Merge branch 'ag/rebase-p'
[git.git] / apply.c
blob959c45791008214cbc971493afaeb5f0e761f486
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->ita_only && (state->check_index || is_not_gitdir))
145 state->ita_only = 0;
146 if (state->check_index)
147 state->unsafe_paths = 0;
149 if (state->apply_verbosity <= verbosity_silent) {
150 state->saved_error_routine = get_error_routine();
151 state->saved_warn_routine = get_warn_routine();
152 set_error_routine(mute_routine);
153 set_warn_routine(mute_routine);
156 return 0;
159 static void set_default_whitespace_mode(struct apply_state *state)
161 if (!state->whitespace_option && !apply_default_whitespace)
162 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
166 * This represents one "hunk" from a patch, starting with
167 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
168 * patch text is pointed at by patch, and its byte length
169 * is stored in size. leading and trailing are the number
170 * of context lines.
172 struct fragment {
173 unsigned long leading, trailing;
174 unsigned long oldpos, oldlines;
175 unsigned long newpos, newlines;
177 * 'patch' is usually borrowed from buf in apply_patch(),
178 * but some codepaths store an allocated buffer.
180 const char *patch;
181 unsigned free_patch:1,
182 rejected:1;
183 int size;
184 int linenr;
185 struct fragment *next;
189 * When dealing with a binary patch, we reuse "leading" field
190 * to store the type of the binary hunk, either deflated "delta"
191 * or deflated "literal".
193 #define binary_patch_method leading
194 #define BINARY_DELTA_DEFLATED 1
195 #define BINARY_LITERAL_DEFLATED 2
198 * This represents a "patch" to a file, both metainfo changes
199 * such as creation/deletion, filemode and content changes represented
200 * as a series of fragments.
202 struct patch {
203 char *new_name, *old_name, *def_name;
204 unsigned int old_mode, new_mode;
205 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
206 int rejected;
207 unsigned ws_rule;
208 int lines_added, lines_deleted;
209 int score;
210 int extension_linenr; /* first line specifying delete/new/rename/copy */
211 unsigned int is_toplevel_relative:1;
212 unsigned int inaccurate_eof:1;
213 unsigned int is_binary:1;
214 unsigned int is_copy:1;
215 unsigned int is_rename:1;
216 unsigned int recount:1;
217 unsigned int conflicted_threeway:1;
218 unsigned int direct_to_threeway:1;
219 unsigned int crlf_in_old:1;
220 struct fragment *fragments;
221 char *result;
222 size_t resultsize;
223 char old_sha1_prefix[41];
224 char new_sha1_prefix[41];
225 struct patch *next;
227 /* three-way fallback result */
228 struct object_id threeway_stage[3];
231 static void free_fragment_list(struct fragment *list)
233 while (list) {
234 struct fragment *next = list->next;
235 if (list->free_patch)
236 free((char *)list->patch);
237 free(list);
238 list = next;
242 static void free_patch(struct patch *patch)
244 free_fragment_list(patch->fragments);
245 free(patch->def_name);
246 free(patch->old_name);
247 free(patch->new_name);
248 free(patch->result);
249 free(patch);
252 static void free_patch_list(struct patch *list)
254 while (list) {
255 struct patch *next = list->next;
256 free_patch(list);
257 list = next;
262 * A line in a file, len-bytes long (includes the terminating LF,
263 * except for an incomplete line at the end if the file ends with
264 * one), and its contents hashes to 'hash'.
266 struct line {
267 size_t len;
268 unsigned hash : 24;
269 unsigned flag : 8;
270 #define LINE_COMMON 1
271 #define LINE_PATCHED 2
275 * This represents a "file", which is an array of "lines".
277 struct image {
278 char *buf;
279 size_t len;
280 size_t nr;
281 size_t alloc;
282 struct line *line_allocated;
283 struct line *line;
286 static uint32_t hash_line(const char *cp, size_t len)
288 size_t i;
289 uint32_t h;
290 for (i = 0, h = 0; i < len; i++) {
291 if (!isspace(cp[i])) {
292 h = h * 3 + (cp[i] & 0xff);
295 return h;
299 * Compare lines s1 of length n1 and s2 of length n2, ignoring
300 * whitespace difference. Returns 1 if they match, 0 otherwise
302 static int fuzzy_matchlines(const char *s1, size_t n1,
303 const char *s2, size_t n2)
305 const char *end1 = s1 + n1;
306 const char *end2 = s2 + n2;
308 /* ignore line endings */
309 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
310 end1--;
311 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
312 end2--;
314 while (s1 < end1 && s2 < end2) {
315 if (isspace(*s1)) {
317 * Skip whitespace. We check on both buffers
318 * because we don't want "a b" to match "ab".
320 if (!isspace(*s2))
321 return 0;
322 while (s1 < end1 && isspace(*s1))
323 s1++;
324 while (s2 < end2 && isspace(*s2))
325 s2++;
326 } else if (*s1++ != *s2++)
327 return 0;
330 /* If we reached the end on one side only, lines don't match. */
331 return s1 == end1 && s2 == end2;
334 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
336 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
337 img->line_allocated[img->nr].len = len;
338 img->line_allocated[img->nr].hash = hash_line(bol, len);
339 img->line_allocated[img->nr].flag = flag;
340 img->nr++;
344 * "buf" has the file contents to be patched (read from various sources).
345 * attach it to "image" and add line-based index to it.
346 * "image" now owns the "buf".
348 static void prepare_image(struct image *image, char *buf, size_t len,
349 int prepare_linetable)
351 const char *cp, *ep;
353 memset(image, 0, sizeof(*image));
354 image->buf = buf;
355 image->len = len;
357 if (!prepare_linetable)
358 return;
360 ep = image->buf + image->len;
361 cp = image->buf;
362 while (cp < ep) {
363 const char *next;
364 for (next = cp; next < ep && *next != '\n'; next++)
366 if (next < ep)
367 next++;
368 add_line_info(image, cp, next - cp, 0);
369 cp = next;
371 image->line = image->line_allocated;
374 static void clear_image(struct image *image)
376 free(image->buf);
377 free(image->line_allocated);
378 memset(image, 0, sizeof(*image));
381 /* fmt must contain _one_ %s and no other substitution */
382 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
384 struct strbuf sb = STRBUF_INIT;
386 if (patch->old_name && patch->new_name &&
387 strcmp(patch->old_name, patch->new_name)) {
388 quote_c_style(patch->old_name, &sb, NULL, 0);
389 strbuf_addstr(&sb, " => ");
390 quote_c_style(patch->new_name, &sb, NULL, 0);
391 } else {
392 const char *n = patch->new_name;
393 if (!n)
394 n = patch->old_name;
395 quote_c_style(n, &sb, NULL, 0);
397 fprintf(output, fmt, sb.buf);
398 fputc('\n', output);
399 strbuf_release(&sb);
402 #define SLOP (16)
404 static int read_patch_file(struct strbuf *sb, int fd)
406 if (strbuf_read(sb, fd, 0) < 0)
407 return error_errno("git apply: failed to read");
410 * Make sure that we have some slop in the buffer
411 * so that we can do speculative "memcmp" etc, and
412 * see to it that it is NUL-filled.
414 strbuf_grow(sb, SLOP);
415 memset(sb->buf + sb->len, 0, SLOP);
416 return 0;
419 static unsigned long linelen(const char *buffer, unsigned long size)
421 unsigned long len = 0;
422 while (size--) {
423 len++;
424 if (*buffer++ == '\n')
425 break;
427 return len;
430 static int is_dev_null(const char *str)
432 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
435 #define TERM_SPACE 1
436 #define TERM_TAB 2
438 static int name_terminate(int c, int terminate)
440 if (c == ' ' && !(terminate & TERM_SPACE))
441 return 0;
442 if (c == '\t' && !(terminate & TERM_TAB))
443 return 0;
445 return 1;
448 /* remove double slashes to make --index work with such filenames */
449 static char *squash_slash(char *name)
451 int i = 0, j = 0;
453 if (!name)
454 return NULL;
456 while (name[i]) {
457 if ((name[j++] = name[i++]) == '/')
458 while (name[i] == '/')
459 i++;
461 name[j] = '\0';
462 return name;
465 static char *find_name_gnu(struct apply_state *state,
466 const char *line,
467 const char *def,
468 int p_value)
470 struct strbuf name = STRBUF_INIT;
471 char *cp;
474 * Proposed "new-style" GNU patch/diff format; see
475 * http://marc.info/?l=git&m=112927316408690&w=2
477 if (unquote_c_style(&name, line, NULL)) {
478 strbuf_release(&name);
479 return NULL;
482 for (cp = name.buf; p_value; p_value--) {
483 cp = strchr(cp, '/');
484 if (!cp) {
485 strbuf_release(&name);
486 return NULL;
488 cp++;
491 strbuf_remove(&name, 0, cp - name.buf);
492 if (state->root.len)
493 strbuf_insert(&name, 0, state->root.buf, state->root.len);
494 return squash_slash(strbuf_detach(&name, NULL));
497 static size_t sane_tz_len(const char *line, size_t len)
499 const char *tz, *p;
501 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
502 return 0;
503 tz = line + len - strlen(" +0500");
505 if (tz[1] != '+' && tz[1] != '-')
506 return 0;
508 for (p = tz + 2; p != line + len; p++)
509 if (!isdigit(*p))
510 return 0;
512 return line + len - tz;
515 static size_t tz_with_colon_len(const char *line, size_t len)
517 const char *tz, *p;
519 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
520 return 0;
521 tz = line + len - strlen(" +08:00");
523 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
524 return 0;
525 p = tz + 2;
526 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
527 !isdigit(*p++) || !isdigit(*p++))
528 return 0;
530 return line + len - tz;
533 static size_t date_len(const char *line, size_t len)
535 const char *date, *p;
537 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
538 return 0;
539 p = date = line + len - strlen("72-02-05");
541 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
542 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
543 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
544 return 0;
546 if (date - line >= strlen("19") &&
547 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
548 date -= strlen("19");
550 return line + len - date;
553 static size_t short_time_len(const char *line, size_t len)
555 const char *time, *p;
557 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
558 return 0;
559 p = time = line + len - strlen(" 07:01:32");
561 /* Permit 1-digit hours? */
562 if (*p++ != ' ' ||
563 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
565 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
566 return 0;
568 return line + len - time;
571 static size_t fractional_time_len(const char *line, size_t len)
573 const char *p;
574 size_t n;
576 /* Expected format: 19:41:17.620000023 */
577 if (!len || !isdigit(line[len - 1]))
578 return 0;
579 p = line + len - 1;
581 /* Fractional seconds. */
582 while (p > line && isdigit(*p))
583 p--;
584 if (*p != '.')
585 return 0;
587 /* Hours, minutes, and whole seconds. */
588 n = short_time_len(line, p - line);
589 if (!n)
590 return 0;
592 return line + len - p + n;
595 static size_t trailing_spaces_len(const char *line, size_t len)
597 const char *p;
599 /* Expected format: ' ' x (1 or more) */
600 if (!len || line[len - 1] != ' ')
601 return 0;
603 p = line + len;
604 while (p != line) {
605 p--;
606 if (*p != ' ')
607 return line + len - (p + 1);
610 /* All spaces! */
611 return len;
614 static size_t diff_timestamp_len(const char *line, size_t len)
616 const char *end = line + len;
617 size_t n;
620 * Posix: 2010-07-05 19:41:17
621 * GNU: 2010-07-05 19:41:17.620000023 -0500
624 if (!isdigit(end[-1]))
625 return 0;
627 n = sane_tz_len(line, end - line);
628 if (!n)
629 n = tz_with_colon_len(line, end - line);
630 end -= n;
632 n = short_time_len(line, end - line);
633 if (!n)
634 n = fractional_time_len(line, end - line);
635 end -= n;
637 n = date_len(line, end - line);
638 if (!n) /* No date. Too bad. */
639 return 0;
640 end -= n;
642 if (end == line) /* No space before date. */
643 return 0;
644 if (end[-1] == '\t') { /* Success! */
645 end--;
646 return line + len - end;
648 if (end[-1] != ' ') /* No space before date. */
649 return 0;
651 /* Whitespace damage. */
652 end -= trailing_spaces_len(line, end - line);
653 return line + len - end;
656 static char *find_name_common(struct apply_state *state,
657 const char *line,
658 const char *def,
659 int p_value,
660 const char *end,
661 int terminate)
663 int len;
664 const char *start = NULL;
666 if (p_value == 0)
667 start = line;
668 while (line != end) {
669 char c = *line;
671 if (!end && isspace(c)) {
672 if (c == '\n')
673 break;
674 if (name_terminate(c, terminate))
675 break;
677 line++;
678 if (c == '/' && !--p_value)
679 start = line;
681 if (!start)
682 return squash_slash(xstrdup_or_null(def));
683 len = line - start;
684 if (!len)
685 return squash_slash(xstrdup_or_null(def));
688 * Generally we prefer the shorter name, especially
689 * if the other one is just a variation of that with
690 * something else tacked on to the end (ie "file.orig"
691 * or "file~").
