builtin/apply: move 'line_termination' global into 'struct apply_state'
[alt-git.git] / builtin / apply.c
blob95cd60a3ee73fd810feeec1ae7d4e7da1ee13109
1 /*
2 * apply.c
4 * Copyright (C) Linus Torvalds, 2005
6 * This applies patches on top of some (arbitrary) version of the SCM.
8 */
9 #include "cache.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "builtin.h"
16 #include "string-list.h"
17 #include "dir.h"
18 #include "diff.h"
19 #include "parse-options.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "rerere.h"
24 struct apply_state {
25 const char *prefix;
26 int prefix_length;
28 /* These control what gets looked at and modified */
29 int cached; /* apply to the index only */
30 int check; /* preimage must match working tree, don't actually apply */
31 int check_index; /* preimage must match the indexed version */
32 int update_index; /* check_index && apply */
34 /* These control cosmetic aspect of the output */
35 int diffstat; /* just show a diffstat, and don't actually apply */
36 int numstat; /* just show a numeric diffstat, and don't actually apply */
37 int summary; /* just report creation, deletion, etc, and don't actually apply */
39 /* These boolean parameters control how the apply is done */
40 int allow_overlap;
41 int apply_in_reverse;
42 int apply_with_reject;
43 int apply_verbosely;
44 int no_add;
45 int threeway;
46 int unidiff_zero;
47 int unsafe_paths;
49 /* Other non boolean parameters */
50 int line_termination;
54 * --index-info shows the old and new index info for paths if available.
56 static int newfd = -1;
58 static int state_p_value = 1;
59 static int p_value_known;
60 static int apply = 1;
61 static const char *fake_ancestor;
62 static unsigned int p_context = UINT_MAX;
63 static const char * const apply_usage[] = {
64 N_("git apply [<options>] [<patch>...]"),
65 NULL
68 static enum ws_error_action {
69 nowarn_ws_error,
70 warn_on_ws_error,
71 die_on_ws_error,
72 correct_ws_error
73 } ws_error_action = warn_on_ws_error;
74 static int whitespace_error;
75 static int squelch_whitespace_errors = 5;
76 static int applied_after_fixing_ws;
78 static enum ws_ignore {
79 ignore_ws_none,
80 ignore_ws_change
81 } ws_ignore_action = ignore_ws_none;
84 static const char *patch_input_file;
85 static struct strbuf root = STRBUF_INIT;
87 static void parse_whitespace_option(const char *option)
89 if (!option) {
90 ws_error_action = warn_on_ws_error;
91 return;
93 if (!strcmp(option, "warn")) {
94 ws_error_action = warn_on_ws_error;
95 return;
97 if (!strcmp(option, "nowarn")) {
98 ws_error_action = nowarn_ws_error;
99 return;
101 if (!strcmp(option, "error")) {
102 ws_error_action = die_on_ws_error;
103 return;
105 if (!strcmp(option, "error-all")) {
106 ws_error_action = die_on_ws_error;
107 squelch_whitespace_errors = 0;
108 return;
110 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
111 ws_error_action = correct_ws_error;
112 return;
114 die(_("unrecognized whitespace option '%s'"), option);
117 static void parse_ignorewhitespace_option(const char *option)
119 if (!option || !strcmp(option, "no") ||
120 !strcmp(option, "false") || !strcmp(option, "never") ||
121 !strcmp(option, "none")) {
122 ws_ignore_action = ignore_ws_none;
123 return;
125 if (!strcmp(option, "change")) {
126 ws_ignore_action = ignore_ws_change;
127 return;
129 die(_("unrecognized whitespace ignore option '%s'"), option);
132 static void set_default_whitespace_mode(const char *whitespace_option)
134 if (!whitespace_option && !apply_default_whitespace)
135 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error);
139 * For "diff-stat" like behaviour, we keep track of the biggest change
140 * we've seen, and the longest filename. That allows us to do simple
141 * scaling.
143 static int max_change, max_len;
146 * Various "current state", notably line numbers and what
147 * file (and how) we're patching right now.. The "is_xxxx"
148 * things are flags, where -1 means "don't know yet".
150 static int state_linenr = 1;
153 * This represents one "hunk" from a patch, starting with
154 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
155 * patch text is pointed at by patch, and its byte length
156 * is stored in size. leading and trailing are the number
157 * of context lines.
159 struct fragment {
160 unsigned long leading, trailing;
161 unsigned long oldpos, oldlines;
162 unsigned long newpos, newlines;
164 * 'patch' is usually borrowed from buf in apply_patch(),
165 * but some codepaths store an allocated buffer.
167 const char *patch;
168 unsigned free_patch:1,
169 rejected:1;
170 int size;
171 int linenr;
172 struct fragment *next;
176 * When dealing with a binary patch, we reuse "leading" field
177 * to store the type of the binary hunk, either deflated "delta"
178 * or deflated "literal".
180 #define binary_patch_method leading
181 #define BINARY_DELTA_DEFLATED 1
182 #define BINARY_LITERAL_DEFLATED 2
185 * This represents a "patch" to a file, both metainfo changes
186 * such as creation/deletion, filemode and content changes represented
187 * as a series of fragments.
189 struct patch {
190 char *new_name, *old_name, *def_name;
191 unsigned int old_mode, new_mode;
192 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
193 int rejected;
194 unsigned ws_rule;
195 int lines_added, lines_deleted;
196 int score;
197 unsigned int is_toplevel_relative:1;
198 unsigned int inaccurate_eof:1;
199 unsigned int is_binary:1;
200 unsigned int is_copy:1;
201 unsigned int is_rename:1;
202 unsigned int recount:1;
203 unsigned int conflicted_threeway:1;
204 unsigned int direct_to_threeway:1;
205 struct fragment *fragments;
206 char *result;
207 size_t resultsize;
208 char old_sha1_prefix[41];
209 char new_sha1_prefix[41];
210 struct patch *next;
212 /* three-way fallback result */
213 struct object_id threeway_stage[3];
216 static void free_fragment_list(struct fragment *list)
218 while (list) {
219 struct fragment *next = list->next;
220 if (list->free_patch)
221 free((char *)list->patch);
222 free(list);
223 list = next;
227 static void free_patch(struct patch *patch)
229 free_fragment_list(patch->fragments);
230 free(patch->def_name);
231 free(patch->old_name);
232 free(patch->new_name);
233 free(patch->result);
234 free(patch);
237 static void free_patch_list(struct patch *list)
239 while (list) {
240 struct patch *next = list->next;
241 free_patch(list);
242 list = next;
247 * A line in a file, len-bytes long (includes the terminating LF,
248 * except for an incomplete line at the end if the file ends with
249 * one), and its contents hashes to 'hash'.
251 struct line {
252 size_t len;
253 unsigned hash : 24;
254 unsigned flag : 8;
255 #define LINE_COMMON 1
256 #define LINE_PATCHED 2
260 * This represents a "file", which is an array of "lines".
262 struct image {
263 char *buf;
264 size_t len;
265 size_t nr;
266 size_t alloc;
267 struct line *line_allocated;
268 struct line *line;
272 * Records filenames that have been touched, in order to handle
273 * the case where more than one patches touch the same file.
