stash: remove now superfluos help for "stash push"
[git.git] / apply.c
blobc022af53a24bf56556d136a3456862cf6a2ca7cc
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
10 #include "cache.h"
11 #include "config.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "diff.h"
15 #include "dir.h"
16 #include "xdiff-interface.h"
17 #include "ll-merge.h"
18 #include "lockfile.h"
19 #include "parse-options.h"
20 #include "quote.h"
21 #include "rerere.h"
22 #include "apply.h"
24 static void git_apply_config(void)
26 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
27 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
28 git_config(git_default_config, NULL);
31 static int parse_whitespace_option(struct apply_state *state, const char *option)
33 if (!option) {
34 state->ws_error_action = warn_on_ws_error;
35 return 0;
37 if (!strcmp(option, "warn")) {
38 state->ws_error_action = warn_on_ws_error;
39 return 0;
41 if (!strcmp(option, "nowarn")) {
42 state->ws_error_action = nowarn_ws_error;
43 return 0;
45 if (!strcmp(option, "error")) {
46 state->ws_error_action = die_on_ws_error;
47 return 0;
49 if (!strcmp(option, "error-all")) {
50 state->ws_error_action = die_on_ws_error;
51 state->squelch_whitespace_errors = 0;
52 return 0;
54 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
55 state->ws_error_action = correct_ws_error;
56 return 0;
58 return error(_("unrecognized whitespace option '%s'"), option);
61 static int parse_ignorewhitespace_option(struct apply_state *state,
62 const char *option)
64 if (!option || !strcmp(option, "no") ||
65 !strcmp(option, "false") || !strcmp(option, "never") ||
66 !strcmp(option, "none")) {
67 state->ws_ignore_action = ignore_ws_none;
68 return 0;
70 if (!strcmp(option, "change")) {
71 state->ws_ignore_action = ignore_ws_change;
72 return 0;
74 return error(_("unrecognized whitespace ignore option '%s'"), option);
77 int init_apply_state(struct apply_state *state,
78 const char *prefix,
79 struct lock_file *lock_file)
81 memset(state, 0, sizeof(*state));
82 state->prefix = prefix;
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 int extension_linenr; /* first line specifying delete/new/rename/copy */
214 unsigned int is_toplevel_relative:1;
215 unsigned int inaccurate_eof:1;
216 unsigned int is_binary:1;
217 unsigned int is_copy:1;
218 unsigned int is_rename:1;
219 unsigned int recount:1;
220 unsigned int conflicted_threeway:1;
221 unsigned int direct_to_threeway:1;
222 unsigned int crlf_in_old:1;
223 struct fragment *fragments;
224 char *result;
225 size_t resultsize;
226 char old_sha1_prefix[41];
227 char new_sha1_prefix[41];
228 struct patch *next;
230 /* three-way fallback result */
231 struct object_id threeway_stage[3];
234 static void free_fragment_list(struct fragment *list)
236 while (list) {
237 struct fragment *next = list->next;
238 if (list->free_patch)
239 free((char *)list->patch);
240 free(list);
241 list = next;
245 static void free_patch(struct patch *patch)
247 free_fragment_list(patch->fragments);
248 free(patch->def_name);
249 free(patch->old_name);
250 free(patch->new_name);
251 free(patch->result);
252 free(patch);
255 static void free_patch_list(struct patch *list)
257 while (list) {
258 struct patch *next = list->next;
259 free_patch(list);
260 list = next;
265 * A line in a file, len-bytes long (includes the terminating LF,
266 * except for an incomplete line at the end if the file ends with
267 * one), and its contents hashes to 'hash'.
269 struct line {
270 size_t len;
271 unsigned hash : 24;
272 unsigned flag : 8;
273 #define LINE_COMMON 1
274 #define LINE_PATCHED 2
278 * This represents a "file", which is an array of "lines".
280 struct image {
281 char *buf;
282 size_t len;
283 size_t nr;
284 size_t alloc;
285 struct line *line_allocated;
286 struct line *line;
289 static uint32_t hash_line(const char *cp, size_t len)
291 size_t i;
292 uint32_t h;
293 for (i = 0, h = 0; i < len; i++) {
294 if (!isspace(cp[i])) {
295 h = h * 3 + (cp[i] & 0xff);
298 return h;
302 * Compare lines s1 of length n1 and s2 of length n2, ignoring
303 * whitespace difference. Returns 1 if they match, 0 otherwise
305 static int fuzzy_matchlines(const char *s1, size_t n1,
306 const char *s2, size_t n2)
308 const char *last1 = s1 + n1 - 1;
309 const char *last2 = s2 + n2 - 1;
310 int result = 0;
312 /* ignore line endings */
313 while ((*last1 == '\r') || (*last1 == '\n'))
314 last1--;
315 while ((*last2 == '\r') || (*last2 == '\n'))
316 last2--;
318 /* skip leading whitespaces, if both begin with whitespace */
319 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
320 while (isspace(*s1) && (s1 <= last1))
321 s1++;
322 while (isspace(*s2) && (s2 <= last2))
323 s2++;
325 /* early return if both lines are empty */
326 if ((s1 > last1) && (s2 > last2))
327 return 1;
328 while (!result) {
329 result = *s1++ - *s2++;
331 * Skip whitespace inside. We check for whitespace on
332 * both buffers because we don't want "a b" to match
333 * "ab"
335 if (isspace(*s1) && isspace(*s2)) {
336 while (isspace(*s1) && s1 <= last1)
337 s1++;
338 while (isspace(*s2) && s2 <= last2)
339 s2++;
342 * If we reached the end on one side only,
343 * lines don't match
345 if (
346 ((s2 > last2) && (s1 <= last1)) ||
347 ((s1 > last1) && (s2 <= last2)))
348 return 0;
349 if ((s1 > last1) && (s2 > last2))
350 break;
353 return !result;
356 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
358 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
359 img->line_allocated[img->nr].len = len;
360 img->line_allocated[img->nr].hash = hash_line(bol, len);
361 img->line_allocated[img->nr].flag = flag;
362 img->nr++;
366 * "buf" has the file contents to be patched (read from various sources).
367 * attach it to "image" and add line-based index to it.
368 * "image" now owns the "buf".
370 static void prepare_image(struct image *image, char *buf, size_t len,
371 int prepare_linetable)
373 const char *cp, *ep;
375 memset(image, 0, sizeof(*image));
376 image->buf = buf;
377 image->len = len;
379 if (!prepare_linetable)
380 return;
382 ep = image->buf + image->len;
383 cp = image->buf;
384 while (cp < ep) {
385 const char *next;
386 for (next = cp; next < ep && *next != '\n'; next++)
388 if (next < ep)
389 next++;
390 add_line_info(image, cp, next - cp, 0);
391 cp = next;
393 image->line = image->line_allocated;
396 static void clear_image(struct image *image)
398 free(image->buf);
399 free(image->line_allocated);
400 memset(image, 0, sizeof(*image));
403 /* fmt must contain _one_ %s and no other substitution */
404 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
406 struct strbuf sb = STRBUF_INIT;
408 if (patch->old_name && patch->new_name &&
409 strcmp(patch->old_name, patch->new_name)) {
410 quote_c_style(patch->old_name, &sb, NULL, 0);
411 strbuf_addstr(&sb, " => ");
412 quote_c_style(patch->new_name, &sb, NULL, 0);
413 } else {
414 const char *n = patch->new_name;
415 if (!n)
416 n = patch->old_name;
417 quote_c_style(n, &sb, NULL, 0);
419 fprintf(output, fmt, sb.buf);
420 fputc('\n', output);
421 strbuf_release(&sb);
424 #define SLOP (16)
426 static int read_patch_file(struct strbuf *sb, int fd)
428 if (strbuf_read(sb, fd, 0) < 0)
429 return error_errno("git apply: failed to read");
432 * Make sure that we have some slop in the buffer
433 * so that we can do speculative "memcmp" etc, and
434 * see to it that it is NUL-filled.
436 strbuf_grow(sb, SLOP);
437 memset(sb->buf + sb->len, 0, SLOP);
438 return 0;
441 static unsigned long linelen(const char *buffer, unsigned long size)
443 unsigned long len = 0;
444 while (size--) {
445 len++;
446 if (*buffer++ == '\n')
447 break;
449 return len;
452 static int is_dev_null(const char *str)
454 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
457 #define TERM_SPACE 1
458 #define TERM_TAB 2
460 static int name_terminate(int c, int terminate)
462 if (c == ' ' && !(terminate & TERM_SPACE))
463 return 0;
464 if (c == '\t' && !(terminate & TERM_TAB))
465 return 0;
467 return 1;
470 /* remove double slashes to make --index work with such filenames */
471 static char *squash_slash(char *name)
473 int i = 0, j = 0;
475 if (!name)
476 return NULL;
478 while (name[i]) {
479 if ((name[j++] = name[i++]) == '/')
480 while (name[i] == '/')
481 i++;
483 name[j] = '\0';
484 return name;
487 static char *find_name_gnu(struct apply_state *state,
488 const char *line,
489 const char *def,
490 int p_value)
492 struct strbuf name = STRBUF_INIT;
493 char *cp;
496 * Proposed "new-style" GNU patch/diff format; see
497 * http://marc.info/?l=git&m=112927316408690&w=2
499 if (unquote_c_style(&name, line, NULL)) {
500 strbuf_release(&name);
501 return NULL;
504 for (cp = name.buf; p_value; p_value--) {
505 cp = strchr(cp, '/');
506 if (!cp) {
507 strbuf_release(&name);
508 return NULL;
510 cp++;
513 strbuf_remove(&name, 0, cp - name.buf);
514 if (state->root.len)
515 strbuf_insert(&name, 0, state->root.buf, state->root.len);
516 return squash_slash(strbuf_detach(&name, NULL));
519 static size_t sane_tz_len(const char *line, size_t len)
521 const char *tz, *p;
523 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
524 return 0;
525 tz = line + len - strlen(" +0500");
527 if (tz[1] != '+' && tz[1] != '-')
528 return 0;
530 for (p = tz + 2; p != line + len; p++)
531 if (!isdigit(*p))
532 return 0;
534 return line + len - tz;
537 static size_t tz_with_colon_len(const char *line, size_t len)
539 const char *tz, *p;
541 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
542 return 0;
543 tz = line + len - strlen(" +08:00");
545 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
546 return 0;
547 p = tz + 2;
548 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
549 !isdigit(*p++) || !isdigit(*p++))
550 return 0;
552 return line + len - tz;
555 static size_t date_len(const char *line, size_t len)
557 const char *date, *p;
559 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
560 return 0;
561 p = date = line + len - strlen("72-02-05");
563 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
565 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
566 return 0;
568 if (date - line >= strlen("19") &&
569 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
570 date -= strlen("19");
572 return line + len - date;
575 static size_t short_time_len(const char *line, size_t len)
577 const char *time, *p;
579 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
580 return 0;
581 p = time = line + len - strlen(" 07:01:32");
583 /* Permit 1-digit hours? */
584 if (*p++ != ' ' ||
585 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
586 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
587 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
588 return 0;
590 return line + len - time;
593 static size_t fractional_time_len(const char *line, size_t len)
595 const char *p;
596 size_t n;
598 /* Expected format: 19:41:17.620000023 */
599 if (!len || !isdigit(line[len - 1]))
600 return 0;
601 p = line + len - 1;
603 /* Fractional seconds. */
604 while (p > line && isdigit(*p))
605 p--;
606 if (*p != '.')
607 return 0;
609 /* Hours, minutes, and whole seconds. */
610 n = short_time_len(line, p - line);
611 if (!n)
612 return 0;
614 return line + len - p + n;
617 static size_t trailing_spaces_len(const char *line, size_t len)
619 const char *p;
621 /* Expected format: ' ' x (1 or more) */
622 if (!len || line[len - 1] != ' ')
623 return 0;
625 p = line + len;
626 while (p != line) {
627 p--;
628 if (*p != ' ')
629 return line + len - (p + 1);
632 /* All spaces! */
633 return len;
636 static size_t diff_timestamp_len(const char *line, size_t len)
638 const char *end = line + len;
639 size_t n;
642 * Posix: 2010-07-05 19:41:17
643 * GNU: 2010-07-05 19:41:17.620000023 -0500
646 if (!isdigit(end[-1]))
647 return 0;
649 n = sane_tz_len(line, end - line);
650 if (!n)
651 n = tz_with_colon_len(line, end - line);
652 end -= n;
654 n = short_time_len(line, end - line);
655 if (!n)
656 n = fractional_time_len(line, end - line);
657 end -= n;
659 n = date_len(line, end - line);
660 if (!n) /* No date. Too bad. */
661 return 0;
662 end -= n;
664 if (end == line) /* No space before date. */
665 return 0;
666 if (end[-1] == '\t') { /* Success! */
667 end--;
668 return line + len - end;
670 if (end[-1] != ' ') /* No space before date. */
671 return 0;
673 /* Whitespace damage. */
674 end -= trailing_spaces_len(line, end - line);
675 return line + len - end;
678 static char *find_name_common(struct apply_state *state,
679 const char *line,
680 const char *def,
681 int p_value,
682 const char *end,
683 int terminate)
685 int len;
686 const char *start = NULL;
688 if (p_value == 0)
689 start = line;
690 while (line != end) {
691 char c = *line;
693 if (!end && isspace(c)) {
694 if (c == '\n')
695 break;
696 if (name_terminate(c, terminate))
697 break;
699 line++;
700 if (c == '/' && !--p_value)
701 start = line;
703 if (!start)
704 return squash_slash(xstrdup_or_null(def));
705 len = line - start;
706 if (!len)
707 return squash_slash(xstrdup_or_null(def));
710 * Generally we prefer the shorter name, especially
711 * if the other one is just a variation of that with
712 * something else tacked on to the end (ie "file.orig"
713 * or "file~").
