a creation/deletion event. Is
776 * this such a timestamp?
778 static int has_epoch_timestamp(const char *nameline)
781 * We are only interested in epoch timestamp; any non-zero
782 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
783 * For the same reason, the date must be either 1969-12-31 or
784 * 1970-01-01, and the seconds part must be "00".
786 const char stamp_regexp[] =
787 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
789 "([-+][0-2][0-9]:?[0-5][0-9])\n";
790 const char *timestamp = NULL, *cp, *colon;
791 static regex_t *stamp;
792 regmatch_t m[10];
793 int zoneoffset, epoch_hour, hour, minute;
794 int status;
796 for (cp = nameline; *cp != '\n'; cp++) {
797 if (*cp == '\t')
798 timestamp = cp + 1;
800 if (!timestamp)
801 return 0;
804 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
805 * (west of GMT) or 1970-01-01 (east of GMT)
807 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
808 epoch_hour = 24;
809 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
810 epoch_hour = 0;
811 else
812 return 0;
814 if (!stamp) {
815 stamp = xmalloc(sizeof(*stamp));
816 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
817 warning(_("Cannot prepare timestamp regexp %s"),
818 stamp_regexp);
819 return 0;
823 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
824 if (status) {
825 if (status != REG_NOMATCH)
826 warning(_("regexec returned %d for input: %s"),
827 status, timestamp);
828 return 0;
831 hour = strtol(timestamp, NULL, 10);
832 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
834 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
835 if (*colon == ':')
836 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
837 else
838 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
839 if (timestamp[m[3].rm_so] == '-')
840 zoneoffset = -zoneoffset;
842 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
846 * Get the name etc info from the ---/+++ lines of a traditional patch header
848 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
849 * files, we can happily check the index for a match, but for creating a
850 * new file we should try to match whatever "patch" does. I have no idea.
852 static int parse_traditional_patch(struct apply_state *state,
853 const char *first,
854 const char *second,
855 struct patch *patch)
857 char *name;
859 first += 4; /* skip "--- " */
860 second += 4; /* skip "+++ " */
861 if (!state->p_value_known) {
862 int p, q;
863 p = guess_p_value(state, first);
864 q = guess_p_value(state, second);
865 if (p < 0) p = q;
866 if (0 <= p && p == q) {
867 state->p_value = p;
868 state->p_value_known = 1;
871 if (is_dev_null(first)) {
872 patch->is_new = 1;
873 patch->is_delete = 0;
874 name = find_name_traditional(&state->root, second, NULL, state->p_value);
875 patch->new_name = name;
876 } else if (is_dev_null(second)) {
877 patch->is_new = 0;
878 patch->is_delete = 1;
879 name = find_name_traditional(&state->root, first, NULL, state->p_value);
880 patch->old_name = name;
881 } else {
882 char *first_name;
883 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
884 name = find_name_traditional(&state->root, second, first_name, state->p_value);
885 free(first_name);
886 if (has_epoch_timestamp(first)) {
887 patch->is_new = 1;
888 patch->is_delete = 0;
889 patch->new_name = name;
890 } else if (has_epoch_timestamp(second)) {
891 patch->is_new = 0;
892 patch->is_delete = 1;
893 patch->old_name = name;
894 } else {
895 patch->old_name = name;
896 patch->new_name = xstrdup_or_null(name);
899 if (!name)
900 return error(_("unable to find filename in patch at line %d"), state->linenr);
902 return 0;
905 static int gitdiff_hdrend(struct gitdiff_data *state UNUSED,
906 const char *line UNUSED,
907 struct patch *patch UNUSED)
909 return 1;
913 * We're anal about diff header consistency, to make
914 * sure that we don't end up having strange ambiguous
915 * patches floating around.
917 * As a result, gitdiff_{old|new}name() will check
918 * their names against any previous information, just
919 * to make sure..
921 #define DIFF_OLD_NAME 0
922 #define DIFF_NEW_NAME 1
924 static int gitdiff_verify_name(struct gitdiff_data *state,
925 const char *line,
926 int isnull,
927 char **name,
928 int side)
930 if (!*name && !isnull) {
931 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
932 return 0;
935 if (*name) {
936 char *another;
937 if (isnull)
938 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
939 *name, state->linenr);
940 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
941 if (!another || strcmp(another, *name)) {
942 free(another);
943 return error((side == DIFF_NEW_NAME) ?
944 _("git apply: bad git-diff - inconsistent new filename on line %d") :
945 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
947 free(another);
948 } else {
949 if (!is_dev_null(line))
950 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
953 return 0;
956 static int gitdiff_oldname(struct gitdiff_data *state,
957 const char *line,
958 struct patch *patch)
960 return gitdiff_verify_name(state, line,
961 patch->is_new, &patch->old_name,
962 DIFF_OLD_NAME);
965 static int gitdiff_newname(struct gitdiff_data *state,
966 const char *line,
967 struct patch *patch)
969 return gitdiff_verify_name(state, line,
970 patch->is_delete, &patch->new_name,
971 DIFF_NEW_NAME);
974 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
976 char *end;
977 *mode = strtoul(line, &end, 8);
978 if (end == line || !isspace(*end))
979 return error(_("invalid mode on line %d: %s"), linenr, line);
980 return 0;
983 static int gitdiff_oldmode(struct gitdiff_data *state,
984 const char *line,
985 struct patch *patch)
987 return parse_mode_line(line, state->linenr, &patch->old_mode);
990 static int gitdiff_newmode(struct gitdiff_data *state,
991 const char *line,
992 struct patch *patch)
994 return parse_mode_line(line, state->linenr, &patch->new_mode);
997 static int gitdiff_delete(struct gitdiff_data *state,
998 const char *line,
999 struct patch *patch)
1001 patch->is_delete = 1;
1002 free(patch->old_name);
1003 patch->old_name = xstrdup_or_null(patch->def_name);
1004 return gitdiff_oldmode(state, line, patch);
1007 static int gitdiff_newfile(struct gitdiff_data *state,
1008 const char *line,
1009 struct patch *patch)
1011 patch->is_new = 1;
1012 free(patch->new_name);
1013 patch->new_name = xstrdup_or_null(patch->def_name);
1014 return gitdiff_newmode(state, line, patch);
1017 static int gitdiff_copysrc(struct gitdiff_data *state,
1018 const char *line,
1019 struct patch *patch)
1021 patch->is_copy = 1;
1022 free(patch->old_name);
1023 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1024 return 0;
1027 static int gitdiff_copydst(struct gitdiff_data *state,
1028 const char *line,
1029 struct patch *patch)
1031 patch->is_copy = 1;
1032 free(patch->new_name);
1033 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1034 return 0;
1037 static int gitdiff_renamesrc(struct gitdiff_data *state,
1038 const char *line,
1039 struct patch *patch)
1041 patch->is_rename = 1;
1042 free(patch->old_name);
1043 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1044 return 0;
1047 static int gitdiff_renamedst(struct gitdiff_data *state,
1048 const char *line,
1049 struct patch *patch)
1051 patch->is_rename = 1;
1052 free(patch->new_name);
1053 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1054 return 0;
1057 static int gitdiff_similarity(struct gitdiff_data *state UNUSED,
1058 const char *line,
1059 struct patch *patch)
1061 unsigned long val = strtoul(line, NULL, 10);
1062 if (val <= 100)
1063 patch->score = val;
1064 return 0;
1067 static int gitdiff_dissimilarity(struct gitdiff_data *state UNUSED,
1068 const char *line,
1069 struct patch *patch)
1071 unsigned long val = strtoul(line, NULL, 10);
1072 if (val <= 100)
1073 patch->score = val;
1074 return 0;
1077 static int gitdiff_index(struct gitdiff_data *state,
1078 const char *line,
1079 struct patch *patch)
1082 * index line is N hexadecimal, "..", N hexadecimal,
1083 * and optional space with octal mode.
1085 const char *ptr, *eol;
1086 int len;
1087 const unsigned hexsz = the_hash_algo->hexsz;
1089 ptr = strchr(line, '.');
1090 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1091 return 0;
1092 len = ptr - line;
1093 memcpy(patch->old_oid_prefix, line, len);
1094 patch->old_oid_prefix[len] = 0;
1096 line = ptr + 2;
1097 ptr = strchr(line, ' ');
1098 eol = strchrnul(line, '\n');
1100 if (!ptr || eol < ptr)
1101 ptr = eol;
1102 len = ptr - line;
1104 if (hexsz < len)
1105 return 0;
1106 memcpy(patch->new_oid_prefix, line, len);
1107 patch->new_oid_prefix[len] = 0;
1108 if (*ptr == ' ')
1109 return gitdiff_oldmode(state, ptr + 1, patch);
1110 return 0;
1114 * This is normal for a diff that doesn't change anything: we'll fall through
1115 * into the next diff. Tell the parser to break out.
1117 static int gitdiff_unrecognized(struct gitdiff_data *state UNUSED,
1118 const char *line UNUSED,
1119 struct patch *patch UNUSED)
1121 return 1;
1125 * Skip p_value leading components from "line"; as we do not accept
1126 * absolute paths, return NULL in that case.
1128 static const char *skip_tree_prefix(int p_value,
1129 const char *line,
1130 int llen)
1132 int nslash;
1133 int i;
1135 if (!p_value)
1136 return (llen && line[0] == '/') ? NULL : line;
1138 nslash = p_value;
1139 for (i = 0; i < llen; i++) {
1140 int ch = line[i];
1141 if (ch == '/' && --nslash <= 0)
1142 return (i == 0) ? NULL : &line[i + 1];
1144 return NULL;
1148 * This is to extract the same name that appears on "diff --git"
1149 * line. We do not find and return anything if it is a rename
1150 * patch, and it is OK because we will find the name elsewhere.
1151 * We need to reliably find name only when it is mode-change only,
1152 * creation or deletion of an empty file. In any of these cases,
1153 * both sides are the same name under a/ and b/ respectively.
1155 static char *git_header_name(int p_value,
1156 const char *line,
1157 int llen)
1159 const char *name;
1160 const char *second = NULL;
1161 size_t len, line_len;
1163 line += strlen("diff --git ");
1164 llen -= strlen("diff --git ");
1166 if (*line == '"') {
1167 const char *cp;
1168 struct strbuf first = STRBUF_INIT;
1169 struct strbuf sp = STRBUF_INIT;
1171 if (unquote_c_style(&first, line, &second))
1172 goto free_and_fail1;
1174 /* strip the a/b prefix including trailing slash */
1175 cp = skip_tree_prefix(p_value, first.buf, first.len);
1176 if (!cp)
1177 goto free_and_fail1;
1178 strbuf_remove(&first, 0, cp - first.buf);
1181 * second points at one past closing dq of name.
1182 * find the second name.
1184 while ((second < line + llen) && isspace(*second))
1185 second++;
1187 if (line + llen <= second)
1188 goto free_and_fail1;
1189 if (*second == '"') {
1190 if (unquote_c_style(&sp, second, NULL))
1191 goto free_and_fail1;
1192 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1193 if (!cp)
1194 goto free_and_fail1;
1195 /* They must match, otherwise ignore */
1196 if (strcmp(cp, first.buf))
1197 goto free_and_fail1;
1198 strbuf_release(&sp);
1199 return strbuf_detach(&first, NULL);
1202 /* unquoted second */
1203 cp = skip_tree_prefix(p_value, second, line + llen - second);
1204 if (!cp)
1205 goto free_and_fail1;
1206 if (line + llen - cp != first.len ||
1207 memcmp(first.buf, cp, first.len))
1208 goto free_and_fail1;
1209 return strbuf_detach(&first, NULL);
1211 free_and_fail1:
1212 strbuf_release(&first);
1213 strbuf_release(&sp);
1214 return NULL;
1217 /* unquoted first name */
1218 name = skip_tree_prefix(p_value, line, llen);
1219 if (!name)
1220 return NULL;
1223 * since the first name is unquoted, a dq if exists must be
1224 * the beginning of the second name.
1226 for (second = name; second < line + llen; second++) {
1227 if (*second == '"') {
1228 struct strbuf sp = STRBUF_INIT;
1229 const char *np;
1231 if (unquote_c_style(&sp, second, NULL))
1232 goto free_and_fail2;
1234 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1235 if (!np)
1236 goto free_and_fail2;
1238 len = sp.buf + sp.len - np;
1239 if (len < second - name &&
1240 !strncmp(np, name, len) &&
1241 isspace(name[len])) {
1242 /* Good */
1243 strbuf_remove(&sp, 0, np - sp.buf);
1244 return strbuf_detach(&sp, NULL);
1247 free_and_fail2:
1248 strbuf_release(&sp);
1249 return NULL;
1254 * Accept a name only if it shows up twice, exactly the same
1255 * form.
1257 second = strchr(name, '\n');
1258 if (!second)
1259 return NULL;
1260 line_len = second - name;
1261 for (len = 0 ; ; len++) {
1262 switch (name[len]) {
1263 default:
1264 continue;
1265 case '\n':
1266 return NULL;
1267 case '\t': case ' ':
1269 * Is this the separator between the preimage
1270 * and the postimage pathname? Again, we are
1271 * only interested in the case where there is
1272 * no rename, as this is only to set def_name
1273 * and a rename patch has the names elsewhere
1274 * in an unambiguous form.
1276 if (!name[len + 1])
1277 return NULL; /* no postimage name */
1278 second = skip_tree_prefix(p_value, name + len + 1,
1279 line_len - (len + 1));
1280 if (!second)
1281 return NULL;
1283 * Does len bytes starting at "name" and "second"
1284 * (that are separated by one HT or SP we just
1285 * found) exactly match?
1287 if (second[len] == '\n' && !strncmp(name, second, len))
1288 return xmemdupz(name, len);
1293 static int check_header_line(int linenr, struct patch *patch)
1295 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1296 (patch->is_rename == 1) + (patch->is_copy == 1);
1297 if (extensions > 1)
1298 return error(_("inconsistent header lines %d and %d"),
1299 patch->extension_linenr, linenr);
1300 if (extensions && !patch->extension_linenr)
1301 patch->extension_linenr = linenr;
1302 return 0;
1305 int parse_git_diff_header(struct strbuf *root,
1306 int *linenr,
1307 int p_value,
1308 const char *line,
1309 int len,
1310 unsigned int size,
1311 struct patch *patch)
1313 unsigned long offset;
1314 struct gitdiff_data parse_hdr_state;
1316 /* A git diff has explicit new/delete information, so we don't guess */
1317 patch->is_new = 0;
1318 patch->is_delete = 0;
1321 * Some things may not have the old name in the
1322 * rest of the headers anywhere (pure mode changes,
1323 * or removing or adding empty files), so we get
1324 * the default name from the header.