276 static struct string_list fn_table;
278 static uint32_t hash_line(const char *cp, size_t len)
280 size_t i;
281 uint32_t h;
282 for (i = 0, h = 0; i < len; i++) {
283 if (!isspace(cp[i])) {
284 h = h * 3 + (cp[i] & 0xff);
287 return h;
291 * Compare lines s1 of length n1 and s2 of length n2, ignoring
292 * whitespace difference. Returns 1 if they match, 0 otherwise
294 static int fuzzy_matchlines(const char *s1, size_t n1,
295 const char *s2, size_t n2)
297 const char *last1 = s1 + n1 - 1;
298 const char *last2 = s2 + n2 - 1;
299 int result = 0;
301 /* ignore line endings */
302 while ((*last1 == '\r') || (*last1 == '\n'))
303 last1--;
304 while ((*last2 == '\r') || (*last2 == '\n'))
305 last2--;
307 /* skip leading whitespaces, if both begin with whitespace */
308 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
309 while (isspace(*s1) && (s1 <= last1))
310 s1++;
311 while (isspace(*s2) && (s2 <= last2))
312 s2++;
314 /* early return if both lines are empty */
315 if ((s1 > last1) && (s2 > last2))
316 return 1;
317 while (!result) {
318 result = *s1++ - *s2++;
320 * Skip whitespace inside. We check for whitespace on
321 * both buffers because we don't want "a b" to match
322 * "ab"
324 if (isspace(*s1) && isspace(*s2)) {
325 while (isspace(*s1) && s1 <= last1)
326 s1++;
327 while (isspace(*s2) && s2 <= last2)
328 s2++;
331 * If we reached the end on one side only,
332 * lines don't match
334 if (
335 ((s2 > last2) && (s1 <= last1)) ||
336 ((s1 > last1) && (s2 <= last2)))
337 return 0;
338 if ((s1 > last1) && (s2 > last2))
339 break;
342 return !result;
345 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
347 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
348 img->line_allocated[img->nr].len = len;
349 img->line_allocated[img->nr].hash = hash_line(bol, len);
350 img->line_allocated[img->nr].flag = flag;
351 img->nr++;
355 * "buf" has the file contents to be patched (read from various sources).
356 * attach it to "image" and add line-based index to it.
357 * "image" now owns the "buf".
359 static void prepare_image(struct image *image, char *buf, size_t len,
360 int prepare_linetable)
362 const char *cp, *ep;
364 memset(image, 0, sizeof(*image));
365 image->buf = buf;
366 image->len = len;
368 if (!prepare_linetable)
369 return;
371 ep = image->buf + image->len;
372 cp = image->buf;
373 while (cp < ep) {
374 const char *next;
375 for (next = cp; next < ep && *next != '\n'; next++)
377 if (next < ep)
378 next++;
379 add_line_info(image, cp, next - cp, 0);
380 cp = next;
382 image->line = image->line_allocated;
385 static void clear_image(struct image *image)
387 free(image->buf);
388 free(image->line_allocated);
389 memset(image, 0, sizeof(*image));
392 /* fmt must contain _one_ %s and no other substitution */
393 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
395 struct strbuf sb = STRBUF_INIT;
397 if (patch->old_name && patch->new_name &&
398 strcmp(patch->old_name, patch->new_name)) {
399 quote_c_style(patch->old_name, &sb, NULL, 0);
400 strbuf_addstr(&sb, " => ");
401 quote_c_style(patch->new_name, &sb, NULL, 0);
402 } else {
403 const char *n = patch->new_name;
404 if (!n)
405 n = patch->old_name;
406 quote_c_style(n, &sb, NULL, 0);
408 fprintf(output, fmt, sb.buf);
409 fputc('\n', output);
410 strbuf_release(&sb);
413 #define SLOP (16)
415 static void read_patch_file(struct strbuf *sb, int fd)
417 if (strbuf_read(sb, fd, 0) < 0)
418 die_errno("git apply: failed to read");
421 * Make sure that we have some slop in the buffer
422 * so that we can do speculative "memcmp" etc, and
423 * see to it that it is NUL-filled.
425 strbuf_grow(sb, SLOP);
426 memset(sb->buf + sb->len, 0, SLOP);
429 static unsigned long linelen(const char *buffer, unsigned long size)
431 unsigned long len = 0;
432 while (size--) {
433 len++;
434 if (*buffer++ == '\n')
435 break;
437 return len;
440 static int is_dev_null(const char *str)
442 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
445 #define TERM_SPACE 1
446 #define TERM_TAB 2
448 static int name_terminate(const char *name, int namelen, int c, int terminate)
450 if (c == ' ' && !(terminate & TERM_SPACE))
451 return 0;
452 if (c == '\t' && !(terminate & TERM_TAB))
453 return 0;
455 return 1;
458 /* remove double slashes to make --index work with such filenames */
459 static char *squash_slash(char *name)
461 int i = 0, j = 0;
463 if (!name)
464 return NULL;
466 while (name[i]) {
467 if ((name[j++] = name[i++]) == '/')
468 while (name[i] == '/')
469 i++;
471 name[j] = '\0';
472 return name;
475 static char *find_name_gnu(const char *line, const char *def, int p_value)
477 struct strbuf name = STRBUF_INIT;
478 char *cp;
481 * Proposed "new-style" GNU patch/diff format; see
482 * http://marc.info/?l=git&m=112927316408690&w=2
484 if (unquote_c_style(&name, line, NULL)) {
485 strbuf_release(&name);
486 return NULL;
489 for (cp = name.buf; p_value; p_value--) {
490 cp = strchr(cp, '/');
491 if (!cp) {
492 strbuf_release(&name);
493 return NULL;
495 cp++;
498 strbuf_remove(&name, 0, cp - name.buf);
499 if (root.len)
500 strbuf_insert(&name, 0, root.buf, root.len);
501 return squash_slash(strbuf_detach(&name, NULL));
504 static size_t sane_tz_len(const char *line, size_t len)
506 const char *tz, *p;
508 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
509 return 0;
510 tz = line + len - strlen(" +0500");
512 if (tz[1] != '+' && tz[1] != '-')
513 return 0;
515 for (p = tz + 2; p != line + len; p++)
516 if (!isdigit(*p))
517 return 0;
519 return line + len - tz;
522 static size_t tz_with_colon_len(const char *line, size_t len)
524 const char *tz, *p;
526 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
527 return 0;
528 tz = line + len - strlen(" +08:00");
530 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
531 return 0;
532 p = tz + 2;
533 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
534 !isdigit(*p++) || !isdigit(*p++))
535 return 0;
537 return line + len - tz;
540 static size_t date_len(const char *line, size_t len)
542 const char *date, *p;
544 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
545 return 0;
546 p = date = line + len - strlen("72-02-05");
548 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
549 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
550 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
551 return 0;
553 if (date - line >= strlen("19") &&
554 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
555 date -= strlen("19");
557 return line + len - date;
560 static size_t short_time_len(const char *line, size_t len)
562 const char *time, *p;
564 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
565 return 0;
566 p = time = line + len - strlen(" 07:01:32");
568 /* Permit 1-digit hours? */
569 if (*p++ != ' ' ||
570 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
571 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
572 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
573 return 0;
575 return line + len - time;
578 static size_t fractional_time_len(const char *line, size_t len)
580 const char *p;
581 size_t n;
583 /* Expected format: 19:41:17.620000023 */
584 if (!len || !isdigit(line[len - 1]))
585 return 0;
586 p = line + len - 1;
588 /* Fractional seconds. */
589 while (p > line && isdigit(*p))
590 p--;
591 if (*p != '.')
