stash: teach 'push' (and 'create_stash') to honor pathspec
[git.git] / apply.c
blob705cf562f07098aafcd9f6e27105d2105714751f
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 "blob.h"
12 #include "delta.h"
13 #include "diff.h"
14 #include "dir.h"
15 #include "xdiff-interface.h"
16 #include "ll-merge.h"
17 #include "lockfile.h"
18 #include "parse-options.h"
19 #include "quote.h"
20 #include "rerere.h"
21 #include "apply.h"
23 static void git_apply_config(void)
25 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
26 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
27 git_config(git_default_config, NULL);
30 static int parse_whitespace_option(struct apply_state *state, const char *option)
32 if (!option) {
33 state->ws_error_action = warn_on_ws_error;
34 return 0;
36 if (!strcmp(option, "warn")) {
37 state->ws_error_action = warn_on_ws_error;
38 return 0;
40 if (!strcmp(option, "nowarn")) {
41 state->ws_error_action = nowarn_ws_error;
42 return 0;
44 if (!strcmp(option, "error")) {
45 state->ws_error_action = die_on_ws_error;
46 return 0;
48 if (!strcmp(option, "error-all")) {
49 state->ws_error_action = die_on_ws_error;
50 state->squelch_whitespace_errors = 0;
51 return 0;
53 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
54 state->ws_error_action = correct_ws_error;
55 return 0;
57 return error(_("unrecognized whitespace option '%s'"), option);
60 static int parse_ignorewhitespace_option(struct apply_state *state,
61 const char *option)
63 if (!option || !strcmp(option, "no") ||
64 !strcmp(option, "false") || !strcmp(option, "never") ||
65 !strcmp(option, "none")) {
66 state->ws_ignore_action = ignore_ws_none;
67 return 0;
69 if (!strcmp(option, "change")) {
70 state->ws_ignore_action = ignore_ws_change;
71 return 0;
73 return error(_("unrecognized whitespace ignore option '%s'"), option);
76 int init_apply_state(struct apply_state *state,
77 const char *prefix,
78 struct lock_file *lock_file)
80 memset(state, 0, sizeof(*state));
81 state->prefix = prefix;
82 state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
83 state->lock_file = lock_file;
84 state->newfd = -1;
85 state->apply = 1;
86 state->line_termination = '\n';
87 state->p_value = 1;
88 state->p_context = UINT_MAX;
89 state->squelch_whitespace_errors = 5;
90 state->ws_error_action = warn_on_ws_error;
91 state->ws_ignore_action = ignore_ws_none;
92 state->linenr = 1;
93 string_list_init(&state->fn_table, 0);
94 string_list_init(&state->limit_by_name, 0);
95 string_list_init(&state->symlink_changes, 0);
96 strbuf_init(&state->root, 0);
98 git_apply_config();
99 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
100 return -1;
101 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
102 return -1;
103 return 0;
106 void clear_apply_state(struct apply_state *state)
108 string_list_clear(&state->limit_by_name, 0);
109 string_list_clear(&state->symlink_changes, 0);
110 strbuf_release(&state->root);
112 /* &state->fn_table is cleared at the end of apply_patch() */
115 static void mute_routine(const char *msg, va_list params)
117 /* do nothing */
120 int check_apply_state(struct apply_state *state, int force_apply)
122 int is_not_gitdir = !startup_info->have_repository;
124 if (state->apply_with_reject && state->threeway)
125 return error(_("--reject and --3way cannot be used together."));
126 if (state->cached && state->threeway)
127 return error(_("--cached and --3way cannot be used together."));
128 if (state->threeway) {
129 if (is_not_gitdir)
130 return error(_("--3way outside a repository"));
131 state->check_index = 1;
133 if (state->apply_with_reject) {
134 state->apply = 1;
135 if (state->apply_verbosity == verbosity_normal)
136 state->apply_verbosity = verbosity_verbose;
138 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
139 state->apply = 0;
140 if (state->check_index && is_not_gitdir)
141 return error(_("--index outside a repository"));
142 if (state->cached) {
143 if (is_not_gitdir)
144 return error(_("--cached outside a repository"));
145 state->check_index = 1;
147 if (state->check_index)
148 state->unsafe_paths = 0;
149 if (!state->lock_file)
150 return error("BUG: state->lock_file should not be NULL");
152 if (state->apply_verbosity <= verbosity_silent) {
153 state->saved_error_routine = get_error_routine();
154 state->saved_warn_routine = get_warn_routine();
155 set_error_routine(mute_routine);
156 set_warn_routine(mute_routine);
159 return 0;
162 static void set_default_whitespace_mode(struct apply_state *state)
164 if (!state->whitespace_option && !apply_default_whitespace)
165 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
169 * This represents one "hunk" from a patch, starting with
170 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
171 * patch text is pointed at by patch, and its byte length
172 * is stored in size. leading and trailing are the number
173 * of context lines.
175 struct fragment {
176 unsigned long leading, trailing;
177 unsigned long oldpos, oldlines;
178 unsigned long newpos, newlines;
180 * 'patch' is usually borrowed from buf in apply_patch(),
181 * but some codepaths store an allocated buffer.
183 const char *patch;
184 unsigned free_patch:1,
185 rejected:1;
186 int size;
187 int linenr;
188 struct fragment *next;
192 * When dealing with a binary patch, we reuse "leading" field
193 * to store the type of the binary hunk, either deflated "delta"
194 * or deflated "literal".
196 #define binary_patch_method leading
197 #define BINARY_DELTA_DEFLATED 1
198 #define BINARY_LITERAL_DEFLATED 2
201 * This represents a "patch" to a file, both metainfo changes
202 * such as creation/deletion, filemode and content changes represented
203 * as a series of fragments.
205 struct patch {
206 char *new_name, *old_name, *def_name;
207 unsigned int old_mode, new_mode;
208 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
209 int rejected;
210 unsigned ws_rule;
211 int lines_added, lines_deleted;
212 int score;
213 unsigned int is_toplevel_relative:1;
214 unsigned int inaccurate_eof:1;
215 unsigned int is_binary:1;
216 unsigned int is_copy:1;
217 unsigned int is_rename:1;
218 unsigned int recount:1;
219 unsigned int conflicted_threeway:1;
220 unsigned int direct_to_threeway:1;
221 struct fragment *fragments;
222 char *result;
223 size_t resultsize;
224 char old_sha1_prefix[41];
225 char new_sha1_prefix[41];
226 struct patch *next;
228 /* three-way fallback result */
229 struct object_id threeway_stage[3];
232 static void free_fragment_list(struct fragment *list)
234 while (list) {
235 struct fragment *next = list->next;
236 if (list->free_patch)
237 free((char *)list->patch);
238 free(list);
239 list = next;
243 static void free_patch(struct patch *patch)
245 free_fragment_list(patch->fragments);
246 free(patch->def_name);
247 free(patch->old_name);
248 free(patch->new_name);
249 free(patch->result);
250 free(patch);
253 static void free_patch_list(struct patch *list)
255 while (list) {
256 struct patch *next = list->next;
257 free_patch(list);
258 list = next;
263 * A line in a file, len-bytes long (includes the terminating LF,
264 * except for an incomplete line at the end if the file ends with
265 * one), and its contents hashes to 'hash'.
267 struct line {
268 size_t len;
269 unsigned hash : 24;
270 unsigned flag : 8;
271 #define LINE_COMMON 1
272 #define LINE_PATCHED 2
276 * This represents a "file", which is an array of "lines".
278 struct image {
279 char *buf;
280 size_t len;
281 size_t nr;
282 size_t alloc;
283 struct line *line_allocated;
284 struct line *line;
287 static uint32_t hash_line(const char *cp, size_t len)
289 size_t i;
290 uint32_t h;
291 for (i = 0, h = 0; i < len; i++) {
292 if (!isspace(cp[i])) {
293 h = h * 3 + (cp[i] & 0xff);
296 return h;
300 * Compare lines s1 of length n1 and s2 of length n2, ignoring
301 * whitespace difference. Returns 1 if they match, 0 otherwise
303 static int fuzzy_matchlines(const char *s1, size_t n1,
304 const char *s2, size_t n2)
306 const char *last1 = s1 + n1 - 1;
307 const char *last2 = s2 + n2 - 1;
308 int result = 0;
310 /* ignore line endings */
311 while ((*last1 == '\r') || (*last1 == '\n'))
312 last1--;
313 while ((*last2 == '\r') || (*last2 == '\n'))
314 last2--;
316 /* skip leading whitespaces, if both begin with whitespace */
317 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
318 while (isspace(*s1) && (s1 <= last1))
319 s1++;
320 while (isspace(*s2) && (s2 <= last2))
321 s2++;
323 /* early return if both lines are empty */
324 if ((s1 > last1) && (s2 > last2))
325 return 1;
326 while (!result) {
327 result = *s1++ - *s2++;
329 * Skip whitespace inside. We check for whitespace on
330 * both buffers because we don't want "a b" to match
331 * "ab"
333 if (isspace(*s1) && isspace(*s2)) {
334 while (isspace(*s1) && s1 <= last1)
335 s1++;
336 while (isspace(*s2) && s2 <= last2)
337 s2++;
340 * If we reached the end on one side only,
341 * lines don't match
343 if (
344 ((s2 > last2) && (s1 <= last1)) ||
345 ((s1 > last1) && (s2 <= last2)))
346 return 0;
347 if ((s1 > last1) && (s2 > last2))
348 break;
351 return !result;
354 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
356 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
357 img->line_allocated[img->nr].len = len;
358 img->line_allocated[img->nr].hash = hash_line(bol, len);
359 img->line_allocated[img->nr].flag = flag;
360 img->nr++;
364 * "buf" has the file contents to be patched (read from various sources).
365 * attach it to "image" and add line-based index to it.
366 * "image" now owns the "buf".
368 static void prepare_image(struct image *image, char *buf, size_t len,
369 int prepare_linetable)
371 const char *cp, *ep;
373 memset(image, 0, sizeof(*image));
374 image->buf = buf;
375 image->len = len;
377 if (!prepare_linetable)
378 return;
380 ep = image->buf + image->len;
381 cp = image->buf;
382 while (cp < ep) {
383 const char *next;
384 for (next = cp; next < ep && *next != '\n'; next++)
386 if (next < ep)
387 next++;
388 add_line_info(image, cp, next - cp, 0);
389 cp = next;
391 image->line = image->line_allocated;
394 static void clear_image(struct image *image)
396 free(image->buf);
397 free(image->line_allocated);
398 memset(image, 0, sizeof(*image));
401 /* fmt must contain _one_ %s and no other substitution */
402 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
404 struct strbuf sb = STRBUF_INIT;
406 if (patch->old_name && patch->new_name &&
407 strcmp(patch->old_name, patch->new_name)) {
408 quote_c_style(patch->old_name, &sb, NULL, 0);
409 strbuf_addstr(&sb, " => ");
410 quote_c_style(patch->new_name, &sb, NULL, 0);
411 } else {
412 const char *n = patch->new_name;
413 if (!n)
414 n = patch->old_name;
415 quote_c_style(n, &sb, NULL, 0);
417 fprintf(output, fmt, sb.buf);
418 fputc('\n', output);
419 strbuf_release(&sb);
422 #define SLOP (16)
424 static int read_patch_file(struct strbuf *sb, int fd)
426 if (strbuf_read(sb, fd, 0) < 0)
427 return error_errno("git apply: failed to read");
430 * Make sure that we have some slop in the buffer
431 * so that we can do speculative "memcmp" etc, and
432 * see to it that it is NUL-filled.
434 strbuf_grow(sb, SLOP);
435 memset(sb->buf + sb->len, 0, SLOP);
436 return 0;
439 static unsigned long linelen(const char *buffer, unsigned long size)
441 unsigned long len = 0;
442 while (size--) {
443 len++;
444 if (*buffer++ == '\n')
445 break;
447 return len;
450 static int is_dev_null(const char *str)
452 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
455 #define TERM_SPACE 1
456 #define TERM_TAB 2
458 static int name_terminate(int c, int terminate)
460 if (c == ' ' && !(terminate & TERM_SPACE))
461 return 0;
462 if (c == '\t' && !(terminate & TERM_TAB))
463 return 0;
465 return 1;
468 /* remove double slashes to make --index work with such filenames */
469 static char *squash_slash(char *name)
471 int i = 0, j = 0;
473 if (!name)
474 return NULL;
476 while (name[i]) {
477 if ((name[j++] = name[i++]) == '/')
478 while (name[i] == '/')
479 i++;
481 name[j] = '\0';
482 return name;
485 static char *find_name_gnu(struct apply_state *state,
486 const char *line,
487 const char *def,
488 int p_value)
490 struct strbuf name = STRBUF_INIT;
491 char *cp;
494 * Proposed "new-style" GNU patch/diff format; see
495 * http://marc.info/?l=git&m=112927316408690&w=2
497 if (unquote_c_style(&name, line, NULL)) {
498 strbuf_release(&name);
499 return NULL;
502 for (cp = name.buf; p_value; p_value--) {
503 cp = strchr(cp, '/');
504 if (!cp) {
505 strbuf_release(&name);
506 return NULL;
508 cp++;
511 strbuf_remove(&name, 0, cp - name.buf);
512 if (state->root.len)
513 strbuf_insert(&name, 0, state->root.buf, state->root.len);
514 return squash_slash(strbuf_detach(&name, NULL));
517 static size_t sane_tz_len(const char *line, size_t len)
519 const char *tz, *p;
521 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
522 return 0;
523 tz = line + len - strlen(" +0500");
525 if (tz[1] != '+' && tz[1] != '-')
526 return 0;
528 for (p = tz + 2; p != line + len; p++)
529 if (!isdigit(*p))
530 return 0;
532 return line + len - tz;
535 static size_t tz_with_colon_len(const char *line, size_t len)
537 const char *tz, *p;
539 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
540 return 0;
541 tz = line + len - strlen(" +08:00");
543 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
544 return 0;
545 p = tz + 2;
546 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
547 !isdigit(*p++) || !isdigit(*p++))
548 return 0;
550 return line + len - tz;
553 static size_t date_len(const char *line, size_t len)
555 const char *date, *p;
557 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
558 return 0;
559 p = date = line + len - strlen("72-02-05");
561 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
562 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
563 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
564 return 0;
566 if (date - line >= strlen("19") &&
567 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
568 date -= strlen("19");
570 return line + len - date;
573 static size_t short_time_len(const char *line, size_t len)
575 const char *time, *p;
577 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
578 return 0;
579 p = time = line + len - strlen(" 07:01:32");
581 /* Permit 1-digit hours? */
582 if (*p++ != ' ' ||
583 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
584 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
585 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
586 return 0;
588 return line + len - time;
591 static size_t fractional_time_len(const char *line, size_t len)
593 const char *p;
594 size_t n;
596 /* Expected format: 19:41:17.620000023 */
597 if (!len || !isdigit(line[len - 1]))
598 return 0;
599 p = line + len - 1;
601 /* Fractional seconds. */
602 while (p > line && isdigit(*p))
603 p--;
604 if (*p != '.')
