The third batch
[git.git] / apply.c
blobf15afa9f6af98ae41af60eb2296ee35d38ccc7fd
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;
60 * Please update $__git_whitespacelist in git-completion.bash
61 * when you add new options.
63 return error(_("unrecognized whitespace option '%s'"), option);
66 static int parse_ignorewhitespace_option(struct apply_state *state,
67 const char *option)
69 if (!option || !strcmp(option, "no") ||
70 !strcmp(option, "false") || !strcmp(option, "never") ||
71 !strcmp(option, "none")) {
72 state->ws_ignore_action = ignore_ws_none;
73 return 0;
75 if (!strcmp(option, "change")) {
76 state->ws_ignore_action = ignore_ws_change;
77 return 0;
79 return error(_("unrecognized whitespace ignore option '%s'"), option);
82 int init_apply_state(struct apply_state *state,
83 struct repository *repo,
84 const char *prefix)
86 memset(state, 0, sizeof(*state));
87 state->prefix = prefix;
88 state->repo = repo;
89 state->apply = 1;
90 state->line_termination = '\n';
91 state->p_value = 1;
92 state->p_context = UINT_MAX;
93 state->squelch_whitespace_errors = 5;
94 state->ws_error_action = warn_on_ws_error;
95 state->ws_ignore_action = ignore_ws_none;
96 state->linenr = 1;
97 string_list_init(&state->fn_table, 0);
98 string_list_init(&state->limit_by_name, 0);
99 string_list_init(&state->symlink_changes, 0);
100 strbuf_init(&state->root, 0);
102 git_apply_config();
103 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
104 return -1;
105 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
106 return -1;
107 return 0;
110 void clear_apply_state(struct apply_state *state)
112 string_list_clear(&state->limit_by_name, 0);
113 string_list_clear(&state->symlink_changes, 0);
114 strbuf_release(&state->root);
116 /* &state->fn_table is cleared at the end of apply_patch() */
119 static void mute_routine(const char *msg, va_list params)
121 /* do nothing */
124 int check_apply_state(struct apply_state *state, int force_apply)
126 int is_not_gitdir = !startup_info->have_repository;
128 if (state->apply_with_reject && state->threeway)
129 return error(_("--reject and --3way cannot be used together."));
130 if (state->cached && state->threeway)
131 return error(_("--cached and --3way cannot be used together."));
132 if (state->threeway) {
133 if (is_not_gitdir)
134 return error(_("--3way outside a repository"));
135 state->check_index = 1;
137 if (state->apply_with_reject) {
138 state->apply = 1;
139 if (state->apply_verbosity == verbosity_normal)
140 state->apply_verbosity = verbosity_verbose;
142 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
143 state->apply = 0;
144 if (state->check_index && is_not_gitdir)
145 return error(_("--index outside a repository"));
146 if (state->cached) {
147 if (is_not_gitdir)
148 return error(_("--cached outside a repository"));
149 state->check_index = 1;
151 if (state->ita_only && (state->check_index || is_not_gitdir))
152 state->ita_only = 0;
153 if (state->check_index)
154 state->unsafe_paths = 0;
156 if (state->apply_verbosity <= verbosity_silent) {
157 state->saved_error_routine = get_error_routine();
158 state->saved_warn_routine = get_warn_routine();
159 set_error_routine(mute_routine);
160 set_warn_routine(mute_routine);
163 return 0;
166 static void set_default_whitespace_mode(struct apply_state *state)
168 if (!state->whitespace_option && !apply_default_whitespace)
169 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
173 * This represents one "hunk" from a patch, starting with
174 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
175 * patch text is pointed at by patch, and its byte length
176 * is stored in size. leading and trailing are the number
177 * of context lines.
179 struct fragment {
180 unsigned long leading, trailing;
181 unsigned long oldpos, oldlines;
182 unsigned long newpos, newlines;
184 * 'patch' is usually borrowed from buf in apply_patch(),
185 * but some codepaths store an allocated buffer.
187 const char *patch;
188 unsigned free_patch:1,
189 rejected:1;
190 int size;
191 int linenr;
192 struct fragment *next;
196 * When dealing with a binary patch, we reuse "leading" field
197 * to store the type of the binary hunk, either deflated "delta"
198 * or deflated "literal".
200 #define binary_patch_method leading
201 #define BINARY_DELTA_DEFLATED 1
202 #define BINARY_LITERAL_DEFLATED 2
205 * This represents a "patch" to a file, both metainfo changes
206 * such as creation/deletion, filemode and content changes represented
207 * as a series of fragments.
209 struct patch {
210 char *new_name, *old_name, *def_name;
211 unsigned int old_mode, new_mode;
212 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
213 int rejected;
214 unsigned ws_rule;
215 int lines_added, lines_deleted;
216 int score;
217 int extension_linenr; /* first line specifying delete/new/rename/copy */
218 unsigned int is_toplevel_relative:1;
219 unsigned int inaccurate_eof:1;
220 unsigned int is_binary:1;
221 unsigned int is_copy:1;
222 unsigned int is_rename:1;
223 unsigned int recount:1;
224 unsigned int conflicted_threeway:1;
225 unsigned int direct_to_threeway:1;
226 unsigned int crlf_in_old:1;
227 struct fragment *fragments;
228 char *result;
229 size_t resultsize;
230 char old_oid_prefix[GIT_MAX_HEXSZ + 1];
231 char new_oid_prefix[GIT_MAX_HEXSZ + 1];
232 struct patch *next;
234 /* three-way fallback result */
235 struct object_id threeway_stage[3];
238 static void free_fragment_list(struct fragment *list)
240 while (list) {
241 struct fragment *next = list->next;
242 if (list->free_patch)
243 free((char *)list->patch);
244 free(list);
245 list = next;
249 static void free_patch(struct patch *patch)
251 free_fragment_list(patch->fragments);
252 free(patch->def_name);
253 free(patch->old_name);
254 free(patch->new_name);
255 free(patch->result);
256 free(patch);
259 static void free_patch_list(struct patch *list)
261 while (list) {
262 struct patch *next = list->next;
263 free_patch(list);
264 list = next;
269 * A line in a file, len-bytes long (includes the terminating LF,
270 * except for an incomplete line at the end if the file ends with
271 * one), and its contents hashes to 'hash'.
273 struct line {
274 size_t len;
275 unsigned hash : 24;
276 unsigned flag : 8;
277 #define LINE_COMMON 1
278 #define LINE_PATCHED 2
282 * This represents a "file", which is an array of "lines".
284 struct image {
285 char *buf;
286 size_t len;
287 size_t nr;
288 size_t alloc;
289 struct line *line_allocated;
290 struct line *line;
293 static uint32_t hash_line(const char *cp, size_t len)
295 size_t i;
296 uint32_t h;
297 for (i = 0, h = 0; i < len; i++) {
298 if (!isspace(cp[i])) {
299 h = h * 3 + (cp[i] & 0xff);
302 return h;
306 * Compare lines s1 of length n1 and s2 of length n2, ignoring
307 * whitespace difference. Returns 1 if they match, 0 otherwise
309 static int fuzzy_matchlines(const char *s1, size_t n1,
310 const char *s2, size_t n2)
312 const char *end1 = s1 + n1;
313 const char *end2 = s2 + n2;
315 /* ignore line endings */
316 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
317 end1--;
318 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
319 end2--;
321 while (s1 < end1 && s2 < end2) {
322 if (isspace(*s1)) {
324 * Skip whitespace. We check on both buffers
325 * because we don't want "a b" to match "ab".
327 if (!isspace(*s2))
328 return 0;
329 while (s1 < end1 && isspace(*s1))
330 s1++;
331 while (s2 < end2 && isspace(*s2))
332 s2++;
333 } else if (*s1++ != *s2++)
334 return 0;
337 /* If we reached the end on one side only, lines don't match. */
338 return s1 == end1 && s2 == end2;
341 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
343 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
344 img->line_allocated[img->nr].len = len;
345 img->line_allocated[img->nr].hash = hash_line(bol, len);
346 img->line_allocated[img->nr].flag = flag;
347 img->nr++;
351 * "buf" has the file contents to be patched (read from various sources).
352 * attach it to "image" and add line-based index to it.
353 * "image" now owns the "buf".
355 static void prepare_image(struct image *image, char *buf, size_t len,
356 int prepare_linetable)
358 const char *cp, *ep;
360 memset(image, 0, sizeof(*image));
361 image->buf = buf;
362 image->len = len;
364 if (!prepare_linetable)
365 return;
367 ep = image->buf + image->len;
368 cp = image->buf;
369 while (cp < ep) {
370 const char *next;
371 for (next = cp; next < ep && *next != '\n'; next++)
373 if (next < ep)
374 next++;
375 add_line_info(image, cp, next - cp, 0);
376 cp = next;
378 image->line = image->line_allocated;
381 static void clear_image(struct image *image)
383 free(image->buf);
384 free(image->line_allocated);
385 memset(image, 0, sizeof(*image));
388 /* fmt must contain _one_ %s and no other substitution */
389 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
391 struct strbuf sb = STRBUF_INIT;
393 if (patch->old_name && patch->new_name &&
394 strcmp(patch->old_name, patch->new_name)) {
395 quote_c_style(patch->old_name, &sb, NULL, 0);
396 strbuf_addstr(&sb, " => ");
397 quote_c_style(patch->new_name, &sb, NULL, 0);
398 } else {
399 const char *n = patch->new_name;
400 if (!n)
401 n = patch->old_name;
402 quote_c_style(n, &sb, NULL, 0);
404 fprintf(output, fmt, sb.buf);
405 fputc('\n', output);
406 strbuf_release(&sb);
409 #define SLOP (16)
411 static int read_patch_file(struct strbuf *sb, int fd)
413 if (strbuf_read(sb, fd, 0) < 0)
414 return error_errno("git apply: failed to read");
417 * Make sure that we have some slop in the buffer
418 * so that we can do speculative "memcmp" etc, and
419 * see to it that it is NUL-filled.
421 strbuf_grow(sb, SLOP);
422 memset(sb->buf + sb->len, 0, SLOP);
423 return 0;
426 static unsigned long linelen(const char *buffer, unsigned long size)
428 unsigned long len = 0;
429 while (size--) {
430 len++;
431 if (*buffer++ == '\n')
432 break;
434 return len;
437 static int is_dev_null(const char *str)
439 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
442 #define TERM_SPACE 1
443 #define TERM_TAB 2
445 static int name_terminate(int c, int terminate)
447 if (c == ' ' && !(terminate & TERM_SPACE))
448 return 0;
449 if (c == '\t' && !(terminate & TERM_TAB))
450 return 0;
452 return 1;
455 /* remove double slashes to make --index work with such filenames */
456 static char *squash_slash(char *name)
458 int i = 0, j = 0;
460 if (!name)
461 return NULL;
463 while (name[i]) {
464 if ((name[j++] = name[i++]) == '/')
465 while (name[i] == '/')
466 i++;
468 name[j] = '\0';
469 return name;
472 static char *find_name_gnu(struct apply_state *state,
473 const char *line,
474 int p_value)
476 struct strbuf name = STRBUF_INIT;
477 char *cp;
480 * Proposed "new-style" GNU patch/diff format; see
481 * http://marc.info/?l=git&m=112927316408690&w=2
483 if (unquote_c_style(&name, line, NULL)) {
484 strbuf_release(&name);
485 return NULL;
488 for (cp = name.buf; p_value; p_value--) {
489 cp = strchr(cp, '/');
490 if (!cp) {
491 strbuf_release(&name);
492 return NULL;
494 cp++;
497 strbuf_remove(&name, 0, cp - name.buf);
498 if (state->root.len)
499 strbuf_insert(&name, 0, state->root.buf, state->root.len);
500 return squash_slash(strbuf_detach(&name, NULL));
503 static size_t sane_tz_len(const char *line, size_t len)
505 const char *tz, *p;
507 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
508 return 0;
509 tz = line + len - strlen(" +0500");
511 if (tz[1] != '+' && tz[1] != '-')
512 return 0;
514 for (p = tz + 2; p != line + len; p++)
515 if (!isdigit(*p))
516 return 0;
518 return line + len - tz;
521 static size_t tz_with_colon_len(const char *line, size_t len)
523 const char *tz, *p;
525 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
526 return 0;
527 tz = line + len - strlen(" +08:00");
529 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
530 return 0;
531 p = tz + 2;
532 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
533 !isdigit(*p++) || !isdigit(*p++))
534 return 0;
536 return line + len - tz;
539 static size_t date_len(const char *line, size_t len)
541 const char *date, *p;
543 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
544 return 0;
545 p = date = line + len - strlen("72-02-05");
547 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
548 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
549 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
550 return 0;
552 if (date - line >= strlen("19") &&
553 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
554 date -= strlen("19");
556 return line + len - date;
559 static size_t short_time_len(const char *line, size_t len)
561 const char *time, *p;
563 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
564 return 0;
565 p = time = line + len - strlen(" 07:01:32");
567 /* Permit 1-digit hours? */
568 if (*p++ != ' ' ||
569 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
570 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
571 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
572 return 0;
574 return line + len - time;
577 static size_t fractional_time_len(const char *line, size_t len)
579 const char *p;
580 size_t n;
582 /* Expected format: 19:41:17.620000023 */
583 if (!len || !isdigit(line[len - 1]))
584 return 0;
585 p = line + len - 1;
587 /* Fractional seconds. */
588 while (p > line && isdigit(*p))
589 p--;
590 if (*p != '.')
591 return 0;
593 /* Hours, minutes, and whole seconds. */
594 n = short_time_len(line, p - line);
595 if (!n)
596 return 0;
598 return line + len - p + n;
601 static size_t trailing_spaces_len(const char *line, size_t len)
603 const char *p;
605 /* Expected format: ' ' x (1 or more) */
606 if (!len || line[len - 1] != ' ')
607 return 0;
609 p = line + len;
610 while (p != line) {
611 p--;
612 if (*p != ' ')
613 return line + len - (p + 1);
616 /* All spaces! */
617 return len;
620 static size_t diff_timestamp_len(const char *line, size_t len)
622 const char *end = line + len;
623 size_t n;
626 * Posix: 2010-07-05 19:41:17
627 * GNU: 2010-07-05 19:41:17.620000023 -0500
630 if (!isdigit(end[-1]))
631 return 0;
633 n = sane_tz_len(line, end - line);
634 if (!n)
635 n = tz_with_colon_len(line, end - line);
636 end -= n;
638 n = short_time_len(line, end - line);
639 if (!n)
640 n = fractional_time_len(line, end - line);
641 end -= n;
643 n = date_len(line, end - line);
644 if (!n) /* No date. Too bad. */
645 return 0;
646 end -= n;
648 if (end == line) /* No space before date. */
649 return 0;
650 if (end[-1] == '\t') { /* Success! */
651 end--;
652 return line + len - end;
654 if (end[-1] != ' ') /* No space before date. */
655 return 0;
657 /* Whitespace damage. */
658 end -= trailing_spaces_len(line, end - line);
659 return line + len - end;
662 static char *find_name_common(struct apply_state *state,
663 const char *line,
664 const char *def,
665 int p_value,
666 const char *end,
667 int terminate)
669 int len;
670 const char *start = NULL;
672 if (p_value == 0)
673 start = line;
674 while (line != end) {
675 char c = *line;
677 if (!end && isspace(c)) {
678 if (c == '\n')
679 break;
680 if (name_terminate(c, terminate))
681 break;
683 line++;
684 if (c == '/' && !--p_value)
685 start = line;
687 if (!start)
688 return squash_slash(xstrdup_or_null(def));
689 len = line - start;
690 if (!len)
691 return squash_slash(xstrdup_or_null(def));
694 * Generally we prefer the shorter name, especially
695 * if the other one is just a variation of that with
696 * something else tacked on to the end (ie "file.orig"
697 * or "file~").