693 if (def) {
694 int deflen = strlen(def);
695 if (deflen < len && !strncmp(start, def, deflen))
696 return squash_slash(xstrdup(def));
699 if (state->root.len) {
700 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
701 return squash_slash(ret);
704 return squash_slash(xmemdupz(start, len));
707 static char *find_name(struct apply_state *state,
708 const char *line,
709 char *def,
710 int p_value,
711 int terminate)
713 if (*line == '"') {
714 char *name = find_name_gnu(state, line, def, p_value);
715 if (name)
716 return name;
719 return find_name_common(state, line, def, p_value, NULL, terminate);
722 static char *find_name_traditional(struct apply_state *state,
723 const char *line,
724 char *def,
725 int p_value)
727 size_t len;
728 size_t date_len;
730 if (*line == '"') {
731 char *name = find_name_gnu(state, line, def, p_value);
732 if (name)
733 return name;
736 len = strchrnul(line, '\n') - line;
737 date_len = diff_timestamp_len(line, len);
738 if (!date_len)
739 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
740 len -= date_len;
742 return find_name_common(state, line, def, p_value, line + len, 0);
746 * Given the string after "--- " or "+++ ", guess the appropriate
747 * p_value for the given patch.
749 static int guess_p_value(struct apply_state *state, const char *nameline)
751 char *name, *cp;
752 int val = -1;
754 if (is_dev_null(nameline))
755 return -1;
756 name = find_name_traditional(state, nameline, NULL, 0);
757 if (!name)
758 return -1;
759 cp = strchr(name, '/');
760 if (!cp)
761 val = 0;
762 else if (state->prefix) {
764 * Does it begin with "a/$our-prefix" and such? Then this is
765 * very likely to apply to our directory.
767 if (starts_with(name, state->prefix))
768 val = count_slashes(state->prefix);
769 else {
770 cp++;
771 if (starts_with(cp, state->prefix))
772 val = count_slashes(state->prefix) + 1;
775 free(name);
776 return val;
780 * Does the ---/+++ line have the POSIX timestamp after the last HT?
781 * GNU diff puts epoch there to signal a creation/deletion event. Is
782 * this such a timestamp?
784 static int has_epoch_timestamp(const char *nameline)
787 * We are only interested in epoch timestamp; any non-zero
788 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
789 * For the same reason, the date must be either 1969-12-31 or
790 * 1970-01-01, and the seconds part must be "00".
792 const char stamp_regexp[] =
793 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
795 "([-+][0-2][0-9]:?[0-5][0-9])\n";
796 const char *timestamp = NULL, *cp, *colon;
797 static regex_t *stamp;
798 regmatch_t m[10];
799 int zoneoffset, epoch_hour, hour, minute;
800 int status;
802 for (cp = nameline; *cp != '\n'; cp++) {
803 if (*cp == '\t')
804 timestamp = cp + 1;
806 if (!timestamp)
807 return 0;
810 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
811 * (west of GMT) or 1970-01-01 (east of GMT)
813 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
814 epoch_hour = 24;
815 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
816 epoch_hour = 0;
817 else
818 return 0;
820 if (!stamp) {
821 stamp = xmalloc(sizeof(*stamp));
822 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
823 warning(_("Cannot prepare timestamp regexp %s"),
824 stamp_regexp);
825 return 0;
829 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
830 if (status) {
831 if (status != REG_NOMATCH)
832 warning(_("regexec returned %d for input: %s"),
833 status, timestamp);
834 return 0;
837 hour = strtol(timestamp, NULL, 10);
838 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
840 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
841 if (*colon == ':')
842 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
843 else
844 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
845 if (timestamp[m[3].rm_so] == '-')
846 zoneoffset = -zoneoffset;
848 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
852 * Get the name etc info from the ---/+++ lines of a traditional patch header
854 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
855 * files, we can happily check the index for a match, but for creating a
856 * new file we should try to match whatever "patch" does. I have no idea.
858 static int parse_traditional_patch(struct apply_state *state,
859 const char *first,
860 const char *second,
861 struct patch *patch)
863 char *name;
865 first += 4; /* skip "--- " */
866 second += 4; /* skip "+++ " */
867 if (!state->p_value_known) {
868 int p, q;
869 p = guess_p_value(state, first);
870 q = guess_p_value(state, second);
871 if (p < 0) p = q;
872 if (0 <= p && p == q) {
873 state->p_value = p;
874 state->p_value_known = 1;
877 if (is_dev_null(first)) {
878 patch->is_new = 1;
879 patch->is_delete = 0;
880 name = find_name_traditional(state, second, NULL, state->p_value);
881 patch->new_name = name;
882 } else if (is_dev_null(second)) {
883 patch->is_new = 0;
884 patch->is_delete = 1;
885 name = find_name_traditional(state, first, NULL, state->p_value);
886 patch->old_name = name;
887 } else {
888 char *first_name;
889 first_name = find_name_traditional(state, first, NULL, state->p_value);
890 name = find_name_traditional(state, second, first_name, state->p_value);
891 free(first_name);
892 if (has_epoch_timestamp(first)) {
893 patch->is_new = 1;
894 patch->is_delete = 0;
895 patch->new_name = name;
896 } else if (has_epoch_timestamp(second)) {
897 patch->is_new = 0;
898 patch->is_delete = 1;
899 patch->old_name = name;
900 } else {
901 patch->old_name = name;
902 patch->new_name = xstrdup_or_null(name);
905 if (!name)
906 return error(_("unable to find filename in patch at line %d"), state->linenr);
908 return 0;
911 static int gitdiff_hdrend(struct apply_state *state,
912 const char *line,
913 struct patch *patch)
915 return 1;
919 * We're anal about diff header consistency, to make
920 * sure that we don't end up having strange ambiguous
921 * patches floating around.
923 * As a result, gitdiff_{old|new}name() will check
924 * their names against any previous information, just
925 * to make sure..
927 #define DIFF_OLD_NAME 0
928 #define DIFF_NEW_NAME 1
930 static int gitdiff_verify_name(struct apply_state *state,
931 const char *line,
932 int isnull,
933 char **name,
934 int side)
936 if (!*name && !isnull) {
937 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
938 return 0;
941 if (*name) {
942 char *another;
943 if (isnull)
944 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
945 *name, state->linenr);
946 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
947 if (!another || strcmp(another, *name)) {
948 free(another);
949 return error((side == DIFF_NEW_NAME) ?
950 _("git apply: bad git-diff - inconsistent new filename on line %d") :
951 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
953 free(another);
954 } else {
955 if (!is_dev_null(line))
956 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
959 return 0;
962 static int gitdiff_oldname(struct apply_state *state,
963 const char *line,
964 struct patch *patch)
966 return gitdiff_verify_name(state, line,
967 patch->is_new, &patch->old_name,
968 DIFF_OLD_NAME);
971 static int gitdiff_newname(struct apply_state *state,
972 const char *line,
973 struct patch *patch)
975 return gitdiff_verify_name(state, line,
976 patch->is_delete, &patch->new_name,
977 DIFF_NEW_NAME);
980 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
982 char *end;
983 *mode = strtoul(line, &end, 8);
984 if (end == line || !isspace(*end))
985 return error(_("invalid mode on line %d: %s"), linenr, line);
986 return 0;
989 static int gitdiff_oldmode(struct apply_state *state,
990 const char *line,
991 struct patch *patch)
993 return parse_mode_line(line, state->linenr, &patch->old_mode);
996 static int gitdiff_newmode(struct apply_state *state,
997 const char *line,
998 struct patch *patch)
1000 return parse_mode_line(line, state->linenr, &patch->new_mode);
1003 static int gitdiff_delete(struct apply_state *state,
1004 const char *line,
1005 struct patch *patch)
1007 patch->is_delete = 1;
1008 free(patch->old_name);
1009 patch->old_name = xstrdup_or_null(patch->def_name);
1010 return gitdiff_oldmode(state, line, patch);
1013 static int gitdiff_newfile(struct apply_state *state,
1014 const char *line,
1015 struct patch *patch)
1017 patch->is_new = 1;
1018 free(patch->new_name);
1019 patch->new_name = xstrdup_or_null(patch->def_name);
1020 return gitdiff_newmode(state, line, patch);
1023 static int gitdiff_copysrc(struct apply_state *state,
1024 const char *line,
1025 struct patch *patch)
1027 patch->is_copy = 1;
1028 free(patch->old_name);
1029 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1030 return 0;
1033 static int gitdiff_copydst(struct apply_state *state,
1034 const char *line,
1035 struct patch *patch)
1037 patch->is_copy = 1;
1038 free(patch->new_name);
1039 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1040 return 0;
1043 static int gitdiff_renamesrc(struct apply_state *state,
1044 const char *line,
1045 struct patch *patch)
1047 patch->is_rename = 1;
1048 free(patch->old_name);
1049 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1050 return 0;
1053 static int gitdiff_renamedst(struct apply_state *state,
1054 const char *line,
1055 struct patch *patch)
1057 patch->is_rename = 1;
1058 free(patch->new_name);
1059 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1060 return 0;
1063 static int gitdiff_similarity(struct apply_state *state,
1064 const char *line,
1065 struct patch *patch)
1067 unsigned long val = strtoul(line, NULL, 10);
1068 if (val <= 100)
1069 patch->score = val;
1070 return 0;
1073 static int gitdiff_dissimilarity(struct apply_state *state,
1074 const char *line,
1075 struct patch *patch)
1077 unsigned long val = strtoul(line, NULL, 10);
1078 if (val <= 100)
1079 patch->score = val;
1080 return 0;
1083 static int gitdiff_index(struct apply_state *state,
1084 const char *line,
1085 struct patch *patch)
1088 * index line is N hexadecimal, "..", N hexadecimal,
1089 * and optional space with octal mode.
1091 const char *ptr, *eol;
1092 int len;
1094 ptr = strchr(line, '.');
1095 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1096 return 0;
1097 len = ptr - line;
1098 memcpy(patch->old_sha1_prefix, line, len);
1099 patch->old_sha1_prefix[len] = 0;
1101 line = ptr + 2;
1102 ptr = strchr(line, ' ');
1103 eol = strchrnul(line, '\n');
1105 if (!ptr || eol < ptr)
1106 ptr = eol;
1107 len = ptr - line;
1109 if (40 < len)
1110 return 0;
1111 memcpy(patch->new_sha1_prefix, line, len);
1112 patch->new_sha1_prefix[len] = 0;
1113 if (*ptr == ' ')
1114 return gitdiff_oldmode(state, ptr + 1, patch);
1115 return 0;
1119 * This is normal for a diff that doesn't change anything: we'll fall through
1120 * into the next diff. Tell the parser to break out.
1122 static int gitdiff_unrecognized(struct apply_state *state,
1123 const char *line,
1124 struct patch *patch)
1126 return 1;
1130 * Skip p_value leading components from "line"; as we do not accept
1131 * absolute paths, return NULL in that case.
1133 static const char *skip_tree_prefix(struct apply_state *state,
1134 const char *line,
1135 int llen)
1137 int nslash;
1138 int i;
1140 if (!state->p_value)
1141 return (llen && line[0] == '/') ? NULL : line;
1143 nslash = state->p_value;
1144 for (i = 0; i < llen; i++) {
1145 int ch = line[i];
1146 if (ch == '/' && --nslash <= 0)
1147 return (i == 0) ? NULL : &line[i + 1];
1149 return NULL;
1153 * This is to extract the same name that appears on "diff --git"
1154 * line. We do not find and return anything if it is a rename
1155 * patch, and it is OK because we will find the name elsewhere.
1156 * We need to reliably find name only when it is mode-change only,
1157 * creation or deletion of an empty file. In any of these cases,
1158 * both sides are the same name under a/ and b/ respectively.
1160 static char *git_header_name(struct apply_state *state,
1161 const char *line,
1162 int llen)
1164 const char *name;
1165 const char *second = NULL;
1166 size_t len, line_len;
1168 line += strlen("diff --git ");
1169 llen -= strlen("diff --git ");
1171 if (*line == '"') {
1172 const char *cp;
1173 struct strbuf first = STRBUF_INIT;
1174 struct strbuf sp = STRBUF_INIT;
1176 if (unquote_c_style(&first, line, &second))
1177 goto free_and_fail1;
1179 /* strip the a/b prefix including trailing slash */
1180 cp = skip_tree_prefix(state, first.buf, first.len);
1181 if (!cp)
1182 goto free_and_fail1;
1183 strbuf_remove(&first, 0, cp - first.buf);
1186 * second points at one past closing dq of name.
1187 * find the second name.
1189 while ((second < line + llen) && isspace(*second))
1190 second++;
1192 if (line + llen <= second)
1193 goto free_and_fail1;
1194 if (*second == '"') {
1195 if (unquote_c_style(&sp, second, NULL))
1196 goto free_and_fail1;
1197 cp = skip_tree_prefix(state, sp.buf, sp.len);
1198 if (!cp)
1199 goto free_and_fail1;
1200 /* They must match, otherwise ignore */
1201 if (strcmp(cp, first.buf))
1202 goto free_and_fail1;
1203 strbuf_release(&sp);
1204 return strbuf_detach(&first, NULL);
1207 /* unquoted second */
1208 cp = skip_tree_prefix(state, second, line + llen - second);
1209 if (!cp)
1210 goto free_and_fail1;
1211 if (line + llen - cp != first.len ||
1212 memcmp(first.buf, cp, first.len))
1213 goto free_and_fail1;
1214 return strbuf_detach(&first, NULL);
1216 free_and_fail1:
1217 strbuf_release(&first);
1218 strbuf_release(&sp);
1219 return NULL;
1222 /* unquoted first name */
1223 name = skip_tree_prefix(state, line, llen);
1224 if (!name)
1225 return NULL;
1228 * since the first name is unquoted, a dq if exists must be
1229 * the beginning of the second name.