605 return 0;
607 /* Hours, minutes, and whole seconds. */
608 n = short_time_len(line, p - line);
609 if (!n)
610 return 0;
612 return line + len - p + n;
615 static size_t trailing_spaces_len(const char *line, size_t len)
617 const char *p;
619 /* Expected format: ' ' x (1 or more) */
620 if (!len || line[len - 1] != ' ')
621 return 0;
623 p = line + len;
624 while (p != line) {
625 p--;
626 if (*p != ' ')
627 return line + len - (p + 1);
630 /* All spaces! */
631 return len;
634 static size_t diff_timestamp_len(const char *line, size_t len)
636 const char *end = line + len;
637 size_t n;
640 * Posix: 2010-07-05 19:41:17
641 * GNU: 2010-07-05 19:41:17.620000023 -0500
644 if (!isdigit(end[-1]))
645 return 0;
647 n = sane_tz_len(line, end - line);
648 if (!n)
649 n = tz_with_colon_len(line, end - line);
650 end -= n;
652 n = short_time_len(line, end - line);
653 if (!n)
654 n = fractional_time_len(line, end - line);
655 end -= n;
657 n = date_len(line, end - line);
658 if (!n) /* No date. Too bad. */
659 return 0;
660 end -= n;
662 if (end == line) /* No space before date. */
663 return 0;
664 if (end[-1] == '\t') { /* Success! */
665 end--;
666 return line + len - end;
668 if (end[-1] != ' ') /* No space before date. */
669 return 0;
671 /* Whitespace damage. */
672 end -= trailing_spaces_len(line, end - line);
673 return line + len - end;
676 static char *find_name_common(struct apply_state *state,
677 const char *line,
678 const char *def,
679 int p_value,
680 const char *end,
681 int terminate)
683 int len;
684 const char *start = NULL;
686 if (p_value == 0)
687 start = line;
688 while (line != end) {
689 char c = *line;
691 if (!end && isspace(c)) {
692 if (c == '\n')
693 break;
694 if (name_terminate(c, terminate))
695 break;
697 line++;
698 if (c == '/' && !--p_value)
699 start = line;
701 if (!start)
702 return squash_slash(xstrdup_or_null(def));
703 len = line - start;
704 if (!len)
705 return squash_slash(xstrdup_or_null(def));
708 * Generally we prefer the shorter name, especially
709 * if the other one is just a variation of that with
710 * something else tacked on to the end (ie "file.orig"
711 * or "file~").
713 if (def) {
714 int deflen = strlen(def);
715 if (deflen < len && !strncmp(start, def, deflen))
716 return squash_slash(xstrdup(def));
719 if (state->root.len) {
720 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
721 return squash_slash(ret);
724 return squash_slash(xmemdupz(start, len));
727 static char *find_name(struct apply_state *state,
728 const char *line,
729 char *def,
730 int p_value,
731 int terminate)
733 if (*line == '"') {
734 char *name = find_name_gnu(state, line, def, p_value);
735 if (name)
736 return name;
739 return find_name_common(state, line, def, p_value, NULL, terminate);
742 static char *find_name_traditional(struct apply_state *state,
743 const char *line,
744 char *def,
745 int p_value)
747 size_t len;
748 size_t date_len;
750 if (*line == '"') {
751 char *name = find_name_gnu(state, line, def, p_value);
752 if (name)
753 return name;
756 len = strchrnul(line, '\n') - line;
757 date_len = diff_timestamp_len(line, len);
758 if (!date_len)
759 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
760 len -= date_len;
762 return find_name_common(state, line, def, p_value, line + len, 0);
765 static int count_slashes(const char *cp)
767 int cnt = 0;
768 char ch;
770 while ((ch = *cp++))
771 if (ch == '/')
772 cnt++;
773 return cnt;
777 * Given the string after "--- " or "+++ ", guess the appropriate
778 * p_value for the given patch.
780 static int guess_p_value(struct apply_state *state, const char *nameline)
782 char *name, *cp;
783 int val = -1;
785 if (is_dev_null(nameline))
786 return -1;
787 name = find_name_traditional(state, nameline, NULL, 0);
788 if (!name)
789 return -1;
790 cp = strchr(name, '/');
791 if (!cp)
792 val = 0;
793 else if (state->prefix) {
795 * Does it begin with "a/$our-prefix" and such? Then this is
796 * very likely to apply to our directory.
798 if (!strncmp(name, state->prefix, state->prefix_length))
799 val = count_slashes(state->prefix);
800 else {
801 cp++;
802 if (!strncmp(cp, state->prefix, state->prefix_length))
803 val = count_slashes(state->prefix) + 1;
806 free(name);
807 return val;
811 * Does the ---/+++ line have the POSIX timestamp after the last HT?
812 * GNU diff puts epoch there to signal a creation/deletion event. Is
813 * this such a timestamp?
815 static int has_epoch_timestamp(const char *nameline)
818 * We are only interested in epoch timestamp; any non-zero
819 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
820 * For the same reason, the date must be either 1969-12-31 or
821 * 1970-01-01, and the seconds part must be "00".
823 const char stamp_regexp[] =
824 "^(1969-12-31|1970-01-01)"
826 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
828 "([-+][0-2][0-9]:?[0-5][0-9])\n";
829 const char *timestamp = NULL, *cp, *colon;
830 static regex_t *stamp;
831 regmatch_t m[10];
832 int zoneoffset;
833 int hourminute;
834 int status;
836 for (cp = nameline; *cp != '\n'; cp++) {
837 if (*cp == '\t')
838 timestamp = cp + 1;
840 if (!timestamp)
841 return 0;
842 if (!stamp) {
843 stamp = xmalloc(sizeof(*stamp));
844 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
845 warning(_("Cannot prepare timestamp regexp %s"),
846 stamp_regexp);
847 return 0;
851 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
852 if (status) {
853 if (status != REG_NOMATCH)
854 warning(_("regexec returned %d for input: %s"),
855 status, timestamp);
856 return 0;
859 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
860 if (*colon == ':')
861 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
862 else
863 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
864 if (timestamp[m[3].rm_so] == '-')
865 zoneoffset = -zoneoffset;
868 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
869 * (west of GMT) or 1970-01-01 (east of GMT)
871 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
872 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
873 return 0;
875 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
876 strtol(timestamp + 14, NULL, 10) -
877 zoneoffset);
879 return ((zoneoffset < 0 && hourminute == 1440) ||
880 (0 <= zoneoffset && !hourminute));
884 * Get the name etc info from the ---/+++ lines of a traditional patch header
886 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
887 * files, we can happily check the index for a match, but for creating a
888 * new file we should try to match whatever "patch" does. I have no idea.
890 static int parse_traditional_patch(struct apply_state *state,
891 const char *first,
892 const char *second,
893 struct patch *patch)
895 char *name;
897 first += 4; /* skip "--- " */
898 second += 4; /* skip "+++ " */
899 if (!state->p_value_known) {
900 int p, q;
901 p = guess_p_value(state, first);
902 q = guess_p_value(state, second);
903 if (p < 0) p = q;
904 if (0 <= p && p == q) {
905 state->p_value = p;
906 state->p_value_known = 1;
909 if (is_dev_null(first)) {
910 patch->is_new = 1;
911 patch->is_delete = 0;
912 name = find_name_traditional(state, second, NULL, state->p_value);
913 patch->new_name = name;
914 } else if (is_dev_null(second)) {
915 patch->is_new = 0;
916 patch->is_delete = 1;
917 name = find_name_traditional(state, first, NULL, state->p_value);
918 patch->old_name = name;
919 } else {
920 char *first_name;
921 first_name = find_name_traditional(state, first, NULL, state->p_value);
922 name = find_name_traditional(state, second, first_name, state->p_value);
923 free(first_name);
924 if (has_epoch_timestamp(first)) {
925 patch->is_new = 1;
926 patch->is_delete = 0;
927 patch->new_name = name;
928 } else if (has_epoch_timestamp(second)) {
929 patch->is_new = 0;
930 patch->is_delete = 1;
931 patch->old_name = name;
932 } else {
933 patch->old_name = name;
934 patch->new_name = xstrdup_or_null(name);
937 if (!name)
938 return error(_("unable to find filename in patch at line %d"), state->linenr);
940 return 0;
943 static int gitdiff_hdrend(struct apply_state *state,
944 const char *line,
945 struct patch *patch)
947 return 1;
951 * We're anal about diff header consistency, to make
952 * sure that we don't end up having strange ambiguous
953 * patches floating around.
955 * As a result, gitdiff_{old|new}name() will check
956 * their names against any previous information, just
957 * to make sure..
959 #define DIFF_OLD_NAME 0
960 #define DIFF_NEW_NAME 1
962 static int gitdiff_verify_name(struct apply_state *state,
963 const char *line,
964 int isnull,
965 char **name,
966 int side)
968 if (!*name && !isnull) {
969 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
970 return 0;
973 if (*name) {
974 int len = strlen(*name);
975 char *another;
976 if (isnull)
977 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
978 *name, state->linenr);
979 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
980 if (!another || memcmp(another, *name, len + 1)) {
981 free(another);
982 return error((side == DIFF_NEW_NAME) ?
983 _("git apply: bad git-diff - inconsistent new filename on line %d") :
984 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
986 free(another);
987 } else {
988 /* expect "/dev/null" */
989 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
990 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
993 return 0;
996 static int gitdiff_oldname(struct apply_state *state,
997 const char *line,
998 struct patch *patch)
1000 return gitdiff_verify_name(state, line,
1001 patch->is_new, &patch->old_name,
1002 DIFF_OLD_NAME);
1005 static int gitdiff_newname(struct apply_state *state,
1006 const char *line,
1007 struct patch *patch)
1009 return gitdiff_verify_name(state, line,
1010 patch->is_delete, &patch->new_name,
1011 DIFF_NEW_NAME);
1014 static int gitdiff_oldmode(struct apply_state *state,
1015 const char *line,
1016 struct patch *patch)
1018 patch->old_mode = strtoul(line, NULL, 8);
1019 return 0;
1022 static int gitdiff_newmode(struct apply_state *state,
1023 const char *line,
1024 struct patch *patch)
1026 patch->new_mode = strtoul(line, NULL, 8);
1027 return 0;
1030 static int gitdiff_delete(struct apply_state *state,
1031 const char *line,
1032 struct patch *patch)
1034 patch->is_delete = 1;
1035 free(patch->old_name);
1036 patch->old_name = xstrdup_or_null(patch->def_name);
1037 return gitdiff_oldmode(state, line, patch);
1040 static int gitdiff_newfile(struct apply_state *state,
1041 const char *line,
1042 struct patch *patch)
1044 patch->is_new = 1;
1045 free(patch->new_name);
1046 patch->new_name = xstrdup_or_null(patch->def_name);
1047 return gitdiff_newmode(state, line, patch);
1050 static int gitdiff_copysrc(struct apply_state *state,
1051 const char *line,
1052 struct patch *patch)
1054 patch->is_copy = 1;
1055 free(patch->old_name);
1056 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1057 return 0;
1060 static int gitdiff_copydst(struct apply_state *state,
1061 const char *line,
1062 struct patch *patch)
1064 patch->is_copy = 1;
1065 free(patch->new_name);
1066 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1067 return 0;
1070 static int gitdiff_renamesrc(struct apply_state *state,
1071 const char *line,
1072 struct patch *patch)
1074 patch->is_rename = 1;
1075 free(patch->old_name);
1076 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1077 return 0;
1080 static int gitdiff_renamedst(struct apply_state *state,
1081 const char *line,
1082 struct patch *patch)
1084 patch->is_rename = 1;
1085 free(patch->new_name);
1086 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1087 return 0;
1090 static int gitdiff_similarity(struct apply_state *state,
1091 const char *line,
1092 struct patch *patch)
1094 unsigned long val = strtoul(line, NULL, 10);
1095 if (val <= 100)
1096 patch->score = val;
1097 return 0;
1100 static int gitdiff_dissimilarity(struct apply_state *state,
1101 const char *line,
1102 struct patch *patch)
1104 unsigned long val = strtoul(line, NULL, 10);
1105 if (val <= 100)
1106 patch->score = val;
1107 return 0;
1110 static int gitdiff_index(struct apply_state *state,
1111 const char *line,
1112 struct patch *patch)
1115 * index line is N hexadecimal, "..", N hexadecimal,
1116 * and optional space with octal mode.