699 if (def) {
700 int deflen = strlen(def);
701 if (deflen < len && !strncmp(start, def, deflen))
702 return squash_slash(xstrdup(def));
705 if (state->root.len) {
706 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
707 return squash_slash(ret);
710 return squash_slash(xmemdupz(start, len));
713 static char *find_name(struct apply_state *state,
714 const char *line,
715 char *def,
716 int p_value,
717 int terminate)
719 if (*line == '"') {
720 char *name = find_name_gnu(state, line, p_value);
721 if (name)
722 return name;
725 return find_name_common(state, line, def, p_value, NULL, terminate);
728 static char *find_name_traditional(struct apply_state *state,
729 const char *line,
730 char *def,
731 int p_value)
733 size_t len;
734 size_t date_len;
736 if (*line == '"') {
737 char *name = find_name_gnu(state, line, p_value);
738 if (name)
739 return name;
742 len = strchrnul(line, '\n') - line;
743 date_len = diff_timestamp_len(line, len);
744 if (!date_len)
745 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
746 len -= date_len;
748 return find_name_common(state, line, def, p_value, line + len, 0);
752 * Given the string after "--- " or "+++ ", guess the appropriate
753 * p_value for the given patch.
755 static int guess_p_value(struct apply_state *state, const char *nameline)
757 char *name, *cp;
758 int val = -1;
760 if (is_dev_null(nameline))
761 return -1;
762 name = find_name_traditional(state, nameline, NULL, 0);
763 if (!name)
764 return -1;
765 cp = strchr(name, '/');
766 if (!cp)
767 val = 0;
768 else if (state->prefix) {
770 * Does it begin with "a/$our-prefix" and such? Then this is
771 * very likely to apply to our directory.
773 if (starts_with(name, state->prefix))
774 val = count_slashes(state->prefix);
775 else {
776 cp++;
777 if (starts_with(cp, state->prefix))
778 val = count_slashes(state->prefix) + 1;
781 free(name);
782 return val;
786 * Does the ---/+++ line have the POSIX timestamp after the last HT?
787 * GNU diff puts epoch there to signal a creation/deletion event. Is
788 * this such a timestamp?
790 static int has_epoch_timestamp(const char *nameline)
793 * We are only interested in epoch timestamp; any non-zero
794 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
795 * For the same reason, the date must be either 1969-12-31 or
796 * 1970-01-01, and the seconds part must be "00".
798 const char stamp_regexp[] =
799 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
801 "([-+][0-2][0-9]:?[0-5][0-9])\n";
802 const char *timestamp = NULL, *cp, *colon;
803 static regex_t *stamp;
804 regmatch_t m[10];
805 int zoneoffset, epoch_hour, hour, minute;
806 int status;
808 for (cp = nameline; *cp != '\n'; cp++) {
809 if (*cp == '\t')
810 timestamp = cp + 1;
812 if (!timestamp)
813 return 0;
816 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
817 * (west of GMT) or 1970-01-01 (east of GMT)
819 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
820 epoch_hour = 24;
821 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
822 epoch_hour = 0;
823 else
824 return 0;
826 if (!stamp) {
827 stamp = xmalloc(sizeof(*stamp));
828 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
829 warning(_("Cannot prepare timestamp regexp %s"),
830 stamp_regexp);
831 return 0;
835 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
836 if (status) {
837 if (status != REG_NOMATCH)
838 warning(_("regexec returned %d for input: %s"),
839 status, timestamp);
840 return 0;
843 hour = strtol(timestamp, NULL, 10);
844 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
846 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
847 if (*colon == ':')
848 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
849 else
850 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
851 if (timestamp[m[3].rm_so] == '-')
852 zoneoffset = -zoneoffset;
854 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
858 * Get the name etc info from the ---/+++ lines of a traditional patch header
860 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
861 * files, we can happily check the index for a match, but for creating a
862 * new file we should try to match whatever "patch" does. I have no idea.
864 static int parse_traditional_patch(struct apply_state *state,
865 const char *first,
866 const char *second,
867 struct patch *patch)
869 char *name;
871 first += 4; /* skip "--- " */
872 second += 4; /* skip "+++ " */
873 if (!state->p_value_known) {
874 int p, q;
875 p = guess_p_value(state, first);
876 q = guess_p_value(state, second);
877 if (p < 0) p = q;
878 if (0 <= p && p == q) {
879 state->p_value = p;
880 state->p_value_known = 1;
883 if (is_dev_null(first)) {
884 patch->is_new = 1;
885 patch->is_delete = 0;
886 name = find_name_traditional(state, second, NULL, state->p_value);
887 patch->new_name = name;
888 } else if (is_dev_null(second)) {
889 patch->is_new = 0;
890 patch->is_delete = 1;
891 name = find_name_traditional(state, first, NULL, state->p_value);
892 patch->old_name = name;
893 } else {
894 char *first_name;
895 first_name = find_name_traditional(state, first, NULL, state->p_value);
896 name = find_name_traditional(state, second, first_name, state->p_value);
897 free(first_name);
898 if (has_epoch_timestamp(first)) {
899 patch->is_new = 1;
900 patch->is_delete = 0;
901 patch->new_name = name;
902 } else if (has_epoch_timestamp(second)) {
903 patch->is_new = 0;
904 patch->is_delete = 1;
905 patch->old_name = name;
906 } else {
907 patch->old_name = name;
908 patch->new_name = xstrdup_or_null(name);
911 if (!name)
912 return error(_("unable to find filename in patch at line %d"), state->linenr);
914 return 0;
917 static int gitdiff_hdrend(struct apply_state *state,
918 const char *line,
919 struct patch *patch)
921 return 1;
925 * We're anal about diff header consistency, to make
926 * sure that we don't end up having strange ambiguous
927 * patches floating around.
929 * As a result, gitdiff_{old|new}name() will check
930 * their names against any previous information, just
931 * to make sure..
933 #define DIFF_OLD_NAME 0
934 #define DIFF_NEW_NAME 1
936 static int gitdiff_verify_name(struct apply_state *state,
937 const char *line,
938 int isnull,
939 char **name,
940 int side)
942 if (!*name && !isnull) {
943 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
944 return 0;
947 if (*name) {
948 char *another;
949 if (isnull)
950 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
951 *name, state->linenr);
952 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
953 if (!another || strcmp(another, *name)) {
954 free(another);
955 return error((side == DIFF_NEW_NAME) ?
956 _("git apply: bad git-diff - inconsistent new filename on line %d") :
957 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
959 free(another);
960 } else {
961 if (!is_dev_null(line))
962 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
965 return 0;
968 static int gitdiff_oldname(struct apply_state *state,
969 const char *line,
970 struct patch *patch)
972 return gitdiff_verify_name(state, line,
973 patch->is_new, &patch->old_name,
974 DIFF_OLD_NAME);
977 static int gitdiff_newname(struct apply_state *state,
978 const char *line,
979 struct patch *patch)
981 return gitdiff_verify_name(state, line,
982 patch->is_delete, &patch->new_name,
983 DIFF_NEW_NAME);
986 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
988 char *end;
989 *mode = strtoul(line, &end, 8);
990 if (end == line || !isspace(*end))
991 return error(_("invalid mode on line %d: %s"), linenr, line);
992 return 0;
995 static int gitdiff_oldmode(struct apply_state *state,
996 const char *line,
997 struct patch *patch)
999 return parse_mode_line(line, state->linenr, &patch->old_mode);
1002 static int gitdiff_newmode(struct apply_state *state,
1003 const char *line,
1004 struct patch *patch)
1006 return parse_mode_line(line, state->linenr, &patch->new_mode);
1009 static int gitdiff_delete(struct apply_state *state,
1010 const char *line,
1011 struct patch *patch)
1013 patch->is_delete = 1;
1014 free(patch->old_name);
1015 patch->old_name = xstrdup_or_null(patch->def_name);
1016 return gitdiff_oldmode(state, line, patch);
1019 static int gitdiff_newfile(struct apply_state *state,
1020 const char *line,
1021 struct patch *patch)
1023 patch->is_new = 1;
1024 free(patch->new_name);
1025 patch->new_name = xstrdup_or_null(patch->def_name);
1026 return gitdiff_newmode(state, line, patch);
1029 static int gitdiff_copysrc(struct apply_state *state,
1030 const char *line,
1031 struct patch *patch)
1033 patch->is_copy = 1;
1034 free(patch->old_name);
1035 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1036 return 0;
1039 static int gitdiff_copydst(struct apply_state *state,
1040 const char *line,
1041 struct patch *patch)
1043 patch->is_copy = 1;
1044 free(patch->new_name);
1045 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1046 return 0;
1049 static int gitdiff_renamesrc(struct apply_state *state,
1050 const char *line,
1051 struct patch *patch)
1053 patch->is_rename = 1;
1054 free(patch->old_name);
1055 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1056 return 0;
1059 static int gitdiff_renamedst(struct apply_state *state,
1060 const char *line,
1061 struct patch *patch)
1063 patch->is_rename = 1;
1064 free(patch->new_name);
1065 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1066 return 0;
1069 static int gitdiff_similarity(struct apply_state *state,
1070 const char *line,
1071 struct patch *patch)
1073 unsigned long val = strtoul(line, NULL, 10);
1074 if (val <= 100)
1075 patch->score = val;
1076 return 0;
1079 static int gitdiff_dissimilarity(struct apply_state *state,
1080 const char *line,
1081 struct patch *patch)
1083 unsigned long val = strtoul(line, NULL, 10);
1084 if (val <= 100)
1085 patch->score = val;
1086 return 0;
1089 static int gitdiff_index(struct apply_state *state,
1090 const char *line,
1091 struct patch *patch)
1094 * index line is N hexadecimal, "..", N hexadecimal,
1095 * and optional space with octal mode.
1097 const char *ptr, *eol;
1098 int len;
1099 const unsigned hexsz = the_hash_algo->hexsz;
1101 ptr = strchr(line, '.');
1102 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1103 return 0;
1104 len = ptr - line;
1105 memcpy(patch->old_oid_prefix, line, len);
1106 patch->old_oid_prefix[len] = 0;
1108 line = ptr + 2;
1109 ptr = strchr(line, ' ');
1110 eol = strchrnul(line, '\n');
1112 if (!ptr || eol < ptr)
1113 ptr = eol;
1114 len = ptr - line;
1116 if (hexsz < len)
1117 return 0;
1118 memcpy(patch->new_oid_prefix, line, len);
1119 patch->new_oid_prefix[len] = 0;
1120 if (*ptr == ' ')
1121 return gitdiff_oldmode(state, ptr + 1, patch);
1122 return 0;
1126 * This is normal for a diff that doesn't change anything: we'll fall through
1127 * into the next diff. Tell the parser to break out.
1129 static int gitdiff_unrecognized(struct apply_state *state,
1130 const char *line,
1131 struct patch *patch)
1133 return 1;
1137 * Skip p_value leading components from "line"; as we do not accept
1138 * absolute paths, return NULL in that case.
1140 static const char *skip_tree_prefix(struct apply_state *state,
1141 const char *line,
1142 int llen)
1144 int nslash;
1145 int i;
1147 if (!state->p_value)
1148 return (llen && line[0] == '/') ? NULL : line;
1150 nslash = state->p_value;
1151 for (i = 0; i < llen; i++) {
1152 int ch = line[i];
1153 if (ch == '/' && --nslash <= 0)
1154 return (i == 0) ? NULL : &line[i + 1];
1156 return NULL;
1160 * This is to extract the same name that appears on "diff --git"
1161 * line. We do not find and return anything if it is a rename
1162 * patch, and it is OK because we will find the name elsewhere.
1163 * We need to reliably find name only when it is mode-change only,
1164 * creation or deletion of an empty file. In any of these cases,
1165 * both sides are the same name under a/ and b/ respectively.
1167 static char *git_header_name(struct apply_state *state,
1168 const char *line,
1169 int llen)
1171 const char *name;
1172 const char *second = NULL;
1173 size_t len, line_len;
1175 line += strlen("diff --git ");
1176 llen -= strlen("diff --git ");
1178 if (*line == '"') {
1179 const char *cp;
1180 struct strbuf first = STRBUF_INIT;
1181 struct strbuf sp = STRBUF_INIT;
1183 if (unquote_c_style(&first, line, &second))
1184 goto free_and_fail1;
1186 /* strip the a/b prefix including trailing slash */
1187 cp = skip_tree_prefix(state, first.buf, first.len);
1188 if (!cp)
1189 goto free_and_fail1;
1190 strbuf_remove(&first, 0, cp - first.buf);
1193 * second points at one past closing dq of name.
1194 * find the second name.
1196 while ((second < line + llen) && isspace(*second))
1197 second++;
1199 if (line + llen <= second)
1200 goto free_and_fail1;
1201 if (*second == '"') {
1202 if (unquote_c_style(&sp, second, NULL))
1203 goto free_and_fail1;
1204 cp = skip_tree_prefix(state, sp.buf, sp.len);
1205 if (!cp)
1206 goto free_and_fail1;
1207 /* They must match, otherwise ignore */
1208 if (strcmp(cp, first.buf))
1209 goto free_and_fail1;
1210 strbuf_release(&sp);
1211 return strbuf_detach(&first, NULL);
1214 /* unquoted second */
1215 cp = skip_tree_prefix(state, second, line + llen - second);
1216 if (!cp)
1217 goto free_and_fail1;
1218 if (line + llen - cp != first.len ||
1219 memcmp(first.buf, cp, first.len))
1220 goto free_and_fail1;
1221 return strbuf_detach(&first, NULL);
1223 free_and_fail1:
1224 strbuf_release(&first);
1225 strbuf_release(&sp);
1226 return NULL;
1229 /* unquoted first name */
1230 name = skip_tree_prefix(state, line, llen);
1231 if (!name)
1232 return NULL;
1235 * since the first name is unquoted, a dq if exists must be
1236 * the beginning of the second name.
1238 for (second = name; second < line + llen; second++) {
1239 if (*second == '"') {
1240 struct strbuf sp = STRBUF_INIT;
1241 const char *np;
1243 if (unquote_c_style(&sp, second, NULL))
1244 goto free_and_fail2;
1246 np = skip_tree_prefix(state, sp.buf, sp.len);
1247 if (!np)
1248 goto free_and_fail2;
1250 len = sp.buf + sp.len - np;
1251 if (len < second - name &&
1252 !strncmp(np, name, len) &&
1253 isspace(name[len])) {
1254 /* Good */
1255 strbuf_remove(&sp, 0, np - sp.buf);
1256 return strbuf_detach(&sp, NULL);
1259 free_and_fail2:
1260 strbuf_release(&sp);
1261 return NULL;
1266 * Accept a name only if it shows up twice, exactly the same
1267 * form.
1269 second = strchr(name, '\n');
1270 if (!second)
1271 return NULL;
1272 line_len = second - name;
1273 for (len = 0 ; ; len++) {
1274 switch (name[len]) {
1275 default:
1276 continue;
1277 case '\n':
1278 return NULL;
1279 case '\t': case ' ':
1281 * Is this the separator between the preimage
1282 * and the postimage pathname? Again, we are
1283 * only interested in the case where there is
1284 * no rename, as this is only to set def_name
1285 * and a rename patch has the names elsewhere
1286 * in an unambiguous form.
1288 if (!name[len + 1])
1289 return NULL; /* no postimage name */
1290 second = skip_tree_prefix(state, name + len + 1,
1291 line_len - (len + 1));
1292 if (!second)
1293 return NULL;
1295 * Does len bytes starting at "name" and "second"
1296 * (that are separated by one HT or SP we just
1297 * found) exactly match?