592 return 0;
594 /* Hours, minutes, and whole seconds. */
595 n = short_time_len(line, p - line);
596 if (!n)
597 return 0;
599 return line + len - p + n;
602 static size_t trailing_spaces_len(const char *line, size_t len)
604 const char *p;
606 /* Expected format: ' ' x (1 or more) */
607 if (!len || line[len - 1] != ' ')
608 return 0;
610 p = line + len;
611 while (p != line) {
612 p--;
613 if (*p != ' ')
614 return line + len - (p + 1);
617 /* All spaces! */
618 return len;
621 static size_t diff_timestamp_len(const char *line, size_t len)
623 const char *end = line + len;
624 size_t n;
627 * Posix: 2010-07-05 19:41:17
628 * GNU: 2010-07-05 19:41:17.620000023 -0500
631 if (!isdigit(end[-1]))
632 return 0;
634 n = sane_tz_len(line, end - line);
635 if (!n)
636 n = tz_with_colon_len(line, end - line);
637 end -= n;
639 n = short_time_len(line, end - line);
640 if (!n)
641 n = fractional_time_len(line, end - line);
642 end -= n;
644 n = date_len(line, end - line);
645 if (!n) /* No date. Too bad. */
646 return 0;
647 end -= n;
649 if (end == line) /* No space before date. */
650 return 0;
651 if (end[-1] == '\t') { /* Success! */
652 end--;
653 return line + len - end;
655 if (end[-1] != ' ') /* No space before date. */
656 return 0;
658 /* Whitespace damage. */
659 end -= trailing_spaces_len(line, end - line);
660 return line + len - end;
663 static char *find_name_common(const char *line, const char *def,
664 int p_value, const char *end, int terminate)
666 int len;
667 const char *start = NULL;
669 if (p_value == 0)
670 start = line;
671 while (line != end) {
672 char c = *line;
674 if (!end && isspace(c)) {
675 if (c == '\n')
676 break;
677 if (name_terminate(start, line-start, c, terminate))
678 break;
680 line++;
681 if (c == '/' && !--p_value)
682 start = line;
684 if (!start)
685 return squash_slash(xstrdup_or_null(def));
686 len = line - start;
687 if (!len)
688 return squash_slash(xstrdup_or_null(def));
691 * Generally we prefer the shorter name, especially
692 * if the other one is just a variation of that with
693 * something else tacked on to the end (ie "file.orig"
694 * or "file~").
696 if (def) {
697 int deflen = strlen(def);
698 if (deflen < len && !strncmp(start, def, deflen))
699 return squash_slash(xstrdup(def));
702 if (root.len) {
703 char *ret = xstrfmt("%s%.*s", root.buf, len, start);
704 return squash_slash(ret);
707 return squash_slash(xmemdupz(start, len));
710 static char *find_name(const char *line, char *def, int p_value, int terminate)
712 if (*line == '"') {
713 char *name = find_name_gnu(line, def, p_value);
714 if (name)
715 return name;
718 return find_name_common(line, def, p_value, NULL, terminate);
721 static char *find_name_traditional(const char *line, char *def, int p_value)
723 size_t len;
724 size_t date_len;
726 if (*line == '"') {
727 char *name = find_name_gnu(line, def, p_value);
728 if (name)
729 return name;
732 len = strchrnul(line, '\n') - line;
733 date_len = diff_timestamp_len(line, len);
734 if (!date_len)
735 return find_name_common(line, def, p_value, NULL, TERM_TAB);
736 len -= date_len;
738 return find_name_common(line, def, p_value, line + len, 0);
741 static int count_slashes(const char *cp)
743 int cnt = 0;
744 char ch;
746 while ((ch = *cp++))
747 if (ch == '/')
748 cnt++;
749 return cnt;
753 * Given the string after "--- " or "+++ ", guess the appropriate
754 * p_value for the given patch.
756 static int guess_p_value(struct apply_state *state, const char *nameline)
758 char *name, *cp;
759 int val = -1;
761 if (is_dev_null(nameline))
762 return -1;
763 name = find_name_traditional(nameline, NULL, 0);
764 if (!name)
765 return -1;
766 cp = strchr(name, '/');
767 if (!cp)
768 val = 0;
769 else if (state->prefix) {
771 * Does it begin with "a/$our-prefix" and such? Then this is
772 * very likely to apply to our directory.
774 if (!strncmp(name, state->prefix, state->prefix_length))
775 val = count_slashes(state->prefix);
776 else {
777 cp++;
778 if (!strncmp(cp, state->prefix, state->prefix_length))
779 val = count_slashes(state->prefix) + 1;
782 free(name);
783 return val;
787 * Does the ---/+++ line have the POSIX timestamp after the last HT?
788 * GNU diff puts epoch there to signal a creation/deletion event. Is
789 * this such a timestamp?
791 static int has_epoch_timestamp(const char *nameline)
794 * We are only interested in epoch timestamp; any non-zero
795 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
796 * For the same reason, the date must be either 1969-12-31 or
797 * 1970-01-01, and the seconds part must be "00".
799 const char stamp_regexp[] =
800 "^(1969-12-31|1970-01-01)"
802 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
804 "([-+][0-2][0-9]:?[0-5][0-9])\n";
805 const char *timestamp = NULL, *cp, *colon;
806 static regex_t *stamp;
807 regmatch_t m[10];
808 int zoneoffset;
809 int hourminute;
810 int status;
812 for (cp = nameline; *cp != '\n'; cp++) {
813 if (*cp == '\t')
814 timestamp = cp + 1;
816 if (!timestamp)
817 return 0;
818 if (!stamp) {
819 stamp = xmalloc(sizeof(*stamp));
820 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
821 warning(_("Cannot prepare timestamp regexp %s"),
822 stamp_regexp);
823 return 0;
827 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
828 if (status) {
829 if (status != REG_NOMATCH)
830 warning(_("regexec returned %d for input: %s"),
831 status, timestamp);
832 return 0;
835 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
836 if (*colon == ':')
837 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
838 else
839 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
840 if (timestamp[m[3].rm_so] == '-')
841 zoneoffset = -zoneoffset;
844 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
845 * (west of GMT) or 1970-01-01 (east of GMT)
847 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
848 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
849 return 0;
851 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
852 strtol(timestamp + 14, NULL, 10) -
853 zoneoffset);
855 return ((zoneoffset < 0 && hourminute == 1440) ||
856 (0 <= zoneoffset && !hourminute));
860 * Get the name etc info from the ---/+++ lines of a traditional patch header
862 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
863 * files, we can happily check the index for a match, but for creating a
864 * new file we should try to match whatever "patch" does. I have no idea.