1231 for (second = name; second < line + llen; second++) {
1232 if (*second == '"') {
1233 struct strbuf sp = STRBUF_INIT;
1234 const char *np;
1236 if (unquote_c_style(&sp, second, NULL))
1237 goto free_and_fail2;
1239 np = skip_tree_prefix(state, sp.buf, sp.len);
1240 if (!np)
1241 goto free_and_fail2;
1243 len = sp.buf + sp.len - np;
1244 if (len < second - name &&
1245 !strncmp(np, name, len) &&
1246 isspace(name[len])) {
1247 /* Good */
1248 strbuf_remove(&sp, 0, np - sp.buf);
1249 return strbuf_detach(&sp, NULL);
1252 free_and_fail2:
1253 strbuf_release(&sp);
1254 return NULL;
1259 * Accept a name only if it shows up twice, exactly the same
1260 * form.
1262 second = strchr(name, '\n');
1263 if (!second)
1264 return NULL;
1265 line_len = second - name;
1266 for (len = 0 ; ; len++) {
1267 switch (name[len]) {
1268 default:
1269 continue;
1270 case '\n':
1271 return NULL;
1272 case '\t': case ' ':
1274 * Is this the separator between the preimage
1275 * and the postimage pathname? Again, we are
1276 * only interested in the case where there is
1277 * no rename, as this is only to set def_name
1278 * and a rename patch has the names elsewhere
1279 * in an unambiguous form.
1281 if (!name[len + 1])
1282 return NULL; /* no postimage name */
1283 second = skip_tree_prefix(state, name + len + 1,
1284 line_len - (len + 1));
1285 if (!second)
1286 return NULL;
1288 * Does len bytes starting at "name" and "second"
1289 * (that are separated by one HT or SP we just
1290 * found) exactly match?
1292 if (second[len] == '\n' && !strncmp(name, second, len))
1293 return xmemdupz(name, len);
1298 static int check_header_line(struct apply_state *state, struct patch *patch)
1300 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1301 (patch->is_rename == 1) + (patch->is_copy == 1);
1302 if (extensions > 1)
1303 return error(_("inconsistent header lines %d and %d"),
1304 patch->extension_linenr, state->linenr);
1305 if (extensions && !patch->extension_linenr)
1306 patch->extension_linenr = state->linenr;
1307 return 0;
1310 /* Verify that we recognize the lines following a git header */
1311 static int parse_git_header(struct apply_state *state,
1312 const char *line,
1313 int len,
1314 unsigned int size,
1315 struct patch *patch)
1317 unsigned long offset;
1319 /* A git diff has explicit new/delete information, so we don't guess */
1320 patch->is_new = 0;
1321 patch->is_delete = 0;
1324 * Some things may not have the old name in the
1325 * rest of the headers anywhere (pure mode changes,
1326 * or removing or adding empty files), so we get
1327 * the default name from the header.
1329 patch->def_name = git_header_name(state, line, len);
1330 if (patch->def_name && state->root.len) {
1331 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1332 free(patch->def_name);
1333 patch->def_name = s;
1336 line += len;
1337 size -= len;
1338 state->linenr++;
1339 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1340 static const struct opentry {
1341 const char *str;
1342 int (*fn)(struct apply_state *, const char *, struct patch *);
1343 } optable[] = {
1344 { "@@ -", gitdiff_hdrend },
1345 { "--- ", gitdiff_oldname },
1346 { "+++ ", gitdiff_newname },
1347 { "old mode ", gitdiff_oldmode },
1348 { "new mode ", gitdiff_newmode },
1349 { "deleted file mode ", gitdiff_delete },
1350 { "new file mode ", gitdiff_newfile },
1351 { "copy from ", gitdiff_copysrc },
1352 { "copy to ", gitdiff_copydst },
1353 { "rename old ", gitdiff_renamesrc },
1354 { "rename new ", gitdiff_renamedst },
1355 { "rename from ", gitdiff_renamesrc },
1356 { "rename to ", gitdiff_renamedst },
1357 { "similarity index ", gitdiff_similarity },
1358 { "dissimilarity index ", gitdiff_dissimilarity },
1359 { "index ", gitdiff_index },
1360 { "", gitdiff_unrecognized },
1362 int i;
1364 len = linelen(line, size);
1365 if (!len || line[len-1] != '\n')
1366 break;
1367 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1368 const struct opentry *p = optable + i;
1369 int oplen = strlen(p->str);
1370 int res;
1371 if (len < oplen || memcmp(p->str, line, oplen))
1372 continue;
1373 res = p->fn(state, line + oplen, patch);
1374 if (res < 0)
1375 return -1;
1376 if (check_header_line(state, patch))
1377 return -1;
1378 if (res > 0)
1379 return offset;
1380 break;
1384 return offset;
1387 static int parse_num(const char *line, unsigned long *p)
1389 char *ptr;
1391 if (!isdigit(*line))
1392 return 0;
1393 *p = strtoul(line, &ptr, 10);
1394 return ptr - line;
1397 static int parse_range(const char *line, int len, int offset, const char *expect,
1398 unsigned long *p1, unsigned long *p2)
1400 int digits, ex;
1402 if (offset < 0 || offset >= len)
1403 return -1;
1404 line += offset;
1405 len -= offset;
1407 digits = parse_num(line, p1);
1408 if (!digits)
1409 return -1;
1411 offset += digits;
1412 line += digits;
1413 len -= digits;
1415 *p2 = 1;
1416 if (*line == ',') {
1417 digits = parse_num(line+1, p2);
1418 if (!digits)
1419 return -1;
1421 offset += digits+1;
1422 line += digits+1;
1423 len -= digits+1;
1426 ex = strlen(expect);
1427 if (ex > len)
1428 return -1;
1429 if (memcmp(line, expect, ex))
1430 return -1;
1432 return offset + ex;
1435 static void recount_diff(const char *line, int size, struct fragment *fragment)
1437 int oldlines = 0, newlines = 0, ret = 0;
1439 if (size < 1) {
1440 warning("recount: ignore empty hunk");
1441 return;
1444 for (;;) {
1445 int len = linelen(line, size);
1446 size -= len;
1447 line += len;
1449 if (size < 1)
1450 break;
1452 switch (*line) {
1453 case ' ': case '\n':
1454 newlines++;
1455 /* fall through */
1456 case '-':
1457 oldlines++;
1458 continue;
1459 case '+':
1460 newlines++;
1461 continue;
1462 case '\\':
1463 continue;
1464 case '@':
1465 ret = size < 3 || !starts_with(line, "@@ ");
1466 break;
1467 case 'd':
1468 ret = size < 5 || !starts_with(line, "diff ");
1469 break;
1470 default:
1471 ret = -1;
1472 break;
1474 if (ret) {
1475 warning(_("recount: unexpected line: %.*s"),
1476 (int)linelen(line, size), line);
1477 return;
1479 break;
1481 fragment->oldlines = oldlines;
1482 fragment->newlines = newlines;
1486 * Parse a unified diff fragment header of the
1487 * form "@@ -a,b +c,d @@"
1489 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1491 int offset;
1493 if (!len || line[len-1] != '\n')
1494 return -1;
1496 /* Figure out the number of lines in a fragment */
1497 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1498 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1500 return offset;
1504 * Find file diff header
1506 * Returns:
1507 * -1 if no header was found
1508 * -128 in case of error
1509 * the size of the header in bytes (called "offset") otherwise
1511 static int find_header(struct apply_state *state,
1512 const char *line,
1513 unsigned long size,
1514 int *hdrsize,
1515 struct patch *patch)
1517 unsigned long offset, len;
1519 patch->is_toplevel_relative = 0;
1520 patch->is_rename = patch->is_copy = 0;
1521 patch->is_new = patch->is_delete = -1;
1522 patch->old_mode = patch->new_mode = 0;
1523 patch->old_name = patch->new_name = NULL;
1524 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1525 unsigned long nextlen;
1527 len = linelen(line, size);
1528 if (!len)
1529 break;
1531 /* Testing this early allows us to take a few shortcuts.. */
1532 if (len < 6)
1533 continue;
1536 * Make sure we don't find any unconnected patch fragments.
1537 * That's a sign that we didn't find a header, and that a
1538 * patch has become corrupted/broken up.
1540 if (!memcmp("@@ -", line, 4)) {
1541 struct fragment dummy;
1542 if (parse_fragment_header(line, len, &dummy) < 0)
1543 continue;
1544 error(_("patch fragment without header at line %d: %.*s"),
1545 state->linenr, (int)len-1, line);
1546 return -128;
1549 if (size < len + 6)
1550 break;
1553 * Git patch? It might not have a real patch, just a rename
1554 * or mode change, so we handle that specially
1556 if (!memcmp("diff --git ", line, 11)) {
1557 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1558 if (git_hdr_len < 0)
1559 return -128;
1560 if (git_hdr_len <= len)
1561 continue;
1562 if (!patch->old_name && !patch->new_name) {
1563 if (!patch->def_name) {
1564 error(Q_("git diff header lacks filename information when removing "
1565 "%d leading pathname component (line %d)",
1566 "git diff header lacks filename information when removing "
1567 "%d leading pathname components (line %d)",
1568 state->p_value),
1569 state->p_value, state->linenr);
1570 return -128;
1572 patch->old_name = xstrdup(patch->def_name);
1573 patch->new_name = xstrdup(patch->def_name);
1575 if ((!patch->new_name && !patch->is_delete) ||
1576 (!patch->old_name && !patch->is_new)) {
1577 error(_("git diff header lacks filename information "
1578 "(line %d)"), state->linenr);
1579 return -128;
1581 patch->is_toplevel_relative = 1;
1582 *hdrsize = git_hdr_len;
1583 return offset;
1586 /* --- followed by +++ ? */
1587 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1588 continue;
1591 * We only accept unified patches, so we want it to
1592 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1593 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1595 nextlen = linelen(line + len, size - len);
1596 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1597 continue;
1599 /* Ok, we'll consider it a patch */
1600 if (parse_traditional_patch(state, line, line+len, patch))
1601 return -128;
1602 *hdrsize = len + nextlen;
1603 state->linenr += 2;
1604 return offset;
1606 return -1;
1609 static void record_ws_error(struct apply_state *state,
1610 unsigned result,
1611 const char *line,
1612 int len,
1613 int linenr)
1615 char *err;
1617 if (!result)
1618 return;
1620 state->whitespace_error++;
1621 if (state->squelch_whitespace_errors &&
1622 state->squelch_whitespace_errors < state->whitespace_error)
1623 return;
1625 err = whitespace_error_string(result);
1626 if (state->apply_verbosity > verbosity_silent)
1627 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1628 state->patch_input_file, linenr, err, len, line);
1629 free(err);
1632 static void check_whitespace(struct apply_state *state,
1633 const char *line,
1634 int len,
1635 unsigned ws_rule)
1637 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1639 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1643 * Check if the patch has context lines with CRLF or
1644 * the patch wants to remove lines with CRLF.
1646 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1648 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1649 patch->ws_rule |= WS_CR_AT_EOL;
1650 patch->crlf_in_old = 1;
1656 * Parse a unified diff. Note that this really needs to parse each
1657 * fragment separately, since the only way to know the difference
1658 * between a "---" that is part of a patch, and a "---" that starts
1659 * the next patch is to look at the line counts..
1661 static int parse_fragment(struct apply_state *state,
1662 const char *line,
1663 unsigned long size,
1664 struct patch *patch,
1665 struct fragment *fragment)
1667 int added, deleted;
1668 int len = linelen(line, size), offset;
1669 unsigned long oldlines, newlines;
1670 unsigned long leading, trailing;
1672 offset = parse_fragment_header(line, len, fragment);
1673 if (offset < 0)
1674 return -1;
1675 if (offset > 0 && patch->recount)
1676 recount_diff(line + offset, size - offset, fragment);
1677 oldlines = fragment->oldlines;
1678 newlines = fragment->newlines;
1679 leading = 0;
1680 trailing = 0;
1682 /* Parse the thing.. */
1683 line += len;
1684 size -= len;
1685 state->linenr++;
1686 added = deleted = 0;
1687 for (offset = len;
1688 0 < size;
1689 offset += len, size -= len, line += len, state->linenr++) {
1690 if (!oldlines && !newlines)
1691 break;
1692 len = linelen(line, size);
1693 if (!len || line[len-1] != '\n')
1694 return -1;
1695 switch (*line) {
1696 default:
1697 return -1;
1698 case '\n': /* newer GNU diff, an empty context line */
1699 case ' ':
1700 oldlines--;
1701 newlines--;
1702 if (!deleted && !added)
1703 leading++;
1704 trailing++;
1705 check_old_for_crlf(patch, line, len);
1706 if (!state->apply_in_reverse &&
1707 state->ws_error_action == correct_ws_error)
1708 check_whitespace(state, line, len, patch->ws_rule);
1709 break;
1710 case '-':
1711 if (!state->apply_in_reverse)
1712 check_old_for_crlf(patch, line, len);
1713 if (state->apply_in_reverse &&
1714 state->ws_error_action != nowarn_ws_error)
1715 check_whitespace(state, line, len, patch->ws_rule);
1716 deleted++;
1717 oldlines--;
1718 trailing = 0;
1719 break;
1720 case '+':
1721 if (state->apply_in_reverse)
1722 check_old_for_crlf(patch, line, len);
1723 if (!state->apply_in_reverse &&
1724 state->ws_error_action != nowarn_ws_error)
1725 check_whitespace(state, line, len, patch->ws_rule);
1726 added++;
1727 newlines--;
1728 trailing = 0;
1729 break;
1732 * We allow "\ No newline at end of file". Depending
1733 * on locale settings when the patch was produced we
1734 * don't know what this line looks like. The only
1735 * thing we do know is that it begins with "\ ".
