Merge branch 'jt/http-redact-cookies' into maint
[git.git] / apply.c
blob321a9fa68d491f7e5e89dc9e398b067f67a10c77
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)
80 memset(state, 0, sizeof(*state));
81 state->prefix = prefix;
82 state->apply = 1;
83 state->line_termination = '\n';
84 state->p_value = 1;
85 state->p_context = UINT_MAX;
86 state->squelch_whitespace_errors = 5;
87 state->ws_error_action = warn_on_ws_error;
88 state->ws_ignore_action = ignore_ws_none;
89 state->linenr = 1;
90 string_list_init(&state->fn_table, 0);
91 string_list_init(&state->limit_by_name, 0);
92 string_list_init(&state->symlink_changes, 0);
93 strbuf_init(&state->root, 0);
95 git_apply_config();
96 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
97 return -1;
98 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
99 return -1;
100 return 0;
103 void clear_apply_state(struct apply_state *state)
105 string_list_clear(&state->limit_by_name, 0);
106 string_list_clear(&state->symlink_changes, 0);
107 strbuf_release(&state->root);
109 /* &state->fn_table is cleared at the end of apply_patch() */
112 static void mute_routine(const char *msg, va_list params)
114 /* do nothing */
117 int check_apply_state(struct apply_state *state, int force_apply)
119 int is_not_gitdir = !startup_info->have_repository;
121 if (state->apply_with_reject && state->threeway)
122 return error(_("--reject and --3way cannot be used together."));
123 if (state->cached && state->threeway)
124 return error(_("--cached and --3way cannot be used together."));
125 if (state->threeway) {
126 if (is_not_gitdir)
127 return error(_("--3way outside a repository"));
128 state->check_index = 1;
130 if (state->apply_with_reject) {
131 state->apply = 1;
132 if (state->apply_verbosity == verbosity_normal)
133 state->apply_verbosity = verbosity_verbose;
135 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
136 state->apply = 0;
137 if (state->check_index && is_not_gitdir)
138 return error(_("--index outside a repository"));
139 if (state->cached) {
140 if (is_not_gitdir)
141 return error(_("--cached outside a repository"));
142 state->check_index = 1;
144 if (state->check_index)
145 state->unsafe_paths = 0;
147 if (state->apply_verbosity <= verbosity_silent) {
148 state->saved_error_routine = get_error_routine();
149 state->saved_warn_routine = get_warn_routine();
150 set_error_routine(mute_routine);
151 set_warn_routine(mute_routine);
154 return 0;
157 static void set_default_whitespace_mode(struct apply_state *state)
159 if (!state->whitespace_option && !apply_default_whitespace)
160 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
164 * This represents one "hunk" from a patch, starting with
165 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
166 * patch text is pointed at by patch, and its byte length
167 * is stored in size. leading and trailing are the number
168 * of context lines.
170 struct fragment {
171 unsigned long leading, trailing;
172 unsigned long oldpos, oldlines;
173 unsigned long newpos, newlines;
175 * 'patch' is usually borrowed from buf in apply_patch(),
176 * but some codepaths store an allocated buffer.
178 const char *patch;
179 unsigned free_patch:1,
180 rejected:1;
181 int size;
182 int linenr;
183 struct fragment *next;
187 * When dealing with a binary patch, we reuse "leading" field
188 * to store the type of the binary hunk, either deflated "delta"
189 * or deflated "literal".
191 #define binary_patch_method leading
192 #define BINARY_DELTA_DEFLATED 1
193 #define BINARY_LITERAL_DEFLATED 2
196 * This represents a "patch" to a file, both metainfo changes
197 * such as creation/deletion, filemode and content changes represented
198 * as a series of fragments.
200 struct patch {
201 char *new_name, *old_name, *def_name;
202 unsigned int old_mode, new_mode;
203 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
204 int rejected;
205 unsigned ws_rule;
206 int lines_added, lines_deleted;
207 int score;
208 int extension_linenr; /* first line specifying delete/new/rename/copy */
209 unsigned int is_toplevel_relative:1;
210 unsigned int inaccurate_eof:1;
211 unsigned int is_binary:1;
212 unsigned int is_copy:1;
213 unsigned int is_rename:1;
214 unsigned int recount:1;
215 unsigned int conflicted_threeway:1;
216 unsigned int direct_to_threeway:1;
217 unsigned int crlf_in_old:1;
218 struct fragment *fragments;
219 char *result;
220 size_t resultsize;
221 char old_sha1_prefix[41];
222 char new_sha1_prefix[41];
223 struct patch *next;
225 /* three-way fallback result */
226 struct object_id threeway_stage[3];
229 static void free_fragment_list(struct fragment *list)
231 while (list) {
232 struct fragment *next = list->next;
233 if (list->free_patch)
234 free((char *)list->patch);
235 free(list);
236 list = next;
240 static void free_patch(struct patch *patch)
242 free_fragment_list(patch->fragments);
243 free(patch->def_name);
244 free(patch->old_name);
245 free(patch->new_name);
246 free(patch->result);
247 free(patch);
250 static void free_patch_list(struct patch *list)
252 while (list) {
253 struct patch *next = list->next;
254 free_patch(list);
255 list = next;
260 * A line in a file, len-bytes long (includes the terminating LF,
261 * except for an incomplete line at the end if the file ends with
262 * one), and its contents hashes to 'hash'.
264 struct line {
265 size_t len;
266 unsigned hash : 24;
267 unsigned flag : 8;
268 #define LINE_COMMON 1
269 #define LINE_PATCHED 2
273 * This represents a "file", which is an array of "lines".
275 struct image {
276 char *buf;
277 size_t len;
278 size_t nr;
279 size_t alloc;
280 struct line *line_allocated;
281 struct line *line;
284 static uint32_t hash_line(const char *cp, size_t len)
286 size_t i;
287 uint32_t h;
288 for (i = 0, h = 0; i < len; i++) {
289 if (!isspace(cp[i])) {
290 h = h * 3 + (cp[i] & 0xff);
293 return h;
297 * Compare lines s1 of length n1 and s2 of length n2, ignoring
298 * whitespace difference. Returns 1 if they match, 0 otherwise
300 static int fuzzy_matchlines(const char *s1, size_t n1,
301 const char *s2, size_t n2)
303 const char *end1 = s1 + n1;
304 const char *end2 = s2 + n2;
306 /* ignore line endings */
307 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
308 end1--;
309 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
310 end2--;
312 while (s1 < end1 && s2 < end2) {
313 if (isspace(*s1)) {
315 * Skip whitespace. We check on both buffers
316 * because we don't want "a b" to match "ab".
318 if (!isspace(*s2))
319 return 0;
320 while (s1 < end1 && isspace(*s1))
321 s1++;
322 while (s2 < end2 && isspace(*s2))
323 s2++;
324 } else if (*s1++ != *s2++)
325 return 0;
328 /* If we reached the end on one side only, lines don't match. */
329 return s1 == end1 && s2 == end2;
332 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
334 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
335 img->line_allocated[img->nr].len = len;
336 img->line_allocated[img->nr].hash = hash_line(bol, len);
337 img->line_allocated[img->nr].flag = flag;
338 img->nr++;
342 * "buf" has the file contents to be patched (read from various sources).
343 * attach it to "image" and add line-based index to it.
344 * "image" now owns the "buf".
346 static void prepare_image(struct image *image, char *buf, size_t len,
347 int prepare_linetable)
349 const char *cp, *ep;
351 memset(image, 0, sizeof(*image));
352 image->buf = buf;
353 image->len = len;
355 if (!prepare_linetable)
356 return;
358 ep = image->buf + image->len;
359 cp = image->buf;
360 while (cp < ep) {
361 const char *next;
362 for (next = cp; next < ep && *next != '\n'; next++)
364 if (next < ep)
365 next++;
366 add_line_info(image, cp, next - cp, 0);
367 cp = next;
369 image->line = image->line_allocated;
372 static void clear_image(struct image *image)
374 free(image->buf);
375 free(image->line_allocated);
376 memset(image, 0, sizeof(*image));
379 /* fmt must contain _one_ %s and no other substitution */
380 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
382 struct strbuf sb = STRBUF_INIT;
384 if (patch->old_name && patch->new_name &&
385 strcmp(patch->old_name, patch->new_name)) {
386 quote_c_style(patch->old_name, &sb, NULL, 0);
387 strbuf_addstr(&sb, " => ");
388 quote_c_style(patch->new_name, &sb, NULL, 0);
389 } else {
390 const char *n = patch->new_name;
391 if (!n)
392 n = patch->old_name;
393 quote_c_style(n, &sb, NULL, 0);
395 fprintf(output, fmt, sb.buf);
396 fputc('\n', output);
397 strbuf_release(&sb);
400 #define SLOP (16)
402 static int read_patch_file(struct strbuf *sb, int fd)
404 if (strbuf_read(sb, fd, 0) < 0)
405 return error_errno("git apply: failed to read");
408 * Make sure that we have some slop in the buffer
409 * so that we can do speculative "memcmp" etc, and
410 * see to it that it is NUL-filled.
412 strbuf_grow(sb, SLOP);
413 memset(sb->buf + sb->len, 0, SLOP);
414 return 0;
417 static unsigned long linelen(const char *buffer, unsigned long size)
419 unsigned long len = 0;
420 while (size--) {
421 len++;
422 if (*buffer++ == '\n')
423 break;
425 return len;
428 static int is_dev_null(const char *str)
430 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
433 #define TERM_SPACE 1
434 #define TERM_TAB 2
436 static int name_terminate(int c, int terminate)
438 if (c == ' ' && !(terminate & TERM_SPACE))
439 return 0;
440 if (c == '\t' && !(terminate & TERM_TAB))
441 return 0;
443 return 1;
446 /* remove double slashes to make --index work with such filenames */
447 static char *squash_slash(char *name)
449 int i = 0, j = 0;
451 if (!name)
452 return NULL;
454 while (name[i]) {
455 if ((name[j++] = name[i++]) == '/')
456 while (name[i] == '/')
457 i++;
459 name[j] = '\0';
460 return name;
463 static char *find_name_gnu(struct apply_state *state,
464 const char *line,
465 const char *def,
466 int p_value)
468 struct strbuf name = STRBUF_INIT;
469 char *cp;
472 * Proposed "new-style" GNU patch/diff format; see
473 * http://marc.info/?l=git&m=112927316408690&w=2
475 if (unquote_c_style(&name, line, NULL)) {
476 strbuf_release(&name);
477 return NULL;
480 for (cp = name.buf; p_value; p_value--) {
481 cp = strchr(cp, '/');
482 if (!cp) {
483 strbuf_release(&name);
484 return NULL;
486 cp++;
489 strbuf_remove(&name, 0, cp - name.buf);
490 if (state->root.len)
491 strbuf_insert(&name, 0, state->root.buf, state->root.len);
492 return squash_slash(strbuf_detach(&name, NULL));
495 static size_t sane_tz_len(const char *line, size_t len)
497 const char *tz, *p;
499 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
500 return 0;
501 tz = line + len - strlen(" +0500");
503 if (tz[1] != '+' && tz[1] != '-')
504 return 0;
506 for (p = tz + 2; p != line + len; p++)
507 if (!isdigit(*p))
508 return 0;
510 return line + len - tz;
513 static size_t tz_with_colon_len(const char *line, size_t len)
515 const char *tz, *p;
517 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
518 return 0;
519 tz = line + len - strlen(" +08:00");
521 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
522 return 0;
523 p = tz + 2;
524 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
525 !isdigit(*p++) || !isdigit(*p++))
526 return 0;
528 return line + len - tz;
531 static size_t date_len(const char *line, size_t len)
533 const char *date, *p;
535 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
536 return 0;
537 p = date = line + len - strlen("72-02-05");
539 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
540 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
541 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
542 return 0;
544 if (date - line >= strlen("19") &&
545 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
546 date -= strlen("19");
548 return line + len - date;
551 static size_t short_time_len(const char *line, size_t len)
553 const char *time, *p;
555 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
556 return 0;
557 p = time = line + len - strlen(" 07:01:32");
559 /* Permit 1-digit hours? */
560 if (*p++ != ' ' ||
561 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
562 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
563 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
564 return 0;
566 return line + len - time;
569 static size_t fractional_time_len(const char *line, size_t len)
571 const char *p;
572 size_t n;
574 /* Expected format: 19:41:17.620000023 */
575 if (!len || !isdigit(line[len - 1]))
576 return 0;
577 p = line + len - 1;
579 /* Fractional seconds. */
580 while (p > line && isdigit(*p))
581 p--;
582 if (*p != '.')