866 static void parse_traditional_patch(struct apply_state *state,
867 const char *first,
868 const char *second,
869 struct patch *patch)
871 char *name;
873 first += 4; /* skip "--- " */
874 second += 4; /* skip "+++ " */
875 if (!p_value_known) {
876 int p, q;
877 p = guess_p_value(state, first);
878 q = guess_p_value(state, second);
879 if (p < 0) p = q;
880 if (0 <= p && p == q) {
881 state_p_value = p;
882 p_value_known = 1;
885 if (is_dev_null(first)) {
886 patch->is_new = 1;
887 patch->is_delete = 0;
888 name = find_name_traditional(second, NULL, state_p_value);
889 patch->new_name = name;
890 } else if (is_dev_null(second)) {
891 patch->is_new = 0;
892 patch->is_delete = 1;
893 name = find_name_traditional(first, NULL, state_p_value);
894 patch->old_name = name;
895 } else {
896 char *first_name;
897 first_name = find_name_traditional(first, NULL, state_p_value);
898 name = find_name_traditional(second, first_name, state_p_value);
899 free(first_name);
900 if (has_epoch_timestamp(first)) {
901 patch->is_new = 1;
902 patch->is_delete = 0;
903 patch->new_name = name;
904 } else if (has_epoch_timestamp(second)) {
905 patch->is_new = 0;
906 patch->is_delete = 1;
907 patch->old_name = name;
908 } else {
909 patch->old_name = name;
910 patch->new_name = xstrdup_or_null(name);
913 if (!name)
914 die(_("unable to find filename in patch at line %d"), state_linenr);
917 static int gitdiff_hdrend(const char *line, struct patch *patch)
919 return -1;
923 * We're anal about diff header consistency, to make
924 * sure that we don't end up having strange ambiguous
925 * patches floating around.
927 * As a result, gitdiff_{old|new}name() will check
928 * their names against any previous information, just
929 * to make sure..
931 #define DIFF_OLD_NAME 0
932 #define DIFF_NEW_NAME 1
934 static void gitdiff_verify_name(const char *line, int isnull, char **name, int side)
936 if (!*name && !isnull) {
937 *name = find_name(line, NULL, state_p_value, TERM_TAB);
938 return;
941 if (*name) {
942 int len = strlen(*name);
943 char *another;
944 if (isnull)
945 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
946 *name, state_linenr);
947 another = find_name(line, NULL, state_p_value, TERM_TAB);
948 if (!another || memcmp(another, *name, len + 1))
949 die((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);
952 free(another);
953 } else {
954 /* expect "/dev/null" */
955 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
956 die(_("git apply: bad git-diff - expected /dev/null on line %d"), state_linenr);
960 static int gitdiff_oldname(const char *line, struct patch *patch)
962 gitdiff_verify_name(line, patch->is_new, &patch->old_name,
963 DIFF_OLD_NAME);
964 return 0;
967 static int gitdiff_newname(const char *line, struct patch *patch)
969 gitdiff_verify_name(line, patch->is_delete, &patch->new_name,
970 DIFF_NEW_NAME);
971 return 0;
974 static int gitdiff_oldmode(const char *line, struct patch *patch)
976 patch->old_mode = strtoul(line, NULL, 8);
977 return 0;
980 static int gitdiff_newmode(const char *line, struct patch *patch)
982 patch->new_mode = strtoul(line, NULL, 8);
983 return 0;
986 static int gitdiff_delete(const char *line, struct patch *patch)
988 patch->is_delete = 1;
989 free(patch->old_name);
990 patch->old_name = xstrdup_or_null(patch->def_name);
991 return gitdiff_oldmode(line, patch);
994 static int gitdiff_newfile(const char *line, struct patch *patch)
996 patch->is_new = 1;
997 free(patch->new_name);
998 patch->new_name = xstrdup_or_null(patch->def_name);
999 return gitdiff_newmode(line, patch);
1002 static int gitdiff_copysrc(const char *line, struct patch *patch)
1004 patch->is_copy = 1;
1005 free(patch->old_name);
1006 patch->old_name = find_name(line, NULL, state_p_value ? state_p_value - 1 : 0, 0);
1007 return 0;
1010 static int gitdiff_copydst(const char *line, struct patch *patch)
1012 patch->is_copy = 1;
1013 free(patch->new_name);
1014 patch->new_name = find_name(line, NULL, state_p_value ? state_p_value - 1 : 0, 0);
1015 return 0;
1018 static int gitdiff_renamesrc(const char *line, struct patch *patch)
1020 patch->is_rename = 1;
1021 free(patch->old_name);
1022 patch->old_name = find_name(line, NULL, state_p_value ? state_p_value - 1 : 0, 0);
1023 return 0;
1026 static int gitdiff_renamedst(const char *line, struct patch *patch)
1028 patch->is_rename = 1;
1029 free(patch->new_name);
1030 patch->new_name = find_name(line, NULL, state_p_value ? state_p_value - 1 : 0, 0);
1031 return 0;
1034 static int gitdiff_similarity(const char *line, struct patch *patch)
1036 unsigned long val = strtoul(line, NULL, 10);
1037 if (val <= 100)
1038 patch->score = val;
1039 return 0;
1042 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
1044 unsigned long val = strtoul(line, NULL, 10);
1045 if (val <= 100)
1046 patch->score = val;
1047 return 0;
1050 static int gitdiff_index(const char *line, struct patch *patch)
1053 * index line is N hexadecimal, "..", N hexadecimal,
1054 * and optional space with octal mode.
1056 const char *ptr, *eol;
1057 int len;
1059 ptr = strchr(line, '.');
1060 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1061 return 0;
1062 len = ptr - line;
1063 memcpy(patch->old_sha1_prefix, line, len);
1064 patch->old_sha1_prefix[len] = 0;
1066 line = ptr + 2;
1067 ptr = strchr(line, ' ');
1068 eol = strchrnul(line, '\n');
1070 if (!ptr || eol < ptr)
1071 ptr = eol;
1072 len = ptr - line;
1074 if (40 < len)
1075 return 0;
1076 memcpy(patch->new_sha1_prefix, line, len);
1077 patch->new_sha1_prefix[len] = 0;
1078 if (*ptr == ' ')
1079 patch->old_mode = strtoul(ptr+1, NULL, 8);
1080 return 0;
1084 * This is normal for a diff that doesn't change anything: we'll fall through
1085 * into the next diff. Tell the parser to break out.
1087 static int gitdiff_unrecognized(const char *line, struct patch *patch)
1089 return -1;
1093 * Skip p_value leading components from "line"; as we do not accept
1094 * absolute paths, return NULL in that case.
1096 static const char *skip_tree_prefix(const char *line, int llen)
1098 int nslash;
1099 int i;
1101 if (!state_p_value)
1102 return (llen && line[0] == '/') ? NULL : line;
1104 nslash = state_p_value;
1105 for (i = 0; i < llen; i++) {
1106 int ch = line[i];
1107 if (ch == '/' && --nslash <= 0)
1108 return (i == 0) ? NULL : &line[i + 1];
1110 return NULL;
1114 * This is to extract the same name that appears on "diff --git"
1115 * line. We do not find and return anything if it is a rename
1116 * patch, and it is OK because we will find the name elsewhere.
1117 * We need to reliably find name only when it is mode-change only,
1118 * creation or deletion of an empty file. In any of these cases,
1119 * both sides are the same name under a/ and b/ respectively.
1121 static char *git_header_name(const char *line, int llen)
1123 const char *name;
1124 const char *second = NULL;
1125 size_t len, line_len;
1127 line += strlen("diff --git ");
1128 llen -= strlen("diff --git ");
1130 if (*line == '"') {
1131 const char *cp;
1132 struct strbuf first = STRBUF_INIT;
1133 struct strbuf sp = STRBUF_INIT;
1135 if (unquote_c_style(&first, line, &second))
1136 goto free_and_fail1;
1138 /* strip the a/b prefix including trailing slash */
1139 cp = skip_tree_prefix(first.buf, first.len);
1140 if (!cp)
1141 goto free_and_fail1;
1142 strbuf_remove(&first, 0, cp - first.buf);
1145 * second points at one past closing dq of name.
1146 * find the second name.