715 if (def) {
716 int deflen = strlen(def);
717 if (deflen < len && !strncmp(start, def, deflen))
718 return squash_slash(xstrdup(def));
721 if (state->root.len) {
722 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
723 return squash_slash(ret);
726 return squash_slash(xmemdupz(start, len));
729 static char *find_name(struct apply_state *state,
730 const char *line,
731 char *def,
732 int p_value,
733 int terminate)
735 if (*line == '"') {
736 char *name = find_name_gnu(state, line, def, p_value);
737 if (name)
738 return name;
741 return find_name_common(state, line, def, p_value, NULL, terminate);
744 static char *find_name_traditional(struct apply_state *state,
745 const char *line,
746 char *def,
747 int p_value)
749 size_t len;
750 size_t date_len;
752 if (*line == '"') {
753 char *name = find_name_gnu(state, line, def, p_value);
754 if (name)
755 return name;
758 len = strchrnul(line, '\n') - line;
759 date_len = diff_timestamp_len(line, len);
760 if (!date_len)
761 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
762 len -= date_len;
764 return find_name_common(state, line, def, p_value, line + len, 0);
768 * Given the string after "--- " or "+++ ", guess the appropriate
769 * p_value for the given patch.
771 static int guess_p_value(struct apply_state *state, const char *nameline)
773 char *name, *cp;
774 int val = -1;
776 if (is_dev_null(nameline))
777 return -1;
778 name = find_name_traditional(state, nameline, NULL, 0);
779 if (!name)
780 return -1;
781 cp = strchr(name, '/');
782 if (!cp)
783 val = 0;
784 else if (state->prefix) {
786 * Does it begin with "a/$our-prefix" and such? Then this is
787 * very likely to apply to our directory.
789 if (starts_with(name, state->prefix))
790 val = count_slashes(state->prefix);
791 else {
792 cp++;
793 if (starts_with(cp, state->prefix))
794 val = count_slashes(state->prefix) + 1;
797 free(name);
798 return val;
802 * Does the ---/+++ line have the POSIX timestamp after the last HT?
803 * GNU diff puts epoch there to signal a creation/deletion event. Is
804 * this such a timestamp?
806 static int has_epoch_timestamp(const char *nameline)
809 * We are only interested in epoch timestamp; any non-zero
810 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
811 * For the same reason, the date must be either 1969-12-31 or
812 * 1970-01-01, and the seconds part must be "00".
814 const char stamp_regexp[] =
815 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
817 "([-+][0-2][0-9]:?[0-5][0-9])\n";
818 const char *timestamp = NULL, *cp, *colon;
819 static regex_t *stamp;
820 regmatch_t m[10];
821 int zoneoffset, epoch_hour, hour, minute;
822 int status;
824 for (cp = nameline; *cp != '\n'; cp++) {
825 if (*cp == '\t')
826 timestamp = cp + 1;
828 if (!timestamp)
829 return 0;
832 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
833 * (west of GMT) or 1970-01-01 (east of GMT)
835 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
836 epoch_hour = 24;
837 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
838 epoch_hour = 0;
839 else
840 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 hour = strtol(timestamp, NULL, 10);
860 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
862 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
863 if (*colon == ':')
864 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
865 else
866 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
867 if (timestamp[m[3].rm_so] == '-')
868 zoneoffset = -zoneoffset;
870 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
874 * Get the name etc info from the ---/+++ lines of a traditional patch header
876 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
877 * files, we can happily check the index for a match, but for creating a
878 * new file we should try to match whatever "patch" does. I have no idea.
880 static int parse_traditional_patch(struct apply_state *state,
881 const char *first,
882 const char *second,
883 struct patch *patch)
885 char *name;
887 first += 4; /* skip "--- " */
888 second += 4; /* skip "+++ " */
889 if (!state->p_value_known) {
890 int p, q;
891 p = guess_p_value(state, first);
892 q = guess_p_value(state, second);
893 if (p < 0) p = q;
894 if (0 <= p && p == q) {
895 state->p_value = p;
896 state->p_value_known = 1;
899 if (is_dev_null(first)) {
900 patch->is_new = 1;
901 patch->is_delete = 0;
902 name = find_name_traditional(state, second, NULL, state->p_value);
903 patch->new_name = name;
904 } else if (is_dev_null(second)) {
905 patch->is_new = 0;
906 patch->is_delete = 1;
907 name = find_name_traditional(state, first, NULL, state->p_value);
908 patch->old_name = name;
909 } else {
910 char *first_name;
911 first_name = find_name_traditional(state, first, NULL, state->p_value);
912 name = find_name_traditional(state, second, first_name, state->p_value);
913 free(first_name);
914 if (has_epoch_timestamp(first)) {
915 patch->is_new = 1;
916 patch->is_delete = 0;
917 patch->new_name = name;
918 } else if (has_epoch_timestamp(second)) {
919 patch->is_new = 0;
920 patch->is_delete = 1;
921 patch->old_name = name;
922 } else {
923 patch->old_name = name;
924 patch->new_name = xstrdup_or_null(name);
927 if (!name)
928 return error(_("unable to find filename in patch at line %d"), state->linenr);
930 return 0;
933 static int gitdiff_hdrend(struct apply_state *state,
934 const char *line,
935 struct patch *patch)
937 return 1;
941 * We're anal about diff header consistency, to make
942 * sure that we don't end up having strange ambiguous
943 * patches floating around.
945 * As a result, gitdiff_{old|new}name() will check
946 * their names against any previous information, just
947 * to make sure..
949 #define DIFF_OLD_NAME 0
950 #define DIFF_NEW_NAME 1
952 static int gitdiff_verify_name(struct apply_state *state,
953 const char *line,
954 int isnull,
955 char **name,
956 int side)
958 if (!*name && !isnull) {
959 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
960 return 0;
963 if (*name) {
964 char *another;
965 if (isnull)
966 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
967 *name, state->linenr);
968 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
969 if (!another || strcmp(another, *name)) {
970 free(another);
971 return error((side == DIFF_NEW_NAME) ?
972 _("git apply: bad git-diff - inconsistent new filename on line %d") :
973 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
975 free(another);
976 } else {
977 if (!starts_with(line, "/dev/null\n"))
978 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
981 return 0;
984 static int gitdiff_oldname(struct apply_state *state,
985 const char *line,
986 struct patch *patch)
988 return gitdiff_verify_name(state, line,
989 patch->is_new, &patch->old_name,
990 DIFF_OLD_NAME);
993 static int gitdiff_newname(struct apply_state *state,
994 const char *line,
995 struct patch *patch)
997 return gitdiff_verify_name(state, line,
998 patch->is_delete, &patch->new_name,
999 DIFF_NEW_NAME);
1002 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
1004 char *end;
1005 *mode = strtoul(line, &end, 8);
1006 if (end == line || !isspace(*end))
1007 return error(_("invalid mode on line %d: %s"), linenr, line);
1008 return 0;
1011 static int gitdiff_oldmode(struct apply_state *state,
1012 const char *line,
1013 struct patch *patch)
1015 return parse_mode_line(line, state->linenr, &patch->old_mode);
1018 static int gitdiff_newmode(struct apply_state *state,
1019 const char *line,
1020 struct patch *patch)
1022 return parse_mode_line(line, state->linenr, &patch->new_mode);
1025 static int gitdiff_delete(struct apply_state *state,
1026 const char *line,
1027 struct patch *patch)
1029 patch->is_delete = 1;
1030 free(patch->old_name);
1031 patch->old_name = xstrdup_or_null(patch->def_name);
1032 return gitdiff_oldmode(state, line, patch);
1035 static int gitdiff_newfile(struct apply_state *state,
1036 const char *line,
1037 struct patch *patch)
1039 patch->is_new = 1;
1040 free(patch->new_name);
1041 patch->new_name = xstrdup_or_null(patch->def_name);
1042 return gitdiff_newmode(state, line, patch);
1045 static int gitdiff_copysrc(struct apply_state *state,
1046 const char *line,
1047 struct patch *patch)
1049 patch->is_copy = 1;
1050 free(patch->old_name);
1051 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1052 return 0;
1055 static int gitdiff_copydst(struct apply_state *state,
1056 const char *line,
1057 struct patch *patch)
1059 patch->is_copy = 1;
1060 free(patch->new_name);
1061 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1062 return 0;
1065 static int gitdiff_renamesrc(struct apply_state *state,
1066 const char *line,
1067 struct patch *patch)
1069 patch->is_rename = 1;
1070 free(patch->old_name);
1071 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1072 return 0;
1075 static int gitdiff_renamedst(struct apply_state *state,
1076 const char *line,
1077 struct patch *patch)
1079 patch->is_rename = 1;
1080 free(patch->new_name);
1081 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1082 return 0;
1085 static int gitdiff_similarity(struct apply_state *state,
1086 const char *line,
1087 struct patch *patch)
1089 unsigned long val = strtoul(line, NULL, 10);
1090 if (val <= 100)
1091 patch->score = val;
1092 return 0;
1095 static int gitdiff_dissimilarity(struct apply_state *state,
1096 const char *line,
1097 struct patch *patch)
1099 unsigned long val = strtoul(line, NULL, 10);
1100 if (val <= 100)
1101 patch->score = val;
1102 return 0;
1105 static int gitdiff_index(struct apply_state *state,
1106 const char *line,
1107 struct patch *patch)
1110 * index line is N hexadecimal, "..", N hexadecimal,
1111 * and optional space with octal mode.
1113 const char *ptr, *eol;
1114 int len;
1116 ptr = strchr(line, '.');
1117 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1118 return 0;
1119 len = ptr - line;
1120 memcpy(patch->old_sha1_prefix, line, len);
1121 patch->old_sha1_prefix[len] = 0;
1123 line = ptr + 2;
1124 ptr = strchr(line, ' ');
1125 eol = strchrnul(line, '\n');
1127 if (!ptr || eol < ptr)
1128 ptr = eol;
1129 len = ptr - line;
1131 if (40 < len)
1132 return 0;
1133 memcpy(patch->new_sha1_prefix, line, len);
1134 patch->new_sha1_prefix[len] = 0;
1135 if (*ptr == ' ')
1136 return gitdiff_oldmode(state, ptr + 1, patch);
1137 return 0;
1141 * This is normal for a diff that doesn't change anything: we'll fall through
1142 * into the next diff. Tell the parser to break out.
1144 static int gitdiff_unrecognized(struct apply_state *state,
1145 const char *line,
1146 struct patch *patch)
1148 return 1;
1152 * Skip p_value leading components from "line"; as we do not accept
1153 * absolute paths, return NULL in that case.
1155 static const char *skip_tree_prefix(struct apply_state *state,
1156 const char *line,
1157 int llen)
1159 int nslash;
1160 int i;
1162 if (!state->p_value)
1163 return (llen && line[0] == '/') ? NULL : line;
1165 nslash = state->p_value;
1166 for (i = 0; i < llen; i++) {
1167 int ch = line[i];
1168 if (ch == '/' && --nslash <= 0)
1169 return (i == 0) ? NULL : &line[i + 1];
1171 return NULL;
1175 * This is to extract the same name that appears on "diff --git"
1176 * line. We do not find and return anything if it is a rename
1177 * patch, and it is OK because we will find the name elsewhere.
1178 * We need to reliably find name only when it is mode-change only,
1179 * creation or deletion of an empty file. In any of these cases,
1180 * both sides are the same name under a/ and b/ respectively.
1182 static char *git_header_name(struct apply_state *state,
1183 const char *line,
1184 int llen)
1186 const char *name;
1187 const char *second = NULL;
1188 size_t len, line_len;
1190 line += strlen("diff --git ");
1191 llen -= strlen("diff --git ");
1193 if (*line == '"') {
1194 const char *cp;
1195 struct strbuf first = STRBUF_INIT;
1196 struct strbuf sp = STRBUF_INIT;
1198 if (unquote_c_style(&first, line, &second))
1199 goto free_and_fail1;
1201 /* strip the a/b prefix including trailing slash */
1202 cp = skip_tree_prefix(state, first.buf, first.len);
1203 if (!cp)
1204 goto free_and_fail1;
1205 strbuf_remove(&first, 0, cp - first.buf);
1208 * second points at one past closing dq of name.
1209 * find the second name.