583 return 0;
585 /* Hours, minutes, and whole seconds. */
586 n = short_time_len(line, p - line);
587 if (!n)
588 return 0;
590 return line + len - p + n;
593 static size_t trailing_spaces_len(const char *line, size_t len)
595 const char *p;
597 /* Expected format: ' ' x (1 or more) */
598 if (!len || line[len - 1] != ' ')
599 return 0;
601 p = line + len;
602 while (p != line) {
603 p--;
604 if (*p != ' ')
605 return line + len - (p + 1);
608 /* All spaces! */
609 return len;
612 static size_t diff_timestamp_len(const char *line, size_t len)
614 const char *end = line + len;
615 size_t n;
618 * Posix: 2010-07-05 19:41:17
619 * GNU: 2010-07-05 19:41:17.620000023 -0500
622 if (!isdigit(end[-1]))
623 return 0;
625 n = sane_tz_len(line, end - line);
626 if (!n)
627 n = tz_with_colon_len(line, end - line);
628 end -= n;
630 n = short_time_len(line, end - line);
631 if (!n)
632 n = fractional_time_len(line, end - line);
633 end -= n;
635 n = date_len(line, end - line);
636 if (!n) /* No date. Too bad. */
637 return 0;
638 end -= n;
640 if (end == line) /* No space before date. */
641 return 0;
642 if (end[-1] == '\t') { /* Success! */
643 end--;
644 return line + len - end;
646 if (end[-1] != ' ') /* No space before date. */
647 return 0;
649 /* Whitespace damage. */
650 end -= trailing_spaces_len(line, end - line);
651 return line + len - end;
654 static char *find_name_common(struct apply_state *state,
655 const char *line,
656 const char *def,
657 int p_value,
658 const char *end,
659 int terminate)
661 int len;
662 const char *start = NULL;
664 if (p_value == 0)
665 start = line;
666 while (line != end) {
667 char c = *line;
669 if (!end && isspace(c)) {
670 if (c == '\n')
671 break;
672 if (name_terminate(c, terminate))
673 break;
675 line++;
676 if (c == '/' && !--p_value)
677 start = line;
679 if (!start)
680 return squash_slash(xstrdup_or_null(def));
681 len = line - start;
682 if (!len)
683 return squash_slash(xstrdup_or_null(def));
686 * Generally we prefer the shorter name, especially
687 * if the other one is just a variation of that with
688 * something else tacked on to the end (ie "file.orig"
689 * or "file~").
691 if (def) {
692 int deflen = strlen(def);
693 if (deflen < len && !strncmp(start, def, deflen))
694 return squash_slash(xstrdup(def));
697 if (state->root.len) {
698 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
699 return squash_slash(ret);
702 return squash_slash(xmemdupz(start, len));
705 static char *find_name(struct apply_state *state,
706 const char *line,
707 char *def,
708 int p_value,
709 int terminate)
711 if (*line == '"') {
712 char *name = find_name_gnu(state, line, def, p_value);
713 if (name)
714 return name;
717 return find_name_common(state, line, def, p_value, NULL, terminate);
720 static char *find_name_traditional(struct apply_state *state,
721 const char *line,
722 char *def,
723 int p_value)
725 size_t len;
726 size_t date_len;
728 if (*line == '"') {
729 char *name = find_name_gnu(state, line, def, p_value);
730 if (name)
731 return name;
734 len = strchrnul(line, '\n') - line;
735 date_len = diff_timestamp_len(line, len);
736 if (!date_len)
737 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
738 len -= date_len;
740 return find_name_common(state, line, def, p_value, line + len, 0);
744 * Given the string after "--- " or "+++ ", guess the appropriate
745 * p_value for the given patch.
747 static int guess_p_value(struct apply_state *state, const char *nameline)
749 char *name, *cp;
750 int val = -1;
752 if (is_dev_null(nameline))
753 return -1;
754 name = find_name_traditional(state, nameline, NULL, 0);
755 if (!name)
756 return -1;
757 cp = strchr(name, '/');
758 if (!cp)
759 val = 0;
760 else if (state->prefix) {
762 * Does it begin with "a/$our-prefix" and such? Then this is
763 * very likely to apply to our directory.
765 if (starts_with(name, state->prefix))
766 val = count_slashes(state->prefix);
767 else {
768 cp++;
769 if (starts_with(cp, state->prefix))
770 val = count_slashes(state->prefix) + 1;
773 free(name);
774 return val;
778 * Does the ---/+++ line have the POSIX timestamp after the last HT?
779 * GNU diff puts epoch there to signal a creation/deletion event. Is
780 * this such a timestamp?
782 static int has_epoch_timestamp(const char *nameline)
785 * We are only interested in epoch timestamp; any non-zero
786 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
787 * For the same reason, the date must be either 1969-12-31 or
788 * 1970-01-01, and the seconds part must be "00".
790 const char stamp_regexp[] =
791 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
793 "([-+][0-2][0-9]:?[0-5][0-9])\n";
794 const char *timestamp = NULL, *cp, *colon;
795 static regex_t *stamp;
796 regmatch_t m[10];
797 int zoneoffset, epoch_hour, hour, minute;
798 int status;
800 for (cp = nameline; *cp != '\n'; cp++) {
801 if (*cp == '\t')
802 timestamp = cp + 1;
804 if (!timestamp)
805 return 0;
808 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
809 * (west of GMT) or 1970-01-01 (east of GMT)
811 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
812 epoch_hour = 24;
813 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
814 epoch_hour = 0;
815 else
816 return 0;
818 if (!stamp) {
819 stamp = xmalloc(sizeof(*stamp));
820 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
821 warning(_("Cannot prepare timestamp regexp %s"),
822 stamp_regexp);
823 return 0;
827 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
828 if (status) {
829 if (status != REG_NOMATCH)
830 warning(_("regexec returned %d for input: %s"),
831 status, timestamp);
832 return 0;
835 hour = strtol(timestamp, NULL, 10);
836 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
838 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
839 if (*colon == ':')
840 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
841 else
842 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
843 if (timestamp[m[3].rm_so] == '-')
844 zoneoffset = -zoneoffset;
846 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
850 * Get the name etc info from the ---/+++ lines of a traditional patch header
852 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
853 * files, we can happily check the index for a match, but for creating a
854 * new file we should try to match whatever "patch" does. I have no idea.
856 static int parse_traditional_patch(struct apply_state *state,
857 const char *first,
858 const char *second,
859 struct patch *patch)
861 char *name;
863 first += 4; /* skip "--- " */
864 second += 4; /* skip "+++ " */
865 if (!state->p_value_known) {
866 int p, q;
867 p = guess_p_value(state, first);
868 q = guess_p_value(state, second);
869 if (p < 0) p = q;
870 if (0 <= p && p == q) {
871 state->p_value = p;
872 state->p_value_known = 1;
875 if (is_dev_null(first)) {
876 patch->is_new = 1;
877 patch->is_delete = 0;
878 name = find_name_traditional(state, second, NULL, state->p_value);
879 patch->new_name = name;
880 } else if (is_dev_null(second)) {
881 patch->is_new = 0;
882 patch->is_delete = 1;
883 name = find_name_traditional(state, first, NULL, state->p_value);
884 patch->old_name = name;
885 } else {
886 char *first_name;
887 first_name = find_name_traditional(state, first, NULL, state->p_value);
888 name = find_name_traditional(state, second, first_name, state->p_value);
889 free(first_name);
890 if (has_epoch_timestamp(first)) {
891 patch->is_new = 1;
892 patch->is_delete = 0;
893 patch->new_name = name;
894 } else if (has_epoch_timestamp(second)) {
895 patch->is_new = 0;
896 patch->is_delete = 1;
897 patch->old_name = name;
898 } else {
899 patch->old_name = name;
900 patch->new_name = xstrdup_or_null(name);
903 if (!name)
904 return error(_("unable to find filename in patch at line %d"), state->linenr);
906 return 0;
909 static int gitdiff_hdrend(struct apply_state *state,
910 const char *line,
911 struct patch *patch)
913 return 1;
917 * We're anal about diff header consistency, to make
918 * sure that we don't end up having strange ambiguous
919 * patches floating around.
921 * As a result, gitdiff_{old|new}name() will check
922 * their names against any previous information, just
923 * to make sure..
925 #define DIFF_OLD_NAME 0
926 #define DIFF_NEW_NAME 1
928 static int gitdiff_verify_name(struct apply_state *state,
929 const char *line,
930 int isnull,
931 char **name,
932 int side)
934 if (!*name && !isnull) {
935 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
936 return 0;
939 if (*name) {
940 char *another;
941 if (isnull)
942 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
943 *name, state->linenr);
944 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
945 if (!another || strcmp(another, *name)) {
946 free(another);
947 return error((side == DIFF_NEW_NAME) ?
948 _("git apply: bad git-diff - inconsistent new filename on line %d") :
949 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
951 free(another);
952 } else {
953 if (!starts_with(line, "/dev/null\n"))
954 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
957 return 0;
960 static int gitdiff_oldname(struct apply_state *state,
961 const char *line,
962 struct patch *patch)
964 return gitdiff_verify_name(state, line,
965 patch->is_new, &patch->old_name,
966 DIFF_OLD_NAME);
969 static int gitdiff_newname(struct apply_state *state,
970 const char *line,
971 struct patch *patch)
973 return gitdiff_verify_name(state, line,
974 patch->is_delete, &patch->new_name,
975 DIFF_NEW_NAME);
978 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
980 char *end;
981 *mode = strtoul(line, &end, 8);
982 if (end == line || !isspace(*end))
983 return error(_("invalid mode on line %d: %s"), linenr, line);
984 return 0;
987 static int gitdiff_oldmode(struct apply_state *state,
988 const char *line,
989 struct patch *patch)
991 return parse_mode_line(line, state->linenr, &patch->old_mode);
994 static int gitdiff_newmode(struct apply_state *state,
995 const char *line,
996 struct patch *patch)
998 return parse_mode_line(line, state->linenr, &patch->new_mode);
1001 static int gitdiff_delete(struct apply_state *state,
1002 const char *line,
1003 struct patch *patch)
1005 patch->is_delete = 1;
1006 free(patch->old_name);
1007 patch->old_name = xstrdup_or_null(patch->def_name);
1008 return gitdiff_oldmode(state, line, patch);
1011 static int gitdiff_newfile(struct apply_state *state,
1012 const char *line,
1013 struct patch *patch)
1015 patch->is_new = 1;
1016 free(patch->new_name);
1017 patch->new_name = xstrdup_or_null(patch->def_name);
1018 return gitdiff_newmode(state, line, patch);
1021 static int gitdiff_copysrc(struct apply_state *state,
1022 const char *line,
1023 struct patch *patch)
1025 patch->is_copy = 1;
1026 free(patch->old_name);
1027 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1028 return 0;
1031 static int gitdiff_copydst(struct apply_state *state,
1032 const char *line,
1033 struct patch *patch)
1035 patch->is_copy = 1;
1036 free(patch->new_name);
1037 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1038 return 0;
1041 static int gitdiff_renamesrc(struct apply_state *state,
1042 const char *line,
1043 struct patch *patch)
1045 patch->is_rename = 1;
1046 free(patch->old_name);
1047 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1048 return 0;
1051 static int gitdiff_renamedst(struct apply_state *state,
1052 const char *line,
1053 struct patch *patch)
1055 patch->is_rename = 1;
1056 free(patch->new_name);
1057 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1058 return 0;
1061 static int gitdiff_similarity(struct apply_state *state,
1062 const char *line,
1063 struct patch *patch)
1065 unsigned long val = strtoul(line, NULL, 10);
1066 if (val <= 100)
1067 patch->score = val;
1068 return 0;
1071 static int gitdiff_dissimilarity(struct apply_state *state,
1072 const char *line,
1073 struct patch *patch)
1075 unsigned long val = strtoul(line, NULL, 10);
1076 if (val <= 100)
1077 patch->score = val;
1078 return 0;
1081 static int gitdiff_index(struct apply_state *state,
1082 const char *line,
1083 struct patch *patch)
1086 * index line is N hexadecimal, "..", N hexadecimal,
1087 * and optional space with octal mode.
1089 const char *ptr, *eol;
1090 int len;
1092 ptr = strchr(line, '.');
1093 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1094 return 0;
1095 len = ptr - line;
1096 memcpy(patch->old_sha1_prefix, line, len);
1097 patch->old_sha1_prefix[len] = 0;
1099 line = ptr + 2;
1100 ptr = strchr(line, ' ');
1101 eol = strchrnul(line, '\n');
1103 if (!ptr || eol < ptr)
1104 ptr = eol;
1105 len = ptr - line;
1107 if (40 < len)
1108 return 0;
1109 memcpy(patch->new_sha1_prefix, line, len);
1110 patch->new_sha1_prefix[len] = 0;
1111 if (*ptr == ' ')
1112 return gitdiff_oldmode(state, ptr + 1, patch);
1113 return 0;
1117 * This is normal for a diff that doesn't change anything: we'll fall through
1118 * into the next diff. Tell the parser to break out.
1120 static int gitdiff_unrecognized(struct apply_state *state,
1121 const char *line,
1122 struct patch *patch)
1124 return 1;
1128 * Skip p_value leading components from "line"; as we do not accept
1129 * absolute paths, return NULL in that case.
1131 static const char *skip_tree_prefix(struct apply_state *state,
1132 const char *line,
1133 int llen)
1135 int nslash;
1136 int i;
1138 if (!state->p_value)
1139 return (llen && line[0] == '/') ? NULL : line;
1141 nslash = state->p_value;
1142 for (i = 0; i < llen; i++) {
1143 int ch = line[i];
1144 if (ch == '/' && --nslash <= 0)
1145 return (i == 0) ? NULL : &line[i + 1];
1147 return NULL;
1151 * This is to extract the same name that appears on "diff --git"
1152 * line. We do not find and return anything if it is a rename
1153 * patch, and it is OK because we will find the name elsewhere.
1154 * We need to reliably find name only when it is mode-change only,
1155 * creation or deletion of an empty file. In any of these cases,
1156 * both sides are the same name under a/ and b/ respectively.
1158 static char *git_header_name(struct apply_state *state,
1159 const char *line,
1160 int llen)
1162 const char *name;
1163 const char *second = NULL;
1164 size_t len, line_len;
1166 line += strlen("diff --git ");
1167 llen -= strlen("diff --git ");
1169 if (*line == '"') {
1170 const char *cp;
1171 struct strbuf first = STRBUF_INIT;
1172 struct strbuf sp = STRBUF_INIT;
1174 if (unquote_c_style(&first, line, &second))
1175 goto free_and_fail1;
1177 /* strip the a/b prefix including trailing slash */
1178 cp = skip_tree_prefix(state, first.buf, first.len);
1179 if (!cp)
1180 goto free_and_fail1;
1181 strbuf_remove(&first, 0, cp - first.buf);
1184 * second points at one past closing dq of name.