1299 if (second[len] == '\n' && !strncmp(name, second, len))
1300 return xmemdupz(name, len);
1305 static int check_header_line(struct apply_state *state, struct patch *patch)
1307 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1308 (patch->is_rename == 1) + (patch->is_copy == 1);
1309 if (extensions > 1)
1310 return error(_("inconsistent header lines %d and %d"),
1311 patch->extension_linenr, state->linenr);
1312 if (extensions && !patch->extension_linenr)
1313 patch->extension_linenr = state->linenr;
1314 return 0;
1317 /* Verify that we recognize the lines following a git header */
1318 static int parse_git_header(struct apply_state *state,
1319 const char *line,
1320 int len,
1321 unsigned int size,
1322 struct patch *patch)
1324 unsigned long offset;
1326 /* A git diff has explicit new/delete information, so we don't guess */
1327 patch->is_new = 0;
1328 patch->is_delete = 0;
1331 * Some things may not have the old name in the
1332 * rest of the headers anywhere (pure mode changes,
1333 * or removing or adding empty files), so we get
1334 * the default name from the header.
1336 patch->def_name = git_header_name(state, line, len);
1337 if (patch->def_name && state->root.len) {
1338 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1339 free(patch->def_name);
1340 patch->def_name = s;
1343 line += len;
1344 size -= len;
1345 state->linenr++;
1346 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1347 static const struct opentry {
1348 const char *str;
1349 int (*fn)(struct apply_state *, const char *, struct patch *);
1350 } optable[] = {
1351 { "@@ -", gitdiff_hdrend },
1352 { "--- ", gitdiff_oldname },
1353 { "+++ ", gitdiff_newname },
1354 { "old mode ", gitdiff_oldmode },
1355 { "new mode ", gitdiff_newmode },
1356 { "deleted file mode ", gitdiff_delete },
1357 { "new file mode ", gitdiff_newfile },
1358 { "copy from ", gitdiff_copysrc },
1359 { "copy to ", gitdiff_copydst },
1360 { "rename old ", gitdiff_renamesrc },
1361 { "rename new ", gitdiff_renamedst },
1362 { "rename from ", gitdiff_renamesrc },
1363 { "rename to ", gitdiff_renamedst },
1364 { "similarity index ", gitdiff_similarity },
1365 { "dissimilarity index ", gitdiff_dissimilarity },
1366 { "index ", gitdiff_index },
1367 { "", gitdiff_unrecognized },
1369 int i;
1371 len = linelen(line, size);
1372 if (!len || line[len-1] != '\n')
1373 break;
1374 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1375 const struct opentry *p = optable + i;
1376 int oplen = strlen(p->str);
1377 int res;
1378 if (len < oplen || memcmp(p->str, line, oplen))
1379 continue;
1380 res = p->fn(state, line + oplen, patch);
1381 if (res < 0)
1382 return -1;
1383 if (check_header_line(state, patch))
1384 return -1;
1385 if (res > 0)
1386 return offset;
1387 break;
1391 return offset;
1394 static int parse_num(const char *line, unsigned long *p)
1396 char *ptr;
1398 if (!isdigit(*line))
1399 return 0;
1400 *p = strtoul(line, &ptr, 10);
1401 return ptr - line;
1404 static int parse_range(const char *line, int len, int offset, const char *expect,
1405 unsigned long *p1, unsigned long *p2)
1407 int digits, ex;
1409 if (offset < 0 || offset >= len)
1410 return -1;
1411 line += offset;
1412 len -= offset;
1414 digits = parse_num(line, p1);
1415 if (!digits)
1416 return -1;
1418 offset += digits;
1419 line += digits;
1420 len -= digits;
1422 *p2 = 1;
1423 if (*line == ',') {
1424 digits = parse_num(line+1, p2);
1425 if (!digits)
1426 return -1;
1428 offset += digits+1;
1429 line += digits+1;
1430 len -= digits+1;
1433 ex = strlen(expect);
1434 if (ex > len)
1435 return -1;
1436 if (memcmp(line, expect, ex))
1437 return -1;
1439 return offset + ex;
1442 static void recount_diff(const char *line, int size, struct fragment *fragment)
1444 int oldlines = 0, newlines = 0, ret = 0;
1446 if (size < 1) {
1447 warning("recount: ignore empty hunk");
1448 return;
1451 for (;;) {
1452 int len = linelen(line, size);
1453 size -= len;
1454 line += len;
1456 if (size < 1)
1457 break;
1459 switch (*line) {
1460 case ' ': case '\n':
1461 newlines++;
1462 /* fall through */
1463 case '-':
1464 oldlines++;
1465 continue;
1466 case '+':
1467 newlines++;
1468 continue;
1469 case '\\':
1470 continue;
1471 case '@':
1472 ret = size < 3 || !starts_with(line, "@@ ");
1473 break;
1474 case 'd':
1475 ret = size < 5 || !starts_with(line, "diff ");
1476 break;
1477 default:
1478 ret = -1;
1479 break;
1481 if (ret) {
1482 warning(_("recount: unexpected line: %.*s"),
1483 (int)linelen(line, size), line);
1484 return;
1486 break;
1488 fragment->oldlines = oldlines;
1489 fragment->newlines = newlines;
1493 * Parse a unified diff fragment header of the
1494 * form "@@ -a,b +c,d @@"
1496 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1498 int offset;
1500 if (!len || line[len-1] != '\n')
1501 return -1;
1503 /* Figure out the number of lines in a fragment */
1504 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1505 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1507 return offset;
1511 * Find file diff header
1513 * Returns:
1514 * -1 if no header was found
1515 * -128 in case of error
1516 * the size of the header in bytes (called "offset") otherwise
1518 static int find_header(struct apply_state *state,
1519 const char *line,
1520 unsigned long size,
1521 int *hdrsize,
1522 struct patch *patch)
1524 unsigned long offset, len;
1526 patch->is_toplevel_relative = 0;
1527 patch->is_rename = patch->is_copy = 0;
1528 patch->is_new = patch->is_delete = -1;
1529 patch->old_mode = patch->new_mode = 0;
1530 patch->old_name = patch->new_name = NULL;
1531 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1532 unsigned long nextlen;
1534 len = linelen(line, size);
1535 if (!len)
1536 break;
1538 /* Testing this early allows us to take a few shortcuts.. */
1539 if (len < 6)
1540 continue;
1543 * Make sure we don't find any unconnected patch fragments.
1544 * That's a sign that we didn't find a header, and that a
1545 * patch has become corrupted/broken up.
1547 if (!memcmp("@@ -", line, 4)) {
1548 struct fragment dummy;
1549 if (parse_fragment_header(line, len, &dummy) < 0)
1550 continue;
1551 error(_("patch fragment without header at line %d: %.*s"),
1552 state->linenr, (int)len-1, line);
1553 return -128;
1556 if (size < len + 6)
1557 break;
1560 * Git patch? It might not have a real patch, just a rename
1561 * or mode change, so we handle that specially
1563 if (!memcmp("diff --git ", line, 11)) {
1564 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1565 if (git_hdr_len < 0)
1566 return -128;
1567 if (git_hdr_len <= len)
1568 continue;
1569 if (!patch->old_name && !patch->new_name) {
1570 if (!patch->def_name) {
1571 error(Q_("git diff header lacks filename information when removing "
1572 "%d leading pathname component (line %d)",
1573 "git diff header lacks filename information when removing "
1574 "%d leading pathname components (line %d)",
1575 state->p_value),
1576 state->p_value, state->linenr);
1577 return -128;
1579 patch->old_name = xstrdup(patch->def_name);
1580 patch->new_name = xstrdup(patch->def_name);
1582 if ((!patch->new_name && !patch->is_delete) ||
1583 (!patch->old_name && !patch->is_new)) {
1584 error(_("git diff header lacks filename information "
1585 "(line %d)"), state->linenr);
1586 return -128;
1588 patch->is_toplevel_relative = 1;
1589 *hdrsize = git_hdr_len;
1590 return offset;
1593 /* --- followed by +++ ? */
1594 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1595 continue;
1598 * We only accept unified patches, so we want it to
1599 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1600 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1602 nextlen = linelen(line + len, size - len);
1603 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1604 continue;
1606 /* Ok, we'll consider it a patch */
1607 if (parse_traditional_patch(state, line, line+len, patch))
1608 return -128;
1609 *hdrsize = len + nextlen;
1610 state->linenr += 2;
1611 return offset;
1613 return -1;
1616 static void record_ws_error(struct apply_state *state,
1617 unsigned result,
1618 const char *line,
1619 int len,
1620 int linenr)
1622 char *err;
1624 if (!result)
1625 return;
1627 state->whitespace_error++;
1628 if (state->squelch_whitespace_errors &&
1629 state->squelch_whitespace_errors < state->whitespace_error)
1630 return;
1632 err = whitespace_error_string(result);
1633 if (state->apply_verbosity > verbosity_silent)
1634 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1635 state->patch_input_file, linenr, err, len, line);
1636 free(err);
1639 static void check_whitespace(struct apply_state *state,
1640 const char *line,
1641 int len,
1642 unsigned ws_rule)
1644 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1646 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1650 * Check if the patch has context lines with CRLF or
1651 * the patch wants to remove lines with CRLF.
1653 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1655 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1656 patch->ws_rule |= WS_CR_AT_EOL;
1657 patch->crlf_in_old = 1;
1663 * Parse a unified diff. Note that this really needs to parse each
1664 * fragment separately, since the only way to know the difference
1665 * between a "---" that is part of a patch, and a "---" that starts
1666 * the next patch is to look at the line counts..
1668 static int parse_fragment(struct apply_state *state,
1669 const char *line,
1670 unsigned long size,
1671 struct patch *patch,
1672 struct fragment *fragment)
1674 int added, deleted;
1675 int len = linelen(line, size), offset;
1676 unsigned long oldlines, newlines;
1677 unsigned long leading, trailing;
1679 offset = parse_fragment_header(line, len, fragment);
1680 if (offset < 0)
1681 return -1;
1682 if (offset > 0 && patch->recount)
1683 recount_diff(line + offset, size - offset, fragment);
1684 oldlines = fragment->oldlines;
1685 newlines = fragment->newlines;
1686 leading = 0;
1687 trailing = 0;
1689 /* Parse the thing.. */
1690 line += len;
1691 size -= len;
1692 state->linenr++;
1693 added = deleted = 0;
1694 for (offset = len;
1695 0 < size;
1696 offset += len, size -= len, line += len, state->linenr++) {
1697 if (!oldlines && !newlines)
1698 break;
1699 len = linelen(line, size);
1700 if (!len || line[len-1] != '\n')
1701 return -1;
1702 switch (*line) {
1703 default:
1704 return -1;
1705 case '\n': /* newer GNU diff, an empty context line */
1706 case ' ':
1707 oldlines--;
1708 newlines--;
1709 if (!deleted && !added)
1710 leading++;
1711 trailing++;
1712 check_old_for_crlf(patch, line, len);
1713 if (!state->apply_in_reverse &&
1714 state->ws_error_action == correct_ws_error)
1715 check_whitespace(state, line, len, patch->ws_rule);
1716 break;
1717 case '-':
1718 if (!state->apply_in_reverse)
1719 check_old_for_crlf(patch, line, len);
1720 if (state->apply_in_reverse &&
1721 state->ws_error_action != nowarn_ws_error)
1722 check_whitespace(state, line, len, patch->ws_rule);
1723 deleted++;
1724 oldlines--;
1725 trailing = 0;
1726 break;
1727 case '+':
1728 if (state->apply_in_reverse)
1729 check_old_for_crlf(patch, line, len);
1730 if (!state->apply_in_reverse &&
1731 state->ws_error_action != nowarn_ws_error)
1732 check_whitespace(state, line, len, patch->ws_rule);
1733 added++;
1734 newlines--;
1735 trailing = 0;
1736 break;
1739 * We allow "\ No newline at end of file". Depending
1740 * on locale settings when the patch was produced we
1741 * don't know what this line looks like. The only
1742 * thing we do know is that it begins with "\ ".
1743 * Checking for 12 is just for sanity check -- any
1744 * l10n of "\ No newline..." is at least that long.
1746 case '\\':
1747 if (len < 12 || memcmp(line, "\\ ", 2))
1748 return -1;
1749 break;
1752 if (oldlines || newlines)
1753 return -1;
1754 if (!patch->recount && !deleted && !added)
1755 return -1;
1757 fragment->leading = leading;
1758 fragment->trailing = trailing;
1761 * If a fragment ends with an incomplete line, we failed to include
1762 * it in the above loop because we hit oldlines == newlines == 0
1763 * before seeing it.
1765 if (12 < size && !memcmp(line, "\\ ", 2))
1766 offset += linelen(line, size);
1768 patch->lines_added += added;
1769 patch->lines_deleted += deleted;
1771 if (0 < patch->is_new && oldlines)
1772 return error(_("new file depends on old contents"));
1773 if (0 < patch->is_delete && newlines)
1774 return error(_("deleted file still has contents"));
1775 return offset;
1779 * We have seen "diff --git a/... b/..." header (or a traditional patch
1780 * header). Read hunks that belong to this patch into fragments and hang
1781 * them to the given patch structure.
1783 * The (fragment->patch, fragment->size) pair points into the memory given
1784 * by the caller, not a copy, when we return.
1786 * Returns:
1787 * -1 in case of error,
1788 * the number of bytes in the patch otherwise.
1790 static int parse_single_patch(struct apply_state *state,
1791 const char *line,
1792 unsigned long size,
1793 struct patch *patch)
1795 unsigned long offset = 0;
1796 unsigned long oldlines = 0, newlines = 0, context = 0;
1797 struct fragment **fragp = &patch->fragments;
1799 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1800 struct fragment *fragment;
1801 int len;
1803 fragment = xcalloc(1, sizeof(*fragment));
1804 fragment->linenr = state->linenr;
1805 len = parse_fragment(state, line, size, patch, fragment);
1806 if (len <= 0) {
1807 free(fragment);
1808 return error(_("corrupt patch at line %d"), state->linenr);
1810 fragment->patch = line;
1811 fragment->size = len;
1812 oldlines += fragment->oldlines;
1813 newlines += fragment->newlines;
1814 context += fragment->leading + fragment->trailing;
1816 *fragp = fragment;
1817 fragp = &fragment->next;
1819 offset += len;
1820 line += len;
1821 size -= len;
1825 * If something was removed (i.e. we have old-lines) it cannot
1826 * be creation, and if something was added it cannot be
1827 * deletion. However, the reverse is not true; --unified=0
1828 * patches that only add are not necessarily creation even
1829 * though they do not have any old lines, and ones that only
1830 * delete are not necessarily deletion.
1832 * Unfortunately, a real creation/deletion patch do _not_ have
1833 * any context line by definition, so we cannot safely tell it
1834 * apart with --unified=0 insanity. At least if the patch has
1835 * more than one hunk it is not creation or deletion.
1837 if (patch->is_new < 0 &&
1838 (oldlines || (patch->fragments && patch->fragments->next)))
1839 patch->is_new = 0;
1840 if (patch->is_delete < 0 &&
1841 (newlines || (patch->fragments && patch->fragments->next)))
1842 patch->is_delete = 0;
1844 if (0 < patch->is_new && oldlines)
1845 return error(_("new file %s depends on old contents"), patch->new_name);
1846 if (0 < patch->is_delete && newlines)
1847 return error(_("deleted file %s still has contents"), patch->old_name);
1848 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1849 fprintf_ln(stderr,
1850 _("** warning: "
1851 "file %s becomes empty but is not deleted"),
1852 patch->new_name);
1854 return offset;
1857 static inline int metadata_changes(struct patch *patch)
1859 return patch->is_rename > 0 ||
1860 patch->is_copy > 0 ||
1861 patch->is_new > 0 ||
1862 patch->is_delete ||
1863 (patch->old_mode && patch->new_mode &&
1864 patch->old_mode != patch->new_mode);
1867 static char *inflate_it(const void *data, unsigned long size,
1868 unsigned long inflated_size)
1870 git_zstream stream;
1871 void *out;
1872 int st;
1874 memset(&stream, 0, sizeof(stream));
1876 stream.next_in = (unsigned char *)data;
1877 stream.avail_in = size;
1878 stream.next_out = out = xmalloc(inflated_size);
1879 stream.avail_out = inflated_size;
1880 git_inflate_init(&stream);
1881 st = git_inflate(&stream, Z_FINISH);
1882 git_inflate_end(&stream);
1883 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1884 free(out);
1885 return NULL;
1887 return out;
1891 * Read a binary hunk and return a new fragment; fragment->patch
1892 * points at an allocated memory that the caller must free, so
1893 * it is marked as "->free_patch = 1".