1148 while ((second < line + llen) && isspace(*second))
1149 second++;
1151 if (line + llen <= second)
1152 goto free_and_fail1;
1153 if (*second == '"') {
1154 if (unquote_c_style(&sp, second, NULL))
1155 goto free_and_fail1;
1156 cp = skip_tree_prefix(sp.buf, sp.len);
1157 if (!cp)
1158 goto free_and_fail1;
1159 /* They must match, otherwise ignore */
1160 if (strcmp(cp, first.buf))
1161 goto free_and_fail1;
1162 strbuf_release(&sp);
1163 return strbuf_detach(&first, NULL);
1166 /* unquoted second */
1167 cp = skip_tree_prefix(second, line + llen - second);
1168 if (!cp)
1169 goto free_and_fail1;
1170 if (line + llen - cp != first.len ||
1171 memcmp(first.buf, cp, first.len))
1172 goto free_and_fail1;
1173 return strbuf_detach(&first, NULL);
1175 free_and_fail1:
1176 strbuf_release(&first);
1177 strbuf_release(&sp);
1178 return NULL;
1181 /* unquoted first name */
1182 name = skip_tree_prefix(line, llen);
1183 if (!name)
1184 return NULL;
1187 * since the first name is unquoted, a dq if exists must be
1188 * the beginning of the second name.
1190 for (second = name; second < line + llen; second++) {
1191 if (*second == '"') {
1192 struct strbuf sp = STRBUF_INIT;
1193 const char *np;
1195 if (unquote_c_style(&sp, second, NULL))
1196 goto free_and_fail2;
1198 np = skip_tree_prefix(sp.buf, sp.len);
1199 if (!np)
1200 goto free_and_fail2;
1202 len = sp.buf + sp.len - np;
1203 if (len < second - name &&
1204 !strncmp(np, name, len) &&
1205 isspace(name[len])) {
1206 /* Good */
1207 strbuf_remove(&sp, 0, np - sp.buf);
1208 return strbuf_detach(&sp, NULL);
1211 free_and_fail2:
1212 strbuf_release(&sp);
1213 return NULL;
1218 * Accept a name only if it shows up twice, exactly the same
1219 * form.
1221 second = strchr(name, '\n');
1222 if (!second)
1223 return NULL;
1224 line_len = second - name;
1225 for (len = 0 ; ; len++) {
1226 switch (name[len]) {
1227 default:
1228 continue;
1229 case '\n':
1230 return NULL;
1231 case '\t': case ' ':
1233 * Is this the separator between the preimage
1234 * and the postimage pathname? Again, we are
1235 * only interested in the case where there is
1236 * no rename, as this is only to set def_name
1237 * and a rename patch has the names elsewhere
1238 * in an unambiguous form.
1240 if (!name[len + 1])
1241 return NULL; /* no postimage name */
1242 second = skip_tree_prefix(name + len + 1,
1243 line_len - (len + 1));
1244 if (!second)
1245 return NULL;
1247 * Does len bytes starting at "name" and "second"
1248 * (that are separated by one HT or SP we just
1249 * found) exactly match?
1251 if (second[len] == '\n' && !strncmp(name, second, len))
1252 return xmemdupz(name, len);
1257 /* Verify that we recognize the lines following a git header */
1258 static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)
1260 unsigned long offset;
1262 /* A git diff has explicit new/delete information, so we don't guess */
1263 patch->is_new = 0;
1264 patch->is_delete = 0;
1267 * Some things may not have the old name in the
1268 * rest of the headers anywhere (pure mode changes,
1269 * or removing or adding empty files), so we get
1270 * the default name from the header.
1272 patch->def_name = git_header_name(line, len);
1273 if (patch->def_name && root.len) {
1274 char *s = xstrfmt("%s%s", root.buf, patch->def_name);
1275 free(patch->def_name);
1276 patch->def_name = s;
1279 line += len;
1280 size -= len;
1281 state_linenr++;
1282 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state_linenr++) {
1283 static const struct opentry {
1284 const char *str;
1285 int (*fn)(const char *, struct patch *);
1286 } optable[] = {
1287 { "@@ -", gitdiff_hdrend },
1288 { "--- ", gitdiff_oldname },
1289 { "+++ ", gitdiff_newname },
1290 { "old mode ", gitdiff_oldmode },
1291 { "new mode ", gitdiff_newmode },
1292 { "deleted file mode ", gitdiff_delete },
1293 { "new file mode ", gitdiff_newfile },
1294 { "copy from ", gitdiff_copysrc },
1295 { "copy to ", gitdiff_copydst },
1296 { "rename old ", gitdiff_renamesrc },
1297 { "rename new ", gitdiff_renamedst },
1298 { "rename from ", gitdiff_renamesrc },
1299 { "rename to ", gitdiff_renamedst },
1300 { "similarity index ", gitdiff_similarity },
1301 { "dissimilarity index ", gitdiff_dissimilarity },
1302 { "index ", gitdiff_index },
1303 { "", gitdiff_unrecognized },
1305 int i;
1307 len = linelen(line, size);
1308 if (!len || line[len-1] != '\n')
1309 break;
1310 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1311 const struct opentry *p = optable + i;
1312 int oplen = strlen(p->str);
1313 if (len < oplen || memcmp(p->str, line, oplen))
1314 continue;
1315 if (p->fn(line + oplen, patch) < 0)
1316 return offset;
1317 break;
1321 return offset;
1324 static int parse_num(const char *line, unsigned long *p)
1326 char *ptr;
1328 if (!isdigit(*line))
1329 return 0;
1330 *p = strtoul(line, &ptr, 10);
1331 return ptr - line;
1334 static int parse_range(const char *line, int len, int offset, const char *expect,
1335 unsigned long *p1, unsigned long *p2)
1337 int digits, ex;
1339 if (offset < 0 || offset >= len)
1340 return -1;
1341 line += offset;
1342 len -= offset;
1344 digits = parse_num(line, p1);
1345 if (!digits)
1346 return -1;
1348 offset += digits;
1349 line += digits;
1350 len -= digits;
1352 *p2 = 1;
1353 if (*line == ',') {
1354 digits = parse_num(line+1, p2);
1355 if (!digits)
1356 return -1;
1358 offset += digits+1;
1359 line += digits+1;
1360 len -= digits+1;
1363 ex = strlen(expect);
1364 if (ex > len)
1365 return -1;
1366 if (memcmp(line, expect, ex))
1367 return -1;
1369 return offset + ex;
1372 static void recount_diff(const char *line, int size, struct fragment *fragment)
1374 int oldlines = 0, newlines = 0, ret = 0;
1376 if (size < 1) {
1377 warning("recount: ignore empty hunk");
1378 return;
1381 for (;;) {
1382 int len = linelen(line, size);
1383 size -= len;
1384 line += len;
1386 if (size < 1)
1387 break;
1389 switch (*line) {
1390 case ' ': case '\n':
1391 newlines++;
1392 /* fall through */
1393 case '-':
1394 oldlines++;
1395 continue;
1396 case '+':
1397 newlines++;
1398 continue;
1399 case '\\':
1400 continue;
1401 case '@':
1402 ret = size < 3 || !starts_with(line, "@@ ");
1403 break;
1404 case 'd':
1405 ret = size < 5 || !starts_with(line, "diff ");
1406 break;
1407 default:
1408 ret = -1;
1409 break;
1411 if (ret) {
1412 warning(_("recount: unexpected line: %.*s"),
1413 (int)linelen(line, size), line);
1414 return;
1416 break;
1418 fragment->oldlines = oldlines;
1419 fragment->newlines = newlines;
1423 * Parse a unified diff fragment header of the
1424 * form "@@ -a,b +c,d @@"
1426 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1428 int offset;
1430 if (!len || line[len-1] != '\n')
1431 return -1;
1433 /* Figure out the number of lines in a fragment */
1434 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1435 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1437 return offset;
1440 static int find_header(struct apply_state *state,
1441 const char *line,
1442 unsigned long size,
1443 int *hdrsize,
1444 struct patch *patch)
1446 unsigned long offset, len;
1448 patch->is_toplevel_relative = 0;
1449 patch->is_rename = patch->is_copy = 0;
1450 patch->is_new = patch->is_delete = -1;
1451 patch->old_mode = patch->new_mode = 0;
1452 patch->old_name = patch->new_name = NULL;
1453 for (offset = 0; size > 0; offset += len, size -= len, line += len, state_linenr++) {
1454 unsigned long nextlen;
1456 len = linelen(line, size);
1457 if (!len)
1458 break;
1460 /* Testing this early allows us to take a few shortcuts.. */
1461 if (len < 6)
1462 continue;
1465 * Make sure we don't find any unconnected patch fragments.