1736 * Checking for 12 is just for sanity check -- any
1737 * l10n of "\ No newline..." is at least that long.
1739 case '\\':
1740 if (len < 12 || memcmp(line, "\\ ", 2))
1741 return -1;
1742 break;
1745 if (oldlines || newlines)
1746 return -1;
1747 if (!deleted && !added)
1748 return -1;
1750 fragment->leading = leading;
1751 fragment->trailing = trailing;
1754 * If a fragment ends with an incomplete line, we failed to include
1755 * it in the above loop because we hit oldlines == newlines == 0
1756 * before seeing it.
1758 if (12 < size && !memcmp(line, "\\ ", 2))
1759 offset += linelen(line, size);
1761 patch->lines_added += added;
1762 patch->lines_deleted += deleted;
1764 if (0 < patch->is_new && oldlines)
1765 return error(_("new file depends on old contents"));
1766 if (0 < patch->is_delete && newlines)
1767 return error(_("deleted file still has contents"));
1768 return offset;
1772 * We have seen "diff --git a/... b/..." header (or a traditional patch
1773 * header). Read hunks that belong to this patch into fragments and hang
1774 * them to the given patch structure.
1776 * The (fragment->patch, fragment->size) pair points into the memory given
1777 * by the caller, not a copy, when we return.
1779 * Returns:
1780 * -1 in case of error,
1781 * the number of bytes in the patch otherwise.
1783 static int parse_single_patch(struct apply_state *state,
1784 const char *line,
1785 unsigned long size,
1786 struct patch *patch)
1788 unsigned long offset = 0;
1789 unsigned long oldlines = 0, newlines = 0, context = 0;
1790 struct fragment **fragp = &patch->fragments;
1792 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1793 struct fragment *fragment;
1794 int len;
1796 fragment = xcalloc(1, sizeof(*fragment));
1797 fragment->linenr = state->linenr;
1798 len = parse_fragment(state, line, size, patch, fragment);
1799 if (len <= 0) {
1800 free(fragment);
1801 return error(_("corrupt patch at line %d"), state->linenr);
1803 fragment->patch = line;
1804 fragment->size = len;
1805 oldlines += fragment->oldlines;
1806 newlines += fragment->newlines;
1807 context += fragment->leading + fragment->trailing;
1809 *fragp = fragment;
1810 fragp = &fragment->next;
1812 offset += len;
1813 line += len;
1814 size -= len;
1818 * If something was removed (i.e. we have old-lines) it cannot
1819 * be creation, and if something was added it cannot be
1820 * deletion. However, the reverse is not true; --unified=0
1821 * patches that only add are not necessarily creation even
1822 * though they do not have any old lines, and ones that only
1823 * delete are not necessarily deletion.
1825 * Unfortunately, a real creation/deletion patch do _not_ have
1826 * any context line by definition, so we cannot safely tell it
1827 * apart with --unified=0 insanity. At least if the patch has
1828 * more than one hunk it is not creation or deletion.
1830 if (patch->is_new < 0 &&
1831 (oldlines || (patch->fragments && patch->fragments->next)))
1832 patch->is_new = 0;
1833 if (patch->is_delete < 0 &&
1834 (newlines || (patch->fragments && patch->fragments->next)))
1835 patch->is_delete = 0;
1837 if (0 < patch->is_new && oldlines)
1838 return error(_("new file %s depends on old contents"), patch->new_name);
1839 if (0 < patch->is_delete && newlines)
1840 return error(_("deleted file %s still has contents"), patch->old_name);
1841 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1842 fprintf_ln(stderr,
1843 _("** warning: "
1844 "file %s becomes empty but is not deleted"),
1845 patch->new_name);
1847 return offset;
1850 static inline int metadata_changes(struct patch *patch)
1852 return patch->is_rename > 0 ||
1853 patch->is_copy > 0 ||
1854 patch->is_new > 0 ||
1855 patch->is_delete ||
1856 (patch->old_mode && patch->new_mode &&
1857 patch->old_mode != patch->new_mode);
1860 static char *inflate_it(const void *data, unsigned long size,
1861 unsigned long inflated_size)
1863 git_zstream stream;
1864 void *out;
1865 int st;
1867 memset(&stream, 0, sizeof(stream));
1869 stream.next_in = (unsigned char *)data;
1870 stream.avail_in = size;
1871 stream.next_out = out = xmalloc(inflated_size);
1872 stream.avail_out = inflated_size;
1873 git_inflate_init(&stream);
1874 st = git_inflate(&stream, Z_FINISH);
1875 git_inflate_end(&stream);
1876 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1877 free(out);
1878 return NULL;
1880 return out;
1884 * Read a binary hunk and return a new fragment; fragment->patch
1885 * points at an allocated memory that the caller must free, so
1886 * it is marked as "->free_patch = 1".
1888 static struct fragment *parse_binary_hunk(struct apply_state *state,
1889 char **buf_p,
1890 unsigned long *sz_p,
1891 int *status_p,
1892 int *used_p)
1895 * Expect a line that begins with binary patch method ("literal"
1896 * or "delta"), followed by the length of data before deflating.
1897 * a sequence of 'length-byte' followed by base-85 encoded data
1898 * should follow, terminated by a newline.
1900 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1901 * and we would limit the patch line to 66 characters,
1902 * so one line can fit up to 13 groups that would decode
1903 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1904 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1906 int llen, used;
1907 unsigned long size = *sz_p;
1908 char *buffer = *buf_p;
1909 int patch_method;
1910 unsigned long origlen;
1911 char *data = NULL;
1912 int hunk_size = 0;
1913 struct fragment *frag;
1915 llen = linelen(buffer, size);
1916 used = llen;
1918 *status_p = 0;
1920 if (starts_with(buffer, "delta ")) {
1921 patch_method = BINARY_DELTA_DEFLATED;
1922 origlen = strtoul(buffer + 6, NULL, 10);
1924 else if (starts_with(buffer, "literal ")) {
1925 patch_method = BINARY_LITERAL_DEFLATED;
1926 origlen = strtoul(buffer + 8, NULL, 10);
1928 else
1929 return NULL;
1931 state->linenr++;
1932 buffer += llen;
1933 while (1) {
1934 int byte_length, max_byte_length, newsize;
1935 llen = linelen(buffer, size);
1936 used += llen;
1937 state->linenr++;
1938 if (llen == 1) {
1939 /* consume the blank line */
1940 buffer++;
1941 size--;
1942 break;
1945 * Minimum line is "A00000\n" which is 7-byte long,
1946 * and the line length must be multiple of 5 plus 2.
1948 if ((llen < 7) || (llen-2) % 5)
1949 goto corrupt;
1950 max_byte_length = (llen - 2) / 5 * 4;
1951 byte_length = *buffer;
1952 if ('A' <= byte_length && byte_length <= 'Z')
1953 byte_length = byte_length - 'A' + 1;
1954 else if ('a' <= byte_length && byte_length <= 'z')
1955 byte_length = byte_length - 'a' + 27;
1956 else
1957 goto corrupt;
1958 /* if the input length was not multiple of 4, we would
1959 * have filler at the end but the filler should never
1960 * exceed 3 bytes
1962 if (max_byte_length < byte_length ||
1963 byte_length <= max_byte_length - 4)
1964 goto corrupt;
1965 newsize = hunk_size + byte_length;
1966 data = xrealloc(data, newsize);
1967 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1968 goto corrupt;
1969 hunk_size = newsize;
1970 buffer += llen;
1971 size -= llen;
1974 frag = xcalloc(1, sizeof(*frag));
1975 frag->patch = inflate_it(data, hunk_size, origlen);
1976 frag->free_patch = 1;
1977 if (!frag->patch)
1978 goto corrupt;
1979 free(data);
1980 frag->size = origlen;
1981 *buf_p = buffer;
1982 *sz_p = size;
1983 *used_p = used;
1984 frag->binary_patch_method = patch_method;
1985 return frag;
1987 corrupt:
1988 free(data);
1989 *status_p = -1;
1990 error(_("corrupt binary patch at line %d: %.*s"),
1991 state->linenr-1, llen-1, buffer);
1992 return NULL;
1996 * Returns:
1997 * -1 in case of error,
1998 * the length of the parsed binary patch otherwise
2000 static int parse_binary(struct apply_state *state,
2001 char *buffer,
2002 unsigned long size,
2003 struct patch *patch)
2006 * We have read "GIT binary patch\n"; what follows is a line
2007 * that says the patch method (currently, either "literal" or
2008 * "delta") and the length of data before deflating; a
2009 * sequence of 'length-byte' followed by base-85 encoded data
2010 * follows.
2012 * When a binary patch is reversible, there is another binary
2013 * hunk in the same format, starting with patch method (either
2014 * "literal" or "delta") with the length of data, and a sequence
2015 * of length-byte + base-85 encoded data, terminated with another
2016 * empty line. This data, when applied to the postimage, produces
2017 * the preimage.
2019 struct fragment *forward;
2020 struct fragment *reverse;
2021 int status;
2022 int used, used_1;
2024 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2025 if (!forward && !status)
2026 /* there has to be one hunk (forward hunk) */
2027 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2028 if (status)
2029 /* otherwise we already gave an error message */
2030 return status;
2032 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2033 if (reverse)
2034 used += used_1;
2035 else if (status) {
2037 * Not having reverse hunk is not an error, but having
2038 * a corrupt reverse hunk is.
2040 free((void*) forward->patch);
2041 free(forward);
2042 return status;
2044 forward->next = reverse;
2045 patch->fragments = forward;
2046 patch->is_binary = 1;
2047 return used;
2050 static void prefix_one(struct apply_state *state, char **name)
2052 char *old_name = *name;
2053 if (!old_name)
2054 return;
2055 *name = prefix_filename(state->prefix, *name);
2056 free(old_name);
2059 static void prefix_patch(struct apply_state *state, struct patch *p)
2061 if (!state->prefix || p->is_toplevel_relative)
2062 return;
2063 prefix_one(state, &p->new_name);
2064 prefix_one(state, &p->old_name);
2068 * include/exclude
2071 static void add_name_limit(struct apply_state *state,
2072 const char *name,
2073 int exclude)
2075 struct string_list_item *it;
2077 it = string_list_append(&state->limit_by_name, name);
2078 it->util = exclude ? NULL : (void *) 1;
2081 static int use_patch(struct apply_state *state, struct patch *p)
2083 const char *pathname = p->new_name ? p->new_name : p->old_name;
2084 int i;
2086 /* Paths outside are not touched regardless of "--include" */
2087 if (state->prefix && *state->prefix) {
2088 const char *rest;
2089 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2090 return 0;
2093 /* See if it matches any of exclude/include rule */
2094 for (i = 0; i < state->limit_by_name.nr; i++) {
2095 struct string_list_item *it = &state->limit_by_name.items[i];
2096 if (!wildmatch(it->string, pathname, 0))
2097 return (it->util != NULL);
2101 * If we had any include, a path that does not match any rule is
2102 * not used. Otherwise, we saw bunch of exclude rules (or none)
2103 * and such a path is used.
2105 return !state->has_include;
2109 * Read the patch text in "buffer" that extends for "size" bytes; stop
2110 * reading after seeing a single patch (i.e. changes to a single file).
2111 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2113 * Returns:
2114 * -1 if no header was found or parse_binary() failed,
2115 * -128 on another error,
2116 * the number of bytes consumed otherwise,
2117 * so that the caller can call us again for the next patch.
2119 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2121 int hdrsize, patchsize;
2122 int offset = find_header(state, buffer, size, &hdrsize, patch);
2124 if (offset < 0)
2125 return offset;
2127 prefix_patch(state, patch);
2129 if (!use_patch(state, patch))
2130 patch->ws_rule = 0;
2131 else
2132 patch->ws_rule = whitespace_rule(patch->new_name
2133 ? patch->new_name
2134 : patch->old_name);
2136 patchsize = parse_single_patch(state,
2137 buffer + offset + hdrsize,
2138 size - offset - hdrsize,
2139 patch);
2141 if (patchsize < 0)
2142 return -128;
2144 if (!patchsize) {
2145 static const char git_binary[] = "GIT binary patch\n";
2146 int hd = hdrsize + offset;
2147 unsigned long llen = linelen(buffer + hd, size - hd);
2149 if (llen == sizeof(git_binary) - 1 &&
2150 !memcmp(git_binary, buffer + hd, llen)) {
2151 int used;
2152 state->linenr++;
2153 used = parse_binary(state, buffer + hd + llen,
2154 size - hd - llen, patch);
2155 if (used < 0)
2156 return -1;
2157 if (used)
2158 patchsize = used + llen;
2159 else
2160 patchsize = 0;
2162 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2163 static const char *binhdr[] = {
2164 "Binary files ",
2165 "Files ",
2166 NULL,
2168 int i;
2169 for (i = 0; binhdr[i]; i++) {
2170 int len = strlen(binhdr[i]);
2171 if (len < size - hd &&
2172 !memcmp(binhdr[i], buffer + hd, len)) {
2173 state->linenr++;
2174 patch->is_binary = 1;
2175 patchsize = llen;
2176 break;
2181 /* Empty patch cannot be applied if it is a text patch
2182 * without metadata change. A binary patch appears
2183 * empty to us here.
