Merge branch 'sg/test-bash-version-fix'
[git.git] / apply.c
blob01793d612620246b14fa57f6ce3ed6c33df65d0f
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"
25 static void git_apply_config(void)
27 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
28 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
29 git_config(git_default_config, NULL);
32 static int parse_whitespace_option(struct apply_state *state, const char *option)
34 if (!option) {
35 state->ws_error_action = warn_on_ws_error;
36 return 0;
38 if (!strcmp(option, "warn")) {
39 state->ws_error_action = warn_on_ws_error;
40 return 0;
42 if (!strcmp(option, "nowarn")) {
43 state->ws_error_action = nowarn_ws_error;
44 return 0;
46 if (!strcmp(option, "error")) {
47 state->ws_error_action = die_on_ws_error;
48 return 0;
50 if (!strcmp(option, "error-all")) {
51 state->ws_error_action = die_on_ws_error;
52 state->squelch_whitespace_errors = 0;
53 return 0;
55 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
56 state->ws_error_action = correct_ws_error;
57 return 0;
59 return error(_("unrecognized whitespace option '%s'"), option);
62 static int parse_ignorewhitespace_option(struct apply_state *state,
63 const char *option)
65 if (!option || !strcmp(option, "no") ||
66 !strcmp(option, "false") || !strcmp(option, "never") ||
67 !strcmp(option, "none")) {
68 state->ws_ignore_action = ignore_ws_none;
69 return 0;
71 if (!strcmp(option, "change")) {
72 state->ws_ignore_action = ignore_ws_change;
73 return 0;
75 return error(_("unrecognized whitespace ignore option '%s'"), option);
78 int init_apply_state(struct apply_state *state,
79 struct repository *repo,
80 const char *prefix)
82 memset(state, 0, sizeof(*state));
83 state->prefix = prefix;
84 state->repo = repo;
85 state->apply = 1;
86 state->line_termination = '\n';
87 state->p_value = 1;
88 state->p_context = UINT_MAX;
89 state->squelch_whitespace_errors = 5;
90 state->ws_error_action = warn_on_ws_error;
91 state->ws_ignore_action = ignore_ws_none;
92 state->linenr = 1;
93 string_list_init(&state->fn_table, 0);
94 string_list_init(&state->limit_by_name, 0);
95 string_list_init(&state->symlink_changes, 0);
96 strbuf_init(&state->root, 0);
98 git_apply_config();
99 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
100 return -1;
101 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
102 return -1;
103 return 0;
106 void clear_apply_state(struct apply_state *state)
108 string_list_clear(&state->limit_by_name, 0);
109 string_list_clear(&state->symlink_changes, 0);
110 strbuf_release(&state->root);
112 /* &state->fn_table is cleared at the end of apply_patch() */
115 static void mute_routine(const char *msg, va_list params)
117 /* do nothing */
120 int check_apply_state(struct apply_state *state, int force_apply)
122 int is_not_gitdir = !startup_info->have_repository;
124 if (state->apply_with_reject && state->threeway)
125 return error(_("--reject and --3way cannot be used together."));
126 if (state->cached && state->threeway)
127 return error(_("--cached and --3way cannot be used together."));
128 if (state->threeway) {
129 if (is_not_gitdir)
130 return error(_("--3way outside a repository"));
131 state->check_index = 1;
133 if (state->apply_with_reject) {
134 state->apply = 1;
135 if (state->apply_verbosity == verbosity_normal)
136 state->apply_verbosity = verbosity_verbose;
138 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
139 state->apply = 0;
140 if (state->check_index && is_not_gitdir)
141 return error(_("--index outside a repository"));
142 if (state->cached) {
143 if (is_not_gitdir)
144 return error(_("--cached outside a repository"));
145 state->check_index = 1;
147 if (state->ita_only && (state->check_index || is_not_gitdir))
148 state->ita_only = 0;
149 if (state->check_index)
150 state->unsafe_paths = 0;
152 if (state->apply_verbosity <= verbosity_silent) {
153 state->saved_error_routine = get_error_routine();
154 state->saved_warn_routine = get_warn_routine();
155 set_error_routine(mute_routine);
156 set_warn_routine(mute_routine);
159 return 0;
162 static void set_default_whitespace_mode(struct apply_state *state)
164 if (!state->whitespace_option && !apply_default_whitespace)
165 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
169 * This represents one "hunk" from a patch, starting with
170 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
171 * patch text is pointed at by patch, and its byte length
172 * is stored in size. leading and trailing are the number
173 * of context lines.
175 struct fragment {
176 unsigned long leading, trailing;
177 unsigned long oldpos, oldlines;
178 unsigned long newpos, newlines;
180 * 'patch' is usually borrowed from buf in apply_patch(),
181 * but some codepaths store an allocated buffer.
183 const char *patch;
184 unsigned free_patch:1,
185 rejected:1;
186 int size;
187 int linenr;
188 struct fragment *next;
192 * When dealing with a binary patch, we reuse "leading" field
193 * to store the type of the binary hunk, either deflated "delta"
194 * or deflated "literal".
196 #define binary_patch_method leading
197 #define BINARY_DELTA_DEFLATED 1
198 #define BINARY_LITERAL_DEFLATED 2
201 * This represents a "patch" to a file, both metainfo changes
202 * such as creation/deletion, filemode and content changes represented
203 * as a series of fragments.
205 struct patch {
206 char *new_name, *old_name, *def_name;
207 unsigned int old_mode, new_mode;
208 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
209 int rejected;
210 unsigned ws_rule;
211 int lines_added, lines_deleted;
212 int score;
213 int extension_linenr; /* first line specifying delete/new/rename/copy */
214 unsigned int is_toplevel_relative:1;
215 unsigned int inaccurate_eof:1;
216 unsigned int is_binary:1;
217 unsigned int is_copy:1;
218 unsigned int is_rename:1;
219 unsigned int recount:1;
220 unsigned int conflicted_threeway:1;
221 unsigned int direct_to_threeway:1;
222 unsigned int crlf_in_old:1;
223 struct fragment *fragments;
224 char *result;
225 size_t resultsize;
226 char old_oid_prefix[GIT_MAX_HEXSZ + 1];
227 char new_oid_prefix[GIT_MAX_HEXSZ + 1];
228 struct patch *next;
230 /* three-way fallback result */
231 struct object_id threeway_stage[3];
234 static void free_fragment_list(struct fragment *list)
236 while (list) {
237 struct fragment *next = list->next;
238 if (list->free_patch)
239 free((char *)list->patch);
240 free(list);
241 list = next;
245 static void free_patch(struct patch *patch)
247 free_fragment_list(patch->fragments);
248 free(patch->def_name);
249 free(patch->old_name);
250 free(patch->new_name);
251 free(patch->result);
252 free(patch);
255 static void free_patch_list(struct patch *list)
257 while (list) {
258 struct patch *next = list->next;
259 free_patch(list);
260 list = next;
265 * A line in a file, len-bytes long (includes the terminating LF,
266 * except for an incomplete line at the end if the file ends with
267 * one), and its contents hashes to 'hash'.
269 struct line {
270 size_t len;
271 unsigned hash : 24;
272 unsigned flag : 8;
273 #define LINE_COMMON 1
274 #define LINE_PATCHED 2
278 * This represents a "file", which is an array of "lines".
280 struct image {
281 char *buf;
282 size_t len;
283 size_t nr;
284 size_t alloc;
285 struct line *line_allocated;
286 struct line *line;
289 static uint32_t hash_line(const char *cp, size_t len)
291 size_t i;
292 uint32_t h;
293 for (i = 0, h = 0; i < len; i++) {
294 if (!isspace(cp[i])) {
295 h = h * 3 + (cp[i] & 0xff);
298 return h;
302 * Compare lines s1 of length n1 and s2 of length n2, ignoring
303 * whitespace difference. Returns 1 if they match, 0 otherwise
305 static int fuzzy_matchlines(const char *s1, size_t n1,
306 const char *s2, size_t n2)
308 const char *end1 = s1 + n1;
309 const char *end2 = s2 + n2;
311 /* ignore line endings */
312 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
313 end1--;
314 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
315 end2--;
317 while (s1 < end1 && s2 < end2) {
318 if (isspace(*s1)) {
320 * Skip whitespace. We check on both buffers
321 * because we don't want "a b" to match "ab".
323 if (!isspace(*s2))
324 return 0;
325 while (s1 < end1 && isspace(*s1))
326 s1++;
327 while (s2 < end2 && isspace(*s2))
328 s2++;
329 } else if (*s1++ != *s2++)
330 return 0;
333 /* If we reached the end on one side only, lines don't match. */
334 return s1 == end1 && s2 == end2;
337 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
339 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
340 img->line_allocated[img->nr].len = len;
341 img->line_allocated[img->nr].hash = hash_line(bol, len);
342 img->line_allocated[img->nr].flag = flag;
343 img->nr++;
347 * "buf" has the file contents to be patched (read from various sources).
348 * attach it to "image" and add line-based index to it.
349 * "image" now owns the "buf".
351 static void prepare_image(struct image *image, char *buf, size_t len,
352 int prepare_linetable)
354 const char *cp, *ep;
356 memset(image, 0, sizeof(*image));
357 image->buf = buf;
358 image->len = len;
360 if (!prepare_linetable)
361 return;
363 ep = image->buf + image->len;
364 cp = image->buf;
365 while (cp < ep) {
366 const char *next;
367 for (next = cp; next < ep && *next != '\n'; next++)
369 if (next < ep)
370 next++;
371 add_line_info(image, cp, next - cp, 0);
372 cp = next;
374 image->line = image->line_allocated;
377 static void clear_image(struct image *image)
379 free(image->buf);
380 free(image->line_allocated);
381 memset(image, 0, sizeof(*image));
384 /* fmt must contain _one_ %s and no other substitution */
385 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
387 struct strbuf sb = STRBUF_INIT;
389 if (patch->old_name && patch->new_name &&
390 strcmp(patch->old_name, patch->new_name)) {
391 quote_c_style(patch->old_name, &sb, NULL, 0);
392 strbuf_addstr(&sb, " => ");
393 quote_c_style(patch->new_name, &sb, NULL, 0);
394 } else {
395 const char *n = patch->new_name;
396 if (!n)
397 n = patch->old_name;
398 quote_c_style(n, &sb, NULL, 0);
400 fprintf(output, fmt, sb.buf);
401 fputc('\n', output);
402 strbuf_release(&sb);
405 #define SLOP (16)
407 static int read_patch_file(struct strbuf *sb, int fd)
409 if (strbuf_read(sb, fd, 0) < 0)
410 return error_errno("git apply: failed to read");
413 * Make sure that we have some slop in the buffer
414 * so that we can do speculative "memcmp" etc, and
415 * see to it that it is NUL-filled.
417 strbuf_grow(sb, SLOP);
418 memset(sb->buf + sb->len, 0, SLOP);
419 return 0;
422 static unsigned long linelen(const char *buffer, unsigned long size)
424 unsigned long len = 0;
425 while (size--) {
426 len++;
427 if (*buffer++ == '\n')
428 break;
430 return len;
433 static int is_dev_null(const char *str)
435 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
438 #define TERM_SPACE 1
439 #define TERM_TAB 2
441 static int name_terminate(int c, int terminate)
443 if (c == ' ' && !(terminate & TERM_SPACE))
444 return 0;
445 if (c == '\t' && !(terminate & TERM_TAB))
446 return 0;
448 return 1;
451 /* remove double slashes to make --index work with such filenames */
452 static char *squash_slash(char *name)
454 int i = 0, j = 0;
456 if (!name)
457 return NULL;
459 while (name[i]) {
460 if ((name[j++] = name[i++]) == '/')
461 while (name[i] == '/')
462 i++;
464 name[j] = '\0';
465 return name;
468 static char *find_name_gnu(struct apply_state *state,
469 const char *line,
470 const char *def,
471 int p_value)
473 struct strbuf name = STRBUF_INIT;
474 char *cp;
477 * Proposed "new-style" GNU patch/diff format; see
478 * http://marc.info/?l=git&m=112927316408690&w=2
480 if (unquote_c_style(&name, line, NULL)) {
481 strbuf_release(&name);
482 return NULL;
485 for (cp = name.buf; p_value; p_value--) {
486 cp = strchr(cp, '/');
487 if (!cp) {
488 strbuf_release(&name);
489 return NULL;
491 cp++;
494 strbuf_remove(&name, 0, cp - name.buf);
495 if (state->root.len)
496 strbuf_insert(&name, 0, state->root.buf, state->root.len);
497 return squash_slash(strbuf_detach(&name, NULL));
500 static size_t sane_tz_len(const char *line, size_t len)
502 const char *tz, *p;
504 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
505 return 0;
506 tz = line + len - strlen(" +0500");
508 if (tz[1] != '+' && tz[1] != '-')
509 return 0;
511 for (p = tz + 2; p != line + len; p++)
512 if (!isdigit(*p))
513 return 0;
515 return line + len - tz;
518 static size_t tz_with_colon_len(const char *line, size_t len)
520 const char *tz, *p;
522 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
523 return 0;
524 tz = line + len - strlen(" +08:00");
526 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
527 return 0;
528 p = tz + 2;
529 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
530 !isdigit(*p++) || !isdigit(*p++))
531 return 0;
533 return line + len - tz;
536 static size_t date_len(const char *line, size_t len)
538 const char *date, *p;
540 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
541 return 0;
542 p = date = line + len - strlen("72-02-05");
544 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
545 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
546 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
547 return 0;
549 if (date - line >= strlen("19") &&
550 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
551 date -= strlen("19");
553 return line + len - date;
556 static size_t short_time_len(const char *line, size_t len)
558 const char *time, *p;
560 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
561 return 0;
562 p = time = line + len - strlen(" 07:01:32");
564 /* Permit 1-digit hours? */
565 if (*p++ != ' ' ||
566 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
568 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
569 return 0;
571 return line + len - time;
574 static size_t fractional_time_len(const char *line, size_t len)
576 const char *p;
577 size_t n;
579 /* Expected format: 19:41:17.620000023 */
580 if (!len || !isdigit(line[len - 1]))
581 return 0;
582 p = line + len - 1;
584 /* Fractional seconds. */
585 while (p > line && isdigit(*p))
586 p--;
587 if (*p != '.')