1118 const char *ptr, *eol;
1119 int len;
1121 ptr = strchr(line, '.');
1122 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1123 return 0;
1124 len = ptr - line;
1125 memcpy(patch->old_sha1_prefix, line, len);
1126 patch->old_sha1_prefix[len] = 0;
1128 line = ptr + 2;
1129 ptr = strchr(line, ' ');
1130 eol = strchrnul(line, '\n');
1132 if (!ptr || eol < ptr)
1133 ptr = eol;
1134 len = ptr - line;
1136 if (40 < len)
1137 return 0;
1138 memcpy(patch->new_sha1_prefix, line, len);
1139 patch->new_sha1_prefix[len] = 0;
1140 if (*ptr == ' ')
1141 patch->old_mode = strtoul(ptr+1, NULL, 8);
1142 return 0;
1146 * This is normal for a diff that doesn't change anything: we'll fall through
1147 * into the next diff. Tell the parser to break out.
1149 static int gitdiff_unrecognized(struct apply_state *state,
1150 const char *line,
1151 struct patch *patch)
1153 return 1;
1157 * Skip p_value leading components from "line"; as we do not accept
1158 * absolute paths, return NULL in that case.
1160 static const char *skip_tree_prefix(struct apply_state *state,
1161 const char *line,
1162 int llen)
1164 int nslash;
1165 int i;
1167 if (!state->p_value)
1168 return (llen && line[0] == '/') ? NULL : line;
1170 nslash = state->p_value;
1171 for (i = 0; i < llen; i++) {
1172 int ch = line[i];
1173 if (ch == '/' && --nslash <= 0)
1174 return (i == 0) ? NULL : &line[i + 1];
1176 return NULL;
1180 * This is to extract the same name that appears on "diff --git"
1181 * line. We do not find and return anything if it is a rename
1182 * patch, and it is OK because we will find the name elsewhere.
1183 * We need to reliably find name only when it is mode-change only,
1184 * creation or deletion of an empty file. In any of these cases,
1185 * both sides are the same name under a/ and b/ respectively.
1187 static char *git_header_name(struct apply_state *state,
1188 const char *line,
1189 int llen)
1191 const char *name;
1192 const char *second = NULL;
1193 size_t len, line_len;
1195 line += strlen("diff --git ");
1196 llen -= strlen("diff --git ");
1198 if (*line == '"') {
1199 const char *cp;
1200 struct strbuf first = STRBUF_INIT;
1201 struct strbuf sp = STRBUF_INIT;
1203 if (unquote_c_style(&first, line, &second))
1204 goto free_and_fail1;
1206 /* strip the a/b prefix including trailing slash */
1207 cp = skip_tree_prefix(state, first.buf, first.len);
1208 if (!cp)
1209 goto free_and_fail1;
1210 strbuf_remove(&first, 0, cp - first.buf);
1213 * second points at one past closing dq of name.
1214 * find the second name.
1216 while ((second < line + llen) && isspace(*second))
1217 second++;
1219 if (line + llen <= second)
1220 goto free_and_fail1;
1221 if (*second == '"') {
1222 if (unquote_c_style(&sp, second, NULL))
1223 goto free_and_fail1;
1224 cp = skip_tree_prefix(state, sp.buf, sp.len);
1225 if (!cp)
1226 goto free_and_fail1;
1227 /* They must match, otherwise ignore */
1228 if (strcmp(cp, first.buf))
1229 goto free_and_fail1;
1230 strbuf_release(&sp);
1231 return strbuf_detach(&first, NULL);
1234 /* unquoted second */
1235 cp = skip_tree_prefix(state, second, line + llen - second);
1236 if (!cp)
1237 goto free_and_fail1;
1238 if (line + llen - cp != first.len ||
1239 memcmp(first.buf, cp, first.len))
1240 goto free_and_fail1;
1241 return strbuf_detach(&first, NULL);
1243 free_and_fail1:
1244 strbuf_release(&first);
1245 strbuf_release(&sp);
1246 return NULL;
1249 /* unquoted first name */
1250 name = skip_tree_prefix(state, line, llen);
1251 if (!name)
1252 return NULL;
1255 * since the first name is unquoted, a dq if exists must be
1256 * the beginning of the second name.
1258 for (second = name; second < line + llen; second++) {
1259 if (*second == '"') {
1260 struct strbuf sp = STRBUF_INIT;
1261 const char *np;
1263 if (unquote_c_style(&sp, second, NULL))
1264 goto free_and_fail2;
1266 np = skip_tree_prefix(state, sp.buf, sp.len);
1267 if (!np)
1268 goto free_and_fail2;
1270 len = sp.buf + sp.len - np;
1271 if (len < second - name &&
1272 !strncmp(np, name, len) &&
1273 isspace(name[len])) {
1274 /* Good */
1275 strbuf_remove(&sp, 0, np - sp.buf);
1276 return strbuf_detach(&sp, NULL);
1279 free_and_fail2:
1280 strbuf_release(&sp);
1281 return NULL;
1286 * Accept a name only if it shows up twice, exactly the same
1287 * form.
1289 second = strchr(name, '\n');
1290 if (!second)
1291 return NULL;
1292 line_len = second - name;
1293 for (len = 0 ; ; len++) {
1294 switch (name[len]) {
1295 default:
1296 continue;
1297 case '\n':
1298 return NULL;
1299 case '\t': case ' ':
1301 * Is this the separator between the preimage
1302 * and the postimage pathname? Again, we are
1303 * only interested in the case where there is
1304 * no rename, as this is only to set def_name
1305 * and a rename patch has the names elsewhere
1306 * in an unambiguous form.
1308 if (!name[len + 1])
1309 return NULL; /* no postimage name */
1310 second = skip_tree_prefix(state, name + len + 1,
1311 line_len - (len + 1));
1312 if (!second)
1313 return NULL;
1315 * Does len bytes starting at "name" and "second"
1316 * (that are separated by one HT or SP we just
1317 * found) exactly match?
1319 if (second[len] == '\n' && !strncmp(name, second, len))
1320 return xmemdupz(name, len);
1325 /* Verify that we recognize the lines following a git header */
1326 static int parse_git_header(struct apply_state *state,
1327 const char *line,
1328 int len,
1329 unsigned int size,
1330 struct patch *patch)
1332 unsigned long offset;
1334 /* A git diff has explicit new/delete information, so we don't guess */
1335 patch->is_new = 0;
1336 patch->is_delete = 0;
1339 * Some things may not have the old name in the
1340 * rest of the headers anywhere (pure mode changes,
1341 * or removing or adding empty files), so we get
1342 * the default name from the header.
1344 patch->def_name = git_header_name(state, line, len);
1345 if (patch->def_name && state->root.len) {
1346 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1347 free(patch->def_name);
1348 patch->def_name = s;
1351 line += len;
1352 size -= len;
1353 state->linenr++;
1354 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1355 static const struct opentry {
1356 const char *str;
1357 int (*fn)(struct apply_state *, const char *, struct patch *);
1358 } optable[] = {
1359 { "@@ -", gitdiff_hdrend },
1360 { "--- ", gitdiff_oldname },
1361 { "+++ ", gitdiff_newname },
1362 { "old mode ", gitdiff_oldmode },
1363 { "new mode ", gitdiff_newmode },
1364 { "deleted file mode ", gitdiff_delete },
1365 { "new file mode ", gitdiff_newfile },
1366 { "copy from ", gitdiff_copysrc },
1367 { "copy to ", gitdiff_copydst },
1368 { "rename old ", gitdiff_renamesrc },
1369 { "rename new ", gitdiff_renamedst },
1370 { "rename from ", gitdiff_renamesrc },
1371 { "rename to ", gitdiff_renamedst },
1372 { "similarity index ", gitdiff_similarity },
1373 { "dissimilarity index ", gitdiff_dissimilarity },
1374 { "index ", gitdiff_index },
1375 { "", gitdiff_unrecognized },
1377 int i;
1379 len = linelen(line, size);
1380 if (!len || line[len-1] != '\n')
1381 break;
1382 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1383 const struct opentry *p = optable + i;
1384 int oplen = strlen(p->str);
1385 int res;
1386 if (len < oplen || memcmp(p->str, line, oplen))
1387 continue;
1388 res = p->fn(state, line + oplen, patch);
1389 if (res < 0)
1390 return -1;
1391 if (res > 0)
1392 return offset;
1393 break;
1397 return offset;
1400 static int parse_num(const char *line, unsigned long *p)
1402 char *ptr;
1404 if (!isdigit(*line))
1405 return 0;
1406 *p = strtoul(line, &ptr, 10);
1407 return ptr - line;
1410 static int parse_range(const char *line, int len, int offset, const char *expect,
1411 unsigned long *p1, unsigned long *p2)
1413 int digits, ex;
1415 if (offset < 0 || offset >= len)
1416 return -1;
1417 line += offset;
1418 len -= offset;
1420 digits = parse_num(line, p1);
1421 if (!digits)
1422 return -1;
1424 offset += digits;
1425 line += digits;
1426 len -= digits;
1428 *p2 = 1;
1429 if (*line == ',') {
1430 digits = parse_num(line+1, p2);
1431 if (!digits)
1432 return -1;
1434 offset += digits+1;
1435 line += digits+1;
1436 len -= digits+1;
1439 ex = strlen(expect);
1440 if (ex > len)
1441 return -1;
1442 if (memcmp(line, expect, ex))
1443 return -1;
1445 return offset + ex;
1448 static void recount_diff(const char *line, int size, struct fragment *fragment)
1450 int oldlines = 0, newlines = 0, ret = 0;
1452 if (size < 1) {
1453 warning("recount: ignore empty hunk");
1454 return;
1457 for (;;) {
1458 int len = linelen(line, size);
1459 size -= len;
1460 line += len;
1462 if (size < 1)
1463 break;
1465 switch (*line) {
1466 case ' ': case '\n':
1467 newlines++;
1468 /* fall through */
1469 case '-':
1470 oldlines++;
1471 continue;
1472 case '+':
1473 newlines++;
1474 continue;
1475 case '\\':
1476 continue;
1477 case '@':
1478 ret = size < 3 || !starts_with(line, "@@ ");
1479 break;
1480 case 'd':
1481 ret = size < 5 || !starts_with(line, "diff ");
1482 break;
1483 default:
1484 ret = -1;
1485 break;
1487 if (ret) {
1488 warning(_("recount: unexpected line: %.*s"),
1489 (int)linelen(line, size), line);
1490 return;
1492 break;
1494 fragment->oldlines = oldlines;
1495 fragment->newlines = newlines;
1499 * Parse a unified diff fragment header of the
1500 * form "@@ -a,b +c,d @@"
1502 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1504 int offset;
1506 if (!len || line[len-1] != '\n')
1507 return -1;
1509 /* Figure out the number of lines in a fragment */
1510 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1511 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1513 return offset;
1517 * Find file diff header
1519 * Returns:
1520 * -1 if no header was found
1521 * -128 in case of error
1522 * the size of the header in bytes (called "offset") otherwise
1524 static int find_header(struct apply_state *state,
1525 const char *line,
1526 unsigned long size,
1527 int *hdrsize,
1528 struct patch *patch)
1530 unsigned long offset, len;
1532 patch->is_toplevel_relative = 0;
1533 patch->is_rename = patch->is_copy = 0;
1534 patch->is_new = patch->is_delete = -1;
1535 patch->old_mode = patch->new_mode = 0;
1536 patch->old_name = patch->new_name = NULL;
1537 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1538 unsigned long nextlen;
1540 len = linelen(line, size);
1541 if (!len)
1542 break;
1544 /* Testing this early allows us to take a few shortcuts.. */
1545 if (len < 6)
1546 continue;
1549 * Make sure we don't find any unconnected patch fragments.
1550 * That's a sign that we didn't find a header, and that a
1551 * patch has become corrupted/broken up.
1553 if (!memcmp("@@ -", line, 4)) {
1554 struct fragment dummy;
1555 if (parse_fragment_header(line, len, &dummy) < 0)
1556 continue;
1557 error(_("patch fragment without header at line %d: %.*s"),
1558 state->linenr, (int)len-1, line);
1559 return -128;
1562 if (size < len + 6)
1563 break;
1566 * Git patch? It might not have a real patch, just a rename
1567 * or mode change, so we handle that specially
1569 if (!memcmp("diff --git ", line, 11)) {
1570 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1571 if (git_hdr_len < 0)
1572 return -128;
1573 if (git_hdr_len <= len)
1574 continue;
1575 if (!patch->old_name && !patch->new_name) {
1576 if (!patch->def_name) {
1577 error(Q_("git diff header lacks filename information when removing "
1578 "%d leading pathname component (line %d)",
1579 "git diff header lacks filename information when removing "
1580 "%d leading pathname components (line %d)",
1581 state->p_value),
1582 state->p_value, state->linenr);
1583 return -128;
1585 patch->old_name = xstrdup(patch->def_name);
1586 patch->new_name = xstrdup(patch->def_name);
1588 if (!patch->is_delete && !patch->new_name) {
1589 error(_("git diff header lacks filename information "
1590 "(line %d)"), state->linenr);
1591 return -128;
1593 patch->is_toplevel_relative = 1;
1594 *hdrsize = git_hdr_len;
1595 return offset;
1598 /* --- followed by +++ ? */
1599 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1600 continue;
1603 * We only accept unified patches, so we want it to
1604 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1605 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1607 nextlen = linelen(line + len, size - len);
1608 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1609 continue;
1611 /* Ok, we'll consider it a patch */
1612 if (parse_traditional_patch(state, line, line+len, patch))
1613 return -128;
1614 *hdrsize = len + nextlen;
1615 state->linenr += 2;
1616 return offset;
1618 return -1;
1621 static void record_ws_error(struct apply_state *state,
1622 unsigned result,
1623 const char *line,
1624 int len,
1625 int linenr)
1627 char *err;
1629 if (!result)
1630 return;
1632 state->whitespace_error++;
1633 if (state->squelch_whitespace_errors &&
1634 state->squelch_whitespace_errors < state->whitespace_error)
1635 return;
1637 err = whitespace_error_string(result);
1638 if (state->apply_verbosity > verbosity_silent)
1639 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1640 state->patch_input_file, linenr, err, len, line);
1641 free(err);
1644 static void check_whitespace(struct apply_state *state,
1645 const char *line,
1646 int len,
1647 unsigned ws_rule)
1649 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1651 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1655 * Parse a unified diff. Note that this really needs to parse each
1656 * fragment separately, since the only way to know the difference
1657 * between a "---" that is part of a patch, and a "---" that starts
1658 * the next patch is to look at the line counts..