1895 static struct fragment *parse_binary_hunk(struct apply_state *state,
1896 char **buf_p,
1897 unsigned long *sz_p,
1898 int *status_p,
1899 int *used_p)
1902 * Expect a line that begins with binary patch method ("literal"
1903 * or "delta"), followed by the length of data before deflating.
1904 * a sequence of 'length-byte' followed by base-85 encoded data
1905 * should follow, terminated by a newline.
1907 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1908 * and we would limit the patch line to 66 characters,
1909 * so one line can fit up to 13 groups that would decode
1910 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1911 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1913 int llen, used;
1914 unsigned long size = *sz_p;
1915 char *buffer = *buf_p;
1916 int patch_method;
1917 unsigned long origlen;
1918 char *data = NULL;
1919 int hunk_size = 0;
1920 struct fragment *frag;
1922 llen = linelen(buffer, size);
1923 used = llen;
1925 *status_p = 0;
1927 if (starts_with(buffer, "delta ")) {
1928 patch_method = BINARY_DELTA_DEFLATED;
1929 origlen = strtoul(buffer + 6, NULL, 10);
1931 else if (starts_with(buffer, "literal ")) {
1932 patch_method = BINARY_LITERAL_DEFLATED;
1933 origlen = strtoul(buffer + 8, NULL, 10);
1935 else
1936 return NULL;
1938 state->linenr++;
1939 buffer += llen;
1940 while (1) {
1941 int byte_length, max_byte_length, newsize;
1942 llen = linelen(buffer, size);
1943 used += llen;
1944 state->linenr++;
1945 if (llen == 1) {
1946 /* consume the blank line */
1947 buffer++;
1948 size--;
1949 break;
1952 * Minimum line is "A00000\n" which is 7-byte long,
1953 * and the line length must be multiple of 5 plus 2.
1955 if ((llen < 7) || (llen-2) % 5)
1956 goto corrupt;
1957 max_byte_length = (llen - 2) / 5 * 4;
1958 byte_length = *buffer;
1959 if ('A' <= byte_length && byte_length <= 'Z')
1960 byte_length = byte_length - 'A' + 1;
1961 else if ('a' <= byte_length && byte_length <= 'z')
1962 byte_length = byte_length - 'a' + 27;
1963 else
1964 goto corrupt;
1965 /* if the input length was not multiple of 4, we would
1966 * have filler at the end but the filler should never
1967 * exceed 3 bytes
1969 if (max_byte_length < byte_length ||
1970 byte_length <= max_byte_length - 4)
1971 goto corrupt;
1972 newsize = hunk_size + byte_length;
1973 data = xrealloc(data, newsize);
1974 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1975 goto corrupt;
1976 hunk_size = newsize;
1977 buffer += llen;
1978 size -= llen;
1981 frag = xcalloc(1, sizeof(*frag));
1982 frag->patch = inflate_it(data, hunk_size, origlen);
1983 frag->free_patch = 1;
1984 if (!frag->patch)
1985 goto corrupt;
1986 free(data);
1987 frag->size = origlen;
1988 *buf_p = buffer;
1989 *sz_p = size;
1990 *used_p = used;
1991 frag->binary_patch_method = patch_method;
1992 return frag;
1994 corrupt:
1995 free(data);
1996 *status_p = -1;
1997 error(_("corrupt binary patch at line %d: %.*s"),
1998 state->linenr-1, llen-1, buffer);
1999 return NULL;
2003 * Returns:
2004 * -1 in case of error,
2005 * the length of the parsed binary patch otherwise
2007 static int parse_binary(struct apply_state *state,
2008 char *buffer,
2009 unsigned long size,
2010 struct patch *patch)
2013 * We have read "GIT binary patch\n"; what follows is a line
2014 * that says the patch method (currently, either "literal" or
2015 * "delta") and the length of data before deflating; a
2016 * sequence of 'length-byte' followed by base-85 encoded data
2017 * follows.
2019 * When a binary patch is reversible, there is another binary
2020 * hunk in the same format, starting with patch method (either
2021 * "literal" or "delta") with the length of data, and a sequence
2022 * of length-byte + base-85 encoded data, terminated with another
2023 * empty line. This data, when applied to the postimage, produces
2024 * the preimage.
2026 struct fragment *forward;
2027 struct fragment *reverse;
2028 int status;
2029 int used, used_1;
2031 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2032 if (!forward && !status)
2033 /* there has to be one hunk (forward hunk) */
2034 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2035 if (status)
2036 /* otherwise we already gave an error message */
2037 return status;
2039 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2040 if (reverse)
2041 used += used_1;
2042 else if (status) {
2044 * Not having reverse hunk is not an error, but having
2045 * a corrupt reverse hunk is.
2047 free((void*) forward->patch);
2048 free(forward);
2049 return status;
2051 forward->next = reverse;
2052 patch->fragments = forward;
2053 patch->is_binary = 1;
2054 return used;
2057 static void prefix_one(struct apply_state *state, char **name)
2059 char *old_name = *name;
2060 if (!old_name)
2061 return;
2062 *name = prefix_filename(state->prefix, *name);
2063 free(old_name);
2066 static void prefix_patch(struct apply_state *state, struct patch *p)
2068 if (!state->prefix || p->is_toplevel_relative)
2069 return;
2070 prefix_one(state, &p->new_name);
2071 prefix_one(state, &p->old_name);
2075 * include/exclude
2078 static void add_name_limit(struct apply_state *state,
2079 const char *name,
2080 int exclude)
2082 struct string_list_item *it;
2084 it = string_list_append(&state->limit_by_name, name);
2085 it->util = exclude ? NULL : (void *) 1;
2088 static int use_patch(struct apply_state *state, struct patch *p)
2090 const char *pathname = p->new_name ? p->new_name : p->old_name;
2091 int i;
2093 /* Paths outside are not touched regardless of "--include" */
2094 if (state->prefix && *state->prefix) {
2095 const char *rest;
2096 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2097 return 0;
2100 /* See if it matches any of exclude/include rule */
2101 for (i = 0; i < state->limit_by_name.nr; i++) {
2102 struct string_list_item *it = &state->limit_by_name.items[i];
2103 if (!wildmatch(it->string, pathname, 0))
2104 return (it->util != NULL);
2108 * If we had any include, a path that does not match any rule is
2109 * not used. Otherwise, we saw bunch of exclude rules (or none)
2110 * and such a path is used.
2112 return !state->has_include;
2116 * Read the patch text in "buffer" that extends for "size" bytes; stop
2117 * reading after seeing a single patch (i.e. changes to a single file).
2118 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2120 * Returns:
2121 * -1 if no header was found or parse_binary() failed,
2122 * -128 on another error,
2123 * the number of bytes consumed otherwise,
2124 * so that the caller can call us again for the next patch.
2126 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2128 int hdrsize, patchsize;
2129 int offset = find_header(state, buffer, size, &hdrsize, patch);
2131 if (offset < 0)
2132 return offset;
2134 prefix_patch(state, patch);
2136 if (!use_patch(state, patch))
2137 patch->ws_rule = 0;
2138 else if (patch->new_name)
2139 patch->ws_rule = whitespace_rule(state->repo->index,
2140 patch->new_name);
2141 else
2142 patch->ws_rule = whitespace_rule(state->repo->index,
2143 patch->old_name);
2145 patchsize = parse_single_patch(state,
2146 buffer + offset + hdrsize,
2147 size - offset - hdrsize,
2148 patch);
2150 if (patchsize < 0)
2151 return -128;
2153 if (!patchsize) {
2154 static const char git_binary[] = "GIT binary patch\n";
2155 int hd = hdrsize + offset;
2156 unsigned long llen = linelen(buffer + hd, size - hd);
2158 if (llen == sizeof(git_binary) - 1 &&
2159 !memcmp(git_binary, buffer + hd, llen)) {
2160 int used;
2161 state->linenr++;
2162 used = parse_binary(state, buffer + hd + llen,
2163 size - hd - llen, patch);
2164 if (used < 0)
2165 return -1;
2166 if (used)
2167 patchsize = used + llen;
2168 else
2169 patchsize = 0;
2171 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2172 static const char *binhdr[] = {
2173 "Binary files ",
2174 "Files ",
2175 NULL,
2177 int i;
2178 for (i = 0; binhdr[i]; i++) {
2179 int len = strlen(binhdr[i]);
2180 if (len < size - hd &&
2181 !memcmp(binhdr[i], buffer + hd, len)) {
2182 state->linenr++;
2183 patch->is_binary = 1;
2184 patchsize = llen;
2185 break;
2190 /* Empty patch cannot be applied if it is a text patch
2191 * without metadata change. A binary patch appears
2192 * empty to us here.
2194 if ((state->apply || state->check) &&
2195 (!patch->is_binary && !metadata_changes(patch))) {
2196 error(_("patch with only garbage at line %d"), state->linenr);
2197 return -128;
2201 return offset + hdrsize + patchsize;
2204 static void reverse_patches(struct patch *p)
2206 for (; p; p = p->next) {
2207 struct fragment *frag = p->fragments;
2209 SWAP(p->new_name, p->old_name);
2210 SWAP(p->new_mode, p->old_mode);
2211 SWAP(p->is_new, p->is_delete);
2212 SWAP(p->lines_added, p->lines_deleted);
2213 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2215 for (; frag; frag = frag->next) {
2216 SWAP(frag->newpos, frag->oldpos);
2217 SWAP(frag->newlines, frag->oldlines);
2222 static const char pluses[] =
2223 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2224 static const char minuses[]=
2225 "----------------------------------------------------------------------";
2227 static void show_stats(struct apply_state *state, struct patch *patch)
2229 struct strbuf qname = STRBUF_INIT;
2230 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2231 int max, add, del;
2233 quote_c_style(cp, &qname, NULL, 0);
2236 * "scale" the filename
2238 max = state->max_len;
2239 if (max > 50)
2240 max = 50;
2242 if (qname.len > max) {
2243 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2244 if (!cp)
2245 cp = qname.buf + qname.len + 3 - max;
2246 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2249 if (patch->is_binary) {
2250 printf(" %-*s | Bin\n", max, qname.buf);
2251 strbuf_release(&qname);
2252 return;
2255 printf(" %-*s |", max, qname.buf);
2256 strbuf_release(&qname);
2259 * scale the add/delete
2261 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2262 add = patch->lines_added;
2263 del = patch->lines_deleted;
2265 if (state->max_change > 0) {
2266 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2267 add = (add * max + state->max_change / 2) / state->max_change;
2268 del = total - add;
2270 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2271 add, pluses, del, minuses);
2274 static int read_old_data(struct stat *st, struct patch *patch,
2275 const char *path, struct strbuf *buf)
2277 int conv_flags = patch->crlf_in_old ?
2278 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2279 switch (st->st_mode & S_IFMT) {
2280 case S_IFLNK:
2281 if (strbuf_readlink(buf, path, st->st_size) < 0)
2282 return error(_("unable to read symlink %s"), path);
2283 return 0;
2284 case S_IFREG:
2285 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2286 return error(_("unable to open or read %s"), path);
2288 * "git apply" without "--index/--cached" should never look
2289 * at the index; the target file may not have been added to
2290 * the index yet, and we may not even be in any Git repository.
2291 * Pass NULL to convert_to_git() to stress this; the function
2292 * should never look at the index when explicit crlf option
2293 * is given.
2295 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2296 return 0;
2297 default:
2298 return -1;
2303 * Update the preimage, and the common lines in postimage,
2304 * from buffer buf of length len. If postlen is 0 the postimage
2305 * is updated in place, otherwise it's updated on a new buffer
2306 * of length postlen
2309 static void update_pre_post_images(struct image *preimage,
2310 struct image *postimage,
2311 char *buf,
2312 size_t len, size_t postlen)
2314 int i, ctx, reduced;
2315 char *new_buf, *old_buf, *fixed;
2316 struct image fixed_preimage;
2319 * Update the preimage with whitespace fixes. Note that we
2320 * are not losing preimage->buf -- apply_one_fragment() will
2321 * free "oldlines".
2323 prepare_image(&fixed_preimage, buf, len, 1);
2324 assert(postlen
2325 ? fixed_preimage.nr == preimage->nr
2326 : fixed_preimage.nr <= preimage->nr);
2327 for (i = 0; i < fixed_preimage.nr; i++)
2328 fixed_preimage.line[i].flag = preimage->line[i].flag;
2329 free(preimage->line_allocated);
2330 *preimage = fixed_preimage;
2333 * Adjust the common context lines in postimage. This can be
2334 * done in-place when we are shrinking it with whitespace
2335 * fixing, but needs a new buffer when ignoring whitespace or
2336 * expanding leading tabs to spaces.
2338 * We trust the caller to tell us if the update can be done
2339 * in place (postlen==0) or not.
2341 old_buf = postimage->buf;
2342 if (postlen)
2343 new_buf = postimage->buf = xmalloc(postlen);
2344 else
2345 new_buf = old_buf;
2346 fixed = preimage->buf;
2348 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2349 size_t l_len = postimage->line[i].len;
2350 if (!(postimage->line[i].flag & LINE_COMMON)) {
2351 /* an added line -- no counterparts in preimage */
2352 memmove(new_buf, old_buf, l_len);
2353 old_buf += l_len;
2354 new_buf += l_len;
2355 continue;
2358 /* a common context -- skip it in the original postimage */
2359 old_buf += l_len;
2361 /* and find the corresponding one in the fixed preimage */
2362 while (ctx < preimage->nr &&
2363 !(preimage->line[ctx].flag & LINE_COMMON)) {
2364 fixed += preimage->line[ctx].len;
2365 ctx++;
2369 * preimage is expected to run out, if the caller
2370 * fixed addition of trailing blank lines.
2372 if (preimage->nr <= ctx) {
2373 reduced++;
2374 continue;
2377 /* and copy it in, while fixing the line length */
2378 l_len = preimage->line[ctx].len;
2379 memcpy(new_buf, fixed, l_len);
2380 new_buf += l_len;
2381 fixed += l_len;
2382 postimage->line[i].len = l_len;
2383 ctx++;
2386 if (postlen
2387 ? postlen < new_buf - postimage->buf
2388 : postimage->len < new_buf - postimage->buf)
2389 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2390 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2392 /* Fix the length of the whole thing */
2393 postimage->len = new_buf - postimage->buf;
2394 postimage->nr -= reduced;
2397 static int line_by_line_fuzzy_match(struct image *img,
2398 struct image *preimage,
2399 struct image *postimage,
2400 unsigned long current,
2401 int current_lno,
2402 int preimage_limit)
2404 int i;
2405 size_t imgoff = 0;
2406 size_t preoff = 0;
2407 size_t postlen = postimage->len;
2408 size_t extra_chars;
2409 char *buf;
2410 char *preimage_eof;
2411 char *preimage_end;
2412 struct strbuf fixed;
2413 char *fixed_buf;
2414 size_t fixed_len;
2416 for (i = 0; i < preimage_limit; i++) {
2417 size_t prelen = preimage->line[i].len;
2418 size_t imglen = img->line[current_lno+i].len;
2420 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2421 preimage->buf + preoff, prelen))
2422 return 0;
2423 if (preimage->line[i].flag & LINE_COMMON)
2424 postlen += imglen - prelen;
2425 imgoff += imglen;
2426 preoff += prelen;
2430 * Ok, the preimage matches with whitespace fuzz.
2432 * imgoff now holds the true length of the target that
2433 * matches the preimage before the end of the file.