1326 patch->def_name = git_header_name(p_value, line, len);
1327 if (patch->def_name && root->len) {
1328 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1329 free(patch->def_name);
1330 patch->def_name = s;
1333 line += len;
1334 size -= len;
1335 (*linenr)++;
1336 parse_hdr_state.root = root;
1337 parse_hdr_state.linenr = *linenr;
1338 parse_hdr_state.p_value = p_value;
1340 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1341 static const struct opentry {
1342 const char *str;
1343 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1344 } optable[] = {
1345 { "@@ -", gitdiff_hdrend },
1346 { "--- ", gitdiff_oldname },
1347 { "+++ ", gitdiff_newname },
1348 { "old mode ", gitdiff_oldmode },
1349 { "new mode ", gitdiff_newmode },
1350 { "deleted file mode ", gitdiff_delete },
1351 { "new file mode ", gitdiff_newfile },
1352 { "copy from ", gitdiff_copysrc },
1353 { "copy to ", gitdiff_copydst },
1354 { "rename old ", gitdiff_renamesrc },
1355 { "rename new ", gitdiff_renamedst },
1356 { "rename from ", gitdiff_renamesrc },
1357 { "rename to ", gitdiff_renamedst },
1358 { "similarity index ", gitdiff_similarity },
1359 { "dissimilarity index ", gitdiff_dissimilarity },
1360 { "index ", gitdiff_index },
1361 { "", gitdiff_unrecognized },
1363 int i;
1365 len = linelen(line, size);
1366 if (!len || line[len-1] != '\n')
1367 break;
1368 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1369 const struct opentry *p = optable + i;
1370 int oplen = strlen(p->str);
1371 int res;
1372 if (len < oplen || memcmp(p->str, line, oplen))
1373 continue;
1374 res = p->fn(&parse_hdr_state, line + oplen, patch);
1375 if (res < 0)
1376 return -1;
1377 if (check_header_line(*linenr, patch))
1378 return -1;
1379 if (res > 0)
1380 goto done;
1381 break;
1385 done:
1386 if (!patch->old_name && !patch->new_name) {
1387 if (!patch->def_name) {
1388 error(Q_("git diff header lacks filename information when removing "
1389 "%d leading pathname component (line %d)",
1390 "git diff header lacks filename information when removing "
1391 "%d leading pathname components (line %d)",
1392 parse_hdr_state.p_value),
1393 parse_hdr_state.p_value, *linenr);
1394 return -128;
1396 patch->old_name = xstrdup(patch->def_name);
1397 patch->new_name = xstrdup(patch->def_name);
1399 if ((!patch->new_name && !patch->is_delete) ||
1400 (!patch->old_name && !patch->is_new)) {
1401 error(_("git diff header lacks filename information "
1402 "(line %d)"), *linenr);
1403 return -128;
1405 patch->is_toplevel_relative = 1;
1406 return offset;
1409 static int parse_num(const char *line, unsigned long *p)
1411 char *ptr;
1413 if (!isdigit(*line))
1414 return 0;
1415 *p = strtoul(line, &ptr, 10);
1416 return ptr - line;
1419 static int parse_range(const char *line, int len, int offset, const char *expect,
1420 unsigned long *p1, unsigned long *p2)
1422 int digits, ex;
1424 if (offset < 0 || offset >= len)
1425 return -1;
1426 line += offset;
1427 len -= offset;
1429 digits = parse_num(line, p1);
1430 if (!digits)
1431 return -1;
1433 offset += digits;
1434 line += digits;
1435 len -= digits;
1437 *p2 = 1;
1438 if (*line == ',') {
1439 digits = parse_num(line+1, p2);
1440 if (!digits)
1441 return -1;
1443 offset += digits+1;
1444 line += digits+1;
1445 len -= digits+1;
1448 ex = strlen(expect);
1449 if (ex > len)
1450 return -1;
1451 if (memcmp(line, expect, ex))
1452 return -1;
1454 return offset + ex;
1457 static void recount_diff(const char *line, int size, struct fragment *fragment)
1459 int oldlines = 0, newlines = 0, ret = 0;
1461 if (size < 1) {
1462 warning("recount: ignore empty hunk");
1463 return;
1466 for (;;) {
1467 int len = linelen(line, size);
1468 size -= len;
1469 line += len;
1471 if (size < 1)
1472 break;
1474 switch (*line) {
1475 case ' ': case '\n':
1476 newlines++;
1477 /* fall through */
1478 case '-':
1479 oldlines++;
1480 continue;
1481 case '+':
1482 newlines++;
1483 continue;
1484 case '\\':
1485 continue;
1486 case '@':
1487 ret = size < 3 || !starts_with(line, "@@ ");
1488 break;
1489 case 'd':
1490 ret = size < 5 || !starts_with(line, "diff ");
1491 break;
1492 default:
1493 ret = -1;
1494 break;
1496 if (ret) {
1497 warning(_("recount: unexpected line: %.*s"),
1498 (int)linelen(line, size), line);
1499 return;
1501 break;
1503 fragment->oldlines = oldlines;
1504 fragment->newlines = newlines;
1508 * Parse a unified diff fragment header of the
1509 * form "@@ -a,b +c,d @@"
1511 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1513 int offset;
1515 if (!len || line[len-1] != '\n')
1516 return -1;
1518 /* Figure out the number of lines in a fragment */
1519 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1520 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1522 return offset;
1526 * Find file diff header
1528 * Returns:
1529 * -1 if no header was found
1530 * -128 in case of error
1531 * the size of the header in bytes (called "offset") otherwise
1533 static int find_header(struct apply_state *state,
1534 const char *line,
1535 unsigned long size,
1536 int *hdrsize,
1537 struct patch *patch)
1539 unsigned long offset, len;
1541 patch->is_toplevel_relative = 0;
1542 patch->is_rename = patch->is_copy = 0;
1543 patch->is_new = patch->is_delete = -1;
1544 patch->old_mode = patch->new_mode = 0;
1545 patch->old_name = patch->new_name = NULL;
1546 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1547 unsigned long nextlen;
1549 len = linelen(line, size);
1550 if (!len)
1551 break;
1553 /* Testing this early allows us to take a few shortcuts.. */
1554 if (len < 6)
1555 continue;
1558 * Make sure we don't find any unconnected patch fragments.
1559 * That's a sign that we didn't find a header, and that a
1560 * patch has become corrupted/broken up.
1562 if (!memcmp("@@ -", line, 4)) {
1563 struct fragment dummy;
1564 if (parse_fragment_header(line, len, &dummy) < 0)
1565 continue;
1566 error(_("patch fragment without header at line %d: %.*s"),
1567 state->linenr, (int)len-1, line);
1568 return -128;
1571 if (size < len + 6)
1572 break;
1575 * Git patch? It might not have a real patch, just a rename
1576 * or mode change, so we handle that specially
1578 if (!memcmp("diff --git ", line, 11)) {
1579 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1580 state->p_value, line, len,
1581 size, patch);
1582 if (git_hdr_len < 0)
1583 return -128;
1584 if (git_hdr_len <= len)
1585 continue;
1586 *hdrsize = git_hdr_len;
1587 return offset;
1590 /* --- followed by +++ ? */
1591 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1592 continue;
1595 * We only accept unified patches, so we want it to
1596 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1597 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1599 nextlen = linelen(line + len, size - len);
1600 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1601 continue;
1603 /* Ok, we'll consider it a patch */
1604 if (parse_traditional_patch(state, line, line+len, patch))
1605 return -128;
1606 *hdrsize = len + nextlen;
1607 state->linenr += 2;
1608 return offset;
1610 return -1;
1613 static void record_ws_error(struct apply_state *state,
1614 unsigned result,
1615 const char *line,
1616 int len,
1617 int linenr)
1619 char *err;
1621 if (!result)
1622 return;
1624 state->whitespace_error++;
1625 if (state->squelch_whitespace_errors &&
1626 state->squelch_whitespace_errors < state->whitespace_error)
1627 return;
1629 err = whitespace_error_string(result);
1630 if (state->apply_verbosity > verbosity_silent)
1631 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1632 state->patch_input_file, linenr, err, len, line);
1633 free(err);
1636 static void check_whitespace(struct apply_state *state,
1637 const char *line,
1638 int len,
1639 unsigned ws_rule)
1641 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1643 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1647 * Check if the patch has context lines with CRLF or
1648 * the patch wants to remove lines with CRLF.
1650 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1652 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1653 patch->ws_rule |= WS_CR_AT_EOL;
1654 patch->crlf_in_old = 1;
1660 * Parse a unified diff. Note that this really needs to parse each
1661 * fragment separately, since the only way to know the difference
1662 * between a "---" that is part of a patch, and a "---" that starts
1663 * the next patch is to look at the line counts..
1665 static int parse_fragment(struct apply_state *state,
1666 const char *line,
1667 unsigned long size,
1668 struct patch *patch,
1669 struct fragment *fragment)
1671 int added, deleted;
1672 int len = linelen(line, size), offset;
1673 unsigned long oldlines, newlines;
1674 unsigned long leading, trailing;
1676 offset = parse_fragment_header(line, len, fragment);
1677 if (offset < 0)
1678 return -1;
1679 if (offset > 0 && patch->recount)
1680 recount_diff(line + offset, size - offset, fragment);
1681 oldlines = fragment->oldlines;
1682 newlines = fragment->newlines;
1683 leading = 0;
1684 trailing = 0;
1686 /* Parse the thing.. */
1687 line += len;
1688 size -= len;
1689 state->linenr++;
1690 added = deleted = 0;
1691 for (offset = len;
1692 0 < size;
1693 offset += len, size -= len, line += len, state->linenr++) {
1694 if (!oldlines && !newlines)
1695 break;
1696 len = linelen(line, size);
1697 if (!len || line[len-1] != '\n')
1698 return -1;
1699 switch (*line) {
1700 default:
1701 return -1;
1702 case '\n': /* newer GNU diff, an empty context line */
1703 case ' ':
1704 oldlines--;
1705 newlines--;
1706 if (!deleted && !added)
1707 leading++;
1708 trailing++;
1709 check_old_for_crlf(patch, line, len);
1710 if (!state->apply_in_reverse &&
1711 state->ws_error_action == correct_ws_error)
1712 check_whitespace(state, line, len, patch->ws_rule);
1713 break;
1714 case '-':
1715 if (!state->apply_in_reverse)
1716 check_old_for_crlf(patch, line, len);
1717 if (state->apply_in_reverse &&
1718 state->ws_error_action != nowarn_ws_error)
1719 check_whitespace(state, line, len, patch->ws_rule);
1720 deleted++;
1721 oldlines--;
1722 trailing = 0;
1723 break;
1724 case '+':
1725 if (state->apply_in_reverse)
1726 check_old_for_crlf(patch, line, len);
1727 if (!state->apply_in_reverse &&
1728 state->ws_error_action != nowarn_ws_error)
1729 check_whitespace(state, line, len, patch->ws_rule);
1730 added++;
1731 newlines--;
1732 trailing = 0;
1733 break;
1736 * We allow "\ No newline at end of file". Depending
1737 * on locale settings when the patch was produced we
1738 * don't know what this line looks like. The only
1739 * thing we do know is that it begins with "\ ".
1740 * Checking for 12 is just for sanity check -- any
1741 * l10n of "\ No newline..." is at least that long.
1743 case '\\':
1744 if (len < 12 || memcmp(line, "\\ ", 2))
1745 return -1;
1746 break;
1749 if (oldlines || newlines)
1750 return -1;
1751 if (!patch->recount && !deleted && !added)
1752 return -1;
1754 fragment->leading = leading;
1755 fragment->trailing = trailing;
1758 * If a fragment ends with an incomplete line, we failed to include
1759 * it in the above loop because we hit oldlines == newlines == 0
1760 * before seeing it.
1762 if (12 < size && !memcmp(line, "\\ ", 2))
1763 offset += linelen(line, size);
1765 patch->lines_added += added;
1766 patch->lines_deleted += deleted;
1768 if (0 < patch->is_new && oldlines)
1769 return error(_("new file depends on old contents"));
1770 if (0 < patch->is_delete && newlines)
1771 return error(_("deleted file still has contents"));
1772 return offset;
1776 * We have seen "diff --git a/... b/..." header (or a traditional patch
1777 * header). Read hunks that belong to this patch into fragments and hang
1778 * them to the given patch structure.
1780 * The (fragment->patch, fragment->size) pair points into the memory given
1781 * by the caller, not a copy, when we return.
1783 * Returns:
1784 * -1 in case of error,
1785 * the number of bytes in the patch otherwise.
1787 static int parse_single_patch(struct apply_state *state,
1788 const char *line,
1789 unsigned long size,
1790 struct patch *patch)
1792 unsigned long offset = 0;
1793 unsigned long oldlines = 0, newlines = 0, context = 0;
1794 struct fragment **fragp = &patch->fragments;
1796 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1797 struct fragment *fragment;
1798 int len;
1800 CALLOC_ARRAY(fragment, 1);
1801 fragment->linenr = state->linenr;
1802 len = parse_fragment(state, line, size, patch, fragment);
1803 if (len <= 0) {
1804 free(fragment);
1805 return error(_("corrupt patch at line %d"), state->linenr);
1807 fragment->patch = line;
1808 fragment->size = len;
1809 oldlines += fragment->oldlines;
1810 newlines += fragment->newlines;
1811 context += fragment->leading + fragment->trailing;
1813 *fragp = fragment;
1814 fragp = &fragment->next;
1816 offset += len;
1817 line += len;
1818 size -= len;
1822 * If something was removed (i.e. we have old-lines) it cannot
1823 * be creation, and if something was added it cannot be
1824 * deletion. However, the reverse is not true; --unified=0
1825 * patches that only add are not necessarily creation even
1826 * though they do not have any old lines, and ones that only
1827 * delete are not necessarily deletion.
1829 * Unfortunately, a real creation/deletion patch do _not_ have
1830 * any context line by definition, so we cannot safely tell it
1831 * apart with --unified=0 insanity. At least if the patch has
1832 * more than one hunk it is not creation or deletion.
1834 if (patch->is_new < 0 &&
1835 (oldlines || (patch->fragments && patch->fragments->next)))
1836 patch->is_new = 0;
1837 if (patch->is_delete < 0 &&
1838 (newlines || (patch->fragments && patch->fragments->next)))
1839 patch->is_delete = 0;
1841 if (0 < patch->is_new && oldlines)
1842 return error(_("new file %s depends on old contents"), patch->new_name);
1843 if (0 < patch->is_delete && newlines)
1844 return error(_("deleted file %s still has contents"), patch->old_name);
1845 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1846 fprintf_ln(stderr,
1847 _("** warning: "
1848 "file %s becomes empty but is not deleted"),
1849 patch->new_name);
1851 return offset;
1854 static inline int metadata_changes(struct patch *patch)
1856 return patch->is_rename > 0 ||
1857 patch->is_copy > 0 ||
1858 patch->is_new > 0 ||
1859 patch->is_delete ||
1860 (patch->old_mode && patch->new_mode &&
1861 patch->old_mode != patch->new_mode);
1864 static char *inflate_it(const void *data, unsigned long size,
1865 unsigned long inflated_size)
1867 git_zstream stream;
1868 void *out;
1869 int st;
1871 memset(&stream, 0, sizeof(stream));
1873 stream.next_in = (unsigned char *)data;
1874 stream.avail_in = size;
1875 stream.next_out = out = xmalloc(inflated_size);
1876 stream.avail_out = inflated_size;
1877 git_inflate_init(&stream);
1878 st = git_inflate(&stream, Z_FINISH);
1879 git_inflate_end(&stream);
1880 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1881 free(out);
1882 return NULL;
1884 return out;
1888 * Read a binary hunk and return a new fragment; fragment->patch
1889 * points at an allocated memory that the caller must free, so
1890 * it is marked as "->free_patch = 1".
1892 static struct fragment *parse_binary_hunk(struct apply_state *state,
1893 char **buf_p,
1894 unsigned long *sz_p,
1895 int *status_p,
1896 int *used_p)
1899 * Expect a line that begins with binary patch method ("literal"
1900 * or "delta"), followed by the length of data before deflating.
1901 * a sequence of 'length-byte' followed by base-85 encoded data
1902 * should follow, terminated by a newline.