588 return 0;
590 /* Hours, minutes, and whole seconds. */
591 n = short_time_len(line, p - line);
592 if (!n)
593 return 0;
595 return line + len - p + n;
598 static size_t trailing_spaces_len(const char *line, size_t len)
600 const char *p;
602 /* Expected format: ' ' x (1 or more) */
603 if (!len || line[len - 1] != ' ')
604 return 0;
606 p = line + len;
607 while (p != line) {
608 p--;
609 if (*p != ' ')
610 return line + len - (p + 1);
613 /* All spaces! */
614 return len;
617 static size_t diff_timestamp_len(const char *line, size_t len)
619 const char *end = line + len;
620 size_t n;
623 * Posix: 2010-07-05 19:41:17
624 * GNU: 2010-07-05 19:41:17.620000023 -0500
627 if (!isdigit(end[-1]))
628 return 0;
630 n = sane_tz_len(line, end - line);
631 if (!n)
632 n = tz_with_colon_len(line, end - line);
633 end -= n;
635 n = short_time_len(line, end - line);
636 if (!n)
637 n = fractional_time_len(line, end - line);
638 end -= n;
640 n = date_len(line, end - line);
641 if (!n) /* No date. Too bad. */
642 return 0;
643 end -= n;
645 if (end == line) /* No space before date. */
646 return 0;
647 if (end[-1] == '\t') { /* Success! */
648 end--;
649 return line + len - end;
651 if (end[-1] != ' ') /* No space before date. */
652 return 0;
654 /* Whitespace damage. */
655 end -= trailing_spaces_len(line, end - line);
656 return line + len - end;
659 static char *find_name_common(struct apply_state *state,
660 const char *line,
661 const char *def,
662 int p_value,
663 const char *end,
664 int terminate)
666 int len;
667 const char *start = NULL;
669 if (p_value == 0)
670 start = line;
671 while (line != end) {
672 char c = *line;
674 if (!end && isspace(c)) {
675 if (c == '\n')
676 break;
677 if (name_terminate(c, terminate))
678 break;
680 line++;
681 if (c == '/' && !--p_value)
682 start = line;
684 if (!start)
685 return squash_slash(xstrdup_or_null(def));
686 len = line - start;
687 if (!len)
688 return squash_slash(xstrdup_or_null(def));
691 * Generally we prefer the shorter name, especially
692 * if the other one is just a variation of that with
693 * something else tacked on to the end (ie "file.orig"
694 * or "file~").
696 if (def) {
697 int deflen = strlen(def);
698 if (deflen < len && !strncmp(start, def, deflen))
699 return squash_slash(xstrdup(def));
702 if (state->root.len) {
703 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
704 return squash_slash(ret);
707 return squash_slash(xmemdupz(start, len));
710 static char *find_name(struct apply_state *state,
711 const char *line,
712 char *def,
713 int p_value,
714 int terminate)
716 if (*line == '"') {
717 char *name = find_name_gnu(state, line, def, p_value);
718 if (name)
719 return name;
722 return find_name_common(state, line, def, p_value, NULL, terminate);
725 static char *find_name_traditional(struct apply_state *state,
726 const char *line,
727 char *def,
728 int p_value)
730 size_t len;
731 size_t date_len;
733 if (*line == '"') {
734 char *name = find_name_gnu(state, line, def, p_value);
735 if (name)
736 return name;
739 len = strchrnul(line, '\n') - line;
740 date_len = diff_timestamp_len(line, len);
741 if (!date_len)
742 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
743 len -= date_len;
745 return find_name_common(state, line, def, p_value, line + len, 0);
749 * Given the string after "--- " or "+++ ", guess the appropriate
750 * p_value for the given patch.
752 static int guess_p_value(struct apply_state *state, const char *nameline)
754 char *name, *cp;
755 int val = -1;
757 if (is_dev_null(nameline))
758 return -1;
759 name = find_name_traditional(state, nameline, NULL, 0);
760 if (!name)
761 return -1;
762 cp = strchr(name, '/');
763 if (!cp)
764 val = 0;
765 else if (state->prefix) {
767 * Does it begin with "a/$our-prefix" and such? Then this is
768 * very likely to apply to our directory.
770 if (starts_with(name, state->prefix))
771 val = count_slashes(state->prefix);
772 else {
773 cp++;
774 if (starts_with(cp, state->prefix))
775 val = count_slashes(state->prefix) + 1;
778 free(name);
779 return val;
783 * Does the ---/+++ line have the POSIX timestamp after the last HT?
784 * GNU diff puts epoch there to signal a creation/deletion event. Is
785 * this such a timestamp?
787 static int has_epoch_timestamp(const char *nameline)
790 * We are only interested in epoch timestamp; any non-zero
791 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
792 * For the same reason, the date must be either 1969-12-31 or
793 * 1970-01-01, and the seconds part must be "00".
795 const char stamp_regexp[] =
796 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
798 "([-+][0-2][0-9]:?[0-5][0-9])\n";
799 const char *timestamp = NULL, *cp, *colon;
800 static regex_t *stamp;
801 regmatch_t m[10];
802 int zoneoffset, epoch_hour, hour, minute;
803 int status;
805 for (cp = nameline; *cp != '\n'; cp++) {
806 if (*cp == '\t')
807 timestamp = cp + 1;
809 if (!timestamp)
810 return 0;
813 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
814 * (west of GMT) or 1970-01-01 (east of GMT)
816 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
817 epoch_hour = 24;
818 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
819 epoch_hour = 0;
820 else
821 return 0;
823 if (!stamp) {
824 stamp = xmalloc(sizeof(*stamp));
825 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
826 warning(_("Cannot prepare timestamp regexp %s"),
827 stamp_regexp);
828 return 0;
832 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
833 if (status) {
834 if (status != REG_NOMATCH)
835 warning(_("regexec returned %d for input: %s"),
836 status, timestamp);
837 return 0;
840 hour = strtol(timestamp, NULL, 10);
841 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
843 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
844 if (*colon == ':')
845 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
846 else
847 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
848 if (timestamp[m[3].rm_so] == '-')
849 zoneoffset = -zoneoffset;
851 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
855 * Get the name etc info from the ---/+++ lines of a traditional patch header
857 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
858 * files, we can happily check the index for a match, but for creating a
859 * new file we should try to match whatever "patch" does. I have no idea.
861 static int parse_traditional_patch(struct apply_state *state,
862 const char *first,
863 const char *second,
864 struct patch *patch)
866 char *name;
868 first += 4; /* skip "--- " */
869 second += 4; /* skip "+++ " */
870 if (!state->p_value_known) {
871 int p, q;
872 p = guess_p_value(state, first);
873 q = guess_p_value(state, second);
874 if (p < 0) p = q;
875 if (0 <= p && p == q) {
876 state->p_value = p;
877 state->p_value_known = 1;
880 if (is_dev_null(first)) {
881 patch->is_new = 1;
882 patch->is_delete = 0;
883 name = find_name_traditional(state, second, NULL, state->p_value);
884 patch->new_name = name;
885 } else if (is_dev_null(second)) {
886 patch->is_new = 0;
887 patch->is_delete = 1;
888 name = find_name_traditional(state, first, NULL, state->p_value);
889 patch->old_name = name;
890 } else {
891 char *first_name;
892 first_name = find_name_traditional(state, first, NULL, state->p_value);
893 name = find_name_traditional(state, second, first_name, state->p_value);
894 free(first_name);
895 if (has_epoch_timestamp(first)) {
896 patch->is_new = 1;
897 patch->is_delete = 0;
898 patch->new_name = name;
899 } else if (has_epoch_timestamp(second)) {
900 patch->is_new = 0;
901 patch->is_delete = 1;
902 patch->old_name = name;
903 } else {
904 patch->old_name = name;
905 patch->new_name = xstrdup_or_null(name);
908 if (!name)
909 return error(_("unable to find filename in patch at line %d"), state->linenr);
911 return 0;
914 static int gitdiff_hdrend(struct apply_state *state,
915 const char *line,
916 struct patch *patch)
918 return 1;
922 * We're anal about diff header consistency, to make
923 * sure that we don't end up having strange ambiguous
924 * patches floating around.
926 * As a result, gitdiff_{old|new}name() will check
927 * their names against any previous information, just
928 * to make sure..
930 #define DIFF_OLD_NAME 0
931 #define DIFF_NEW_NAME 1
933 static int gitdiff_verify_name(struct apply_state *state,
934 const char *line,
935 int isnull,
936 char **name,
937 int side)
939 if (!*name && !isnull) {
940 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
941 return 0;
944 if (*name) {
945 char *another;
946 if (isnull)
947 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
948 *name, state->linenr);
949 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
950 if (!another || strcmp(another, *name)) {
951 free(another);
952 return error((side == DIFF_NEW_NAME) ?
953 _("git apply: bad git-diff - inconsistent new filename on line %d") :
954 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
956 free(another);
957 } else {
958 if (!is_dev_null(line))
959 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
962 return 0;
965 static int gitdiff_oldname(struct apply_state *state,
966 const char *line,
967 struct patch *patch)
969 return gitdiff_verify_name(state, line,
970 patch->is_new, &patch->old_name,
971 DIFF_OLD_NAME);
974 static int gitdiff_newname(struct apply_state *state,
975 const char *line,
976 struct patch *patch)
978 return gitdiff_verify_name(state, line,
979 patch->is_delete, &patch->new_name,
980 DIFF_NEW_NAME);
983 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
985 char *end;
986 *mode = strtoul(line, &end, 8);
987 if (end == line || !isspace(*end))
988 return error(_("invalid mode on line %d: %s"), linenr, line);
989 return 0;
992 static int gitdiff_oldmode(struct apply_state *state,
993 const char *line,
994 struct patch *patch)
996 return parse_mode_line(line, state->linenr, &patch->old_mode);
999 static int gitdiff_newmode(struct apply_state *state,
1000 const char *line,
1001 struct patch *patch)
1003 return parse_mode_line(line, state->linenr, &patch->new_mode);
1006 static int gitdiff_delete(struct apply_state *state,
1007 const char *line,
1008 struct patch *patch)
1010 patch->is_delete = 1;
1011 free(patch->old_name);
1012 patch->old_name = xstrdup_or_null(patch->def_name);
1013 return gitdiff_oldmode(state, line, patch);
1016 static int gitdiff_newfile(struct apply_state *state,
1017 const char *line,
1018 struct patch *patch)
1020 patch->is_new = 1;
1021 free(patch->new_name);
1022 patch->new_name = xstrdup_or_null(patch->def_name);
1023 return gitdiff_newmode(state, line, patch);
1026 static int gitdiff_copysrc(struct apply_state *state,
1027 const char *line,
1028 struct patch *patch)
1030 patch->is_copy = 1;
1031 free(patch->old_name);
1032 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1033 return 0;
1036 static int gitdiff_copydst(struct apply_state *state,
1037 const char *line,
1038 struct patch *patch)
1040 patch->is_copy = 1;
1041 free(patch->new_name);
1042 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1043 return 0;
1046 static int gitdiff_renamesrc(struct apply_state *state,
1047 const char *line,
1048 struct patch *patch)
1050 patch->is_rename = 1;
1051 free(patch->old_name);
1052 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1053 return 0;
1056 static int gitdiff_renamedst(struct apply_state *state,
1057 const char *line,
1058 struct patch *patch)
1060 patch->is_rename = 1;
1061 free(patch->new_name);
1062 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1063 return 0;
1066 static int gitdiff_similarity(struct apply_state *state,
1067 const char *line,
1068 struct patch *patch)
1070 unsigned long val = strtoul(line, NULL, 10);
1071 if (val <= 100)
1072 patch->score = val;
1073 return 0;
1076 static int gitdiff_dissimilarity(struct apply_state *state,
1077 const char *line,
1078 struct patch *patch)
1080 unsigned long val = strtoul(line, NULL, 10);
1081 if (val <= 100)
1082 patch->score = val;
1083 return 0;
1086 static int gitdiff_index(struct apply_state *state,
1087 const char *line,
1088 struct patch *patch)
1091 * index line is N hexadecimal, "..", N hexadecimal,
1092 * and optional space with octal mode.