1185 * find the second name.
1187 while ((second < line + llen) && isspace(*second))
1188 second++;
1190 if (line + llen <= second)
1191 goto free_and_fail1;
1192 if (*second == '"') {
1193 if (unquote_c_style(&sp, second, NULL))
1194 goto free_and_fail1;
1195 cp = skip_tree_prefix(state, sp.buf, sp.len);
1196 if (!cp)
1197 goto free_and_fail1;
1198 /* They must match, otherwise ignore */
1199 if (strcmp(cp, first.buf))
1200 goto free_and_fail1;
1201 strbuf_release(&sp);
1202 return strbuf_detach(&first, NULL);
1205 /* unquoted second */
1206 cp = skip_tree_prefix(state, second, line + llen - second);
1207 if (!cp)
1208 goto free_and_fail1;
1209 if (line + llen - cp != first.len ||
1210 memcmp(first.buf, cp, first.len))
1211 goto free_and_fail1;
1212 return strbuf_detach(&first, NULL);
1214 free_and_fail1:
1215 strbuf_release(&first);
1216 strbuf_release(&sp);
1217 return NULL;
1220 /* unquoted first name */
1221 name = skip_tree_prefix(state, line, llen);
1222 if (!name)
1223 return NULL;
1226 * since the first name is unquoted, a dq if exists must be
1227 * the beginning of the second name.
1229 for (second = name; second < line + llen; second++) {
1230 if (*second == '"') {
1231 struct strbuf sp = STRBUF_INIT;
1232 const char *np;
1234 if (unquote_c_style(&sp, second, NULL))
1235 goto free_and_fail2;
1237 np = skip_tree_prefix(state, sp.buf, sp.len);
1238 if (!np)
1239 goto free_and_fail2;
1241 len = sp.buf + sp.len - np;
1242 if (len < second - name &&
1243 !strncmp(np, name, len) &&
1244 isspace(name[len])) {
1245 /* Good */
1246 strbuf_remove(&sp, 0, np - sp.buf);
1247 return strbuf_detach(&sp, NULL);
1250 free_and_fail2:
1251 strbuf_release(&sp);
1252 return NULL;
1257 * Accept a name only if it shows up twice, exactly the same
1258 * form.
1260 second = strchr(name, '\n');
1261 if (!second)
1262 return NULL;
1263 line_len = second - name;
1264 for (len = 0 ; ; len++) {
1265 switch (name[len]) {
1266 default:
1267 continue;
1268 case '\n':
1269 return NULL;
1270 case '\t': case ' ':
1272 * Is this the separator between the preimage
1273 * and the postimage pathname? Again, we are
1274 * only interested in the case where there is
1275 * no rename, as this is only to set def_name
1276 * and a rename patch has the names elsewhere
1277 * in an unambiguous form.
1279 if (!name[len + 1])
1280 return NULL; /* no postimage name */
1281 second = skip_tree_prefix(state, name + len + 1,
1282 line_len - (len + 1));
1283 if (!second)
1284 return NULL;
1286 * Does len bytes starting at "name" and "second"
1287 * (that are separated by one HT or SP we just
1288 * found) exactly match?
1290 if (second[len] == '\n' && !strncmp(name, second, len))
1291 return xmemdupz(name, len);
1296 static int check_header_line(struct apply_state *state, struct patch *patch)
1298 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1299 (patch->is_rename == 1) + (patch->is_copy == 1);
1300 if (extensions > 1)
1301 return error(_("inconsistent header lines %d and %d"),
1302 patch->extension_linenr, state->linenr);
1303 if (extensions && !patch->extension_linenr)
1304 patch->extension_linenr = state->linenr;
1305 return 0;
1308 /* Verify that we recognize the lines following a git header */
1309 static int parse_git_header(struct apply_state *state,
1310 const char *line,
1311 int len,
1312 unsigned int size,
1313 struct patch *patch)
1315 unsigned long offset;
1317 /* A git diff has explicit new/delete information, so we don't guess */
1318 patch->is_new = 0;
1319 patch->is_delete = 0;
1322 * Some things may not have the old name in the
1323 * rest of the headers anywhere (pure mode changes,
1324 * or removing or adding empty files), so we get
1325 * the default name from the header.
1327 patch->def_name = git_header_name(state, line, len);
1328 if (patch->def_name && state->root.len) {
1329 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1330 free(patch->def_name);
1331 patch->def_name = s;
1334 line += len;
1335 size -= len;
1336 state->linenr++;
1337 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1338 static const struct opentry {
1339 const char *str;
1340 int (*fn)(struct apply_state *, const char *, struct patch *);
1341 } optable[] = {
1342 { "@@ -", gitdiff_hdrend },
1343 { "--- ", gitdiff_oldname },
1344 { "+++ ", gitdiff_newname },
1345 { "old mode ", gitdiff_oldmode },
1346 { "new mode ", gitdiff_newmode },
1347 { "deleted file mode ", gitdiff_delete },
1348 { "new file mode ", gitdiff_newfile },
1349 { "copy from ", gitdiff_copysrc },
1350 { "copy to ", gitdiff_copydst },
1351 { "rename old ", gitdiff_renamesrc },
1352 { "rename new ", gitdiff_renamedst },
1353 { "rename from ", gitdiff_renamesrc },
1354 { "rename to ", gitdiff_renamedst },
1355 { "similarity index ", gitdiff_similarity },
1356 { "dissimilarity index ", gitdiff_dissimilarity },
1357 { "index ", gitdiff_index },
1358 { "", gitdiff_unrecognized },
1360 int i;
1362 len = linelen(line, size);
1363 if (!len || line[len-1] != '\n')
1364 break;
1365 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1366 const struct opentry *p = optable + i;
1367 int oplen = strlen(p->str);
1368 int res;
1369 if (len < oplen || memcmp(p->str, line, oplen))
1370 continue;
1371 res = p->fn(state, line + oplen, patch);
1372 if (res < 0)
1373 return -1;
1374 if (check_header_line(state, patch))
1375 return -1;
1376 if (res > 0)
1377 return offset;
1378 break;
1382 return offset;
1385 static int parse_num(const char *line, unsigned long *p)
1387 char *ptr;
1389 if (!isdigit(*line))
1390 return 0;
1391 *p = strtoul(line, &ptr, 10);
1392 return ptr - line;
1395 static int parse_range(const char *line, int len, int offset, const char *expect,
1396 unsigned long *p1, unsigned long *p2)
1398 int digits, ex;
1400 if (offset < 0 || offset >= len)
1401 return -1;
1402 line += offset;
1403 len -= offset;
1405 digits = parse_num(line, p1);
1406 if (!digits)
1407 return -1;
1409 offset += digits;
1410 line += digits;
1411 len -= digits;
1413 *p2 = 1;
1414 if (*line == ',') {
1415 digits = parse_num(line+1, p2);
1416 if (!digits)
1417 return -1;
1419 offset += digits+1;
1420 line += digits+1;
1421 len -= digits+1;
1424 ex = strlen(expect);
1425 if (ex > len)
1426 return -1;
1427 if (memcmp(line, expect, ex))
1428 return -1;
1430 return offset + ex;
1433 static void recount_diff(const char *line, int size, struct fragment *fragment)
1435 int oldlines = 0, newlines = 0, ret = 0;
1437 if (size < 1) {
1438 warning("recount: ignore empty hunk");
1439 return;
1442 for (;;) {
1443 int len = linelen(line, size);
1444 size -= len;
1445 line += len;
1447 if (size < 1)
1448 break;
1450 switch (*line) {
1451 case ' ': case '\n':
1452 newlines++;
1453 /* fall through */
1454 case '-':
1455 oldlines++;
1456 continue;
1457 case '+':
1458 newlines++;
1459 continue;
1460 case '\\':
1461 continue;
1462 case '@':
1463 ret = size < 3 || !starts_with(line, "@@ ");
1464 break;
1465 case 'd':
1466 ret = size < 5 || !starts_with(line, "diff ");
1467 break;
1468 default:
1469 ret = -1;
1470 break;
1472 if (ret) {
1473 warning(_("recount: unexpected line: %.*s"),
1474 (int)linelen(line, size), line);
1475 return;
1477 break;
1479 fragment->oldlines = oldlines;
1480 fragment->newlines = newlines;
1484 * Parse a unified diff fragment header of the
1485 * form "@@ -a,b +c,d @@"
1487 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1489 int offset;
1491 if (!len || line[len-1] != '\n')
1492 return -1;
1494 /* Figure out the number of lines in a fragment */
1495 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1496 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1498 return offset;
1502 * Find file diff header
1504 * Returns:
1505 * -1 if no header was found
1506 * -128 in case of error
1507 * the size of the header in bytes (called "offset") otherwise
1509 static int find_header(struct apply_state *state,
1510 const char *line,
1511 unsigned long size,
1512 int *hdrsize,
1513 struct patch *patch)
1515 unsigned long offset, len;
1517 patch->is_toplevel_relative = 0;
1518 patch->is_rename = patch->is_copy = 0;
1519 patch->is_new = patch->is_delete = -1;
1520 patch->old_mode = patch->new_mode = 0;
1521 patch->old_name = patch->new_name = NULL;
1522 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1523 unsigned long nextlen;
1525 len = linelen(line, size);
1526 if (!len)
1527 break;
1529 /* Testing this early allows us to take a few shortcuts.. */
1530 if (len < 6)
1531 continue;
1534 * Make sure we don't find any unconnected patch fragments.
1535 * That's a sign that we didn't find a header, and that a
1536 * patch has become corrupted/broken up.
1538 if (!memcmp("@@ -", line, 4)) {
1539 struct fragment dummy;
1540 if (parse_fragment_header(line, len, &dummy) < 0)
1541 continue;
1542 error(_("patch fragment without header at line %d: %.*s"),
1543 state->linenr, (int)len-1, line);
1544 return -128;
1547 if (size < len + 6)
1548 break;
1551 * Git patch? It might not have a real patch, just a rename
1552 * or mode change, so we handle that specially
1554 if (!memcmp("diff --git ", line, 11)) {
1555 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1556 if (git_hdr_len < 0)
1557 return -128;
1558 if (git_hdr_len <= len)
1559 continue;
1560 if (!patch->old_name && !patch->new_name) {
1561 if (!patch->def_name) {
1562 error(Q_("git diff header lacks filename information when removing "
1563 "%d leading pathname component (line %d)",
1564 "git diff header lacks filename information when removing "
1565 "%d leading pathname components (line %d)",
1566 state->p_value),
1567 state->p_value, state->linenr);
1568 return -128;
1570 patch->old_name = xstrdup(patch->def_name);
1571 patch->new_name = xstrdup(patch->def_name);
1573 if ((!patch->new_name && !patch->is_delete) ||
1574 (!patch->old_name && !patch->is_new)) {
1575 error(_("git diff header lacks filename information "
1576 "(line %d)"), state->linenr);
1577 return -128;
1579 patch->is_toplevel_relative = 1;
1580 *hdrsize = git_hdr_len;
1581 return offset;
1584 /* --- followed by +++ ? */
1585 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1586 continue;
1589 * We only accept unified patches, so we want it to
1590 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1591 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1593 nextlen = linelen(line + len, size - len);
1594 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1595 continue;
1597 /* Ok, we'll consider it a patch */
1598 if (parse_traditional_patch(state, line, line+len, patch))
1599 return -128;
1600 *hdrsize = len + nextlen;
1601 state->linenr += 2;
1602 return offset;
1604 return -1;
1607 static void record_ws_error(struct apply_state *state,
1608 unsigned result,
1609 const char *line,
1610 int len,
1611 int linenr)
1613 char *err;
1615 if (!result)
1616 return;
1618 state->whitespace_error++;
1619 if (state->squelch_whitespace_errors &&
1620 state->squelch_whitespace_errors < state->whitespace_error)
1621 return;
1623 err = whitespace_error_string(result);
1624 if (state->apply_verbosity > verbosity_silent)
1625 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1626 state->patch_input_file, linenr, err, len, line);
1627 free(err);
1630 static void check_whitespace(struct apply_state *state,
1631 const char *line,
1632 int len,
1633 unsigned ws_rule)
1635 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1637 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1641 * Check if the patch has context lines with CRLF or
1642 * the patch wants to remove lines with CRLF.
1644 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1646 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1647 patch->ws_rule |= WS_CR_AT_EOL;
1648 patch->crlf_in_old = 1;
1654 * Parse a unified diff. Note that this really needs to parse each
1655 * fragment separately, since the only way to know the difference
1656 * between a "---" that is part of a patch, and a "---" that starts
1657 * the next patch is to look at the line counts..