1904 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1905 * and we would limit the patch line to 66 characters,
1906 * so one line can fit up to 13 groups that would decode
1907 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1908 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1910 int llen, used;
1911 unsigned long size = *sz_p;
1912 char *buffer = *buf_p;
1913 int patch_method;
1914 unsigned long origlen;
1915 char *data = NULL;
1916 int hunk_size = 0;
1917 struct fragment *frag;
1919 llen = linelen(buffer, size);
1920 used = llen;
1922 *status_p = 0;
1924 if (starts_with(buffer, "delta ")) {
1925 patch_method = BINARY_DELTA_DEFLATED;
1926 origlen = strtoul(buffer + 6, NULL, 10);
1928 else if (starts_with(buffer, "literal ")) {
1929 patch_method = BINARY_LITERAL_DEFLATED;
1930 origlen = strtoul(buffer + 8, NULL, 10);
1932 else
1933 return NULL;
1935 state->linenr++;
1936 buffer += llen;
1937 size -= llen;
1938 while (1) {
1939 int byte_length, max_byte_length, newsize;
1940 llen = linelen(buffer, size);
1941 used += llen;
1942 state->linenr++;
1943 if (llen == 1) {
1944 /* consume the blank line */
1945 buffer++;
1946 size--;
1947 break;
1950 * Minimum line is "A00000\n" which is 7-byte long,
1951 * and the line length must be multiple of 5 plus 2.
1953 if ((llen < 7) || (llen-2) % 5)
1954 goto corrupt;
1955 max_byte_length = (llen - 2) / 5 * 4;
1956 byte_length = *buffer;
1957 if ('A' <= byte_length && byte_length <= 'Z')
1958 byte_length = byte_length - 'A' + 1;
1959 else if ('a' <= byte_length && byte_length <= 'z')
1960 byte_length = byte_length - 'a' + 27;
1961 else
1962 goto corrupt;
1963 /* if the input length was not multiple of 4, we would
1964 * have filler at the end but the filler should never
1965 * exceed 3 bytes
1967 if (max_byte_length < byte_length ||
1968 byte_length <= max_byte_length - 4)
1969 goto corrupt;
1970 newsize = hunk_size + byte_length;
1971 data = xrealloc(data, newsize);
1972 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1973 goto corrupt;
1974 hunk_size = newsize;
1975 buffer += llen;
1976 size -= llen;
1979 CALLOC_ARRAY(frag, 1);
1980 frag->patch = inflate_it(data, hunk_size, origlen);
1981 frag->free_patch = 1;
1982 if (!frag->patch)
1983 goto corrupt;
1984 free(data);
1985 frag->size = origlen;
1986 *buf_p = buffer;
1987 *sz_p = size;
1988 *used_p = used;
1989 frag->binary_patch_method = patch_method;
1990 return frag;
1992 corrupt:
1993 free(data);
1994 *status_p = -1;
1995 error(_("corrupt binary patch at line %d: %.*s"),
1996 state->linenr-1, llen-1, buffer);
1997 return NULL;
2001 * Returns:
2002 * -1 in case of error,
2003 * the length of the parsed binary patch otherwise
2005 static int parse_binary(struct apply_state *state,
2006 char *buffer,
2007 unsigned long size,
2008 struct patch *patch)
2011 * We have read "GIT binary patch\n"; what follows is a line
2012 * that says the patch method (currently, either "literal" or
2013 * "delta") and the length of data before deflating; a
2014 * sequence of 'length-byte' followed by base-85 encoded data
2015 * follows.
2017 * When a binary patch is reversible, there is another binary
2018 * hunk in the same format, starting with patch method (either
2019 * "literal" or "delta") with the length of data, and a sequence
2020 * of length-byte + base-85 encoded data, terminated with another
2021 * empty line. This data, when applied to the postimage, produces
2022 * the preimage.
2024 struct fragment *forward;
2025 struct fragment *reverse;
2026 int status;
2027 int used, used_1;
2029 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2030 if (!forward && !status)
2031 /* there has to be one hunk (forward hunk) */
2032 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2033 if (status)
2034 /* otherwise we already gave an error message */
2035 return status;
2037 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2038 if (reverse)
2039 used += used_1;
2040 else if (status) {
2042 * Not having reverse hunk is not an error, but having
2043 * a corrupt reverse hunk is.
2045 free((void*) forward->patch);
2046 free(forward);
2047 return status;
2049 forward->next = reverse;
2050 patch->fragments = forward;
2051 patch->is_binary = 1;
2052 return used;
2055 static void prefix_one(struct apply_state *state, char **name)
2057 char *old_name = *name;
2058 if (!old_name)
2059 return;
2060 *name = prefix_filename(state->prefix, *name);
2061 free(old_name);
2064 static void prefix_patch(struct apply_state *state, struct patch *p)
2066 if (!state->prefix || p->is_toplevel_relative)
2067 return;
2068 prefix_one(state, &p->new_name);
2069 prefix_one(state, &p->old_name);
2073 * include/exclude
2076 static void add_name_limit(struct apply_state *state,
2077 const char *name,
2078 int exclude)
2080 struct string_list_item *it;
2082 it = string_list_append(&state->limit_by_name, name);
2083 it->util = exclude ? NULL : (void *) 1;
2086 static int use_patch(struct apply_state *state, struct patch *p)
2088 const char *pathname = p->new_name ? p->new_name : p->old_name;
2089 int i;
2091 /* Paths outside are not touched regardless of "--include" */
2092 if (state->prefix && *state->prefix) {
2093 const char *rest;
2094 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2095 return 0;
2098 /* See if it matches any of exclude/include rule */
2099 for (i = 0; i < state->limit_by_name.nr; i++) {
2100 struct string_list_item *it = &state->limit_by_name.items[i];
2101 if (!wildmatch(it->string, pathname, 0))
2102 return (it->util != NULL);
2106 * If we had any include, a path that does not match any rule is
2107 * not used. Otherwise, we saw bunch of exclude rules (or none)
2108 * and such a path is used.
2110 return !state->has_include;
2114 * Read the patch text in "buffer" that extends for "size" bytes; stop
2115 * reading after seeing a single patch (i.e. changes to a single file).
2116 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2118 * Returns:
2119 * -1 if no header was found or parse_binary() failed,
2120 * -128 on another error,
2121 * the number of bytes consumed otherwise,
2122 * so that the caller can call us again for the next patch.
2124 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2126 int hdrsize, patchsize;
2127 int offset = find_header(state, buffer, size, &hdrsize, patch);
2129 if (offset < 0)
2130 return offset;
2132 prefix_patch(state, patch);
2134 if (!use_patch(state, patch))
2135 patch->ws_rule = 0;
2136 else if (patch->new_name)
2137 patch->ws_rule = whitespace_rule(state->repo->index,
2138 patch->new_name);
2139 else
2140 patch->ws_rule = whitespace_rule(state->repo->index,
2141 patch->old_name);
2143 patchsize = parse_single_patch(state,
2144 buffer + offset + hdrsize,
2145 size - offset - hdrsize,
2146 patch);
2148 if (patchsize < 0)
2149 return -128;
2151 if (!patchsize) {
2152 static const char git_binary[] = "GIT binary patch\n";
2153 int hd = hdrsize + offset;
2154 unsigned long llen = linelen(buffer + hd, size - hd);
2156 if (llen == sizeof(git_binary) - 1 &&
2157 !memcmp(git_binary, buffer + hd, llen)) {
2158 int used;
2159 state->linenr++;
2160 used = parse_binary(state, buffer + hd + llen,
2161 size - hd - llen, patch);
2162 if (used < 0)
2163 return -1;
2164 if (used)
2165 patchsize = used + llen;
2166 else
2167 patchsize = 0;
2169 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2170 static const char *binhdr[] = {
2171 "Binary files ",
2172 "Files ",
2173 NULL,
2175 int i;
2176 for (i = 0; binhdr[i]; i++) {
2177 int len = strlen(binhdr[i]);
2178 if (len < size - hd &&
2179 !memcmp(binhdr[i], buffer + hd, len)) {
2180 state->linenr++;
2181 patch->is_binary = 1;
2182 patchsize = llen;
2183 break;
2188 /* Empty patch cannot be applied if it is a text patch
2189 * without metadata change. A binary patch appears
2190 * empty to us here.
2192 if ((state->apply || state->check) &&
2193 (!patch->is_binary && !metadata_changes(patch))) {
2194 error(_("patch with only garbage at line %d"), state->linenr);
2195 return -128;
2199 return offset + hdrsize + patchsize;
2202 static void reverse_patches(struct patch *p)
2204 for (; p; p = p->next) {
2205 struct fragment *frag = p->fragments;
2207 SWAP(p->new_name, p->old_name);
2208 SWAP(p->new_mode, p->old_mode);
2209 SWAP(p->is_new, p->is_delete);
2210 SWAP(p->lines_added, p->lines_deleted);
2211 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2213 for (; frag; frag = frag->next) {
2214 SWAP(frag->newpos, frag->oldpos);
2215 SWAP(frag->newlines, frag->oldlines);
2220 static const char pluses[] =
2221 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2222 static const char minuses[]=
2223 "----------------------------------------------------------------------";
2225 static void show_stats(struct apply_state *state, struct patch *patch)
2227 struct strbuf qname = STRBUF_INIT;
2228 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2229 int max, add, del;
2231 quote_c_style(cp, &qname, NULL, 0);
2234 * "scale" the filename
2236 max = state->max_len;
2237 if (max > 50)
2238 max = 50;
2240 if (qname.len > max) {
2241 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2242 if (!cp)
2243 cp = qname.buf + qname.len + 3 - max;
2244 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2247 if (patch->is_binary) {
2248 printf(" %-*s | Bin\n", max, qname.buf);
2249 strbuf_release(&qname);
2250 return;
2253 printf(" %-*s |", max, qname.buf);
2254 strbuf_release(&qname);
2257 * scale the add/delete
2259 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2260 add = patch->lines_added;
2261 del = patch->lines_deleted;
2263 if (state->max_change > 0) {
2264 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2265 add = (add * max + state->max_change / 2) / state->max_change;
2266 del = total - add;
2268 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2269 add, pluses, del, minuses);
2272 static int read_old_data(struct stat *st, struct patch *patch,
2273 const char *path, struct strbuf *buf)
2275 int conv_flags = patch->crlf_in_old ?
2276 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2277 switch (st->st_mode & S_IFMT) {
2278 case S_IFLNK:
2279 if (strbuf_readlink(buf, path, st->st_size) < 0)
2280 return error(_("unable to read symlink %s"), path);
2281 return 0;
2282 case S_IFREG:
2283 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2284 return error(_("unable to open or read %s"), path);
2286 * "git apply" without "--index/--cached" should never look
2287 * at the index; the target file may not have been added to
2288 * the index yet, and we may not even be in any Git repository.
2289 * Pass NULL to convert_to_git() to stress this; the function
2290 * should never look at the index when explicit crlf option
2291 * is given.
2293 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2294 return 0;
2295 default:
2296 return -1;
2301 * Update the preimage, and the common lines in postimage,
2302 * from buffer buf of length len. If postlen is 0 the postimage
2303 * is updated in place, otherwise it's updated on a new buffer
2304 * of length postlen
2307 static void update_pre_post_images(struct image *preimage,
2308 struct image *postimage,
2309 char *buf,
2310 size_t len, size_t postlen)
2312 int i, ctx, reduced;
2313 char *new_buf, *old_buf, *fixed;
2314 struct image fixed_preimage;
2317 * Update the preimage with whitespace fixes. Note that we
2318 * are not losing preimage->buf -- apply_one_fragment() will
2319 * free "oldlines".
2321 prepare_image(&fixed_preimage, buf, len, 1);
2322 assert(postlen
2323 ? fixed_preimage.nr == preimage->nr
2324 : fixed_preimage.nr <= preimage->nr);
2325 for (i = 0; i < fixed_preimage.nr; i++)
2326 fixed_preimage.line[i].flag = preimage->line[i].flag;
2327 free(preimage->line_allocated);
2328 *preimage = fixed_preimage;
2331 * Adjust the common context lines in postimage. This can be
2332 * done in-place when we are shrinking it with whitespace
2333 * fixing, but needs a new buffer when ignoring whitespace or
2334 * expanding leading tabs to spaces.
2336 * We trust the caller to tell us if the update can be done
2337 * in place (postlen==0) or not.
2339 old_buf = postimage->buf;
2340 if (postlen)
2341 new_buf = postimage->buf = xmalloc(postlen);
2342 else
2343 new_buf = old_buf;
2344 fixed = preimage->buf;
2346 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2347 size_t l_len = postimage->line[i].len;
2348 if (!(postimage->line[i].flag & LINE_COMMON)) {
2349 /* an added line -- no counterparts in preimage */
2350 memmove(new_buf, old_buf, l_len);
2351 old_buf += l_len;
2352 new_buf += l_len;
2353 continue;
2356 /* a common context -- skip it in the original postimage */
2357 old_buf += l_len;
2359 /* and find the corresponding one in the fixed preimage */
2360 while (ctx < preimage->nr &&
2361 !(preimage->line[ctx].flag & LINE_COMMON)) {
2362 fixed += preimage->line[ctx].len;
2363 ctx++;
2367 * preimage is expected to run out, if the caller
2368 * fixed addition of trailing blank lines.
2370 if (preimage->nr <= ctx) {
2371 reduced++;
2372 continue;
2375 /* and copy it in, while fixing the line length */
2376 l_len = preimage->line[ctx].len;
2377 memcpy(new_buf, fixed, l_len);
2378 new_buf += l_len;
2379 fixed += l_len;
2380 postimage->line[i].len = l_len;
2381 ctx++;
2384 if (postlen
2385 ? postlen < new_buf - postimage->buf
2386 : postimage->len < new_buf - postimage->buf)
2387 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2388 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2390 /* Fix the length of the whole thing */
2391 postimage->len = new_buf - postimage->buf;
2392 postimage->nr -= reduced;
2395 static int line_by_line_fuzzy_match(struct image *img,
2396 struct image *preimage,
2397 struct image *postimage,
2398 unsigned long current,
2399 int current_lno,
2400 int preimage_limit)
2402 int i;
2403 size_t imgoff = 0;
2404 size_t preoff = 0;
2405 size_t postlen = postimage->len;
2406 size_t extra_chars;
2407 char *buf;
2408 char *preimage_eof;
2409 char *preimage_end;
2410 struct strbuf fixed;
2411 char *fixed_buf;
2412 size_t fixed_len;
2414 for (i = 0; i < preimage_limit; i++) {
2415 size_t prelen = preimage->line[i].len;
2416 size_t imglen = img->line[current_lno+i].len;
2418 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2419 preimage->buf + preoff, prelen))
2420 return 0;
2421 if (preimage->line[i].flag & LINE_COMMON)
2422 postlen += imglen - prelen;
2423 imgoff += imglen;
2424 preoff += prelen;
2428 * Ok, the preimage matches with whitespace fuzz.
2430 * imgoff now holds the true length of the target that
2431 * matches the preimage before the end of the file.
2433 * Count the number of characters in the preimage that fall
2434 * beyond the end of the file and make sure that all of them
2435 * are whitespace characters. (This can only happen if
2436 * we are removing blank lines at the end of the file.)
2438 buf = preimage_eof = preimage->buf + preoff;
2439 for ( ; i < preimage->nr; i++)
2440 preoff += preimage->line[i].len;
2441 preimage_end = preimage->buf + preoff;
2442 for ( ; buf < preimage_end; buf++)
2443 if (!isspace(*buf))
2444 return 0;
2447 * Update the preimage and the common postimage context
2448 * lines to use the same whitespace as the target.
2449 * If whitespace is missing in the target (i.e.
2450 * if the preimage extends beyond the end of the file),
2451 * use the whitespace from the preimage.
2453 extra_chars = preimage_end - preimage_eof;
2454 strbuf_init(&fixed, imgoff + extra_chars);
2455 strbuf_add(&fixed, img->buf + current, imgoff);
2456 strbuf_add(&fixed, preimage_eof, extra_chars);
2457 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2458 update_pre_post_images(preimage, postimage,
2459 fixed_buf, fixed_len, postlen);
2460 return 1;
2463 static int match_fragment(struct apply_state *state,
2464 struct image *img,
2465 struct image *preimage,
2466 struct image *postimage,
2467 unsigned long current,
2468 int current_lno,
2469 unsigned ws_rule,
2470 int match_beginning, int match_end)
2472 int i;
2473 char *fixed_buf, *buf, *orig, *target;
2474 struct strbuf fixed;
2475 size_t fixed_len, postlen;
2476 int preimage_limit;
2478 if (preimage->nr + current_lno <= img->nr) {
2480 * The hunk falls within the boundaries of img.