2185 if ((state->apply || state->check) &&
2186 (!patch->is_binary && !metadata_changes(patch))) {
2187 error(_("patch with only garbage at line %d"), state->linenr);
2188 return -128;
2192 return offset + hdrsize + patchsize;
2195 static void reverse_patches(struct patch *p)
2197 for (; p; p = p->next) {
2198 struct fragment *frag = p->fragments;
2200 SWAP(p->new_name, p->old_name);
2201 SWAP(p->new_mode, p->old_mode);
2202 SWAP(p->is_new, p->is_delete);
2203 SWAP(p->lines_added, p->lines_deleted);
2204 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2206 for (; frag; frag = frag->next) {
2207 SWAP(frag->newpos, frag->oldpos);
2208 SWAP(frag->newlines, frag->oldlines);
2213 static const char pluses[] =
2214 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2215 static const char minuses[]=
2216 "----------------------------------------------------------------------";
2218 static void show_stats(struct apply_state *state, struct patch *patch)
2220 struct strbuf qname = STRBUF_INIT;
2221 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2222 int max, add, del;
2224 quote_c_style(cp, &qname, NULL, 0);
2227 * "scale" the filename
2229 max = state->max_len;
2230 if (max > 50)
2231 max = 50;
2233 if (qname.len > max) {
2234 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2235 if (!cp)
2236 cp = qname.buf + qname.len + 3 - max;
2237 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2240 if (patch->is_binary) {
2241 printf(" %-*s | Bin\n", max, qname.buf);
2242 strbuf_release(&qname);
2243 return;
2246 printf(" %-*s |", max, qname.buf);
2247 strbuf_release(&qname);
2250 * scale the add/delete
2252 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2253 add = patch->lines_added;
2254 del = patch->lines_deleted;
2256 if (state->max_change > 0) {
2257 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2258 add = (add * max + state->max_change / 2) / state->max_change;
2259 del = total - add;
2261 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2262 add, pluses, del, minuses);
2265 static int read_old_data(struct stat *st, struct patch *patch,
2266 const char *path, struct strbuf *buf)
2268 int conv_flags = patch->crlf_in_old ?
2269 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2270 switch (st->st_mode & S_IFMT) {
2271 case S_IFLNK:
2272 if (strbuf_readlink(buf, path, st->st_size) < 0)
2273 return error(_("unable to read symlink %s"), path);
2274 return 0;
2275 case S_IFREG:
2276 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2277 return error(_("unable to open or read %s"), path);
2279 * "git apply" without "--index/--cached" should never look
2280 * at the index; the target file may not have been added to
2281 * the index yet, and we may not even be in any Git repository.
2282 * Pass NULL to convert_to_git() to stress this; the function
2283 * should never look at the index when explicit crlf option
2284 * is given.
2286 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2287 return 0;
2288 default:
2289 return -1;
2294 * Update the preimage, and the common lines in postimage,
2295 * from buffer buf of length len. If postlen is 0 the postimage
2296 * is updated in place, otherwise it's updated on a new buffer
2297 * of length postlen
2300 static void update_pre_post_images(struct image *preimage,
2301 struct image *postimage,
2302 char *buf,
2303 size_t len, size_t postlen)
2305 int i, ctx, reduced;
2306 char *new_buf, *old_buf, *fixed;
2307 struct image fixed_preimage;
2310 * Update the preimage with whitespace fixes. Note that we
2311 * are not losing preimage->buf -- apply_one_fragment() will
2312 * free "oldlines".
2314 prepare_image(&fixed_preimage, buf, len, 1);
2315 assert(postlen
2316 ? fixed_preimage.nr == preimage->nr
2317 : fixed_preimage.nr <= preimage->nr);
2318 for (i = 0; i < fixed_preimage.nr; i++)
2319 fixed_preimage.line[i].flag = preimage->line[i].flag;
2320 free(preimage->line_allocated);
2321 *preimage = fixed_preimage;
2324 * Adjust the common context lines in postimage. This can be
2325 * done in-place when we are shrinking it with whitespace
2326 * fixing, but needs a new buffer when ignoring whitespace or
2327 * expanding leading tabs to spaces.
2329 * We trust the caller to tell us if the update can be done
2330 * in place (postlen==0) or not.
2332 old_buf = postimage->buf;
2333 if (postlen)
2334 new_buf = postimage->buf = xmalloc(postlen);
2335 else
2336 new_buf = old_buf;
2337 fixed = preimage->buf;
2339 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2340 size_t l_len = postimage->line[i].len;
2341 if (!(postimage->line[i].flag & LINE_COMMON)) {
2342 /* an added line -- no counterparts in preimage */
2343 memmove(new_buf, old_buf, l_len);
2344 old_buf += l_len;
2345 new_buf += l_len;
2346 continue;
2349 /* a common context -- skip it in the original postimage */
2350 old_buf += l_len;
2352 /* and find the corresponding one in the fixed preimage */
2353 while (ctx < preimage->nr &&
2354 !(preimage->line[ctx].flag & LINE_COMMON)) {
2355 fixed += preimage->line[ctx].len;
2356 ctx++;
2360 * preimage is expected to run out, if the caller
2361 * fixed addition of trailing blank lines.
2363 if (preimage->nr <= ctx) {
2364 reduced++;
2365 continue;
2368 /* and copy it in, while fixing the line length */
2369 l_len = preimage->line[ctx].len;
2370 memcpy(new_buf, fixed, l_len);
2371 new_buf += l_len;
2372 fixed += l_len;
2373 postimage->line[i].len = l_len;
2374 ctx++;
2377 if (postlen
2378 ? postlen < new_buf - postimage->buf
2379 : postimage->len < new_buf - postimage->buf)
2380 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2381 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2383 /* Fix the length of the whole thing */
2384 postimage->len = new_buf - postimage->buf;
2385 postimage->nr -= reduced;
2388 static int line_by_line_fuzzy_match(struct image *img,
2389 struct image *preimage,
2390 struct image *postimage,
2391 unsigned long current,
2392 int current_lno,
2393 int preimage_limit)
2395 int i;
2396 size_t imgoff = 0;
2397 size_t preoff = 0;
2398 size_t postlen = postimage->len;
2399 size_t extra_chars;
2400 char *buf;
2401 char *preimage_eof;
2402 char *preimage_end;
2403 struct strbuf fixed;
2404 char *fixed_buf;
2405 size_t fixed_len;
2407 for (i = 0; i < preimage_limit; i++) {
2408 size_t prelen = preimage->line[i].len;
2409 size_t imglen = img->line[current_lno+i].len;
2411 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2412 preimage->buf + preoff, prelen))
2413 return 0;
2414 if (preimage->line[i].flag & LINE_COMMON)
2415 postlen += imglen - prelen;
2416 imgoff += imglen;
2417 preoff += prelen;
2421 * Ok, the preimage matches with whitespace fuzz.
2423 * imgoff now holds the true length of the target that
2424 * matches the preimage before the end of the file.
2426 * Count the number of characters in the preimage that fall
2427 * beyond the end of the file and make sure that all of them
2428 * are whitespace characters. (This can only happen if
2429 * we are removing blank lines at the end of the file.)
2431 buf = preimage_eof = preimage->buf + preoff;
2432 for ( ; i < preimage->nr; i++)
2433 preoff += preimage->line[i].len;
2434 preimage_end = preimage->buf + preoff;
2435 for ( ; buf < preimage_end; buf++)
2436 if (!isspace(*buf))
2437 return 0;
2440 * Update the preimage and the common postimage context
2441 * lines to use the same whitespace as the target.
2442 * If whitespace is missing in the target (i.e.
2443 * if the preimage extends beyond the end of the file),
2444 * use the whitespace from the preimage.
2446 extra_chars = preimage_end - preimage_eof;
2447 strbuf_init(&fixed, imgoff + extra_chars);
2448 strbuf_add(&fixed, img->buf + current, imgoff);
2449 strbuf_add(&fixed, preimage_eof, extra_chars);
2450 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2451 update_pre_post_images(preimage, postimage,
2452 fixed_buf, fixed_len, postlen);
2453 return 1;
2456 static int match_fragment(struct apply_state *state,
2457 struct image *img,
2458 struct image *preimage,
2459 struct image *postimage,
2460 unsigned long current,
2461 int current_lno,
2462 unsigned ws_rule,
2463 int match_beginning, int match_end)
2465 int i;
2466 char *fixed_buf, *buf, *orig, *target;
2467 struct strbuf fixed;
2468 size_t fixed_len, postlen;
2469 int preimage_limit;
2471 if (preimage->nr + current_lno <= img->nr) {
2473 * The hunk falls within the boundaries of img.
2475 preimage_limit = preimage->nr;
2476 if (match_end && (preimage->nr + current_lno != img->nr))
2477 return 0;
2478 } else if (state->ws_error_action == correct_ws_error &&
2479 (ws_rule & WS_BLANK_AT_EOF)) {
2481 * This hunk extends beyond the end of img, and we are
2482 * removing blank lines at the end of the file. This
2483 * many lines from the beginning of the preimage must
2484 * match with img, and the remainder of the preimage
2485 * must be blank.
2487 preimage_limit = img->nr - current_lno;
2488 } else {
2490 * The hunk extends beyond the end of the img and
2491 * we are not removing blanks at the end, so we
2492 * should reject the hunk at this position.
2494 return 0;
2497 if (match_beginning && current_lno)
2498 return 0;
2500 /* Quick hash check */
2501 for (i = 0; i < preimage_limit; i++)
2502 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2503 (preimage->line[i].hash != img->line[current_lno + i].hash))
2504 return 0;
2506 if (preimage_limit == preimage->nr) {
2508 * Do we have an exact match? If we were told to match
2509 * at the end, size must be exactly at current+fragsize,
2510 * otherwise current+fragsize must be still within the preimage,
2511 * and either case, the old piece should match the preimage
2512 * exactly.
2514 if ((match_end
2515 ? (current + preimage->len == img->len)
2516 : (current + preimage->len <= img->len)) &&
2517 !memcmp(img->buf + current, preimage->buf, preimage->len))
2518 return 1;
2519 } else {
2521 * The preimage extends beyond the end of img, so
2522 * there cannot be an exact match.
2524 * There must be one non-blank context line that match
2525 * a line before the end of img.
2527 char *buf_end;
2529 buf = preimage->buf;
2530 buf_end = buf;
2531 for (i = 0; i < preimage_limit; i++)
2532 buf_end += preimage->line[i].len;
2534 for ( ; buf < buf_end; buf++)
2535 if (!isspace(*buf))
2536 break;
2537 if (buf == buf_end)
2538 return 0;
2542 * No exact match. If we are ignoring whitespace, run a line-by-line
2543 * fuzzy matching. We collect all the line length information because
2544 * we need it to adjust whitespace if we match.
2546 if (state->ws_ignore_action == ignore_ws_change)
2547 return line_by_line_fuzzy_match(img, preimage, postimage,
2548 current, current_lno, preimage_limit);
2550 if (state->ws_error_action != correct_ws_error)
2551 return 0;
2554 * The hunk does not apply byte-by-byte, but the hash says
2555 * it might with whitespace fuzz. We weren't asked to
2556 * ignore whitespace, we were asked to correct whitespace
2557 * errors, so let's try matching after whitespace correction.
2559 * While checking the preimage against the target, whitespace
2560 * errors in both fixed, we count how large the corresponding
2561 * postimage needs to be. The postimage prepared by
2562 * apply_one_fragment() has whitespace errors fixed on added
2563 * lines already, but the common lines were propagated as-is,
2564 * which may become longer when their whitespace errors are
2565 * fixed.
2568 /* First count added lines in postimage */
2569 postlen = 0;
2570 for (i = 0; i < postimage->nr; i++) {
2571 if (!(postimage->line[i].flag & LINE_COMMON))
2572 postlen += postimage->line[i].len;
2576 * The preimage may extend beyond the end of the file,
2577 * but in this loop we will only handle the part of the
2578 * preimage that falls within the file.
2580 strbuf_init(&fixed, preimage->len + 1);
2581 orig = preimage->buf;
2582 target = img->buf + current;
2583 for (i = 0; i < preimage_limit; i++) {
2584 size_t oldlen = preimage->line[i].len;
2585 size_t tgtlen = img->line[current_lno + i].len;
2586 size_t fixstart = fixed.len;
2587 struct strbuf tgtfix;
2588 int match;
2590 /* Try fixing the line in the preimage */
2591 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2593 /* Try fixing the line in the target */
2594 strbuf_init(&tgtfix, tgtlen);
2595 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2598 * If they match, either the preimage was based on
2599 * a version before our tree fixed whitespace breakage,
2600 * or we are lacking a whitespace-fix patch the tree
2601 * the preimage was based on already had (i.e. target
2602 * has whitespace breakage, the preimage doesn't).
2603 * In either case, we are fixing the whitespace breakages
2604 * so we might as well take the fix together with their
2605 * real change.