1211 while ((second < line + llen) && isspace(*second))
1212 second++;
1214 if (line + llen <= second)
1215 goto free_and_fail1;
1216 if (*second == '"') {
1217 if (unquote_c_style(&sp, second, NULL))
1218 goto free_and_fail1;
1219 cp = skip_tree_prefix(state, sp.buf, sp.len);
1220 if (!cp)
1221 goto free_and_fail1;
1222 /* They must match, otherwise ignore */
1223 if (strcmp(cp, first.buf))
1224 goto free_and_fail1;
1225 strbuf_release(&sp);
1226 return strbuf_detach(&first, NULL);
1229 /* unquoted second */
1230 cp = skip_tree_prefix(state, second, line + llen - second);
1231 if (!cp)
1232 goto free_and_fail1;
1233 if (line + llen - cp != first.len ||
1234 memcmp(first.buf, cp, first.len))
1235 goto free_and_fail1;
1236 return strbuf_detach(&first, NULL);
1238 free_and_fail1:
1239 strbuf_release(&first);
1240 strbuf_release(&sp);
1241 return NULL;
1244 /* unquoted first name */
1245 name = skip_tree_prefix(state, line, llen);
1246 if (!name)
1247 return NULL;
1250 * since the first name is unquoted, a dq if exists must be
1251 * the beginning of the second name.
1253 for (second = name; second < line + llen; second++) {
1254 if (*second == '"') {
1255 struct strbuf sp = STRBUF_INIT;
1256 const char *np;
1258 if (unquote_c_style(&sp, second, NULL))
1259 goto free_and_fail2;
1261 np = skip_tree_prefix(state, sp.buf, sp.len);
1262 if (!np)
1263 goto free_and_fail2;
1265 len = sp.buf + sp.len - np;
1266 if (len < second - name &&
1267 !strncmp(np, name, len) &&
1268 isspace(name[len])) {
1269 /* Good */
1270 strbuf_remove(&sp, 0, np - sp.buf);
1271 return strbuf_detach(&sp, NULL);
1274 free_and_fail2:
1275 strbuf_release(&sp);
1276 return NULL;
1281 * Accept a name only if it shows up twice, exactly the same
1282 * form.
1284 second = strchr(name, '\n');
1285 if (!second)
1286 return NULL;
1287 line_len = second - name;
1288 for (len = 0 ; ; len++) {
1289 switch (name[len]) {
1290 default:
1291 continue;
1292 case '\n':
1293 return NULL;
1294 case '\t': case ' ':
1296 * Is this the separator between the preimage
1297 * and the postimage pathname? Again, we are
1298 * only interested in the case where there is
1299 * no rename, as this is only to set def_name
1300 * and a rename patch has the names elsewhere
1301 * in an unambiguous form.
1303 if (!name[len + 1])
1304 return NULL; /* no postimage name */
1305 second = skip_tree_prefix(state, name + len + 1,
1306 line_len - (len + 1));
1307 if (!second)
1308 return NULL;
1310 * Does len bytes starting at "name" and "second"
1311 * (that are separated by one HT or SP we just
1312 * found) exactly match?
1314 if (second[len] == '\n' && !strncmp(name, second, len))
1315 return xmemdupz(name, len);
1320 static int check_header_line(struct apply_state *state, struct patch *patch)
1322 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1323 (patch->is_rename == 1) + (patch->is_copy == 1);
1324 if (extensions > 1)
1325 return error(_("inconsistent header lines %d and %d"),
1326 patch->extension_linenr, state->linenr);
1327 if (extensions && !patch->extension_linenr)
1328 patch->extension_linenr = state->linenr;
1329 return 0;
1332 /* Verify that we recognize the lines following a git header */
1333 static int parse_git_header(struct apply_state *state,
1334 const char *line,
1335 int len,
1336 unsigned int size,
1337 struct patch *patch)
1339 unsigned long offset;
1341 /* A git diff has explicit new/delete information, so we don't guess */
1342 patch->is_new = 0;
1343 patch->is_delete = 0;
1346 * Some things may not have the old name in the
1347 * rest of the headers anywhere (pure mode changes,
1348 * or removing or adding empty files), so we get
1349 * the default name from the header.
1351 patch->def_name = git_header_name(state, line, len);
1352 if (patch->def_name && state->root.len) {
1353 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1354 free(patch->def_name);
1355 patch->def_name = s;
1358 line += len;
1359 size -= len;
1360 state->linenr++;
1361 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1362 static const struct opentry {
1363 const char *str;
1364 int (*fn)(struct apply_state *, const char *, struct patch *);
1365 } optable[] = {
1366 { "@@ -", gitdiff_hdrend },
1367 { "--- ", gitdiff_oldname },
1368 { "+++ ", gitdiff_newname },
1369 { "old mode ", gitdiff_oldmode },
1370 { "new mode ", gitdiff_newmode },
1371 { "deleted file mode ", gitdiff_delete },
1372 { "new file mode ", gitdiff_newfile },
1373 { "copy from ", gitdiff_copysrc },
1374 { "copy to ", gitdiff_copydst },
1375 { "rename old ", gitdiff_renamesrc },
1376 { "rename new ", gitdiff_renamedst },
1377 { "rename from ", gitdiff_renamesrc },
1378 { "rename to ", gitdiff_renamedst },
1379 { "similarity index ", gitdiff_similarity },
1380 { "dissimilarity index ", gitdiff_dissimilarity },
1381 { "index ", gitdiff_index },
1382 { "", gitdiff_unrecognized },
1384 int i;
1386 len = linelen(line, size);
1387 if (!len || line[len-1] != '\n')
1388 break;
1389 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1390 const struct opentry *p = optable + i;
1391 int oplen = strlen(p->str);
1392 int res;
1393 if (len < oplen || memcmp(p->str, line, oplen))
1394 continue;
1395 res = p->fn(state, line + oplen, patch);
1396 if (res < 0)
1397 return -1;
1398 if (check_header_line(state, patch))
1399 return -1;
1400 if (res > 0)
1401 return offset;
1402 break;
1406 return offset;
1409 static int parse_num(const char *line, unsigned long *p)
1411 char *ptr;
1413 if (!isdigit(*line))
1414 return 0;
1415 *p = strtoul(line, &ptr, 10);
1416 return ptr - line;
1419 static int parse_range(const char *line, int len, int offset, const char *expect,
1420 unsigned long *p1, unsigned long *p2)
1422 int digits, ex;
1424 if (offset < 0 || offset >= len)
1425 return -1;
1426 line += offset;
1427 len -= offset;
1429 digits = parse_num(line, p1);
1430 if (!digits)
1431 return -1;
1433 offset += digits;
1434 line += digits;
1435 len -= digits;
1437 *p2 = 1;
1438 if (*line == ',') {
1439 digits = parse_num(line+1, p2);
1440 if (!digits)
1441 return -1;
1443 offset += digits+1;
1444 line += digits+1;
1445 len -= digits+1;
1448 ex = strlen(expect);
1449 if (ex > len)
1450 return -1;
1451 if (memcmp(line, expect, ex))
1452 return -1;
1454 return offset + ex;
1457 static void recount_diff(const char *line, int size, struct fragment *fragment)
1459 int oldlines = 0, newlines = 0, ret = 0;
1461 if (size < 1) {
1462 warning("recount: ignore empty hunk");
1463 return;
1466 for (;;) {
1467 int len = linelen(line, size);
1468 size -= len;
1469 line += len;
1471 if (size < 1)
1472 break;
1474 switch (*line) {
1475 case ' ': case '\n':
1476 newlines++;
1477 /* fall through */
1478 case '-':
1479 oldlines++;
1480 continue;
1481 case '+':
1482 newlines++;
1483 continue;
1484 case '\\':
1485 continue;
1486 case '@':
1487 ret = size < 3 || !starts_with(line, "@@ ");
1488 break;
1489 case 'd':
1490 ret = size < 5 || !starts_with(line, "diff ");
1491 break;
1492 default:
1493 ret = -1;
1494 break;
1496 if (ret) {
1497 warning(_("recount: unexpected line: %.*s"),
1498 (int)linelen(line, size), line);
1499 return;
1501 break;
1503 fragment->oldlines = oldlines;
1504 fragment->newlines = newlines;
1508 * Parse a unified diff fragment header of the
1509 * form "@@ -a,b +c,d @@"
1511 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1513 int offset;
1515 if (!len || line[len-1] != '\n')
1516 return -1;
1518 /* Figure out the number of lines in a fragment */
1519 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1520 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1522 return offset;
1526 * Find file diff header
1528 * Returns:
1529 * -1 if no header was found
1530 * -128 in case of error
1531 * the size of the header in bytes (called "offset") otherwise
1533 static int find_header(struct apply_state *state,
1534 const char *line,
1535 unsigned long size,
1536 int *hdrsize,
1537 struct patch *patch)
1539 unsigned long offset, len;
1541 patch->is_toplevel_relative = 0;
1542 patch->is_rename = patch->is_copy = 0;
1543 patch->is_new = patch->is_delete = -1;
1544 patch->old_mode = patch->new_mode = 0;
1545 patch->old_name = patch->new_name = NULL;
1546 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1547 unsigned long nextlen;
1549 len = linelen(line, size);
1550 if (!len)
1551 break;
1553 /* Testing this early allows us to take a few shortcuts.. */
1554 if (len < 6)
1555 continue;
1558 * Make sure we don't find any unconnected patch fragments.
1559 * That's a sign that we didn't find a header, and that a
1560 * patch has become corrupted/broken up.
1562 if (!memcmp("@@ -", line, 4)) {
1563 struct fragment dummy;
1564 if (parse_fragment_header(line, len, &dummy) < 0)
1565 continue;
1566 error(_("patch fragment without header at line %d: %.*s"),
1567 state->linenr, (int)len-1, line);
1568 return -128;
1571 if (size < len + 6)
1572 break;
1575 * Git patch? It might not have a real patch, just a rename
1576 * or mode change, so we handle that specially
1578 if (!memcmp("diff --git ", line, 11)) {
1579 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1580 if (git_hdr_len < 0)
1581 return -128;
1582 if (git_hdr_len <= len)
1583 continue;
1584 if (!patch->old_name && !patch->new_name) {
1585 if (!patch->def_name) {
1586 error(Q_("git diff header lacks filename information when removing "
1587 "%d leading pathname component (line %d)",
1588 "git diff header lacks filename information when removing "
1589 "%d leading pathname components (line %d)",
1590 state->p_value),
1591 state->p_value, state->linenr);
1592 return -128;
1594 patch->old_name = xstrdup(patch->def_name);
1595 patch->new_name = xstrdup(patch->def_name);
1597 if ((!patch->new_name && !patch->is_delete) ||
1598 (!patch->old_name && !patch->is_new)) {
1599 error(_("git diff header lacks filename information "
1600 "(line %d)"), state->linenr);
1601 return -128;
1603 patch->is_toplevel_relative = 1;
1604 *hdrsize = git_hdr_len;
1605 return offset;
1608 /* --- followed by +++ ? */
1609 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1610 continue;
1613 * We only accept unified patches, so we want it to
1614 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1615 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1617 nextlen = linelen(line + len, size - len);
1618 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1619 continue;
1621 /* Ok, we'll consider it a patch */
1622 if (parse_traditional_patch(state, line, line+len, patch))
1623 return -128;
1624 *hdrsize = len + nextlen;
1625 state->linenr += 2;
1626 return offset;
1628 return -1;
1631 static void record_ws_error(struct apply_state *state,
1632 unsigned result,
1633 const char *line,
1634 int len,
1635 int linenr)
1637 char *err;
1639 if (!result)
1640 return;
1642 state->whitespace_error++;
1643 if (state->squelch_whitespace_errors &&
1644 state->squelch_whitespace_errors < state->whitespace_error)
1645 return;
1647 err = whitespace_error_string(result);
1648 if (state->apply_verbosity > verbosity_silent)
1649 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1650 state->patch_input_file, linenr, err, len, line);
1651 free(err);
1654 static void check_whitespace(struct apply_state *state,
1655 const char *line,
1656 int len,
1657 unsigned ws_rule)
1659 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1661 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1665 * Check if the patch has context lines with CRLF or
1666 * the patch wants to remove lines with CRLF.
1668 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1670 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1671 patch->ws_rule |= WS_CR_AT_EOL;
1672 patch->crlf_in_old = 1;
1678 * Parse a unified diff. Note that this really needs to parse each
1679 * fragment separately, since the only way to know the difference
1680 * between a "---" that is part of a patch, and a "---" that starts
1681 * the next patch is to look at the line counts..
1683 static int parse_fragment(struct apply_state *state,
1684 const char *line,
1685 unsigned long size,
1686 struct patch *patch,
1687 struct fragment *fragment)
1689 int added, deleted;
1690 int len = linelen(line, size), offset;
1691 unsigned long oldlines, newlines;
1692 unsigned long leading, trailing;
1694 offset = parse_fragment_header(line, len, fragment);
1695 if (offset < 0)
1696 return -1;
1697 if (offset > 0 && patch->recount)
1698 recount_diff(line + offset, size - offset, fragment);
1699 oldlines = fragment->oldlines;
1700 newlines = fragment->newlines;
1701 leading = 0;
1702 trailing = 0;
1704 /* Parse the thing.. */
1705 line += len;
1706 size -= len;
1707 state->linenr++;
1708 added = deleted = 0;
1709 for (offset = len;
1710 0 < size;
1711 offset += len, size -= len, line += len, state->linenr++) {
1712 if (!oldlines && !newlines)
1713 break;
1714 len = linelen(line, size);
1715 if (!len || line[len-1] != '\n')
1716 return -1;
1717 switch (*line) {
1718 default:
1719 return -1;
1720 case '\n': /* newer GNU diff, an empty context line */
1721 case ' ':
1722 oldlines--;
1723 newlines--;
1724 if (!deleted && !added)
1725 leading++;
1726 trailing++;
1727 check_old_for_crlf(patch, line, len);
1728 if (!state->apply_in_reverse &&
1729 state->ws_error_action == correct_ws_error)
1730 check_whitespace(state, line, len, patch->ws_rule);
1731 break;
1732 case '-':
1733 if (!state->apply_in_reverse)
1734 check_old_for_crlf(patch, line, len);
1735 if (state->apply_in_reverse &&
1736 state->ws_error_action != nowarn_ws_error)
1737 check_whitespace(state, line, len, patch->ws_rule);
1738 deleted++;
1739 oldlines--;
1740 trailing = 0;
1741 break;
1742 case '+':
1743 if (state->apply_in_reverse)
1744 check_old_for_crlf(patch, line, len);
1745 if (!state->apply_in_reverse &&
1746 state->ws_error_action != nowarn_ws_error)
1747 check_whitespace(state, line, len, patch->ws_rule);
1748 added++;
1749 newlines--;
1750 trailing = 0;
1751 break;
1754 * We allow "\ No newline at end of file". Depending
1755 * on locale settings when the patch was produced we
1756 * don't know what this line looks like. The only
1757 * thing we do know is that it begins with "\ ".