2435 * Count the number of characters in the preimage that fall
2436 * beyond the end of the file and make sure that all of them
2437 * are whitespace characters. (This can only happen if
2438 * we are removing blank lines at the end of the file.)
2440 buf = preimage_eof = preimage->buf + preoff;
2441 for ( ; i < preimage->nr; i++)
2442 preoff += preimage->line[i].len;
2443 preimage_end = preimage->buf + preoff;
2444 for ( ; buf < preimage_end; buf++)
2445 if (!isspace(*buf))
2446 return 0;
2449 * Update the preimage and the common postimage context
2450 * lines to use the same whitespace as the target.
2451 * If whitespace is missing in the target (i.e.
2452 * if the preimage extends beyond the end of the file),
2453 * use the whitespace from the preimage.
2455 extra_chars = preimage_end - preimage_eof;
2456 strbuf_init(&fixed, imgoff + extra_chars);
2457 strbuf_add(&fixed, img->buf + current, imgoff);
2458 strbuf_add(&fixed, preimage_eof, extra_chars);
2459 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2460 update_pre_post_images(preimage, postimage,
2461 fixed_buf, fixed_len, postlen);
2462 return 1;
2465 static int match_fragment(struct apply_state *state,
2466 struct image *img,
2467 struct image *preimage,
2468 struct image *postimage,
2469 unsigned long current,
2470 int current_lno,
2471 unsigned ws_rule,
2472 int match_beginning, int match_end)
2474 int i;
2475 char *fixed_buf, *buf, *orig, *target;
2476 struct strbuf fixed;
2477 size_t fixed_len, postlen;
2478 int preimage_limit;
2480 if (preimage->nr + current_lno <= img->nr) {
2482 * The hunk falls within the boundaries of img.
2484 preimage_limit = preimage->nr;
2485 if (match_end && (preimage->nr + current_lno != img->nr))
2486 return 0;
2487 } else if (state->ws_error_action == correct_ws_error &&
2488 (ws_rule & WS_BLANK_AT_EOF)) {
2490 * This hunk extends beyond the end of img, and we are
2491 * removing blank lines at the end of the file. This
2492 * many lines from the beginning of the preimage must
2493 * match with img, and the remainder of the preimage
2494 * must be blank.
2496 preimage_limit = img->nr - current_lno;
2497 } else {
2499 * The hunk extends beyond the end of the img and
2500 * we are not removing blanks at the end, so we
2501 * should reject the hunk at this position.
2503 return 0;
2506 if (match_beginning && current_lno)
2507 return 0;
2509 /* Quick hash check */
2510 for (i = 0; i < preimage_limit; i++)
2511 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2512 (preimage->line[i].hash != img->line[current_lno + i].hash))
2513 return 0;
2515 if (preimage_limit == preimage->nr) {
2517 * Do we have an exact match? If we were told to match
2518 * at the end, size must be exactly at current+fragsize,
2519 * otherwise current+fragsize must be still within the preimage,
2520 * and either case, the old piece should match the preimage
2521 * exactly.
2523 if ((match_end
2524 ? (current + preimage->len == img->len)
2525 : (current + preimage->len <= img->len)) &&
2526 !memcmp(img->buf + current, preimage->buf, preimage->len))
2527 return 1;
2528 } else {
2530 * The preimage extends beyond the end of img, so
2531 * there cannot be an exact match.
2533 * There must be one non-blank context line that match
2534 * a line before the end of img.
2536 char *buf_end;
2538 buf = preimage->buf;
2539 buf_end = buf;
2540 for (i = 0; i < preimage_limit; i++)
2541 buf_end += preimage->line[i].len;
2543 for ( ; buf < buf_end; buf++)
2544 if (!isspace(*buf))
2545 break;
2546 if (buf == buf_end)
2547 return 0;
2551 * No exact match. If we are ignoring whitespace, run a line-by-line
2552 * fuzzy matching. We collect all the line length information because
2553 * we need it to adjust whitespace if we match.
2555 if (state->ws_ignore_action == ignore_ws_change)
2556 return line_by_line_fuzzy_match(img, preimage, postimage,
2557 current, current_lno, preimage_limit);
2559 if (state->ws_error_action != correct_ws_error)
2560 return 0;
2563 * The hunk does not apply byte-by-byte, but the hash says
2564 * it might with whitespace fuzz. We weren't asked to
2565 * ignore whitespace, we were asked to correct whitespace
2566 * errors, so let's try matching after whitespace correction.
2568 * While checking the preimage against the target, whitespace
2569 * errors in both fixed, we count how large the corresponding
2570 * postimage needs to be. The postimage prepared by
2571 * apply_one_fragment() has whitespace errors fixed on added
2572 * lines already, but the common lines were propagated as-is,
2573 * which may become longer when their whitespace errors are
2574 * fixed.
2577 /* First count added lines in postimage */
2578 postlen = 0;
2579 for (i = 0; i < postimage->nr; i++) {
2580 if (!(postimage->line[i].flag & LINE_COMMON))
2581 postlen += postimage->line[i].len;
2585 * The preimage may extend beyond the end of the file,
2586 * but in this loop we will only handle the part of the
2587 * preimage that falls within the file.
2589 strbuf_init(&fixed, preimage->len + 1);
2590 orig = preimage->buf;
2591 target = img->buf + current;
2592 for (i = 0; i < preimage_limit; i++) {
2593 size_t oldlen = preimage->line[i].len;
2594 size_t tgtlen = img->line[current_lno + i].len;
2595 size_t fixstart = fixed.len;
2596 struct strbuf tgtfix;
2597 int match;
2599 /* Try fixing the line in the preimage */
2600 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2602 /* Try fixing the line in the target */
2603 strbuf_init(&tgtfix, tgtlen);
2604 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2607 * If they match, either the preimage was based on
2608 * a version before our tree fixed whitespace breakage,
2609 * or we are lacking a whitespace-fix patch the tree
2610 * the preimage was based on already had (i.e. target
2611 * has whitespace breakage, the preimage doesn't).
2612 * In either case, we are fixing the whitespace breakages
2613 * so we might as well take the fix together with their
2614 * real change.
2616 match = (tgtfix.len == fixed.len - fixstart &&
2617 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2618 fixed.len - fixstart));
2620 /* Add the length if this is common with the postimage */
2621 if (preimage->line[i].flag & LINE_COMMON)
2622 postlen += tgtfix.len;
2624 strbuf_release(&tgtfix);
2625 if (!match)
2626 goto unmatch_exit;
2628 orig += oldlen;
2629 target += tgtlen;
2634 * Now handle the lines in the preimage that falls beyond the
2635 * end of the file (if any). They will only match if they are
2636 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2637 * false).
2639 for ( ; i < preimage->nr; i++) {
2640 size_t fixstart = fixed.len; /* start of the fixed preimage */
2641 size_t oldlen = preimage->line[i].len;
2642 int j;
2644 /* Try fixing the line in the preimage */
2645 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2647 for (j = fixstart; j < fixed.len; j++)
2648 if (!isspace(fixed.buf[j]))
2649 goto unmatch_exit;
2651 orig += oldlen;
2655 * Yes, the preimage is based on an older version that still
2656 * has whitespace breakages unfixed, and fixing them makes the
2657 * hunk match. Update the context lines in the postimage.
2659 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2660 if (postlen < postimage->len)
2661 postlen = 0;
2662 update_pre_post_images(preimage, postimage,
2663 fixed_buf, fixed_len, postlen);
2664 return 1;
2666 unmatch_exit:
2667 strbuf_release(&fixed);
2668 return 0;
2671 static int find_pos(struct apply_state *state,
2672 struct image *img,
2673 struct image *preimage,
2674 struct image *postimage,
2675 int line,
2676 unsigned ws_rule,
2677 int match_beginning, int match_end)
2679 int i;
2680 unsigned long backwards, forwards, current;
2681 int backwards_lno, forwards_lno, current_lno;
2684 * If match_beginning or match_end is specified, there is no
2685 * point starting from a wrong line that will never match and
2686 * wander around and wait for a match at the specified end.
2688 if (match_beginning)
2689 line = 0;
2690 else if (match_end)
2691 line = img->nr - preimage->nr;
2694 * Because the comparison is unsigned, the following test
2695 * will also take care of a negative line number that can
2696 * result when match_end and preimage is larger than the target.
2698 if ((size_t) line > img->nr)
2699 line = img->nr;
2701 current = 0;
2702 for (i = 0; i < line; i++)
2703 current += img->line[i].len;
2706 * There's probably some smart way to do this, but I'll leave
2707 * that to the smart and beautiful people. I'm simple and stupid.
2709 backwards = current;
2710 backwards_lno = line;
2711 forwards = current;
2712 forwards_lno = line;
2713 current_lno = line;
2715 for (i = 0; ; i++) {
2716 if (match_fragment(state, img, preimage, postimage,
2717 current, current_lno, ws_rule,
2718 match_beginning, match_end))
2719 return current_lno;
2721 again:
2722 if (backwards_lno == 0 && forwards_lno == img->nr)
2723 break;
2725 if (i & 1) {
2726 if (backwards_lno == 0) {
2727 i++;
2728 goto again;
2730 backwards_lno--;
2731 backwards -= img->line[backwards_lno].len;
2732 current = backwards;
2733 current_lno = backwards_lno;
2734 } else {
2735 if (forwards_lno == img->nr) {
2736 i++;
2737 goto again;
2739 forwards += img->line[forwards_lno].len;
2740 forwards_lno++;
2741 current = forwards;
2742 current_lno = forwards_lno;
2746 return -1;
2749 static void remove_first_line(struct image *img)
2751 img->buf += img->line[0].len;
2752 img->len -= img->line[0].len;
2753 img->line++;
2754 img->nr--;
2757 static void remove_last_line(struct image *img)
2759 img->len -= img->line[--img->nr].len;
2763 * The change from "preimage" and "postimage" has been found to
2764 * apply at applied_pos (counts in line numbers) in "img".
2765 * Update "img" to remove "preimage" and replace it with "postimage".
2767 static void update_image(struct apply_state *state,
2768 struct image *img,
2769 int applied_pos,
2770 struct image *preimage,
2771 struct image *postimage)
2774 * remove the copy of preimage at offset in img
2775 * and replace it with postimage
2777 int i, nr;
2778 size_t remove_count, insert_count, applied_at = 0;
2779 char *result;
2780 int preimage_limit;
2783 * If we are removing blank lines at the end of img,
2784 * the preimage may extend beyond the end.
2785 * If that is the case, we must be careful only to
2786 * remove the part of the preimage that falls within
2787 * the boundaries of img. Initialize preimage_limit
2788 * to the number of lines in the preimage that falls
2789 * within the boundaries.
2791 preimage_limit = preimage->nr;
2792 if (preimage_limit > img->nr - applied_pos)
2793 preimage_limit = img->nr - applied_pos;
2795 for (i = 0; i < applied_pos; i++)
2796 applied_at += img->line[i].len;
2798 remove_count = 0;
2799 for (i = 0; i < preimage_limit; i++)
2800 remove_count += img->line[applied_pos + i].len;
2801 insert_count = postimage->len;
2803 /* Adjust the contents */
2804 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2805 memcpy(result, img->buf, applied_at);
2806 memcpy(result + applied_at, postimage->buf, postimage->len);
2807 memcpy(result + applied_at + postimage->len,
2808 img->buf + (applied_at + remove_count),
2809 img->len - (applied_at + remove_count));
2810 free(img->buf);
2811 img->buf = result;
2812 img->len += insert_count - remove_count;
2813 result[img->len] = '\0';
2815 /* Adjust the line table */
2816 nr = img->nr + postimage->nr - preimage_limit;
2817 if (preimage_limit < postimage->nr) {
2819 * NOTE: this knows that we never call remove_first_line()
2820 * on anything other than pre/post image.
2822 REALLOC_ARRAY(img->line, nr);
2823 img->line_allocated = img->line;
2825 if (preimage_limit != postimage->nr)
2826 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2827 img->line + applied_pos + preimage_limit,
2828 img->nr - (applied_pos + preimage_limit));
2829 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2830 if (!state->allow_overlap)
2831 for (i = 0; i < postimage->nr; i++)
2832 img->line[applied_pos + i].flag |= LINE_PATCHED;
2833 img->nr = nr;
2837 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2838 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2839 * replace the part of "img" with "postimage" text.
2841 static int apply_one_fragment(struct apply_state *state,
2842 struct image *img, struct fragment *frag,
2843 int inaccurate_eof, unsigned ws_rule,
2844 int nth_fragment)
2846 int match_beginning, match_end;
2847 const char *patch = frag->patch;
2848 int size = frag->size;
2849 char *old, *oldlines;
2850 struct strbuf newlines;
2851 int new_blank_lines_at_end = 0;
2852 int found_new_blank_lines_at_end = 0;
2853 int hunk_linenr = frag->linenr;
2854 unsigned long leading, trailing;
2855 int pos, applied_pos;
2856 struct image preimage;
2857 struct image postimage;
2859 memset(&preimage, 0, sizeof(preimage));
2860 memset(&postimage, 0, sizeof(postimage));
2861 oldlines = xmalloc(size);
2862 strbuf_init(&newlines, size);
2864 old = oldlines;
2865 while (size > 0) {
2866 char first;
2867 int len = linelen(patch, size);
2868 int plen;
2869 int added_blank_line = 0;
2870 int is_blank_context = 0;
2871 size_t start;
2873 if (!len)
2874 break;
2877 * "plen" is how much of the line we should use for
2878 * the actual patch data. Normally we just remove the
2879 * first character on the line, but if the line is
2880 * followed by "\ No newline", then we also remove the
2881 * last one (which is the newline, of course).
2883 plen = len - 1;
2884 if (len < size && patch[len] == '\\')
2885 plen--;
2886 first = *patch;
2887 if (state->apply_in_reverse) {
2888 if (first == '-')
2889 first = '+';
2890 else if (first == '+')
2891 first = '-';
2894 switch (first) {
2895 case '\n':
2896 /* Newer GNU diff, empty context line */
2897 if (plen < 0)
2898 /* ... followed by '\No newline'; nothing */
2899 break;
2900 *old++ = '\n';
2901 strbuf_addch(&newlines, '\n');
2902 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2903 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2904 is_blank_context = 1;
2905 break;
2906 case ' ':
2907 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2908 ws_blank_line(patch + 1, plen, ws_rule))
2909 is_blank_context = 1;
2910 /* fallthrough */
2911 case '-':
2912 memcpy(old, patch + 1, plen);
2913 add_line_info(&preimage, old, plen,
2914 (first == ' ' ? LINE_COMMON : 0));
2915 old += plen;
2916 if (first == '-')
2917 break;
2918 /* fallthrough */
2919 case '+':
2920 /* --no-add does not add new lines */
2921 if (first == '+' && state->no_add)
2922 break;
2924 start = newlines.len;
2925 if (first != '+' ||
2926 !state->whitespace_error ||
2927 state->ws_error_action != correct_ws_error) {
2928 strbuf_add(&newlines, patch + 1, plen);
2930 else {
2931 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2933 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2934 (first == '+' ? 0 : LINE_COMMON));
2935 if (first == '+' &&
2936 (ws_rule & WS_BLANK_AT_EOF) &&
2937 ws_blank_line(patch + 1, plen, ws_rule))
2938 added_blank_line = 1;
2939 break;
2940 case '@': case '\\':
2941 /* Ignore it, we already handled it */
2942 break;
2943 default:
2944 if (state->apply_verbosity > verbosity_normal)
2945 error(_("invalid start of line: '%c'"), first);
2946 applied_pos = -1;
2947 goto out;
2949 if (added_blank_line) {
2950 if (!new_blank_lines_at_end)
2951 found_new_blank_lines_at_end = hunk_linenr;
2952 new_blank_lines_at_end++;
2954 else if (is_blank_context)
2956 else
2957 new_blank_lines_at_end = 0;
2958 patch += len;
2959 size -= len;
2960 hunk_linenr++;
2962 if (inaccurate_eof &&
2963 old > oldlines && old[-1] == '\n' &&
2964 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2965 old--;
2966 strbuf_setlen(&newlines, newlines.len - 1);
2967 preimage.line_allocated[preimage.nr - 1].len--;
2968 postimage.line_allocated[postimage.nr - 1].len--;
2971 leading = frag->leading;
2972 trailing = frag->trailing;
2975 * A hunk to change lines at the beginning would begin with
2976 * @@ -1,L +N,M @@
2977 * but we need to be careful. -U0 that inserts before the second
2978 * line also has this pattern.