2607 match = (tgtfix.len == fixed.len - fixstart &&
2608 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2609 fixed.len - fixstart));
2611 /* Add the length if this is common with the postimage */
2612 if (preimage->line[i].flag & LINE_COMMON)
2613 postlen += tgtfix.len;
2615 strbuf_release(&tgtfix);
2616 if (!match)
2617 goto unmatch_exit;
2619 orig += oldlen;
2620 target += tgtlen;
2625 * Now handle the lines in the preimage that falls beyond the
2626 * end of the file (if any). They will only match if they are
2627 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2628 * false).
2630 for ( ; i < preimage->nr; i++) {
2631 size_t fixstart = fixed.len; /* start of the fixed preimage */
2632 size_t oldlen = preimage->line[i].len;
2633 int j;
2635 /* Try fixing the line in the preimage */
2636 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2638 for (j = fixstart; j < fixed.len; j++)
2639 if (!isspace(fixed.buf[j]))
2640 goto unmatch_exit;
2642 orig += oldlen;
2646 * Yes, the preimage is based on an older version that still
2647 * has whitespace breakages unfixed, and fixing them makes the
2648 * hunk match. Update the context lines in the postimage.
2650 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2651 if (postlen < postimage->len)
2652 postlen = 0;
2653 update_pre_post_images(preimage, postimage,
2654 fixed_buf, fixed_len, postlen);
2655 return 1;
2657 unmatch_exit:
2658 strbuf_release(&fixed);
2659 return 0;
2662 static int find_pos(struct apply_state *state,
2663 struct image *img,
2664 struct image *preimage,
2665 struct image *postimage,
2666 int line,
2667 unsigned ws_rule,
2668 int match_beginning, int match_end)
2670 int i;
2671 unsigned long backwards, forwards, current;
2672 int backwards_lno, forwards_lno, current_lno;
2675 * If match_beginning or match_end is specified, there is no
2676 * point starting from a wrong line that will never match and
2677 * wander around and wait for a match at the specified end.
2679 if (match_beginning)
2680 line = 0;
2681 else if (match_end)
2682 line = img->nr - preimage->nr;
2685 * Because the comparison is unsigned, the following test
2686 * will also take care of a negative line number that can
2687 * result when match_end and preimage is larger than the target.
2689 if ((size_t) line > img->nr)
2690 line = img->nr;
2692 current = 0;
2693 for (i = 0; i < line; i++)
2694 current += img->line[i].len;
2697 * There's probably some smart way to do this, but I'll leave
2698 * that to the smart and beautiful people. I'm simple and stupid.
2700 backwards = current;
2701 backwards_lno = line;
2702 forwards = current;
2703 forwards_lno = line;
2704 current_lno = line;
2706 for (i = 0; ; i++) {
2707 if (match_fragment(state, img, preimage, postimage,
2708 current, current_lno, ws_rule,
2709 match_beginning, match_end))
2710 return current_lno;
2712 again:
2713 if (backwards_lno == 0 && forwards_lno == img->nr)
2714 break;
2716 if (i & 1) {
2717 if (backwards_lno == 0) {
2718 i++;
2719 goto again;
2721 backwards_lno--;
2722 backwards -= img->line[backwards_lno].len;
2723 current = backwards;
2724 current_lno = backwards_lno;
2725 } else {
2726 if (forwards_lno == img->nr) {
2727 i++;
2728 goto again;
2730 forwards += img->line[forwards_lno].len;
2731 forwards_lno++;
2732 current = forwards;
2733 current_lno = forwards_lno;
2737 return -1;
2740 static void remove_first_line(struct image *img)
2742 img->buf += img->line[0].len;
2743 img->len -= img->line[0].len;
2744 img->line++;
2745 img->nr--;
2748 static void remove_last_line(struct image *img)
2750 img->len -= img->line[--img->nr].len;
2754 * The change from "preimage" and "postimage" has been found to
2755 * apply at applied_pos (counts in line numbers) in "img".
2756 * Update "img" to remove "preimage" and replace it with "postimage".
2758 static void update_image(struct apply_state *state,
2759 struct image *img,
2760 int applied_pos,
2761 struct image *preimage,
2762 struct image *postimage)
2765 * remove the copy of preimage at offset in img
2766 * and replace it with postimage
2768 int i, nr;
2769 size_t remove_count, insert_count, applied_at = 0;
2770 char *result;
2771 int preimage_limit;
2774 * If we are removing blank lines at the end of img,
2775 * the preimage may extend beyond the end.
2776 * If that is the case, we must be careful only to
2777 * remove the part of the preimage that falls within
2778 * the boundaries of img. Initialize preimage_limit
2779 * to the number of lines in the preimage that falls
2780 * within the boundaries.
2782 preimage_limit = preimage->nr;
2783 if (preimage_limit > img->nr - applied_pos)
2784 preimage_limit = img->nr - applied_pos;
2786 for (i = 0; i < applied_pos; i++)
2787 applied_at += img->line[i].len;
2789 remove_count = 0;
2790 for (i = 0; i < preimage_limit; i++)
2791 remove_count += img->line[applied_pos + i].len;
2792 insert_count = postimage->len;
2794 /* Adjust the contents */
2795 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2796 memcpy(result, img->buf, applied_at);
2797 memcpy(result + applied_at, postimage->buf, postimage->len);
2798 memcpy(result + applied_at + postimage->len,
2799 img->buf + (applied_at + remove_count),
2800 img->len - (applied_at + remove_count));
2801 free(img->buf);
2802 img->buf = result;
2803 img->len += insert_count - remove_count;
2804 result[img->len] = '\0';
2806 /* Adjust the line table */
2807 nr = img->nr + postimage->nr - preimage_limit;
2808 if (preimage_limit < postimage->nr) {
2810 * NOTE: this knows that we never call remove_first_line()
2811 * on anything other than pre/post image.
2813 REALLOC_ARRAY(img->line, nr);
2814 img->line_allocated = img->line;
2816 if (preimage_limit != postimage->nr)
2817 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2818 img->line + applied_pos + preimage_limit,
2819 img->nr - (applied_pos + preimage_limit));
2820 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2821 if (!state->allow_overlap)
2822 for (i = 0; i < postimage->nr; i++)
2823 img->line[applied_pos + i].flag |= LINE_PATCHED;
2824 img->nr = nr;
2828 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2829 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2830 * replace the part of "img" with "postimage" text.
2832 static int apply_one_fragment(struct apply_state *state,
2833 struct image *img, struct fragment *frag,
2834 int inaccurate_eof, unsigned ws_rule,
2835 int nth_fragment)
2837 int match_beginning, match_end;
2838 const char *patch = frag->patch;
2839 int size = frag->size;
2840 char *old, *oldlines;
2841 struct strbuf newlines;
2842 int new_blank_lines_at_end = 0;
2843 int found_new_blank_lines_at_end = 0;
2844 int hunk_linenr = frag->linenr;
2845 unsigned long leading, trailing;
2846 int pos, applied_pos;
2847 struct image preimage;
2848 struct image postimage;
2850 memset(&preimage, 0, sizeof(preimage));
2851 memset(&postimage, 0, sizeof(postimage));
2852 oldlines = xmalloc(size);
2853 strbuf_init(&newlines, size);
2855 old = oldlines;
2856 while (size > 0) {
2857 char first;
2858 int len = linelen(patch, size);
2859 int plen;
2860 int added_blank_line = 0;
2861 int is_blank_context = 0;
2862 size_t start;
2864 if (!len)
2865 break;
2868 * "plen" is how much of the line we should use for
2869 * the actual patch data. Normally we just remove the
2870 * first character on the line, but if the line is
2871 * followed by "\ No newline", then we also remove the
2872 * last one (which is the newline, of course).
2874 plen = len - 1;
2875 if (len < size && patch[len] == '\\')
2876 plen--;
2877 first = *patch;
2878 if (state->apply_in_reverse) {
2879 if (first == '-')
2880 first = '+';
2881 else if (first == '+')
2882 first = '-';
2885 switch (first) {
2886 case '\n':
2887 /* Newer GNU diff, empty context line */
2888 if (plen < 0)
2889 /* ... followed by '\No newline'; nothing */
2890 break;
2891 *old++ = '\n';
2892 strbuf_addch(&newlines, '\n');
2893 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2894 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2895 is_blank_context = 1;
2896 break;
2897 case ' ':
2898 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2899 ws_blank_line(patch + 1, plen, ws_rule))
2900 is_blank_context = 1;
2901 /* fallthrough */
2902 case '-':
2903 memcpy(old, patch + 1, plen);
2904 add_line_info(&preimage, old, plen,
2905 (first == ' ' ? LINE_COMMON : 0));
2906 old += plen;
2907 if (first == '-')
2908 break;
2909 /* fallthrough */
2910 case '+':
2911 /* --no-add does not add new lines */
2912 if (first == '+' && state->no_add)
2913 break;
2915 start = newlines.len;
2916 if (first != '+' ||
2917 !state->whitespace_error ||
2918 state->ws_error_action != correct_ws_error) {
2919 strbuf_add(&newlines, patch + 1, plen);
2921 else {
2922 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2924 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2925 (first == '+' ? 0 : LINE_COMMON));
2926 if (first == '+' &&
2927 (ws_rule & WS_BLANK_AT_EOF) &&
2928 ws_blank_line(patch + 1, plen, ws_rule))
2929 added_blank_line = 1;
2930 break;
2931 case '@': case '\\':
2932 /* Ignore it, we already handled it */
2933 break;
2934 default:
2935 if (state->apply_verbosity > verbosity_normal)
2936 error(_("invalid start of line: '%c'"), first);
2937 applied_pos = -1;
2938 goto out;
2940 if (added_blank_line) {
2941 if (!new_blank_lines_at_end)
2942 found_new_blank_lines_at_end = hunk_linenr;
2943 new_blank_lines_at_end++;
2945 else if (is_blank_context)
2947 else
2948 new_blank_lines_at_end = 0;
2949 patch += len;
2950 size -= len;
2951 hunk_linenr++;
2953 if (inaccurate_eof &&
2954 old > oldlines && old[-1] == '\n' &&
2955 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2956 old--;
2957 strbuf_setlen(&newlines, newlines.len - 1);
2958 preimage.line_allocated[preimage.nr - 1].len--;
2959 postimage.line_allocated[postimage.nr - 1].len--;
2962 leading = frag->leading;
2963 trailing = frag->trailing;
2966 * A hunk to change lines at the beginning would begin with
2967 * @@ -1,L +N,M @@
2968 * but we need to be careful. -U0 that inserts before the second
2969 * line also has this pattern.
2971 * And a hunk to add to an empty file would begin with
2972 * @@ -0,0 +N,M @@
2974 * In other words, a hunk that is (frag->oldpos <= 1) with or
2975 * without leading context must match at the beginning.
2977 match_beginning = (!frag->oldpos ||
2978 (frag->oldpos == 1 && !state->unidiff_zero));
2981 * A hunk without trailing lines must match at the end.
2982 * However, we simply cannot tell if a hunk must match end
2983 * from the lack of trailing lines if the patch was generated
2984 * with unidiff without any context.
2986 match_end = !state->unidiff_zero && !trailing;
2988 pos = frag->newpos ? (frag->newpos - 1) : 0;
2989 preimage.buf = oldlines;
2990 preimage.len = old - oldlines;
2991 postimage.buf = newlines.buf;
2992 postimage.len = newlines.len;
2993 preimage.line = preimage.line_allocated;
2994 postimage.line = postimage.line_allocated;
2996 for (;;) {
2998 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2999 ws_rule, match_beginning, match_end);
3001 if (applied_pos >= 0)
3002 break;
3004 /* Am I at my context limits? */
3005 if ((leading <= state->p_context) && (trailing <= state->p_context))
3006 break;
3007 if (match_beginning || match_end) {
3008 match_beginning = match_end = 0;
3009 continue;
3013 * Reduce the number of context lines; reduce both
3014 * leading and trailing if they are equal otherwise
3015 * just reduce the larger context.
3017 if (leading >= trailing) {
3018 remove_first_line(&preimage);
3019 remove_first_line(&postimage);
3020 pos--;
3021 leading--;
3023 if (trailing > leading) {
3024 remove_last_line(&preimage);
3025 remove_last_line(&postimage);
3026 trailing--;
3030 if (applied_pos >= 0) {
3031 if (new_blank_lines_at_end &&
3032 preimage.nr + applied_pos >= img->nr &&
3033 (ws_rule & WS_BLANK_AT_EOF) &&
3034 state->ws_error_action != nowarn_ws_error) {
3035 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3036 found_new_blank_lines_at_end);
3037 if (state->ws_error_action == correct_ws_error) {
3038 while (new_blank_lines_at_end--)
3039 remove_last_line(&postimage);
3042 * We would want to prevent write_out_results()
3043 * from taking place in apply_patch() that follows
3044 * the callchain led us here, which is:
3045 * apply_patch->check_patch_list->check_patch->
3046 * apply_data->apply_fragments->apply_one_fragment
3048 if (state->ws_error_action == die_on_ws_error)
3049 state->apply = 0;
3052 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3053 int offset = applied_pos - pos;
3054 if (state->apply_in_reverse)
3055 offset = 0 - offset;
3056 fprintf_ln(stderr,
3057 Q_("Hunk #%d succeeded at %d (offset %d line).",
3058 "Hunk #%d succeeded at %d (offset %d lines).",
3059 offset),
3060 nth_fragment, applied_pos + 1, offset);
3064 * Warn if it was necessary to reduce the number
3065 * of context lines.