1660 static int parse_fragment(struct apply_state *state,
1661 const char *line,
1662 unsigned long size,
1663 struct patch *patch,
1664 struct fragment *fragment)
1666 int added, deleted;
1667 int len = linelen(line, size), offset;
1668 unsigned long oldlines, newlines;
1669 unsigned long leading, trailing;
1671 offset = parse_fragment_header(line, len, fragment);
1672 if (offset < 0)
1673 return -1;
1674 if (offset > 0 && patch->recount)
1675 recount_diff(line + offset, size - offset, fragment);
1676 oldlines = fragment->oldlines;
1677 newlines = fragment->newlines;
1678 leading = 0;
1679 trailing = 0;
1681 /* Parse the thing.. */
1682 line += len;
1683 size -= len;
1684 state->linenr++;
1685 added = deleted = 0;
1686 for (offset = len;
1687 0 < size;
1688 offset += len, size -= len, line += len, state->linenr++) {
1689 if (!oldlines && !newlines)
1690 break;
1691 len = linelen(line, size);
1692 if (!len || line[len-1] != '\n')
1693 return -1;
1694 switch (*line) {
1695 default:
1696 return -1;
1697 case '\n': /* newer GNU diff, an empty context line */
1698 case ' ':
1699 oldlines--;
1700 newlines--;
1701 if (!deleted && !added)
1702 leading++;
1703 trailing++;
1704 if (!state->apply_in_reverse &&
1705 state->ws_error_action == correct_ws_error)
1706 check_whitespace(state, line, len, patch->ws_rule);
1707 break;
1708 case '-':
1709 if (state->apply_in_reverse &&
1710 state->ws_error_action != nowarn_ws_error)
1711 check_whitespace(state, line, len, patch->ws_rule);
1712 deleted++;
1713 oldlines--;
1714 trailing = 0;
1715 break;
1716 case '+':
1717 if (!state->apply_in_reverse &&
1718 state->ws_error_action != nowarn_ws_error)
1719 check_whitespace(state, line, len, patch->ws_rule);
1720 added++;
1721 newlines--;
1722 trailing = 0;
1723 break;
1726 * We allow "\ No newline at end of file". Depending
1727 * on locale settings when the patch was produced we
1728 * don't know what this line looks like. The only
1729 * thing we do know is that it begins with "\ ".
1730 * Checking for 12 is just for sanity check -- any
1731 * l10n of "\ No newline..." is at least that long.
1733 case '\\':
1734 if (len < 12 || memcmp(line, "\\ ", 2))
1735 return -1;
1736 break;
1739 if (oldlines || newlines)
1740 return -1;
1741 if (!deleted && !added)
1742 return -1;
1744 fragment->leading = leading;
1745 fragment->trailing = trailing;
1748 * If a fragment ends with an incomplete line, we failed to include
1749 * it in the above loop because we hit oldlines == newlines == 0
1750 * before seeing it.
1752 if (12 < size && !memcmp(line, "\\ ", 2))
1753 offset += linelen(line, size);
1755 patch->lines_added += added;
1756 patch->lines_deleted += deleted;
1758 if (0 < patch->is_new && oldlines)
1759 return error(_("new file depends on old contents"));
1760 if (0 < patch->is_delete && newlines)
1761 return error(_("deleted file still has contents"));
1762 return offset;
1766 * We have seen "diff --git a/... b/..." header (or a traditional patch
1767 * header). Read hunks that belong to this patch into fragments and hang
1768 * them to the given patch structure.
1770 * The (fragment->patch, fragment->size) pair points into the memory given
1771 * by the caller, not a copy, when we return.
1773 * Returns:
1774 * -1 in case of error,
1775 * the number of bytes in the patch otherwise.
1777 static int parse_single_patch(struct apply_state *state,
1778 const char *line,
1779 unsigned long size,
1780 struct patch *patch)
1782 unsigned long offset = 0;
1783 unsigned long oldlines = 0, newlines = 0, context = 0;
1784 struct fragment **fragp = &patch->fragments;
1786 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1787 struct fragment *fragment;
1788 int len;
1790 fragment = xcalloc(1, sizeof(*fragment));
1791 fragment->linenr = state->linenr;
1792 len = parse_fragment(state, line, size, patch, fragment);
1793 if (len <= 0) {
1794 free(fragment);
1795 return error(_("corrupt patch at line %d"), state->linenr);
1797 fragment->patch = line;
1798 fragment->size = len;
1799 oldlines += fragment->oldlines;
1800 newlines += fragment->newlines;
1801 context += fragment->leading + fragment->trailing;
1803 *fragp = fragment;
1804 fragp = &fragment->next;
1806 offset += len;
1807 line += len;
1808 size -= len;
1812 * If something was removed (i.e. we have old-lines) it cannot
1813 * be creation, and if something was added it cannot be
1814 * deletion. However, the reverse is not true; --unified=0
1815 * patches that only add are not necessarily creation even
1816 * though they do not have any old lines, and ones that only
1817 * delete are not necessarily deletion.
1819 * Unfortunately, a real creation/deletion patch do _not_ have
1820 * any context line by definition, so we cannot safely tell it
1821 * apart with --unified=0 insanity. At least if the patch has
1822 * more than one hunk it is not creation or deletion.
1824 if (patch->is_new < 0 &&
1825 (oldlines || (patch->fragments && patch->fragments->next)))
1826 patch->is_new = 0;
1827 if (patch->is_delete < 0 &&
1828 (newlines || (patch->fragments && patch->fragments->next)))
1829 patch->is_delete = 0;
1831 if (0 < patch->is_new && oldlines)
1832 return error(_("new file %s depends on old contents"), patch->new_name);
1833 if (0 < patch->is_delete && newlines)
1834 return error(_("deleted file %s still has contents"), patch->old_name);
1835 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1836 fprintf_ln(stderr,
1837 _("** warning: "
1838 "file %s becomes empty but is not deleted"),
1839 patch->new_name);
1841 return offset;
1844 static inline int metadata_changes(struct patch *patch)
1846 return patch->is_rename > 0 ||
1847 patch->is_copy > 0 ||
1848 patch->is_new > 0 ||
1849 patch->is_delete ||
1850 (patch->old_mode && patch->new_mode &&
1851 patch->old_mode != patch->new_mode);
1854 static char *inflate_it(const void *data, unsigned long size,
1855 unsigned long inflated_size)
1857 git_zstream stream;
1858 void *out;
1859 int st;
1861 memset(&stream, 0, sizeof(stream));
1863 stream.next_in = (unsigned char *)data;
1864 stream.avail_in = size;
1865 stream.next_out = out = xmalloc(inflated_size);
1866 stream.avail_out = inflated_size;
1867 git_inflate_init(&stream);
1868 st = git_inflate(&stream, Z_FINISH);
1869 git_inflate_end(&stream);
1870 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1871 free(out);
1872 return NULL;
1874 return out;
1878 * Read a binary hunk and return a new fragment; fragment->patch
1879 * points at an allocated memory that the caller must free, so
1880 * it is marked as "->free_patch = 1".
1882 static struct fragment *parse_binary_hunk(struct apply_state *state,
1883 char **buf_p,
1884 unsigned long *sz_p,
1885 int *status_p,
1886 int *used_p)
1889 * Expect a line that begins with binary patch method ("literal"
1890 * or "delta"), followed by the length of data before deflating.
1891 * a sequence of 'length-byte' followed by base-85 encoded data
1892 * should follow, terminated by a newline.
1894 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1895 * and we would limit the patch line to 66 characters,
1896 * so one line can fit up to 13 groups that would decode
1897 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1898 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1900 int llen, used;
1901 unsigned long size = *sz_p;
1902 char *buffer = *buf_p;
1903 int patch_method;
1904 unsigned long origlen;
1905 char *data = NULL;
1906 int hunk_size = 0;
1907 struct fragment *frag;
1909 llen = linelen(buffer, size);
1910 used = llen;
1912 *status_p = 0;
1914 if (starts_with(buffer, "delta ")) {
1915 patch_method = BINARY_DELTA_DEFLATED;
1916 origlen = strtoul(buffer + 6, NULL, 10);
1918 else if (starts_with(buffer, "literal ")) {
1919 patch_method = BINARY_LITERAL_DEFLATED;
1920 origlen = strtoul(buffer + 8, NULL, 10);
1922 else
1923 return NULL;
1925 state->linenr++;
1926 buffer += llen;
1927 while (1) {
1928 int byte_length, max_byte_length, newsize;
1929 llen = linelen(buffer, size);
1930 used += llen;
1931 state->linenr++;
1932 if (llen == 1) {
1933 /* consume the blank line */
1934 buffer++;
1935 size--;
1936 break;
1939 * Minimum line is "A00000\n" which is 7-byte long,
1940 * and the line length must be multiple of 5 plus 2.
1942 if ((llen < 7) || (llen-2) % 5)
1943 goto corrupt;
1944 max_byte_length = (llen - 2) / 5 * 4;
1945 byte_length = *buffer;
1946 if ('A' <= byte_length && byte_length <= 'Z')
1947 byte_length = byte_length - 'A' + 1;
1948 else if ('a' <= byte_length && byte_length <= 'z')
1949 byte_length = byte_length - 'a' + 27;
1950 else
1951 goto corrupt;
1952 /* if the input length was not multiple of 4, we would
1953 * have filler at the end but the filler should never
1954 * exceed 3 bytes
1956 if (max_byte_length < byte_length ||
1957 byte_length <= max_byte_length - 4)
1958 goto corrupt;
1959 newsize = hunk_size + byte_length;
1960 data = xrealloc(data, newsize);
1961 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1962 goto corrupt;
1963 hunk_size = newsize;
1964 buffer += llen;
1965 size -= llen;
1968 frag = xcalloc(1, sizeof(*frag));
1969 frag->patch = inflate_it(data, hunk_size, origlen);
1970 frag->free_patch = 1;
1971 if (!frag->patch)
1972 goto corrupt;
1973 free(data);
1974 frag->size = origlen;
1975 *buf_p = buffer;
1976 *sz_p = size;
1977 *used_p = used;
1978 frag->binary_patch_method = patch_method;
1979 return frag;
1981 corrupt:
1982 free(data);
1983 *status_p = -1;
1984 error(_("corrupt binary patch at line %d: %.*s"),
1985 state->linenr-1, llen-1, buffer);
1986 return NULL;
1990 * Returns:
1991 * -1 in case of error,
1992 * the length of the parsed binary patch otherwise
1994 static int parse_binary(struct apply_state *state,
1995 char *buffer,
1996 unsigned long size,
1997 struct patch *patch)
2000 * We have read "GIT binary patch\n"; what follows is a line
2001 * that says the patch method (currently, either "literal" or
2002 * "delta") and the length of data before deflating; a
2003 * sequence of 'length-byte' followed by base-85 encoded data
2004 * follows.
2006 * When a binary patch is reversible, there is another binary
2007 * hunk in the same format, starting with patch method (either
2008 * "literal" or "delta") with the length of data, and a sequence
2009 * of length-byte + base-85 encoded data, terminated with another
2010 * empty line. This data, when applied to the postimage, produces
2011 * the preimage.
2013 struct fragment *forward;
2014 struct fragment *reverse;
2015 int status;
2016 int used, used_1;
2018 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2019 if (!forward && !status)
2020 /* there has to be one hunk (forward hunk) */
2021 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2022 if (status)
2023 /* otherwise we already gave an error message */
2024 return status;
2026 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2027 if (reverse)
2028 used += used_1;
2029 else if (status) {
2031 * Not having reverse hunk is not an error, but having
2032 * a corrupt reverse hunk is.
2034 free((void*) forward->patch);
2035 free(forward);
2036 return status;
2038 forward->next = reverse;
2039 patch->fragments = forward;
2040 patch->is_binary = 1;
2041 return used;
2044 static void prefix_one(struct apply_state *state, char **name)
2046 char *old_name = *name;
2047 if (!old_name)
2048 return;
2049 *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));
2050 free(old_name);
2053 static void prefix_patch(struct apply_state *state, struct patch *p)
2055 if (!state->prefix || p->is_toplevel_relative)
2056 return;
2057 prefix_one(state, &p->new_name);
2058 prefix_one(state, &p->old_name);
2062 * include/exclude
2065 static void add_name_limit(struct apply_state *state,
2066 const char *name,
2067 int exclude)
2069 struct string_list_item *it;
2071 it = string_list_append(&state->limit_by_name, name);
2072 it->util = exclude ? NULL : (void *) 1;
2075 static int use_patch(struct apply_state *state, struct patch *p)
2077 const char *pathname = p->new_name ? p->new_name : p->old_name;
2078 int i;
2080 /* Paths outside are not touched regardless of "--include" */
2081 if (0 < state->prefix_length) {
2082 int pathlen = strlen(pathname);
2083 if (pathlen <= state->prefix_length ||
2084 memcmp(state->prefix, pathname, state->prefix_length))
2085 return 0;
2088 /* See if it matches any of exclude/include rule */
2089 for (i = 0; i < state->limit_by_name.nr; i++) {
2090 struct string_list_item *it = &state->limit_by_name.items[i];
2091 if (!wildmatch(it->string, pathname, 0, NULL))
2092 return (it->util != NULL);
2096 * If we had any include, a path that does not match any rule is
2097 * not used. Otherwise, we saw bunch of exclude rules (or none)
2098 * and such a path is used.