2482 preimage_limit = preimage->nr;
2483 if (match_end && (preimage->nr + current_lno != img->nr))
2484 return 0;
2485 } else if (state->ws_error_action == correct_ws_error &&
2486 (ws_rule & WS_BLANK_AT_EOF)) {
2488 * This hunk extends beyond the end of img, and we are
2489 * removing blank lines at the end of the file. This
2490 * many lines from the beginning of the preimage must
2491 * match with img, and the remainder of the preimage
2492 * must be blank.
2494 preimage_limit = img->nr - current_lno;
2495 } else {
2497 * The hunk extends beyond the end of the img and
2498 * we are not removing blanks at the end, so we
2499 * should reject the hunk at this position.
2501 return 0;
2504 if (match_beginning && current_lno)
2505 return 0;
2507 /* Quick hash check */
2508 for (i = 0; i < preimage_limit; i++)
2509 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2510 (preimage->line[i].hash != img->line[current_lno + i].hash))
2511 return 0;
2513 if (preimage_limit == preimage->nr) {
2515 * Do we have an exact match? If we were told to match
2516 * at the end, size must be exactly at current+fragsize,
2517 * otherwise current+fragsize must be still within the preimage,
2518 * and either case, the old piece should match the preimage
2519 * exactly.
2521 if ((match_end
2522 ? (current + preimage->len == img->len)
2523 : (current + preimage->len <= img->len)) &&
2524 !memcmp(img->buf + current, preimage->buf, preimage->len))
2525 return 1;
2526 } else {
2528 * The preimage extends beyond the end of img, so
2529 * there cannot be an exact match.
2531 * There must be one non-blank context line that match
2532 * a line before the end of img.
2534 char *buf_end;
2536 buf = preimage->buf;
2537 buf_end = buf;
2538 for (i = 0; i < preimage_limit; i++)
2539 buf_end += preimage->line[i].len;
2541 for ( ; buf < buf_end; buf++)
2542 if (!isspace(*buf))
2543 break;
2544 if (buf == buf_end)
2545 return 0;
2549 * No exact match. If we are ignoring whitespace, run a line-by-line
2550 * fuzzy matching. We collect all the line length information because
2551 * we need it to adjust whitespace if we match.
2553 if (state->ws_ignore_action == ignore_ws_change)
2554 return line_by_line_fuzzy_match(img, preimage, postimage,
2555 current, current_lno, preimage_limit);
2557 if (state->ws_error_action != correct_ws_error)
2558 return 0;
2561 * The hunk does not apply byte-by-byte, but the hash says
2562 * it might with whitespace fuzz. We weren't asked to
2563 * ignore whitespace, we were asked to correct whitespace
2564 * errors, so let's try matching after whitespace correction.
2566 * While checking the preimage against the target, whitespace
2567 * errors in both fixed, we count how large the corresponding
2568 * postimage needs to be. The postimage prepared by
2569 * apply_one_fragment() has whitespace errors fixed on added
2570 * lines already, but the common lines were propagated as-is,
2571 * which may become longer when their whitespace errors are
2572 * fixed.
2575 /* First count added lines in postimage */
2576 postlen = 0;
2577 for (i = 0; i < postimage->nr; i++) {
2578 if (!(postimage->line[i].flag & LINE_COMMON))
2579 postlen += postimage->line[i].len;
2583 * The preimage may extend beyond the end of the file,
2584 * but in this loop we will only handle the part of the
2585 * preimage that falls within the file.
2587 strbuf_init(&fixed, preimage->len + 1);
2588 orig = preimage->buf;
2589 target = img->buf + current;
2590 for (i = 0; i < preimage_limit; i++) {
2591 size_t oldlen = preimage->line[i].len;
2592 size_t tgtlen = img->line[current_lno + i].len;
2593 size_t fixstart = fixed.len;
2594 struct strbuf tgtfix;
2595 int match;
2597 /* Try fixing the line in the preimage */
2598 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2600 /* Try fixing the line in the target */
2601 strbuf_init(&tgtfix, tgtlen);
2602 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2605 * If they match, either the preimage was based on
2606 * a version before our tree fixed whitespace breakage,
2607 * or we are lacking a whitespace-fix patch the tree
2608 * the preimage was based on already had (i.e. target
2609 * has whitespace breakage, the preimage doesn't).
2610 * In either case, we are fixing the whitespace breakages
2611 * so we might as well take the fix together with their
2612 * real change.
2614 match = (tgtfix.len == fixed.len - fixstart &&
2615 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2616 fixed.len - fixstart));
2618 /* Add the length if this is common with the postimage */
2619 if (preimage->line[i].flag & LINE_COMMON)
2620 postlen += tgtfix.len;
2622 strbuf_release(&tgtfix);
2623 if (!match)
2624 goto unmatch_exit;
2626 orig += oldlen;
2627 target += tgtlen;
2632 * Now handle the lines in the preimage that falls beyond the
2633 * end of the file (if any). They will only match if they are
2634 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2635 * false).
2637 for ( ; i < preimage->nr; i++) {
2638 size_t fixstart = fixed.len; /* start of the fixed preimage */
2639 size_t oldlen = preimage->line[i].len;
2640 int j;
2642 /* Try fixing the line in the preimage */
2643 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2645 for (j = fixstart; j < fixed.len; j++)
2646 if (!isspace(fixed.buf[j]))
2647 goto unmatch_exit;
2649 orig += oldlen;
2653 * Yes, the preimage is based on an older version that still
2654 * has whitespace breakages unfixed, and fixing them makes the
2655 * hunk match. Update the context lines in the postimage.
2657 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2658 if (postlen < postimage->len)
2659 postlen = 0;
2660 update_pre_post_images(preimage, postimage,
2661 fixed_buf, fixed_len, postlen);
2662 return 1;
2664 unmatch_exit:
2665 strbuf_release(&fixed);
2666 return 0;
2669 static int find_pos(struct apply_state *state,
2670 struct image *img,
2671 struct image *preimage,
2672 struct image *postimage,
2673 int line,
2674 unsigned ws_rule,
2675 int match_beginning, int match_end)
2677 int i;
2678 unsigned long backwards, forwards, current;
2679 int backwards_lno, forwards_lno, current_lno;
2682 * When running with --allow-overlap, it is possible that a hunk is
2683 * seen that pretends to start at the beginning (but no longer does),
2684 * and that *still* needs to match the end. So trust `match_end` more
2685 * than `match_beginning`.
2687 if (state->allow_overlap && match_beginning && match_end &&
2688 img->nr - preimage->nr != 0)
2689 match_beginning = 0;
2692 * If match_beginning or match_end is specified, there is no
2693 * point starting from a wrong line that will never match and
2694 * wander around and wait for a match at the specified end.
2696 if (match_beginning)
2697 line = 0;
2698 else if (match_end)
2699 line = img->nr - preimage->nr;
2702 * Because the comparison is unsigned, the following test
2703 * will also take care of a negative line number that can
2704 * result when match_end and preimage is larger than the target.
2706 if ((size_t) line > img->nr)
2707 line = img->nr;
2709 current = 0;
2710 for (i = 0; i < line; i++)
2711 current += img->line[i].len;
2714 * There's probably some smart way to do this, but I'll leave
2715 * that to the smart and beautiful people. I'm simple and stupid.
2717 backwards = current;
2718 backwards_lno = line;
2719 forwards = current;
2720 forwards_lno = line;
2721 current_lno = line;
2723 for (i = 0; ; i++) {
2724 if (match_fragment(state, img, preimage, postimage,
2725 current, current_lno, ws_rule,
2726 match_beginning, match_end))
2727 return current_lno;
2729 again:
2730 if (backwards_lno == 0 && forwards_lno == img->nr)
2731 break;
2733 if (i & 1) {
2734 if (backwards_lno == 0) {
2735 i++;
2736 goto again;
2738 backwards_lno--;
2739 backwards -= img->line[backwards_lno].len;
2740 current = backwards;
2741 current_lno = backwards_lno;
2742 } else {
2743 if (forwards_lno == img->nr) {
2744 i++;
2745 goto again;
2747 forwards += img->line[forwards_lno].len;
2748 forwards_lno++;
2749 current = forwards;
2750 current_lno = forwards_lno;
2754 return -1;
2757 static void remove_first_line(struct image *img)
2759 img->buf += img->line[0].len;
2760 img->len -= img->line[0].len;
2761 img->line++;
2762 img->nr--;
2765 static void remove_last_line(struct image *img)
2767 img->len -= img->line[--img->nr].len;
2771 * The change from "preimage" and "postimage" has been found to
2772 * apply at applied_pos (counts in line numbers) in "img".
2773 * Update "img" to remove "preimage" and replace it with "postimage".
2775 static void update_image(struct apply_state *state,
2776 struct image *img,
2777 int applied_pos,
2778 struct image *preimage,
2779 struct image *postimage)
2782 * remove the copy of preimage at offset in img
2783 * and replace it with postimage
2785 int i, nr;
2786 size_t remove_count, insert_count, applied_at = 0;
2787 char *result;
2788 int preimage_limit;
2791 * If we are removing blank lines at the end of img,
2792 * the preimage may extend beyond the end.
2793 * If that is the case, we must be careful only to
2794 * remove the part of the preimage that falls within
2795 * the boundaries of img. Initialize preimage_limit
2796 * to the number of lines in the preimage that falls
2797 * within the boundaries.
2799 preimage_limit = preimage->nr;
2800 if (preimage_limit > img->nr - applied_pos)
2801 preimage_limit = img->nr - applied_pos;
2803 for (i = 0; i < applied_pos; i++)
2804 applied_at += img->line[i].len;
2806 remove_count = 0;
2807 for (i = 0; i < preimage_limit; i++)
2808 remove_count += img->line[applied_pos + i].len;
2809 insert_count = postimage->len;
2811 /* Adjust the contents */
2812 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2813 memcpy(result, img->buf, applied_at);
2814 memcpy(result + applied_at, postimage->buf, postimage->len);
2815 memcpy(result + applied_at + postimage->len,
2816 img->buf + (applied_at + remove_count),
2817 img->len - (applied_at + remove_count));
2818 free(img->buf);
2819 img->buf = result;
2820 img->len += insert_count - remove_count;
2821 result[img->len] = '\0';
2823 /* Adjust the line table */
2824 nr = img->nr + postimage->nr - preimage_limit;
2825 if (preimage_limit < postimage->nr) {
2827 * NOTE: this knows that we never call remove_first_line()
2828 * on anything other than pre/post image.
2830 REALLOC_ARRAY(img->line, nr);
2831 img->line_allocated = img->line;
2833 if (preimage_limit != postimage->nr)
2834 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2835 img->line + applied_pos + preimage_limit,
2836 img->nr - (applied_pos + preimage_limit));
2837 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2838 if (!state->allow_overlap)
2839 for (i = 0; i < postimage->nr; i++)
2840 img->line[applied_pos + i].flag |= LINE_PATCHED;
2841 img->nr = nr;
2845 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2846 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2847 * replace the part of "img" with "postimage" text.
2849 static int apply_one_fragment(struct apply_state *state,
2850 struct image *img, struct fragment *frag,
2851 int inaccurate_eof, unsigned ws_rule,
2852 int nth_fragment)
2854 int match_beginning, match_end;
2855 const char *patch = frag->patch;
2856 int size = frag->size;
2857 char *old, *oldlines;
2858 struct strbuf newlines;
2859 int new_blank_lines_at_end = 0;
2860 int found_new_blank_lines_at_end = 0;
2861 int hunk_linenr = frag->linenr;
2862 unsigned long leading, trailing;
2863 int pos, applied_pos;
2864 struct image preimage;
2865 struct image postimage;
2867 memset(&preimage, 0, sizeof(preimage));
2868 memset(&postimage, 0, sizeof(postimage));
2869 oldlines = xmalloc(size);
2870 strbuf_init(&newlines, size);
2872 old = oldlines;
2873 while (size > 0) {
2874 char first;
2875 int len = linelen(patch, size);
2876 int plen;
2877 int added_blank_line = 0;
2878 int is_blank_context = 0;
2879 size_t start;
2881 if (!len)
2882 break;
2885 * "plen" is how much of the line we should use for
2886 * the actual patch data. Normally we just remove the
2887 * first character on the line, but if the line is
2888 * followed by "\ No newline", then we also remove the
2889 * last one (which is the newline, of course).
2891 plen = len - 1;
2892 if (len < size && patch[len] == '\\')
2893 plen--;
2894 first = *patch;
2895 if (state->apply_in_reverse) {
2896 if (first == '-')
2897 first = '+';
2898 else if (first == '+')
2899 first = '-';
2902 switch (first) {
2903 case '\n':
2904 /* Newer GNU diff, empty context line */
2905 if (plen < 0)
2906 /* ... followed by '\No newline'; nothing */
2907 break;
2908 *old++ = '\n';
2909 strbuf_addch(&newlines, '\n');
2910 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2911 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2912 is_blank_context = 1;
2913 break;
2914 case ' ':
2915 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2916 ws_blank_line(patch + 1, plen))
2917 is_blank_context = 1;
2918 /* fallthrough */
2919 case '-':
2920 memcpy(old, patch + 1, plen);
2921 add_line_info(&preimage, old, plen,
2922 (first == ' ' ? LINE_COMMON : 0));
2923 old += plen;
2924 if (first == '-')
2925 break;
2926 /* fallthrough */
2927 case '+':
2928 /* --no-add does not add new lines */
2929 if (first == '+' && state->no_add)
2930 break;
2932 start = newlines.len;
2933 if (first != '+' ||
2934 !state->whitespace_error ||
2935 state->ws_error_action != correct_ws_error) {
2936 strbuf_add(&newlines, patch + 1, plen);
2938 else {
2939 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2941 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2942 (first == '+' ? 0 : LINE_COMMON));
2943 if (first == '+' &&
2944 (ws_rule & WS_BLANK_AT_EOF) &&
2945 ws_blank_line(patch + 1, plen))
2946 added_blank_line = 1;
2947 break;
2948 case '@': case '\\':
2949 /* Ignore it, we already handled it */
2950 break;
2951 default:
2952 if (state->apply_verbosity > verbosity_normal)
2953 error(_("invalid start of line: '%c'"), first);
2954 applied_pos = -1;
2955 goto out;
2957 if (added_blank_line) {
2958 if (!new_blank_lines_at_end)
2959 found_new_blank_lines_at_end = hunk_linenr;
2960 new_blank_lines_at_end++;
2962 else if (is_blank_context)
2964 else
2965 new_blank_lines_at_end = 0;
2966 patch += len;
2967 size -= len;
2968 hunk_linenr++;
2970 if (inaccurate_eof &&
2971 old > oldlines && old[-1] == '\n' &&
2972 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2973 old--;
2974 strbuf_setlen(&newlines, newlines.len - 1);
2975 preimage.line_allocated[preimage.nr - 1].len--;
2976 postimage.line_allocated[postimage.nr - 1].len--;
2979 leading = frag->leading;
2980 trailing = frag->trailing;
2983 * A hunk to change lines at the beginning would begin with
2984 * @@ -1,L +N,M @@
2985 * but we need to be careful. -U0 that inserts before the second
2986 * line also has this pattern.
2988 * And a hunk to add to an empty file would begin with
2989 * @@ -0,0 +N,M @@
2991 * In other words, a hunk that is (frag->oldpos <= 1) with or
2992 * without leading context must match at the beginning.
2994 match_beginning = (!frag->oldpos ||
2995 (frag->oldpos == 1 && !state->unidiff_zero));
2998 * A hunk without trailing lines must match at the end.