3067 if ((leading != frag->leading ||
3068 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3069 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3070 " to apply fragment at %d"),
3071 leading, trailing, applied_pos+1);
3072 update_image(state, img, applied_pos, &preimage, &postimage);
3073 } else {
3074 if (state->apply_verbosity > verbosity_normal)
3075 error(_("while searching for:\n%.*s"),
3076 (int)(old - oldlines), oldlines);
3079 out:
3080 free(oldlines);
3081 strbuf_release(&newlines);
3082 free(preimage.line_allocated);
3083 free(postimage.line_allocated);
3085 return (applied_pos < 0);
3088 static int apply_binary_fragment(struct apply_state *state,
3089 struct image *img,
3090 struct patch *patch)
3092 struct fragment *fragment = patch->fragments;
3093 unsigned long len;
3094 void *dst;
3096 if (!fragment)
3097 return error(_("missing binary patch data for '%s'"),
3098 patch->new_name ?
3099 patch->new_name :
3100 patch->old_name);
3102 /* Binary patch is irreversible without the optional second hunk */
3103 if (state->apply_in_reverse) {
3104 if (!fragment->next)
3105 return error(_("cannot reverse-apply a binary patch "
3106 "without the reverse hunk to '%s'"),
3107 patch->new_name
3108 ? patch->new_name : patch->old_name);
3109 fragment = fragment->next;
3111 switch (fragment->binary_patch_method) {
3112 case BINARY_DELTA_DEFLATED:
3113 dst = patch_delta(img->buf, img->len, fragment->patch,
3114 fragment->size, &len);
3115 if (!dst)
3116 return -1;
3117 clear_image(img);
3118 img->buf = dst;
3119 img->len = len;
3120 return 0;
3121 case BINARY_LITERAL_DEFLATED:
3122 clear_image(img);
3123 img->len = fragment->size;
3124 img->buf = xmemdupz(fragment->patch, img->len);
3125 return 0;
3127 return -1;
3131 * Replace "img" with the result of applying the binary patch.
3132 * The binary patch data itself in patch->fragment is still kept
3133 * but the preimage prepared by the caller in "img" is freed here
3134 * or in the helper function apply_binary_fragment() this calls.
3136 static int apply_binary(struct apply_state *state,
3137 struct image *img,
3138 struct patch *patch)
3140 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3141 struct object_id oid;
3144 * For safety, we require patch index line to contain
3145 * full 40-byte textual SHA1 for old and new, at least for now.
3147 if (strlen(patch->old_sha1_prefix) != 40 ||
3148 strlen(patch->new_sha1_prefix) != 40 ||
3149 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3150 get_oid_hex(patch->new_sha1_prefix, &oid))
3151 return error(_("cannot apply binary patch to '%s' "
3152 "without full index line"), name);
3154 if (patch->old_name) {
3156 * See if the old one matches what the patch
3157 * applies to.
3159 hash_object_file(img->buf, img->len, blob_type, &oid);
3160 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3161 return error(_("the patch applies to '%s' (%s), "
3162 "which does not match the "
3163 "current contents."),
3164 name, oid_to_hex(&oid));
3166 else {
3167 /* Otherwise, the old one must be empty. */
3168 if (img->len)
3169 return error(_("the patch applies to an empty "
3170 "'%s' but it is not empty"), name);
3173 get_oid_hex(patch->new_sha1_prefix, &oid);
3174 if (is_null_oid(&oid)) {
3175 clear_image(img);
3176 return 0; /* deletion patch */
3179 if (has_sha1_file(oid.hash)) {
3180 /* We already have the postimage */
3181 enum object_type type;
3182 unsigned long size;
3183 char *result;
3185 result = read_object_file(&oid, &type, &size);
3186 if (!result)
3187 return error(_("the necessary postimage %s for "
3188 "'%s' cannot be read"),
3189 patch->new_sha1_prefix, name);
3190 clear_image(img);
3191 img->buf = result;
3192 img->len = size;
3193 } else {
3195 * We have verified buf matches the preimage;
3196 * apply the patch data to it, which is stored
3197 * in the patch->fragments->{patch,size}.
3199 if (apply_binary_fragment(state, img, patch))
3200 return error(_("binary patch does not apply to '%s'"),
3201 name);
3203 /* verify that the result matches */
3204 hash_object_file(img->buf, img->len, blob_type, &oid);
3205 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3206 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3207 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3210 return 0;
3213 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3215 struct fragment *frag = patch->fragments;
3216 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3217 unsigned ws_rule = patch->ws_rule;
3218 unsigned inaccurate_eof = patch->inaccurate_eof;
3219 int nth = 0;
3221 if (patch->is_binary)
3222 return apply_binary(state, img, patch);
3224 while (frag) {
3225 nth++;
3226 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3227 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3228 if (!state->apply_with_reject)
3229 return -1;
3230 frag->rejected = 1;
3232 frag = frag->next;
3234 return 0;
3237 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3239 if (S_ISGITLINK(mode)) {
3240 strbuf_grow(buf, 100);
3241 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3242 } else {
3243 enum object_type type;
3244 unsigned long sz;
3245 char *result;
3247 result = read_object_file(oid, &type, &sz);
3248 if (!result)
3249 return -1;
3250 /* XXX read_sha1_file NUL-terminates */
3251 strbuf_attach(buf, result, sz, sz + 1);
3253 return 0;
3256 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3258 if (!ce)
3259 return 0;
3260 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3263 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3265 struct string_list_item *item;
3267 if (name == NULL)
3268 return NULL;
3270 item = string_list_lookup(&state->fn_table, name);
3271 if (item != NULL)
3272 return (struct patch *)item->util;
3274 return NULL;
3278 * item->util in the filename table records the status of the path.
3279 * Usually it points at a patch (whose result records the contents
3280 * of it after applying it), but it could be PATH_WAS_DELETED for a
3281 * path that a previously applied patch has already removed, or
3282 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3284 * The latter is needed to deal with a case where two paths A and B
3285 * are swapped by first renaming A to B and then renaming B to A;
3286 * moving A to B should not be prevented due to presence of B as we
3287 * will remove it in a later patch.
3289 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3290 #define PATH_WAS_DELETED ((struct patch *) -1)
3292 static int to_be_deleted(struct patch *patch)
3294 return patch == PATH_TO_BE_DELETED;
3297 static int was_deleted(struct patch *patch)
3299 return patch == PATH_WAS_DELETED;
3302 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3304 struct string_list_item *item;
3307 * Always add new_name unless patch is a deletion
3308 * This should cover the cases for normal diffs,
3309 * file creations and copies
3311 if (patch->new_name != NULL) {
3312 item = string_list_insert(&state->fn_table, patch->new_name);
3313 item->util = patch;
3317 * store a failure on rename/deletion cases because
3318 * later chunks shouldn't patch old names
3320 if ((patch->new_name == NULL) || (patch->is_rename)) {
3321 item = string_list_insert(&state->fn_table, patch->old_name);
3322 item->util = PATH_WAS_DELETED;
3326 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3329 * store information about incoming file deletion
3331 while (patch) {
3332 if ((patch->new_name == NULL) || (patch->is_rename)) {
3333 struct string_list_item *item;
3334 item = string_list_insert(&state->fn_table, patch->old_name);
3335 item->util = PATH_TO_BE_DELETED;
3337 patch = patch->next;
3341 static int checkout_target(struct index_state *istate,
3342 struct cache_entry *ce, struct stat *st)
3344 struct checkout costate = CHECKOUT_INIT;
3346 costate.refresh_cache = 1;
3347 costate.istate = istate;
3348 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3349 return error(_("cannot checkout %s"), ce->name);
3350 return 0;
3353 static struct patch *previous_patch(struct apply_state *state,
3354 struct patch *patch,
3355 int *gone)
3357 struct patch *previous;
3359 *gone = 0;
3360 if (patch->is_copy || patch->is_rename)
3361 return NULL; /* "git" patches do not depend on the order */
3363 previous = in_fn_table(state, patch->old_name);
3364 if (!previous)
3365 return NULL;
3367 if (to_be_deleted(previous))
3368 return NULL; /* the deletion hasn't happened yet */
3370 if (was_deleted(previous))
3371 *gone = 1;
3373 return previous;
3376 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3378 if (S_ISGITLINK(ce->ce_mode)) {
3379 if (!S_ISDIR(st->st_mode))
3380 return -1;
3381 return 0;
3383 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3386 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3388 static int load_patch_target(struct apply_state *state,
3389 struct strbuf *buf,
3390 const struct cache_entry *ce,
3391 struct stat *st,
3392 struct patch *patch,
3393 const char *name,
3394 unsigned expected_mode)
3396 if (state->cached || state->check_index) {
3397 if (read_file_or_gitlink(ce, buf))
3398 return error(_("failed to read %s"), name);
3399 } else if (name) {
3400 if (S_ISGITLINK(expected_mode)) {
3401 if (ce)
3402 return read_file_or_gitlink(ce, buf);
3403 else
3404 return SUBMODULE_PATCH_WITHOUT_INDEX;
3405 } else if (has_symlink_leading_path(name, strlen(name))) {
3406 return error(_("reading from '%s' beyond a symbolic link"), name);
3407 } else {
3408 if (read_old_data(st, patch, name, buf))
3409 return error(_("failed to read %s"), name);
3412 return 0;
3416 * We are about to apply "patch"; populate the "image" with the
3417 * current version we have, from the working tree or from the index,
3418 * depending on the situation e.g. --cached/--index. If we are
3419 * applying a non-git patch that incrementally updates the tree,
3420 * we read from the result of a previous diff.
3422 static int load_preimage(struct apply_state *state,
3423 struct image *image,
3424 struct patch *patch, struct stat *st,
3425 const struct cache_entry *ce)
3427 struct strbuf buf = STRBUF_INIT;
3428 size_t len;
3429 char *img;
3430 struct patch *previous;
3431 int status;
3433 previous = previous_patch(state, patch, &status);
3434 if (status)
3435 return error(_("path %s has been renamed/deleted"),
3436 patch->old_name);
3437 if (previous) {
3438 /* We have a patched copy in memory; use that. */
3439 strbuf_add(&buf, previous->result, previous->resultsize);
3440 } else {
3441 status = load_patch_target(state, &buf, ce, st, patch,
3442 patch->old_name, patch->old_mode);
3443 if (status < 0)
3444 return status;
3445 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3447 * There is no way to apply subproject
3448 * patch without looking at the index.
3449 * NEEDSWORK: shouldn't this be flagged
3450 * as an error???
3452 free_fragment_list(patch->fragments);
3453 patch->fragments = NULL;
3454 } else if (status) {
3455 return error(_("failed to read %s"), patch->old_name);
3459 img = strbuf_detach(&buf, &len);
3460 prepare_image(image, img, len, !patch->is_binary);
3461 return 0;
3464 static int three_way_merge(struct image *image,
3465 char *path,
3466 const struct object_id *base,
3467 const struct object_id *ours,
3468 const struct object_id *theirs)
3470 mmfile_t base_file, our_file, their_file;
3471 mmbuffer_t result = { NULL };
3472 int status;
3474 read_mmblob(&base_file, base);
3475 read_mmblob(&our_file, ours);
3476 read_mmblob(&their_file, theirs);
3477 status = ll_merge(&result, path,
3478 &base_file, "base",
3479 &our_file, "ours",
3480 &their_file, "theirs", NULL);
3481 free(base_file.ptr);
3482 free(our_file.ptr);
3483 free(their_file.ptr);
3484 if (status < 0 || !result.ptr) {
3485 free(result.ptr);
3486 return -1;
3488 clear_image(image);
3489 image->buf = result.ptr;
3490 image->len = result.size;
3492 return status;
3496 * When directly falling back to add/add three-way merge, we read from
3497 * the current contents of the new_name. In no cases other than that
3498 * this function will be called.