1094 const char *ptr, *eol;
1095 int len;
1096 const unsigned hexsz = the_hash_algo->hexsz;
1098 ptr = strchr(line, '.');
1099 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1100 return 0;
1101 len = ptr - line;
1102 memcpy(patch->old_oid_prefix, line, len);
1103 patch->old_oid_prefix[len] = 0;
1105 line = ptr + 2;
1106 ptr = strchr(line, ' ');
1107 eol = strchrnul(line, '\n');
1109 if (!ptr || eol < ptr)
1110 ptr = eol;
1111 len = ptr - line;
1113 if (hexsz < len)
1114 return 0;
1115 memcpy(patch->new_oid_prefix, line, len);
1116 patch->new_oid_prefix[len] = 0;
1117 if (*ptr == ' ')
1118 return gitdiff_oldmode(state, ptr + 1, patch);
1119 return 0;
1123 * This is normal for a diff that doesn't change anything: we'll fall through
1124 * into the next diff. Tell the parser to break out.
1126 static int gitdiff_unrecognized(struct apply_state *state,
1127 const char *line,
1128 struct patch *patch)
1130 return 1;
1134 * Skip p_value leading components from "line"; as we do not accept
1135 * absolute paths, return NULL in that case.
1137 static const char *skip_tree_prefix(struct apply_state *state,
1138 const char *line,
1139 int llen)
1141 int nslash;
1142 int i;
1144 if (!state->p_value)
1145 return (llen && line[0] == '/') ? NULL : line;
1147 nslash = state->p_value;
1148 for (i = 0; i < llen; i++) {
1149 int ch = line[i];
1150 if (ch == '/' && --nslash <= 0)
1151 return (i == 0) ? NULL : &line[i + 1];
1153 return NULL;
1157 * This is to extract the same name that appears on "diff --git"
1158 * line. We do not find and return anything if it is a rename
1159 * patch, and it is OK because we will find the name elsewhere.
1160 * We need to reliably find name only when it is mode-change only,
1161 * creation or deletion of an empty file. In any of these cases,
1162 * both sides are the same name under a/ and b/ respectively.
1164 static char *git_header_name(struct apply_state *state,
1165 const char *line,
1166 int llen)
1168 const char *name;
1169 const char *second = NULL;
1170 size_t len, line_len;
1172 line += strlen("diff --git ");
1173 llen -= strlen("diff --git ");
1175 if (*line == '"') {
1176 const char *cp;
1177 struct strbuf first = STRBUF_INIT;
1178 struct strbuf sp = STRBUF_INIT;
1180 if (unquote_c_style(&first, line, &second))
1181 goto free_and_fail1;
1183 /* strip the a/b prefix including trailing slash */
1184 cp = skip_tree_prefix(state, first.buf, first.len);
1185 if (!cp)
1186 goto free_and_fail1;
1187 strbuf_remove(&first, 0, cp - first.buf);
1190 * second points at one past closing dq of name.
1191 * find the second name.
1193 while ((second < line + llen) && isspace(*second))
1194 second++;
1196 if (line + llen <= second)
1197 goto free_and_fail1;
1198 if (*second == '"') {
1199 if (unquote_c_style(&sp, second, NULL))
1200 goto free_and_fail1;
1201 cp = skip_tree_prefix(state, sp.buf, sp.len);
1202 if (!cp)
1203 goto free_and_fail1;
1204 /* They must match, otherwise ignore */
1205 if (strcmp(cp, first.buf))
1206 goto free_and_fail1;
1207 strbuf_release(&sp);
1208 return strbuf_detach(&first, NULL);
1211 /* unquoted second */
1212 cp = skip_tree_prefix(state, second, line + llen - second);
1213 if (!cp)
1214 goto free_and_fail1;
1215 if (line + llen - cp != first.len ||
1216 memcmp(first.buf, cp, first.len))
1217 goto free_and_fail1;
1218 return strbuf_detach(&first, NULL);
1220 free_and_fail1:
1221 strbuf_release(&first);
1222 strbuf_release(&sp);
1223 return NULL;
1226 /* unquoted first name */
1227 name = skip_tree_prefix(state, line, llen);
1228 if (!name)
1229 return NULL;
1232 * since the first name is unquoted, a dq if exists must be
1233 * the beginning of the second name.
1235 for (second = name; second < line + llen; second++) {
1236 if (*second == '"') {
1237 struct strbuf sp = STRBUF_INIT;
1238 const char *np;
1240 if (unquote_c_style(&sp, second, NULL))
1241 goto free_and_fail2;
1243 np = skip_tree_prefix(state, sp.buf, sp.len);
1244 if (!np)
1245 goto free_and_fail2;
1247 len = sp.buf + sp.len - np;
1248 if (len < second - name &&
1249 !strncmp(np, name, len) &&
1250 isspace(name[len])) {
1251 /* Good */
1252 strbuf_remove(&sp, 0, np - sp.buf);
1253 return strbuf_detach(&sp, NULL);
1256 free_and_fail2:
1257 strbuf_release(&sp);
1258 return NULL;
1263 * Accept a name only if it shows up twice, exactly the same
1264 * form.
1266 second = strchr(name, '\n');
1267 if (!second)
1268 return NULL;
1269 line_len = second - name;
1270 for (len = 0 ; ; len++) {
1271 switch (name[len]) {
1272 default:
1273 continue;
1274 case '\n':
1275 return NULL;
1276 case '\t': case ' ':
1278 * Is this the separator between the preimage
1279 * and the postimage pathname? Again, we are
1280 * only interested in the case where there is
1281 * no rename, as this is only to set def_name
1282 * and a rename patch has the names elsewhere
1283 * in an unambiguous form.
1285 if (!name[len + 1])
1286 return NULL; /* no postimage name */
1287 second = skip_tree_prefix(state, name + len + 1,
1288 line_len - (len + 1));
1289 if (!second)
1290 return NULL;
1292 * Does len bytes starting at "name" and "second"
1293 * (that are separated by one HT or SP we just
1294 * found) exactly match?
1296 if (second[len] == '\n' && !strncmp(name, second, len))
1297 return xmemdupz(name, len);
1302 static int check_header_line(struct apply_state *state, struct patch *patch)
1304 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1305 (patch->is_rename == 1) + (patch->is_copy == 1);
1306 if (extensions > 1)
1307 return error(_("inconsistent header lines %d and %d"),
1308 patch->extension_linenr, state->linenr);
1309 if (extensions && !patch->extension_linenr)
1310 patch->extension_linenr = state->linenr;
1311 return 0;
1314 /* Verify that we recognize the lines following a git header */
1315 static int parse_git_header(struct apply_state *state,
1316 const char *line,
1317 int len,
1318 unsigned int size,
1319 struct patch *patch)
1321 unsigned long offset;
1323 /* A git diff has explicit new/delete information, so we don't guess */
1324 patch->is_new = 0;
1325 patch->is_delete = 0;
1328 * Some things may not have the old name in the
1329 * rest of the headers anywhere (pure mode changes,
1330 * or removing or adding empty files), so we get
1331 * the default name from the header.
1333 patch->def_name = git_header_name(state, line, len);
1334 if (patch->def_name && state->root.len) {
1335 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1336 free(patch->def_name);
1337 patch->def_name = s;
1340 line += len;
1341 size -= len;
1342 state->linenr++;
1343 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1344 static const struct opentry {
1345 const char *str;
1346 int (*fn)(struct apply_state *, const char *, struct patch *);
1347 } optable[] = {
1348 { "@@ -", gitdiff_hdrend },
1349 { "--- ", gitdiff_oldname },
1350 { "+++ ", gitdiff_newname },
1351 { "old mode ", gitdiff_oldmode },
1352 { "new mode ", gitdiff_newmode },
1353 { "deleted file mode ", gitdiff_delete },
1354 { "new file mode ", gitdiff_newfile },
1355 { "copy from ", gitdiff_copysrc },
1356 { "copy to ", gitdiff_copydst },
1357 { "rename old ", gitdiff_renamesrc },
1358 { "rename new ", gitdiff_renamedst },
1359 { "rename from ", gitdiff_renamesrc },
1360 { "rename to ", gitdiff_renamedst },
1361 { "similarity index ", gitdiff_similarity },
1362 { "dissimilarity index ", gitdiff_dissimilarity },
1363 { "index ", gitdiff_index },
1364 { "", gitdiff_unrecognized },
1366 int i;
1368 len = linelen(line, size);
1369 if (!len || line[len-1] != '\n')
1370 break;
1371 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1372 const struct opentry *p = optable + i;
1373 int oplen = strlen(p->str);
1374 int res;
1375 if (len < oplen || memcmp(p->str, line, oplen))
1376 continue;
1377 res = p->fn(state, line + oplen, patch);
1378 if (res < 0)
1379 return -1;
1380 if (check_header_line(state, patch))
1381 return -1;
1382 if (res > 0)
1383 return offset;
1384 break;
1388 return offset;
1391 static int parse_num(const char *line, unsigned long *p)
1393 char *ptr;
1395 if (!isdigit(*line))
1396 return 0;
1397 *p = strtoul(line, &ptr, 10);
1398 return ptr - line;
1401 static int parse_range(const char *line, int len, int offset, const char *expect,
1402 unsigned long *p1, unsigned long *p2)
1404 int digits, ex;
1406 if (offset < 0 || offset >= len)
1407 return -1;
1408 line += offset;
1409 len -= offset;
1411 digits = parse_num(line, p1);
1412 if (!digits)
1413 return -1;
1415 offset += digits;
1416 line += digits;
1417 len -= digits;
1419 *p2 = 1;
1420 if (*line == ',') {
1421 digits = parse_num(line+1, p2);
1422 if (!digits)
1423 return -1;
1425 offset += digits+1;
1426 line += digits+1;
1427 len -= digits+1;
1430 ex = strlen(expect);
1431 if (ex > len)
1432 return -1;
1433 if (memcmp(line, expect, ex))
1434 return -1;
1436 return offset + ex;
1439 static void recount_diff(const char *line, int size, struct fragment *fragment)
1441 int oldlines = 0, newlines = 0, ret = 0;
1443 if (size < 1) {
1444 warning("recount: ignore empty hunk");
1445 return;
1448 for (;;) {
1449 int len = linelen(line, size);
1450 size -= len;
1451 line += len;
1453 if (size < 1)
1454 break;
1456 switch (*line) {
1457 case ' ': case '\n':
1458 newlines++;
1459 /* fall through */
1460 case '-':
1461 oldlines++;
1462 continue;
1463 case '+':
1464 newlines++;
1465 continue;
1466 case '\\':
1467 continue;
1468 case '@':
1469 ret = size < 3 || !starts_with(line, "@@ ");
1470 break;
1471 case 'd':
1472 ret = size < 5 || !starts_with(line, "diff ");
1473 break;
1474 default:
1475 ret = -1;
1476 break;
1478 if (ret) {
1479 warning(_("recount: unexpected line: %.*s"),
1480 (int)linelen(line, size), line);
1481 return;
1483 break;
1485 fragment->oldlines = oldlines;
1486 fragment->newlines = newlines;
1490 * Parse a unified diff fragment header of the
1491 * form "@@ -a,b +c,d @@"
1493 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1495 int offset;
1497 if (!len || line[len-1] != '\n')
1498 return -1;
1500 /* Figure out the number of lines in a fragment */
1501 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1502 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1504 return offset;
1508 * Find file diff header
1510 * Returns:
1511 * -1 if no header was found
1512 * -128 in case of error
1513 * the size of the header in bytes (called "offset") otherwise
1515 static int find_header(struct apply_state *state,
1516 const char *line,
1517 unsigned long size,
1518 int *hdrsize,
1519 struct patch *patch)
1521 unsigned long offset, len;
1523 patch->is_toplevel_relative = 0;
1524 patch->is_rename = patch->is_copy = 0;
1525 patch->is_new = patch->is_delete = -1;
1526 patch->old_mode = patch->new_mode = 0;
1527 patch->old_name = patch->new_name = NULL;
1528 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1529 unsigned long nextlen;
1531 len = linelen(line, size);
1532 if (!len)
1533 break;
1535 /* Testing this early allows us to take a few shortcuts.. */
1536 if (len < 6)
1537 continue;
1540 * Make sure we don't find any unconnected patch fragments.