1758 * Checking for 12 is just for sanity check -- any
1759 * l10n of "\ No newline..." is at least that long.
1761 case '\\':
1762 if (len < 12 || memcmp(line, "\\ ", 2))
1763 return -1;
1764 break;
1767 if (oldlines || newlines)
1768 return -1;
1769 if (!deleted && !added)
1770 return -1;
1772 fragment->leading = leading;
1773 fragment->trailing = trailing;
1776 * If a fragment ends with an incomplete line, we failed to include
1777 * it in the above loop because we hit oldlines == newlines == 0
1778 * before seeing it.
1780 if (12 < size && !memcmp(line, "\\ ", 2))
1781 offset += linelen(line, size);
1783 patch->lines_added += added;
1784 patch->lines_deleted += deleted;
1786 if (0 < patch->is_new && oldlines)
1787 return error(_("new file depends on old contents"));
1788 if (0 < patch->is_delete && newlines)
1789 return error(_("deleted file still has contents"));
1790 return offset;
1794 * We have seen "diff --git a/... b/..." header (or a traditional patch
1795 * header). Read hunks that belong to this patch into fragments and hang
1796 * them to the given patch structure.
1798 * The (fragment->patch, fragment->size) pair points into the memory given
1799 * by the caller, not a copy, when we return.
1801 * Returns:
1802 * -1 in case of error,
1803 * the number of bytes in the patch otherwise.
1805 static int parse_single_patch(struct apply_state *state,
1806 const char *line,
1807 unsigned long size,
1808 struct patch *patch)
1810 unsigned long offset = 0;
1811 unsigned long oldlines = 0, newlines = 0, context = 0;
1812 struct fragment **fragp = &patch->fragments;
1814 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1815 struct fragment *fragment;
1816 int len;
1818 fragment = xcalloc(1, sizeof(*fragment));
1819 fragment->linenr = state->linenr;
1820 len = parse_fragment(state, line, size, patch, fragment);
1821 if (len <= 0) {
1822 free(fragment);
1823 return error(_("corrupt patch at line %d"), state->linenr);
1825 fragment->patch = line;
1826 fragment->size = len;
1827 oldlines += fragment->oldlines;
1828 newlines += fragment->newlines;
1829 context += fragment->leading + fragment->trailing;
1831 *fragp = fragment;
1832 fragp = &fragment->next;
1834 offset += len;
1835 line += len;
1836 size -= len;
1840 * If something was removed (i.e. we have old-lines) it cannot
1841 * be creation, and if something was added it cannot be
1842 * deletion. However, the reverse is not true; --unified=0
1843 * patches that only add are not necessarily creation even
1844 * though they do not have any old lines, and ones that only
1845 * delete are not necessarily deletion.
1847 * Unfortunately, a real creation/deletion patch do _not_ have
1848 * any context line by definition, so we cannot safely tell it
1849 * apart with --unified=0 insanity. At least if the patch has
1850 * more than one hunk it is not creation or deletion.
1852 if (patch->is_new < 0 &&
1853 (oldlines || (patch->fragments && patch->fragments->next)))
1854 patch->is_new = 0;
1855 if (patch->is_delete < 0 &&
1856 (newlines || (patch->fragments && patch->fragments->next)))
1857 patch->is_delete = 0;
1859 if (0 < patch->is_new && oldlines)
1860 return error(_("new file %s depends on old contents"), patch->new_name);
1861 if (0 < patch->is_delete && newlines)
1862 return error(_("deleted file %s still has contents"), patch->old_name);
1863 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1864 fprintf_ln(stderr,
1865 _("** warning: "
1866 "file %s becomes empty but is not deleted"),
1867 patch->new_name);
1869 return offset;
1872 static inline int metadata_changes(struct patch *patch)
1874 return patch->is_rename > 0 ||
1875 patch->is_copy > 0 ||
1876 patch->is_new > 0 ||
1877 patch->is_delete ||
1878 (patch->old_mode && patch->new_mode &&
1879 patch->old_mode != patch->new_mode);
1882 static char *inflate_it(const void *data, unsigned long size,
1883 unsigned long inflated_size)
1885 git_zstream stream;
1886 void *out;
1887 int st;
1889 memset(&stream, 0, sizeof(stream));
1891 stream.next_in = (unsigned char *)data;
1892 stream.avail_in = size;
1893 stream.next_out = out = xmalloc(inflated_size);
1894 stream.avail_out = inflated_size;
1895 git_inflate_init(&stream);
1896 st = git_inflate(&stream, Z_FINISH);
1897 git_inflate_end(&stream);
1898 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1899 free(out);
1900 return NULL;
1902 return out;
1906 * Read a binary hunk and return a new fragment; fragment->patch
1907 * points at an allocated memory that the caller must free, so
1908 * it is marked as "->free_patch = 1".
1910 static struct fragment *parse_binary_hunk(struct apply_state *state,
1911 char **buf_p,
1912 unsigned long *sz_p,
1913 int *status_p,
1914 int *used_p)
1917 * Expect a line that begins with binary patch method ("literal"
1918 * or "delta"), followed by the length of data before deflating.
1919 * a sequence of 'length-byte' followed by base-85 encoded data
1920 * should follow, terminated by a newline.
1922 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1923 * and we would limit the patch line to 66 characters,
1924 * so one line can fit up to 13 groups that would decode
1925 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1926 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1928 int llen, used;
1929 unsigned long size = *sz_p;
1930 char *buffer = *buf_p;
1931 int patch_method;
1932 unsigned long origlen;
1933 char *data = NULL;
1934 int hunk_size = 0;
1935 struct fragment *frag;
1937 llen = linelen(buffer, size);
1938 used = llen;
1940 *status_p = 0;
1942 if (starts_with(buffer, "delta ")) {
1943 patch_method = BINARY_DELTA_DEFLATED;
1944 origlen = strtoul(buffer + 6, NULL, 10);
1946 else if (starts_with(buffer, "literal ")) {
1947 patch_method = BINARY_LITERAL_DEFLATED;
1948 origlen = strtoul(buffer + 8, NULL, 10);
1950 else
1951 return NULL;
1953 state->linenr++;
1954 buffer += llen;
1955 while (1) {
1956 int byte_length, max_byte_length, newsize;
1957 llen = linelen(buffer, size);
1958 used += llen;
1959 state->linenr++;
1960 if (llen == 1) {
1961 /* consume the blank line */
1962 buffer++;
1963 size--;
1964 break;
1967 * Minimum line is "A00000\n" which is 7-byte long,
1968 * and the line length must be multiple of 5 plus 2.
1970 if ((llen < 7) || (llen-2) % 5)
1971 goto corrupt;
1972 max_byte_length = (llen - 2) / 5 * 4;
1973 byte_length = *buffer;
1974 if ('A' <= byte_length && byte_length <= 'Z')
1975 byte_length = byte_length - 'A' + 1;
1976 else if ('a' <= byte_length && byte_length <= 'z')
1977 byte_length = byte_length - 'a' + 27;
1978 else
1979 goto corrupt;
1980 /* if the input length was not multiple of 4, we would
1981 * have filler at the end but the filler should never
1982 * exceed 3 bytes
1984 if (max_byte_length < byte_length ||
1985 byte_length <= max_byte_length - 4)
1986 goto corrupt;
1987 newsize = hunk_size + byte_length;
1988 data = xrealloc(data, newsize);
1989 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1990 goto corrupt;
1991 hunk_size = newsize;
1992 buffer += llen;
1993 size -= llen;
1996 frag = xcalloc(1, sizeof(*frag));
1997 frag->patch = inflate_it(data, hunk_size, origlen);
1998 frag->free_patch = 1;
1999 if (!frag->patch)
2000 goto corrupt;
2001 free(data);
2002 frag->size = origlen;
2003 *buf_p = buffer;
2004 *sz_p = size;
2005 *used_p = used;
2006 frag->binary_patch_method = patch_method;
2007 return frag;
2009 corrupt:
2010 free(data);
2011 *status_p = -1;
2012 error(_("corrupt binary patch at line %d: %.*s"),
2013 state->linenr-1, llen-1, buffer);
2014 return NULL;
2018 * Returns:
2019 * -1 in case of error,
2020 * the length of the parsed binary patch otherwise
2022 static int parse_binary(struct apply_state *state,
2023 char *buffer,
2024 unsigned long size,
2025 struct patch *patch)
2028 * We have read "GIT binary patch\n"; what follows is a line
2029 * that says the patch method (currently, either "literal" or
2030 * "delta") and the length of data before deflating; a
2031 * sequence of 'length-byte' followed by base-85 encoded data
2032 * follows.
2034 * When a binary patch is reversible, there is another binary
2035 * hunk in the same format, starting with patch method (either
2036 * "literal" or "delta") with the length of data, and a sequence
2037 * of length-byte + base-85 encoded data, terminated with another
2038 * empty line. This data, when applied to the postimage, produces
2039 * the preimage.
2041 struct fragment *forward;
2042 struct fragment *reverse;
2043 int status;
2044 int used, used_1;
2046 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2047 if (!forward && !status)
2048 /* there has to be one hunk (forward hunk) */
2049 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2050 if (status)
2051 /* otherwise we already gave an error message */
2052 return status;
2054 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2055 if (reverse)
2056 used += used_1;
2057 else if (status) {
2059 * Not having reverse hunk is not an error, but having
2060 * a corrupt reverse hunk is.
2062 free((void*) forward->patch);
2063 free(forward);
2064 return status;
2066 forward->next = reverse;
2067 patch->fragments = forward;
2068 patch->is_binary = 1;
2069 return used;
2072 static void prefix_one(struct apply_state *state, char **name)
2074 char *old_name = *name;
2075 if (!old_name)
2076 return;
2077 *name = prefix_filename(state->prefix, *name);
2078 free(old_name);
2081 static void prefix_patch(struct apply_state *state, struct patch *p)
2083 if (!state->prefix || p->is_toplevel_relative)
2084 return;
2085 prefix_one(state, &p->new_name);
2086 prefix_one(state, &p->old_name);
2090 * include/exclude
2093 static void add_name_limit(struct apply_state *state,
2094 const char *name,
2095 int exclude)
2097 struct string_list_item *it;
2099 it = string_list_append(&state->limit_by_name, name);
2100 it->util = exclude ? NULL : (void *) 1;
2103 static int use_patch(struct apply_state *state, struct patch *p)
2105 const char *pathname = p->new_name ? p->new_name : p->old_name;
2106 int i;
2108 /* Paths outside are not touched regardless of "--include" */
2109 if (state->prefix && *state->prefix) {
2110 const char *rest;
2111 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2112 return 0;
2115 /* See if it matches any of exclude/include rule */
2116 for (i = 0; i < state->limit_by_name.nr; i++) {
2117 struct string_list_item *it = &state->limit_by_name.items[i];
2118 if (!wildmatch(it->string, pathname, 0))
2119 return (it->util != NULL);
2123 * If we had any include, a path that does not match any rule is
2124 * not used. Otherwise, we saw bunch of exclude rules (or none)
2125 * and such a path is used.
2127 return !state->has_include;
2131 * Read the patch text in "buffer" that extends for "size" bytes; stop
2132 * reading after seeing a single patch (i.e. changes to a single file).
2133 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2135 * Returns:
2136 * -1 if no header was found or parse_binary() failed,
2137 * -128 on another error,
2138 * the number of bytes consumed otherwise,
2139 * so that the caller can call us again for the next patch.