1659 static int parse_fragment(struct apply_state *state,
1660 const char *line,
1661 unsigned long size,
1662 struct patch *patch,
1663 struct fragment *fragment)
1665 int added, deleted;
1666 int len = linelen(line, size), offset;
1667 unsigned long oldlines, newlines;
1668 unsigned long leading, trailing;
1670 offset = parse_fragment_header(line, len, fragment);
1671 if (offset < 0)
1672 return -1;
1673 if (offset > 0 && patch->recount)
1674 recount_diff(line + offset, size - offset, fragment);
1675 oldlines = fragment->oldlines;
1676 newlines = fragment->newlines;
1677 leading = 0;
1678 trailing = 0;
1680 /* Parse the thing.. */
1681 line += len;
1682 size -= len;
1683 state->linenr++;
1684 added = deleted = 0;
1685 for (offset = len;
1686 0 < size;
1687 offset += len, size -= len, line += len, state->linenr++) {
1688 if (!oldlines && !newlines)
1689 break;
1690 len = linelen(line, size);
1691 if (!len || line[len-1] != '\n')
1692 return -1;
1693 switch (*line) {
1694 default:
1695 return -1;
1696 case '\n': /* newer GNU diff, an empty context line */
1697 case ' ':
1698 oldlines--;
1699 newlines--;
1700 if (!deleted && !added)
1701 leading++;
1702 trailing++;
1703 check_old_for_crlf(patch, line, len);
1704 if (!state->apply_in_reverse &&
1705 state->ws_error_action == correct_ws_error)
1706 check_whitespace(state, line, len, patch->ws_rule);
1707 break;
1708 case '-':
1709 if (!state->apply_in_reverse)
1710 check_old_for_crlf(patch, line, len);
1711 if (state->apply_in_reverse &&
1712 state->ws_error_action != nowarn_ws_error)
1713 check_whitespace(state, line, len, patch->ws_rule);
1714 deleted++;
1715 oldlines--;
1716 trailing = 0;
1717 break;
1718 case '+':
1719 if (state->apply_in_reverse)
1720 check_old_for_crlf(patch, line, len);
1721 if (!state->apply_in_reverse &&
1722 state->ws_error_action != nowarn_ws_error)
1723 check_whitespace(state, line, len, patch->ws_rule);
1724 added++;
1725 newlines--;
1726 trailing = 0;
1727 break;
1730 * We allow "\ No newline at end of file". Depending
1731 * on locale settings when the patch was produced we
1732 * don't know what this line looks like. The only
1733 * thing we do know is that it begins with "\ ".
1734 * Checking for 12 is just for sanity check -- any
1735 * l10n of "\ No newline..." is at least that long.
1737 case '\\':
1738 if (len < 12 || memcmp(line, "\\ ", 2))
1739 return -1;
1740 break;
1743 if (oldlines || newlines)
1744 return -1;
1745 if (!deleted && !added)
1746 return -1;
1748 fragment->leading = leading;
1749 fragment->trailing = trailing;
1752 * If a fragment ends with an incomplete line, we failed to include
1753 * it in the above loop because we hit oldlines == newlines == 0
1754 * before seeing it.
1756 if (12 < size && !memcmp(line, "\\ ", 2))
1757 offset += linelen(line, size);
1759 patch->lines_added += added;
1760 patch->lines_deleted += deleted;
1762 if (0 < patch->is_new && oldlines)
1763 return error(_("new file depends on old contents"));
1764 if (0 < patch->is_delete && newlines)
1765 return error(_("deleted file still has contents"));
1766 return offset;
1770 * We have seen "diff --git a/... b/..." header (or a traditional patch
1771 * header). Read hunks that belong to this patch into fragments and hang
1772 * them to the given patch structure.
1774 * The (fragment->patch, fragment->size) pair points into the memory given
1775 * by the caller, not a copy, when we return.
1777 * Returns:
1778 * -1 in case of error,
1779 * the number of bytes in the patch otherwise.
1781 static int parse_single_patch(struct apply_state *state,
1782 const char *line,
1783 unsigned long size,
1784 struct patch *patch)
1786 unsigned long offset = 0;
1787 unsigned long oldlines = 0, newlines = 0, context = 0;
1788 struct fragment **fragp = &patch->fragments;
1790 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1791 struct fragment *fragment;
1792 int len;
1794 fragment = xcalloc(1, sizeof(*fragment));
1795 fragment->linenr = state->linenr;
1796 len = parse_fragment(state, line, size, patch, fragment);
1797 if (len <= 0) {
1798 free(fragment);
1799 return error(_("corrupt patch at line %d"), state->linenr);
1801 fragment->patch = line;
1802 fragment->size = len;
1803 oldlines += fragment->oldlines;
1804 newlines += fragment->newlines;
1805 context += fragment->leading + fragment->trailing;
1807 *fragp = fragment;
1808 fragp = &fragment->next;
1810 offset += len;
1811 line += len;
1812 size -= len;
1816 * If something was removed (i.e. we have old-lines) it cannot
1817 * be creation, and if something was added it cannot be
1818 * deletion. However, the reverse is not true; --unified=0
1819 * patches that only add are not necessarily creation even
1820 * though they do not have any old lines, and ones that only
1821 * delete are not necessarily deletion.
1823 * Unfortunately, a real creation/deletion patch do _not_ have
1824 * any context line by definition, so we cannot safely tell it
1825 * apart with --unified=0 insanity. At least if the patch has
1826 * more than one hunk it is not creation or deletion.
1828 if (patch->is_new < 0 &&
1829 (oldlines || (patch->fragments && patch->fragments->next)))
1830 patch->is_new = 0;
1831 if (patch->is_delete < 0 &&
1832 (newlines || (patch->fragments && patch->fragments->next)))
1833 patch->is_delete = 0;
1835 if (0 < patch->is_new && oldlines)
1836 return error(_("new file %s depends on old contents"), patch->new_name);
1837 if (0 < patch->is_delete && newlines)
1838 return error(_("deleted file %s still has contents"), patch->old_name);
1839 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1840 fprintf_ln(stderr,
1841 _("** warning: "
1842 "file %s becomes empty but is not deleted"),
1843 patch->new_name);
1845 return offset;
1848 static inline int metadata_changes(struct patch *patch)
1850 return patch->is_rename > 0 ||
1851 patch->is_copy > 0 ||
1852 patch->is_new > 0 ||
1853 patch->is_delete ||
1854 (patch->old_mode && patch->new_mode &&
1855 patch->old_mode != patch->new_mode);
1858 static char *inflate_it(const void *data, unsigned long size,
1859 unsigned long inflated_size)
1861 git_zstream stream;
1862 void *out;
1863 int st;
1865 memset(&stream, 0, sizeof(stream));
1867 stream.next_in = (unsigned char *)data;
1868 stream.avail_in = size;
1869 stream.next_out = out = xmalloc(inflated_size);
1870 stream.avail_out = inflated_size;
1871 git_inflate_init(&stream);
1872 st = git_inflate(&stream, Z_FINISH);
1873 git_inflate_end(&stream);
1874 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1875 free(out);
1876 return NULL;
1878 return out;
1882 * Read a binary hunk and return a new fragment; fragment->patch
1883 * points at an allocated memory that the caller must free, so
1884 * it is marked as "->free_patch = 1".
1886 static struct fragment *parse_binary_hunk(struct apply_state *state,
1887 char **buf_p,
1888 unsigned long *sz_p,
1889 int *status_p,
1890 int *used_p)
1893 * Expect a line that begins with binary patch method ("literal"
1894 * or "delta"), followed by the length of data before deflating.
1895 * a sequence of 'length-byte' followed by base-85 encoded data
1896 * should follow, terminated by a newline.
1898 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1899 * and we would limit the patch line to 66 characters,
1900 * so one line can fit up to 13 groups that would decode
1901 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1902 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1904 int llen, used;
1905 unsigned long size = *sz_p;
1906 char *buffer = *buf_p;
1907 int patch_method;
1908 unsigned long origlen;
1909 char *data = NULL;
1910 int hunk_size = 0;
1911 struct fragment *frag;
1913 llen = linelen(buffer, size);
1914 used = llen;
1916 *status_p = 0;
1918 if (starts_with(buffer, "delta ")) {
1919 patch_method = BINARY_DELTA_DEFLATED;
1920 origlen = strtoul(buffer + 6, NULL, 10);
1922 else if (starts_with(buffer, "literal ")) {
1923 patch_method = BINARY_LITERAL_DEFLATED;
1924 origlen = strtoul(buffer + 8, NULL, 10);
1926 else
1927 return NULL;
1929 state->linenr++;
1930 buffer += llen;
1931 while (1) {
1932 int byte_length, max_byte_length, newsize;
1933 llen = linelen(buffer, size);
1934 used += llen;
1935 state->linenr++;
1936 if (llen == 1) {
1937 /* consume the blank line */
1938 buffer++;
1939 size--;
1940 break;
1943 * Minimum line is "A00000\n" which is 7-byte long,
1944 * and the line length must be multiple of 5 plus 2.
1946 if ((llen < 7) || (llen-2) % 5)
1947 goto corrupt;
1948 max_byte_length = (llen - 2) / 5 * 4;
1949 byte_length = *buffer;
1950 if ('A' <= byte_length && byte_length <= 'Z')
1951 byte_length = byte_length - 'A' + 1;
1952 else if ('a' <= byte_length && byte_length <= 'z')
1953 byte_length = byte_length - 'a' + 27;
1954 else
1955 goto corrupt;
1956 /* if the input length was not multiple of 4, we would
1957 * have filler at the end but the filler should never
1958 * exceed 3 bytes
1960 if (max_byte_length < byte_length ||
1961 byte_length <= max_byte_length - 4)
1962 goto corrupt;
1963 newsize = hunk_size + byte_length;
1964 data = xrealloc(data, newsize);
1965 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1966 goto corrupt;
1967 hunk_size = newsize;
1968 buffer += llen;
1969 size -= llen;
1972 frag = xcalloc(1, sizeof(*frag));
1973 frag->patch = inflate_it(data, hunk_size, origlen);
1974 frag->free_patch = 1;
1975 if (!frag->patch)
1976 goto corrupt;
1977 free(data);
1978 frag->size = origlen;
1979 *buf_p = buffer;
1980 *sz_p = size;
1981 *used_p = used;
1982 frag->binary_patch_method = patch_method;
1983 return frag;
1985 corrupt:
1986 free(data);
1987 *status_p = -1;
1988 error(_("corrupt binary patch at line %d: %.*s"),
1989 state->linenr-1, llen-1, buffer);
1990 return NULL;
1994 * Returns:
1995 * -1 in case of error,
1996 * the length of the parsed binary patch otherwise
1998 static int parse_binary(struct apply_state *state,
1999 char *buffer,
2000 unsigned long size,
2001 struct patch *patch)
2004 * We have read "GIT binary patch\n"; what follows is a line
2005 * that says the patch method (currently, either "literal" or
2006 * "delta") and the length of data before deflating; a
2007 * sequence of 'length-byte' followed by base-85 encoded data
2008 * follows.
2010 * When a binary patch is reversible, there is another binary
2011 * hunk in the same format, starting with patch method (either
2012 * "literal" or "delta") with the length of data, and a sequence
2013 * of length-byte + base-85 encoded data, terminated with another
2014 * empty line. This data, when applied to the postimage, produces
2015 * the preimage.
2017 struct fragment *forward;
2018 struct fragment *reverse;
2019 int status;
2020 int used, used_1;
2022 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2023 if (!forward && !status)
2024 /* there has to be one hunk (forward hunk) */
2025 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2026 if (status)
2027 /* otherwise we already gave an error message */
2028 return status;
2030 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2031 if (reverse)
2032 used += used_1;
2033 else if (status) {
2035 * Not having reverse hunk is not an error, but having
2036 * a corrupt reverse hunk is.
2038 free((void*) forward->patch);
2039 free(forward);
2040 return status;
2042 forward->next = reverse;
2043 patch->fragments = forward;
2044 patch->is_binary = 1;
2045 return used;
2048 static void prefix_one(struct apply_state *state, char **name)
2050 char *old_name = *name;
2051 if (!old_name)
2052 return;
2053 *name = prefix_filename(state->prefix, *name);
2054 free(old_name);
2057 static void prefix_patch(struct apply_state *state, struct patch *p)
2059 if (!state->prefix || p->is_toplevel_relative)
2060 return;
2061 prefix_one(state, &p->new_name);
2062 prefix_one(state, &p->old_name);
2066 * include/exclude
2069 static void add_name_limit(struct apply_state *state,
2070 const char *name,
2071 int exclude)
2073 struct string_list_item *it;
2075 it = string_list_append(&state->limit_by_name, name);
2076 it->util = exclude ? NULL : (void *) 1;
2079 static int use_patch(struct apply_state *state, struct patch *p)
2081 const char *pathname = p->new_name ? p->new_name : p->old_name;
2082 int i;
2084 /* Paths outside are not touched regardless of "--include" */
2085 if (state->prefix && *state->prefix) {
2086 const char *rest;
2087 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2088 return 0;
2091 /* See if it matches any of exclude/include rule */
2092 for (i = 0; i < state->limit_by_name.nr; i++) {
2093 struct string_list_item *it = &state->limit_by_name.items[i];
2094 if (!wildmatch(it->string, pathname, 0))
2095 return (it->util != NULL);
2099 * If we had any include, a path that does not match any rule is
2100 * not used. Otherwise, we saw bunch of exclude rules (or none)
2101 * and such a path is used.
2103 return !state->has_include;
2107 * Read the patch text in "buffer" that extends for "size" bytes; stop
2108 * reading after seeing a single patch (i.e. changes to a single file).
2109 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2111 * Returns:
2112 * -1 if no header was found or parse_binary() failed,
2113 * -128 on another error,
2114 * the number of bytes consumed otherwise,
2115 * so that the caller can call us again for the next patch.