1541 * That's a sign that we didn't find a header, and that a
1542 * patch has become corrupted/broken up.
1544 if (!memcmp("@@ -", line, 4)) {
1545 struct fragment dummy;
1546 if (parse_fragment_header(line, len, &dummy) < 0)
1547 continue;
1548 error(_("patch fragment without header at line %d: %.*s"),
1549 state->linenr, (int)len-1, line);
1550 return -128;
1553 if (size < len + 6)
1554 break;
1557 * Git patch? It might not have a real patch, just a rename
1558 * or mode change, so we handle that specially
1560 if (!memcmp("diff --git ", line, 11)) {
1561 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1562 if (git_hdr_len < 0)
1563 return -128;
1564 if (git_hdr_len <= len)
1565 continue;
1566 if (!patch->old_name && !patch->new_name) {
1567 if (!patch->def_name) {
1568 error(Q_("git diff header lacks filename information when removing "
1569 "%d leading pathname component (line %d)",
1570 "git diff header lacks filename information when removing "
1571 "%d leading pathname components (line %d)",
1572 state->p_value),
1573 state->p_value, state->linenr);
1574 return -128;
1576 patch->old_name = xstrdup(patch->def_name);
1577 patch->new_name = xstrdup(patch->def_name);
1579 if ((!patch->new_name && !patch->is_delete) ||
1580 (!patch->old_name && !patch->is_new)) {
1581 error(_("git diff header lacks filename information "
1582 "(line %d)"), state->linenr);
1583 return -128;
1585 patch->is_toplevel_relative = 1;
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 fragment = xcalloc(1, sizeof(*fragment));
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 while (1) {
1938 int byte_length, max_byte_length, newsize;
1939 llen = linelen(buffer, size);
1940 used += llen;
1941 state->linenr++;
1942 if (llen == 1) {
1943 /* consume the blank line */
1944 buffer++;
1945 size--;
1946 break;
1949 * Minimum line is "A00000\n" which is 7-byte long,
1950 * and the line length must be multiple of 5 plus 2.
1952 if ((llen < 7) || (llen-2) % 5)
1953 goto corrupt;
1954 max_byte_length = (llen - 2) / 5 * 4;
1955 byte_length = *buffer;
1956 if ('A' <= byte_length && byte_length <= 'Z')
1957 byte_length = byte_length - 'A' + 1;
1958 else if ('a' <= byte_length && byte_length <= 'z')
1959 byte_length = byte_length - 'a' + 27;
1960 else
1961 goto corrupt;
1962 /* if the input length was not multiple of 4, we would
1963 * have filler at the end but the filler should never
1964 * exceed 3 bytes
1966 if (max_byte_length < byte_length ||
1967 byte_length <= max_byte_length - 4)
1968 goto corrupt;
1969 newsize = hunk_size + byte_length;
1970 data = xrealloc(data, newsize);
1971 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1972 goto corrupt;
1973 hunk_size = newsize;
1974 buffer += llen;
1975 size -= llen;
1978 frag = xcalloc(1, sizeof(*frag));
1979 frag->patch = inflate_it(data, hunk_size, origlen);
1980 frag->free_patch = 1;
1981 if (!frag->patch)
1982 goto corrupt;
1983 free(data);
1984 frag->size = origlen;
1985 *buf_p = buffer;
1986 *sz_p = size;
1987 *used_p = used;
1988 frag->binary_patch_method = patch_method;
1989 return frag;
1991 corrupt:
1992 free(data);
1993 *status_p = -1;
1994 error(_("corrupt binary patch at line %d: %.*s"),
1995 state->linenr-1, llen-1, buffer);
1996 return NULL;
2000 * Returns:
2001 * -1 in case of error,
2002 * the length of the parsed binary patch otherwise
2004 static int parse_binary(struct apply_state *state,
2005 char *buffer,
2006 unsigned long size,
2007 struct patch *patch)
2010 * We have read "GIT binary patch\n"; what follows is a line
2011 * that says the patch method (currently, either "literal" or
2012 * "delta") and the length of data before deflating; a
2013 * sequence of 'length-byte' followed by base-85 encoded data
2014 * follows.
2016 * When a binary patch is reversible, there is another binary
2017 * hunk in the same format, starting with patch method (either
2018 * "literal" or "delta") with the length of data, and a sequence
2019 * of length-byte + base-85 encoded data, terminated with another
2020 * empty line. This data, when applied to the postimage, produces
2021 * the preimage.
2023 struct fragment *forward;
2024 struct fragment *reverse;
2025 int status;
2026 int used, used_1;
2028 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2029 if (!forward && !status)
2030 /* there has to be one hunk (forward hunk) */
2031 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2032 if (status)
2033 /* otherwise we already gave an error message */
2034 return status;
2036 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2037 if (reverse)
2038 used += used_1;
2039 else if (status) {
2041 * Not having reverse hunk is not an error, but having
2042 * a corrupt reverse hunk is.
2044 free((void*) forward->patch);
2045 free(forward);
2046 return status;
2048 forward->next = reverse;
2049 patch->fragments = forward;
2050 patch->is_binary = 1;
2051 return used;
2054 static void prefix_one(struct apply_state *state, char **name)
2056 char *old_name = *name;
2057 if (!old_name)
2058 return;
2059 *name = prefix_filename(state->prefix, *name);
2060 free(old_name);
2063 static void prefix_patch(struct apply_state *state, struct patch *p)
2065 if (!state->prefix || p->is_toplevel_relative)
2066 return;
2067 prefix_one(state, &p->new_name);
2068 prefix_one(state, &p->old_name);
2072 * include/exclude
2075 static void add_name_limit(struct apply_state *state,
2076 const char *name,
2077 int exclude)
2079 struct string_list_item *it;
2081 it = string_list_append(&state->limit_by_name, name);
2082 it->util = exclude ? NULL : (void *) 1;
2085 static int use_patch(struct apply_state *state, struct patch *p)
2087 const char *pathname = p->new_name ? p->new_name : p->old_name;
2088 int i;
2090 /* Paths outside are not touched regardless of "--include" */
2091 if (state->prefix && *state->prefix) {
2092 const char *rest;
2093 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2094 return 0;
2097 /* See if it matches any of exclude/include rule */
2098 for (i = 0; i < state->limit_by_name.nr; i++) {
2099 struct string_list_item *it = &state->limit_by_name.items[i];
2100 if (!wildmatch(it->string, pathname, 0))
2101 return (it->util != NULL);
2105 * If we had any include, a path that does not match any rule is
2106 * not used. Otherwise, we saw bunch of exclude rules (or none)
2107 * and such a path is used.
2109 return !state->has_include;
2113 * Read the patch text in "buffer" that extends for "size" bytes; stop
2114 * reading after seeing a single patch (i.e. changes to a single file).
2115 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2117 * Returns:
2118 * -1 if no header was found or parse_binary() failed,
2119 * -128 on another error,
2120 * the number of bytes consumed otherwise,
2121 * so that the caller can call us again for the next patch.
2123 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2125 int hdrsize, patchsize;
2126 int offset = find_header(state, buffer, size, &hdrsize, patch);
2128 if (offset < 0)
2129 return offset;
2131 prefix_patch(state, patch);
2133 if (!use_patch(state, patch))
2134 patch->ws_rule = 0;
2135 else if (patch->new_name)
2136 patch->ws_rule = whitespace_rule(state->repo->index,
2137 patch->new_name);
2138 else
2139 patch->ws_rule = whitespace_rule(state->repo->index,
2140 patch->old_name);
2142 patchsize = parse_single_patch(state,
2143 buffer + offset + hdrsize,
2144 size - offset - hdrsize,
2145 patch);
2147 if (patchsize < 0)
2148 return -128;
2150 if (!patchsize) {
2151 static const char git_binary[] = "GIT binary patch\n";
2152 int hd = hdrsize + offset;
2153 unsigned long llen = linelen(buffer + hd, size - hd);
2155 if (llen == sizeof(git_binary) - 1 &&
2156 !memcmp(git_binary, buffer + hd, llen)) {
2157 int used;
2158 state->linenr++;
2159 used = parse_binary(state, buffer + hd + llen,
2160 size - hd - llen, patch);
2161 if (used < 0)
2162 return -1;
2163 if (used)
2164 patchsize = used + llen;
2165 else
2166 patchsize = 0;
2168 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2169 static const char *binhdr[] = {
2170 "Binary files ",
2171 "Files ",
2172 NULL,
2174 int i;
2175 for (i = 0; binhdr[i]; i++) {
2176 int len = strlen(binhdr[i]);
2177 if (len < size - hd &&
2178 !memcmp(binhdr[i], buffer + hd, len)) {
2179 state->linenr++;
2180 patch->is_binary = 1;
2181 patchsize = llen;
2182 break;
2187 /* Empty patch cannot be applied if it is a text patch
2188 * without metadata change. A binary patch appears
2189 * empty to us here.
2191 if ((state->apply || state->check) &&
2192 (!patch->is_binary && !metadata_changes(patch))) {
2193 error(_("patch with only garbage at line %d"), state->linenr);
2194 return -128;
2198 return offset + hdrsize + patchsize;
2201 static void reverse_patches(struct patch *p)
2203 for (; p; p = p->next) {
2204 struct fragment *frag = p->fragments;
2206 SWAP(p->new_name, p->old_name);
2207 SWAP(p->new_mode, p->old_mode);
2208 SWAP(p->is_new, p->is_delete);
2209 SWAP(p->lines_added, p->lines_deleted);
2210 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2212 for (; frag; frag = frag->next) {
2213 SWAP(frag->newpos, frag->oldpos);
2214 SWAP(frag->newlines, frag->oldlines);
2219 static const char pluses[] =
2220 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2221 static const char minuses[]=
2222 "----------------------------------------------------------------------";
2224 static void show_stats(struct apply_state *state, struct patch *patch)
2226 struct strbuf qname = STRBUF_INIT;
2227 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2228 int max, add, del;
2230 quote_c_style(cp, &qname, NULL, 0);
2233 * "scale" the filename
2235 max = state->max_len;
2236 if (max > 50)
2237 max = 50;
2239 if (qname.len > max) {
2240 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2241 if (!cp)
2242 cp = qname.buf + qname.len + 3 - max;
2243 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2246 if (patch->is_binary) {
2247 printf(" %-*s | Bin\n", max, qname.buf);
2248 strbuf_release(&qname);
2249 return;
2252 printf(" %-*s |", max, qname.buf);
2253 strbuf_release(&qname);
2256 * scale the add/delete
2258 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2259 add = patch->lines_added;
2260 del = patch->lines_deleted;
2262 if (state->max_change > 0) {
2263 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2264 add = (add * max + state->max_change / 2) / state->max_change;
2265 del = total - add;
2267 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2268 add, pluses, del, minuses);
2271 static int read_old_data(struct stat *st, struct patch *patch,
2272 const char *path, struct strbuf *buf)
2274 int conv_flags = patch->crlf_in_old ?
2275 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2276 switch (st->st_mode & S_IFMT) {
2277 case S_IFLNK:
2278 if (strbuf_readlink(buf, path, st->st_size) < 0)
2279 return error(_("unable to read symlink %s"), path);
2280 return 0;
2281 case S_IFREG:
2282 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2283 return error(_("unable to open or read %s"), path);
2285 * "git apply" without "--index/--cached" should never look
2286 * at the index; the target file may not have been added to
2287 * the index yet, and we may not even be in any Git repository.
2288 * Pass NULL to convert_to_git() to stress this; the function
2289 * should never look at the index when explicit crlf option
2290 * is given.