2100 return !state->has_include;
2104 * Read the patch text in "buffer" that extends for "size" bytes; stop
2105 * reading after seeing a single patch (i.e. changes to a single file).
2106 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2108 * Returns:
2109 * -1 if no header was found or parse_binary() failed,
2110 * -128 on another error,
2111 * the number of bytes consumed otherwise,
2112 * so that the caller can call us again for the next patch.
2114 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2116 int hdrsize, patchsize;
2117 int offset = find_header(state, buffer, size, &hdrsize, patch);
2119 if (offset < 0)
2120 return offset;
2122 prefix_patch(state, patch);
2124 if (!use_patch(state, patch))
2125 patch->ws_rule = 0;
2126 else
2127 patch->ws_rule = whitespace_rule(patch->new_name
2128 ? patch->new_name
2129 : patch->old_name);
2131 patchsize = parse_single_patch(state,
2132 buffer + offset + hdrsize,
2133 size - offset - hdrsize,
2134 patch);
2136 if (patchsize < 0)
2137 return -128;
2139 if (!patchsize) {
2140 static const char git_binary[] = "GIT binary patch\n";
2141 int hd = hdrsize + offset;
2142 unsigned long llen = linelen(buffer + hd, size - hd);
2144 if (llen == sizeof(git_binary) - 1 &&
2145 !memcmp(git_binary, buffer + hd, llen)) {
2146 int used;
2147 state->linenr++;
2148 used = parse_binary(state, buffer + hd + llen,
2149 size - hd - llen, patch);
2150 if (used < 0)
2151 return -1;
2152 if (used)
2153 patchsize = used + llen;
2154 else
2155 patchsize = 0;
2157 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2158 static const char *binhdr[] = {
2159 "Binary files ",
2160 "Files ",
2161 NULL,
2163 int i;
2164 for (i = 0; binhdr[i]; i++) {
2165 int len = strlen(binhdr[i]);
2166 if (len < size - hd &&
2167 !memcmp(binhdr[i], buffer + hd, len)) {
2168 state->linenr++;
2169 patch->is_binary = 1;
2170 patchsize = llen;
2171 break;
2176 /* Empty patch cannot be applied if it is a text patch
2177 * without metadata change. A binary patch appears
2178 * empty to us here.
2180 if ((state->apply || state->check) &&
2181 (!patch->is_binary && !metadata_changes(patch))) {
2182 error(_("patch with only garbage at line %d"), state->linenr);
2183 return -128;
2187 return offset + hdrsize + patchsize;
2190 #define swap(a,b) myswap((a),(b),sizeof(a))
2192 #define myswap(a, b, size) do { \
2193 unsigned char mytmp[size]; \
2194 memcpy(mytmp, &a, size); \
2195 memcpy(&a, &b, size); \
2196 memcpy(&b, mytmp, size); \
2197 } while (0)
2199 static void reverse_patches(struct patch *p)
2201 for (; p; p = p->next) {
2202 struct fragment *frag = p->fragments;
2204 swap(p->new_name, p->old_name);
2205 swap(p->new_mode, p->old_mode);
2206 swap(p->is_new, p->is_delete);
2207 swap(p->lines_added, p->lines_deleted);
2208 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2210 for (; frag; frag = frag->next) {
2211 swap(frag->newpos, frag->oldpos);
2212 swap(frag->newlines, frag->oldlines);
2217 static const char pluses[] =
2218 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2219 static const char minuses[]=
2220 "----------------------------------------------------------------------";
2222 static void show_stats(struct apply_state *state, struct patch *patch)
2224 struct strbuf qname = STRBUF_INIT;
2225 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2226 int max, add, del;
2228 quote_c_style(cp, &qname, NULL, 0);
2231 * "scale" the filename
2233 max = state->max_len;
2234 if (max > 50)
2235 max = 50;
2237 if (qname.len > max) {
2238 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2239 if (!cp)
2240 cp = qname.buf + qname.len + 3 - max;
2241 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2244 if (patch->is_binary) {
2245 printf(" %-*s | Bin\n", max, qname.buf);
2246 strbuf_release(&qname);
2247 return;
2250 printf(" %-*s |", max, qname.buf);
2251 strbuf_release(&qname);
2254 * scale the add/delete
2256 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2257 add = patch->lines_added;
2258 del = patch->lines_deleted;
2260 if (state->max_change > 0) {
2261 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2262 add = (add * max + state->max_change / 2) / state->max_change;
2263 del = total - add;
2265 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2266 add, pluses, del, minuses);
2269 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2271 switch (st->st_mode & S_IFMT) {
2272 case S_IFLNK:
2273 if (strbuf_readlink(buf, path, st->st_size) < 0)
2274 return error(_("unable to read symlink %s"), path);
2275 return 0;
2276 case S_IFREG:
2277 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2278 return error(_("unable to open or read %s"), path);
2279 convert_to_git(path, buf->buf, buf->len, buf, 0);
2280 return 0;
2281 default:
2282 return -1;
2287 * Update the preimage, and the common lines in postimage,
2288 * from buffer buf of length len. If postlen is 0 the postimage
2289 * is updated in place, otherwise it's updated on a new buffer
2290 * of length postlen
2293 static void update_pre_post_images(struct image *preimage,
2294 struct image *postimage,
2295 char *buf,
2296 size_t len, size_t postlen)
2298 int i, ctx, reduced;
2299 char *new, *old, *fixed;
2300 struct image fixed_preimage;
2303 * Update the preimage with whitespace fixes. Note that we
2304 * are not losing preimage->buf -- apply_one_fragment() will
2305 * free "oldlines".
2307 prepare_image(&fixed_preimage, buf, len, 1);
2308 assert(postlen
2309 ? fixed_preimage.nr == preimage->nr
2310 : fixed_preimage.nr <= preimage->nr);
2311 for (i = 0; i < fixed_preimage.nr; i++)
2312 fixed_preimage.line[i].flag = preimage->line[i].flag;
2313 free(preimage->line_allocated);
2314 *preimage = fixed_preimage;
2317 * Adjust the common context lines in postimage. This can be
2318 * done in-place when we are shrinking it with whitespace
2319 * fixing, but needs a new buffer when ignoring whitespace or
2320 * expanding leading tabs to spaces.
2322 * We trust the caller to tell us if the update can be done
2323 * in place (postlen==0) or not.
2325 old = postimage->buf;
2326 if (postlen)
2327 new = postimage->buf = xmalloc(postlen);
2328 else
2329 new = old;
2330 fixed = preimage->buf;
2332 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2333 size_t l_len = postimage->line[i].len;
2334 if (!(postimage->line[i].flag & LINE_COMMON)) {
2335 /* an added line -- no counterparts in preimage */
2336 memmove(new, old, l_len);
2337 old += l_len;
2338 new += l_len;
2339 continue;
2342 /* a common context -- skip it in the original postimage */
2343 old += l_len;
2345 /* and find the corresponding one in the fixed preimage */
2346 while (ctx < preimage->nr &&
2347 !(preimage->line[ctx].flag & LINE_COMMON)) {
2348 fixed += preimage->line[ctx].len;
2349 ctx++;
2353 * preimage is expected to run out, if the caller
2354 * fixed addition of trailing blank lines.
2356 if (preimage->nr <= ctx) {
2357 reduced++;
2358 continue;
2361 /* and copy it in, while fixing the line length */
2362 l_len = preimage->line[ctx].len;
2363 memcpy(new, fixed, l_len);
2364 new += l_len;
2365 fixed += l_len;
2366 postimage->line[i].len = l_len;
2367 ctx++;
2370 if (postlen
2371 ? postlen < new - postimage->buf
2372 : postimage->len < new - postimage->buf)
2373 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2374 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2376 /* Fix the length of the whole thing */
2377 postimage->len = new - postimage->buf;
2378 postimage->nr -= reduced;
2381 static int line_by_line_fuzzy_match(struct image *img,
2382 struct image *preimage,
2383 struct image *postimage,
2384 unsigned long try,
2385 int try_lno,
2386 int preimage_limit)
2388 int i;
2389 size_t imgoff = 0;
2390 size_t preoff = 0;
2391 size_t postlen = postimage->len;
2392 size_t extra_chars;
2393 char *buf;
2394 char *preimage_eof;
2395 char *preimage_end;
2396 struct strbuf fixed;
2397 char *fixed_buf;
2398 size_t fixed_len;
2400 for (i = 0; i < preimage_limit; i++) {
2401 size_t prelen = preimage->line[i].len;
2402 size_t imglen = img->line[try_lno+i].len;
2404 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2405 preimage->buf + preoff, prelen))
2406 return 0;
2407 if (preimage->line[i].flag & LINE_COMMON)
2408 postlen += imglen - prelen;
2409 imgoff += imglen;
2410 preoff += prelen;
2414 * Ok, the preimage matches with whitespace fuzz.
2416 * imgoff now holds the true length of the target that
2417 * matches the preimage before the end of the file.
2419 * Count the number of characters in the preimage that fall
2420 * beyond the end of the file and make sure that all of them
2421 * are whitespace characters. (This can only happen if
2422 * we are removing blank lines at the end of the file.)
2424 buf = preimage_eof = preimage->buf + preoff;
2425 for ( ; i < preimage->nr; i++)
2426 preoff += preimage->line[i].len;
2427 preimage_end = preimage->buf + preoff;
2428 for ( ; buf < preimage_end; buf++)
2429 if (!isspace(*buf))
2430 return 0;
2433 * Update the preimage and the common postimage context
2434 * lines to use the same whitespace as the target.
2435 * If whitespace is missing in the target (i.e.
2436 * if the preimage extends beyond the end of the file),
2437 * use the whitespace from the preimage.
2439 extra_chars = preimage_end - preimage_eof;
2440 strbuf_init(&fixed, imgoff + extra_chars);
2441 strbuf_add(&fixed, img->buf + try, imgoff);
2442 strbuf_add(&fixed, preimage_eof, extra_chars);
2443 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2444 update_pre_post_images(preimage, postimage,
2445 fixed_buf, fixed_len, postlen);
2446 return 1;
2449 static int match_fragment(struct apply_state *state,
2450 struct image *img,
2451 struct image *preimage,
2452 struct image *postimage,
2453 unsigned long try,
2454 int try_lno,
2455 unsigned ws_rule,
2456 int match_beginning, int match_end)
2458 int i;
2459 char *fixed_buf, *buf, *orig, *target;
2460 struct strbuf fixed;
2461 size_t fixed_len, postlen;
2462 int preimage_limit;
2464 if (preimage->nr + try_lno <= img->nr) {
2466 * The hunk falls within the boundaries of img.
2468 preimage_limit = preimage->nr;
2469 if (match_end && (preimage->nr + try_lno != img->nr))
2470 return 0;
2471 } else if (state->ws_error_action == correct_ws_error &&
2472 (ws_rule & WS_BLANK_AT_EOF)) {
2474 * This hunk extends beyond the end of img, and we are
2475 * removing blank lines at the end of the file. This
2476 * many lines from the beginning of the preimage must
2477 * match with img, and the remainder of the preimage
2478 * must be blank.
2480 preimage_limit = img->nr - try_lno;
2481 } else {
2483 * The hunk extends beyond the end of the img and
2484 * we are not removing blanks at the end, so we
2485 * should reject the hunk at this position.
2487 return 0;
2490 if (match_beginning && try_lno)
2491 return 0;
2493 /* Quick hash check */
2494 for (i = 0; i < preimage_limit; i++)
2495 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2496 (preimage->line[i].hash != img->line[try_lno + i].hash))
2497 return 0;
2499 if (preimage_limit == preimage->nr) {
2501 * Do we have an exact match? If we were told to match
2502 * at the end, size must be exactly at try+fragsize,
2503 * otherwise try+fragsize must be still within the preimage,
2504 * and either case, the old piece should match the preimage
2505 * exactly.
2507 if ((match_end
2508 ? (try + preimage->len == img->len)
2509 : (try + preimage->len <= img->len)) &&
2510 !memcmp(img->buf + try, preimage->buf, preimage->len))
2511 return 1;
2512 } else {
2514 * The preimage extends beyond the end of img, so
2515 * there cannot be an exact match.
2517 * There must be one non-blank context line that match
2518 * a line before the end of img.
2520 char *buf_end;
2522 buf = preimage->buf;
2523 buf_end = buf;
2524 for (i = 0; i < preimage_limit; i++)
2525 buf_end += preimage->line[i].len;
2527 for ( ; buf < buf_end; buf++)
2528 if (!isspace(*buf))
2529 break;
2530 if (buf == buf_end)
2531 return 0;
2535 * No exact match. If we are ignoring whitespace, run a line-by-line
2536 * fuzzy matching. We collect all the line length information because
2537 * we need it to adjust whitespace if we match.
2539 if (state->ws_ignore_action == ignore_ws_change)
2540 return line_by_line_fuzzy_match(img, preimage, postimage,
2541 try, try_lno, preimage_limit);
2543 if (state->ws_error_action != correct_ws_error)
2544 return 0;
2547 * The hunk does not apply byte-by-byte, but the hash says
2548 * it might with whitespace fuzz. We weren't asked to
2549 * ignore whitespace, we were asked to correct whitespace
2550 * errors, so let's try matching after whitespace correction.
2552 * While checking the preimage against the target, whitespace
2553 * errors in both fixed, we count how large the corresponding
2554 * postimage needs to be. The postimage prepared by
2555 * apply_one_fragment() has whitespace errors fixed on added
2556 * lines already, but the common lines were propagated as-is,
2557 * which may become longer when their whitespace errors are
2558 * fixed.
2561 /* First count added lines in postimage */
2562 postlen = 0;
2563 for (i = 0; i < postimage->nr; i++) {
2564 if (!(postimage->line[i].flag & LINE_COMMON))
2565 postlen += postimage->line[i].len;
2569 * The preimage may extend beyond the end of the file,
2570 * but in this loop we will only handle the part of the
2571 * preimage that falls within the file.