2999 * However, we simply cannot tell if a hunk must match end
3000 * from the lack of trailing lines if the patch was generated
3001 * with unidiff without any context.
3003 match_end = !state->unidiff_zero && !trailing;
3005 pos = frag->newpos ? (frag->newpos - 1) : 0;
3006 preimage.buf = oldlines;
3007 preimage.len = old - oldlines;
3008 postimage.buf = newlines.buf;
3009 postimage.len = newlines.len;
3010 preimage.line = preimage.line_allocated;
3011 postimage.line = postimage.line_allocated;
3013 for (;;) {
3015 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3016 ws_rule, match_beginning, match_end);
3018 if (applied_pos >= 0)
3019 break;
3021 /* Am I at my context limits? */
3022 if ((leading <= state->p_context) && (trailing <= state->p_context))
3023 break;
3024 if (match_beginning || match_end) {
3025 match_beginning = match_end = 0;
3026 continue;
3030 * Reduce the number of context lines; reduce both
3031 * leading and trailing if they are equal otherwise
3032 * just reduce the larger context.
3034 if (leading >= trailing) {
3035 remove_first_line(&preimage);
3036 remove_first_line(&postimage);
3037 pos--;
3038 leading--;
3040 if (trailing > leading) {
3041 remove_last_line(&preimage);
3042 remove_last_line(&postimage);
3043 trailing--;
3047 if (applied_pos >= 0) {
3048 if (new_blank_lines_at_end &&
3049 preimage.nr + applied_pos >= img->nr &&
3050 (ws_rule & WS_BLANK_AT_EOF) &&
3051 state->ws_error_action != nowarn_ws_error) {
3052 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3053 found_new_blank_lines_at_end);
3054 if (state->ws_error_action == correct_ws_error) {
3055 while (new_blank_lines_at_end--)
3056 remove_last_line(&postimage);
3059 * We would want to prevent write_out_results()
3060 * from taking place in apply_patch() that follows
3061 * the callchain led us here, which is:
3062 * apply_patch->check_patch_list->check_patch->
3063 * apply_data->apply_fragments->apply_one_fragment
3065 if (state->ws_error_action == die_on_ws_error)
3066 state->apply = 0;
3069 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3070 int offset = applied_pos - pos;
3071 if (state->apply_in_reverse)
3072 offset = 0 - offset;
3073 fprintf_ln(stderr,
3074 Q_("Hunk #%d succeeded at %d (offset %d line).",
3075 "Hunk #%d succeeded at %d (offset %d lines).",
3076 offset),
3077 nth_fragment, applied_pos + 1, offset);
3081 * Warn if it was necessary to reduce the number
3082 * of context lines.
3084 if ((leading != frag->leading ||
3085 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3086 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3087 " to apply fragment at %d"),
3088 leading, trailing, applied_pos+1);
3089 update_image(state, img, applied_pos, &preimage, &postimage);
3090 } else {
3091 if (state->apply_verbosity > verbosity_normal)
3092 error(_("while searching for:\n%.*s"),
3093 (int)(old - oldlines), oldlines);
3096 out:
3097 free(oldlines);
3098 strbuf_release(&newlines);
3099 free(preimage.line_allocated);
3100 free(postimage.line_allocated);
3102 return (applied_pos < 0);
3105 static int apply_binary_fragment(struct apply_state *state,
3106 struct image *img,
3107 struct patch *patch)
3109 struct fragment *fragment = patch->fragments;
3110 unsigned long len;
3111 void *dst;
3113 if (!fragment)
3114 return error(_("missing binary patch data for '%s'"),
3115 patch->new_name ?
3116 patch->new_name :
3117 patch->old_name);
3119 /* Binary patch is irreversible without the optional second hunk */
3120 if (state->apply_in_reverse) {
3121 if (!fragment->next)
3122 return error(_("cannot reverse-apply a binary patch "
3123 "without the reverse hunk to '%s'"),
3124 patch->new_name
3125 ? patch->new_name : patch->old_name);
3126 fragment = fragment->next;
3128 switch (fragment->binary_patch_method) {
3129 case BINARY_DELTA_DEFLATED:
3130 dst = patch_delta(img->buf, img->len, fragment->patch,
3131 fragment->size, &len);
3132 if (!dst)
3133 return -1;
3134 clear_image(img);
3135 img->buf = dst;
3136 img->len = len;
3137 return 0;
3138 case BINARY_LITERAL_DEFLATED:
3139 clear_image(img);
3140 img->len = fragment->size;
3141 img->buf = xmemdupz(fragment->patch, img->len);
3142 return 0;
3144 return -1;
3148 * Replace "img" with the result of applying the binary patch.
3149 * The binary patch data itself in patch->fragment is still kept
3150 * but the preimage prepared by the caller in "img" is freed here
3151 * or in the helper function apply_binary_fragment() this calls.
3153 static int apply_binary(struct apply_state *state,
3154 struct image *img,
3155 struct patch *patch)
3157 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3158 struct object_id oid;
3159 const unsigned hexsz = the_hash_algo->hexsz;
3162 * For safety, we require patch index line to contain
3163 * full hex textual object ID for old and new, at least for now.
3165 if (strlen(patch->old_oid_prefix) != hexsz ||
3166 strlen(patch->new_oid_prefix) != hexsz ||
3167 get_oid_hex(patch->old_oid_prefix, &oid) ||
3168 get_oid_hex(patch->new_oid_prefix, &oid))
3169 return error(_("cannot apply binary patch to '%s' "
3170 "without full index line"), name);
3172 if (patch->old_name) {
3174 * See if the old one matches what the patch
3175 * applies to.
3177 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3178 &oid);
3179 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3180 return error(_("the patch applies to '%s' (%s), "
3181 "which does not match the "
3182 "current contents."),
3183 name, oid_to_hex(&oid));
3185 else {
3186 /* Otherwise, the old one must be empty. */
3187 if (img->len)
3188 return error(_("the patch applies to an empty "
3189 "'%s' but it is not empty"), name);
3192 get_oid_hex(patch->new_oid_prefix, &oid);
3193 if (is_null_oid(&oid)) {
3194 clear_image(img);
3195 return 0; /* deletion patch */
3198 if (has_object(the_repository, &oid, 0)) {
3199 /* We already have the postimage */
3200 enum object_type type;
3201 unsigned long size;
3202 char *result;
3204 result = read_object_file(&oid, &type, &size);
3205 if (!result)
3206 return error(_("the necessary postimage %s for "
3207 "'%s' cannot be read"),
3208 patch->new_oid_prefix, name);
3209 clear_image(img);
3210 img->buf = result;
3211 img->len = size;
3212 } else {
3214 * We have verified buf matches the preimage;
3215 * apply the patch data to it, which is stored
3216 * in the patch->fragments->{patch,size}.
3218 if (apply_binary_fragment(state, img, patch))
3219 return error(_("binary patch does not apply to '%s'"),
3220 name);
3222 /* verify that the result matches */
3223 hash_object_file(the_hash_algo, img->buf, img->len, OBJ_BLOB,
3224 &oid);
3225 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3226 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3227 name, patch->new_oid_prefix, oid_to_hex(&oid));
3230 return 0;
3233 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3235 struct fragment *frag = patch->fragments;
3236 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3237 unsigned ws_rule = patch->ws_rule;
3238 unsigned inaccurate_eof = patch->inaccurate_eof;
3239 int nth = 0;
3241 if (patch->is_binary)
3242 return apply_binary(state, img, patch);
3244 while (frag) {
3245 nth++;
3246 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3247 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3248 if (!state->apply_with_reject)
3249 return -1;
3250 frag->rejected = 1;
3252 frag = frag->next;
3254 return 0;
3257 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3259 if (S_ISGITLINK(mode)) {
3260 strbuf_grow(buf, 100);
3261 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3262 } else {
3263 enum object_type type;
3264 unsigned long sz;
3265 char *result;
3267 result = read_object_file(oid, &type, &sz);
3268 if (!result)
3269 return -1;
3270 /* XXX read_sha1_file NUL-terminates */
3271 strbuf_attach(buf, result, sz, sz + 1);
3273 return 0;
3276 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3278 if (!ce)
3279 return 0;
3280 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3283 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3285 struct string_list_item *item;
3287 if (!name)
3288 return NULL;
3290 item = string_list_lookup(&state->fn_table, name);
3291 if (item)
3292 return (struct patch *)item->util;
3294 return NULL;
3298 * item->util in the filename table records the status of the path.
3299 * Usually it points at a patch (whose result records the contents
3300 * of it after applying it), but it could be PATH_WAS_DELETED for a
3301 * path that a previously applied patch has already removed, or
3302 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3304 * The latter is needed to deal with a case where two paths A and B
3305 * are swapped by first renaming A to B and then renaming B to A;
3306 * moving A to B should not be prevented due to presence of B as we
3307 * will remove it in a later patch.
3309 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3310 #define PATH_WAS_DELETED ((struct patch *) -1)
3312 static int to_be_deleted(struct patch *patch)
3314 return patch == PATH_TO_BE_DELETED;
3317 static int was_deleted(struct patch *patch)
3319 return patch == PATH_WAS_DELETED;
3322 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3324 struct string_list_item *item;
3327 * Always add new_name unless patch is a deletion
3328 * This should cover the cases for normal diffs,
3329 * file creations and copies
3331 if (patch->new_name) {
3332 item = string_list_insert(&state->fn_table, patch->new_name);
3333 item->util = patch;
3337 * store a failure on rename/deletion cases because
3338 * later chunks shouldn't patch old names
3340 if ((patch->new_name == NULL) || (patch->is_rename)) {
3341 item = string_list_insert(&state->fn_table, patch->old_name);
3342 item->util = PATH_WAS_DELETED;
3346 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3349 * store information about incoming file deletion
3351 while (patch) {
3352 if ((patch->new_name == NULL) || (patch->is_rename)) {
3353 struct string_list_item *item;
3354 item = string_list_insert(&state->fn_table, patch->old_name);
3355 item->util = PATH_TO_BE_DELETED;
3357 patch = patch->next;
3361 static int checkout_target(struct index_state *istate,
3362 struct cache_entry *ce, struct stat *st)
3364 struct checkout costate = CHECKOUT_INIT;
3366 costate.refresh_cache = 1;
3367 costate.istate = istate;
3368 if (checkout_entry(ce, &costate, NULL, NULL) ||
3369 lstat(ce->name, st))
3370 return error(_("cannot checkout %s"), ce->name);
3371 return 0;
3374 static struct patch *previous_patch(struct apply_state *state,
3375 struct patch *patch,
3376 int *gone)
3378 struct patch *previous;
3380 *gone = 0;
3381 if (patch->is_copy || patch->is_rename)
3382 return NULL; /* "git" patches do not depend on the order */
3384 previous = in_fn_table(state, patch->old_name);
3385 if (!previous)
3386 return NULL;
3388 if (to_be_deleted(previous))
3389 return NULL; /* the deletion hasn't happened yet */
3391 if (was_deleted(previous))
3392 *gone = 1;
3394 return previous;
3397 static int verify_index_match(struct apply_state *state,
3398 const struct cache_entry *ce,
3399 struct stat *st)
3401 if (S_ISGITLINK(ce->ce_mode)) {
3402 if (!S_ISDIR(st->st_mode))
3403 return -1;
3404 return 0;
3406 return ie_match_stat(state->repo->index, ce, st,
3407 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3410 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3412 static int load_patch_target(struct apply_state *state,
3413 struct strbuf *buf,
3414 const struct cache_entry *ce,
3415 struct stat *st,
3416 struct patch *patch,
3417 const char *name,
3418 unsigned expected_mode)
3420 if (state->cached || state->check_index) {
3421 if (read_file_or_gitlink(ce, buf))
3422 return error(_("failed to read %s"), name);
3423 } else if (name) {
3424 if (S_ISGITLINK(expected_mode)) {
3425 if (ce)
3426 return read_file_or_gitlink(ce, buf);
3427 else
3428 return SUBMODULE_PATCH_WITHOUT_INDEX;
3429 } else if (has_symlink_leading_path(name, strlen(name))) {
3430 return error(_("reading from '%s' beyond a symbolic link"), name);
3431 } else {
3432 if (read_old_data(st, patch, name, buf))
3433 return error(_("failed to read %s"), name);
3436 return 0;
3440 * We are about to apply "patch"; populate the "image" with the
3441 * current version we have, from the working tree or from the index,
3442 * depending on the situation e.g. --cached/--index. If we are
3443 * applying a non-git patch that incrementally updates the tree,
3444 * we read from the result of a previous diff.
3446 static int load_preimage(struct apply_state *state,
3447 struct image *image,
3448 struct patch *patch, struct stat *st,
3449 const struct cache_entry *ce)
3451 struct strbuf buf = STRBUF_INIT;
3452 size_t len;
3453 char *img;
3454 struct patch *previous;
3455 int status;
3457 previous = previous_patch(state, patch, &status);
3458 if (status)
3459 return error(_("path %s has been renamed/deleted"),
3460 patch->old_name);
3461 if (previous) {
3462 /* We have a patched copy in memory; use that. */
3463 strbuf_add(&buf, previous->result, previous->resultsize);
3464 } else {
3465 status = load_patch_target(state, &buf, ce, st, patch,
3466 patch->old_name, patch->old_mode);
3467 if (status < 0)
3468 return status;
3469 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3471 * There is no way to apply subproject
3472 * patch without looking at the index.
3473 * NEEDSWORK: shouldn't this be flagged
3474 * as an error???
3476 free_fragment_list(patch->fragments);
3477 patch->fragments = NULL;
3478 } else if (status) {
3479 return error(_("failed to read %s"), patch->old_name);
3483 img = strbuf_detach(&buf, &len);
3484 prepare_image(image, img, len, !patch->is_binary);
3485 return 0;
3488 static int resolve_to(struct image *image, const struct object_id *result_id)
3490 unsigned long size;
3491 enum object_type type;
3493 clear_image(image);
3495 image->buf = read_object_file(result_id, &type, &size);
3496 if (!image->buf || type != OBJ_BLOB)
3497 die("unable to read blob object %s", oid_to_hex(result_id));
3498 image->len = size;
3500 return 0;
3503 static int three_way_merge(struct apply_state *state,
3504 struct image *image,
3505 char *path,
3506 const struct object_id *base,
3507 const struct object_id *ours,
3508 const struct object_id *theirs)
3510 mmfile_t base_file, our_file, their_file;
3511 mmbuffer_t result = { NULL };
3512 enum ll_merge_result status;
3514 /* resolve trivial cases first */
3515 if (oideq(base, ours))
3516 return resolve_to(image, theirs);
3517 else if (oideq(base, theirs) || oideq(ours, theirs))
3518 return resolve_to(image, ours);
3520 read_mmblob(&base_file, base);
3521 read_mmblob(&our_file, ours);
3522 read_mmblob(&their_file, theirs);
3523 status = ll_merge(&result, path,
3524 &base_file, "base",
3525 &our_file, "ours",
3526 &their_file, "theirs",
3527 state->repo->index,
3528 NULL);
3529 if (status == LL_MERGE_BINARY_CONFLICT)
3530 warning("Cannot merge binary files: %s (%s vs. %s)",
3531 path, "ours", "theirs");
3532 free(base_file.ptr);
3533 free(our_file.ptr);
3534 free(their_file.ptr);
3535 if (status < 0 || !result.ptr) {
3536 free(result.ptr);
3537 return -1;
3539 clear_image(image);
3540 image->buf = result.ptr;
3541 image->len = result.size;
3543 return status;
3547 * When directly falling back to add/add three-way merge, we read from
3548 * the current contents of the new_name. In no cases other than that
3549 * this function will be called.