1466 * That's a sign that we didn't find a header, and that a
1467 * patch has become corrupted/broken up.
1469 if (!memcmp("@@ -", line, 4)) {
1470 struct fragment dummy;
1471 if (parse_fragment_header(line, len, &dummy) < 0)
1472 continue;
1473 die(_("patch fragment without header at line %d: %.*s"),
1474 state_linenr, (int)len-1, line);
1477 if (size < len + 6)
1478 break;
1481 * Git patch? It might not have a real patch, just a rename
1482 * or mode change, so we handle that specially
1484 if (!memcmp("diff --git ", line, 11)) {
1485 int git_hdr_len = parse_git_header(line, len, size, patch);
1486 if (git_hdr_len <= len)
1487 continue;
1488 if (!patch->old_name && !patch->new_name) {
1489 if (!patch->def_name)
1490 die(Q_("git diff header lacks filename information when removing "
1491 "%d leading pathname component (line %d)",
1492 "git diff header lacks filename information when removing "
1493 "%d leading pathname components (line %d)",
1494 state_p_value),
1495 state_p_value, state_linenr);
1496 patch->old_name = xstrdup(patch->def_name);
1497 patch->new_name = xstrdup(patch->def_name);
1499 if (!patch->is_delete && !patch->new_name)
1500 die("git diff header lacks filename information "
1501 "(line %d)", state_linenr);
1502 patch->is_toplevel_relative = 1;
1503 *hdrsize = git_hdr_len;
1504 return offset;
1507 /* --- followed by +++ ? */
1508 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1509 continue;
1512 * We only accept unified patches, so we want it to
1513 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1514 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1516 nextlen = linelen(line + len, size - len);
1517 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1518 continue;
1520 /* Ok, we'll consider it a patch */
1521 parse_traditional_patch(state, line, line+len, patch);
1522 *hdrsize = len + nextlen;
1523 state_linenr += 2;
1524 return offset;
1526 return -1;
1529 static void record_ws_error(unsigned result, const char *line, int len, int linenr)
1531 char *err;
1533 if (!result)
1534 return;
1536 whitespace_error++;
1537 if (squelch_whitespace_errors &&
1538 squelch_whitespace_errors < whitespace_error)
1539 return;
1541 err = whitespace_error_string(result);
1542 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1543 patch_input_file, linenr, err, len, line);
1544 free(err);
1547 static void check_whitespace(const char *line, int len, unsigned ws_rule)
1549 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1551 record_ws_error(result, line + 1, len - 2, state_linenr);
1555 * Parse a unified diff. Note that this really needs to parse each
1556 * fragment separately, since the only way to know the difference
1557 * between a "---" that is part of a patch, and a "---" that starts
1558 * the next patch is to look at the line counts..
1560 static int parse_fragment(struct apply_state *state,
1561 const char *line,
1562 unsigned long size,
1563 struct patch *patch,
1564 struct fragment *fragment)
1566 int added, deleted;
1567 int len = linelen(line, size), offset;
1568 unsigned long oldlines, newlines;
1569 unsigned long leading, trailing;
1571 offset = parse_fragment_header(line, len, fragment);
1572 if (offset < 0)
1573 return -1;
1574 if (offset > 0 && patch->recount)
1575 recount_diff(line + offset, size - offset, fragment);
1576 oldlines = fragment->oldlines;
1577 newlines = fragment->newlines;
1578 leading = 0;
1579 trailing = 0;
1581 /* Parse the thing.. */
1582 line += len;
1583 size -= len;
1584 state_linenr++;
1585 added = deleted = 0;
1586 for (offset = len;
1587 0 < size;
1588 offset += len, size -= len, line += len, state_linenr++) {
1589 if (!oldlines && !newlines)
1590 break;
1591 len = linelen(line, size);
1592 if (!len || line[len-1] != '\n')
1593 return -1;
1594 switch (*line) {
1595 default:
1596 return -1;
1597 case '\n': /* newer GNU diff, an empty context line */
1598 case ' ':
1599 oldlines--;
1600 newlines--;
1601 if (!deleted && !added)
1602 leading++;
1603 trailing++;
1604 if (!state->apply_in_reverse &&
1605 ws_error_action == correct_ws_error)
1606 check_whitespace(line, len, patch->ws_rule);
1607 break;
1608 case '-':
1609 if (state->apply_in_reverse &&
1610 ws_error_action != nowarn_ws_error)
1611 check_whitespace(line, len, patch->ws_rule);
1612 deleted++;
1613 oldlines--;
1614 trailing = 0;
1615 break;
1616 case '+':
1617 if (!state->apply_in_reverse &&
1618 ws_error_action != nowarn_ws_error)
1619 check_whitespace(line, len, patch->ws_rule);
1620 added++;
1621 newlines--;
1622 trailing = 0;
1623 break;
1626 * We allow "\ No newline at end of file". Depending
1627 * on locale settings when the patch was produced we
1628 * don't know what this line looks like. The only
1629 * thing we do know is that it begins with "\ ".
1630 * Checking for 12 is just for sanity check -- any
1631 * l10n of "\ No newline..." is at least that long.
1633 case '\\':
1634 if (len < 12 || memcmp(line, "\\ ", 2))
1635 return -1;
1636 break;
1639 if (oldlines || newlines)
1640 return -1;
1641 if (!deleted && !added)
1642 return -1;
1644 fragment->leading = leading;
1645 fragment->trailing = trailing;
1648 * If a fragment ends with an incomplete line, we failed to include
1649 * it in the above loop because we hit oldlines == newlines == 0
1650 * before seeing it.
1652 if (12 < size && !memcmp(line, "\\ ", 2))
1653 offset += linelen(line, size);
1655 patch->lines_added += added;
1656 patch->lines_deleted += deleted;
1658 if (0 < patch->is_new && oldlines)
1659 return error(_("new file depends on old contents"));
1660 if (0 < patch->is_delete && newlines)
1661 return error(_("deleted file still has contents"));
1662 return offset;
1666 * We have seen "diff --git a/... b/..." header (or a traditional patch
1667 * header). Read hunks that belong to this patch into fragments and hang
1668 * them to the given patch structure.
1670 * The (fragment->patch, fragment->size) pair points into the memory given
1671 * by the caller, not a copy, when we return.