2573 strbuf_init(&fixed, preimage->len + 1);
2574 orig = preimage->buf;
2575 target = img->buf + try;
2576 for (i = 0; i < preimage_limit; i++) {
2577 size_t oldlen = preimage->line[i].len;
2578 size_t tgtlen = img->line[try_lno + i].len;
2579 size_t fixstart = fixed.len;
2580 struct strbuf tgtfix;
2581 int match;
2583 /* Try fixing the line in the preimage */
2584 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2586 /* Try fixing the line in the target */
2587 strbuf_init(&tgtfix, tgtlen);
2588 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2591 * If they match, either the preimage was based on
2592 * a version before our tree fixed whitespace breakage,
2593 * or we are lacking a whitespace-fix patch the tree
2594 * the preimage was based on already had (i.e. target
2595 * has whitespace breakage, the preimage doesn't).
2596 * In either case, we are fixing the whitespace breakages
2597 * so we might as well take the fix together with their
2598 * real change.
2600 match = (tgtfix.len == fixed.len - fixstart &&
2601 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2602 fixed.len - fixstart));
2604 /* Add the length if this is common with the postimage */
2605 if (preimage->line[i].flag & LINE_COMMON)
2606 postlen += tgtfix.len;
2608 strbuf_release(&tgtfix);
2609 if (!match)
2610 goto unmatch_exit;
2612 orig += oldlen;
2613 target += tgtlen;
2618 * Now handle the lines in the preimage that falls beyond the
2619 * end of the file (if any). They will only match if they are
2620 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2621 * false).
2623 for ( ; i < preimage->nr; i++) {
2624 size_t fixstart = fixed.len; /* start of the fixed preimage */
2625 size_t oldlen = preimage->line[i].len;
2626 int j;
2628 /* Try fixing the line in the preimage */
2629 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2631 for (j = fixstart; j < fixed.len; j++)
2632 if (!isspace(fixed.buf[j]))
2633 goto unmatch_exit;
2635 orig += oldlen;
2639 * Yes, the preimage is based on an older version that still
2640 * has whitespace breakages unfixed, and fixing them makes the
2641 * hunk match. Update the context lines in the postimage.
2643 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2644 if (postlen < postimage->len)
2645 postlen = 0;
2646 update_pre_post_images(preimage, postimage,
2647 fixed_buf, fixed_len, postlen);
2648 return 1;
2650 unmatch_exit:
2651 strbuf_release(&fixed);
2652 return 0;
2655 static int find_pos(struct apply_state *state,
2656 struct image *img,
2657 struct image *preimage,
2658 struct image *postimage,
2659 int line,
2660 unsigned ws_rule,
2661 int match_beginning, int match_end)
2663 int i;
2664 unsigned long backwards, forwards, try;
2665 int backwards_lno, forwards_lno, try_lno;
2668 * If match_beginning or match_end is specified, there is no
2669 * point starting from a wrong line that will never match and
2670 * wander around and wait for a match at the specified end.
2672 if (match_beginning)
2673 line = 0;
2674 else if (match_end)
2675 line = img->nr - preimage->nr;
2678 * Because the comparison is unsigned, the following test
2679 * will also take care of a negative line number that can
2680 * result when match_end and preimage is larger than the target.
2682 if ((size_t) line > img->nr)
2683 line = img->nr;
2685 try = 0;
2686 for (i = 0; i < line; i++)
2687 try += img->line[i].len;
2690 * There's probably some smart way to do this, but I'll leave
2691 * that to the smart and beautiful people. I'm simple and stupid.
2693 backwards = try;
2694 backwards_lno = line;
2695 forwards = try;
2696 forwards_lno = line;
2697 try_lno = line;
2699 for (i = 0; ; i++) {
2700 if (match_fragment(state, img, preimage, postimage,
2701 try, try_lno, ws_rule,
2702 match_beginning, match_end))
2703 return try_lno;
2705 again:
2706 if (backwards_lno == 0 && forwards_lno == img->nr)
2707 break;
2709 if (i & 1) {
2710 if (backwards_lno == 0) {
2711 i++;
2712 goto again;
2714 backwards_lno--;
2715 backwards -= img->line[backwards_lno].len;
2716 try = backwards;
2717 try_lno = backwards_lno;
2718 } else {
2719 if (forwards_lno == img->nr) {
2720 i++;
2721 goto again;
2723 forwards += img->line[forwards_lno].len;
2724 forwards_lno++;
2725 try = forwards;
2726 try_lno = forwards_lno;
2730 return -1;
2733 static void remove_first_line(struct image *img)
2735 img->buf += img->line[0].len;
2736 img->len -= img->line[0].len;
2737 img->line++;
2738 img->nr--;
2741 static void remove_last_line(struct image *img)
2743 img->len -= img->line[--img->nr].len;
2747 * The change from "preimage" and "postimage" has been found to
2748 * apply at applied_pos (counts in line numbers) in "img".
2749 * Update "img" to remove "preimage" and replace it with "postimage".
2751 static void update_image(struct apply_state *state,
2752 struct image *img,
2753 int applied_pos,
2754 struct image *preimage,
2755 struct image *postimage)
2758 * remove the copy of preimage at offset in img
2759 * and replace it with postimage
2761 int i, nr;
2762 size_t remove_count, insert_count, applied_at = 0;
2763 char *result;
2764 int preimage_limit;
2767 * If we are removing blank lines at the end of img,
2768 * the preimage may extend beyond the end.
2769 * If that is the case, we must be careful only to
2770 * remove the part of the preimage that falls within
2771 * the boundaries of img. Initialize preimage_limit
2772 * to the number of lines in the preimage that falls
2773 * within the boundaries.
2775 preimage_limit = preimage->nr;
2776 if (preimage_limit > img->nr - applied_pos)
2777 preimage_limit = img->nr - applied_pos;
2779 for (i = 0; i < applied_pos; i++)
2780 applied_at += img->line[i].len;
2782 remove_count = 0;
2783 for (i = 0; i < preimage_limit; i++)
2784 remove_count += img->line[applied_pos + i].len;
2785 insert_count = postimage->len;
2787 /* Adjust the contents */
2788 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2789 memcpy(result, img->buf, applied_at);
2790 memcpy(result + applied_at, postimage->buf, postimage->len);
2791 memcpy(result + applied_at + postimage->len,
2792 img->buf + (applied_at + remove_count),
2793 img->len - (applied_at + remove_count));
2794 free(img->buf);
2795 img->buf = result;
2796 img->len += insert_count - remove_count;
2797 result[img->len] = '\0';
2799 /* Adjust the line table */
2800 nr = img->nr + postimage->nr - preimage_limit;
2801 if (preimage_limit < postimage->nr) {
2803 * NOTE: this knows that we never call remove_first_line()
2804 * on anything other than pre/post image.
2806 REALLOC_ARRAY(img->line, nr);
2807 img->line_allocated = img->line;
2809 if (preimage_limit != postimage->nr)
2810 memmove(img->line + applied_pos + postimage->nr,
2811 img->line + applied_pos + preimage_limit,
2812 (img->nr - (applied_pos + preimage_limit)) *
2813 sizeof(*img->line));
2814 memcpy(img->line + applied_pos,
2815 postimage->line,
2816 postimage->nr * sizeof(*img->line));
2817 if (!state->allow_overlap)
2818 for (i = 0; i < postimage->nr; i++)
2819 img->line[applied_pos + i].flag |= LINE_PATCHED;
2820 img->nr = nr;
2824 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2825 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2826 * replace the part of "img" with "postimage" text.
2828 static int apply_one_fragment(struct apply_state *state,
2829 struct image *img, struct fragment *frag,
2830 int inaccurate_eof, unsigned ws_rule,
2831 int nth_fragment)
2833 int match_beginning, match_end;
2834 const char *patch = frag->patch;
2835 int size = frag->size;
2836 char *old, *oldlines;
2837 struct strbuf newlines;
2838 int new_blank_lines_at_end = 0;
2839 int found_new_blank_lines_at_end = 0;
2840 int hunk_linenr = frag->linenr;
2841 unsigned long leading, trailing;
2842 int pos, applied_pos;
2843 struct image preimage;
2844 struct image postimage;
2846 memset(&preimage, 0, sizeof(preimage));
2847 memset(&postimage, 0, sizeof(postimage));
2848 oldlines = xmalloc(size);
2849 strbuf_init(&newlines, size);
2851 old = oldlines;
2852 while (size > 0) {
2853 char first;
2854 int len = linelen(patch, size);
2855 int plen;
2856 int added_blank_line = 0;
2857 int is_blank_context = 0;
2858 size_t start;
2860 if (!len)
2861 break;
2864 * "plen" is how much of the line we should use for
2865 * the actual patch data. Normally we just remove the
2866 * first character on the line, but if the line is
2867 * followed by "\ No newline", then we also remove the
2868 * last one (which is the newline, of course).
2870 plen = len - 1;
2871 if (len < size && patch[len] == '\\')
2872 plen--;
2873 first = *patch;
2874 if (state->apply_in_reverse) {
2875 if (first == '-')
2876 first = '+';
2877 else if (first == '+')
2878 first = '-';
2881 switch (first) {
2882 case '\n':
2883 /* Newer GNU diff, empty context line */
2884 if (plen < 0)
2885 /* ... followed by '\No newline'; nothing */
2886 break;
2887 *old++ = '\n';
2888 strbuf_addch(&newlines, '\n');
2889 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2890 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2891 is_blank_context = 1;
2892 break;
2893 case ' ':
2894 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2895 ws_blank_line(patch + 1, plen, ws_rule))
2896 is_blank_context = 1;
2897 case '-':
2898 memcpy(old, patch + 1, plen);
2899 add_line_info(&preimage, old, plen,
2900 (first == ' ' ? LINE_COMMON : 0));
2901 old += plen;
2902 if (first == '-')
2903 break;
2904 /* Fall-through for ' ' */
2905 case '+':
2906 /* --no-add does not add new lines */
2907 if (first == '+' && state->no_add)
2908 break;
2910 start = newlines.len;
2911 if (first != '+' ||
2912 !state->whitespace_error ||
2913 state->ws_error_action != correct_ws_error) {
2914 strbuf_add(&newlines, patch + 1, plen);
2916 else {
2917 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2919 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2920 (first == '+' ? 0 : LINE_COMMON));
2921 if (first == '+' &&
2922 (ws_rule & WS_BLANK_AT_EOF) &&
2923 ws_blank_line(patch + 1, plen, ws_rule))
2924 added_blank_line = 1;
2925 break;
2926 case '@': case '\\':
2927 /* Ignore it, we already handled it */
2928 break;
2929 default:
2930 if (state->apply_verbosity > verbosity_normal)
2931 error(_("invalid start of line: '%c'"), first);
2932 applied_pos = -1;
2933 goto out;
2935 if (added_blank_line) {
2936 if (!new_blank_lines_at_end)
2937 found_new_blank_lines_at_end = hunk_linenr;
2938 new_blank_lines_at_end++;
2940 else if (is_blank_context)
2942 else
2943 new_blank_lines_at_end = 0;
2944 patch += len;
2945 size -= len;
2946 hunk_linenr++;
2948 if (inaccurate_eof &&
2949 old > oldlines && old[-1] == '\n' &&
2950 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2951 old--;
2952 strbuf_setlen(&newlines, newlines.len - 1);
2955 leading = frag->leading;
2956 trailing = frag->trailing;
2959 * A hunk to change lines at the beginning would begin with
2960 * @@ -1,L +N,M @@
2961 * but we need to be careful. -U0 that inserts before the second
2962 * line also has this pattern.
2964 * And a hunk to add to an empty file would begin with
2965 * @@ -0,0 +N,M @@
2967 * In other words, a hunk that is (frag->oldpos <= 1) with or
2968 * without leading context must match at the beginning.
2970 match_beginning = (!frag->oldpos ||
2971 (frag->oldpos == 1 && !state->unidiff_zero));
2974 * A hunk without trailing lines must match at the end.
2975 * However, we simply cannot tell if a hunk must match end
2976 * from the lack of trailing lines if the patch was generated
2977 * with unidiff without any context.
2979 match_end = !state->unidiff_zero && !trailing;
2981 pos = frag->newpos ? (frag->newpos - 1) : 0;
2982 preimage.buf = oldlines;
2983 preimage.len = old - oldlines;
2984 postimage.buf = newlines.buf;
2985 postimage.len = newlines.len;
2986 preimage.line = preimage.line_allocated;
2987 postimage.line = postimage.line_allocated;
2989 for (;;) {
2991 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2992 ws_rule, match_beginning, match_end);
2994 if (applied_pos >= 0)
2995 break;
2997 /* Am I at my context limits? */
2998 if ((leading <= state->p_context) && (trailing <= state->p_context))
2999 break;
3000 if (match_beginning || match_end) {
3001 match_beginning = match_end = 0;
3002 continue;
3006 * Reduce the number of context lines; reduce both
3007 * leading and trailing if they are equal otherwise
3008 * just reduce the larger context.
3010 if (leading >= trailing) {
3011 remove_first_line(&preimage);
3012 remove_first_line(&postimage);
3013 pos--;
3014 leading--;
3016 if (trailing > leading) {
3017 remove_last_line(&preimage);
3018 remove_last_line(&postimage);
3019 trailing--;
3023 if (applied_pos >= 0) {
3024 if (new_blank_lines_at_end &&
3025 preimage.nr + applied_pos >= img->nr &&
3026 (ws_rule & WS_BLANK_AT_EOF) &&
3027 state->ws_error_action != nowarn_ws_error) {
3028 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3029 found_new_blank_lines_at_end);
3030 if (state->ws_error_action == correct_ws_error) {
3031 while (new_blank_lines_at_end--)
3032 remove_last_line(&postimage);
3035 * We would want to prevent write_out_results()
3036 * from taking place in apply_patch() that follows
3037 * the callchain led us here, which is:
3038 * apply_patch->check_patch_list->check_patch->
3039 * apply_data->apply_fragments->apply_one_fragment
3041 if (state->ws_error_action == die_on_ws_error)
3042 state->apply = 0;
3045 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3046 int offset = applied_pos - pos;
3047 if (state->apply_in_reverse)
3048 offset = 0 - offset;
3049 fprintf_ln(stderr,
3050 Q_("Hunk #%d succeeded at %d (offset %d line).",
3051 "Hunk #%d succeeded at %d (offset %d lines).",
3052 offset),
3053 nth_fragment, applied_pos + 1, offset);
3057 * Warn if it was necessary to reduce the number
3058 * of context lines.