3500 static int load_current(struct apply_state *state,
3501 struct image *image,
3502 struct patch *patch)
3504 struct strbuf buf = STRBUF_INIT;
3505 int status, pos;
3506 size_t len;
3507 char *img;
3508 struct stat st;
3509 struct cache_entry *ce;
3510 char *name = patch->new_name;
3511 unsigned mode = patch->new_mode;
3513 if (!patch->is_new)
3514 BUG("patch to %s is not a creation", patch->old_name);
3516 pos = cache_name_pos(name, strlen(name));
3517 if (pos < 0)
3518 return error(_("%s: does not exist in index"), name);
3519 ce = active_cache[pos];
3520 if (lstat(name, &st)) {
3521 if (errno != ENOENT)
3522 return error_errno("%s", name);
3523 if (checkout_target(&the_index, ce, &st))
3524 return -1;
3526 if (verify_index_match(ce, &st))
3527 return error(_("%s: does not match index"), name);
3529 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3530 if (status < 0)
3531 return status;
3532 else if (status)
3533 return -1;
3534 img = strbuf_detach(&buf, &len);
3535 prepare_image(image, img, len, !patch->is_binary);
3536 return 0;
3539 static int try_threeway(struct apply_state *state,
3540 struct image *image,
3541 struct patch *patch,
3542 struct stat *st,
3543 const struct cache_entry *ce)
3545 struct object_id pre_oid, post_oid, our_oid;
3546 struct strbuf buf = STRBUF_INIT;
3547 size_t len;
3548 int status;
3549 char *img;
3550 struct image tmp_image;
3552 /* No point falling back to 3-way merge in these cases */
3553 if (patch->is_delete ||
3554 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3555 return -1;
3557 /* Preimage the patch was prepared for */
3558 if (patch->is_new)
3559 write_object_file("", 0, blob_type, &pre_oid);
3560 else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
3561 read_blob_object(&buf, &pre_oid, patch->old_mode))
3562 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3564 if (state->apply_verbosity > verbosity_silent)
3565 fprintf(stderr, _("Falling back to three-way merge...\n"));
3567 img = strbuf_detach(&buf, &len);
3568 prepare_image(&tmp_image, img, len, 1);
3569 /* Apply the patch to get the post image */
3570 if (apply_fragments(state, &tmp_image, patch) < 0) {
3571 clear_image(&tmp_image);
3572 return -1;
3574 /* post_oid is theirs */
3575 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &post_oid);
3576 clear_image(&tmp_image);
3578 /* our_oid is ours */
3579 if (patch->is_new) {
3580 if (load_current(state, &tmp_image, patch))
3581 return error(_("cannot read the current contents of '%s'"),
3582 patch->new_name);
3583 } else {
3584 if (load_preimage(state, &tmp_image, patch, st, ce))
3585 return error(_("cannot read the current contents of '%s'"),
3586 patch->old_name);
3588 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &our_oid);
3589 clear_image(&tmp_image);
3591 /* in-core three-way merge between post and our using pre as base */
3592 status = three_way_merge(image, patch->new_name,
3593 &pre_oid, &our_oid, &post_oid);
3594 if (status < 0) {
3595 if (state->apply_verbosity > verbosity_silent)
3596 fprintf(stderr,
3597 _("Failed to fall back on three-way merge...\n"));
3598 return status;
3601 if (status) {
3602 patch->conflicted_threeway = 1;
3603 if (patch->is_new)
3604 oidclr(&patch->threeway_stage[0]);
3605 else
3606 oidcpy(&patch->threeway_stage[0], &pre_oid);
3607 oidcpy(&patch->threeway_stage[1], &our_oid);
3608 oidcpy(&patch->threeway_stage[2], &post_oid);
3609 if (state->apply_verbosity > verbosity_silent)
3610 fprintf(stderr,
3611 _("Applied patch to '%s' with conflicts.\n"),
3612 patch->new_name);
3613 } else {
3614 if (state->apply_verbosity > verbosity_silent)
3615 fprintf(stderr,
3616 _("Applied patch to '%s' cleanly.\n"),
3617 patch->new_name);
3619 return 0;
3622 static int apply_data(struct apply_state *state, struct patch *patch,
3623 struct stat *st, const struct cache_entry *ce)
3625 struct image image;
3627 if (load_preimage(state, &image, patch, st, ce) < 0)
3628 return -1;
3630 if (patch->direct_to_threeway ||
3631 apply_fragments(state, &image, patch) < 0) {
3632 /* Note: with --reject, apply_fragments() returns 0 */
3633 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3634 return -1;
3636 patch->result = image.buf;
3637 patch->resultsize = image.len;
3638 add_to_fn_table(state, patch);
3639 free(image.line_allocated);
3641 if (0 < patch->is_delete && patch->resultsize)
3642 return error(_("removal patch leaves file contents"));
3644 return 0;
3648 * If "patch" that we are looking at modifies or deletes what we have,
3649 * we would want it not to lose any local modification we have, either
3650 * in the working tree or in the index.
3652 * This also decides if a non-git patch is a creation patch or a
3653 * modification to an existing empty file. We do not check the state
3654 * of the current tree for a creation patch in this function; the caller
3655 * check_patch() separately makes sure (and errors out otherwise) that
3656 * the path the patch creates does not exist in the current tree.
3658 static int check_preimage(struct apply_state *state,
3659 struct patch *patch,
3660 struct cache_entry **ce,
3661 struct stat *st)
3663 const char *old_name = patch->old_name;
3664 struct patch *previous = NULL;
3665 int stat_ret = 0, status;
3666 unsigned st_mode = 0;
3668 if (!old_name)
3669 return 0;
3671 assert(patch->is_new <= 0);
3672 previous = previous_patch(state, patch, &status);
3674 if (status)
3675 return error(_("path %s has been renamed/deleted"), old_name);
3676 if (previous) {
3677 st_mode = previous->new_mode;
3678 } else if (!state->cached) {
3679 stat_ret = lstat(old_name, st);
3680 if (stat_ret && errno != ENOENT)
3681 return error_errno("%s", old_name);
3684 if (state->check_index && !previous) {
3685 int pos = cache_name_pos(old_name, strlen(old_name));
3686 if (pos < 0) {
3687 if (patch->is_new < 0)
3688 goto is_new;
3689 return error(_("%s: does not exist in index"), old_name);
3691 *ce = active_cache[pos];
3692 if (stat_ret < 0) {
3693 if (checkout_target(&the_index, *ce, st))
3694 return -1;
3696 if (!state->cached && verify_index_match(*ce, st))
3697 return error(_("%s: does not match index"), old_name);
3698 if (state->cached)
3699 st_mode = (*ce)->ce_mode;
3700 } else if (stat_ret < 0) {
3701 if (patch->is_new < 0)
3702 goto is_new;
3703 return error_errno("%s", old_name);
3706 if (!state->cached && !previous)
3707 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3709 if (patch->is_new < 0)
3710 patch->is_new = 0;
3711 if (!patch->old_mode)
3712 patch->old_mode = st_mode;
3713 if ((st_mode ^ patch->old_mode) & S_IFMT)
3714 return error(_("%s: wrong type"), old_name);
3715 if (st_mode != patch->old_mode)
3716 warning(_("%s has type %o, expected %o"),
3717 old_name, st_mode, patch->old_mode);
3718 if (!patch->new_mode && !patch->is_delete)
3719 patch->new_mode = st_mode;
3720 return 0;
3722 is_new:
3723 patch->is_new = 1;
3724 patch->is_delete = 0;
3725 FREE_AND_NULL(patch->old_name);
3726 return 0;
3730 #define EXISTS_IN_INDEX 1
3731 #define EXISTS_IN_WORKTREE 2
3733 static int check_to_create(struct apply_state *state,
3734 const char *new_name,
3735 int ok_if_exists)
3737 struct stat nst;
3739 if (state->check_index &&
3740 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3741 !ok_if_exists)
3742 return EXISTS_IN_INDEX;
3743 if (state->cached)
3744 return 0;
3746 if (!lstat(new_name, &nst)) {
3747 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3748 return 0;
3750 * A leading component of new_name might be a symlink
3751 * that is going to be removed with this patch, but
3752 * still pointing at somewhere that has the path.
3753 * In such a case, path "new_name" does not exist as
3754 * far as git is concerned.
3756 if (has_symlink_leading_path(new_name, strlen(new_name)))
3757 return 0;
3759 return EXISTS_IN_WORKTREE;
3760 } else if (!is_missing_file_error(errno)) {
3761 return error_errno("%s", new_name);
3763 return 0;
3766 static uintptr_t register_symlink_changes(struct apply_state *state,
3767 const char *path,
3768 uintptr_t what)
3770 struct string_list_item *ent;
3772 ent = string_list_lookup(&state->symlink_changes, path);
3773 if (!ent) {
3774 ent = string_list_insert(&state->symlink_changes, path);
3775 ent->util = (void *)0;
3777 ent->util = (void *)(what | ((uintptr_t)ent->util));
3778 return (uintptr_t)ent->util;
3781 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3783 struct string_list_item *ent;
3785 ent = string_list_lookup(&state->symlink_changes, path);
3786 if (!ent)
3787 return 0;
3788 return (uintptr_t)ent->util;
3791 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3793 for ( ; patch; patch = patch->next) {
3794 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3795 (patch->is_rename || patch->is_delete))
3796 /* the symlink at patch->old_name is removed */
3797 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3799 if (patch->new_name && S_ISLNK(patch->new_mode))
3800 /* the symlink at patch->new_name is created or remains */
3801 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3805 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3807 do {
3808 unsigned int change;
3810 while (--name->len && name->buf[name->len] != '/')
3811 ; /* scan backwards */
3812 if (!name->len)
3813 break;
3814 name->buf[name->len] = '\0';
3815 change = check_symlink_changes(state, name->buf);
3816 if (change & APPLY_SYMLINK_IN_RESULT)
3817 return 1;
3818 if (change & APPLY_SYMLINK_GOES_AWAY)
3820 * This cannot be "return 0", because we may
3821 * see a new one created at a higher level.
3823 continue;
3825 /* otherwise, check the preimage */
3826 if (state->check_index) {
3827 struct cache_entry *ce;
3829 ce = cache_file_exists(name->buf, name->len, ignore_case);
3830 if (ce && S_ISLNK(ce->ce_mode))
3831 return 1;
3832 } else {
3833 struct stat st;
3834 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3835 return 1;
3837 } while (1);
3838 return 0;
3841 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3843 int ret;
3844 struct strbuf name = STRBUF_INIT;
3846 assert(*name_ != '\0');
3847 strbuf_addstr(&name, name_);
3848 ret = path_is_beyond_symlink_1(state, &name);
3849 strbuf_release(&name);
3851 return ret;
3854 static int check_unsafe_path(struct patch *patch)
3856 const char *old_name = NULL;
3857 const char *new_name = NULL;
3858 if (patch->is_delete)
3859 old_name = patch->old_name;
3860 else if (!patch->is_new && !patch->is_copy)
3861 old_name = patch->old_name;
3862 if (!patch->is_delete)
3863 new_name = patch->new_name;
3865 if (old_name && !verify_path(old_name, patch->old_mode))
3866 return error(_("invalid path '%s'"), old_name);
3867 if (new_name && !verify_path(new_name, patch->new_mode))
3868 return error(_("invalid path '%s'"), new_name);
3869 return 0;
3873 * Check and apply the patch in-core; leave the result in patch->result
3874 * for the caller to write it out to the final destination.
3876 static int check_patch(struct apply_state *state, struct patch *patch)
3878 struct stat st;
3879 const char *old_name = patch->old_name;
3880 const char *new_name = patch->new_name;
3881 const char *name = old_name ? old_name : new_name;
3882 struct cache_entry *ce = NULL;
3883 struct patch *tpatch;
3884 int ok_if_exists;
3885 int status;
3887 patch->rejected = 1; /* we will drop this after we succeed */
3889 status = check_preimage(state, patch, &ce, &st);
3890 if (status)
3891 return status;
3892 old_name = patch->old_name;
3895 * A type-change diff is always split into a patch to delete
3896 * old, immediately followed by a patch to create new (see
3897 * diff.c::run_diff()); in such a case it is Ok that the entry
3898 * to be deleted by the previous patch is still in the working
3899 * tree and in the index.
3901 * A patch to swap-rename between A and B would first rename A
3902 * to B and then rename B to A. While applying the first one,
3903 * the presence of B should not stop A from getting renamed to
3904 * B; ask to_be_deleted() about the later rename. Removal of
3905 * B and rename from A to B is handled the same way by asking
3906 * was_deleted().
3908 if ((tpatch = in_fn_table(state, new_name)) &&
3909 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3910 ok_if_exists = 1;
3911 else
3912 ok_if_exists = 0;
3914 if (new_name &&
3915 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3916 int err = check_to_create(state, new_name, ok_if_exists);
3918 if (err && state->threeway) {
3919 patch->direct_to_threeway = 1;
3920 } else switch (err) {
3921 case 0:
3922 break; /* happy */
3923 case EXISTS_IN_INDEX:
3924 return error(_("%s: already exists in index"), new_name);
3925 break;
3926 case EXISTS_IN_WORKTREE:
3927 return error(_("%s: already exists in working directory"),
3928 new_name);
3929 default:
3930 return err;
3933 if (!patch->new_mode) {
3934 if (0 < patch->is_new)
3935 patch->new_mode = S_IFREG | 0644;
3936 else
3937 patch->new_mode = patch->old_mode;
3941 if (new_name && old_name) {
3942 int same = !strcmp(old_name, new_name);
3943 if (!patch->new_mode)
3944 patch->new_mode = patch->old_mode;
3945 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3946 if (same)
3947 return error(_("new mode (%o) of %s does not "
3948 "match old mode (%o)"),
3949 patch->new_mode, new_name,
3950 patch->old_mode);
3951 else
3952 return error(_("new mode (%o) of %s does not "
3953 "match old mode (%o) of %s"),
3954 patch->new_mode, new_name,
3955 patch->old_mode, old_name);
3959 if (!state->unsafe_paths && check_unsafe_path(patch))
3960 return -128;
3963 * An attempt to read from or delete a path that is beyond a
3964 * symbolic link will be prevented by load_patch_target() that
3965 * is called at the beginning of apply_data() so we do not
3966 * have to worry about a patch marked with "is_delete" bit
3967 * here. We however need to make sure that the patch result
3968 * is not deposited to a path that is beyond a symbolic link
3969 * here.
3971 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3972 return error(_("affected file '%s' is beyond a symbolic link"),
3973 patch->new_name);
3975 if (apply_data(state, patch, &st, ce) < 0)
3976 return error(_("%s: patch does not apply"), name);
3977 patch->rejected = 0;
3978 return 0;
3981 static int check_patch_list(struct apply_state *state, struct patch *patch)
3983 int err = 0;
3985 prepare_symlink_changes(state, patch);
3986 prepare_fn_table(state, patch);
3987 while (patch) {
3988 int res;