2292 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2293 return 0;
2294 default:
2295 return -1;
2300 * Update the preimage, and the common lines in postimage,
2301 * from buffer buf of length len. If postlen is 0 the postimage
2302 * is updated in place, otherwise it's updated on a new buffer
2303 * of length postlen
2306 static void update_pre_post_images(struct image *preimage,
2307 struct image *postimage,
2308 char *buf,
2309 size_t len, size_t postlen)
2311 int i, ctx, reduced;
2312 char *new_buf, *old_buf, *fixed;
2313 struct image fixed_preimage;
2316 * Update the preimage with whitespace fixes. Note that we
2317 * are not losing preimage->buf -- apply_one_fragment() will
2318 * free "oldlines".
2320 prepare_image(&fixed_preimage, buf, len, 1);
2321 assert(postlen
2322 ? fixed_preimage.nr == preimage->nr
2323 : fixed_preimage.nr <= preimage->nr);
2324 for (i = 0; i < fixed_preimage.nr; i++)
2325 fixed_preimage.line[i].flag = preimage->line[i].flag;
2326 free(preimage->line_allocated);
2327 *preimage = fixed_preimage;
2330 * Adjust the common context lines in postimage. This can be
2331 * done in-place when we are shrinking it with whitespace
2332 * fixing, but needs a new buffer when ignoring whitespace or
2333 * expanding leading tabs to spaces.
2335 * We trust the caller to tell us if the update can be done
2336 * in place (postlen==0) or not.
2338 old_buf = postimage->buf;
2339 if (postlen)
2340 new_buf = postimage->buf = xmalloc(postlen);
2341 else
2342 new_buf = old_buf;
2343 fixed = preimage->buf;
2345 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2346 size_t l_len = postimage->line[i].len;
2347 if (!(postimage->line[i].flag & LINE_COMMON)) {
2348 /* an added line -- no counterparts in preimage */
2349 memmove(new_buf, old_buf, l_len);
2350 old_buf += l_len;
2351 new_buf += l_len;
2352 continue;
2355 /* a common context -- skip it in the original postimage */
2356 old_buf += l_len;
2358 /* and find the corresponding one in the fixed preimage */
2359 while (ctx < preimage->nr &&
2360 !(preimage->line[ctx].flag & LINE_COMMON)) {
2361 fixed += preimage->line[ctx].len;
2362 ctx++;
2366 * preimage is expected to run out, if the caller
2367 * fixed addition of trailing blank lines.
2369 if (preimage->nr <= ctx) {
2370 reduced++;
2371 continue;
2374 /* and copy it in, while fixing the line length */
2375 l_len = preimage->line[ctx].len;
2376 memcpy(new_buf, fixed, l_len);
2377 new_buf += l_len;
2378 fixed += l_len;
2379 postimage->line[i].len = l_len;
2380 ctx++;
2383 if (postlen
2384 ? postlen < new_buf - postimage->buf
2385 : postimage->len < new_buf - postimage->buf)
2386 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2387 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2389 /* Fix the length of the whole thing */
2390 postimage->len = new_buf - postimage->buf;
2391 postimage->nr -= reduced;
2394 static int line_by_line_fuzzy_match(struct image *img,
2395 struct image *preimage,
2396 struct image *postimage,
2397 unsigned long current,
2398 int current_lno,
2399 int preimage_limit)
2401 int i;
2402 size_t imgoff = 0;
2403 size_t preoff = 0;
2404 size_t postlen = postimage->len;
2405 size_t extra_chars;
2406 char *buf;
2407 char *preimage_eof;
2408 char *preimage_end;
2409 struct strbuf fixed;
2410 char *fixed_buf;
2411 size_t fixed_len;
2413 for (i = 0; i < preimage_limit; i++) {
2414 size_t prelen = preimage->line[i].len;
2415 size_t imglen = img->line[current_lno+i].len;
2417 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2418 preimage->buf + preoff, prelen))
2419 return 0;
2420 if (preimage->line[i].flag & LINE_COMMON)
2421 postlen += imglen - prelen;
2422 imgoff += imglen;
2423 preoff += prelen;
2427 * Ok, the preimage matches with whitespace fuzz.
2429 * imgoff now holds the true length of the target that
2430 * matches the preimage before the end of the file.
2432 * Count the number of characters in the preimage that fall
2433 * beyond the end of the file and make sure that all of them
2434 * are whitespace characters. (This can only happen if
2435 * we are removing blank lines at the end of the file.)
2437 buf = preimage_eof = preimage->buf + preoff;
2438 for ( ; i < preimage->nr; i++)
2439 preoff += preimage->line[i].len;
2440 preimage_end = preimage->buf + preoff;
2441 for ( ; buf < preimage_end; buf++)
2442 if (!isspace(*buf))
2443 return 0;
2446 * Update the preimage and the common postimage context
2447 * lines to use the same whitespace as the target.
2448 * If whitespace is missing in the target (i.e.
2449 * if the preimage extends beyond the end of the file),
2450 * use the whitespace from the preimage.
2452 extra_chars = preimage_end - preimage_eof;
2453 strbuf_init(&fixed, imgoff + extra_chars);
2454 strbuf_add(&fixed, img->buf + current, imgoff);
2455 strbuf_add(&fixed, preimage_eof, extra_chars);
2456 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2457 update_pre_post_images(preimage, postimage,
2458 fixed_buf, fixed_len, postlen);
2459 return 1;
2462 static int match_fragment(struct apply_state *state,
2463 struct image *img,
2464 struct image *preimage,
2465 struct image *postimage,
2466 unsigned long current,
2467 int current_lno,
2468 unsigned ws_rule,
2469 int match_beginning, int match_end)
2471 int i;
2472 char *fixed_buf, *buf, *orig, *target;
2473 struct strbuf fixed;
2474 size_t fixed_len, postlen;
2475 int preimage_limit;
2477 if (preimage->nr + current_lno <= img->nr) {
2479 * The hunk falls within the boundaries of img.
2481 preimage_limit = preimage->nr;
2482 if (match_end && (preimage->nr + current_lno != img->nr))
2483 return 0;
2484 } else if (state->ws_error_action == correct_ws_error &&
2485 (ws_rule & WS_BLANK_AT_EOF)) {
2487 * This hunk extends beyond the end of img, and we are
2488 * removing blank lines at the end of the file. This
2489 * many lines from the beginning of the preimage must
2490 * match with img, and the remainder of the preimage
2491 * must be blank.
2493 preimage_limit = img->nr - current_lno;
2494 } else {
2496 * The hunk extends beyond the end of the img and
2497 * we are not removing blanks at the end, so we
2498 * should reject the hunk at this position.
2500 return 0;
2503 if (match_beginning && current_lno)
2504 return 0;
2506 /* Quick hash check */
2507 for (i = 0; i < preimage_limit; i++)
2508 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2509 (preimage->line[i].hash != img->line[current_lno + i].hash))
2510 return 0;
2512 if (preimage_limit == preimage->nr) {
2514 * Do we have an exact match? If we were told to match
2515 * at the end, size must be exactly at current+fragsize,
2516 * otherwise current+fragsize must be still within the preimage,
2517 * and either case, the old piece should match the preimage
2518 * exactly.
2520 if ((match_end
2521 ? (current + preimage->len == img->len)
2522 : (current + preimage->len <= img->len)) &&
2523 !memcmp(img->buf + current, preimage->buf, preimage->len))
2524 return 1;
2525 } else {
2527 * The preimage extends beyond the end of img, so
2528 * there cannot be an exact match.
2530 * There must be one non-blank context line that match
2531 * a line before the end of img.
2533 char *buf_end;
2535 buf = preimage->buf;
2536 buf_end = buf;
2537 for (i = 0; i < preimage_limit; i++)
2538 buf_end += preimage->line[i].len;
2540 for ( ; buf < buf_end; buf++)
2541 if (!isspace(*buf))
2542 break;
2543 if (buf == buf_end)
2544 return 0;
2548 * No exact match. If we are ignoring whitespace, run a line-by-line
2549 * fuzzy matching. We collect all the line length information because
2550 * we need it to adjust whitespace if we match.
2552 if (state->ws_ignore_action == ignore_ws_change)
2553 return line_by_line_fuzzy_match(img, preimage, postimage,
2554 current, current_lno, preimage_limit);
2556 if (state->ws_error_action != correct_ws_error)
2557 return 0;
2560 * The hunk does not apply byte-by-byte, but the hash says
2561 * it might with whitespace fuzz. We weren't asked to
2562 * ignore whitespace, we were asked to correct whitespace
2563 * errors, so let's try matching after whitespace correction.
2565 * While checking the preimage against the target, whitespace
2566 * errors in both fixed, we count how large the corresponding
2567 * postimage needs to be. The postimage prepared by
2568 * apply_one_fragment() has whitespace errors fixed on added
2569 * lines already, but the common lines were propagated as-is,
2570 * which may become longer when their whitespace errors are
2571 * fixed.
2574 /* First count added lines in postimage */
2575 postlen = 0;
2576 for (i = 0; i < postimage->nr; i++) {
2577 if (!(postimage->line[i].flag & LINE_COMMON))
2578 postlen += postimage->line[i].len;
2582 * The preimage may extend beyond the end of the file,
2583 * but in this loop we will only handle the part of the
2584 * preimage that falls within the file.
2586 strbuf_init(&fixed, preimage->len + 1);
2587 orig = preimage->buf;
2588 target = img->buf + current;
2589 for (i = 0; i < preimage_limit; i++) {
2590 size_t oldlen = preimage->line[i].len;
2591 size_t tgtlen = img->line[current_lno + i].len;
2592 size_t fixstart = fixed.len;
2593 struct strbuf tgtfix;
2594 int match;
2596 /* Try fixing the line in the preimage */
2597 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2599 /* Try fixing the line in the target */
2600 strbuf_init(&tgtfix, tgtlen);
2601 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2604 * If they match, either the preimage was based on
2605 * a version before our tree fixed whitespace breakage,
2606 * or we are lacking a whitespace-fix patch the tree
2607 * the preimage was based on already had (i.e. target
2608 * has whitespace breakage, the preimage doesn't).
2609 * In either case, we are fixing the whitespace breakages
2610 * so we might as well take the fix together with their
2611 * real change.
2613 match = (tgtfix.len == fixed.len - fixstart &&
2614 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2615 fixed.len - fixstart));
2617 /* Add the length if this is common with the postimage */
2618 if (preimage->line[i].flag & LINE_COMMON)
2619 postlen += tgtfix.len;
2621 strbuf_release(&tgtfix);
2622 if (!match)
2623 goto unmatch_exit;
2625 orig += oldlen;
2626 target += tgtlen;
2631 * Now handle the lines in the preimage that falls beyond the
2632 * end of the file (if any). They will only match if they are
2633 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2634 * false).
2636 for ( ; i < preimage->nr; i++) {
2637 size_t fixstart = fixed.len; /* start of the fixed preimage */
2638 size_t oldlen = preimage->line[i].len;
2639 int j;
2641 /* Try fixing the line in the preimage */
2642 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2644 for (j = fixstart; j < fixed.len; j++)
2645 if (!isspace(fixed.buf[j]))
2646 goto unmatch_exit;
2648 orig += oldlen;
2652 * Yes, the preimage is based on an older version that still
2653 * has whitespace breakages unfixed, and fixing them makes the
2654 * hunk match. Update the context lines in the postimage.
2656 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2657 if (postlen < postimage->len)
2658 postlen = 0;
2659 update_pre_post_images(preimage, postimage,
2660 fixed_buf, fixed_len, postlen);
2661 return 1;
2663 unmatch_exit:
2664 strbuf_release(&fixed);
2665 return 0;
2668 static int find_pos(struct apply_state *state,
2669 struct image *img,
2670 struct image *preimage,
2671 struct image *postimage,
2672 int line,
2673 unsigned ws_rule,
2674 int match_beginning, int match_end)
2676 int i;
2677 unsigned long backwards, forwards, current;
2678 int backwards_lno, forwards_lno, current_lno;
2681 * If match_beginning or match_end is specified, there is no
2682 * point starting from a wrong line that will never match and
2683 * wander around and wait for a match at the specified end.
2685 if (match_beginning)
2686 line = 0;
2687 else if (match_end)
2688 line = img->nr - preimage->nr;
2691 * Because the comparison is unsigned, the following test
2692 * will also take care of a negative line number that can
2693 * result when match_end and preimage is larger than the target.
2695 if ((size_t) line > img->nr)
2696 line = img->nr;
2698 current = 0;
2699 for (i = 0; i < line; i++)
2700 current += img->line[i].len;
2703 * There's probably some smart way to do this, but I'll leave
2704 * that to the smart and beautiful people. I'm simple and stupid.