3060 if ((leading != frag->leading ||
3061 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3062 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3063 " to apply fragment at %d"),
3064 leading, trailing, applied_pos+1);
3065 update_image(state, img, applied_pos, &preimage, &postimage);
3066 } else {
3067 if (state->apply_verbosity > verbosity_normal)
3068 error(_("while searching for:\n%.*s"),
3069 (int)(old - oldlines), oldlines);
3072 out:
3073 free(oldlines);
3074 strbuf_release(&newlines);
3075 free(preimage.line_allocated);
3076 free(postimage.line_allocated);
3078 return (applied_pos < 0);
3081 static int apply_binary_fragment(struct apply_state *state,
3082 struct image *img,
3083 struct patch *patch)
3085 struct fragment *fragment = patch->fragments;
3086 unsigned long len;
3087 void *dst;
3089 if (!fragment)
3090 return error(_("missing binary patch data for '%s'"),
3091 patch->new_name ?
3092 patch->new_name :
3093 patch->old_name);
3095 /* Binary patch is irreversible without the optional second hunk */
3096 if (state->apply_in_reverse) {
3097 if (!fragment->next)
3098 return error(_("cannot reverse-apply a binary patch "
3099 "without the reverse hunk to '%s'"),
3100 patch->new_name
3101 ? patch->new_name : patch->old_name);
3102 fragment = fragment->next;
3104 switch (fragment->binary_patch_method) {
3105 case BINARY_DELTA_DEFLATED:
3106 dst = patch_delta(img->buf, img->len, fragment->patch,
3107 fragment->size, &len);
3108 if (!dst)
3109 return -1;
3110 clear_image(img);
3111 img->buf = dst;
3112 img->len = len;
3113 return 0;
3114 case BINARY_LITERAL_DEFLATED:
3115 clear_image(img);
3116 img->len = fragment->size;
3117 img->buf = xmemdupz(fragment->patch, img->len);
3118 return 0;
3120 return -1;
3124 * Replace "img" with the result of applying the binary patch.
3125 * The binary patch data itself in patch->fragment is still kept
3126 * but the preimage prepared by the caller in "img" is freed here
3127 * or in the helper function apply_binary_fragment() this calls.
3129 static int apply_binary(struct apply_state *state,
3130 struct image *img,
3131 struct patch *patch)
3133 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3134 struct object_id oid;
3137 * For safety, we require patch index line to contain
3138 * full 40-byte textual SHA1 for old and new, at least for now.
3140 if (strlen(patch->old_sha1_prefix) != 40 ||
3141 strlen(patch->new_sha1_prefix) != 40 ||
3142 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3143 get_oid_hex(patch->new_sha1_prefix, &oid))
3144 return error(_("cannot apply binary patch to '%s' "
3145 "without full index line"), name);
3147 if (patch->old_name) {
3149 * See if the old one matches what the patch
3150 * applies to.
3152 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3153 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3154 return error(_("the patch applies to '%s' (%s), "
3155 "which does not match the "
3156 "current contents."),
3157 name, oid_to_hex(&oid));
3159 else {
3160 /* Otherwise, the old one must be empty. */
3161 if (img->len)
3162 return error(_("the patch applies to an empty "
3163 "'%s' but it is not empty"), name);
3166 get_oid_hex(patch->new_sha1_prefix, &oid);
3167 if (is_null_oid(&oid)) {
3168 clear_image(img);
3169 return 0; /* deletion patch */
3172 if (has_sha1_file(oid.hash)) {
3173 /* We already have the postimage */
3174 enum object_type type;
3175 unsigned long size;
3176 char *result;
3178 result = read_sha1_file(oid.hash, &type, &size);
3179 if (!result)
3180 return error(_("the necessary postimage %s for "
3181 "'%s' cannot be read"),
3182 patch->new_sha1_prefix, name);
3183 clear_image(img);
3184 img->buf = result;
3185 img->len = size;
3186 } else {
3188 * We have verified buf matches the preimage;
3189 * apply the patch data to it, which is stored
3190 * in the patch->fragments->{patch,size}.
3192 if (apply_binary_fragment(state, img, patch))
3193 return error(_("binary patch does not apply to '%s'"),
3194 name);
3196 /* verify that the result matches */
3197 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3198 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3199 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3200 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3203 return 0;
3206 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3208 struct fragment *frag = patch->fragments;
3209 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3210 unsigned ws_rule = patch->ws_rule;
3211 unsigned inaccurate_eof = patch->inaccurate_eof;
3212 int nth = 0;
3214 if (patch->is_binary)
3215 return apply_binary(state, img, patch);
3217 while (frag) {
3218 nth++;
3219 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3220 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3221 if (!state->apply_with_reject)
3222 return -1;
3223 frag->rejected = 1;
3225 frag = frag->next;
3227 return 0;
3230 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3232 if (S_ISGITLINK(mode)) {
3233 strbuf_grow(buf, 100);
3234 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3235 } else {
3236 enum object_type type;
3237 unsigned long sz;
3238 char *result;
3240 result = read_sha1_file(oid->hash, &type, &sz);
3241 if (!result)
3242 return -1;
3243 /* XXX read_sha1_file NUL-terminates */
3244 strbuf_attach(buf, result, sz, sz + 1);
3246 return 0;
3249 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3251 if (!ce)
3252 return 0;
3253 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3256 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3258 struct string_list_item *item;
3260 if (name == NULL)
3261 return NULL;
3263 item = string_list_lookup(&state->fn_table, name);
3264 if (item != NULL)
3265 return (struct patch *)item->util;
3267 return NULL;
3271 * item->util in the filename table records the status of the path.
3272 * Usually it points at a patch (whose result records the contents
3273 * of it after applying it), but it could be PATH_WAS_DELETED for a
3274 * path that a previously applied patch has already removed, or
3275 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3277 * The latter is needed to deal with a case where two paths A and B
3278 * are swapped by first renaming A to B and then renaming B to A;
3279 * moving A to B should not be prevented due to presence of B as we
3280 * will remove it in a later patch.
3282 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3283 #define PATH_WAS_DELETED ((struct patch *) -1)
3285 static int to_be_deleted(struct patch *patch)
3287 return patch == PATH_TO_BE_DELETED;
3290 static int was_deleted(struct patch *patch)
3292 return patch == PATH_WAS_DELETED;
3295 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3297 struct string_list_item *item;
3300 * Always add new_name unless patch is a deletion
3301 * This should cover the cases for normal diffs,
3302 * file creations and copies
3304 if (patch->new_name != NULL) {
3305 item = string_list_insert(&state->fn_table, patch->new_name);
3306 item->util = patch;
3310 * store a failure on rename/deletion cases because
3311 * later chunks shouldn't patch old names
3313 if ((patch->new_name == NULL) || (patch->is_rename)) {
3314 item = string_list_insert(&state->fn_table, patch->old_name);
3315 item->util = PATH_WAS_DELETED;
3319 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3322 * store information about incoming file deletion
3324 while (patch) {
3325 if ((patch->new_name == NULL) || (patch->is_rename)) {
3326 struct string_list_item *item;
3327 item = string_list_insert(&state->fn_table, patch->old_name);
3328 item->util = PATH_TO_BE_DELETED;
3330 patch = patch->next;
3334 static int checkout_target(struct index_state *istate,
3335 struct cache_entry *ce, struct stat *st)
3337 struct checkout costate = CHECKOUT_INIT;
3339 costate.refresh_cache = 1;
3340 costate.istate = istate;
3341 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3342 return error(_("cannot checkout %s"), ce->name);
3343 return 0;
3346 static struct patch *previous_patch(struct apply_state *state,
3347 struct patch *patch,
3348 int *gone)
3350 struct patch *previous;
3352 *gone = 0;
3353 if (patch->is_copy || patch->is_rename)
3354 return NULL; /* "git" patches do not depend on the order */
3356 previous = in_fn_table(state, patch->old_name);
3357 if (!previous)
3358 return NULL;
3360 if (to_be_deleted(previous))
3361 return NULL; /* the deletion hasn't happened yet */
3363 if (was_deleted(previous))
3364 *gone = 1;
3366 return previous;
3369 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3371 if (S_ISGITLINK(ce->ce_mode)) {
3372 if (!S_ISDIR(st->st_mode))
3373 return -1;
3374 return 0;
3376 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3379 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3381 static int load_patch_target(struct apply_state *state,
3382 struct strbuf *buf,
3383 const struct cache_entry *ce,
3384 struct stat *st,
3385 const char *name,
3386 unsigned expected_mode)
3388 if (state->cached || state->check_index) {
3389 if (read_file_or_gitlink(ce, buf))
3390 return error(_("failed to read %s"), name);
3391 } else if (name) {
3392 if (S_ISGITLINK(expected_mode)) {
3393 if (ce)
3394 return read_file_or_gitlink(ce, buf);
3395 else
3396 return SUBMODULE_PATCH_WITHOUT_INDEX;
3397 } else if (has_symlink_leading_path(name, strlen(name))) {
3398 return error(_("reading from '%s' beyond a symbolic link"), name);
3399 } else {
3400 if (read_old_data(st, name, buf))
3401 return error(_("failed to read %s"), name);
3404 return 0;
3408 * We are about to apply "patch"; populate the "image" with the
3409 * current version we have, from the working tree or from the index,
3410 * depending on the situation e.g. --cached/--index. If we are
3411 * applying a non-git patch that incrementally updates the tree,
3412 * we read from the result of a previous diff.
3414 static int load_preimage(struct apply_state *state,
3415 struct image *image,
3416 struct patch *patch, struct stat *st,
3417 const struct cache_entry *ce)
3419 struct strbuf buf = STRBUF_INIT;
3420 size_t len;
3421 char *img;
3422 struct patch *previous;
3423 int status;
3425 previous = previous_patch(state, patch, &status);
3426 if (status)
3427 return error(_("path %s has been renamed/deleted"),
3428 patch->old_name);
3429 if (previous) {
3430 /* We have a patched copy in memory; use that. */
3431 strbuf_add(&buf, previous->result, previous->resultsize);
3432 } else {
3433 status = load_patch_target(state, &buf, ce, st,
3434 patch->old_name, patch->old_mode);
3435 if (status < 0)
3436 return status;
3437 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3439 * There is no way to apply subproject
3440 * patch without looking at the index.
3441 * NEEDSWORK: shouldn't this be flagged
3442 * as an error???
3444 free_fragment_list(patch->fragments);
3445 patch->fragments = NULL;
3446 } else if (status) {
3447 return error(_("failed to read %s"), patch->old_name);
3451 img = strbuf_detach(&buf, &len);
3452 prepare_image(image, img, len, !patch->is_binary);
3453 return 0;
3456 static int three_way_merge(struct image *image,
3457 char *path,
3458 const struct object_id *base,
3459 const struct object_id *ours,
3460 const struct object_id *theirs)
3462 mmfile_t base_file, our_file, their_file;
3463 mmbuffer_t result = { NULL };
3464 int status;
3466 read_mmblob(&base_file, base);
3467 read_mmblob(&our_file, ours);
3468 read_mmblob(&their_file, theirs);
3469 status = ll_merge(&result, path,
3470 &base_file, "base",
3471 &our_file, "ours",
3472 &their_file, "theirs", NULL);
3473 free(base_file.ptr);
3474 free(our_file.ptr);
3475 free(their_file.ptr);
3476 if (status < 0 || !result.ptr) {
3477 free(result.ptr);
3478 return -1;
3480 clear_image(image);
3481 image->buf = result.ptr;
3482 image->len = result.size;
3484 return status;
3488 * When directly falling back to add/add three-way merge, we read from
3489 * the current contents of the new_name. In no cases other than that
3490 * this function will be called.