1673 static int parse_single_patch(struct apply_state *state,
1674 const char *line,
1675 unsigned long size,
1676 struct patch *patch)
1678 unsigned long offset = 0;
1679 unsigned long oldlines = 0, newlines = 0, context = 0;
1680 struct fragment **fragp = &patch->fragments;
1682 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1683 struct fragment *fragment;
1684 int len;
1686 fragment = xcalloc(1, sizeof(*fragment));
1687 fragment->linenr = state_linenr;
1688 len = parse_fragment(state, line, size, patch, fragment);
1689 if (len <= 0)
1690 die(_("corrupt patch at line %d"), state_linenr);
1691 fragment->patch = line;
1692 fragment->size = len;
1693 oldlines += fragment->oldlines;
1694 newlines += fragment->newlines;
1695 context += fragment->leading + fragment->trailing;
1697 *fragp = fragment;
1698 fragp = &fragment->next;
1700 offset += len;
1701 line += len;
1702 size -= len;
1706 * If something was removed (i.e. we have old-lines) it cannot
1707 * be creation, and if something was added it cannot be
1708 * deletion. However, the reverse is not true; --unified=0
1709 * patches that only add are not necessarily creation even
1710 * though they do not have any old lines, and ones that only
1711 * delete are not necessarily deletion.
1713 * Unfortunately, a real creation/deletion patch do _not_ have
1714 * any context line by definition, so we cannot safely tell it
1715 * apart with --unified=0 insanity. At least if the patch has
1716 * more than one hunk it is not creation or deletion.
1718 if (patch->is_new < 0 &&
1719 (oldlines || (patch->fragments && patch->fragments->next)))
1720 patch->is_new = 0;
1721 if (patch->is_delete < 0 &&
1722 (newlines || (patch->fragments && patch->fragments->next)))
1723 patch->is_delete = 0;
1725 if (0 < patch->is_new && oldlines)
1726 die(_("new file %s depends on old contents"), patch->new_name);
1727 if (0 < patch->is_delete && newlines)
1728 die(_("deleted file %s still has contents"), patch->old_name);
1729 if (!patch->is_delete && !newlines && context)
1730 fprintf_ln(stderr,
1731 _("** warning: "
1732 "file %s becomes empty but is not deleted"),
1733 patch->new_name);
1735 return offset;
1738 static inline int metadata_changes(struct patch *patch)
1740 return patch->is_rename > 0 ||
1741 patch->is_copy > 0 ||
1742 patch->is_new > 0 ||
1743 patch->is_delete ||
1744 (patch->old_mode && patch->new_mode &&
1745 patch->old_mode != patch->new_mode);
1748 static char *inflate_it(const void *data, unsigned long size,
1749 unsigned long inflated_size)
1751 git_zstream stream;
1752 void *out;
1753 int st;
1755 memset(&stream, 0, sizeof(stream));
1757 stream.next_in = (unsigned char *)data;
1758 stream.avail_in = size;
1759 stream.next_out = out = xmalloc(inflated_size);
1760 stream.avail_out = inflated_size;
1761 git_inflate_init(&stream);
1762 st = git_inflate(&stream, Z_FINISH);
1763 git_inflate_end(&stream);
1764 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1765 free(out);
1766 return NULL;
1768 return out;
1772 * Read a binary hunk and return a new fragment; fragment->patch
1773 * points at an allocated memory that the caller must free, so
1774 * it is marked as "->free_patch = 1".
1776 static struct fragment *parse_binary_hunk(char **buf_p,
1777 unsigned long *sz_p,
1778 int *status_p,
1779 int *used_p)
1782 * Expect a line that begins with binary patch method ("literal"
1783 * or "delta"), followed by the length of data before deflating.
1784 * a sequence of 'length-byte' followed by base-85 encoded data
1785 * should follow, terminated by a newline.
1787 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1788 * and we would limit the patch line to 66 characters,
1789 * so one line can fit up to 13 groups that would decode
1790 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1791 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1793 int llen, used;
1794 unsigned long size = *sz_p;
1795 char *buffer = *buf_p;
1796 int patch_method;
1797 unsigned long origlen;
1798 char *data = NULL;
1799 int hunk_size = 0;
1800 struct fragment *frag;
1802 llen = linelen(buffer, size);
1803 used = llen;
1805 *status_p = 0;
1807 if (starts_with(buffer, "delta ")) {
1808 patch_method = BINARY_DELTA_DEFLATED;
1809 origlen = strtoul(buffer + 6, NULL, 10);
1811 else if (starts_with(buffer, "literal ")) {
1812 patch_method = BINARY_LITERAL_DEFLATED;
1813 origlen = strtoul(buffer + 8, NULL, 10);
1815 else
1816 return NULL;
1818 state_linenr++;
1819 buffer += llen;
1820 while (1) {
1821 int byte_length, max_byte_length, newsize;
1822 llen = linelen(buffer, size);
1823 used += llen;
1824 state_linenr++;
1825 if (llen == 1) {
1826 /* consume the blank line */
1827 buffer++;
1828 size--;
1829 break;
1832 * Minimum line is "A00000\n" which is 7-byte long,
1833 * and the line length must be multiple of 5 plus 2.
1835 if ((llen < 7) || (llen-2) % 5)
1836 goto corrupt;
1837 max_byte_length = (llen - 2) / 5 * 4;
1838 byte_length = *buffer;
1839 if ('A' <= byte_length && byte_length <= 'Z')
1840 byte_length = byte_length - 'A' + 1;
1841 else if ('a' <= byte_length && byte_length <= 'z')
1842 byte_length = byte_length - 'a' + 27;
1843 else
1844 goto corrupt;
1845 /* if the input length was not multiple of 4, we would
1846 * have filler at the end but the filler should never
1847 * exceed 3 bytes
1849 if (max_byte_length < byte_length ||
1850 byte_length <= max_byte_length - 4)
1851 goto corrupt;
1852 newsize = hunk_size + byte_length;
1853 data = xrealloc(data, newsize);
1854 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1855 goto corrupt;
1856 hunk_size = newsize;
1857 buffer += llen;
1858 size -= llen;
1861 frag = xcalloc(1, sizeof(*frag));
1862 frag->patch = inflate_it(data, hunk_size, origlen);
1863 frag->free_patch = 1;
1864 if (!frag->patch)
1865 goto corrupt;
1866 free(data);
1867 frag->size = origlen;
1868 *buf_p = buffer;
1869 *sz_p = size;
1870 *used_p = used;
1871 frag->binary_patch_method = patch_method;
1872 return frag;
1874 corrupt:
1875 free(data);
1876 *status_p = -1;
1877 error(_("corrupt binary patch at line %d: %.*s"),
1878 state_linenr-1, llen-1, buffer);
1879 return NULL;
1883 * Returns:
1884 * -1 in case of error,
1885 * the length of the parsed binary patch otherwise
1887 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1890 * We have read "GIT binary patch\n"; what follows is a line
1891 * that says the patch method (currently, either "literal" or
1892 * "delta") and the length of data before deflating; a
1893 * sequence of 'length-byte' followed by base-85 encoded data
1894 * follows.
1896 * When a binary patch is reversible, there is another binary
1897 * hunk in the same format, starting with patch method (either
1898 * "literal" or "delta") with the length of data, and a sequence
1899 * of length-byte + base-85 encoded data, terminated with another
1900 * empty line. This data, when applied to the postimage, produces
1901 * the preimage.