2980 * And a hunk to add to an empty file would begin with
2981 * @@ -0,0 +N,M @@
2983 * In other words, a hunk that is (frag->oldpos <= 1) with or
2984 * without leading context must match at the beginning.
2986 match_beginning = (!frag->oldpos ||
2987 (frag->oldpos == 1 && !state->unidiff_zero));
2990 * A hunk without trailing lines must match at the end.
2991 * However, we simply cannot tell if a hunk must match end
2992 * from the lack of trailing lines if the patch was generated
2993 * with unidiff without any context.
2995 match_end = !state->unidiff_zero && !trailing;
2997 pos = frag->newpos ? (frag->newpos - 1) : 0;
2998 preimage.buf = oldlines;
2999 preimage.len = old - oldlines;
3000 postimage.buf = newlines.buf;
3001 postimage.len = newlines.len;
3002 preimage.line = preimage.line_allocated;
3003 postimage.line = postimage.line_allocated;
3005 for (;;) {
3007 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3008 ws_rule, match_beginning, match_end);
3010 if (applied_pos >= 0)
3011 break;
3013 /* Am I at my context limits? */
3014 if ((leading <= state->p_context) && (trailing <= state->p_context))
3015 break;
3016 if (match_beginning || match_end) {
3017 match_beginning = match_end = 0;
3018 continue;
3022 * Reduce the number of context lines; reduce both
3023 * leading and trailing if they are equal otherwise
3024 * just reduce the larger context.
3026 if (leading >= trailing) {
3027 remove_first_line(&preimage);
3028 remove_first_line(&postimage);
3029 pos--;
3030 leading--;
3032 if (trailing > leading) {
3033 remove_last_line(&preimage);
3034 remove_last_line(&postimage);
3035 trailing--;
3039 if (applied_pos >= 0) {
3040 if (new_blank_lines_at_end &&
3041 preimage.nr + applied_pos >= img->nr &&
3042 (ws_rule & WS_BLANK_AT_EOF) &&
3043 state->ws_error_action != nowarn_ws_error) {
3044 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3045 found_new_blank_lines_at_end);
3046 if (state->ws_error_action == correct_ws_error) {
3047 while (new_blank_lines_at_end--)
3048 remove_last_line(&postimage);
3051 * We would want to prevent write_out_results()
3052 * from taking place in apply_patch() that follows
3053 * the callchain led us here, which is:
3054 * apply_patch->check_patch_list->check_patch->
3055 * apply_data->apply_fragments->apply_one_fragment
3057 if (state->ws_error_action == die_on_ws_error)
3058 state->apply = 0;
3061 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3062 int offset = applied_pos - pos;
3063 if (state->apply_in_reverse)
3064 offset = 0 - offset;
3065 fprintf_ln(stderr,
3066 Q_("Hunk #%d succeeded at %d (offset %d line).",
3067 "Hunk #%d succeeded at %d (offset %d lines).",
3068 offset),
3069 nth_fragment, applied_pos + 1, offset);
3073 * Warn if it was necessary to reduce the number
3074 * of context lines.
3076 if ((leading != frag->leading ||
3077 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3078 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3079 " to apply fragment at %d"),
3080 leading, trailing, applied_pos+1);
3081 update_image(state, img, applied_pos, &preimage, &postimage);
3082 } else {
3083 if (state->apply_verbosity > verbosity_normal)
3084 error(_("while searching for:\n%.*s"),
3085 (int)(old - oldlines), oldlines);
3088 out:
3089 free(oldlines);
3090 strbuf_release(&newlines);
3091 free(preimage.line_allocated);
3092 free(postimage.line_allocated);
3094 return (applied_pos < 0);
3097 static int apply_binary_fragment(struct apply_state *state,
3098 struct image *img,
3099 struct patch *patch)
3101 struct fragment *fragment = patch->fragments;
3102 unsigned long len;
3103 void *dst;
3105 if (!fragment)
3106 return error(_("missing binary patch data for '%s'"),
3107 patch->new_name ?
3108 patch->new_name :
3109 patch->old_name);
3111 /* Binary patch is irreversible without the optional second hunk */
3112 if (state->apply_in_reverse) {
3113 if (!fragment->next)
3114 return error(_("cannot reverse-apply a binary patch "
3115 "without the reverse hunk to '%s'"),
3116 patch->new_name
3117 ? patch->new_name : patch->old_name);
3118 fragment = fragment->next;
3120 switch (fragment->binary_patch_method) {
3121 case BINARY_DELTA_DEFLATED:
3122 dst = patch_delta(img->buf, img->len, fragment->patch,
3123 fragment->size, &len);
3124 if (!dst)
3125 return -1;
3126 clear_image(img);
3127 img->buf = dst;
3128 img->len = len;
3129 return 0;
3130 case BINARY_LITERAL_DEFLATED:
3131 clear_image(img);
3132 img->len = fragment->size;
3133 img->buf = xmemdupz(fragment->patch, img->len);
3134 return 0;
3136 return -1;
3140 * Replace "img" with the result of applying the binary patch.
3141 * The binary patch data itself in patch->fragment is still kept
3142 * but the preimage prepared by the caller in "img" is freed here
3143 * or in the helper function apply_binary_fragment() this calls.
3145 static int apply_binary(struct apply_state *state,
3146 struct image *img,
3147 struct patch *patch)
3149 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3150 struct object_id oid;
3151 const unsigned hexsz = the_hash_algo->hexsz;
3154 * For safety, we require patch index line to contain
3155 * full hex textual object ID for old and new, at least for now.
3157 if (strlen(patch->old_oid_prefix) != hexsz ||
3158 strlen(patch->new_oid_prefix) != hexsz ||
3159 get_oid_hex(patch->old_oid_prefix, &oid) ||
3160 get_oid_hex(patch->new_oid_prefix, &oid))
3161 return error(_("cannot apply binary patch to '%s' "
3162 "without full index line"), name);
3164 if (patch->old_name) {
3166 * See if the old one matches what the patch
3167 * applies to.
3169 hash_object_file(img->buf, img->len, blob_type, &oid);
3170 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3171 return error(_("the patch applies to '%s' (%s), "
3172 "which does not match the "
3173 "current contents."),
3174 name, oid_to_hex(&oid));
3176 else {
3177 /* Otherwise, the old one must be empty. */
3178 if (img->len)
3179 return error(_("the patch applies to an empty "
3180 "'%s' but it is not empty"), name);
3183 get_oid_hex(patch->new_oid_prefix, &oid);
3184 if (is_null_oid(&oid)) {
3185 clear_image(img);
3186 return 0; /* deletion patch */
3189 if (has_object_file(&oid)) {
3190 /* We already have the postimage */
3191 enum object_type type;
3192 unsigned long size;
3193 char *result;
3195 result = read_object_file(&oid, &type, &size);
3196 if (!result)
3197 return error(_("the necessary postimage %s for "
3198 "'%s' cannot be read"),
3199 patch->new_oid_prefix, name);
3200 clear_image(img);
3201 img->buf = result;
3202 img->len = size;
3203 } else {
3205 * We have verified buf matches the preimage;
3206 * apply the patch data to it, which is stored
3207 * in the patch->fragments->{patch,size}.
3209 if (apply_binary_fragment(state, img, patch))
3210 return error(_("binary patch does not apply to '%s'"),
3211 name);
3213 /* verify that the result matches */
3214 hash_object_file(img->buf, img->len, blob_type, &oid);
3215 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3216 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3217 name, patch->new_oid_prefix, oid_to_hex(&oid));
3220 return 0;
3223 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3225 struct fragment *frag = patch->fragments;
3226 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3227 unsigned ws_rule = patch->ws_rule;
3228 unsigned inaccurate_eof = patch->inaccurate_eof;
3229 int nth = 0;
3231 if (patch->is_binary)
3232 return apply_binary(state, img, patch);
3234 while (frag) {
3235 nth++;
3236 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3237 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3238 if (!state->apply_with_reject)
3239 return -1;
3240 frag->rejected = 1;
3242 frag = frag->next;
3244 return 0;
3247 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3249 if (S_ISGITLINK(mode)) {
3250 strbuf_grow(buf, 100);
3251 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3252 } else {
3253 enum object_type type;
3254 unsigned long sz;
3255 char *result;
3257 result = read_object_file(oid, &type, &sz);
3258 if (!result)
3259 return -1;
3260 /* XXX read_sha1_file NUL-terminates */
3261 strbuf_attach(buf, result, sz, sz + 1);
3263 return 0;
3266 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3268 if (!ce)
3269 return 0;
3270 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3273 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3275 struct string_list_item *item;
3277 if (name == NULL)
3278 return NULL;
3280 item = string_list_lookup(&state->fn_table, name);
3281 if (item != NULL)
3282 return (struct patch *)item->util;
3284 return NULL;
3288 * item->util in the filename table records the status of the path.
3289 * Usually it points at a patch (whose result records the contents
3290 * of it after applying it), but it could be PATH_WAS_DELETED for a
3291 * path that a previously applied patch has already removed, or
3292 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3294 * The latter is needed to deal with a case where two paths A and B
3295 * are swapped by first renaming A to B and then renaming B to A;
3296 * moving A to B should not be prevented due to presence of B as we
3297 * will remove it in a later patch.
3299 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3300 #define PATH_WAS_DELETED ((struct patch *) -1)
3302 static int to_be_deleted(struct patch *patch)
3304 return patch == PATH_TO_BE_DELETED;
3307 static int was_deleted(struct patch *patch)
3309 return patch == PATH_WAS_DELETED;
3312 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3314 struct string_list_item *item;
3317 * Always add new_name unless patch is a deletion
3318 * This should cover the cases for normal diffs,
3319 * file creations and copies
3321 if (patch->new_name != NULL) {
3322 item = string_list_insert(&state->fn_table, patch->new_name);
3323 item->util = patch;
3327 * store a failure on rename/deletion cases because
3328 * later chunks shouldn't patch old names
3330 if ((patch->new_name == NULL) || (patch->is_rename)) {
3331 item = string_list_insert(&state->fn_table, patch->old_name);
3332 item->util = PATH_WAS_DELETED;
3336 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3339 * store information about incoming file deletion
3341 while (patch) {
3342 if ((patch->new_name == NULL) || (patch->is_rename)) {
3343 struct string_list_item *item;
3344 item = string_list_insert(&state->fn_table, patch->old_name);
3345 item->util = PATH_TO_BE_DELETED;
3347 patch = patch->next;
3351 static int checkout_target(struct index_state *istate,
3352 struct cache_entry *ce, struct stat *st)
3354 struct checkout costate = CHECKOUT_INIT;
3356 costate.refresh_cache = 1;
3357 costate.istate = istate;
3358 if (checkout_entry(ce, &costate, NULL, NULL) ||
3359 lstat(ce->name, st))
3360 return error(_("cannot checkout %s"), ce->name);
3361 return 0;
3364 static struct patch *previous_patch(struct apply_state *state,
3365 struct patch *patch,
3366 int *gone)
3368 struct patch *previous;
3370 *gone = 0;
3371 if (patch->is_copy || patch->is_rename)
3372 return NULL; /* "git" patches do not depend on the order */
3374 previous = in_fn_table(state, patch->old_name);
3375 if (!previous)
3376 return NULL;
3378 if (to_be_deleted(previous))
3379 return NULL; /* the deletion hasn't happened yet */
3381 if (was_deleted(previous))
3382 *gone = 1;
3384 return previous;
3387 static int verify_index_match(struct apply_state *state,
3388 const struct cache_entry *ce,
3389 struct stat *st)
3391 if (S_ISGITLINK(ce->ce_mode)) {
3392 if (!S_ISDIR(st->st_mode))
3393 return -1;
3394 return 0;
3396 return ie_match_stat(state->repo->index, ce, st,
3397 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3400 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3402 static int load_patch_target(struct apply_state *state,
3403 struct strbuf *buf,
3404 const struct cache_entry *ce,
3405 struct stat *st,
3406 struct patch *patch,
3407 const char *name,
3408 unsigned expected_mode)
3410 if (state->cached || state->check_index) {
3411 if (read_file_or_gitlink(ce, buf))
3412 return error(_("failed to read %s"), name);
3413 } else if (name) {
3414 if (S_ISGITLINK(expected_mode)) {
3415 if (ce)
3416 return read_file_or_gitlink(ce, buf);
3417 else
3418 return SUBMODULE_PATCH_WITHOUT_INDEX;
3419 } else if (has_symlink_leading_path(name, strlen(name))) {
3420 return error(_("reading from '%s' beyond a symbolic link"), name);
3421 } else {
3422 if (read_old_data(st, patch, name, buf))
3423 return error(_("failed to read %s"), name);
3426 return 0;
3430 * We are about to apply "patch"; populate the "image" with the
3431 * current version we have, from the working tree or from the index,
3432 * depending on the situation e.g. --cached/--index. If we are
3433 * applying a non-git patch that incrementally updates the tree,
3434 * we read from the result of a previous diff.
3436 static int load_preimage(struct apply_state *state,
3437 struct image *image,
3438 struct patch *patch, struct stat *st,
3439 const struct cache_entry *ce)
3441 struct strbuf buf = STRBUF_INIT;
3442 size_t len;
3443 char *img;
3444 struct patch *previous;
3445 int status;
3447 previous = previous_patch(state, patch, &status);
3448 if (status)
3449 return error(_("path %s has been renamed/deleted"),
3450 patch->old_name);
3451 if (previous) {
3452 /* We have a patched copy in memory; use that. */
3453 strbuf_add(&buf, previous->result, previous->resultsize);
3454 } else {
3455 status = load_patch_target(state, &buf, ce, st, patch,
3456 patch->old_name, patch->old_mode);
3457 if (status < 0)
3458 return status;
3459 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3461 * There is no way to apply subproject
3462 * patch without looking at the index.
3463 * NEEDSWORK: shouldn't this be flagged
3464 * as an error???
3466 free_fragment_list(patch->fragments);
3467 patch->fragments = NULL;
3468 } else if (status) {
3469 return error(_("failed to read %s"), patch->old_name);
3473 img = strbuf_detach(&buf, &len);
3474 prepare_image(image, img, len, !patch->is_binary);
3475 return 0;
3478 static int three_way_merge(struct apply_state *state,
3479 struct image *image,
3480 char *path,
3481 const struct object_id *base,
3482 const struct object_id *ours,
3483 const struct object_id *theirs)
3485 mmfile_t base_file, our_file, their_file;
3486 mmbuffer_t result = { NULL };
3487 int status;
3489 read_mmblob(&base_file, base);
3490 read_mmblob(&our_file, ours);
3491 read_mmblob(&their_file, theirs);
3492 status = ll_merge(&result, path,
3493 &base_file, "base",
3494 &our_file, "ours",
3495 &their_file, "theirs",
3496 state->repo->index,
3497 NULL);
3498 free(base_file.ptr);
3499 free(our_file.ptr);
3500 free(their_file.ptr);
3501 if (status < 0 || !result.ptr) {
3502 free(result.ptr);
3503 return -1;
3505 clear_image(image);
3506 image->buf = result.ptr;
3507 image->len = result.size;
3509 return status;
3513 * When directly falling back to add/add three-way merge, we read from
3514 * the current contents of the new_name. In no cases other than that
3515 * this function will be called.