2706 backwards = current;
2707 backwards_lno = line;
2708 forwards = current;
2709 forwards_lno = line;
2710 current_lno = line;
2712 for (i = 0; ; i++) {
2713 if (match_fragment(state, img, preimage, postimage,
2714 current, current_lno, ws_rule,
2715 match_beginning, match_end))
2716 return current_lno;
2718 again:
2719 if (backwards_lno == 0 && forwards_lno == img->nr)
2720 break;
2722 if (i & 1) {
2723 if (backwards_lno == 0) {
2724 i++;
2725 goto again;
2727 backwards_lno--;
2728 backwards -= img->line[backwards_lno].len;
2729 current = backwards;
2730 current_lno = backwards_lno;
2731 } else {
2732 if (forwards_lno == img->nr) {
2733 i++;
2734 goto again;
2736 forwards += img->line[forwards_lno].len;
2737 forwards_lno++;
2738 current = forwards;
2739 current_lno = forwards_lno;
2743 return -1;
2746 static void remove_first_line(struct image *img)
2748 img->buf += img->line[0].len;
2749 img->len -= img->line[0].len;
2750 img->line++;
2751 img->nr--;
2754 static void remove_last_line(struct image *img)
2756 img->len -= img->line[--img->nr].len;
2760 * The change from "preimage" and "postimage" has been found to
2761 * apply at applied_pos (counts in line numbers) in "img".
2762 * Update "img" to remove "preimage" and replace it with "postimage".
2764 static void update_image(struct apply_state *state,
2765 struct image *img,
2766 int applied_pos,
2767 struct image *preimage,
2768 struct image *postimage)
2771 * remove the copy of preimage at offset in img
2772 * and replace it with postimage
2774 int i, nr;
2775 size_t remove_count, insert_count, applied_at = 0;
2776 char *result;
2777 int preimage_limit;
2780 * If we are removing blank lines at the end of img,
2781 * the preimage may extend beyond the end.
2782 * If that is the case, we must be careful only to
2783 * remove the part of the preimage that falls within
2784 * the boundaries of img. Initialize preimage_limit
2785 * to the number of lines in the preimage that falls
2786 * within the boundaries.
2788 preimage_limit = preimage->nr;
2789 if (preimage_limit > img->nr - applied_pos)
2790 preimage_limit = img->nr - applied_pos;
2792 for (i = 0; i < applied_pos; i++)
2793 applied_at += img->line[i].len;
2795 remove_count = 0;
2796 for (i = 0; i < preimage_limit; i++)
2797 remove_count += img->line[applied_pos + i].len;
2798 insert_count = postimage->len;
2800 /* Adjust the contents */
2801 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2802 memcpy(result, img->buf, applied_at);
2803 memcpy(result + applied_at, postimage->buf, postimage->len);
2804 memcpy(result + applied_at + postimage->len,
2805 img->buf + (applied_at + remove_count),
2806 img->len - (applied_at + remove_count));
2807 free(img->buf);
2808 img->buf = result;
2809 img->len += insert_count - remove_count;
2810 result[img->len] = '\0';
2812 /* Adjust the line table */
2813 nr = img->nr + postimage->nr - preimage_limit;
2814 if (preimage_limit < postimage->nr) {
2816 * NOTE: this knows that we never call remove_first_line()
2817 * on anything other than pre/post image.
2819 REALLOC_ARRAY(img->line, nr);
2820 img->line_allocated = img->line;
2822 if (preimage_limit != postimage->nr)
2823 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2824 img->line + applied_pos + preimage_limit,
2825 img->nr - (applied_pos + preimage_limit));
2826 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2827 if (!state->allow_overlap)
2828 for (i = 0; i < postimage->nr; i++)
2829 img->line[applied_pos + i].flag |= LINE_PATCHED;
2830 img->nr = nr;
2834 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2835 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2836 * replace the part of "img" with "postimage" text.
2838 static int apply_one_fragment(struct apply_state *state,
2839 struct image *img, struct fragment *frag,
2840 int inaccurate_eof, unsigned ws_rule,
2841 int nth_fragment)
2843 int match_beginning, match_end;
2844 const char *patch = frag->patch;
2845 int size = frag->size;
2846 char *old, *oldlines;
2847 struct strbuf newlines;
2848 int new_blank_lines_at_end = 0;
2849 int found_new_blank_lines_at_end = 0;
2850 int hunk_linenr = frag->linenr;
2851 unsigned long leading, trailing;
2852 int pos, applied_pos;
2853 struct image preimage;
2854 struct image postimage;
2856 memset(&preimage, 0, sizeof(preimage));
2857 memset(&postimage, 0, sizeof(postimage));
2858 oldlines = xmalloc(size);
2859 strbuf_init(&newlines, size);
2861 old = oldlines;
2862 while (size > 0) {
2863 char first;
2864 int len = linelen(patch, size);
2865 int plen;
2866 int added_blank_line = 0;
2867 int is_blank_context = 0;
2868 size_t start;
2870 if (!len)
2871 break;
2874 * "plen" is how much of the line we should use for
2875 * the actual patch data. Normally we just remove the
2876 * first character on the line, but if the line is
2877 * followed by "\ No newline", then we also remove the
2878 * last one (which is the newline, of course).
2880 plen = len - 1;
2881 if (len < size && patch[len] == '\\')
2882 plen--;
2883 first = *patch;
2884 if (state->apply_in_reverse) {
2885 if (first == '-')
2886 first = '+';
2887 else if (first == '+')
2888 first = '-';
2891 switch (first) {
2892 case '\n':
2893 /* Newer GNU diff, empty context line */
2894 if (plen < 0)
2895 /* ... followed by '\No newline'; nothing */
2896 break;
2897 *old++ = '\n';
2898 strbuf_addch(&newlines, '\n');
2899 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2900 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2901 is_blank_context = 1;
2902 break;
2903 case ' ':
2904 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2905 ws_blank_line(patch + 1, plen, ws_rule))
2906 is_blank_context = 1;
2907 /* fallthrough */
2908 case '-':
2909 memcpy(old, patch + 1, plen);
2910 add_line_info(&preimage, old, plen,
2911 (first == ' ' ? LINE_COMMON : 0));
2912 old += plen;
2913 if (first == '-')
2914 break;
2915 /* fallthrough */
2916 case '+':
2917 /* --no-add does not add new lines */
2918 if (first == '+' && state->no_add)
2919 break;
2921 start = newlines.len;
2922 if (first != '+' ||
2923 !state->whitespace_error ||
2924 state->ws_error_action != correct_ws_error) {
2925 strbuf_add(&newlines, patch + 1, plen);
2927 else {
2928 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2930 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2931 (first == '+' ? 0 : LINE_COMMON));
2932 if (first == '+' &&
2933 (ws_rule & WS_BLANK_AT_EOF) &&
2934 ws_blank_line(patch + 1, plen, ws_rule))
2935 added_blank_line = 1;
2936 break;
2937 case '@': case '\\':
2938 /* Ignore it, we already handled it */
2939 break;
2940 default:
2941 if (state->apply_verbosity > verbosity_normal)
2942 error(_("invalid start of line: '%c'"), first);
2943 applied_pos = -1;
2944 goto out;
2946 if (added_blank_line) {
2947 if (!new_blank_lines_at_end)
2948 found_new_blank_lines_at_end = hunk_linenr;
2949 new_blank_lines_at_end++;
2951 else if (is_blank_context)
2953 else
2954 new_blank_lines_at_end = 0;
2955 patch += len;
2956 size -= len;
2957 hunk_linenr++;
2959 if (inaccurate_eof &&
2960 old > oldlines && old[-1] == '\n' &&
2961 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2962 old--;
2963 strbuf_setlen(&newlines, newlines.len - 1);
2964 preimage.line_allocated[preimage.nr - 1].len--;
2965 postimage.line_allocated[postimage.nr - 1].len--;
2968 leading = frag->leading;
2969 trailing = frag->trailing;
2972 * A hunk to change lines at the beginning would begin with
2973 * @@ -1,L +N,M @@
2974 * but we need to be careful. -U0 that inserts before the second
2975 * line also has this pattern.
2977 * And a hunk to add to an empty file would begin with
2978 * @@ -0,0 +N,M @@
2980 * In other words, a hunk that is (frag->oldpos <= 1) with or
2981 * without leading context must match at the beginning.
2983 match_beginning = (!frag->oldpos ||
2984 (frag->oldpos == 1 && !state->unidiff_zero));
2987 * A hunk without trailing lines must match at the end.
2988 * However, we simply cannot tell if a hunk must match end
2989 * from the lack of trailing lines if the patch was generated
2990 * with unidiff without any context.
2992 match_end = !state->unidiff_zero && !trailing;
2994 pos = frag->newpos ? (frag->newpos - 1) : 0;
2995 preimage.buf = oldlines;
2996 preimage.len = old - oldlines;
2997 postimage.buf = newlines.buf;
2998 postimage.len = newlines.len;
2999 preimage.line = preimage.line_allocated;
3000 postimage.line = postimage.line_allocated;
3002 for (;;) {
3004 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3005 ws_rule, match_beginning, match_end);
3007 if (applied_pos >= 0)
3008 break;
3010 /* Am I at my context limits? */
3011 if ((leading <= state->p_context) && (trailing <= state->p_context))
3012 break;
3013 if (match_beginning || match_end) {
3014 match_beginning = match_end = 0;
3015 continue;
3019 * Reduce the number of context lines; reduce both
3020 * leading and trailing if they are equal otherwise
3021 * just reduce the larger context.
3023 if (leading >= trailing) {
3024 remove_first_line(&preimage);
3025 remove_first_line(&postimage);
3026 pos--;
3027 leading--;
3029 if (trailing > leading) {
3030 remove_last_line(&preimage);
3031 remove_last_line(&postimage);
3032 trailing--;
3036 if (applied_pos >= 0) {
3037 if (new_blank_lines_at_end &&
3038 preimage.nr + applied_pos >= img->nr &&
3039 (ws_rule & WS_BLANK_AT_EOF) &&
3040 state->ws_error_action != nowarn_ws_error) {
3041 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3042 found_new_blank_lines_at_end);
3043 if (state->ws_error_action == correct_ws_error) {
3044 while (new_blank_lines_at_end--)
3045 remove_last_line(&postimage);
3048 * We would want to prevent write_out_results()
3049 * from taking place in apply_patch() that follows
3050 * the callchain led us here, which is:
3051 * apply_patch->check_patch_list->check_patch->
3052 * apply_data->apply_fragments->apply_one_fragment
3054 if (state->ws_error_action == die_on_ws_error)
3055 state->apply = 0;
3058 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3059 int offset = applied_pos - pos;
3060 if (state->apply_in_reverse)
3061 offset = 0 - offset;
3062 fprintf_ln(stderr,
3063 Q_("Hunk #%d succeeded at %d (offset %d line).",
3064 "Hunk #%d succeeded at %d (offset %d lines).",
3065 offset),
3066 nth_fragment, applied_pos + 1, offset);
3070 * Warn if it was necessary to reduce the number
3071 * of context lines.
3073 if ((leading != frag->leading ||
3074 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3075 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3076 " to apply fragment at %d"),
3077 leading, trailing, applied_pos+1);
3078 update_image(state, img, applied_pos, &preimage, &postimage);
3079 } else {
3080 if (state->apply_verbosity > verbosity_normal)
3081 error(_("while searching for:\n%.*s"),
3082 (int)(old - oldlines), oldlines);
3085 out:
3086 free(oldlines);
3087 strbuf_release(&newlines);
3088 free(preimage.line_allocated);
3089 free(postimage.line_allocated);
3091 return (applied_pos < 0);
3094 static int apply_binary_fragment(struct apply_state *state,
3095 struct image *img,
3096 struct patch *patch)
3098 struct fragment *fragment = patch->fragments;
3099 unsigned long len;
3100 void *dst;
3102 if (!fragment)
3103 return error(_("missing binary patch data for '%s'"),
3104 patch->new_name ?
3105 patch->new_name :
3106 patch->old_name);
3108 /* Binary patch is irreversible without the optional second hunk */
3109 if (state->apply_in_reverse) {
3110 if (!fragment->next)
3111 return error(_("cannot reverse-apply a binary patch "
3112 "without the reverse hunk to '%s'"),
3113 patch->new_name
3114 ? patch->new_name : patch->old_name);
3115 fragment = fragment->next;
3117 switch (fragment->binary_patch_method) {
3118 case BINARY_DELTA_DEFLATED:
3119 dst = patch_delta(img->buf, img->len, fragment->patch,
3120 fragment->size, &len);
3121 if (!dst)
3122 return -1;
3123 clear_image(img);
3124 img->buf = dst;
3125 img->len = len;
3126 return 0;
3127 case BINARY_LITERAL_DEFLATED:
3128 clear_image(img);
3129 img->len = fragment->size;
3130 img->buf = xmemdupz(fragment->patch, img->len);
3131 return 0;
3133 return -1;
3137 * Replace "img" with the result of applying the binary patch.