2141 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2143 int hdrsize, patchsize;
2144 int offset = find_header(state, buffer, size, &hdrsize, patch);
2146 if (offset < 0)
2147 return offset;
2149 prefix_patch(state, patch);
2151 if (!use_patch(state, patch))
2152 patch->ws_rule = 0;
2153 else
2154 patch->ws_rule = whitespace_rule(patch->new_name
2155 ? patch->new_name
2156 : patch->old_name);
2158 patchsize = parse_single_patch(state,
2159 buffer + offset + hdrsize,
2160 size - offset - hdrsize,
2161 patch);
2163 if (patchsize < 0)
2164 return -128;
2166 if (!patchsize) {
2167 static const char git_binary[] = "GIT binary patch\n";
2168 int hd = hdrsize + offset;
2169 unsigned long llen = linelen(buffer + hd, size - hd);
2171 if (llen == sizeof(git_binary) - 1 &&
2172 !memcmp(git_binary, buffer + hd, llen)) {
2173 int used;
2174 state->linenr++;
2175 used = parse_binary(state, buffer + hd + llen,
2176 size - hd - llen, patch);
2177 if (used < 0)
2178 return -1;
2179 if (used)
2180 patchsize = used + llen;
2181 else
2182 patchsize = 0;
2184 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2185 static const char *binhdr[] = {
2186 "Binary files ",
2187 "Files ",
2188 NULL,
2190 int i;
2191 for (i = 0; binhdr[i]; i++) {
2192 int len = strlen(binhdr[i]);
2193 if (len < size - hd &&
2194 !memcmp(binhdr[i], buffer + hd, len)) {
2195 state->linenr++;
2196 patch->is_binary = 1;
2197 patchsize = llen;
2198 break;
2203 /* Empty patch cannot be applied if it is a text patch
2204 * without metadata change. A binary patch appears
2205 * empty to us here.
2207 if ((state->apply || state->check) &&
2208 (!patch->is_binary && !metadata_changes(patch))) {
2209 error(_("patch with only garbage at line %d"), state->linenr);
2210 return -128;
2214 return offset + hdrsize + patchsize;
2217 static void reverse_patches(struct patch *p)
2219 for (; p; p = p->next) {
2220 struct fragment *frag = p->fragments;
2222 SWAP(p->new_name, p->old_name);
2223 SWAP(p->new_mode, p->old_mode);
2224 SWAP(p->is_new, p->is_delete);
2225 SWAP(p->lines_added, p->lines_deleted);
2226 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2228 for (; frag; frag = frag->next) {
2229 SWAP(frag->newpos, frag->oldpos);
2230 SWAP(frag->newlines, frag->oldlines);
2235 static const char pluses[] =
2236 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2237 static const char minuses[]=
2238 "----------------------------------------------------------------------";
2240 static void show_stats(struct apply_state *state, struct patch *patch)
2242 struct strbuf qname = STRBUF_INIT;
2243 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2244 int max, add, del;
2246 quote_c_style(cp, &qname, NULL, 0);
2249 * "scale" the filename
2251 max = state->max_len;
2252 if (max > 50)
2253 max = 50;
2255 if (qname.len > max) {
2256 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2257 if (!cp)
2258 cp = qname.buf + qname.len + 3 - max;
2259 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2262 if (patch->is_binary) {
2263 printf(" %-*s | Bin\n", max, qname.buf);
2264 strbuf_release(&qname);
2265 return;
2268 printf(" %-*s |", max, qname.buf);
2269 strbuf_release(&qname);
2272 * scale the add/delete
2274 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2275 add = patch->lines_added;
2276 del = patch->lines_deleted;
2278 if (state->max_change > 0) {
2279 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2280 add = (add * max + state->max_change / 2) / state->max_change;
2281 del = total - add;
2283 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2284 add, pluses, del, minuses);
2287 static int read_old_data(struct stat *st, struct patch *patch,
2288 const char *path, struct strbuf *buf)
2290 enum safe_crlf safe_crlf = patch->crlf_in_old ?
2291 SAFE_CRLF_KEEP_CRLF : SAFE_CRLF_RENORMALIZE;
2292 switch (st->st_mode & S_IFMT) {
2293 case S_IFLNK:
2294 if (strbuf_readlink(buf, path, st->st_size) < 0)
2295 return error(_("unable to read symlink %s"), path);
2296 return 0;
2297 case S_IFREG:
2298 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2299 return error(_("unable to open or read %s"), path);
2301 * "git apply" without "--index/--cached" should never look
2302 * at the index; the target file may not have been added to
2303 * the index yet, and we may not even be in any Git repository.
2304 * Pass NULL to convert_to_git() to stress this; the function
2305 * should never look at the index when explicit crlf option
2306 * is given.
2308 convert_to_git(NULL, path, buf->buf, buf->len, buf, safe_crlf);
2309 return 0;
2310 default:
2311 return -1;
2316 * Update the preimage, and the common lines in postimage,
2317 * from buffer buf of length len. If postlen is 0 the postimage
2318 * is updated in place, otherwise it's updated on a new buffer
2319 * of length postlen
2322 static void update_pre_post_images(struct image *preimage,
2323 struct image *postimage,
2324 char *buf,
2325 size_t len, size_t postlen)
2327 int i, ctx, reduced;
2328 char *new, *old, *fixed;
2329 struct image fixed_preimage;
2332 * Update the preimage with whitespace fixes. Note that we
2333 * are not losing preimage->buf -- apply_one_fragment() will
2334 * free "oldlines".
2336 prepare_image(&fixed_preimage, buf, len, 1);
2337 assert(postlen
2338 ? fixed_preimage.nr == preimage->nr
2339 : fixed_preimage.nr <= preimage->nr);
2340 for (i = 0; i < fixed_preimage.nr; i++)
2341 fixed_preimage.line[i].flag = preimage->line[i].flag;
2342 free(preimage->line_allocated);
2343 *preimage = fixed_preimage;
2346 * Adjust the common context lines in postimage. This can be
2347 * done in-place when we are shrinking it with whitespace
2348 * fixing, but needs a new buffer when ignoring whitespace or
2349 * expanding leading tabs to spaces.
2351 * We trust the caller to tell us if the update can be done
2352 * in place (postlen==0) or not.
2354 old = postimage->buf;
2355 if (postlen)
2356 new = postimage->buf = xmalloc(postlen);
2357 else
2358 new = old;
2359 fixed = preimage->buf;
2361 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2362 size_t l_len = postimage->line[i].len;
2363 if (!(postimage->line[i].flag & LINE_COMMON)) {
2364 /* an added line -- no counterparts in preimage */
2365 memmove(new, old, l_len);
2366 old += l_len;
2367 new += l_len;
2368 continue;
2371 /* a common context -- skip it in the original postimage */
2372 old += l_len;
2374 /* and find the corresponding one in the fixed preimage */
2375 while (ctx < preimage->nr &&
2376 !(preimage->line[ctx].flag & LINE_COMMON)) {
2377 fixed += preimage->line[ctx].len;
2378 ctx++;
2382 * preimage is expected to run out, if the caller
2383 * fixed addition of trailing blank lines.
2385 if (preimage->nr <= ctx) {
2386 reduced++;
2387 continue;
2390 /* and copy it in, while fixing the line length */
2391 l_len = preimage->line[ctx].len;
2392 memcpy(new, fixed, l_len);
2393 new += l_len;
2394 fixed += l_len;
2395 postimage->line[i].len = l_len;
2396 ctx++;
2399 if (postlen
2400 ? postlen < new - postimage->buf
2401 : postimage->len < new - postimage->buf)
2402 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2403 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2405 /* Fix the length of the whole thing */
2406 postimage->len = new - postimage->buf;
2407 postimage->nr -= reduced;
2410 static int line_by_line_fuzzy_match(struct image *img,
2411 struct image *preimage,
2412 struct image *postimage,
2413 unsigned long try,
2414 int try_lno,
2415 int preimage_limit)
2417 int i;
2418 size_t imgoff = 0;
2419 size_t preoff = 0;
2420 size_t postlen = postimage->len;
2421 size_t extra_chars;
2422 char *buf;
2423 char *preimage_eof;
2424 char *preimage_end;
2425 struct strbuf fixed;
2426 char *fixed_buf;
2427 size_t fixed_len;
2429 for (i = 0; i < preimage_limit; i++) {
2430 size_t prelen = preimage->line[i].len;
2431 size_t imglen = img->line[try_lno+i].len;
2433 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2434 preimage->buf + preoff, prelen))
2435 return 0;
2436 if (preimage->line[i].flag & LINE_COMMON)
2437 postlen += imglen - prelen;
2438 imgoff += imglen;
2439 preoff += prelen;
2443 * Ok, the preimage matches with whitespace fuzz.
2445 * imgoff now holds the true length of the target that
2446 * matches the preimage before the end of the file.
2448 * Count the number of characters in the preimage that fall
2449 * beyond the end of the file and make sure that all of them
2450 * are whitespace characters. (This can only happen if
2451 * we are removing blank lines at the end of the file.)
2453 buf = preimage_eof = preimage->buf + preoff;
2454 for ( ; i < preimage->nr; i++)
2455 preoff += preimage->line[i].len;
2456 preimage_end = preimage->buf + preoff;
2457 for ( ; buf < preimage_end; buf++)
2458 if (!isspace(*buf))
2459 return 0;
2462 * Update the preimage and the common postimage context
2463 * lines to use the same whitespace as the target.
2464 * If whitespace is missing in the target (i.e.
2465 * if the preimage extends beyond the end of the file),
2466 * use the whitespace from the preimage.
2468 extra_chars = preimage_end - preimage_eof;
2469 strbuf_init(&fixed, imgoff + extra_chars);
2470 strbuf_add(&fixed, img->buf + try, imgoff);
2471 strbuf_add(&fixed, preimage_eof, extra_chars);
2472 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2473 update_pre_post_images(preimage, postimage,
2474 fixed_buf, fixed_len, postlen);
2475 return 1;
2478 static int match_fragment(struct apply_state *state,
2479 struct image *img,
2480 struct image *preimage,
2481 struct image *postimage,
2482 unsigned long try,
2483 int try_lno,
2484 unsigned ws_rule,
2485 int match_beginning, int match_end)
2487 int i;
2488 char *fixed_buf, *buf, *orig, *target;
2489 struct strbuf fixed;
2490 size_t fixed_len, postlen;
2491 int preimage_limit;
2493 if (preimage->nr + try_lno <= img->nr) {
2495 * The hunk falls within the boundaries of img.
2497 preimage_limit = preimage->nr;
2498 if (match_end && (preimage->nr + try_lno != img->nr))
2499 return 0;
2500 } else if (state->ws_error_action == correct_ws_error &&
2501 (ws_rule & WS_BLANK_AT_EOF)) {
2503 * This hunk extends beyond the end of img, and we are
2504 * removing blank lines at the end of the file. This
2505 * many lines from the beginning of the preimage must
2506 * match with img, and the remainder of the preimage
2507 * must be blank.
2509 preimage_limit = img->nr - try_lno;
2510 } else {
2512 * The hunk extends beyond the end of the img and
2513 * we are not removing blanks at the end, so we
2514 * should reject the hunk at this position.
2516 return 0;
2519 if (match_beginning && try_lno)
2520 return 0;
2522 /* Quick hash check */
2523 for (i = 0; i < preimage_limit; i++)
2524 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2525 (preimage->line[i].hash != img->line[try_lno + i].hash))
2526 return 0;
2528 if (preimage_limit == preimage->nr) {
2530 * Do we have an exact match? If we were told to match
2531 * at the end, size must be exactly at try+fragsize,
2532 * otherwise try+fragsize must be still within the preimage,
2533 * and either case, the old piece should match the preimage
2534 * exactly.
2536 if ((match_end
2537 ? (try + preimage->len == img->len)
2538 : (try + preimage->len <= img->len)) &&
2539 !memcmp(img->buf + try, preimage->buf, preimage->len))
2540 return 1;
2541 } else {
2543 * The preimage extends beyond the end of img, so
2544 * there cannot be an exact match.
2546 * There must be one non-blank context line that match
2547 * a line before the end of img.
2549 char *buf_end;
2551 buf = preimage->buf;
2552 buf_end = buf;
2553 for (i = 0; i < preimage_limit; i++)
2554 buf_end += preimage->line[i].len;
2556 for ( ; buf < buf_end; buf++)
2557 if (!isspace(*buf))
2558 break;
2559 if (buf == buf_end)
2560 return 0;
2564 * No exact match. If we are ignoring whitespace, run a line-by-line
2565 * fuzzy matching. We collect all the line length information because
2566 * we need it to adjust whitespace if we match.
2568 if (state->ws_ignore_action == ignore_ws_change)
2569 return line_by_line_fuzzy_match(img, preimage, postimage,
2570 try, try_lno, preimage_limit);
2572 if (state->ws_error_action != correct_ws_error)
2573 return 0;
2576 * The hunk does not apply byte-by-byte, but the hash says
2577 * it might with whitespace fuzz. We weren't asked to
2578 * ignore whitespace, we were asked to correct whitespace
2579 * errors, so let's try matching after whitespace correction.
2581 * While checking the preimage against the target, whitespace
2582 * errors in both fixed, we count how large the corresponding
2583 * postimage needs to be. The postimage prepared by
2584 * apply_one_fragment() has whitespace errors fixed on added
2585 * lines already, but the common lines were propagated as-is,
2586 * which may become longer when their whitespace errors are
2587 * fixed.
2590 /* First count added lines in postimage */
2591 postlen = 0;
2592 for (i = 0; i < postimage->nr; i++) {
2593 if (!(postimage->line[i].flag & LINE_COMMON))
2594 postlen += postimage->line[i].len;
2598 * The preimage may extend beyond the end of the file,
2599 * but in this loop we will only handle the part of the
2600 * preimage that falls within the file.