3517 static int load_current(struct apply_state *state,
3518 struct image *image,
3519 struct patch *patch)
3521 struct strbuf buf = STRBUF_INIT;
3522 int status, pos;
3523 size_t len;
3524 char *img;
3525 struct stat st;
3526 struct cache_entry *ce;
3527 char *name = patch->new_name;
3528 unsigned mode = patch->new_mode;
3530 if (!patch->is_new)
3531 BUG("patch to %s is not a creation", patch->old_name);
3533 pos = index_name_pos(state->repo->index, name, strlen(name));
3534 if (pos < 0)
3535 return error(_("%s: does not exist in index"), name);
3536 ce = state->repo->index->cache[pos];
3537 if (lstat(name, &st)) {
3538 if (errno != ENOENT)
3539 return error_errno("%s", name);
3540 if (checkout_target(state->repo->index, ce, &st))
3541 return -1;
3543 if (verify_index_match(state, ce, &st))
3544 return error(_("%s: does not match index"), name);
3546 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3547 if (status < 0)
3548 return status;
3549 else if (status)
3550 return -1;
3551 img = strbuf_detach(&buf, &len);
3552 prepare_image(image, img, len, !patch->is_binary);
3553 return 0;
3556 static int try_threeway(struct apply_state *state,
3557 struct image *image,
3558 struct patch *patch,
3559 struct stat *st,
3560 const struct cache_entry *ce)
3562 struct object_id pre_oid, post_oid, our_oid;
3563 struct strbuf buf = STRBUF_INIT;
3564 size_t len;
3565 int status;
3566 char *img;
3567 struct image tmp_image;
3569 /* No point falling back to 3-way merge in these cases */
3570 if (patch->is_delete ||
3571 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3572 return -1;
3574 /* Preimage the patch was prepared for */
3575 if (patch->is_new)
3576 write_object_file("", 0, blob_type, &pre_oid);
3577 else if (get_oid(patch->old_oid_prefix, &pre_oid) ||
3578 read_blob_object(&buf, &pre_oid, patch->old_mode))
3579 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3581 if (state->apply_verbosity > verbosity_silent)
3582 fprintf(stderr, _("Falling back to three-way merge...\n"));
3584 img = strbuf_detach(&buf, &len);
3585 prepare_image(&tmp_image, img, len, 1);
3586 /* Apply the patch to get the post image */
3587 if (apply_fragments(state, &tmp_image, patch) < 0) {
3588 clear_image(&tmp_image);
3589 return -1;
3591 /* post_oid is theirs */
3592 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &post_oid);
3593 clear_image(&tmp_image);
3595 /* our_oid is ours */
3596 if (patch->is_new) {
3597 if (load_current(state, &tmp_image, patch))
3598 return error(_("cannot read the current contents of '%s'"),
3599 patch->new_name);
3600 } else {
3601 if (load_preimage(state, &tmp_image, patch, st, ce))
3602 return error(_("cannot read the current contents of '%s'"),
3603 patch->old_name);
3605 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &our_oid);
3606 clear_image(&tmp_image);
3608 /* in-core three-way merge between post and our using pre as base */
3609 status = three_way_merge(state, image, patch->new_name,
3610 &pre_oid, &our_oid, &post_oid);
3611 if (status < 0) {
3612 if (state->apply_verbosity > verbosity_silent)
3613 fprintf(stderr,
3614 _("Failed to fall back on three-way merge...\n"));
3615 return status;
3618 if (status) {
3619 patch->conflicted_threeway = 1;
3620 if (patch->is_new)
3621 oidclr(&patch->threeway_stage[0]);
3622 else
3623 oidcpy(&patch->threeway_stage[0], &pre_oid);
3624 oidcpy(&patch->threeway_stage[1], &our_oid);
3625 oidcpy(&patch->threeway_stage[2], &post_oid);
3626 if (state->apply_verbosity > verbosity_silent)
3627 fprintf(stderr,
3628 _("Applied patch to '%s' with conflicts.\n"),
3629 patch->new_name);
3630 } else {
3631 if (state->apply_verbosity > verbosity_silent)
3632 fprintf(stderr,
3633 _("Applied patch to '%s' cleanly.\n"),
3634 patch->new_name);
3636 return 0;
3639 static int apply_data(struct apply_state *state, struct patch *patch,
3640 struct stat *st, const struct cache_entry *ce)
3642 struct image image;
3644 if (load_preimage(state, &image, patch, st, ce) < 0)
3645 return -1;
3647 if (patch->direct_to_threeway ||
3648 apply_fragments(state, &image, patch) < 0) {
3649 /* Note: with --reject, apply_fragments() returns 0 */
3650 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3651 return -1;
3653 patch->result = image.buf;
3654 patch->resultsize = image.len;
3655 add_to_fn_table(state, patch);
3656 free(image.line_allocated);
3658 if (0 < patch->is_delete && patch->resultsize)
3659 return error(_("removal patch leaves file contents"));
3661 return 0;
3665 * If "patch" that we are looking at modifies or deletes what we have,
3666 * we would want it not to lose any local modification we have, either
3667 * in the working tree or in the index.
3669 * This also decides if a non-git patch is a creation patch or a
3670 * modification to an existing empty file. We do not check the state
3671 * of the current tree for a creation patch in this function; the caller
3672 * check_patch() separately makes sure (and errors out otherwise) that
3673 * the path the patch creates does not exist in the current tree.
3675 static int check_preimage(struct apply_state *state,
3676 struct patch *patch,
3677 struct cache_entry **ce,
3678 struct stat *st)
3680 const char *old_name = patch->old_name;
3681 struct patch *previous = NULL;
3682 int stat_ret = 0, status;
3683 unsigned st_mode = 0;
3685 if (!old_name)
3686 return 0;
3688 assert(patch->is_new <= 0);
3689 previous = previous_patch(state, patch, &status);
3691 if (status)
3692 return error(_("path %s has been renamed/deleted"), old_name);
3693 if (previous) {
3694 st_mode = previous->new_mode;
3695 } else if (!state->cached) {
3696 stat_ret = lstat(old_name, st);
3697 if (stat_ret && errno != ENOENT)
3698 return error_errno("%s", old_name);
3701 if (state->check_index && !previous) {
3702 int pos = index_name_pos(state->repo->index, old_name,
3703 strlen(old_name));
3704 if (pos < 0) {
3705 if (patch->is_new < 0)
3706 goto is_new;
3707 return error(_("%s: does not exist in index"), old_name);
3709 *ce = state->repo->index->cache[pos];
3710 if (stat_ret < 0) {
3711 if (checkout_target(state->repo->index, *ce, st))
3712 return -1;
3714 if (!state->cached && verify_index_match(state, *ce, st))
3715 return error(_("%s: does not match index"), old_name);
3716 if (state->cached)
3717 st_mode = (*ce)->ce_mode;
3718 } else if (stat_ret < 0) {
3719 if (patch->is_new < 0)
3720 goto is_new;
3721 return error_errno("%s", old_name);
3724 if (!state->cached && !previous)
3725 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3727 if (patch->is_new < 0)
3728 patch->is_new = 0;
3729 if (!patch->old_mode)
3730 patch->old_mode = st_mode;
3731 if ((st_mode ^ patch->old_mode) & S_IFMT)
3732 return error(_("%s: wrong type"), old_name);
3733 if (st_mode != patch->old_mode)
3734 warning(_("%s has type %o, expected %o"),
3735 old_name, st_mode, patch->old_mode);
3736 if (!patch->new_mode && !patch->is_delete)
3737 patch->new_mode = st_mode;
3738 return 0;
3740 is_new:
3741 patch->is_new = 1;
3742 patch->is_delete = 0;
3743 FREE_AND_NULL(patch->old_name);
3744 return 0;
3748 #define EXISTS_IN_INDEX 1
3749 #define EXISTS_IN_WORKTREE 2
3751 static int check_to_create(struct apply_state *state,
3752 const char *new_name,
3753 int ok_if_exists)
3755 struct stat nst;
3757 if (state->check_index &&
3758 index_name_pos(state->repo->index, new_name, strlen(new_name)) >= 0 &&
3759 !ok_if_exists)
3760 return EXISTS_IN_INDEX;
3761 if (state->cached)
3762 return 0;
3764 if (!lstat(new_name, &nst)) {
3765 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3766 return 0;
3768 * A leading component of new_name might be a symlink
3769 * that is going to be removed with this patch, but
3770 * still pointing at somewhere that has the path.
3771 * In such a case, path "new_name" does not exist as
3772 * far as git is concerned.
3774 if (has_symlink_leading_path(new_name, strlen(new_name)))
3775 return 0;
3777 return EXISTS_IN_WORKTREE;
3778 } else if (!is_missing_file_error(errno)) {
3779 return error_errno("%s", new_name);
3781 return 0;
3784 static uintptr_t register_symlink_changes(struct apply_state *state,
3785 const char *path,
3786 uintptr_t what)
3788 struct string_list_item *ent;
3790 ent = string_list_lookup(&state->symlink_changes, path);
3791 if (!ent) {
3792 ent = string_list_insert(&state->symlink_changes, path);
3793 ent->util = (void *)0;
3795 ent->util = (void *)(what | ((uintptr_t)ent->util));
3796 return (uintptr_t)ent->util;
3799 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3801 struct string_list_item *ent;
3803 ent = string_list_lookup(&state->symlink_changes, path);
3804 if (!ent)
3805 return 0;
3806 return (uintptr_t)ent->util;
3809 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3811 for ( ; patch; patch = patch->next) {
3812 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3813 (patch->is_rename || patch->is_delete))
3814 /* the symlink at patch->old_name is removed */
3815 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3817 if (patch->new_name && S_ISLNK(patch->new_mode))
3818 /* the symlink at patch->new_name is created or remains */
3819 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3823 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3825 do {
3826 unsigned int change;
3828 while (--name->len && name->buf[name->len] != '/')
3829 ; /* scan backwards */
3830 if (!name->len)
3831 break;
3832 name->buf[name->len] = '\0';
3833 change = check_symlink_changes(state, name->buf);
3834 if (change & APPLY_SYMLINK_IN_RESULT)
3835 return 1;
3836 if (change & APPLY_SYMLINK_GOES_AWAY)
3838 * This cannot be "return 0", because we may
3839 * see a new one created at a higher level.
3841 continue;
3843 /* otherwise, check the preimage */
3844 if (state->check_index) {
3845 struct cache_entry *ce;
3847 ce = index_file_exists(state->repo->index, name->buf,
3848 name->len, ignore_case);
3849 if (ce && S_ISLNK(ce->ce_mode))
3850 return 1;
3851 } else {
3852 struct stat st;
3853 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3854 return 1;
3856 } while (1);
3857 return 0;
3860 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3862 int ret;
3863 struct strbuf name = STRBUF_INIT;
3865 assert(*name_ != '\0');
3866 strbuf_addstr(&name, name_);
3867 ret = path_is_beyond_symlink_1(state, &name);
3868 strbuf_release(&name);
3870 return ret;
3873 static int check_unsafe_path(struct patch *patch)
3875 const char *old_name = NULL;
3876 const char *new_name = NULL;
3877 if (patch->is_delete)
3878 old_name = patch->old_name;
3879 else if (!patch->is_new && !patch->is_copy)
3880 old_name = patch->old_name;
3881 if (!patch->is_delete)
3882 new_name = patch->new_name;
3884 if (old_name && !verify_path(old_name, patch->old_mode))
3885 return error(_("invalid path '%s'"), old_name);
3886 if (new_name && !verify_path(new_name, patch->new_mode))
3887 return error(_("invalid path '%s'"), new_name);
3888 return 0;
3892 * Check and apply the patch in-core; leave the result in patch->result
3893 * for the caller to write it out to the final destination.
3895 static int check_patch(struct apply_state *state, struct patch *patch)
3897 struct stat st;
3898 const char *old_name = patch->old_name;
3899 const char *new_name = patch->new_name;
3900 const char *name = old_name ? old_name : new_name;
3901 struct cache_entry *ce = NULL;
3902 struct patch *tpatch;
3903 int ok_if_exists;
3904 int status;
3906 patch->rejected = 1; /* we will drop this after we succeed */
3908 status = check_preimage(state, patch, &ce, &st);
3909 if (status)
3910 return status;
3911 old_name = patch->old_name;
3914 * A type-change diff is always split into a patch to delete
3915 * old, immediately followed by a patch to create new (see
3916 * diff.c::run_diff()); in such a case it is Ok that the entry
3917 * to be deleted by the previous patch is still in the working
3918 * tree and in the index.
3920 * A patch to swap-rename between A and B would first rename A
3921 * to B and then rename B to A. While applying the first one,
3922 * the presence of B should not stop A from getting renamed to
3923 * B; ask to_be_deleted() about the later rename. Removal of
3924 * B and rename from A to B is handled the same way by asking
3925 * was_deleted().
3927 if ((tpatch = in_fn_table(state, new_name)) &&
3928 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3929 ok_if_exists = 1;
3930 else
3931 ok_if_exists = 0;
3933 if (new_name &&
3934 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3935 int err = check_to_create(state, new_name, ok_if_exists);
3937 if (err && state->threeway) {
3938 patch->direct_to_threeway = 1;
3939 } else switch (err) {
3940 case 0:
3941 break; /* happy */
3942 case EXISTS_IN_INDEX:
3943 return error(_("%s: already exists in index"), new_name);
3944 break;
3945 case EXISTS_IN_WORKTREE:
3946 return error(_("%s: already exists in working directory"),
3947 new_name);
3948 default:
3949 return err;
3952 if (!patch->new_mode) {
3953 if (0 < patch->is_new)
3954 patch->new_mode = S_IFREG | 0644;
3955 else
3956 patch->new_mode = patch->old_mode;
3960 if (new_name && old_name) {
3961 int same = !strcmp(old_name, new_name);
3962 if (!patch->new_mode)
3963 patch->new_mode = patch->old_mode;
3964 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3965 if (same)
3966 return error(_("new mode (%o) of %s does not "
3967 "match old mode (%o)"),
3968 patch->new_mode, new_name,
3969 patch->old_mode);
3970 else
3971 return error(_("new mode (%o) of %s does not "
3972 "match old mode (%o) of %s"),
3973 patch->new_mode, new_name,
3974 patch->old_mode, old_name);
3978 if (!state->unsafe_paths && check_unsafe_path(patch))
3979 return -128;
3982 * An attempt to read from or delete a path that is beyond a
3983 * symbolic link will be prevented by load_patch_target() that
3984 * is called at the beginning of apply_data() so we do not
3985 * have to worry about a patch marked with "is_delete" bit
3986 * here. We however need to make sure that the patch result
3987 * is not deposited to a path that is beyond a symbolic link
3988 * here.
3990 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3991 return error(_("affected file '%s' is beyond a symbolic link"),
3992 patch->new_name);
3994 if (apply_data(state, patch, &st, ce) < 0)
3995 return error(_("%s: patch does not apply"), name);
3996 patch->rejected = 0;
3997 return 0;
4000 static int check_patch_list(struct apply_state *state, struct patch *patch)
4002 int err = 0;
4004 prepare_symlink_changes(state, patch);
4005 prepare_fn_table(state, patch);
4006 while (patch) {
4007 int res;
4008 if (state->apply_verbosity > verbosity_normal)
4009 say_patch_name(stderr,
4010 _("Checking patch %s..."), patch);
4011 res = check_patch(state, patch);
4012 if (res == -128)
4013 return -128;
4014 err |= res;
4015 patch = patch->next;
4017 return err;
4020 static int read_apply_cache(struct apply_state *state)
4022 if (state->index_file)
4023 return read_index_from(state->repo->index, state->index_file,
4024 get_git_dir());
4025 else
4026 return repo_read_index(state->repo);
4029 /* This function tries to read the object name from the current index */
4030 static int get_current_oid(struct apply_state *state, const char *path,
4031 struct object_id *oid)
4033 int pos;
4035 if (read_apply_cache(state) < 0)
4036 return -1;
4037 pos = index_name_pos(state->repo->index, path, strlen(path));
4038 if (pos < 0)
4039 return -1;
4040 oidcpy(oid, &state->repo->index->cache[pos]->oid);
4041 return 0;
4044 static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
4047 * A usable gitlink patch has only one fragment (hunk) that looks like:
4048 * @@ -1 +1 @@
4049 * -Subproject commit <old sha1>
4050 * +Subproject commit <new sha1>
4051 * or
4052 * @@ -1 +0,0 @@
4053 * -Subproject commit <old sha1>
4054 * for a removal patch.