2117 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2119 int hdrsize, patchsize;
2120 int offset = find_header(state, buffer, size, &hdrsize, patch);
2122 if (offset < 0)
2123 return offset;
2125 prefix_patch(state, patch);
2127 if (!use_patch(state, patch))
2128 patch->ws_rule = 0;
2129 else
2130 patch->ws_rule = whitespace_rule(patch->new_name
2131 ? patch->new_name
2132 : patch->old_name);
2134 patchsize = parse_single_patch(state,
2135 buffer + offset + hdrsize,
2136 size - offset - hdrsize,
2137 patch);
2139 if (patchsize < 0)
2140 return -128;
2142 if (!patchsize) {
2143 static const char git_binary[] = "GIT binary patch\n";
2144 int hd = hdrsize + offset;
2145 unsigned long llen = linelen(buffer + hd, size - hd);
2147 if (llen == sizeof(git_binary) - 1 &&
2148 !memcmp(git_binary, buffer + hd, llen)) {
2149 int used;
2150 state->linenr++;
2151 used = parse_binary(state, buffer + hd + llen,
2152 size - hd - llen, patch);
2153 if (used < 0)
2154 return -1;
2155 if (used)
2156 patchsize = used + llen;
2157 else
2158 patchsize = 0;
2160 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2161 static const char *binhdr[] = {
2162 "Binary files ",
2163 "Files ",
2164 NULL,
2166 int i;
2167 for (i = 0; binhdr[i]; i++) {
2168 int len = strlen(binhdr[i]);
2169 if (len < size - hd &&
2170 !memcmp(binhdr[i], buffer + hd, len)) {
2171 state->linenr++;
2172 patch->is_binary = 1;
2173 patchsize = llen;
2174 break;
2179 /* Empty patch cannot be applied if it is a text patch
2180 * without metadata change. A binary patch appears
2181 * empty to us here.
2183 if ((state->apply || state->check) &&
2184 (!patch->is_binary && !metadata_changes(patch))) {
2185 error(_("patch with only garbage at line %d"), state->linenr);
2186 return -128;
2190 return offset + hdrsize + patchsize;
2193 static void reverse_patches(struct patch *p)
2195 for (; p; p = p->next) {
2196 struct fragment *frag = p->fragments;
2198 SWAP(p->new_name, p->old_name);
2199 SWAP(p->new_mode, p->old_mode);
2200 SWAP(p->is_new, p->is_delete);
2201 SWAP(p->lines_added, p->lines_deleted);
2202 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2204 for (; frag; frag = frag->next) {
2205 SWAP(frag->newpos, frag->oldpos);
2206 SWAP(frag->newlines, frag->oldlines);
2211 static const char pluses[] =
2212 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2213 static const char minuses[]=
2214 "----------------------------------------------------------------------";
2216 static void show_stats(struct apply_state *state, struct patch *patch)
2218 struct strbuf qname = STRBUF_INIT;
2219 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2220 int max, add, del;
2222 quote_c_style(cp, &qname, NULL, 0);
2225 * "scale" the filename
2227 max = state->max_len;
2228 if (max > 50)
2229 max = 50;
2231 if (qname.len > max) {
2232 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2233 if (!cp)
2234 cp = qname.buf + qname.len + 3 - max;
2235 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2238 if (patch->is_binary) {
2239 printf(" %-*s | Bin\n", max, qname.buf);
2240 strbuf_release(&qname);
2241 return;
2244 printf(" %-*s |", max, qname.buf);
2245 strbuf_release(&qname);
2248 * scale the add/delete
2250 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2251 add = patch->lines_added;
2252 del = patch->lines_deleted;
2254 if (state->max_change > 0) {
2255 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2256 add = (add * max + state->max_change / 2) / state->max_change;
2257 del = total - add;
2259 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2260 add, pluses, del, minuses);
2263 static int read_old_data(struct stat *st, struct patch *patch,
2264 const char *path, struct strbuf *buf)
2266 enum safe_crlf safe_crlf = patch->crlf_in_old ?
2267 SAFE_CRLF_KEEP_CRLF : SAFE_CRLF_RENORMALIZE;
2268 switch (st->st_mode & S_IFMT) {
2269 case S_IFLNK:
2270 if (strbuf_readlink(buf, path, st->st_size) < 0)
2271 return error(_("unable to read symlink %s"), path);
2272 return 0;
2273 case S_IFREG:
2274 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2275 return error(_("unable to open or read %s"), path);
2277 * "git apply" without "--index/--cached" should never look
2278 * at the index; the target file may not have been added to
2279 * the index yet, and we may not even be in any Git repository.
2280 * Pass NULL to convert_to_git() to stress this; the function
2281 * should never look at the index when explicit crlf option
2282 * is given.
2284 convert_to_git(NULL, path, buf->buf, buf->len, buf, safe_crlf);
2285 return 0;
2286 default:
2287 return -1;
2292 * Update the preimage, and the common lines in postimage,
2293 * from buffer buf of length len. If postlen is 0 the postimage
2294 * is updated in place, otherwise it's updated on a new buffer
2295 * of length postlen
2298 static void update_pre_post_images(struct image *preimage,
2299 struct image *postimage,
2300 char *buf,
2301 size_t len, size_t postlen)
2303 int i, ctx, reduced;
2304 char *new, *old, *fixed;
2305 struct image fixed_preimage;
2308 * Update the preimage with whitespace fixes. Note that we
2309 * are not losing preimage->buf -- apply_one_fragment() will
2310 * free "oldlines".
2312 prepare_image(&fixed_preimage, buf, len, 1);
2313 assert(postlen
2314 ? fixed_preimage.nr == preimage->nr
2315 : fixed_preimage.nr <= preimage->nr);
2316 for (i = 0; i < fixed_preimage.nr; i++)
2317 fixed_preimage.line[i].flag = preimage->line[i].flag;
2318 free(preimage->line_allocated);
2319 *preimage = fixed_preimage;
2322 * Adjust the common context lines in postimage. This can be
2323 * done in-place when we are shrinking it with whitespace
2324 * fixing, but needs a new buffer when ignoring whitespace or
2325 * expanding leading tabs to spaces.
2327 * We trust the caller to tell us if the update can be done
2328 * in place (postlen==0) or not.
2330 old = postimage->buf;
2331 if (postlen)
2332 new = postimage->buf = xmalloc(postlen);
2333 else
2334 new = old;
2335 fixed = preimage->buf;
2337 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2338 size_t l_len = postimage->line[i].len;
2339 if (!(postimage->line[i].flag & LINE_COMMON)) {
2340 /* an added line -- no counterparts in preimage */
2341 memmove(new, old, l_len);
2342 old += l_len;
2343 new += l_len;
2344 continue;
2347 /* a common context -- skip it in the original postimage */
2348 old += l_len;
2350 /* and find the corresponding one in the fixed preimage */
2351 while (ctx < preimage->nr &&
2352 !(preimage->line[ctx].flag & LINE_COMMON)) {
2353 fixed += preimage->line[ctx].len;
2354 ctx++;
2358 * preimage is expected to run out, if the caller
2359 * fixed addition of trailing blank lines.
2361 if (preimage->nr <= ctx) {
2362 reduced++;
2363 continue;
2366 /* and copy it in, while fixing the line length */
2367 l_len = preimage->line[ctx].len;
2368 memcpy(new, fixed, l_len);
2369 new += l_len;
2370 fixed += l_len;
2371 postimage->line[i].len = l_len;
2372 ctx++;
2375 if (postlen
2376 ? postlen < new - postimage->buf
2377 : postimage->len < new - postimage->buf)
2378 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2379 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2381 /* Fix the length of the whole thing */
2382 postimage->len = new - postimage->buf;
2383 postimage->nr -= reduced;
2386 static int line_by_line_fuzzy_match(struct image *img,
2387 struct image *preimage,
2388 struct image *postimage,
2389 unsigned long try,
2390 int try_lno,
2391 int preimage_limit)
2393 int i;
2394 size_t imgoff = 0;
2395 size_t preoff = 0;
2396 size_t postlen = postimage->len;
2397 size_t extra_chars;
2398 char *buf;
2399 char *preimage_eof;
2400 char *preimage_end;
2401 struct strbuf fixed;
2402 char *fixed_buf;
2403 size_t fixed_len;
2405 for (i = 0; i < preimage_limit; i++) {
2406 size_t prelen = preimage->line[i].len;
2407 size_t imglen = img->line[try_lno+i].len;
2409 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2410 preimage->buf + preoff, prelen))
2411 return 0;
2412 if (preimage->line[i].flag & LINE_COMMON)
2413 postlen += imglen - prelen;
2414 imgoff += imglen;
2415 preoff += prelen;
2419 * Ok, the preimage matches with whitespace fuzz.
2421 * imgoff now holds the true length of the target that
2422 * matches the preimage before the end of the file.
2424 * Count the number of characters in the preimage that fall
2425 * beyond the end of the file and make sure that all of them
2426 * are whitespace characters. (This can only happen if
2427 * we are removing blank lines at the end of the file.)
2429 buf = preimage_eof = preimage->buf + preoff;
2430 for ( ; i < preimage->nr; i++)
2431 preoff += preimage->line[i].len;
2432 preimage_end = preimage->buf + preoff;
2433 for ( ; buf < preimage_end; buf++)
2434 if (!isspace(*buf))
2435 return 0;
2438 * Update the preimage and the common postimage context
2439 * lines to use the same whitespace as the target.
2440 * If whitespace is missing in the target (i.e.
2441 * if the preimage extends beyond the end of the file),
2442 * use the whitespace from the preimage.
2444 extra_chars = preimage_end - preimage_eof;
2445 strbuf_init(&fixed, imgoff + extra_chars);
2446 strbuf_add(&fixed, img->buf + try, imgoff);
2447 strbuf_add(&fixed, preimage_eof, extra_chars);
2448 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2449 update_pre_post_images(preimage, postimage,
2450 fixed_buf, fixed_len, postlen);
2451 return 1;
2454 static int match_fragment(struct apply_state *state,
2455 struct image *img,
2456 struct image *preimage,
2457 struct image *postimage,
2458 unsigned long try,
2459 int try_lno,
2460 unsigned ws_rule,
2461 int match_beginning, int match_end)
2463 int i;
2464 char *fixed_buf, *buf, *orig, *target;
2465 struct strbuf fixed;
2466 size_t fixed_len, postlen;
2467 int preimage_limit;
2469 if (preimage->nr + try_lno <= img->nr) {
2471 * The hunk falls within the boundaries of img.
2473 preimage_limit = preimage->nr;
2474 if (match_end && (preimage->nr + try_lno != img->nr))
2475 return 0;
2476 } else if (state->ws_error_action == correct_ws_error &&
2477 (ws_rule & WS_BLANK_AT_EOF)) {
2479 * This hunk extends beyond the end of img, and we are
2480 * removing blank lines at the end of the file. This
2481 * many lines from the beginning of the preimage must
2482 * match with img, and the remainder of the preimage
2483 * must be blank.
2485 preimage_limit = img->nr - try_lno;
2486 } else {
2488 * The hunk extends beyond the end of the img and
2489 * we are not removing blanks at the end, so we
2490 * should reject the hunk at this position.
2492 return 0;
2495 if (match_beginning && try_lno)
2496 return 0;
2498 /* Quick hash check */
2499 for (i = 0; i < preimage_limit; i++)
2500 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2501 (preimage->line[i].hash != img->line[try_lno + i].hash))
2502 return 0;
2504 if (preimage_limit == preimage->nr) {
2506 * Do we have an exact match? If we were told to match
2507 * at the end, size must be exactly at try+fragsize,
2508 * otherwise try+fragsize must be still within the preimage,
2509 * and either case, the old piece should match the preimage
2510 * exactly.
2512 if ((match_end
2513 ? (try + preimage->len == img->len)
2514 : (try + preimage->len <= img->len)) &&
2515 !memcmp(img->buf + try, preimage->buf, preimage->len))
2516 return 1;
2517 } else {
2519 * The preimage extends beyond the end of img, so
2520 * there cannot be an exact match.
2522 * There must be one non-blank context line that match
2523 * a line before the end of img.
2525 char *buf_end;
2527 buf = preimage->buf;
2528 buf_end = buf;
2529 for (i = 0; i < preimage_limit; i++)
2530 buf_end += preimage->line[i].len;
2532 for ( ; buf < buf_end; buf++)
2533 if (!isspace(*buf))
2534 break;
2535 if (buf == buf_end)
2536 return 0;
2540 * No exact match. If we are ignoring whitespace, run a line-by-line
2541 * fuzzy matching. We collect all the line length information because
2542 * we need it to adjust whitespace if we match.
2544 if (state->ws_ignore_action == ignore_ws_change)
2545 return line_by_line_fuzzy_match(img, preimage, postimage,
2546 try, try_lno, preimage_limit);
2548 if (state->ws_error_action != correct_ws_error)
2549 return 0;
2552 * The hunk does not apply byte-by-byte, but the hash says
2553 * it might with whitespace fuzz. We weren't asked to
2554 * ignore whitespace, we were asked to correct whitespace
2555 * errors, so let's try matching after whitespace correction.
2557 * While checking the preimage against the target, whitespace
2558 * errors in both fixed, we count how large the corresponding
2559 * postimage needs to be. The postimage prepared by
2560 * apply_one_fragment() has whitespace errors fixed on added
2561 * lines already, but the common lines were propagated as-is,
2562 * which may become longer when their whitespace errors are
2563 * fixed.
2566 /* First count added lines in postimage */
2567 postlen = 0;
2568 for (i = 0; i < postimage->nr; i++) {
2569 if (!(postimage->line[i].flag & LINE_COMMON))
2570 postlen += postimage->line[i].len;
2574 * The preimage may extend beyond the end of the file,
2575 * but in this loop we will only handle the part of the
2576 * preimage that falls within the file.