2602 strbuf_init(&fixed, preimage->len + 1);
2603 orig = preimage->buf;
2604 target = img->buf + try;
2605 for (i = 0; i < preimage_limit; i++) {
2606 size_t oldlen = preimage->line[i].len;
2607 size_t tgtlen = img->line[try_lno + i].len;
2608 size_t fixstart = fixed.len;
2609 struct strbuf tgtfix;
2610 int match;
2612 /* Try fixing the line in the preimage */
2613 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2615 /* Try fixing the line in the target */
2616 strbuf_init(&tgtfix, tgtlen);
2617 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2620 * If they match, either the preimage was based on
2621 * a version before our tree fixed whitespace breakage,
2622 * or we are lacking a whitespace-fix patch the tree
2623 * the preimage was based on already had (i.e. target
2624 * has whitespace breakage, the preimage doesn't).
2625 * In either case, we are fixing the whitespace breakages
2626 * so we might as well take the fix together with their
2627 * real change.
2629 match = (tgtfix.len == fixed.len - fixstart &&
2630 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2631 fixed.len - fixstart));
2633 /* Add the length if this is common with the postimage */
2634 if (preimage->line[i].flag & LINE_COMMON)
2635 postlen += tgtfix.len;
2637 strbuf_release(&tgtfix);
2638 if (!match)
2639 goto unmatch_exit;
2641 orig += oldlen;
2642 target += tgtlen;
2647 * Now handle the lines in the preimage that falls beyond the
2648 * end of the file (if any). They will only match if they are
2649 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2650 * false).
2652 for ( ; i < preimage->nr; i++) {
2653 size_t fixstart = fixed.len; /* start of the fixed preimage */
2654 size_t oldlen = preimage->line[i].len;
2655 int j;
2657 /* Try fixing the line in the preimage */
2658 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2660 for (j = fixstart; j < fixed.len; j++)
2661 if (!isspace(fixed.buf[j]))
2662 goto unmatch_exit;
2664 orig += oldlen;
2668 * Yes, the preimage is based on an older version that still
2669 * has whitespace breakages unfixed, and fixing them makes the
2670 * hunk match. Update the context lines in the postimage.
2672 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2673 if (postlen < postimage->len)
2674 postlen = 0;
2675 update_pre_post_images(preimage, postimage,
2676 fixed_buf, fixed_len, postlen);
2677 return 1;
2679 unmatch_exit:
2680 strbuf_release(&fixed);
2681 return 0;
2684 static int find_pos(struct apply_state *state,
2685 struct image *img,
2686 struct image *preimage,
2687 struct image *postimage,
2688 int line,
2689 unsigned ws_rule,
2690 int match_beginning, int match_end)
2692 int i;
2693 unsigned long backwards, forwards, try;
2694 int backwards_lno, forwards_lno, try_lno;
2697 * If match_beginning or match_end is specified, there is no
2698 * point starting from a wrong line that will never match and
2699 * wander around and wait for a match at the specified end.
2701 if (match_beginning)
2702 line = 0;
2703 else if (match_end)
2704 line = img->nr - preimage->nr;
2707 * Because the comparison is unsigned, the following test
2708 * will also take care of a negative line number that can
2709 * result when match_end and preimage is larger than the target.
2711 if ((size_t) line > img->nr)
2712 line = img->nr;
2714 try = 0;
2715 for (i = 0; i < line; i++)
2716 try += img->line[i].len;
2719 * There's probably some smart way to do this, but I'll leave
2720 * that to the smart and beautiful people. I'm simple and stupid.
2722 backwards = try;
2723 backwards_lno = line;
2724 forwards = try;
2725 forwards_lno = line;
2726 try_lno = line;
2728 for (i = 0; ; i++) {
2729 if (match_fragment(state, img, preimage, postimage,
2730 try, try_lno, ws_rule,
2731 match_beginning, match_end))
2732 return try_lno;
2734 again:
2735 if (backwards_lno == 0 && forwards_lno == img->nr)
2736 break;
2738 if (i & 1) {
2739 if (backwards_lno == 0) {
2740 i++;
2741 goto again;
2743 backwards_lno--;
2744 backwards -= img->line[backwards_lno].len;
2745 try = backwards;
2746 try_lno = backwards_lno;
2747 } else {
2748 if (forwards_lno == img->nr) {
2749 i++;
2750 goto again;
2752 forwards += img->line[forwards_lno].len;
2753 forwards_lno++;
2754 try = forwards;
2755 try_lno = forwards_lno;
2759 return -1;
2762 static void remove_first_line(struct image *img)
2764 img->buf += img->line[0].len;
2765 img->len -= img->line[0].len;
2766 img->line++;
2767 img->nr--;
2770 static void remove_last_line(struct image *img)
2772 img->len -= img->line[--img->nr].len;
2776 * The change from "preimage" and "postimage" has been found to
2777 * apply at applied_pos (counts in line numbers) in "img".
2778 * Update "img" to remove "preimage" and replace it with "postimage".
2780 static void update_image(struct apply_state *state,
2781 struct image *img,
2782 int applied_pos,
2783 struct image *preimage,
2784 struct image *postimage)
2787 * remove the copy of preimage at offset in img
2788 * and replace it with postimage
2790 int i, nr;
2791 size_t remove_count, insert_count, applied_at = 0;
2792 char *result;
2793 int preimage_limit;
2796 * If we are removing blank lines at the end of img,
2797 * the preimage may extend beyond the end.
2798 * If that is the case, we must be careful only to
2799 * remove the part of the preimage that falls within
2800 * the boundaries of img. Initialize preimage_limit
2801 * to the number of lines in the preimage that falls
2802 * within the boundaries.
2804 preimage_limit = preimage->nr;
2805 if (preimage_limit > img->nr - applied_pos)
2806 preimage_limit = img->nr - applied_pos;
2808 for (i = 0; i < applied_pos; i++)
2809 applied_at += img->line[i].len;
2811 remove_count = 0;
2812 for (i = 0; i < preimage_limit; i++)
2813 remove_count += img->line[applied_pos + i].len;
2814 insert_count = postimage->len;
2816 /* Adjust the contents */
2817 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2818 memcpy(result, img->buf, applied_at);
2819 memcpy(result + applied_at, postimage->buf, postimage->len);
2820 memcpy(result + applied_at + postimage->len,
2821 img->buf + (applied_at + remove_count),
2822 img->len - (applied_at + remove_count));
2823 free(img->buf);
2824 img->buf = result;
2825 img->len += insert_count - remove_count;
2826 result[img->len] = '\0';
2828 /* Adjust the line table */
2829 nr = img->nr + postimage->nr - preimage_limit;
2830 if (preimage_limit < postimage->nr) {
2832 * NOTE: this knows that we never call remove_first_line()
2833 * on anything other than pre/post image.
2835 REALLOC_ARRAY(img->line, nr);
2836 img->line_allocated = img->line;
2838 if (preimage_limit != postimage->nr)
2839 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2840 img->line + applied_pos + preimage_limit,
2841 img->nr - (applied_pos + preimage_limit));
2842 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2843 if (!state->allow_overlap)
2844 for (i = 0; i < postimage->nr; i++)
2845 img->line[applied_pos + i].flag |= LINE_PATCHED;
2846 img->nr = nr;
2850 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2851 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2852 * replace the part of "img" with "postimage" text.
2854 static int apply_one_fragment(struct apply_state *state,
2855 struct image *img, struct fragment *frag,
2856 int inaccurate_eof, unsigned ws_rule,
2857 int nth_fragment)
2859 int match_beginning, match_end;
2860 const char *patch = frag->patch;
2861 int size = frag->size;
2862 char *old, *oldlines;
2863 struct strbuf newlines;
2864 int new_blank_lines_at_end = 0;
2865 int found_new_blank_lines_at_end = 0;
2866 int hunk_linenr = frag->linenr;
2867 unsigned long leading, trailing;
2868 int pos, applied_pos;
2869 struct image preimage;
2870 struct image postimage;
2872 memset(&preimage, 0, sizeof(preimage));
2873 memset(&postimage, 0, sizeof(postimage));
2874 oldlines = xmalloc(size);
2875 strbuf_init(&newlines, size);
2877 old = oldlines;
2878 while (size > 0) {
2879 char first;
2880 int len = linelen(patch, size);
2881 int plen;
2882 int added_blank_line = 0;
2883 int is_blank_context = 0;
2884 size_t start;
2886 if (!len)
2887 break;
2890 * "plen" is how much of the line we should use for
2891 * the actual patch data. Normally we just remove the
2892 * first character on the line, but if the line is
2893 * followed by "\ No newline", then we also remove the
2894 * last one (which is the newline, of course).
2896 plen = len - 1;
2897 if (len < size && patch[len] == '\\')
2898 plen--;
2899 first = *patch;
2900 if (state->apply_in_reverse) {
2901 if (first == '-')
2902 first = '+';
2903 else if (first == '+')
2904 first = '-';
2907 switch (first) {
2908 case '\n':
2909 /* Newer GNU diff, empty context line */
2910 if (plen < 0)
2911 /* ... followed by '\No newline'; nothing */
2912 break;
2913 *old++ = '\n';
2914 strbuf_addch(&newlines, '\n');
2915 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2916 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2917 is_blank_context = 1;
2918 break;
2919 case ' ':
2920 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2921 ws_blank_line(patch + 1, plen, ws_rule))
2922 is_blank_context = 1;
2923 /* fallthrough */
2924 case '-':
2925 memcpy(old, patch + 1, plen);
2926 add_line_info(&preimage, old, plen,
2927 (first == ' ' ? LINE_COMMON : 0));
2928 old += plen;
2929 if (first == '-')
2930 break;
2931 /* fallthrough */
2932 case '+':
2933 /* --no-add does not add new lines */
2934 if (first == '+' && state->no_add)
2935 break;
2937 start = newlines.len;
2938 if (first != '+' ||
2939 !state->whitespace_error ||
2940 state->ws_error_action != correct_ws_error) {
2941 strbuf_add(&newlines, patch + 1, plen);
2943 else {
2944 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2946 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2947 (first == '+' ? 0 : LINE_COMMON));
2948 if (first == '+' &&
2949 (ws_rule & WS_BLANK_AT_EOF) &&
2950 ws_blank_line(patch + 1, plen, ws_rule))
2951 added_blank_line = 1;
2952 break;
2953 case '@': case '\\':
2954 /* Ignore it, we already handled it */
2955 break;
2956 default:
2957 if (state->apply_verbosity > verbosity_normal)
2958 error(_("invalid start of line: '%c'"), first);
2959 applied_pos = -1;
2960 goto out;
2962 if (added_blank_line) {
2963 if (!new_blank_lines_at_end)
2964 found_new_blank_lines_at_end = hunk_linenr;
2965 new_blank_lines_at_end++;
2967 else if (is_blank_context)
2969 else
2970 new_blank_lines_at_end = 0;
2971 patch += len;
2972 size -= len;
2973 hunk_linenr++;
2975 if (inaccurate_eof &&
2976 old > oldlines && old[-1] == '\n' &&
2977 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2978 old--;
2979 strbuf_setlen(&newlines, newlines.len - 1);
2982 leading = frag->leading;
2983 trailing = frag->trailing;
2986 * A hunk to change lines at the beginning would begin with
2987 * @@ -1,L +N,M @@
2988 * but we need to be careful. -U0 that inserts before the second
2989 * line also has this pattern.
2991 * And a hunk to add to an empty file would begin with
2992 * @@ -0,0 +N,M @@
2994 * In other words, a hunk that is (frag->oldpos <= 1) with or
2995 * without leading context must match at the beginning.
2997 match_beginning = (!frag->oldpos ||
2998 (frag->oldpos == 1 && !state->unidiff_zero));
3001 * A hunk without trailing lines must match at the end.
3002 * However, we simply cannot tell if a hunk must match end
3003 * from the lack of trailing lines if the patch was generated
3004 * with unidiff without any context.
3006 match_end = !state->unidiff_zero && !trailing;
3008 pos = frag->newpos ? (frag->newpos - 1) : 0;
3009 preimage.buf = oldlines;
3010 preimage.len = old - oldlines;
3011 postimage.buf = newlines.buf;
3012 postimage.len = newlines.len;
3013 preimage.line = preimage.line_allocated;
3014 postimage.line = postimage.line_allocated;
3016 for (;;) {
3018 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3019 ws_rule, match_beginning, match_end);
3021 if (applied_pos >= 0)
3022 break;
3024 /* Am I at my context limits? */
3025 if ((leading <= state->p_context) && (trailing <= state->p_context))
3026 break;
3027 if (match_beginning || match_end) {
3028 match_beginning = match_end = 0;
3029 continue;
3033 * Reduce the number of context lines; reduce both
3034 * leading and trailing if they are equal otherwise
3035 * just reduce the larger context.
3037 if (leading >= trailing) {
3038 remove_first_line(&preimage);
3039 remove_first_line(&postimage);
3040 pos--;
3041 leading--;
3043 if (trailing > leading) {
3044 remove_last_line(&preimage);
3045 remove_last_line(&postimage);
3046 trailing--;
3050 if (applied_pos >= 0) {
3051 if (new_blank_lines_at_end &&
3052 preimage.nr + applied_pos >= img->nr &&
3053 (ws_rule & WS_BLANK_AT_EOF) &&
3054 state->ws_error_action != nowarn_ws_error) {
3055 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3056 found_new_blank_lines_at_end);
3057 if (state->ws_error_action == correct_ws_error) {
3058 while (new_blank_lines_at_end--)
3059 remove_last_line(&postimage);
3062 * We would want to prevent write_out_results()
3063 * from taking place in apply_patch() that follows
3064 * the callchain led us here, which is:
3065 * apply_patch->check_patch_list->check_patch->
3066 * apply_data->apply_fragments->apply_one_fragment
3068 if (state->ws_error_action == die_on_ws_error)
3069 state->apply = 0;
3072 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3073 int offset = applied_pos - pos;
3074 if (state->apply_in_reverse)
3075 offset = 0 - offset;
3076 fprintf_ln(stderr,
3077 Q_("Hunk #%d succeeded at %d (offset %d line).",
3078 "Hunk #%d succeeded at %d (offset %d lines).",
3079 offset),
3080 nth_fragment, applied_pos + 1, offset);
3084 * Warn if it was necessary to reduce the number
3085 * of context lines.