4056 struct fragment *hunk = p->fragments;
4057 static const char heading[] = "-Subproject commit ";
4058 char *preimage;
4060 if (/* does the patch have only one hunk? */
4061 hunk && !hunk->next &&
4062 /* is its preimage one line? */
4063 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4064 /* does preimage begin with the heading? */
4065 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4066 starts_with(++preimage, heading) &&
4067 /* does it record full SHA-1? */
4068 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4069 preimage[sizeof(heading) + the_hash_algo->hexsz - 1] == '\n' &&
4070 /* does the abbreviated name on the index line agree with it? */
4071 starts_with(preimage + sizeof(heading) - 1, p->old_oid_prefix))
4072 return 0; /* it all looks fine */
4074 /* we may have full object name on the index line */
4075 return get_oid_hex(p->old_oid_prefix, oid);
4078 /* Build an index that contains just the files needed for a 3way merge */
4079 static int build_fake_ancestor(struct apply_state *state, struct patch *list)
4081 struct patch *patch;
4082 struct index_state result = { NULL };
4083 struct lock_file lock = LOCK_INIT;
4084 int res;
4086 /* Once we start supporting the reverse patch, it may be
4087 * worth showing the new sha1 prefix, but until then...
4089 for (patch = list; patch; patch = patch->next) {
4090 struct object_id oid;
4091 struct cache_entry *ce;
4092 const char *name;
4094 name = patch->old_name ? patch->old_name : patch->new_name;
4095 if (0 < patch->is_new)
4096 continue;
4098 if (S_ISGITLINK(patch->old_mode)) {
4099 if (!preimage_oid_in_gitlink_patch(patch, &oid))
4100 ; /* ok, the textual part looks sane */
4101 else
4102 return error(_("sha1 information is lacking or "
4103 "useless for submodule %s"), name);
4104 } else if (!get_oid_blob(patch->old_oid_prefix, &oid)) {
4105 ; /* ok */
4106 } else if (!patch->lines_added && !patch->lines_deleted) {
4107 /* mode-only change: update the current */
4108 if (get_current_oid(state, patch->old_name, &oid))
4109 return error(_("mode change for %s, which is not "
4110 "in current HEAD"), name);
4111 } else
4112 return error(_("sha1 information is lacking or useless "
4113 "(%s)."), name);
4115 ce = make_cache_entry(&result, patch->old_mode, &oid, name, 0, 0);
4116 if (!ce)
4117 return error(_("make_cache_entry failed for path '%s'"),
4118 name);
4119 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4120 discard_cache_entry(ce);
4121 return error(_("could not add %s to temporary index"),
4122 name);
4126 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
4127 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4128 discard_index(&result);
4130 if (res)
4131 return error(_("could not write temporary index to %s"),
4132 state->fake_ancestor);
4134 return 0;
4137 static void stat_patch_list(struct apply_state *state, struct patch *patch)
4139 int files, adds, dels;
4141 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4142 files++;
4143 adds += patch->lines_added;
4144 dels += patch->lines_deleted;
4145 show_stats(state, patch);
4148 print_stat_summary(stdout, files, adds, dels);
4151 static void numstat_patch_list(struct apply_state *state,
4152 struct patch *patch)
4154 for ( ; patch; patch = patch->next) {
4155 const char *name;
4156 name = patch->new_name ? patch->new_name : patch->old_name;
4157 if (patch->is_binary)
4158 printf("-\t-\t");
4159 else
4160 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4161 write_name_quoted(name, stdout, state->line_termination);
4165 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4167 if (mode)
4168 printf(" %s mode %06o %s\n", newdelete, mode, name);
4169 else
4170 printf(" %s %s\n", newdelete, name);
4173 static void show_mode_change(struct patch *p, int show_name)
4175 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4176 if (show_name)
4177 printf(" mode change %06o => %06o %s\n",
4178 p->old_mode, p->new_mode, p->new_name);
4179 else
4180 printf(" mode change %06o => %06o\n",
4181 p->old_mode, p->new_mode);
4185 static void show_rename_copy(struct patch *p)
4187 const char *renamecopy = p->is_rename ? "rename" : "copy";
4188 const char *old_name, *new_name;
4190 /* Find common prefix */
4191 old_name = p->old_name;
4192 new_name = p->new_name;
4193 while (1) {
4194 const char *slash_old, *slash_new;
4195 slash_old = strchr(old_name, '/');
4196 slash_new = strchr(new_name, '/');
4197 if (!slash_old ||
4198 !slash_new ||
4199 slash_old - old_name != slash_new - new_name ||
4200 memcmp(old_name, new_name, slash_new - new_name))
4201 break;
4202 old_name = slash_old + 1;
4203 new_name = slash_new + 1;
4205 /* p->old_name thru old_name is the common prefix, and old_name and new_name
4206 * through the end of names are renames
4208 if (old_name != p->old_name)
4209 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4210 (int)(old_name - p->old_name), p->old_name,
4211 old_name, new_name, p->score);
4212 else
4213 printf(" %s %s => %s (%d%%)\n", renamecopy,
4214 p->old_name, p->new_name, p->score);
4215 show_mode_change(p, 0);
4218 static void summary_patch_list(struct patch *patch)
4220 struct patch *p;
4222 for (p = patch; p; p = p->next) {
4223 if (p->is_new)
4224 show_file_mode_name("create", p->new_mode, p->new_name);
4225 else if (p->is_delete)
4226 show_file_mode_name("delete", p->old_mode, p->old_name);
4227 else {
4228 if (p->is_rename || p->is_copy)
4229 show_rename_copy(p);
4230 else {
4231 if (p->score) {
4232 printf(" rewrite %s (%d%%)\n",
4233 p->new_name, p->score);
4234 show_mode_change(p, 0);
4236 else
4237 show_mode_change(p, 1);
4243 static void patch_stats(struct apply_state *state, struct patch *patch)
4245 int lines = patch->lines_added + patch->lines_deleted;
4247 if (lines > state->max_change)
4248 state->max_change = lines;
4249 if (patch->old_name) {
4250 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4251 if (!len)
4252 len = strlen(patch->old_name);
4253 if (len > state->max_len)
4254 state->max_len = len;
4256 if (patch->new_name) {
4257 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4258 if (!len)
4259 len = strlen(patch->new_name);
4260 if (len > state->max_len)
4261 state->max_len = len;
4265 static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4267 if (state->update_index && !state->ita_only) {
4268 if (remove_file_from_index(state->repo->index, patch->old_name) < 0)
4269 return error(_("unable to remove %s from index"), patch->old_name);
4271 if (!state->cached) {
4272 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4273 remove_path(patch->old_name);
4276 return 0;
4279 static int add_index_file(struct apply_state *state,
4280 const char *path,
4281 unsigned mode,
4282 void *buf,
4283 unsigned long size)
4285 struct stat st;
4286 struct cache_entry *ce;
4287 int namelen = strlen(path);
4289 ce = make_empty_cache_entry(state->repo->index, namelen);
4290 memcpy(ce->name, path, namelen);
4291 ce->ce_mode = create_ce_mode(mode);
4292 ce->ce_flags = create_ce_flags(0);
4293 ce->ce_namelen = namelen;
4294 if (state->ita_only) {
4295 ce->ce_flags |= CE_INTENT_TO_ADD;
4296 set_object_name_for_intent_to_add_entry(ce);
4297 } else if (S_ISGITLINK(mode)) {
4298 const char *s;
4300 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4301 get_oid_hex(s, &ce->oid)) {
4302 discard_cache_entry(ce);
4303 return error(_("corrupt patch for submodule %s"), path);
4305 } else {
4306 if (!state->cached) {
4307 if (lstat(path, &st) < 0) {
4308 discard_cache_entry(ce);
4309 return error_errno(_("unable to stat newly "
4310 "created file '%s'"),
4311 path);
4313 fill_stat_cache_info(ce, &st);
4315 if (write_object_file(buf, size, blob_type, &ce->oid) < 0) {
4316 discard_cache_entry(ce);
4317 return error(_("unable to create backing store "
4318 "for newly created file %s"), path);
4321 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4322 discard_cache_entry(ce);
4323 return error(_("unable to add cache entry for %s"), path);
4326 return 0;
4330 * Returns:
4331 * -1 if an unrecoverable error happened
4332 * 0 if everything went well
4333 * 1 if a recoverable error happened
4335 static int try_create_file(struct apply_state *state, const char *path,
4336 unsigned int mode, const char *buf,
4337 unsigned long size)
4339 int fd, res;
4340 struct strbuf nbuf = STRBUF_INIT;
4342 if (S_ISGITLINK(mode)) {
4343 struct stat st;
4344 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4345 return 0;
4346 return !!mkdir(path, 0777);
4349 if (has_symlinks && S_ISLNK(mode))
4350 /* Although buf:size is counted string, it also is NUL
4351 * terminated.
4353 return !!symlink(buf, path);
4355 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4356 if (fd < 0)
4357 return 1;
4359 if (convert_to_working_tree(state->repo->index, path, buf, size, &nbuf)) {
4360 size = nbuf.len;
4361 buf = nbuf.buf;
4364 res = write_in_full(fd, buf, size) < 0;
4365 if (res)
4366 error_errno(_("failed to write to '%s'"), path);
4367 strbuf_release(&nbuf);
4369 if (close(fd) < 0 && !res)
4370 return error_errno(_("closing file '%s'"), path);
4372 return res ? -1 : 0;
4376 * We optimistically assume that the directories exist,
4377 * which is true 99% of the time anyway. If they don't,
4378 * we create them and try again.
4380 * Returns:
4381 * -1 on error
4382 * 0 otherwise
4384 static int create_one_file(struct apply_state *state,
4385 char *path,
4386 unsigned mode,
4387 const char *buf,
4388 unsigned long size)
4390 int res;
4392 if (state->cached)
4393 return 0;
4395 res = try_create_file(state, path, mode, buf, size);
4396 if (res < 0)
4397 return -1;
4398 if (!res)
4399 return 0;
4401 if (errno == ENOENT) {
4402 if (safe_create_leading_directories(path))
4403 return 0;
4404 res = try_create_file(state, path, mode, buf, size);
4405 if (res < 0)
4406 return -1;
4407 if (!res)
4408 return 0;
4411 if (errno == EEXIST || errno == EACCES) {
4412 /* We may be trying to create a file where a directory
4413 * used to be.
4415 struct stat st;
4416 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4417 errno = EEXIST;
4420 if (errno == EEXIST) {
4421 unsigned int nr = getpid();
4423 for (;;) {
4424 char newpath[PATH_MAX];
4425 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4426 res = try_create_file(state, newpath, mode, buf, size);
4427 if (res < 0)
4428 return -1;
4429 if (!res) {
4430 if (!rename(newpath, path))
4431 return 0;
4432 unlink_or_warn(newpath);
4433 break;
4435 if (errno != EEXIST)
4436 break;
4437 ++nr;
4440 return error_errno(_("unable to write file '%s' mode %o"),
4441 path, mode);
4444 static int add_conflicted_stages_file(struct apply_state *state,
4445 struct patch *patch)
4447 int stage, namelen;
4448 unsigned mode;
4449 struct cache_entry *ce;
4451 if (!state->update_index)
4452 return 0;
4453 namelen = strlen(patch->new_name);
4454 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4456 remove_file_from_index(state->repo->index, patch->new_name);
4457 for (stage = 1; stage < 4; stage++) {
4458 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4459 continue;
4460 ce = make_empty_cache_entry(state->repo->index, namelen);
4461 memcpy(ce->name, patch->new_name, namelen);
4462 ce->ce_mode = create_ce_mode(mode);
4463 ce->ce_flags = create_ce_flags(stage);
4464 ce->ce_namelen = namelen;
4465 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
4466 if (add_index_entry(state->repo->index, ce, ADD_CACHE_OK_TO_ADD) < 0) {
4467 discard_cache_entry(ce);
4468 return error(_("unable to add cache entry for %s"),
4469 patch->new_name);
4473 return 0;
4476 static int create_file(struct apply_state *state, struct patch *patch)
4478 char *path = patch->new_name;
4479 unsigned mode = patch->new_mode;
4480 unsigned long size = patch->resultsize;
4481 char *buf = patch->result;
4483 if (!mode)
4484 mode = S_IFREG | 0644;
4485 if (create_one_file(state, path, mode, buf, size))
4486 return -1;
4488 if (patch->conflicted_threeway)
4489 return add_conflicted_stages_file(state, patch);
4490 else if (state->update_index)
4491 return add_index_file(state, path, mode, buf, size);
4492 return 0;
4495 /* phase zero is to remove, phase one is to create */
4496 static int write_out_one_result(struct apply_state *state,
4497 struct patch *patch,
4498 int phase)
4500 if (patch->is_delete > 0) {
4501 if (phase == 0)
4502 return remove_file(state, patch, 1);
4503 return 0;
4505 if (patch->is_new > 0 || patch->is_copy) {
4506 if (phase == 1)
4507 return create_file(state, patch);
4508 return 0;
4511 * Rename or modification boils down to the same
4512 * thing: remove the old, write the new
4514 if (phase == 0)
4515 return remove_file(state, patch, patch->is_rename);
4516 if (phase == 1)
4517 return create_file(state, patch);
4518 return 0;
4521 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4523 FILE *rej;
4524 char namebuf[PATH_MAX];
4525 struct fragment *frag;
4526 int cnt = 0;
4527 struct strbuf sb = STRBUF_INIT;
4529 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4530 if (!frag->rejected)
4531 continue;
4532 cnt++;
4535 if (!cnt) {
4536 if (state->apply_verbosity > verbosity_normal)
4537 say_patch_name(stderr,
4538 _("Applied patch %s cleanly."), patch);
4539 return 0;
4542 /* This should not happen, because a removal patch that leaves
4543 * contents are marked "rejected" at the patch level.
4545 if (!patch->new_name)
4546 die(_("internal error"));
4548 /* Say this even without --verbose */
4549 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4550 "Applying patch %%s with %d rejects...",
4551 cnt),
4552 cnt);
4553 if (state->apply_verbosity > verbosity_silent)
4554 say_patch_name(stderr, sb.buf, patch);
4555 strbuf_release(&sb);
4557 cnt = strlen(patch->new_name);
4558 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4559 cnt = ARRAY_SIZE(namebuf) - 5;
4560 warning(_("truncating .rej filename to %.*s.rej"),
4561 cnt - 1, patch->new_name);
4563 memcpy(namebuf, patch->new_name, cnt);
4564 memcpy(namebuf + cnt, ".rej", 5);
4566 rej = fopen(namebuf, "w");
4567 if (!rej)
4568 return error_errno(_("cannot open %s"), namebuf);
4570 /* Normal git tools never deal with .rej, so do not pretend
4571 * this is a git patch by saying --git or giving extended
4572 * headers. While at it, maybe please "kompare" that wants
4573 * the trailing TAB and some garbage at the end of line ;-).
4575 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4576 patch->new_name, patch->new_name);
4577 for (cnt = 1, frag = patch->fragments;
4578 frag;
4579 cnt++, frag = frag->next) {
4580 if (!frag->rejected) {
4581 if (state->apply_verbosity > verbosity_silent)
4582 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4583 continue;
4585 if (state->apply_verbosity > verbosity_silent)
4586 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4587 fprintf(rej, "%.*s", frag->size, frag->patch);