3492 static int load_current(struct apply_state *state,
3493 struct image *image,
3494 struct patch *patch)
3496 struct strbuf buf = STRBUF_INIT;
3497 int status, pos;
3498 size_t len;
3499 char *img;
3500 struct stat st;
3501 struct cache_entry *ce;
3502 char *name = patch->new_name;
3503 unsigned mode = patch->new_mode;
3505 if (!patch->is_new)
3506 die("BUG: patch to %s is not a creation", patch->old_name);
3508 pos = cache_name_pos(name, strlen(name));
3509 if (pos < 0)
3510 return error(_("%s: does not exist in index"), name);
3511 ce = active_cache[pos];
3512 if (lstat(name, &st)) {
3513 if (errno != ENOENT)
3514 return error_errno("%s", name);
3515 if (checkout_target(&the_index, ce, &st))
3516 return -1;
3518 if (verify_index_match(ce, &st))
3519 return error(_("%s: does not match index"), name);
3521 status = load_patch_target(state, &buf, ce, &st, name, mode);
3522 if (status < 0)
3523 return status;
3524 else if (status)
3525 return -1;
3526 img = strbuf_detach(&buf, &len);
3527 prepare_image(image, img, len, !patch->is_binary);
3528 return 0;
3531 static int try_threeway(struct apply_state *state,
3532 struct image *image,
3533 struct patch *patch,
3534 struct stat *st,
3535 const struct cache_entry *ce)
3537 struct object_id pre_oid, post_oid, our_oid;
3538 struct strbuf buf = STRBUF_INIT;
3539 size_t len;
3540 int status;
3541 char *img;
3542 struct image tmp_image;
3544 /* No point falling back to 3-way merge in these cases */
3545 if (patch->is_delete ||
3546 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3547 return -1;
3549 /* Preimage the patch was prepared for */
3550 if (patch->is_new)
3551 write_sha1_file("", 0, blob_type, pre_oid.hash);
3552 else if (get_sha1(patch->old_sha1_prefix, pre_oid.hash) ||
3553 read_blob_object(&buf, &pre_oid, patch->old_mode))
3554 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3556 if (state->apply_verbosity > verbosity_silent)
3557 fprintf(stderr, _("Falling back to three-way merge...\n"));
3559 img = strbuf_detach(&buf, &len);
3560 prepare_image(&tmp_image, img, len, 1);
3561 /* Apply the patch to get the post image */
3562 if (apply_fragments(state, &tmp_image, patch) < 0) {
3563 clear_image(&tmp_image);
3564 return -1;
3566 /* post_oid is theirs */
3567 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3568 clear_image(&tmp_image);
3570 /* our_oid is ours */
3571 if (patch->is_new) {
3572 if (load_current(state, &tmp_image, patch))
3573 return error(_("cannot read the current contents of '%s'"),
3574 patch->new_name);
3575 } else {
3576 if (load_preimage(state, &tmp_image, patch, st, ce))
3577 return error(_("cannot read the current contents of '%s'"),
3578 patch->old_name);
3580 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3581 clear_image(&tmp_image);
3583 /* in-core three-way merge between post and our using pre as base */
3584 status = three_way_merge(image, patch->new_name,
3585 &pre_oid, &our_oid, &post_oid);
3586 if (status < 0) {
3587 if (state->apply_verbosity > verbosity_silent)
3588 fprintf(stderr,
3589 _("Failed to fall back on three-way merge...\n"));
3590 return status;
3593 if (status) {
3594 patch->conflicted_threeway = 1;
3595 if (patch->is_new)
3596 oidclr(&patch->threeway_stage[0]);
3597 else
3598 oidcpy(&patch->threeway_stage[0], &pre_oid);
3599 oidcpy(&patch->threeway_stage[1], &our_oid);
3600 oidcpy(&patch->threeway_stage[2], &post_oid);
3601 if (state->apply_verbosity > verbosity_silent)
3602 fprintf(stderr,
3603 _("Applied patch to '%s' with conflicts.\n"),
3604 patch->new_name);
3605 } else {
3606 if (state->apply_verbosity > verbosity_silent)
3607 fprintf(stderr,
3608 _("Applied patch to '%s' cleanly.\n"),
3609 patch->new_name);
3611 return 0;
3614 static int apply_data(struct apply_state *state, struct patch *patch,
3615 struct stat *st, const struct cache_entry *ce)
3617 struct image image;
3619 if (load_preimage(state, &image, patch, st, ce) < 0)
3620 return -1;
3622 if (patch->direct_to_threeway ||
3623 apply_fragments(state, &image, patch) < 0) {
3624 /* Note: with --reject, apply_fragments() returns 0 */
3625 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3626 return -1;
3628 patch->result = image.buf;
3629 patch->resultsize = image.len;
3630 add_to_fn_table(state, patch);
3631 free(image.line_allocated);
3633 if (0 < patch->is_delete && patch->resultsize)
3634 return error(_("removal patch leaves file contents"));
3636 return 0;
3640 * If "patch" that we are looking at modifies or deletes what we have,
3641 * we would want it not to lose any local modification we have, either
3642 * in the working tree or in the index.
3644 * This also decides if a non-git patch is a creation patch or a
3645 * modification to an existing empty file. We do not check the state
3646 * of the current tree for a creation patch in this function; the caller
3647 * check_patch() separately makes sure (and errors out otherwise) that
3648 * the path the patch creates does not exist in the current tree.
3650 static int check_preimage(struct apply_state *state,
3651 struct patch *patch,
3652 struct cache_entry **ce,
3653 struct stat *st)
3655 const char *old_name = patch->old_name;
3656 struct patch *previous = NULL;
3657 int stat_ret = 0, status;
3658 unsigned st_mode = 0;
3660 if (!old_name)
3661 return 0;
3663 assert(patch->is_new <= 0);
3664 previous = previous_patch(state, patch, &status);
3666 if (status)
3667 return error(_("path %s has been renamed/deleted"), old_name);
3668 if (previous) {
3669 st_mode = previous->new_mode;
3670 } else if (!state->cached) {
3671 stat_ret = lstat(old_name, st);
3672 if (stat_ret && errno != ENOENT)
3673 return error_errno("%s", old_name);
3676 if (state->check_index && !previous) {
3677 int pos = cache_name_pos(old_name, strlen(old_name));
3678 if (pos < 0) {
3679 if (patch->is_new < 0)
3680 goto is_new;
3681 return error(_("%s: does not exist in index"), old_name);
3683 *ce = active_cache[pos];
3684 if (stat_ret < 0) {
3685 if (checkout_target(&the_index, *ce, st))
3686 return -1;
3688 if (!state->cached && verify_index_match(*ce, st))
3689 return error(_("%s: does not match index"), old_name);
3690 if (state->cached)
3691 st_mode = (*ce)->ce_mode;
3692 } else if (stat_ret < 0) {
3693 if (patch->is_new < 0)
3694 goto is_new;
3695 return error_errno("%s", old_name);
3698 if (!state->cached && !previous)
3699 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3701 if (patch->is_new < 0)
3702 patch->is_new = 0;
3703 if (!patch->old_mode)
3704 patch->old_mode = st_mode;
3705 if ((st_mode ^ patch->old_mode) & S_IFMT)
3706 return error(_("%s: wrong type"), old_name);
3707 if (st_mode != patch->old_mode)
3708 warning(_("%s has type %o, expected %o"),
3709 old_name, st_mode, patch->old_mode);
3710 if (!patch->new_mode && !patch->is_delete)
3711 patch->new_mode = st_mode;
3712 return 0;
3714 is_new:
3715 patch->is_new = 1;
3716 patch->is_delete = 0;
3717 free(patch->old_name);
3718 patch->old_name = NULL;
3719 return 0;
3723 #define EXISTS_IN_INDEX 1
3724 #define EXISTS_IN_WORKTREE 2
3726 static int check_to_create(struct apply_state *state,
3727 const char *new_name,
3728 int ok_if_exists)
3730 struct stat nst;
3732 if (state->check_index &&
3733 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3734 !ok_if_exists)
3735 return EXISTS_IN_INDEX;
3736 if (state->cached)
3737 return 0;
3739 if (!lstat(new_name, &nst)) {
3740 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3741 return 0;
3743 * A leading component of new_name might be a symlink
3744 * that is going to be removed with this patch, but
3745 * still pointing at somewhere that has the path.
3746 * In such a case, path "new_name" does not exist as
3747 * far as git is concerned.
3749 if (has_symlink_leading_path(new_name, strlen(new_name)))
3750 return 0;
3752 return EXISTS_IN_WORKTREE;
3753 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3754 return error_errno("%s", new_name);
3756 return 0;
3759 static uintptr_t register_symlink_changes(struct apply_state *state,
3760 const char *path,
3761 uintptr_t what)
3763 struct string_list_item *ent;
3765 ent = string_list_lookup(&state->symlink_changes, path);
3766 if (!ent) {
3767 ent = string_list_insert(&state->symlink_changes, path);
3768 ent->util = (void *)0;
3770 ent->util = (void *)(what | ((uintptr_t)ent->util));
3771 return (uintptr_t)ent->util;
3774 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3776 struct string_list_item *ent;
3778 ent = string_list_lookup(&state->symlink_changes, path);
3779 if (!ent)
3780 return 0;
3781 return (uintptr_t)ent->util;
3784 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3786 for ( ; patch; patch = patch->next) {
3787 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3788 (patch->is_rename || patch->is_delete))
3789 /* the symlink at patch->old_name is removed */
3790 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3792 if (patch->new_name && S_ISLNK(patch->new_mode))
3793 /* the symlink at patch->new_name is created or remains */
3794 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3798 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3800 do {
3801 unsigned int change;
3803 while (--name->len && name->buf[name->len] != '/')
3804 ; /* scan backwards */
3805 if (!name->len)
3806 break;
3807 name->buf[name->len] = '\0';
3808 change = check_symlink_changes(state, name->buf);
3809 if (change & APPLY_SYMLINK_IN_RESULT)
3810 return 1;
3811 if (change & APPLY_SYMLINK_GOES_AWAY)
3813 * This cannot be "return 0", because we may
3814 * see a new one created at a higher level.
3816 continue;
3818 /* otherwise, check the preimage */
3819 if (state->check_index) {
3820 struct cache_entry *ce;
3822 ce = cache_file_exists(name->buf, name->len, ignore_case);
3823 if (ce && S_ISLNK(ce->ce_mode))
3824 return 1;
3825 } else {
3826 struct stat st;
3827 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3828 return 1;
3830 } while (1);
3831 return 0;
3834 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3836 int ret;
3837 struct strbuf name = STRBUF_INIT;
3839 assert(*name_ != '\0');
3840 strbuf_addstr(&name, name_);
3841 ret = path_is_beyond_symlink_1(state, &name);
3842 strbuf_release(&name);
3844 return ret;
3847 static int check_unsafe_path(struct patch *patch)
3849 const char *old_name = NULL;
3850 const char *new_name = NULL;
3851 if (patch->is_delete)
3852 old_name = patch->old_name;
3853 else if (!patch->is_new && !patch->is_copy)
3854 old_name = patch->old_name;
3855 if (!patch->is_delete)
3856 new_name = patch->new_name;
3858 if (old_name && !verify_path(old_name))
3859 return error(_("invalid path '%s'"), old_name);
3860 if (new_name && !verify_path(new_name))
3861 return error(_("invalid path '%s'"), new_name);
3862 return 0;
3866 * Check and apply the patch in-core; leave the result in patch->result
3867 * for the caller to write it out to the final destination.
3869 static int check_patch(struct apply_state *state, struct patch *patch)
3871 struct stat st;
3872 const char *old_name = patch->old_name;
3873 const char *new_name = patch->new_name;
3874 const char *name = old_name ? old_name : new_name;
3875 struct cache_entry *ce = NULL;
3876 struct patch *tpatch;
3877 int ok_if_exists;
3878 int status;
3880 patch->rejected = 1; /* we will drop this after we succeed */
3882 status = check_preimage(state, patch, &ce, &st);
3883 if (status)
3884 return status;
3885 old_name = patch->old_name;
3888 * A type-change diff is always split into a patch to delete
3889 * old, immediately followed by a patch to create new (see
3890 * diff.c::run_diff()); in such a case it is Ok that the entry
3891 * to be deleted by the previous patch is still in the working
3892 * tree and in the index.
3894 * A patch to swap-rename between A and B would first rename A
3895 * to B and then rename B to A. While applying the first one,
3896 * the presence of B should not stop A from getting renamed to
3897 * B; ask to_be_deleted() about the later rename. Removal of
3898 * B and rename from A to B is handled the same way by asking
3899 * was_deleted().
3901 if ((tpatch = in_fn_table(state, new_name)) &&
3902 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3903 ok_if_exists = 1;
3904 else
3905 ok_if_exists = 0;
3907 if (new_name &&
3908 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3909 int err = check_to_create(state, new_name, ok_if_exists);
3911 if (err && state->threeway) {
3912 patch->direct_to_threeway = 1;
3913 } else switch (err) {
3914 case 0:
3915 break; /* happy */
3916 case EXISTS_IN_INDEX:
3917 return error(_("%s: already exists in index"), new_name);
3918 break;
3919 case EXISTS_IN_WORKTREE:
3920 return error(_("%s: already exists in working directory"),
3921 new_name);
3922 default:
3923 return err;
3926 if (!patch->new_mode) {
3927 if (0 < patch->is_new)
3928 patch->new_mode = S_IFREG | 0644;
3929 else
3930 patch->new_mode = patch->old_mode;
3934 if (new_name && old_name) {
3935 int same = !strcmp(old_name, new_name);
3936 if (!patch->new_mode)
3937 patch->new_mode = patch->old_mode;
3938 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3939 if (same)
3940 return error(_("new mode (%o) of %s does not "
3941 "match old mode (%o)"),
3942 patch->new_mode, new_name,
3943 patch->old_mode);
3944 else
3945 return error(_("new mode (%o) of %s does not "
3946 "match old mode (%o) of %s"),
3947 patch->new_mode, new_name,
3948 patch->old_mode, old_name);
3952 if (!state->unsafe_paths && check_unsafe_path(patch))
3953 return -128;
3956 * An attempt to read from or delete a path that is beyond a
3957 * symbolic link will be prevented by load_patch_target() that
3958 * is called at the beginning of apply_data() so we do not
3959 * have to worry about a patch marked with "is_delete" bit
3960 * here. We however need to make sure that the patch result
3961 * is not deposited to a path that is beyond a symbolic link
3962 * here.
3964 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3965 return error(_("affected file '%s' is beyond a symbolic link"),
3966 patch->new_name);
3968 if (apply_data(state, patch, &st, ce) < 0)
3969 return error(_("%s: patch does not apply"), name);
3970 patch->rejected = 0;
3971 return 0;
3974 static int check_patch_list(struct apply_state *state, struct patch *patch)
3976 int err = 0;
3978 prepare_symlink_changes(state, patch);
3979 prepare_fn_table(state, patch);
3980 while (patch) {
3981 int res;
3982 if (state->apply_verbosity > verbosity_normal)
3983 say_patch_name(stderr,
3984 _("Checking patch %s..."), patch);
3985 res = check_patch(state, patch);
3986 if (res == -128)
3987 return -128;
3988 err |= res;
3989 patch = patch->next;
3991 return err;