3087 if ((leading != frag->leading ||
3088 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3089 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3090 " to apply fragment at %d"),
3091 leading, trailing, applied_pos+1);
3092 update_image(state, img, applied_pos, &preimage, &postimage);
3093 } else {
3094 if (state->apply_verbosity > verbosity_normal)
3095 error(_("while searching for:\n%.*s"),
3096 (int)(old - oldlines), oldlines);
3099 out:
3100 free(oldlines);
3101 strbuf_release(&newlines);
3102 free(preimage.line_allocated);
3103 free(postimage.line_allocated);
3105 return (applied_pos < 0);
3108 static int apply_binary_fragment(struct apply_state *state,
3109 struct image *img,
3110 struct patch *patch)
3112 struct fragment *fragment = patch->fragments;
3113 unsigned long len;
3114 void *dst;
3116 if (!fragment)
3117 return error(_("missing binary patch data for '%s'"),
3118 patch->new_name ?
3119 patch->new_name :
3120 patch->old_name);
3122 /* Binary patch is irreversible without the optional second hunk */
3123 if (state->apply_in_reverse) {
3124 if (!fragment->next)
3125 return error(_("cannot reverse-apply a binary patch "
3126 "without the reverse hunk to '%s'"),
3127 patch->new_name
3128 ? patch->new_name : patch->old_name);
3129 fragment = fragment->next;
3131 switch (fragment->binary_patch_method) {
3132 case BINARY_DELTA_DEFLATED:
3133 dst = patch_delta(img->buf, img->len, fragment->patch,
3134 fragment->size, &len);
3135 if (!dst)
3136 return -1;
3137 clear_image(img);
3138 img->buf = dst;
3139 img->len = len;
3140 return 0;
3141 case BINARY_LITERAL_DEFLATED:
3142 clear_image(img);
3143 img->len = fragment->size;
3144 img->buf = xmemdupz(fragment->patch, img->len);
3145 return 0;
3147 return -1;
3151 * Replace "img" with the result of applying the binary patch.
3152 * The binary patch data itself in patch->fragment is still kept
3153 * but the preimage prepared by the caller in "img" is freed here
3154 * or in the helper function apply_binary_fragment() this calls.
3156 static int apply_binary(struct apply_state *state,
3157 struct image *img,
3158 struct patch *patch)
3160 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3161 struct object_id oid;
3164 * For safety, we require patch index line to contain
3165 * full 40-byte textual SHA1 for old and new, at least for now.
3167 if (strlen(patch->old_sha1_prefix) != 40 ||
3168 strlen(patch->new_sha1_prefix) != 40 ||
3169 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3170 get_oid_hex(patch->new_sha1_prefix, &oid))
3171 return error(_("cannot apply binary patch to '%s' "
3172 "without full index line"), name);
3174 if (patch->old_name) {
3176 * See if the old one matches what the patch
3177 * applies to.
3179 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3180 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3181 return error(_("the patch applies to '%s' (%s), "
3182 "which does not match the "
3183 "current contents."),
3184 name, oid_to_hex(&oid));
3186 else {
3187 /* Otherwise, the old one must be empty. */
3188 if (img->len)
3189 return error(_("the patch applies to an empty "
3190 "'%s' but it is not empty"), name);
3193 get_oid_hex(patch->new_sha1_prefix, &oid);
3194 if (is_null_oid(&oid)) {
3195 clear_image(img);
3196 return 0; /* deletion patch */
3199 if (has_sha1_file(oid.hash)) {
3200 /* We already have the postimage */
3201 enum object_type type;
3202 unsigned long size;
3203 char *result;
3205 result = read_sha1_file(oid.hash, &type, &size);
3206 if (!result)
3207 return error(_("the necessary postimage %s for "
3208 "'%s' cannot be read"),
3209 patch->new_sha1_prefix, name);
3210 clear_image(img);
3211 img->buf = result;
3212 img->len = size;
3213 } else {
3215 * We have verified buf matches the preimage;
3216 * apply the patch data to it, which is stored
3217 * in the patch->fragments->{patch,size}.
3219 if (apply_binary_fragment(state, img, patch))
3220 return error(_("binary patch does not apply to '%s'"),
3221 name);
3223 /* verify that the result matches */
3224 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3225 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3226 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3227 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3230 return 0;
3233 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3235 struct fragment *frag = patch->fragments;
3236 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3237 unsigned ws_rule = patch->ws_rule;
3238 unsigned inaccurate_eof = patch->inaccurate_eof;
3239 int nth = 0;
3241 if (patch->is_binary)
3242 return apply_binary(state, img, patch);
3244 while (frag) {
3245 nth++;
3246 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3247 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3248 if (!state->apply_with_reject)
3249 return -1;
3250 frag->rejected = 1;
3252 frag = frag->next;
3254 return 0;
3257 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3259 if (S_ISGITLINK(mode)) {
3260 strbuf_grow(buf, 100);
3261 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3262 } else {
3263 enum object_type type;
3264 unsigned long sz;
3265 char *result;
3267 result = read_sha1_file(oid->hash, &type, &sz);
3268 if (!result)
3269 return -1;
3270 /* XXX read_sha1_file NUL-terminates */
3271 strbuf_attach(buf, result, sz, sz + 1);
3273 return 0;
3276 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3278 if (!ce)
3279 return 0;
3280 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3283 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3285 struct string_list_item *item;
3287 if (name == NULL)
3288 return NULL;
3290 item = string_list_lookup(&state->fn_table, name);
3291 if (item != NULL)
3292 return (struct patch *)item->util;
3294 return NULL;
3298 * item->util in the filename table records the status of the path.
3299 * Usually it points at a patch (whose result records the contents
3300 * of it after applying it), but it could be PATH_WAS_DELETED for a
3301 * path that a previously applied patch has already removed, or
3302 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3304 * The latter is needed to deal with a case where two paths A and B
3305 * are swapped by first renaming A to B and then renaming B to A;
3306 * moving A to B should not be prevented due to presence of B as we
3307 * will remove it in a later patch.
3309 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3310 #define PATH_WAS_DELETED ((struct patch *) -1)
3312 static int to_be_deleted(struct patch *patch)
3314 return patch == PATH_TO_BE_DELETED;
3317 static int was_deleted(struct patch *patch)
3319 return patch == PATH_WAS_DELETED;
3322 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3324 struct string_list_item *item;
3327 * Always add new_name unless patch is a deletion
3328 * This should cover the cases for normal diffs,
3329 * file creations and copies
3331 if (patch->new_name != NULL) {
3332 item = string_list_insert(&state->fn_table, patch->new_name);
3333 item->util = patch;
3337 * store a failure on rename/deletion cases because
3338 * later chunks shouldn't patch old names
3340 if ((patch->new_name == NULL) || (patch->is_rename)) {
3341 item = string_list_insert(&state->fn_table, patch->old_name);
3342 item->util = PATH_WAS_DELETED;
3346 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3349 * store information about incoming file deletion
3351 while (patch) {
3352 if ((patch->new_name == NULL) || (patch->is_rename)) {
3353 struct string_list_item *item;
3354 item = string_list_insert(&state->fn_table, patch->old_name);
3355 item->util = PATH_TO_BE_DELETED;
3357 patch = patch->next;
3361 static int checkout_target(struct index_state *istate,
3362 struct cache_entry *ce, struct stat *st)
3364 struct checkout costate = CHECKOUT_INIT;
3366 costate.refresh_cache = 1;
3367 costate.istate = istate;
3368 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3369 return error(_("cannot checkout %s"), ce->name);
3370 return 0;
3373 static struct patch *previous_patch(struct apply_state *state,
3374 struct patch *patch,
3375 int *gone)
3377 struct patch *previous;
3379 *gone = 0;
3380 if (patch->is_copy || patch->is_rename)
3381 return NULL; /* "git" patches do not depend on the order */
3383 previous = in_fn_table(state, patch->old_name);
3384 if (!previous)
3385 return NULL;
3387 if (to_be_deleted(previous))
3388 return NULL; /* the deletion hasn't happened yet */
3390 if (was_deleted(previous))
3391 *gone = 1;
3393 return previous;
3396 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3398 if (S_ISGITLINK(ce->ce_mode)) {
3399 if (!S_ISDIR(st->st_mode))
3400 return -1;
3401 return 0;
3403 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3406 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3408 static int load_patch_target(struct apply_state *state,
3409 struct strbuf *buf,
3410 const struct cache_entry *ce,
3411 struct stat *st,
3412 struct patch *patch,
3413 const char *name,
3414 unsigned expected_mode)
3416 if (state->cached || state->check_index) {
3417 if (read_file_or_gitlink(ce, buf))
3418 return error(_("failed to read %s"), name);
3419 } else if (name) {
3420 if (S_ISGITLINK(expected_mode)) {
3421 if (ce)
3422 return read_file_or_gitlink(ce, buf);
3423 else
3424 return SUBMODULE_PATCH_WITHOUT_INDEX;
3425 } else if (has_symlink_leading_path(name, strlen(name))) {
3426 return error(_("reading from '%s' beyond a symbolic link"), name);
3427 } else {
3428 if (read_old_data(st, patch, name, buf))
3429 return error(_("failed to read %s"), name);
3432 return 0;
3436 * We are about to apply "patch"; populate the "image" with the
3437 * current version we have, from the working tree or from the index,
3438 * depending on the situation e.g. --cached/--index. If we are
3439 * applying a non-git patch that incrementally updates the tree,
3440 * we read from the result of a previous diff.
3442 static int load_preimage(struct apply_state *state,
3443 struct image *image,
3444 struct patch *patch, struct stat *st,
3445 const struct cache_entry *ce)
3447 struct strbuf buf = STRBUF_INIT;
3448 size_t len;
3449 char *img;
3450 struct patch *previous;
3451 int status;
3453 previous = previous_patch(state, patch, &status);
3454 if (status)
3455 return error(_("path %s has been renamed/deleted"),
3456 patch->old_name);
3457 if (previous) {
3458 /* We have a patched copy in memory; use that. */
3459 strbuf_add(&buf, previous->result, previous->resultsize);
3460 } else {
3461 status = load_patch_target(state, &buf, ce, st, patch,
3462 patch->old_name, patch->old_mode);
3463 if (status < 0)
3464 return status;
3465 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3467 * There is no way to apply subproject
3468 * patch without looking at the index.
3469 * NEEDSWORK: shouldn't this be flagged
3470 * as an error???
3472 free_fragment_list(patch->fragments);
3473 patch->fragments = NULL;
3474 } else if (status) {
3475 return error(_("failed to read %s"), patch->old_name);
3479 img = strbuf_detach(&buf, &len);
3480 prepare_image(image, img, len, !patch->is_binary);
3481 return 0;
3484 static int three_way_merge(struct image *image,
3485 char *path,
3486 const struct object_id *base,
3487 const struct object_id *ours,
3488 const struct object_id *theirs)
3490 mmfile_t base_file, our_file, their_file;
3491 mmbuffer_t result = { NULL };
3492 int status;
3494 read_mmblob(&base_file, base);
3495 read_mmblob(&our_file, ours);
3496 read_mmblob(&their_file, theirs);
3497 status = ll_merge(&result, path,
3498 &base_file, "base",
3499 &our_file, "ours",
3500 &their_file, "theirs", NULL);
3501 free(base_file.ptr);
3502 free(our_file.ptr);
3503 free(their_file.ptr);
3504 if (status < 0 || !result.ptr) {
3505 free(result.ptr);
3506 return -1;
3508 clear_image(image);
3509 image->buf = result.ptr;
3510 image->len = result.size;
3512 return status;
3516 * When directly falling back to add/add three-way merge, we read from
3517 * the current contents of the new_name. In no cases other than that
3518 * this function will be called.