2578 strbuf_init(&fixed, preimage->len + 1);
2579 orig = preimage->buf;
2580 target = img->buf + try;
2581 for (i = 0; i < preimage_limit; i++) {
2582 size_t oldlen = preimage->line[i].len;
2583 size_t tgtlen = img->line[try_lno + i].len;
2584 size_t fixstart = fixed.len;
2585 struct strbuf tgtfix;
2586 int match;
2588 /* Try fixing the line in the preimage */
2589 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2591 /* Try fixing the line in the target */
2592 strbuf_init(&tgtfix, tgtlen);
2593 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2596 * If they match, either the preimage was based on
2597 * a version before our tree fixed whitespace breakage,
2598 * or we are lacking a whitespace-fix patch the tree
2599 * the preimage was based on already had (i.e. target
2600 * has whitespace breakage, the preimage doesn't).
2601 * In either case, we are fixing the whitespace breakages
2602 * so we might as well take the fix together with their
2603 * real change.
2605 match = (tgtfix.len == fixed.len - fixstart &&
2606 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2607 fixed.len - fixstart));
2609 /* Add the length if this is common with the postimage */
2610 if (preimage->line[i].flag & LINE_COMMON)
2611 postlen += tgtfix.len;
2613 strbuf_release(&tgtfix);
2614 if (!match)
2615 goto unmatch_exit;
2617 orig += oldlen;
2618 target += tgtlen;
2623 * Now handle the lines in the preimage that falls beyond the
2624 * end of the file (if any). They will only match if they are
2625 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2626 * false).
2628 for ( ; i < preimage->nr; i++) {
2629 size_t fixstart = fixed.len; /* start of the fixed preimage */
2630 size_t oldlen = preimage->line[i].len;
2631 int j;
2633 /* Try fixing the line in the preimage */
2634 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2636 for (j = fixstart; j < fixed.len; j++)
2637 if (!isspace(fixed.buf[j]))
2638 goto unmatch_exit;
2640 orig += oldlen;
2644 * Yes, the preimage is based on an older version that still
2645 * has whitespace breakages unfixed, and fixing them makes the
2646 * hunk match. Update the context lines in the postimage.
2648 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2649 if (postlen < postimage->len)
2650 postlen = 0;
2651 update_pre_post_images(preimage, postimage,
2652 fixed_buf, fixed_len, postlen);
2653 return 1;
2655 unmatch_exit:
2656 strbuf_release(&fixed);
2657 return 0;
2660 static int find_pos(struct apply_state *state,
2661 struct image *img,
2662 struct image *preimage,
2663 struct image *postimage,
2664 int line,
2665 unsigned ws_rule,
2666 int match_beginning, int match_end)
2668 int i;
2669 unsigned long backwards, forwards, try;
2670 int backwards_lno, forwards_lno, try_lno;
2673 * If match_beginning or match_end is specified, there is no
2674 * point starting from a wrong line that will never match and
2675 * wander around and wait for a match at the specified end.
2677 if (match_beginning)
2678 line = 0;
2679 else if (match_end)
2680 line = img->nr - preimage->nr;
2683 * Because the comparison is unsigned, the following test
2684 * will also take care of a negative line number that can
2685 * result when match_end and preimage is larger than the target.
2687 if ((size_t) line > img->nr)
2688 line = img->nr;
2690 try = 0;
2691 for (i = 0; i < line; i++)
2692 try += img->line[i].len;
2695 * There's probably some smart way to do this, but I'll leave
2696 * that to the smart and beautiful people. I'm simple and stupid.
2698 backwards = try;
2699 backwards_lno = line;
2700 forwards = try;
2701 forwards_lno = line;
2702 try_lno = line;
2704 for (i = 0; ; i++) {
2705 if (match_fragment(state, img, preimage, postimage,
2706 try, try_lno, ws_rule,
2707 match_beginning, match_end))
2708 return try_lno;
2710 again:
2711 if (backwards_lno == 0 && forwards_lno == img->nr)
2712 break;
2714 if (i & 1) {
2715 if (backwards_lno == 0) {
2716 i++;
2717 goto again;
2719 backwards_lno--;
2720 backwards -= img->line[backwards_lno].len;
2721 try = backwards;
2722 try_lno = backwards_lno;
2723 } else {
2724 if (forwards_lno == img->nr) {
2725 i++;
2726 goto again;
2728 forwards += img->line[forwards_lno].len;
2729 forwards_lno++;
2730 try = forwards;
2731 try_lno = forwards_lno;
2735 return -1;
2738 static void remove_first_line(struct image *img)
2740 img->buf += img->line[0].len;
2741 img->len -= img->line[0].len;
2742 img->line++;
2743 img->nr--;
2746 static void remove_last_line(struct image *img)
2748 img->len -= img->line[--img->nr].len;
2752 * The change from "preimage" and "postimage" has been found to
2753 * apply at applied_pos (counts in line numbers) in "img".
2754 * Update "img" to remove "preimage" and replace it with "postimage".
2756 static void update_image(struct apply_state *state,
2757 struct image *img,
2758 int applied_pos,
2759 struct image *preimage,
2760 struct image *postimage)
2763 * remove the copy of preimage at offset in img
2764 * and replace it with postimage
2766 int i, nr;
2767 size_t remove_count, insert_count, applied_at = 0;
2768 char *result;
2769 int preimage_limit;
2772 * If we are removing blank lines at the end of img,
2773 * the preimage may extend beyond the end.
2774 * If that is the case, we must be careful only to
2775 * remove the part of the preimage that falls within
2776 * the boundaries of img. Initialize preimage_limit
2777 * to the number of lines in the preimage that falls
2778 * within the boundaries.
2780 preimage_limit = preimage->nr;
2781 if (preimage_limit > img->nr - applied_pos)
2782 preimage_limit = img->nr - applied_pos;
2784 for (i = 0; i < applied_pos; i++)
2785 applied_at += img->line[i].len;
2787 remove_count = 0;
2788 for (i = 0; i < preimage_limit; i++)
2789 remove_count += img->line[applied_pos + i].len;
2790 insert_count = postimage->len;
2792 /* Adjust the contents */
2793 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2794 memcpy(result, img->buf, applied_at);
2795 memcpy(result + applied_at, postimage->buf, postimage->len);
2796 memcpy(result + applied_at + postimage->len,
2797 img->buf + (applied_at + remove_count),
2798 img->len - (applied_at + remove_count));
2799 free(img->buf);
2800 img->buf = result;
2801 img->len += insert_count - remove_count;
2802 result[img->len] = '\0';
2804 /* Adjust the line table */
2805 nr = img->nr + postimage->nr - preimage_limit;
2806 if (preimage_limit < postimage->nr) {
2808 * NOTE: this knows that we never call remove_first_line()
2809 * on anything other than pre/post image.
2811 REALLOC_ARRAY(img->line, nr);
2812 img->line_allocated = img->line;
2814 if (preimage_limit != postimage->nr)
2815 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2816 img->line + applied_pos + preimage_limit,
2817 img->nr - (applied_pos + preimage_limit));
2818 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2819 if (!state->allow_overlap)
2820 for (i = 0; i < postimage->nr; i++)
2821 img->line[applied_pos + i].flag |= LINE_PATCHED;
2822 img->nr = nr;
2826 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2827 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2828 * replace the part of "img" with "postimage" text.
2830 static int apply_one_fragment(struct apply_state *state,
2831 struct image *img, struct fragment *frag,
2832 int inaccurate_eof, unsigned ws_rule,
2833 int nth_fragment)
2835 int match_beginning, match_end;
2836 const char *patch = frag->patch;
2837 int size = frag->size;
2838 char *old, *oldlines;
2839 struct strbuf newlines;
2840 int new_blank_lines_at_end = 0;
2841 int found_new_blank_lines_at_end = 0;
2842 int hunk_linenr = frag->linenr;
2843 unsigned long leading, trailing;
2844 int pos, applied_pos;
2845 struct image preimage;
2846 struct image postimage;
2848 memset(&preimage, 0, sizeof(preimage));
2849 memset(&postimage, 0, sizeof(postimage));
2850 oldlines = xmalloc(size);
2851 strbuf_init(&newlines, size);
2853 old = oldlines;
2854 while (size > 0) {
2855 char first;
2856 int len = linelen(patch, size);
2857 int plen;
2858 int added_blank_line = 0;
2859 int is_blank_context = 0;
2860 size_t start;
2862 if (!len)
2863 break;
2866 * "plen" is how much of the line we should use for
2867 * the actual patch data. Normally we just remove the
2868 * first character on the line, but if the line is
2869 * followed by "\ No newline", then we also remove the
2870 * last one (which is the newline, of course).
2872 plen = len - 1;
2873 if (len < size && patch[len] == '\\')
2874 plen--;
2875 first = *patch;
2876 if (state->apply_in_reverse) {
2877 if (first == '-')
2878 first = '+';
2879 else if (first == '+')
2880 first = '-';
2883 switch (first) {
2884 case '\n':
2885 /* Newer GNU diff, empty context line */
2886 if (plen < 0)
2887 /* ... followed by '\No newline'; nothing */
2888 break;
2889 *old++ = '\n';
2890 strbuf_addch(&newlines, '\n');
2891 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2892 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2893 is_blank_context = 1;
2894 break;
2895 case ' ':
2896 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2897 ws_blank_line(patch + 1, plen, ws_rule))
2898 is_blank_context = 1;
2899 /* fallthrough */
2900 case '-':
2901 memcpy(old, patch + 1, plen);
2902 add_line_info(&preimage, old, plen,
2903 (first == ' ' ? LINE_COMMON : 0));
2904 old += plen;
2905 if (first == '-')
2906 break;
2907 /* fallthrough */
2908 case '+':
2909 /* --no-add does not add new lines */
2910 if (first == '+' && state->no_add)
2911 break;
2913 start = newlines.len;
2914 if (first != '+' ||
2915 !state->whitespace_error ||
2916 state->ws_error_action != correct_ws_error) {
2917 strbuf_add(&newlines, patch + 1, plen);
2919 else {
2920 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2922 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2923 (first == '+' ? 0 : LINE_COMMON));
2924 if (first == '+' &&
2925 (ws_rule & WS_BLANK_AT_EOF) &&
2926 ws_blank_line(patch + 1, plen, ws_rule))
2927 added_blank_line = 1;
2928 break;
2929 case '@': case '\\':
2930 /* Ignore it, we already handled it */
2931 break;
2932 default:
2933 if (state->apply_verbosity > verbosity_normal)
2934 error(_("invalid start of line: '%c'"), first);
2935 applied_pos = -1;
2936 goto out;
2938 if (added_blank_line) {
2939 if (!new_blank_lines_at_end)
2940 found_new_blank_lines_at_end = hunk_linenr;
2941 new_blank_lines_at_end++;
2943 else if (is_blank_context)
2945 else
2946 new_blank_lines_at_end = 0;
2947 patch += len;
2948 size -= len;
2949 hunk_linenr++;
2951 if (inaccurate_eof &&
2952 old > oldlines && old[-1] == '\n' &&
2953 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2954 old--;
2955 strbuf_setlen(&newlines, newlines.len - 1);
2956 preimage.line_allocated[preimage.nr - 1].len--;
2957 postimage.line_allocated[postimage.nr - 1].len--;
2960 leading = frag->leading;
2961 trailing = frag->trailing;
2964 * A hunk to change lines at the beginning would begin with
2965 * @@ -1,L +N,M @@
2966 * but we need to be careful. -U0 that inserts before the second
2967 * line also has this pattern.
2969 * And a hunk to add to an empty file would begin with
2970 * @@ -0,0 +N,M @@
2972 * In other words, a hunk that is (frag->oldpos <= 1) with or
2973 * without leading context must match at the beginning.
2975 match_beginning = (!frag->oldpos ||
2976 (frag->oldpos == 1 && !state->unidiff_zero));
2979 * A hunk without trailing lines must match at the end.
2980 * However, we simply cannot tell if a hunk must match end
2981 * from the lack of trailing lines if the patch was generated
2982 * with unidiff without any context.
2984 match_end = !state->unidiff_zero && !trailing;
2986 pos = frag->newpos ? (frag->newpos - 1) : 0;
2987 preimage.buf = oldlines;
2988 preimage.len = old - oldlines;
2989 postimage.buf = newlines.buf;
2990 postimage.len = newlines.len;
2991 preimage.line = preimage.line_allocated;
2992 postimage.line = postimage.line_allocated;
2994 for (;;) {
2996 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2997 ws_rule, match_beginning, match_end);
2999 if (applied_pos >= 0)
3000 break;
3002 /* Am I at my context limits? */
3003 if ((leading <= state->p_context) && (trailing <= state->p_context))
3004 break;
3005 if (match_beginning || match_end) {
3006 match_beginning = match_end = 0;
3007 continue;
3011 * Reduce the number of context lines; reduce both
3012 * leading and trailing if they are equal otherwise
3013 * just reduce the larger context.
3015 if (leading >= trailing) {
3016 remove_first_line(&preimage);
3017 remove_first_line(&postimage);
3018 pos--;
3019 leading--;
3021 if (trailing > leading) {
3022 remove_last_line(&preimage);
3023 remove_last_line(&postimage);
3024 trailing--;
3028 if (applied_pos >= 0) {
3029 if (new_blank_lines_at_end &&
3030 preimage.nr + applied_pos >= img->nr &&
3031 (ws_rule & WS_BLANK_AT_EOF) &&
3032 state->ws_error_action != nowarn_ws_error) {
3033 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3034 found_new_blank_lines_at_end);
3035 if (state->ws_error_action == correct_ws_error) {
3036 while (new_blank_lines_at_end--)
3037 remove_last_line(&postimage);
3040 * We would want to prevent write_out_results()
3041 * from taking place in apply_patch() that follows
3042 * the callchain led us here, which is:
3043 * apply_patch->check_patch_list->check_patch->
3044 * apply_data->apply_fragments->apply_one_fragment
3046 if (state->ws_error_action == die_on_ws_error)
3047 state->apply = 0;
3050 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3051 int offset = applied_pos - pos;
3052 if (state->apply_in_reverse)
3053 offset = 0 - offset;
3054 fprintf_ln(stderr,
3055 Q_("Hunk #%d succeeded at %d (offset %d line).",
3056 "Hunk #%d succeeded at %d (offset %d lines).",
3057 offset),
3058 nth_fragment, applied_pos + 1, offset);
3062 * Warn if it was necessary to reduce the number
3063 * of context lines.