3551 static int load_current(struct apply_state *state,
3552 struct image *image,
3553 struct patch *patch)
3555 struct strbuf buf = STRBUF_INIT;
3556 int status, pos;
3557 size_t len;
3558 char *img;
3559 struct stat st;
3560 struct cache_entry *ce;
3561 char *name = patch->new_name;
3562 unsigned mode = patch->new_mode;
3564 if (!patch->is_new)
3565 BUG("patch to %s is not a creation", patch->old_name);
3567 pos = index_name_pos(state->repo->index, name, strlen(name));
3568 if (pos < 0)
3569 return error(_("%s: does not exist in index"), name);
3570 ce = state->repo->index->cache[pos];
3571 if (lstat(name, &st)) {
3572 if (errno != ENOENT)
3573 return error_errno("%s", name);
3574 if (checkout_target(state->repo->index, ce, &st))
3575 return -1;
3577 if (verify_index_match(state, ce, &st))
3578 return error(_("%s: does not match index"), name);
3580 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3581 if (status < 0)
3582 return status;
3583 else if (status)
3584 return -1;
3585 img = strbuf_detach(&buf, &len);
3586 prepare_image(image, img, len, !patch->is_binary);
3587 return 0;
3590 static int try_threeway(struct apply_state *state,
3591 struct image *image,
3592 struct patch *patch,
3593 struct stat *st,
3594 const struct cache_entry *ce)
3596 struct object_id pre_oid, post_oid, our_oid;
3597 struct strbuf buf = STRBUF_INIT;
3598 size_t len;
3599 int status;
3600 char *img;
3601 struct image tmp_image;
3603 /* No point falling back to 3-way merge in these cases */
3604 if (patch->is_delete ||
3605 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode) ||
3606 (patch->is_new && !patch->direct_to_threeway) ||
3607 (patch->is_rename && !patch->lines_added && !patch->lines_deleted))
3608 return -1;
3610 /* Preimage the patch was prepared for */
3611 if (patch->is_new)
3612 write_object_file("", 0, OBJ_BLOB, &pre_oid);
3613 else if (get_oid(patch->old_oid_prefix, &pre_oid) ||
3614 read_blob_object(&buf, &pre_oid, patch->old_mode))
3615 return error(_("repository lacks the necessary blob to perform 3-way merge."));
3617 if (state->apply_verbosity > verbosity_silent && patch->direct_to_threeway)
3618 fprintf(stderr, _("Performing three-way merge...\n"));
3620 img = strbuf_detach(&buf, &len);
3621 prepare_image(&tmp_image, img, len, 1);
3622 /* Apply the patch to get the post image */
3623 if (apply_fragments(state, &tmp_image, patch) < 0) {
3624 clear_image(&tmp_image);
3625 return -1;
3627 /* post_oid is theirs */
3628 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &post_oid);
3629 clear_image(&tmp_image);
3631 /* our_oid is ours */
3632 if (patch->is_new) {
3633 if (load_current(state, &tmp_image, patch))
3634 return error(_("cannot read the current contents of '%s'"),
3635 patch->new_name);
3636 } else {
3637 if (load_preimage(state, &tmp_image, patch, st, ce))
3638 return error(_("cannot read the current contents of '%s'"),
3639 patch->old_name);
3641 write_object_file(tmp_image.buf, tmp_image.len, OBJ_BLOB, &our_oid);
3642 clear_image(&tmp_image);
3644 /* in-core three-way merge between post and our using pre as base */
3645 status = three_way_merge(state, image, patch->new_name,
3646 &pre_oid, &our_oid, &post_oid);
3647 if (status < 0) {
3648 if (state->apply_verbosity > verbosity_silent)
3649 fprintf(stderr,
3650 _("Failed to perform three-way merge...\n"));
3651 return status;
3654 if (status) {
3655 patch->conflicted_threeway = 1;
3656 if (patch->is_new)
3657 oidclr(&patch->threeway_stage[0]);
3658 else
3659 oidcpy(&patch->threeway_stage[0], &pre_oid);
3660 oidcpy(&patch->threeway_stage[1], &our_oid);
3661 oidcpy(&patch->threeway_stage[2], &post_oid);
3662 if (state->apply_verbosity > verbosity_silent)
3663 fprintf(stderr,
3664 _("Applied patch to '%s' with conflicts.\n"),
3665 patch->new_name);
3666 } else {
3667 if (state->apply_verbosity > verbosity_silent)
3668 fprintf(stderr,
3669 _("Applied patch to '%s' cleanly.\n"),
3670 patch->new_name);
3672 return 0;
3675 static int apply_data(struct apply_state *state, struct patch *patch,
3676 struct stat *st, const struct cache_entry *ce)
3678 struct image image;
3680 if (load_preimage(state, &image, patch, st, ce) < 0)
3681 return -1;
3683 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0) {
3684 if (state->apply_verbosity > verbosity_silent &&
3685 state->threeway && !patch->direct_to_threeway)
3686 fprintf(stderr, _("Falling back to direct application...\n"));
3688 /* Note: with --reject, apply_fragments() returns 0 */
3689 if (patch->direct_to_threeway || apply_fragments(state, &image, patch) < 0)
3690 return -1;
3692 patch->result = image.buf;
3693 patch->resultsize = image.len;
3694 add_to_fn_table(state, patch);
3695 free(image.line_allocated);
3697 if (0 < patch->is_delete && patch->resultsize)
3698 return error(_("removal patch leaves file contents"));
3700 return 0;
3704 * If "patch" that we are looking at modifies or deletes what we have,
3705 * we would want it not to lose any local modification we have, either
3706 * in the working tree or in the index.
3708 * This also decides if a non-git patch is a creation patch or a
3709 * modification to an existing empty file. We do not check the state
3710 * of the current tree for a creation patch in this function; the caller
3711 * check_patch() separately makes sure (and errors out otherwise) that
3712 * the path the patch creates does not exist in the current tree.
3714 static int check_preimage(struct apply_state *state,
3715 struct patch *patch,
3716 struct cache_entry **ce,
3717 struct stat *st)
3719 const char *old_name = patch->old_name;
3720 struct patch *previous = NULL;
3721 int stat_ret = 0, status;
3722 unsigned st_mode = 0;
3724 if (!old_name)
3725 return 0;
3727 assert(patch->is_new <= 0);
3728 previous = previous_patch(state, patch, &status);
3730 if (status)
3731 return error(_("path %s has been renamed/deleted"), old_name);
3732 if (previous) {
3733 st_mode = previous->new_mode;
3734 } else if (!state->cached) {
3735 stat_ret = lstat(old_name, st);
3736 if (stat_ret && errno != ENOENT)
3737 return error_errno("%s", old_name);
3740 if (state->check_index && !previous) {
3741 int pos = index_name_pos(state->repo->index, old_name,
3742 strlen(old_name));
3743 if (pos < 0) {
3744 if (patch->is_new < 0)
3745 goto is_new;
3746 return error(_("%s: does not exist in index"), old_name);
3748 *ce = state->repo->index->cache[pos];
3749 if (stat_ret < 0) {
3750 if (checkout_target(state->repo->index, *ce, st))
3751 return -1;
3753 if (!state->cached && verify_index_match(state, *ce, st))
3754 return error(_("%s: does not match index"), old_name);
3755 if (state->cached)
3756 st_mode = (*ce)->ce_mode;
3757 } else if (stat_ret < 0) {
3758 if (patch->is_new < 0)
3759 goto is_new;
3760 return error_errno("%s", old_name);
3763 if (!state->cached && !previous)
3764 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3766 if (patch->is_new < 0)
3767 patch->is_new = 0;
3768 if (!patch->old_mode)
3769 patch->old_mode = st_mode;
3770 if ((st_mode ^ patch->old_mode) & S_IFMT)
3771 return error(_("%s: wrong type"), old_name);
3772 if (st_mode != patch->old_mode)
3773 warning(_("%s has type %o, expected %o"),
3774 old_name, st_mode, patch->old_mode);
3775 if (!patch->new_mode && !patch->is_delete)
3776 patch->new_mode = st_mode;
3777 return 0;
3779 is_new:
3780 patch->is_new = 1;
3781 patch->is_delete = 0;
3782 FREE_AND_NULL(patch->old_name);
3783 return 0;
3787 #define EXISTS_IN_INDEX 1
3788 #define EXISTS_IN_WORKTREE 2
3789 #define EXISTS_IN_INDEX_AS_ITA 3
3791 static int check_to_create(struct apply_state *state,
3792 const char *new_name,
3793 int ok_if_exists)
3795 struct stat nst;
3797 if (state->check_index && (!ok_if_exists || !state->cached)) {
3798 int pos;
3800 pos = index_name_pos(state->repo->index, new_name, strlen(new_name));
3801 if (pos >= 0) {
3802 struct cache_entry *ce = state->repo->index->cache[pos];
3804 /* allow ITA, as they do not yet exist in the index */
3805 if (!ok_if_exists && !(ce->ce_flags & CE_INTENT_TO_ADD))
3806 return EXISTS_IN_INDEX;
3808 /* ITA entries can never match working tree files */
3809 if (!state->cached && (ce->ce_flags & CE_INTENT_TO_ADD))
3810 return EXISTS_IN_INDEX_AS_ITA;
3814 if (state->cached)
3815 return 0;
3817 if (!lstat(new_name, &nst)) {
3818 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3819 return 0;
3821 * A leading component of new_name might be a symlink
3822 * that is going to be removed with this patch, but
3823 * still pointing at somewhere that has the path.
3824 * In such a case, path "new_name" does not exist as
3825 * far as git is concerned.
3827 if (has_symlink_leading_path(new_name, strlen(new_name)))
3828 return 0;
3830 return EXISTS_IN_WORKTREE;
3831 } else if (!is_missing_file_error(errno)) {
3832 return error_errno("%s", new_name);
3834 return 0;
3837 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3839 for ( ; patch; patch = patch->next) {
3840 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3841 (patch->is_rename || patch->is_delete))
3842 /* the symlink at patch->old_name is removed */
3843 strset_add(&state->removed_symlinks, patch->old_name);
3845 if (patch->new_name && S_ISLNK(patch->new_mode))
3846 /* the symlink at patch->new_name is created or remains */
3847 strset_add(&state->kept_symlinks, patch->new_name);
3851 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3853 do {
3854 while (--name->len && name->buf[name->len] != '/')
3855 ; /* scan backwards */
3856 if (!name->len)
3857 break;
3858 name->buf[name->len] = '\0';
3859 if (strset_contains(&state->kept_symlinks, name->buf))
3860 return 1;
3861 if (strset_contains(&state->removed_symlinks, name->buf))
3863 * This cannot be "return 0", because we may
3864 * see a new one created at a higher level.
3866 continue;
3868 /* otherwise, check the preimage */
3869 if (state->check_index) {
3870 struct cache_entry *ce;
3872 ce = index_file_exists(state->repo->index, name->buf,
3873 name->len, ignore_case);
3874 if (ce && S_ISLNK(ce->ce_mode))
3875 return 1;
3876 } else {
3877 struct stat st;
3878 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3879 return 1;
3881 } while (1);
3882 return 0;
3885 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3887 int ret;
3888 struct strbuf name = STRBUF_INIT;
3890 assert(*name_ != '\0');
3891 strbuf_addstr(&name, name_);
3892 ret = path_is_beyond_symlink_1(state, &name);
3893 strbuf_release(&name);
3895 return ret;
3898 static int check_unsafe_path(struct patch *patch)
3900 const char *old_name = NULL;
3901 const char *new_name = NULL;
3902 if (patch->is_delete)
3903 old_name = patch->old_name;
3904 else if (!patch->is_new && !patch->is_copy)
3905 old_name = patch->old_name;
3906 if (!patch->is_delete)
3907 new_name = patch->new_name;
3909 if (old_name && !verify_path(old_name, patch->old_mode))
3910 return error(_("invalid path '%s'"), old_name);
3911 if (new_name && !verify_path(new_name, patch->new_mode))
3912 return error(_("invalid path '%s'"), new_name);
3913 return 0;
3917 * Check and apply the patch in-core; leave the result in patch->result
3918 * for the caller to write it out to the final destination.
3920 static int check_patch(struct apply_state *state, struct patch *patch)
3922 struct stat st;
3923 const char *old_name = patch->old_name;
3924 const char *new_name = patch->new_name;
3925 const char *name = old_name ? old_name : new_name;
3926 struct cache_entry *ce = NULL;
3927 struct patch *tpatch;
3928 int ok_if_exists;
3929 int status;
3931 patch->rejected = 1; /* we will drop this after we succeed */
3933 status = check_preimage(state, patch, &ce, &st);
3934 if (status)
3935 return status;
3936 old_name = patch->old_name;
3939 * A type-change diff is always split into a patch to delete
3940 * old, immediately followed by a patch to create new (see
3941 * diff.c::run_diff()); in such a case it is Ok that the entry
3942 * to be deleted by the previous patch is still in the working
3943 * tree and in the index.
3945 * A patch to swap-rename between A and B would first rename A
3946 * to B and then rename B to A. While applying the first one,
3947 * the presence of B should not stop A from getting renamed to
3948 * B; ask to_be_deleted() about the later rename. Removal of
3949 * B and rename from A to B is handled the same way by asking
3950 * was_deleted().
3952 if ((tpatch = in_fn_table(state, new_name)) &&
3953 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3954 ok_if_exists = 1;
3955 else
3956 ok_if_exists = 0;
3958 if (new_name &&
3959 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3960 int err = check_to_create(state, new_name, ok_if_exists);
3962 if (err && state->threeway) {
3963 patch->direct_to_threeway = 1;
3964 } else switch (err) {
3965 case 0:
3966 break; /* happy */
3967 case EXISTS_IN_INDEX:
3968 return error(_("%s: already exists in index"), new_name);
3969 case EXISTS_IN_INDEX_AS_ITA:
3970 return error(_("%s: does not match index"), new_name);
3971 case EXISTS_IN_WORKTREE:
3972 return error(_("%s: already exists in working directory"),
3973 new_name);
3974 default:
3975 return err;
3978 if (!patch->new_mode) {
3979 if (0 < patch->is_new)
3980 patch->new_mode = S_IFREG | 0644;
3981 else
3982 patch->new_mode = patch->old_mode;
3986 if (new_name && old_name) {
3987 int same = !strcmp(old_name, new_name);
3988 if (!patch->new_mode)
3989 patch->new_mode = patch->old_mode;
3990 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3991 if (same)
3992 return error(_("new mode (%o) of %s does not "
3993 "match old mode (%o)"),
3994 patch->new_mode, new_name,
3995 patch->old_mode);
3996 else
3997 return error(_("new mode (%o) of %s does not "
3998 "match old mode (%o) of %s"),
3999 patch->new_mode, new_name,
4000 patch->old_mode, old_name);
4004 if (!state->unsafe_paths && check_unsafe_path(patch))
4005 return -128;
4008 * An attempt to read from or delete a path that is beyond a
4009 * symbolic link will be prevented by load_patch_target() that
4010 * is called at the beginning of apply_data() so we do not
4011 * have to worry about a patch marked with "is_delete" bit
4012 * here. We however need to make sure that the patch result
4013 * is not deposited to a path that is beyond a symbolic link
4014 * here.