3138 * The binary patch data itself in patch->fragment is still kept
3139 * but the preimage prepared by the caller in "img" is freed here
3140 * or in the helper function apply_binary_fragment() this calls.
3142 static int apply_binary(struct apply_state *state,
3143 struct image *img,
3144 struct patch *patch)
3146 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3147 struct object_id oid;
3148 const unsigned hexsz = the_hash_algo->hexsz;
3151 * For safety, we require patch index line to contain
3152 * full hex textual object ID for old and new, at least for now.
3154 if (strlen(patch->old_oid_prefix) != hexsz ||
3155 strlen(patch->new_oid_prefix) != hexsz ||
3156 get_oid_hex(patch->old_oid_prefix, &oid) ||
3157 get_oid_hex(patch->new_oid_prefix, &oid))
3158 return error(_("cannot apply binary patch to '%s' "
3159 "without full index line"), name);
3161 if (patch->old_name) {
3163 * See if the old one matches what the patch
3164 * applies to.
3166 hash_object_file(img->buf, img->len, blob_type, &oid);
3167 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3168 return error(_("the patch applies to '%s' (%s), "
3169 "which does not match the "
3170 "current contents."),
3171 name, oid_to_hex(&oid));
3173 else {
3174 /* Otherwise, the old one must be empty. */
3175 if (img->len)
3176 return error(_("the patch applies to an empty "
3177 "'%s' but it is not empty"), name);
3180 get_oid_hex(patch->new_oid_prefix, &oid);
3181 if (is_null_oid(&oid)) {
3182 clear_image(img);
3183 return 0; /* deletion patch */
3186 if (has_sha1_file(oid.hash)) {
3187 /* We already have the postimage */
3188 enum object_type type;
3189 unsigned long size;
3190 char *result;
3192 result = read_object_file(&oid, &type, &size);
3193 if (!result)
3194 return error(_("the necessary postimage %s for "
3195 "'%s' cannot be read"),
3196 patch->new_oid_prefix, name);
3197 clear_image(img);
3198 img->buf = result;
3199 img->len = size;
3200 } else {
3202 * We have verified buf matches the preimage;
3203 * apply the patch data to it, which is stored
3204 * in the patch->fragments->{patch,size}.
3206 if (apply_binary_fragment(state, img, patch))
3207 return error(_("binary patch does not apply to '%s'"),
3208 name);
3210 /* verify that the result matches */
3211 hash_object_file(img->buf, img->len, blob_type, &oid);
3212 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3213 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3214 name, patch->new_oid_prefix, oid_to_hex(&oid));
3217 return 0;
3220 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3222 struct fragment *frag = patch->fragments;
3223 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3224 unsigned ws_rule = patch->ws_rule;
3225 unsigned inaccurate_eof = patch->inaccurate_eof;
3226 int nth = 0;
3228 if (patch->is_binary)
3229 return apply_binary(state, img, patch);
3231 while (frag) {
3232 nth++;
3233 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3234 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3235 if (!state->apply_with_reject)
3236 return -1;
3237 frag->rejected = 1;
3239 frag = frag->next;
3241 return 0;
3244 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3246 if (S_ISGITLINK(mode)) {
3247 strbuf_grow(buf, 100);
3248 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3249 } else {
3250 enum object_type type;
3251 unsigned long sz;
3252 char *result;
3254 result = read_object_file(oid, &type, &sz);
3255 if (!result)
3256 return -1;
3257 /* XXX read_sha1_file NUL-terminates */
3258 strbuf_attach(buf, result, sz, sz + 1);
3260 return 0;
3263 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3265 if (!ce)
3266 return 0;
3267 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3270 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3272 struct string_list_item *item;
3274 if (name == NULL)
3275 return NULL;
3277 item = string_list_lookup(&state->fn_table, name);
3278 if (item != NULL)
3279 return (struct patch *)item->util;
3281 return NULL;
3285 * item->util in the filename table records the status of the path.
3286 * Usually it points at a patch (whose result records the contents
3287 * of it after applying it), but it could be PATH_WAS_DELETED for a
3288 * path that a previously applied patch has already removed, or
3289 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3291 * The latter is needed to deal with a case where two paths A and B
3292 * are swapped by first renaming A to B and then renaming B to A;
3293 * moving A to B should not be prevented due to presence of B as we
3294 * will remove it in a later patch.
3296 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3297 #define PATH_WAS_DELETED ((struct patch *) -1)
3299 static int to_be_deleted(struct patch *patch)
3301 return patch == PATH_TO_BE_DELETED;
3304 static int was_deleted(struct patch *patch)
3306 return patch == PATH_WAS_DELETED;
3309 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3311 struct string_list_item *item;
3314 * Always add new_name unless patch is a deletion
3315 * This should cover the cases for normal diffs,
3316 * file creations and copies
3318 if (patch->new_name != NULL) {
3319 item = string_list_insert(&state->fn_table, patch->new_name);
3320 item->util = patch;
3324 * store a failure on rename/deletion cases because
3325 * later chunks shouldn't patch old names
3327 if ((patch->new_name == NULL) || (patch->is_rename)) {
3328 item = string_list_insert(&state->fn_table, patch->old_name);
3329 item->util = PATH_WAS_DELETED;
3333 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3336 * store information about incoming file deletion
3338 while (patch) {
3339 if ((patch->new_name == NULL) || (patch->is_rename)) {
3340 struct string_list_item *item;
3341 item = string_list_insert(&state->fn_table, patch->old_name);
3342 item->util = PATH_TO_BE_DELETED;
3344 patch = patch->next;
3348 static int checkout_target(struct index_state *istate,
3349 struct cache_entry *ce, struct stat *st)
3351 struct checkout costate = CHECKOUT_INIT;
3353 costate.refresh_cache = 1;
3354 costate.istate = istate;
3355 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3356 return error(_("cannot checkout %s"), ce->name);
3357 return 0;
3360 static struct patch *previous_patch(struct apply_state *state,
3361 struct patch *patch,
3362 int *gone)
3364 struct patch *previous;
3366 *gone = 0;
3367 if (patch->is_copy || patch->is_rename)
3368 return NULL; /* "git" patches do not depend on the order */
3370 previous = in_fn_table(state, patch->old_name);
3371 if (!previous)
3372 return NULL;
3374 if (to_be_deleted(previous))
3375 return NULL; /* the deletion hasn't happened yet */
3377 if (was_deleted(previous))
3378 *gone = 1;
3380 return previous;
3383 static int verify_index_match(struct apply_state *state,
3384 const struct cache_entry *ce,
3385 struct stat *st)
3387 if (S_ISGITLINK(ce->ce_mode)) {
3388 if (!S_ISDIR(st->st_mode))
3389 return -1;
3390 return 0;
3392 return ie_match_stat(state->repo->index, ce, st,
3393 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3396 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3398 static int load_patch_target(struct apply_state *state,
3399 struct strbuf *buf,
3400 const struct cache_entry *ce,
3401 struct stat *st,
3402 struct patch *patch,
3403 const char *name,
3404 unsigned expected_mode)
3406 if (state->cached || state->check_index) {
3407 if (read_file_or_gitlink(ce, buf))
3408 return error(_("failed to read %s"), name);
3409 } else if (name) {
3410 if (S_ISGITLINK(expected_mode)) {
3411 if (ce)
3412 return read_file_or_gitlink(ce, buf);
3413 else
3414 return SUBMODULE_PATCH_WITHOUT_INDEX;
3415 } else if (has_symlink_leading_path(name, strlen(name))) {
3416 return error(_("reading from '%s' beyond a symbolic link"), name);
3417 } else {
3418 if (read_old_data(st, patch, name, buf))
3419 return error(_("failed to read %s"), name);
3422 return 0;
3426 * We are about to apply "patch"; populate the "image" with the
3427 * current version we have, from the working tree or from the index,
3428 * depending on the situation e.g. --cached/--index. If we are
3429 * applying a non-git patch that incrementally updates the tree,
3430 * we read from the result of a previous diff.
3432 static int load_preimage(struct apply_state *state,
3433 struct image *image,
3434 struct patch *patch, struct stat *st,
3435 const struct cache_entry *ce)
3437 struct strbuf buf = STRBUF_INIT;
3438 size_t len;
3439 char *img;
3440 struct patch *previous;
3441 int status;
3443 previous = previous_patch(state, patch, &status);
3444 if (status)
3445 return error(_("path %s has been renamed/deleted"),
3446 patch->old_name);
3447 if (previous) {
3448 /* We have a patched copy in memory; use that. */
3449 strbuf_add(&buf, previous->result, previous->resultsize);
3450 } else {
3451 status = load_patch_target(state, &buf, ce, st, patch,
3452 patch->old_name, patch->old_mode);
3453 if (status < 0)
3454 return status;
3455 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3457 * There is no way to apply subproject
3458 * patch without looking at the index.
3459 * NEEDSWORK: shouldn't this be flagged
3460 * as an error???
3462 free_fragment_list(patch->fragments);
3463 patch->fragments = NULL;
3464 } else if (status) {
3465 return error(_("failed to read %s"), patch->old_name);
3469 img = strbuf_detach(&buf, &len);
3470 prepare_image(image, img, len, !patch->is_binary);
3471 return 0;
3474 static int three_way_merge(struct apply_state *state,
3475 struct image *image,
3476 char *path,
3477 const struct object_id *base,
3478 const struct object_id *ours,
3479 const struct object_id *theirs)
3481 mmfile_t base_file, our_file, their_file;
3482 mmbuffer_t result = { NULL };
3483 int status;
3485 read_mmblob(&base_file, base);
3486 read_mmblob(&our_file, ours);
3487 read_mmblob(&their_file, theirs);
3488 status = ll_merge(&result, path,
3489 &base_file, "base",
3490 &our_file, "ours",
3491 &their_file, "theirs",
3492 state->repo->index,
3493 NULL);
3494 free(base_file.ptr);
3495 free(our_file.ptr);
3496 free(their_file.ptr);
3497 if (status < 0 || !result.ptr) {
3498 free(result.ptr);
3499 return -1;
3501 clear_image(image);
3502 image->buf = result.ptr;
3503 image->len = result.size;
3505 return status;
3509 * When directly falling back to add/add three-way merge, we read from
3510 * the current contents of the new_name. In no cases other than that
3511 * this function will be called.