3065 if ((leading != frag->leading ||
3066 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3067 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3068 " to apply fragment at %d"),
3069 leading, trailing, applied_pos+1);
3070 update_image(state, img, applied_pos, &preimage, &postimage);
3071 } else {
3072 if (state->apply_verbosity > verbosity_normal)
3073 error(_("while searching for:\n%.*s"),
3074 (int)(old - oldlines), oldlines);
3077 out:
3078 free(oldlines);
3079 strbuf_release(&newlines);
3080 free(preimage.line_allocated);
3081 free(postimage.line_allocated);
3083 return (applied_pos < 0);
3086 static int apply_binary_fragment(struct apply_state *state,
3087 struct image *img,
3088 struct patch *patch)
3090 struct fragment *fragment = patch->fragments;
3091 unsigned long len;
3092 void *dst;
3094 if (!fragment)
3095 return error(_("missing binary patch data for '%s'"),
3096 patch->new_name ?
3097 patch->new_name :
3098 patch->old_name);
3100 /* Binary patch is irreversible without the optional second hunk */
3101 if (state->apply_in_reverse) {
3102 if (!fragment->next)
3103 return error(_("cannot reverse-apply a binary patch "
3104 "without the reverse hunk to '%s'"),
3105 patch->new_name
3106 ? patch->new_name : patch->old_name);
3107 fragment = fragment->next;
3109 switch (fragment->binary_patch_method) {
3110 case BINARY_DELTA_DEFLATED:
3111 dst = patch_delta(img->buf, img->len, fragment->patch,
3112 fragment->size, &len);
3113 if (!dst)
3114 return -1;
3115 clear_image(img);
3116 img->buf = dst;
3117 img->len = len;
3118 return 0;
3119 case BINARY_LITERAL_DEFLATED:
3120 clear_image(img);
3121 img->len = fragment->size;
3122 img->buf = xmemdupz(fragment->patch, img->len);
3123 return 0;
3125 return -1;
3129 * Replace "img" with the result of applying the binary patch.
3130 * The binary patch data itself in patch->fragment is still kept
3131 * but the preimage prepared by the caller in "img" is freed here
3132 * or in the helper function apply_binary_fragment() this calls.
3134 static int apply_binary(struct apply_state *state,
3135 struct image *img,
3136 struct patch *patch)
3138 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3139 struct object_id oid;
3142 * For safety, we require patch index line to contain
3143 * full 40-byte textual SHA1 for old and new, at least for now.
3145 if (strlen(patch->old_sha1_prefix) != 40 ||
3146 strlen(patch->new_sha1_prefix) != 40 ||
3147 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3148 get_oid_hex(patch->new_sha1_prefix, &oid))
3149 return error(_("cannot apply binary patch to '%s' "
3150 "without full index line"), name);
3152 if (patch->old_name) {
3154 * See if the old one matches what the patch
3155 * applies to.
3157 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3158 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3159 return error(_("the patch applies to '%s' (%s), "
3160 "which does not match the "
3161 "current contents."),
3162 name, oid_to_hex(&oid));
3164 else {
3165 /* Otherwise, the old one must be empty. */
3166 if (img->len)
3167 return error(_("the patch applies to an empty "
3168 "'%s' but it is not empty"), name);
3171 get_oid_hex(patch->new_sha1_prefix, &oid);
3172 if (is_null_oid(&oid)) {
3173 clear_image(img);
3174 return 0; /* deletion patch */
3177 if (has_sha1_file(oid.hash)) {
3178 /* We already have the postimage */
3179 enum object_type type;
3180 unsigned long size;
3181 char *result;
3183 result = read_sha1_file(oid.hash, &type, &size);
3184 if (!result)
3185 return error(_("the necessary postimage %s for "
3186 "'%s' cannot be read"),
3187 patch->new_sha1_prefix, name);
3188 clear_image(img);
3189 img->buf = result;
3190 img->len = size;
3191 } else {
3193 * We have verified buf matches the preimage;
3194 * apply the patch data to it, which is stored
3195 * in the patch->fragments->{patch,size}.
3197 if (apply_binary_fragment(state, img, patch))
3198 return error(_("binary patch does not apply to '%s'"),
3199 name);
3201 /* verify that the result matches */
3202 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3203 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3204 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3205 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3208 return 0;
3211 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3213 struct fragment *frag = patch->fragments;
3214 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3215 unsigned ws_rule = patch->ws_rule;
3216 unsigned inaccurate_eof = patch->inaccurate_eof;
3217 int nth = 0;
3219 if (patch->is_binary)
3220 return apply_binary(state, img, patch);
3222 while (frag) {
3223 nth++;
3224 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3225 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3226 if (!state->apply_with_reject)
3227 return -1;
3228 frag->rejected = 1;
3230 frag = frag->next;
3232 return 0;
3235 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3237 if (S_ISGITLINK(mode)) {
3238 strbuf_grow(buf, 100);
3239 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3240 } else {
3241 enum object_type type;
3242 unsigned long sz;
3243 char *result;
3245 result = read_sha1_file(oid->hash, &type, &sz);
3246 if (!result)
3247 return -1;
3248 /* XXX read_sha1_file NUL-terminates */
3249 strbuf_attach(buf, result, sz, sz + 1);
3251 return 0;
3254 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3256 if (!ce)
3257 return 0;
3258 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3261 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3263 struct string_list_item *item;
3265 if (name == NULL)
3266 return NULL;
3268 item = string_list_lookup(&state->fn_table, name);
3269 if (item != NULL)
3270 return (struct patch *)item->util;
3272 return NULL;
3276 * item->util in the filename table records the status of the path.
3277 * Usually it points at a patch (whose result records the contents
3278 * of it after applying it), but it could be PATH_WAS_DELETED for a
3279 * path that a previously applied patch has already removed, or
3280 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3282 * The latter is needed to deal with a case where two paths A and B
3283 * are swapped by first renaming A to B and then renaming B to A;
3284 * moving A to B should not be prevented due to presence of B as we
3285 * will remove it in a later patch.
3287 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3288 #define PATH_WAS_DELETED ((struct patch *) -1)
3290 static int to_be_deleted(struct patch *patch)
3292 return patch == PATH_TO_BE_DELETED;
3295 static int was_deleted(struct patch *patch)
3297 return patch == PATH_WAS_DELETED;
3300 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3302 struct string_list_item *item;
3305 * Always add new_name unless patch is a deletion
3306 * This should cover the cases for normal diffs,
3307 * file creations and copies
3309 if (patch->new_name != NULL) {
3310 item = string_list_insert(&state->fn_table, patch->new_name);
3311 item->util = patch;
3315 * store a failure on rename/deletion cases because
3316 * later chunks shouldn't patch old names
3318 if ((patch->new_name == NULL) || (patch->is_rename)) {
3319 item = string_list_insert(&state->fn_table, patch->old_name);
3320 item->util = PATH_WAS_DELETED;
3324 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3327 * store information about incoming file deletion
3329 while (patch) {
3330 if ((patch->new_name == NULL) || (patch->is_rename)) {
3331 struct string_list_item *item;
3332 item = string_list_insert(&state->fn_table, patch->old_name);
3333 item->util = PATH_TO_BE_DELETED;
3335 patch = patch->next;
3339 static int checkout_target(struct index_state *istate,
3340 struct cache_entry *ce, struct stat *st)
3342 struct checkout costate = CHECKOUT_INIT;
3344 costate.refresh_cache = 1;
3345 costate.istate = istate;
3346 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3347 return error(_("cannot checkout %s"), ce->name);
3348 return 0;
3351 static struct patch *previous_patch(struct apply_state *state,
3352 struct patch *patch,
3353 int *gone)
3355 struct patch *previous;
3357 *gone = 0;
3358 if (patch->is_copy || patch->is_rename)
3359 return NULL; /* "git" patches do not depend on the order */
3361 previous = in_fn_table(state, patch->old_name);
3362 if (!previous)
3363 return NULL;
3365 if (to_be_deleted(previous))
3366 return NULL; /* the deletion hasn't happened yet */
3368 if (was_deleted(previous))
3369 *gone = 1;
3371 return previous;
3374 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3376 if (S_ISGITLINK(ce->ce_mode)) {
3377 if (!S_ISDIR(st->st_mode))
3378 return -1;
3379 return 0;
3381 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3384 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3386 static int load_patch_target(struct apply_state *state,
3387 struct strbuf *buf,
3388 const struct cache_entry *ce,
3389 struct stat *st,
3390 struct patch *patch,
3391 const char *name,
3392 unsigned expected_mode)
3394 if (state->cached || state->check_index) {
3395 if (read_file_or_gitlink(ce, buf))
3396 return error(_("failed to read %s"), name);
3397 } else if (name) {
3398 if (S_ISGITLINK(expected_mode)) {
3399 if (ce)
3400 return read_file_or_gitlink(ce, buf);
3401 else
3402 return SUBMODULE_PATCH_WITHOUT_INDEX;
3403 } else if (has_symlink_leading_path(name, strlen(name))) {
3404 return error(_("reading from '%s' beyond a symbolic link"), name);
3405 } else {
3406 if (read_old_data(st, patch, name, buf))
3407 return error(_("failed to read %s"), name);
3410 return 0;
3414 * We are about to apply "patch"; populate the "image" with the
3415 * current version we have, from the working tree or from the index,
3416 * depending on the situation e.g. --cached/--index. If we are
3417 * applying a non-git patch that incrementally updates the tree,
3418 * we read from the result of a previous diff.
3420 static int load_preimage(struct apply_state *state,
3421 struct image *image,
3422 struct patch *patch, struct stat *st,
3423 const struct cache_entry *ce)
3425 struct strbuf buf = STRBUF_INIT;
3426 size_t len;
3427 char *img;
3428 struct patch *previous;
3429 int status;
3431 previous = previous_patch(state, patch, &status);
3432 if (status)
3433 return error(_("path %s has been renamed/deleted"),
3434 patch->old_name);
3435 if (previous) {
3436 /* We have a patched copy in memory; use that. */
3437 strbuf_add(&buf, previous->result, previous->resultsize);
3438 } else {
3439 status = load_patch_target(state, &buf, ce, st, patch,
3440 patch->old_name, patch->old_mode);
3441 if (status < 0)
3442 return status;
3443 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3445 * There is no way to apply subproject
3446 * patch without looking at the index.
3447 * NEEDSWORK: shouldn't this be flagged
3448 * as an error???
3450 free_fragment_list(patch->fragments);
3451 patch->fragments = NULL;
3452 } else if (status) {
3453 return error(_("failed to read %s"), patch->old_name);
3457 img = strbuf_detach(&buf, &len);
3458 prepare_image(image, img, len, !patch->is_binary);
3459 return 0;
3462 static int three_way_merge(struct image *image,
3463 char *path,
3464 const struct object_id *base,
3465 const struct object_id *ours,
3466 const struct object_id *theirs)
3468 mmfile_t base_file, our_file, their_file;
3469 mmbuffer_t result = { NULL };
3470 int status;
3472 read_mmblob(&base_file, base);
3473 read_mmblob(&our_file, ours);
3474 read_mmblob(&their_file, theirs);
3475 status = ll_merge(&result, path,
3476 &base_file, "base",
3477 &our_file, "ours",
3478 &their_file, "theirs", NULL);
3479 free(base_file.ptr);
3480 free(our_file.ptr);
3481 free(their_file.ptr);
3482 if (status < 0 || !result.ptr) {
3483 free(result.ptr);
3484 return -1;
3486 clear_image(image);
3487 image->buf = result.ptr;
3488 image->len = result.size;
3490 return status;
3494 * When directly falling back to add/add three-way merge, we read from
3495 * the current contents of the new_name. In no cases other than that
3496 * this function will be called.