4016 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
4017 return error(_("affected file '%s' is beyond a symbolic link"),
4018 patch->new_name);
4020 if (apply_data(state, patch, &st, ce) < 0)
4021 return error(_("%s: patch does not apply"), name);
4022 patch->rejected = 0;
4023 return 0;
4026 static int check_patch_list(struct apply_state *state, struct patch *patch)
4028 int err = 0;
4030 prepare_symlink_changes(state, patch);
4031 prepare_fn_table(state, patch);
4032 while (patch) {
4033 int res;
4034 if (state->apply_verbosity > verbosity_normal)
4035 say_patch_name(stderr,
4036 _("Checking patch %s..."), patch);
4037 res = check_patch(state, patch);
4038 if (res == -128)
4039 return -128;
4040 err |= res;
4041 patch = patch->next;
4043 return err;
4046 static int read_apply_cache(struct apply_state *state)
4048 if (state->index_file)
4049 return read_index_from(state->repo->index, state->index_file,
4050 get_git_dir());
4051 else
4052 return repo_read_index(state->repo);
4055 /* This function tries to read the object name from the current index */
4056 static int get_current_oid(struct apply_state *state, const char *path,
4057 struct object_id *oid)
4059 int pos;
4061 if (read_apply_cache(state) < 0)
4062 return -1;
4063 pos = index_name_pos(state->repo->index, path, strlen(path));
4064 if (pos < 0)
4065 return -1;
4066 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4067 return 0;
4070 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4073 * A usable gitlink patch has only one fragment (hunk) that looks like:
4074 * @@ -1 +1 @@
4075 * -Subproject commit <old sha1>
4076 * +Subproject commit <new sha1>
4077 * or
4078 * @@ -1 +0,0 @@
4079 * -Subproject commit <old sha1>
4080 * for a removal patch.
4082 struct fragment *hunk = p->fragments;
4083 static const char heading[] = "-Subproject commit ";
4084 char *preimage;
4086 if (/* does the patch have only one hunk? */
4087 hunk && !hunk->next &&
4088 /* is its preimage one line? */
4089 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4090 /* does preimage begin with the heading? */
4091 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4092 starts_with(++preimage, heading) &&
4093 /* does it record full SHA-1? */
4094 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4095 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4096 /* does the abbreviated name on the index line agree with it? */
4097 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4098 return 0; /* it all looks fine */
4100 /* we may have full object name on the index line */
4101 return get_oid_hex(p->old_oid_prefix, oid);
4104 /* Build an index that contains just the files needed for a 3way merge */
4105 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4107 struct patch *patch;
4108 struct index_state result = { NULL };
4109 struct lock_file lock = LOCK_INIT;
4110 int res;
4112 /* Once we start supporting the reverse patch, it may be
4113 * worth showing the new sha1 prefix, but until then...
4115 for (patch = list; patch; patch = patch->next) {
4116 struct object_id oid;
4117 struct cache_entry *ce;
4118 const char *name;
4120 name = patch->old_name ? patch->old_name : patch->new_name;
4121 if (0 < patch->is_new)
4122 continue;
4124 if (S_ISGITLINK(patch->old_mode)) {
4125 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4126 ; /* ok, the textual part looks sane */
4127 else
4128 return error(_("sha1 information is lacking or "
4129 "useless for submodule %s"), name);
4130 } else if (!get_oid_blob(patch->old_oid_prefix, &oid)) {
4131 ; /* ok */
4132 } else if (!patch->lines_added && !patch->lines_deleted) {
4133 /* mode-only change: update the current */
4134 if (get_current_oid(state, patch->old_name, &oid))
4135 return error(_("mode change for %s, which is not "
4136 "in current HEAD"), name);
4137 } else
4138 return error(_("sha1 information is lacking or useless "
4139 "(%s)."), name);
4141 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4142 if (!ce)
4143 return error(_("make_cache_entry failed for path '%s'"),
4144 name);
4145 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4146 discard_cache_entry(ce);
4147 return error(_("could not add %s to temporary index"),
4148 name);
4152 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4153 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4154 discard_index(&result);
4156 if (res)
4157 return error(_("could not write temporary index to %s"),
4158 state->fake_ancestor);
4160 return 0;
4163 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4165 int files, adds, dels;
4167 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4168 files++;
4169 adds += patch->lines_added;
4170 dels += patch->lines_deleted;
4171 show_stats(state, patch);
4174 print_stat_summary(stdout, files, adds, dels);
4177 static void numstat_patch_list(struct apply_state *state,
4178 struct patch *patch)
4180 for ( ; patch; patch = patch->next) {
4181 const char *name;
4182 name = patch->new_name ? patch->new_name : patch->old_name;
4183 if (patch->is_binary)
4184 printf("-\t-\t");
4185 else
4186 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4187 write_name_quoted(name, stdout, state->line_termination);
4191 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4193 if (mode)
4194 printf(" %s mode %06o %s\n", newdelete, mode, name);
4195 else
4196 printf(" %s %s\n", newdelete, name);
4199 static void show_mode_change(struct patch *p, int show_name)
4201 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4202 if (show_name)
4203 printf(" mode change %06o => %06o %s\n",
4204 p->old_mode, p->new_mode, p->new_name);
4205 else
4206 printf(" mode change %06o => %06o\n",
4207 p->old_mode, p->new_mode);
4211 static void show_rename_copy(struct patch *p)
4213 const char *renamecopy = p->is_rename ? "rename" : "copy";
4214 const char *old_name, *new_name;
4216 /* Find common prefix */
4217 old_name = p->old_name;
4218 new_name = p->new_name;
4219 while (1) {
4220 const char *slash_old, *slash_new;
4221 slash_old = strchr(old_name, '/');
4222 slash_new = strchr(new_name, '/');
4223 if (!slash_old ||
4224 !slash_new ||
4225 slash_old - old_name != slash_new - new_name ||
4226 memcmp(old_name, new_name, slash_new - new_name))
4227 break;
4228 old_name = slash_old + 1;
4229 new_name = slash_new + 1;
4231 /* p->old_name through old_name is the common prefix, and old_name and
4232 * new_name through the end of names are renames
4234 if (old_name != p->old_name)
4235 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4236 (int)(old_name - p->old_name), p->old_name,
4237 old_name, new_name, p->score);
4238 else
4239 printf(" %s %s => %s (%d%%)\n", renamecopy,
4240 p->old_name, p->new_name, p->score);
4241 show_mode_change(p, 0);
4244 static void summary_patch_list(struct patch *patch)
4246 struct patch *p;
4248 for (p = patch; p; p = p->next) {
4249 if (p->is_new)
4250 show_file_mode_name("create", p->new_mode, p->new_name);
4251 else if (p->is_delete)
4252 show_file_mode_name("delete", p->old_mode, p->old_name);
4253 else {
4254 if (p->is_rename || p->is_copy)
4255 show_rename_copy(p);
4256 else {
4257 if (p->score) {
4258 printf(" rewrite %s (%d%%)\n",
4259 p->new_name, p->score);
4260 show_mode_change(p, 0);
4262 else
4263 show_mode_change(p, 1);
4269 static void patch_stats(struct apply_state *state, struct patch *patch)
4271 int lines = patch->lines_added + patch->lines_deleted;
4273 if (lines > state->max_change)
4274 state->max_change = lines;
4275 if (patch->old_name) {
4276 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4277 if (!len)
4278 len = strlen(patch->old_name);
4279 if (len > state->max_len)
4280 state->max_len = len;
4282 if (patch->new_name) {
4283 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4284 if (!len)
4285 len = strlen(patch->new_name);
4286 if (len > state->max_len)
4287 state->max_len = len;
4291 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4293 if (state->update_index && !state->ita_only) {
4294 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4295 return error(_("unable to remove %s from index"), patch->old_name);
4297 if (!state->cached) {
4298 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4299 remove_path(patch->old_name);
4302 return 0;
4305 static int add_index_file(struct apply_state *state,
4306 const char *path,
4307 unsigned mode,
4308 void *buf,
4309 unsigned long size)
4311 struct stat st;
4312 struct cache_entry *ce;
4313 int namelen = strlen(path);
4315 ce = make_empty_cache_entry(state->repo->index, namelen);
4316 memcpy(ce->name, path, namelen);
4317 ce->ce_mode = create_ce_mode(mode);
4318 ce->ce_flags = create_ce_flags(0);
4319 ce->ce_namelen = namelen;
4320 if (state->ita_only) {
4321 ce->ce_flags |= CE_INTENT_TO_ADD;
4322 set_object_name_for_intent_to_add_entry(ce);
4323 } else if (S_ISGITLINK(mode)) {
4324 const char *s;
4326 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4327 get_oid_hex(s, &ce->oid)) {
4328 discard_cache_entry(ce);
4329 return error(_("corrupt patch for submodule %s"), path);
4331 } else {
4332 if (!state->cached) {
4333 if (lstat(path, &st) < 0) {
4334 discard_cache_entry(ce);
4335 return error_errno(_("unable to stat newly "
4336 "created file '%s'"),
4337 path);
4339 fill_stat_cache_info(state->repo->index, ce, &st);
4341 if (write_object_file(buf, size, OBJ_BLOB, &ce->oid) < 0) {
4342 discard_cache_entry(ce);
4343 return error(_("unable to create backing store "
4344 "for newly created file %s"), path);
4347 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4348 discard_cache_entry(ce);
4349 return error(_("unable to add cache entry for %s"), path);
4352 return 0;
4356 * Returns:
4357 * -1 if an unrecoverable error happened
4358 * 0 if everything went well
4359 * 1 if a recoverable error happened
4361 static int try_create_file(struct apply_state *state, const char *path,
4362 unsigned int mode, const char *buf,
4363 unsigned long size)
4365 int fd, res;
4366 struct strbuf nbuf = STRBUF_INIT;
4368 if (S_ISGITLINK(mode)) {
4369 struct stat st;
4370 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4371 return 0;
4372 return !!mkdir(path, 0777);
4375 if (has_symlinks && S_ISLNK(mode))
4376 /* Although buf:size is counted string, it also is NUL
4377 * terminated.
4379 return !!symlink(buf, path);
4381 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4382 if (fd < 0)
4383 return 1;
4385 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf, NULL)) {
4386 size = nbuf.len;
4387 buf = nbuf.buf;
4390 res = write_in_full(fd, buf, size) < 0;
4391 if (res)
4392 error_errno(_("failed to write to '%s'"), path);
4393 strbuf_release(&nbuf);
4395 if (close(fd) < 0 && !res)
4396 return error_errno(_("closing file '%s'"), path);
4398 return res ? -1 : 0;
4402 * We optimistically assume that the directories exist,
4403 * which is true 99% of the time anyway. If they don't,
4404 * we create them and try again.
4406 * Returns:
4407 * -1 on error
4408 * 0 otherwise
4410 static int create_one_file(struct apply_state *state,
4411 char *path,
4412 unsigned mode,
4413 const char *buf,
4414 unsigned long size)
4416 int res;
4418 if (state->cached)
4419 return 0;
4422 * We already try to detect whether files are beyond a symlink in our
4423 * up-front checks. But in the case where symlinks are created by any
4424 * of the intermediate hunks it can happen that our up-front checks
4425 * didn't yet see the symlink, but at the point of arriving here there
4426 * in fact is one. We thus repeat the check for symlinks here.
4428 * Note that this does not make the up-front check obsolete as the
4429 * failure mode is different:
4431 * - The up-front checks cause us to abort before we have written
4432 * anything into the working directory. So when we exit this way the
4433 * working directory remains clean.
4435 * - The checks here happen in the middle of the action where we have
4436 * already started to apply the patch. The end result will be a dirty
4437 * working directory.
4439 * Ideally, we should update the up-front checks to catch what would
4440 * happen when we apply the patch before we damage the working tree.
4441 * We have all the information necessary to do so. But for now, as a
4442 * part of embargoed security work, having this check would serve as a
4443 * reasonable first step.
4445 if (path_is_beyond_symlink(state, path))
4446 return error(_("affected file '%s' is beyond a symbolic link"), path);
4448 res = try_create_file(state, path, mode, buf, size);
4449 if (res < 0)
4450 return -1;
4451 if (!res)
4452 return 0;
4454 if (errno == ENOENT) {
4455 if (safe_create_leading_directories_no_share(path))
4456 return 0;
4457 res = try_create_file(state, path, mode, buf, size);
4458 if (res < 0)
4459 return -1;
4460 if (!res)
4461 return 0;
4464 if (errno == EEXIST || errno == EACCES) {
4465 /* We may be trying to create a file where a directory
4466 * used to be.
4468 struct stat st;
4469 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4470 errno = EEXIST;
4473 if (errno == EEXIST) {
4474 unsigned int nr = getpid();
4476 for (;;) {
4477 char newpath[PATH_MAX];
4478 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4479 res = try_create_file(state, newpath, mode, buf, size);
4480 if (res < 0)
4481 return -1;
4482 if (!res) {
4483 if (!rename(newpath, path))
4484 return 0;
4485 unlink_or_warn(newpath);
4486 break;
4488 if (errno != EEXIST)
4489 break;
4490 ++nr;
4493 return error_errno(_("unable to write file '%s' mode %o"),
4494 path, mode);
4497 static int add_conflicted_stages_file(struct apply_state *state,
4498 struct patch *patch)
4500 int stage, namelen;
4501 unsigned mode;
4502 struct cache_entry *ce;
4504 if (!state->update_index)
4505 return 0;
4506 namelen = strlen(patch->new_name);
4507 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4509 remove_file_from_index(state->repo->index, patch->new_name);
4510 for (stage = 1; stage < 4; stage++) {
4511 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4512 continue;
4513 ce = make_empty_cache_entry(state->repo->index, namelen);
4514 memcpy(ce->name, patch->new_name, namelen);
4515 ce->ce_mode = create_ce_mode(mode);
4516 ce->ce_flags = create_ce_flags(stage);
4517 ce->ce_namelen = namelen;
4518 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4519 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4520 discard_cache_entry(ce);
4521 return error(_("unable to add cache entry for %s"),
4522 patch->new_name);
4526 return 0;
4529 static int create_file(struct apply_state *state, struct patch *patch)
4531 char *path = patch->new_name;
4532 unsigned mode = patch->new_mode;
4533 unsigned long size = patch->resultsize;
4534 char *buf = patch->result;
4536 if (!mode)
4537 mode = S_IFREG | 0644;
4538 if (create_one_file(state, path, mode, buf, size))
4539 return -1;
4541 if (patch->conflicted_threeway)
4542 return add_conflicted_stages_file(state, patch);
4543 else if (state->update_index)
4544 return add_index_file(state, path, mode, buf, size);
4545 return 0;
4548 /* phase zero is to remove, phase one is to create */
4549 static int write_out_one_result(struct apply_state *state,
4550 struct patch *patch,
4551 int phase)
4553 if (patch->is_delete > 0) {
4554 if (phase == 0)
4555 return remove_file(state, patch, 1);
4556 return 0;
4558 if (patch->is_new > 0 || patch->is_copy) {
4559 if (phase == 1)
4560 return create_file(state, patch);
4561 return 0;
4564 * Rename or modification boils down to the same
4565 * thing: remove the old, write the new
4567 if (phase == 0)
4568 return remove_file(state, patch, patch->is_rename);
4569 if (phase == 1)
4570 return create_file(state, patch);
4571 return 0;
4574 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4576 FILE *rej;
4577 char namebuf[PATH_MAX];
4578 struct fragment *frag;
4579 int cnt = 0;
4580 struct strbuf sb = STRBUF_INIT;
4582 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4583 if (!frag->rejected)
4584 continue;
4585 cnt++;
4588 if (!cnt) {
4589 if (state->apply_verbosity > verbosity_normal)
4590 say_patch_name(stderr,
4591 _("Applied patch %s cleanly."), patch);
4592 return 0;
4595 /* This should not happen, because a removal patch that leaves
4596 * contents are marked "rejected" at the patch level.
4598 if (!patch->new_name)
4599 die(_("internal error"));
4601 /* Say this even without --verbose */
4602 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4603 "Applying patch %%s with %d rejects...",
4604 cnt),
4605 cnt);
4606 if (state->apply_verbosity > verbosity_silent)
4607 say_patch_name(stderr, sb.buf, patch);
4608 strbuf_release(&sb);
4610 cnt = strlen(patch->new_name);
4611 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4612 cnt = ARRAY_SIZE(namebuf) - 5;
4613 warning(_("truncating .rej filename to %.*s.rej"),
4614 cnt - 1, patch->new_name);
4616 memcpy(namebuf, patch->new_name, cnt);
4617 memcpy(namebuf + cnt, ".rej", 5);
4619 rej = fopen(namebuf, "w");
4620 if (!rej)
4621 return error_errno(_("cannot open %s"), namebuf);
4623 /* Normal git tools never deal with .rej, so do not pretend
4624 * this is a git patch by saying --git or giving extended
4625 * headers. While at it, maybe please "kompare" that wants
4626 * the trailing TAB and some garbage at the end of line ;-).