3520 static int load_current(struct apply_state *state,
3521 struct image *image,
3522 struct patch *patch)
3524 struct strbuf buf = STRBUF_INIT;
3525 int status, pos;
3526 size_t len;
3527 char *img;
3528 struct stat st;
3529 struct cache_entry *ce;
3530 char *name = patch->new_name;
3531 unsigned mode = patch->new_mode;
3533 if (!patch->is_new)
3534 die("BUG: patch to %s is not a creation", patch->old_name);
3536 pos = cache_name_pos(name, strlen(name));
3537 if (pos < 0)
3538 return error(_("%s: does not exist in index"), name);
3539 ce = active_cache[pos];
3540 if (lstat(name, &st)) {
3541 if (errno != ENOENT)
3542 return error_errno("%s", name);
3543 if (checkout_target(&the_index, ce, &st))
3544 return -1;
3546 if (verify_index_match(ce, &st))
3547 return error(_("%s: does not match index"), name);
3549 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3550 if (status < 0)
3551 return status;
3552 else if (status)
3553 return -1;
3554 img = strbuf_detach(&buf, &len);
3555 prepare_image(image, img, len, !patch->is_binary);
3556 return 0;
3559 static int try_threeway(struct apply_state *state,
3560 struct image *image,
3561 struct patch *patch,
3562 struct stat *st,
3563 const struct cache_entry *ce)
3565 struct object_id pre_oid, post_oid, our_oid;
3566 struct strbuf buf = STRBUF_INIT;
3567 size_t len;
3568 int status;
3569 char *img;
3570 struct image tmp_image;
3572 /* No point falling back to 3-way merge in these cases */
3573 if (patch->is_delete ||
3574 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3575 return -1;
3577 /* Preimage the patch was prepared for */
3578 if (patch->is_new)
3579 write_sha1_file("", 0, blob_type, pre_oid.hash);
3580 else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3581 read_blob_object(&buf, &pre_oid, patch->old_mode))
3582 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3584 if (state->apply_verbosity > verbosity_silent)
3585 fprintf(stderr, _("Falling back to three-way merge...\n"));
3587 img = strbuf_detach(&buf, &len);
3588 prepare_image(&tmp_image, img, len, 1);
3589 /* Apply the patch to get the post image */
3590 if (apply_fragments(state, &tmp_image, patch) < 0) {
3591 clear_image(&tmp_image);
3592 return -1;
3594 /* post_oid is theirs */
3595 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3596 clear_image(&tmp_image);
3598 /* our_oid is ours */
3599 if (patch->is_new) {
3600 if (load_current(state, &tmp_image, patch))
3601 return error(_("cannot read the current contents of '%s'"),
3602 patch->new_name);
3603 } else {
3604 if (load_preimage(state, &tmp_image, patch, st, ce))
3605 return error(_("cannot read the current contents of '%s'"),
3606 patch->old_name);
3608 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3609 clear_image(&tmp_image);
3611 /* in-core three-way merge between post and our using pre as base */
3612 status = three_way_merge(image, patch->new_name,
3613 &pre_oid, &our_oid, &post_oid);
3614 if (status < 0) {
3615 if (state->apply_verbosity > verbosity_silent)
3616 fprintf(stderr,
3617 _("Failed to fall back on three-way merge...\n"));
3618 return status;
3621 if (status) {
3622 patch->conflicted_threeway = 1;
3623 if (patch->is_new)
3624 oidclr(&patch->threeway_stage[0]);
3625 else
3626 oidcpy(&patch->threeway_stage[0], &pre_oid);
3627 oidcpy(&patch->threeway_stage[1], &our_oid);
3628 oidcpy(&patch->threeway_stage[2], &post_oid);
3629 if (state->apply_verbosity > verbosity_silent)
3630 fprintf(stderr,
3631 _("Applied patch to '%s' with conflicts.\n"),
3632 patch->new_name);
3633 } else {
3634 if (state->apply_verbosity > verbosity_silent)
3635 fprintf(stderr,
3636 _("Applied patch to '%s' cleanly.\n"),
3637 patch->new_name);
3639 return 0;
3642 static int apply_data(struct apply_state *state, struct patch *patch,
3643 struct stat *st, const struct cache_entry *ce)
3645 struct image image;
3647 if (load_preimage(state, &image, patch, st, ce) < 0)
3648 return -1;
3650 if (patch->direct_to_threeway ||
3651 apply_fragments(state, &image, patch) < 0) {
3652 /* Note: with --reject, apply_fragments() returns 0 */
3653 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3654 return -1;
3656 patch->result = image.buf;
3657 patch->resultsize = image.len;
3658 add_to_fn_table(state, patch);
3659 free(image.line_allocated);
3661 if (0 < patch->is_delete && patch->resultsize)
3662 return error(_("removal patch leaves file contents"));
3664 return 0;
3668 * If "patch" that we are looking at modifies or deletes what we have,
3669 * we would want it not to lose any local modification we have, either
3670 * in the working tree or in the index.
3672 * This also decides if a non-git patch is a creation patch or a
3673 * modification to an existing empty file. We do not check the state
3674 * of the current tree for a creation patch in this function; the caller
3675 * check_patch() separately makes sure (and errors out otherwise) that
3676 * the path the patch creates does not exist in the current tree.
3678 static int check_preimage(struct apply_state *state,
3679 struct patch *patch,
3680 struct cache_entry **ce,
3681 struct stat *st)
3683 const char *old_name = patch->old_name;
3684 struct patch *previous = NULL;
3685 int stat_ret = 0, status;
3686 unsigned st_mode = 0;
3688 if (!old_name)
3689 return 0;
3691 assert(patch->is_new <= 0);
3692 previous = previous_patch(state, patch, &status);
3694 if (status)
3695 return error(_("path %s has been renamed/deleted"), old_name);
3696 if (previous) {
3697 st_mode = previous->new_mode;
3698 } else if (!state->cached) {
3699 stat_ret = lstat(old_name, st);
3700 if (stat_ret && errno != ENOENT)
3701 return error_errno("%s", old_name);
3704 if (state->check_index && !previous) {
3705 int pos = cache_name_pos(old_name, strlen(old_name));
3706 if (pos < 0) {
3707 if (patch->is_new < 0)
3708 goto is_new;
3709 return error(_("%s: does not exist in index"), old_name);
3711 *ce = active_cache[pos];
3712 if (stat_ret < 0) {
3713 if (checkout_target(&the_index, *ce, st))
3714 return -1;
3716 if (!state->cached && verify_index_match(*ce, st))
3717 return error(_("%s: does not match index"), old_name);
3718 if (state->cached)
3719 st_mode = (*ce)->ce_mode;
3720 } else if (stat_ret < 0) {
3721 if (patch->is_new < 0)
3722 goto is_new;
3723 return error_errno("%s", old_name);
3726 if (!state->cached && !previous)
3727 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3729 if (patch->is_new < 0)
3730 patch->is_new = 0;
3731 if (!patch->old_mode)
3732 patch->old_mode = st_mode;
3733 if ((st_mode ^ patch->old_mode) & S_IFMT)
3734 return error(_("%s: wrong type"), old_name);
3735 if (st_mode != patch->old_mode)
3736 warning(_("%s has type %o, expected %o"),
3737 old_name, st_mode, patch->old_mode);
3738 if (!patch->new_mode && !patch->is_delete)
3739 patch->new_mode = st_mode;
3740 return 0;
3742 is_new:
3743 patch->is_new = 1;
3744 patch->is_delete = 0;
3745 FREE_AND_NULL(patch->old_name);
3746 return 0;
3750 #define EXISTS_IN_INDEX 1
3751 #define EXISTS_IN_WORKTREE 2
3753 static int check_to_create(struct apply_state *state,
3754 const char *new_name,
3755 int ok_if_exists)
3757 struct stat nst;
3759 if (state->check_index &&
3760 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3761 !ok_if_exists)
3762 return EXISTS_IN_INDEX;
3763 if (state->cached)
3764 return 0;
3766 if (!lstat(new_name, &nst)) {
3767 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3768 return 0;
3770 * A leading component of new_name might be a symlink
3771 * that is going to be removed with this patch, but
3772 * still pointing at somewhere that has the path.
3773 * In such a case, path "new_name" does not exist as
3774 * far as git is concerned.
3776 if (has_symlink_leading_path(new_name, strlen(new_name)))
3777 return 0;
3779 return EXISTS_IN_WORKTREE;
3780 } else if (!is_missing_file_error(errno)) {
3781 return error_errno("%s", new_name);
3783 return 0;
3786 static uintptr_t register_symlink_changes(struct apply_state *state,
3787 const char *path,
3788 uintptr_t what)
3790 struct string_list_item *ent;
3792 ent = string_list_lookup(&state->symlink_changes, path);
3793 if (!ent) {
3794 ent = string_list_insert(&state->symlink_changes, path);
3795 ent->util = (void *)0;
3797 ent->util = (void *)(what | ((uintptr_t)ent->util));
3798 return (uintptr_t)ent->util;
3801 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3803 struct string_list_item *ent;
3805 ent = string_list_lookup(&state->symlink_changes, path);
3806 if (!ent)
3807 return 0;
3808 return (uintptr_t)ent->util;
3811 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3813 for ( ; patch; patch = patch->next) {
3814 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3815 (patch->is_rename || patch->is_delete))
3816 /* the symlink at patch->old_name is removed */
3817 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3819 if (patch->new_name && S_ISLNK(patch->new_mode))
3820 /* the symlink at patch->new_name is created or remains */
3821 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3825 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3827 do {
3828 unsigned int change;
3830 while (--name->len && name->buf[name->len] != '/')
3831 ; /* scan backwards */
3832 if (!name->len)
3833 break;
3834 name->buf[name->len] = '\0';
3835 change = check_symlink_changes(state, name->buf);
3836 if (change & APPLY_SYMLINK_IN_RESULT)
3837 return 1;
3838 if (change & APPLY_SYMLINK_GOES_AWAY)
3840 * This cannot be "return 0", because we may
3841 * see a new one created at a higher level.
3843 continue;
3845 /* otherwise, check the preimage */
3846 if (state->check_index) {
3847 struct cache_entry *ce;
3849 ce = cache_file_exists(name->buf, name->len, ignore_case);
3850 if (ce && S_ISLNK(ce->ce_mode))
3851 return 1;
3852 } else {
3853 struct stat st;
3854 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3855 return 1;
3857 } while (1);
3858 return 0;
3861 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3863 int ret;
3864 struct strbuf name = STRBUF_INIT;
3866 assert(*name_ != '\0');
3867 strbuf_addstr(&name, name_);
3868 ret = path_is_beyond_symlink_1(state, &name);
3869 strbuf_release(&name);
3871 return ret;
3874 static int check_unsafe_path(struct patch *patch)
3876 const char *old_name = NULL;
3877 const char *new_name = NULL;
3878 if (patch->is_delete)
3879 old_name = patch->old_name;
3880 else if (!patch->is_new && !patch->is_copy)
3881 old_name = patch->old_name;
3882 if (!patch->is_delete)
3883 new_name = patch->new_name;
3885 if (old_name && !verify_path(old_name))
3886 return error(_("invalid path '%s'"), old_name);
3887 if (new_name && !verify_path(new_name))
3888 return error(_("invalid path '%s'"), new_name);
3889 return 0;
3893 * Check and apply the patch in-core; leave the result in patch->result
3894 * for the caller to write it out to the final destination.
3896 static int check_patch(struct apply_state *state, struct patch *patch)
3898 struct stat st;
3899 const char *old_name = patch->old_name;
3900 const char *new_name = patch->new_name;
3901 const char *name = old_name ? old_name : new_name;
3902 struct cache_entry *ce = NULL;
3903 struct patch *tpatch;
3904 int ok_if_exists;
3905 int status;
3907 patch->rejected = 1; /* we will drop this after we succeed */
3909 status = check_preimage(state, patch, &ce, &st);
3910 if (status)
3911 return status;
3912 old_name = patch->old_name;
3915 * A type-change diff is always split into a patch to delete
3916 * old, immediately followed by a patch to create new (see
3917 * diff.c::run_diff()); in such a case it is Ok that the entry
3918 * to be deleted by the previous patch is still in the working
3919 * tree and in the index.
3921 * A patch to swap-rename between A and B would first rename A
3922 * to B and then rename B to A. While applying the first one,
3923 * the presence of B should not stop A from getting renamed to
3924 * B; ask to_be_deleted() about the later rename. Removal of
3925 * B and rename from A to B is handled the same way by asking
3926 * was_deleted().
3928 if ((tpatch = in_fn_table(state, new_name)) &&
3929 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3930 ok_if_exists = 1;
3931 else
3932 ok_if_exists = 0;
3934 if (new_name &&
3935 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3936 int err = check_to_create(state, new_name, ok_if_exists);
3938 if (err && state->threeway) {
3939 patch->direct_to_threeway = 1;
3940 } else switch (err) {
3941 case 0:
3942 break; /* happy */
3943 case EXISTS_IN_INDEX:
3944 return error(_("%s: already exists in index"), new_name);
3945 break;
3946 case EXISTS_IN_WORKTREE:
3947 return error(_("%s: already exists in working directory"),
3948 new_name);
3949 default:
3950 return err;
3953 if (!patch->new_mode) {
3954 if (0 < patch->is_new)
3955 patch->new_mode = S_IFREG | 0644;
3956 else
3957 patch->new_mode = patch->old_mode;
3961 if (new_name && old_name) {
3962 int same = !strcmp(old_name, new_name);
3963 if (!patch->new_mode)
3964 patch->new_mode = patch->old_mode;
3965 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3966 if (same)
3967 return error(_("new mode (%o) of %s does not "
3968 "match old mode (%o)"),
3969 patch->new_mode, new_name,
3970 patch->old_mode);
3971 else
3972 return error(_("new mode (%o) of %s does not "
3973 "match old mode (%o) of %s"),
3974 patch->new_mode, new_name,
3975 patch->old_mode, old_name);
3979 if (!state->unsafe_paths && check_unsafe_path(patch))
3980 return -128;
3983 * An attempt to read from or delete a path that is beyond a
3984 * symbolic link will be prevented by load_patch_target() that
3985 * is called at the beginning of apply_data() so we do not
3986 * have to worry about a patch marked with "is_delete" bit
3987 * here. We however need to make sure that the patch result
3988 * is not deposited to a path that is beyond a symbolic link