3498 static int load_current(struct apply_state *state,
3499 struct image *image,
3500 struct patch *patch)
3502 struct strbuf buf = STRBUF_INIT;
3503 int status, pos;
3504 size_t len;
3505 char *img;
3506 struct stat st;
3507 struct cache_entry *ce;
3508 char *name = patch->new_name;
3509 unsigned mode = patch->new_mode;
3511 if (!patch->is_new)
3512 die("BUG: patch to %s is not a creation", patch->old_name);
3514 pos = cache_name_pos(name, strlen(name));
3515 if (pos < 0)
3516 return error(_("%s: does not exist in index"), name);
3517 ce = active_cache[pos];
3518 if (lstat(name, &st)) {
3519 if (errno != ENOENT)
3520 return error_errno("%s", name);
3521 if (checkout_target(&the_index, ce, &st))
3522 return -1;
3524 if (verify_index_match(ce, &st))
3525 return error(_("%s: does not match index"), name);
3527 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3528 if (status < 0)
3529 return status;
3530 else if (status)
3531 return -1;
3532 img = strbuf_detach(&buf, &len);
3533 prepare_image(image, img, len, !patch->is_binary);
3534 return 0;
3537 static int try_threeway(struct apply_state *state,
3538 struct image *image,
3539 struct patch *patch,
3540 struct stat *st,
3541 const struct cache_entry *ce)
3543 struct object_id pre_oid, post_oid, our_oid;
3544 struct strbuf buf = STRBUF_INIT;
3545 size_t len;
3546 int status;
3547 char *img;
3548 struct image tmp_image;
3550 /* No point falling back to 3-way merge in these cases */
3551 if (patch->is_delete ||
3552 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3553 return -1;
3555 /* Preimage the patch was prepared for */
3556 if (patch->is_new)
3557 write_sha1_file("", 0, blob_type, pre_oid.hash);
3558 else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3559 read_blob_object(&buf, &pre_oid, patch->old_mode))
3560 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3562 if (state->apply_verbosity > verbosity_silent)
3563 fprintf(stderr, _("Falling back to three-way merge...\n"));
3565 img = strbuf_detach(&buf, &len);
3566 prepare_image(&tmp_image, img, len, 1);
3567 /* Apply the patch to get the post image */
3568 if (apply_fragments(state, &tmp_image, patch) < 0) {
3569 clear_image(&tmp_image);
3570 return -1;
3572 /* post_oid is theirs */
3573 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3574 clear_image(&tmp_image);
3576 /* our_oid is ours */
3577 if (patch->is_new) {
3578 if (load_current(state, &tmp_image, patch))
3579 return error(_("cannot read the current contents of '%s'"),
3580 patch->new_name);
3581 } else {
3582 if (load_preimage(state, &tmp_image, patch, st, ce))
3583 return error(_("cannot read the current contents of '%s'"),
3584 patch->old_name);
3586 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3587 clear_image(&tmp_image);
3589 /* in-core three-way merge between post and our using pre as base */
3590 status = three_way_merge(image, patch->new_name,
3591 &pre_oid, &our_oid, &post_oid);
3592 if (status < 0) {
3593 if (state->apply_verbosity > verbosity_silent)
3594 fprintf(stderr,
3595 _("Failed to fall back on three-way merge...\n"));
3596 return status;
3599 if (status) {
3600 patch->conflicted_threeway = 1;
3601 if (patch->is_new)
3602 oidclr(&patch->threeway_stage[0]);
3603 else
3604 oidcpy(&patch->threeway_stage[0], &pre_oid);
3605 oidcpy(&patch->threeway_stage[1], &our_oid);
3606 oidcpy(&patch->threeway_stage[2], &post_oid);
3607 if (state->apply_verbosity > verbosity_silent)
3608 fprintf(stderr,
3609 _("Applied patch to '%s' with conflicts.\n"),
3610 patch->new_name);
3611 } else {
3612 if (state->apply_verbosity > verbosity_silent)
3613 fprintf(stderr,
3614 _("Applied patch to '%s' cleanly.\n"),
3615 patch->new_name);
3617 return 0;
3620 static int apply_data(struct apply_state *state, struct patch *patch,
3621 struct stat *st, const struct cache_entry *ce)
3623 struct image image;
3625 if (load_preimage(state, &image, patch, st, ce) < 0)
3626 return -1;
3628 if (patch->direct_to_threeway ||
3629 apply_fragments(state, &image, patch) < 0) {
3630 /* Note: with --reject, apply_fragments() returns 0 */
3631 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3632 return -1;
3634 patch->result = image.buf;
3635 patch->resultsize = image.len;
3636 add_to_fn_table(state, patch);
3637 free(image.line_allocated);
3639 if (0 < patch->is_delete && patch->resultsize)
3640 return error(_("removal patch leaves file contents"));
3642 return 0;
3646 * If "patch" that we are looking at modifies or deletes what we have,
3647 * we would want it not to lose any local modification we have, either
3648 * in the working tree or in the index.
3650 * This also decides if a non-git patch is a creation patch or a
3651 * modification to an existing empty file. We do not check the state
3652 * of the current tree for a creation patch in this function; the caller
3653 * check_patch() separately makes sure (and errors out otherwise) that
3654 * the path the patch creates does not exist in the current tree.
3656 static int check_preimage(struct apply_state *state,
3657 struct patch *patch,
3658 struct cache_entry **ce,
3659 struct stat *st)
3661 const char *old_name = patch->old_name;
3662 struct patch *previous = NULL;
3663 int stat_ret = 0, status;
3664 unsigned st_mode = 0;
3666 if (!old_name)
3667 return 0;
3669 assert(patch->is_new <= 0);
3670 previous = previous_patch(state, patch, &status);
3672 if (status)
3673 return error(_("path %s has been renamed/deleted"), old_name);
3674 if (previous) {
3675 st_mode = previous->new_mode;
3676 } else if (!state->cached) {
3677 stat_ret = lstat(old_name, st);
3678 if (stat_ret && errno != ENOENT)
3679 return error_errno("%s", old_name);
3682 if (state->check_index && !previous) {
3683 int pos = cache_name_pos(old_name, strlen(old_name));
3684 if (pos < 0) {
3685 if (patch->is_new < 0)
3686 goto is_new;
3687 return error(_("%s: does not exist in index"), old_name);
3689 *ce = active_cache[pos];
3690 if (stat_ret < 0) {
3691 if (checkout_target(&the_index, *ce, st))
3692 return -1;
3694 if (!state->cached && verify_index_match(*ce, st))
3695 return error(_("%s: does not match index"), old_name);
3696 if (state->cached)
3697 st_mode = (*ce)->ce_mode;
3698 } else if (stat_ret < 0) {
3699 if (patch->is_new < 0)
3700 goto is_new;
3701 return error_errno("%s", old_name);
3704 if (!state->cached && !previous)
3705 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3707 if (patch->is_new < 0)
3708 patch->is_new = 0;
3709 if (!patch->old_mode)
3710 patch->old_mode = st_mode;
3711 if ((st_mode ^ patch->old_mode) & S_IFMT)
3712 return error(_("%s: wrong type"), old_name);
3713 if (st_mode != patch->old_mode)
3714 warning(_("%s has type %o, expected %o"),
3715 old_name, st_mode, patch->old_mode);
3716 if (!patch->new_mode && !patch->is_delete)
3717 patch->new_mode = st_mode;
3718 return 0;
3720 is_new:
3721 patch->is_new = 1;
3722 patch->is_delete = 0;
3723 FREE_AND_NULL(patch->old_name);
3724 return 0;
3728 #define EXISTS_IN_INDEX 1
3729 #define EXISTS_IN_WORKTREE 2
3731 static int check_to_create(struct apply_state *state,
3732 const char *new_name,
3733 int ok_if_exists)
3735 struct stat nst;
3737 if (state->check_index &&
3738 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3739 !ok_if_exists)
3740 return EXISTS_IN_INDEX;
3741 if (state->cached)
3742 return 0;
3744 if (!lstat(new_name, &nst)) {
3745 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3746 return 0;
3748 * A leading component of new_name might be a symlink
3749 * that is going to be removed with this patch, but
3750 * still pointing at somewhere that has the path.
3751 * In such a case, path "new_name" does not exist as
3752 * far as git is concerned.
3754 if (has_symlink_leading_path(new_name, strlen(new_name)))
3755 return 0;
3757 return EXISTS_IN_WORKTREE;
3758 } else if (!is_missing_file_error(errno)) {
3759 return error_errno("%s", new_name);
3761 return 0;
3764 static uintptr_t register_symlink_changes(struct apply_state *state,
3765 const char *path,
3766 uintptr_t what)
3768 struct string_list_item *ent;
3770 ent = string_list_lookup(&state->symlink_changes, path);
3771 if (!ent) {
3772 ent = string_list_insert(&state->symlink_changes, path);
3773 ent->util = (void *)0;
3775 ent->util = (void *)(what | ((uintptr_t)ent->util));
3776 return (uintptr_t)ent->util;
3779 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3781 struct string_list_item *ent;
3783 ent = string_list_lookup(&state->symlink_changes, path);
3784 if (!ent)
3785 return 0;
3786 return (uintptr_t)ent->util;
3789 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3791 for ( ; patch; patch = patch->next) {
3792 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3793 (patch->is_rename || patch->is_delete))
3794 /* the symlink at patch->old_name is removed */
3795 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3797 if (patch->new_name && S_ISLNK(patch->new_mode))
3798 /* the symlink at patch->new_name is created or remains */
3799 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3803 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3805 do {
3806 unsigned int change;
3808 while (--name->len && name->buf[name->len] != '/')
3809 ; /* scan backwards */
3810 if (!name->len)
3811 break;
3812 name->buf[name->len] = '\0';
3813 change = check_symlink_changes(state, name->buf);
3814 if (change & APPLY_SYMLINK_IN_RESULT)
3815 return 1;
3816 if (change & APPLY_SYMLINK_GOES_AWAY)
3818 * This cannot be "return 0", because we may
3819 * see a new one created at a higher level.
3821 continue;
3823 /* otherwise, check the preimage */
3824 if (state->check_index) {
3825 struct cache_entry *ce;
3827 ce = cache_file_exists(name->buf, name->len, ignore_case);
3828 if (ce && S_ISLNK(ce->ce_mode))
3829 return 1;
3830 } else {
3831 struct stat st;
3832 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3833 return 1;
3835 } while (1);
3836 return 0;
3839 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3841 int ret;
3842 struct strbuf name = STRBUF_INIT;
3844 assert(*name_ != '\0');
3845 strbuf_addstr(&name, name_);
3846 ret = path_is_beyond_symlink_1(state, &name);
3847 strbuf_release(&name);
3849 return ret;
3852 static int check_unsafe_path(struct patch *patch)
3854 const char *old_name = NULL;
3855 const char *new_name = NULL;
3856 if (patch->is_delete)
3857 old_name = patch->old_name;
3858 else if (!patch->is_new && !patch->is_copy)
3859 old_name = patch->old_name;
3860 if (!patch->is_delete)
3861 new_name = patch->new_name;
3863 if (old_name && !verify_path(old_name))
3864 return error(_("invalid path '%s'"), old_name);
3865 if (new_name && !verify_path(new_name))
3866 return error(_("invalid path '%s'"), new_name);
3867 return 0;
3871 * Check and apply the patch in-core; leave the result in patch->result
3872 * for the caller to write it out to the final destination.
3874 static int check_patch(struct apply_state *state, struct patch *patch)
3876 struct stat st;
3877 const char *old_name = patch->old_name;
3878 const char *new_name = patch->new_name;
3879 const char *name = old_name ? old_name : new_name;
3880 struct cache_entry *ce = NULL;
3881 struct patch *tpatch;
3882 int ok_if_exists;
3883 int status;
3885 patch->rejected = 1; /* we will drop this after we succeed */
3887 status = check_preimage(state, patch, &ce, &st);
3888 if (status)
3889 return status;
3890 old_name = patch->old_name;
3893 * A type-change diff is always split into a patch to delete
3894 * old, immediately followed by a patch to create new (see
3895 * diff.c::run_diff()); in such a case it is Ok that the entry
3896 * to be deleted by the previous patch is still in the working
3897 * tree and in the index.
3899 * A patch to swap-rename between A and B would first rename A
3900 * to B and then rename B to A. While applying the first one,
3901 * the presence of B should not stop A from getting renamed to
3902 * B; ask to_be_deleted() about the later rename. Removal of
3903 * B and rename from A to B is handled the same way by asking
3904 * was_deleted().
3906 if ((tpatch = in_fn_table(state, new_name)) &&
3907 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3908 ok_if_exists = 1;
3909 else
3910 ok_if_exists = 0;
3912 if (new_name &&
3913 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3914 int err = check_to_create(state, new_name, ok_if_exists);
3916 if (err && state->threeway) {
3917 patch->direct_to_threeway = 1;
3918 } else switch (err) {
3919 case 0:
3920 break; /* happy */
3921 case EXISTS_IN_INDEX:
3922 return error(_("%s: already exists in index"), new_name);
3923 break;
3924 case EXISTS_IN_WORKTREE:
3925 return error(_("%s: already exists in working directory"),
3926 new_name);
3927 default:
3928 return err;
3931 if (!patch->new_mode) {
3932 if (0 < patch->is_new)
3933 patch->new_mode = S_IFREG | 0644;
3934 else
3935 patch->new_mode = patch->old_mode;
3939 if (new_name && old_name) {
3940 int same = !strcmp(old_name, new_name);
3941 if (!patch->new_mode)
3942 patch->new_mode = patch->old_mode;
3943 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3944 if (same)
3945 return error(_("new mode (%o) of %s does not "
3946 "match old mode (%o)"),
3947 patch->new_mode, new_name,
3948 patch->old_mode);
3949 else
3950 return error(_("new mode (%o) of %s does not "
3951 "match old mode (%o) of %s"),
3952 patch->new_mode, new_name,
3953 patch->old_mode, old_name);
3957 if (!state->unsafe_paths && check_unsafe_path(patch))
3958 return -128;
3961 * An attempt to read from or delete a path that is beyond a
3962 * symbolic link will be prevented by load_patch_target() that
3963 * is called at the beginning of apply_data() so we do not
3964 * have to worry about a patch marked with "is_delete" bit
3965 * here. We however need to make sure that the patch result
3966 * is not deposited to a path that is beyond a symbolic link
3967 * here.
3969 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3970 return error(_("affected file '%s' is beyond a symbolic link"),
3971 patch->new_name);
3973 if (apply_data(state, patch, &st, ce) < 0)
3974 return error(_("%s: patch does not apply"), name);
3975 patch->rejected = 0;
3976 return 0;
3979 static int check_patch_list(struct apply_state *state, struct patch *patch)
3981 int err = 0;
3983 prepare_symlink_changes(state, patch);
3984 prepare_fn_table(state, patch);
3985 while (patch) {
3986 int res;