4628 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4629 patch->new_name, patch->new_name);
4630 for (cnt = 1, frag = patch->fragments;
4631 frag;
4632 cnt++, frag = frag->next) {
4633 if (!frag->rejected) {
4634 if (state->apply_verbosity > verbosity_silent)
4635 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4636 continue;
4638 if (state->apply_verbosity > verbosity_silent)
4639 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4640 fprintf(rej, "%.*s", frag->size, frag->patch);
4641 if (frag->patch[frag->size-1] != '\n')
4642 fputc('\n', rej);
4644 fclose(rej);
4645 return -1;
4649 * Returns:
4650 * -1 if an error happened
4651 * 0 if the patch applied cleanly
4652 * 1 if the patch did not apply cleanly
4654 static int write_out_results(struct apply_state *state, struct patch *list)
4656 int phase;
4657 int errs = 0;
4658 struct patch *l;
4659 struct string_list cpath = STRING_LIST_INIT_DUP;
4661 for (phase = 0; phase < 2; phase++) {
4662 l = list;
4663 while (l) {
4664 if (l->rejected)
4665 errs = 1;
4666 else {
4667 if (write_out_one_result(state, l, phase)) {
4668 string_list_clear(&cpath, 0);
4669 return -1;
4671 if (phase == 1) {
4672 if (write_out_one_reject(state, l))
4673 errs = 1;
4674 if (l->conflicted_threeway) {
4675 string_list_append(&cpath, l->new_name);
4676 errs = 1;
4680 l = l->next;
4684 if (cpath.nr) {
4685 struct string_list_item *item;
4687 string_list_sort(&cpath);
4688 if (state->apply_verbosity > verbosity_silent) {
4689 for_each_string_list_item(item, &cpath)
4690 fprintf(stderr, "U %s\n", item->string);
4692 string_list_clear(&cpath, 0);
4695 * rerere relies on the partially merged result being in the working
4696 * tree with conflict markers, but that isn't written with --cached.
4698 if (!state->cached)
4699 repo_rerere(state->repo, 0);
4702 return errs;
4706 * Try to apply a patch.
4708 * Returns:
4709 * -128 if a bad error happened (like patch unreadable)
4710 * -1 if patch did not apply and user cannot deal with it
4711 * 0 if the patch applied
4712 * 1 if the patch did not apply but user might fix it
4714 static int apply_patch(struct apply_state *state,
4715 int fd,
4716 const char *filename,
4717 int options)
4719 size_t offset;
4720 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4721 struct patch *list = NULL, **listp = &list;
4722 int skipped_patch = 0;
4723 int res = 0;
4724 int flush_attributes = 0;
4726 state->patch_input_file = filename;
4727 if (read_patch_file(&buf, fd) < 0)
4728 return -128;
4729 offset = 0;
4730 while (offset < buf.len) {
4731 struct patch *patch;
4732 int nr;
4734 CALLOC_ARRAY(patch, 1);
4735 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4736 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4737 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4738 if (nr < 0) {
4739 free_patch(patch);
4740 if (nr == -128) {
4741 res = -128;
4742 goto end;
4744 break;
4746 if (state->apply_in_reverse)
4747 reverse_patches(patch);
4748 if (use_patch(state, patch)) {
4749 patch_stats(state, patch);
4750 if (!list || !state->apply_in_reverse) {
4751 *listp = patch;
4752 listp = &patch->next;
4753 } else {
4754 patch->next = list;
4755 list = patch;
4758 if ((patch->new_name &&
4759 ends_with_path_components(patch->new_name,
4760 GITATTRIBUTES_FILE)) ||
4761 (patch->old_name &&
4762 ends_with_path_components(patch->old_name,
4763 GITATTRIBUTES_FILE)))
4764 flush_attributes = 1;
4766 else {
4767 if (state->apply_verbosity > verbosity_normal)
4768 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4769 free_patch(patch);
4770 skipped_patch++;
4772 offset += nr;
4775 if (!list && !skipped_patch) {
4776 if (!state->allow_empty) {
4777 error(_("No valid patches in input (allow with \"--allow-empty\")"));
4778 res = -128;
4780 goto end;
4783 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4784 state->apply = 0;
4786 state->update_index = (state->check_index || state->ita_only) && state->apply;
4787 if (state->update_index && !is_lock_file_locked(&state->lock_file)) {
4788 if (state->index_file)
4789 hold_lock_file_for_update(&state->lock_file,
4790 state->index_file,
4791 LOCK_DIE_ON_ERROR);
4792 else
4793 repo_hold_locked_index(state->repo, &state->lock_file,
4794 LOCK_DIE_ON_ERROR);
4797 if (state->check_index && read_apply_cache(state) < 0) {
4798 error(_("unable to read index file"));
4799 res = -128;
4800 goto end;
4803 if (state->check || state->apply) {
4804 int r = check_patch_list(state, list);
4805 if (r == -128) {
4806 res = -128;
4807 goto end;
4809 if (r < 0 && !state->apply_with_reject) {
4810 res = -1;
4811 goto end;
4815 if (state->apply) {
4816 int write_res = write_out_results(state, list);
4817 if (write_res < 0) {
4818 res = -128;
4819 goto end;
4821 if (write_res > 0) {
4822 /* with --3way, we still need to write the index out */
4823 res = state->apply_with_reject ? -1 : 1;
4824 goto end;
4828 if (state->fake_ancestor &&
4829 build_fake_ancestor(state, list)) {
4830 res = -128;
4831 goto end;
4834 if (state->diffstat && state->apply_verbosity > verbosity_silent)
4835 stat_patch_list(state, list);
4837 if (state->numstat && state->apply_verbosity > verbosity_silent)
4838 numstat_patch_list(state, list);
4840 if (state->summary && state->apply_verbosity > verbosity_silent)
4841 summary_patch_list(list);
4843 if (flush_attributes)
4844 reset_parsed_attributes();
4845 end:
4846 free_patch_list(list);
4847 strbuf_release(&buf);
4848 string_list_clear(&state->fn_table, 0);
4849 return res;
4852 static int apply_option_parse_exclude(const struct option *opt,
4853 const char *arg, int unset)
4855 struct apply_state *state = opt->value;
4857 BUG_ON_OPT_NEG(unset);
4859 add_name_limit(state, arg, 1);
4860 return 0;
4863 static int apply_option_parse_include(const struct option *opt,
4864 const char *arg, int unset)
4866 struct apply_state *state = opt->value;
4868 BUG_ON_OPT_NEG(unset);
4870 add_name_limit(state, arg, 0);
4871 state->has_include = 1;
4872 return 0;
4875 static int apply_option_parse_p(const struct option *opt,
4876 const char *arg,
4877 int unset)
4879 struct apply_state *state = opt->value;
4881 BUG_ON_OPT_NEG(unset);
4883 state->p_value = atoi(arg);
4884 state->p_value_known = 1;
4885 return 0;
4888 static int apply_option_parse_space_change(const struct option *opt,
4889 const char *arg, int unset)
4891 struct apply_state *state = opt->value;
4893 BUG_ON_OPT_ARG(arg);
4895 if (unset)
4896 state->ws_ignore_action = ignore_ws_none;
4897 else
4898 state->ws_ignore_action = ignore_ws_change;
4899 return 0;
4902 static int apply_option_parse_whitespace(const struct option *opt,
4903 const char *arg, int unset)
4905 struct apply_state *state = opt->value;
4907 BUG_ON_OPT_NEG(unset);
4909 state->whitespace_option = arg;
4910 if (parse_whitespace_option(state, arg))
4911 return -1;
4912 return 0;
4915 static int apply_option_parse_directory(const struct option *opt,
4916 const char *arg, int unset)
4918 struct apply_state *state = opt->value;
4920 BUG_ON_OPT_NEG(unset);
4922 strbuf_reset(&state->root);
4923 strbuf_addstr(&state->root, arg);
4924 strbuf_complete(&state->root, '/');
4925 return 0;
4928 int apply_all_patches(struct apply_state *state,
4929 int argc,
4930 const char **argv,
4931 int options)
4933 int i;
4934 int res;
4935 int errs = 0;
4936 int read_stdin = 1;
4938 for (i = 0; i < argc; i++) {
4939 const char *arg = argv[i];
4940 char *to_free = NULL;
4941 int fd;
4943 if (!strcmp(arg, "-")) {
4944 res = apply_patch(state, 0, "<stdin>", options);
4945 if (res < 0)
4946 goto end;
4947 errs |= res;
4948 read_stdin = 0;
4949 continue;
4950 } else
4951 arg = to_free = prefix_filename(state->prefix, arg);
4953 fd = open(arg, O_RDONLY);
4954 if (fd < 0) {
4955 error(_("can't open patch '%s': %s"), arg, strerror(errno));
4956 res = -128;
4957 free(to_free);
4958 goto end;
4960 read_stdin = 0;
4961 set_default_whitespace_mode(state);
4962 res = apply_patch(state, fd, arg, options);
4963 close(fd);
4964 free(to_free);
4965 if (res < 0)
4966 goto end;
4967 errs |= res;
4969 set_default_whitespace_mode(state);
4970 if (read_stdin) {
4971 res = apply_patch(state, 0, "<stdin>", options);
4972 if (res < 0)
4973 goto end;
4974 errs |= res;
4977 if (state->whitespace_error) {
4978 if (state->squelch_whitespace_errors &&
4979 state->squelch_whitespace_errors < state->whitespace_error) {
4980 int squelched =
4981 state->whitespace_error - state->squelch_whitespace_errors;
4982 warning(Q_("squelched %d whitespace error",
4983 "squelched %d whitespace errors",
4984 squelched),
4985 squelched);
4987 if (state->ws_error_action == die_on_ws_error) {
4988 error(Q_("%d line adds whitespace errors.",
4989 "%d lines add whitespace errors.",
4990 state->whitespace_error),
4991 state->whitespace_error);
4992 res = -128;
4993 goto end;
4995 if (state->applied_after_fixing_ws && state->apply)
4996 warning(Q_("%d line applied after"
4997 " fixing whitespace errors.",
4998 "%d lines applied after"
4999 " fixing whitespace errors.",
5000 state->applied_after_fixing_ws),
5001 state->applied_after_fixing_ws);
5002 else if (state->whitespace_error)
5003 warning(Q_("%d line adds whitespace errors.",
5004 "%d lines add whitespace errors.",
5005 state->whitespace_error),
5006 state->whitespace_error);
5009 if (state->update_index) {
5010 res = write_locked_index(state->repo->index, &state->lock_file, COMMIT_LOCK);
5011 if (res) {
5012 error(_("Unable to write new index file"));
5013 res = -128;
5014 goto end;
5018 res = !!errs;
5020 end:
5021 rollback_lock_file(&state->lock_file);
5023 if (state->apply_verbosity <= verbosity_silent) {
5024 set_error_routine(state->saved_error_routine);
5025 set_warn_routine(state->saved_warn_routine);
5028 if (res > -1)
5029 return res;
5030 return (res == -1 ? 1 : 128);
5033 int apply_parse_options(int argc, const char **argv,
5034 struct apply_state *state,
5035 int *force_apply, int *options,
5036 const char * const *apply_usage)
5038 struct option builtin_apply_options[] = {
5039 OPT_CALLBACK_F(0, "exclude", state, N_("path"),
5040 N_("don't apply changes matching the given path"),
5041 PARSE_OPT_NONEG, apply_option_parse_exclude),
5042 OPT_CALLBACK_F(0, "include", state, N_("path"),
5043 N_("apply changes matching the given path"),
5044 PARSE_OPT_NONEG, apply_option_parse_include),
5045 OPT_CALLBACK('p', NULL, state, N_("num"),
5046 N_("remove <num> leading slashes from traditional diff paths"),
5047 apply_option_parse_p),
5048 OPT_BOOL(0, "no-add", &state->no_add,
5049 N_("ignore additions made by the patch")),
5050 OPT_BOOL(0, "stat", &state->diffstat,
5051 N_("instead of applying the patch, output diffstat for the input")),
5052 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
5053 OPT_NOOP_NOARG(0, "binary"),
5054 OPT_BOOL(0, "numstat", &state->numstat,
5055 N_("show number of added and deleted lines in decimal notation")),
5056 OPT_BOOL(0, "summary", &state->summary,
5057 N_("instead of applying the patch, output a summary for the input")),
5058 OPT_BOOL(0, "check", &state->check,
5059 N_("instead of applying the patch, see if the patch is applicable")),
5060 OPT_BOOL(0, "index", &state->check_index,
5061 N_("make sure the patch is applicable to the current index")),
5062 OPT_BOOL('N', "intent-to-add", &state->ita_only,
5063 N_("mark new files with `git add --intent-to-add`")),
5064 OPT_BOOL(0, "cached", &state->cached,
5065 N_("apply a patch without touching the working tree")),
5066 OPT_BOOL_F(0, "unsafe-paths", &state->unsafe_paths,
5067 N_("accept a patch that touches outside the working area"),
5068 PARSE_OPT_NOCOMPLETE),
5069 OPT_BOOL(0, "apply", force_apply,
5070 N_("also apply the patch (use with --stat/--summary/--check)")),
5071 OPT_BOOL('3', "3way", &state->threeway,
5072 N_( "attempt three-way merge, fall back on normal patch if that fails")),
5073 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
5074 N_("build a temporary index based on embedded index information")),
5075 /* Think twice before adding "--nul" synonym to this */
5076 OPT_SET_INT('z', NULL, &state->line_termination,
5077 N_("paths are separated with NUL character"), '\0'),
5078 OPT_INTEGER('C', NULL, &state->p_context,
5079 N_("ensure at least <n> lines of context match")),
5080 OPT_CALLBACK(0, "whitespace", state, N_("action"),
5081 N_("detect new or modified lines that have whitespace errors"),
5082 apply_option_parse_whitespace),
5083 OPT_CALLBACK_F(0, "ignore-space-change", state, NULL,
5084 N_("ignore changes in whitespace when finding context"),
5085 PARSE_OPT_NOARG, apply_option_parse_space_change),
5086 OPT_CALLBACK_F(0, "ignore-whitespace", state, NULL,
5087 N_("ignore changes in whitespace when finding context"),
5088 PARSE_OPT_NOARG, apply_option_parse_space_change),
5089 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
5090 N_("apply the patch in reverse")),
5091 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
5092 N_("don't expect at least one line of context")),
5093 OPT_BOOL(0, "reject", &state->apply_with_reject,
5094 N_("leave the rejected hunks in corresponding *.rej files")),
5095 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
5096 N_("allow overlapping hunks")),
5097 OPT__VERBOSITY(&state->apply_verbosity),
5098 OPT_BIT(0, "inaccurate-eof", options,
5099 N_("tolerate incorrectly detected missing new-line at the end of file"),
5100 APPLY_OPT_INACCURATE_EOF),
5101 OPT_BIT(0, "recount", options,
5102 N_("do not trust the line counts in the hunk headers"),
5103 APPLY_OPT_RECOUNT),
5104 OPT_CALLBACK(0, "directory", state, N_("root"),
5105 N_("prepend <root> to all filenames"),
5106 apply_option_parse_directory),
5107 OPT_BOOL(0, "allow-empty", &state->allow_empty,
5108 N_("don't return error for empty patches")),
5109 OPT_END()
5112 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);