3513 static int load_current(struct apply_state *state,
3514 struct image *image,
3515 struct patch *patch)
3517 struct strbuf buf = STRBUF_INIT;
3518 int status, pos;
3519 size_t len;
3520 char *img;
3521 struct stat st;
3522 struct cache_entry *ce;
3523 char *name = patch->new_name;
3524 unsigned mode = patch->new_mode;
3526 if (!patch->is_new)
3527 BUG("patch to %s is not a creation", patch->old_name);
3529 pos = index_name_pos(state->repo->index, name, strlen(name));
3530 if (pos < 0)
3531 return error(_("%s: does not exist in index"), name);
3532 ce = state->repo->index->cache[pos];
3533 if (lstat(name, &st)) {
3534 if (errno != ENOENT)
3535 return error_errno("%s", name);
3536 if (checkout_target(state->repo->index, ce, &st))
3537 return -1;
3539 if (verify_index_match(state, ce, &st))
3540 return error(_("%s: does not match index"), name);
3542 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3543 if (status < 0)
3544 return status;
3545 else if (status)
3546 return -1;
3547 img = strbuf_detach(&buf, &len);
3548 prepare_image(image, img, len, !patch->is_binary);
3549 return 0;
3552 static int try_threeway(struct apply_state *state,
3553 struct image *image,
3554 struct patch *patch,
3555 struct stat *st,
3556 const struct cache_entry *ce)
3558 struct object_id pre_oid, post_oid, our_oid;
3559 struct strbuf buf = STRBUF_INIT;
3560 size_t len;
3561 int status;
3562 char *img;
3563 struct image tmp_image;
3565 /* No point falling back to 3-way merge in these cases */
3566 if (patch->is_delete ||
3567 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3568 return -1;
3570 /* Preimage the patch was prepared for */
3571 if (patch->is_new)
3572 write_object_file("", 0, blob_type, &pre_oid);
3573 else if (get_oid(patch->old_oid_prefix, &pre_oid) ||
3574 read_blob_object(&buf, &pre_oid, patch->old_mode))
3575 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3577 if (state->apply_verbosity > verbosity_silent)
3578 fprintf(stderr, _("Falling back to three-way merge...\n"));
3580 img = strbuf_detach(&buf, &len);
3581 prepare_image(&tmp_image, img, len, 1);
3582 /* Apply the patch to get the post image */
3583 if (apply_fragments(state, &tmp_image, patch) < 0) {
3584 clear_image(&tmp_image);
3585 return -1;
3587 /* post_oid is theirs */
3588 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &post_oid);
3589 clear_image(&tmp_image);
3591 /* our_oid is ours */
3592 if (patch->is_new) {
3593 if (load_current(state, &tmp_image, patch))
3594 return error(_("cannot read the current contents of '%s'"),
3595 patch->new_name);
3596 } else {
3597 if (load_preimage(state, &tmp_image, patch, st, ce))
3598 return error(_("cannot read the current contents of '%s'"),
3599 patch->old_name);
3601 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &our_oid);
3602 clear_image(&tmp_image);
3604 /* in-core three-way merge between post and our using pre as base */
3605 status = three_way_merge(state, image, patch->new_name,
3606 &pre_oid, &our_oid, &post_oid);
3607 if (status < 0) {
3608 if (state->apply_verbosity > verbosity_silent)
3609 fprintf(stderr,
3610 _("Failed to fall back on three-way merge...\n"));
3611 return status;
3614 if (status) {
3615 patch->conflicted_threeway = 1;
3616 if (patch->is_new)
3617 oidclr(&patch->threeway_stage[0]);
3618 else
3619 oidcpy(&patch->threeway_stage[0], &pre_oid);
3620 oidcpy(&patch->threeway_stage[1], &our_oid);
3621 oidcpy(&patch->threeway_stage[2], &post_oid);
3622 if (state->apply_verbosity > verbosity_silent)
3623 fprintf(stderr,
3624 _("Applied patch to '%s' with conflicts.\n"),
3625 patch->new_name);
3626 } else {
3627 if (state->apply_verbosity > verbosity_silent)
3628 fprintf(stderr,
3629 _("Applied patch to '%s' cleanly.\n"),
3630 patch->new_name);
3632 return 0;
3635 static int apply_data(struct apply_state *state, struct patch *patch,
3636 struct stat *st, const struct cache_entry *ce)
3638 struct image image;
3640 if (load_preimage(state, &image, patch, st, ce) < 0)
3641 return -1;
3643 if (patch->direct_to_threeway ||
3644 apply_fragments(state, &image, patch) < 0) {
3645 /* Note: with --reject, apply_fragments() returns 0 */
3646 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3647 return -1;
3649 patch->result = image.buf;
3650 patch->resultsize = image.len;
3651 add_to_fn_table(state, patch);
3652 free(image.line_allocated);
3654 if (0 < patch->is_delete && patch->resultsize)
3655 return error(_("removal patch leaves file contents"));
3657 return 0;
3661 * If "patch" that we are looking at modifies or deletes what we have,
3662 * we would want it not to lose any local modification we have, either
3663 * in the working tree or in the index.
3665 * This also decides if a non-git patch is a creation patch or a
3666 * modification to an existing empty file. We do not check the state
3667 * of the current tree for a creation patch in this function; the caller
3668 * check_patch() separately makes sure (and errors out otherwise) that
3669 * the path the patch creates does not exist in the current tree.
3671 static int check_preimage(struct apply_state *state,
3672 struct patch *patch,
3673 struct cache_entry **ce,
3674 struct stat *st)
3676 const char *old_name = patch->old_name;
3677 struct patch *previous = NULL;
3678 int stat_ret = 0, status;
3679 unsigned st_mode = 0;
3681 if (!old_name)
3682 return 0;
3684 assert(patch->is_new <= 0);
3685 previous = previous_patch(state, patch, &status);
3687 if (status)
3688 return error(_("path %s has been renamed/deleted"), old_name);
3689 if (previous) {
3690 st_mode = previous->new_mode;
3691 } else if (!state->cached) {
3692 stat_ret = lstat(old_name, st);
3693 if (stat_ret && errno != ENOENT)
3694 return error_errno("%s", old_name);
3697 if (state->check_index && !previous) {
3698 int pos = index_name_pos(state->repo->index, old_name,
3699 strlen(old_name));
3700 if (pos < 0) {
3701 if (patch->is_new < 0)
3702 goto is_new;
3703 return error(_("%s: does not exist in index"), old_name);
3705 *ce = state->repo->index->cache[pos];
3706 if (stat_ret < 0) {
3707 if (checkout_target(state->repo->index, *ce, st))
3708 return -1;
3710 if (!state->cached && verify_index_match(state, *ce, st))
3711 return error(_("%s: does not match index"), old_name);
3712 if (state->cached)
3713 st_mode = (*ce)->ce_mode;
3714 } else if (stat_ret < 0) {
3715 if (patch->is_new < 0)
3716 goto is_new;
3717 return error_errno("%s", old_name);
3720 if (!state->cached && !previous)
3721 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3723 if (patch->is_new < 0)
3724 patch->is_new = 0;
3725 if (!patch->old_mode)
3726 patch->old_mode = st_mode;
3727 if ((st_mode ^ patch->old_mode) & S_IFMT)
3728 return error(_("%s: wrong type"), old_name);
3729 if (st_mode != patch->old_mode)
3730 warning(_("%s has type %o, expected %o"),
3731 old_name, st_mode, patch->old_mode);
3732 if (!patch->new_mode && !patch->is_delete)
3733 patch->new_mode = st_mode;
3734 return 0;
3736 is_new:
3737 patch->is_new = 1;
3738 patch->is_delete = 0;
3739 FREE_AND_NULL(patch->old_name);
3740 return 0;
3744 #define EXISTS_IN_INDEX 1
3745 #define EXISTS_IN_WORKTREE 2
3747 static int check_to_create(struct apply_state *state,
3748 const char *new_name,
3749 int ok_if_exists)
3751 struct stat nst;
3753 if (state->check_index &&
3754 index_name_pos(state->repo->index, new_name, strlen(new_name)) >= 0 &&
3755 !ok_if_exists)
3756 return EXISTS_IN_INDEX;
3757 if (state->cached)
3758 return 0;
3760 if (!lstat(new_name, &nst)) {
3761 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3762 return 0;
3764 * A leading component of new_name might be a symlink
3765 * that is going to be removed with this patch, but
3766 * still pointing at somewhere that has the path.
3767 * In such a case, path "new_name" does not exist as
3768 * far as git is concerned.
3770 if (has_symlink_leading_path(new_name, strlen(new_name)))
3771 return 0;
3773 return EXISTS_IN_WORKTREE;
3774 } else if (!is_missing_file_error(errno)) {
3775 return error_errno("%s", new_name);
3777 return 0;
3780 static uintptr_t register_symlink_changes(struct apply_state *state,
3781 const char *path,
3782 uintptr_t what)
3784 struct string_list_item *ent;
3786 ent = string_list_lookup(&state->symlink_changes, path);
3787 if (!ent) {
3788 ent = string_list_insert(&state->symlink_changes, path);
3789 ent->util = (void *)0;
3791 ent->util = (void *)(what | ((uintptr_t)ent->util));
3792 return (uintptr_t)ent->util;
3795 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3797 struct string_list_item *ent;
3799 ent = string_list_lookup(&state->symlink_changes, path);
3800 if (!ent)
3801 return 0;
3802 return (uintptr_t)ent->util;
3805 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3807 for ( ; patch; patch = patch->next) {
3808 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3809 (patch->is_rename || patch->is_delete))
3810 /* the symlink at patch->old_name is removed */
3811 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3813 if (patch->new_name && S_ISLNK(patch->new_mode))
3814 /* the symlink at patch->new_name is created or remains */
3815 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3819 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3821 do {
3822 unsigned int change;
3824 while (--name->len && name->buf[name->len] != '/')
3825 ; /* scan backwards */
3826 if (!name->len)
3827 break;
3828 name->buf[name->len] = '\0';
3829 change = check_symlink_changes(state, name->buf);
3830 if (change & APPLY_SYMLINK_IN_RESULT)
3831 return 1;
3832 if (change & APPLY_SYMLINK_GOES_AWAY)
3834 * This cannot be "return 0", because we may
3835 * see a new one created at a higher level.
3837 continue;
3839 /* otherwise, check the preimage */
3840 if (state->check_index) {
3841 struct cache_entry *ce;
3843 ce = index_file_exists(state->repo->index, name->buf,
3844 name->len, ignore_case);
3845 if (ce && S_ISLNK(ce->ce_mode))
3846 return 1;
3847 } else {
3848 struct stat st;
3849 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3850 return 1;
3852 } while (1);
3853 return 0;
3856 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3858 int ret;
3859 struct strbuf name = STRBUF_INIT;
3861 assert(*name_ != '\0');
3862 strbuf_addstr(&name, name_);
3863 ret = path_is_beyond_symlink_1(state, &name);
3864 strbuf_release(&name);
3866 return ret;
3869 static int check_unsafe_path(struct patch *patch)
3871 const char *old_name = NULL;
3872 const char *new_name = NULL;
3873 if (patch->is_delete)
3874 old_name = patch->old_name;
3875 else if (!patch->is_new && !patch->is_copy)
3876 old_name = patch->old_name;
3877 if (!patch->is_delete)
3878 new_name = patch->new_name;
3880 if (old_name && !verify_path(old_name, patch->old_mode))
3881 return error(_("invalid path '%s'"), old_name);
3882 if (new_name && !verify_path(new_name, patch->new_mode))
3883 return error(_("invalid path '%s'"), new_name);
3884 return 0;
3888 * Check and apply the patch in-core; leave the result in patch->result
3889 * for the caller to write it out to the final destination.
3891 static int check_patch(struct apply_state *state, struct patch *patch)
3893 struct stat st;
3894 const char *old_name = patch->old_name;
3895 const char *new_name = patch->new_name;
3896 const char *name = old_name ? old_name : new_name;
3897 struct cache_entry *ce = NULL;
3898 struct patch *tpatch;
3899 int ok_if_exists;
3900 int status;
3902 patch->rejected = 1; /* we will drop this after we succeed */
3904 status = check_preimage(state, patch, &ce, &st);
3905 if (status)
3906 return status;
3907 old_name = patch->old_name;
3910 * A type-change diff is always split into a patch to delete
3911 * old, immediately followed by a patch to create new (see
3912 * diff.c::run_diff()); in such a case it is Ok that the entry
3913 * to be deleted by the previous patch is still in the working
3914 * tree and in the index.
3916 * A patch to swap-rename between A and B would first rename A
3917 * to B and then rename B to A. While applying the first one,
3918 * the presence of B should not stop A from getting renamed to
3919 * B; ask to_be_deleted() about the later rename. Removal of
3920 * B and rename from A to B is handled the same way by asking
3921 * was_deleted().
3923 if ((tpatch = in_fn_table(state, new_name)) &&
3924 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3925 ok_if_exists = 1;
3926 else
3927 ok_if_exists = 0;
3929 if (new_name &&
3930 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3931 int err = check_to_create(state, new_name, ok_if_exists);
3933 if (err && state->threeway) {
3934 patch->direct_to_threeway = 1;
3935 } else switch (err) {
3936 case 0:
3937 break; /* happy */
3938 case EXISTS_IN_INDEX:
3939 return error(_("%s: already exists in index"), new_name);
3940 break;
3941 case EXISTS_IN_WORKTREE:
3942 return error(_("%s: already exists in working directory"),
3943 new_name);
3944 default:
3945 return err;
3948 if (!patch->new_mode) {
3949 if (0 < patch->is_new)
3950 patch->new_mode = S_IFREG | 0644;
3951 else
3952 patch->new_mode = patch->old_mode;
3956 if (new_name && old_name) {
3957 int same = !strcmp(old_name, new_name);
3958 if (!patch->new_mode)
3959 patch->new_mode = patch->old_mode;
3960 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3961 if (same)
3962 return error(_("new mode (%o) of %s does not "
3963 "match old mode (%o)"),
3964 patch->new_mode, new_name,
3965 patch->old_mode);
3966 else
3967 return error(_("new mode (%o) of %s does not "
3968 "match old mode (%o) of %s"),
3969 patch->new_mode, new_name,
3970 patch->old_mode, old_name);
3974 if (!state->unsafe_paths && check_unsafe_path(patch))
3975 return -128;
3978 * An attempt to read from or delete a path that is beyond a
3979 * symbolic link will be prevented by load_patch_target() that
3980 * is called at the beginning of apply_data() so we do not
3981 * have to worry about a patch marked with "is_delete" bit
3982 * here. We however need to make sure that the patch result
3983 * is not deposited to a path that is beyond a symbolic link
3984 * here.
3986 if (!patch->is_delete && path_is_beyond_symlin