1903 struct fragment *forward;
1904 struct fragment *reverse;
1905 int status;
1906 int used, used_1;
1908 forward = parse_binary_hunk(&buffer, &size, &status, &used);
1909 if (!forward && !status)
1910 /* there has to be one hunk (forward hunk) */
1911 return error(_("unrecognized binary patch at line %d"), state_linenr-1);
1912 if (status)
1913 /* otherwise we already gave an error message */
1914 return status;
1916 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1917 if (reverse)
1918 used += used_1;
1919 else if (status) {
1921 * Not having reverse hunk is not an error, but having
1922 * a corrupt reverse hunk is.
1924 free((void*) forward->patch);
1925 free(forward);
1926 return status;
1928 forward->next = reverse;
1929 patch->fragments = forward;
1930 patch->is_binary = 1;
1931 return used;
1934 static void prefix_one(struct apply_state *state, char **name)
1936 char *old_name = *name;
1937 if (!old_name)
1938 return;
1939 *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));
1940 free(old_name);
1943 static void prefix_patch(struct apply_state *state, struct patch *p)
1945 if (!state->prefix || p->is_toplevel_relative)
1946 return;
1947 prefix_one(state, &p->new_name);
1948 prefix_one(state, &p->old_name);
1952 * include/exclude
1955 static struct string_list limit_by_name;
1956 static int has_include;
1957 static void add_name_limit(const char *name, int exclude)
1959 struct string_list_item *it;
1961 it = string_list_append(&limit_by_name, name);
1962 it->util = exclude ? NULL : (void *) 1;
1965 static int use_patch(struct apply_state *state, struct patch *p)
1967 const char *pathname = p->new_name ? p->new_name : p->old_name;
1968 int i;
1970 /* Paths outside are not touched regardless of "--include" */
1971 if (0 < state->prefix_length) {
1972 int pathlen = strlen(pathname);
1973 if (pathlen <= state->prefix_length ||
1974 memcmp(state->prefix, pathname, state->prefix_length))
1975 return 0;
1978 /* See if it matches any of exclude/include rule */
1979 for (i = 0; i < limit_by_name.nr; i++) {
1980 struct string_list_item *it = &limit_by_name.items[i];
1981 if (!wildmatch(it->string, pathname, 0, NULL))
1982 return (it->util != NULL);
1986 * If we had any include, a path that does not match any rule is
1987 * not used. Otherwise, we saw bunch of exclude rules (or none)
1988 * and such a path is used.
1990 return !has_include;
1995 * Read the patch text in "buffer" that extends for "size" bytes; stop
1996 * reading after seeing a single patch (i.e. changes to a single file).
1997 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1998 * Return the number of bytes consumed, so that the caller can call us
1999 * again for the next patch.
2001 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2003 int hdrsize, patchsize;
2004 int offset = find_header(state, buffer, size, &hdrsize, patch);
2006 if (offset < 0)
2007 return offset;
2009 prefix_patch(state, patch);
2011 if (!use_patch(state, patch))
2012 patch->ws_rule = 0;
2013 else
2014 patch->ws_rule = whitespace_rule(patch->new_name
2015 ? patch->new_name
2016 : patch->old_name);
2018 patchsize = parse_single_patch(state,
2019 buffer + offset + hdrsize,
2020 size - offset - hdrsize,
2021 patch);
2023 if (!patchsize) {
2024 static const char git_binary[] = "GIT binary patch\n";
2025 int hd = hdrsize + offset;
2026 unsigned long llen = linelen(buffer + hd, size - hd);
2028 if (llen == sizeof(git_binary) - 1 &&
2029 !memcmp(git_binary, buffer + hd, llen)) {
2030 int used;
2031 state_linenr++;
2032 used = parse_binary(buffer + hd + llen,
2033 size - hd - llen, patch);
2034 if (used < 0)
2035 return -1;
2036 if (used)
2037 patchsize = used + llen;
2038 else
2039 patchsize = 0;
2041 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2042 static const char *binhdr[] = {
2043 "Binary files ",
2044 "Files ",
2045 NULL,
2047 int i;
2048 for (i = 0; binhdr[i]; i++) {
2049 int len = strlen(binhdr[i]);
2050 if (len < size - hd &&
2051 !memcmp(binhdr[i], buffer + hd, len)) {
2052 state_linenr++;
2053 patch->is_binary = 1;
2054 patchsize = llen;
2055 break;
2060 /* Empty patch cannot be applied if it is a text patch
2061 * without metadata change. A binary patch appears
2062 * empty to us here.
2064 if ((apply || state->check) &&
2065 (!patch->is_binary && !metadata_changes(patch)))
2066 die(_("patch with only garbage at line %d"), state_linenr);
2069 return offset + hdrsize + patchsize;
2072 #define swap(a,b) myswap((a),(b),sizeof(a))
2074 #define myswap(a, b, size) do { \
2075 unsigned char mytmp[size]; \
2076 memcpy(mytmp, &a, size); \
2077 memcpy(&a, &b, size); \
2078 memcpy(&b, mytmp, size); \
2079 } while (0)
2081 static void reverse_patches(struct patch *p)
2083 for (; p; p = p->next) {
2084 struct fragment *frag = p->fragments;
2086 swap(p->new_name, p->old_name);
2087 swap(p->new_mode, p->old_mode);
2088 swap(p->is_new, p->is_delete);
2089 swap(p->lines_added, p->lines_deleted);
2090 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2092 for (; frag; frag = frag->next) {
2093 swap(frag->newpos, frag->oldpos);
2094 swap(frag->newlines, frag->oldlines);
2099 static const char pluses[] =
2100 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2101 static const char minuses[]=
2102 "----------------------------------------------------------------------";
2104 static void show_stats(struct patch *patch)
2106 struct strbuf qname = STRBUF_INIT;
2107 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2108 int max, add, del;
2110 quote_c_style(cp, &qname, NULL, 0);
2113 * "scale" the filename
2115 max = max_len;
2116 if (max > 50)
2117 max = 50;
2119 if (qname.len > max) {
2120 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2121 if (!cp)
2122 cp = qname.buf + qname.len + 3 - max;
2123 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2126 if (patch->is_binary) {
2127 printf(" %-*s | Bin\n", max, qname.buf);
2128 strbuf_release(&qname);
2129 return;
2132 printf(" %-*s |", max, qname.buf);
2133 strbuf_release(&qname);
2136 * scale the add/delete
2138 max = max + max_change > 70 ? 70 - max : max_change;
2139 add = patch->lines_added;
2140 del = patch->lines_deleted;
2142 if (max_change > 0) {
2143 int total = ((add + del) * max + max_change / 2) / max_change;
2144 add = (add * max + max_change / 2) / max_change;
2145 del = total - add;
2147 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2148 add, pluses, del, minuses);
2151 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2153 switch (st->st_mode & S_IFMT) {
2154 case S_IFLNK:
2155 if (strbuf_readlink(buf, path, st->st_size) < 0)
2156 return error(_("unable to read symlink %s"), path);
2157 return 0;
2158 case S_IFREG:
2159 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2160 return error(_("unable to open or read %s"), path);
2161 convert_to_git(path, buf->buf, buf->len, buf, 0);
2162 return 0;
2163 default:
2164 return -1;
2169 * Update the preimage, and the common lines in postimage,
2170 * from buffer buf of length len. If postlen is 0 the postimage
2171 * is updated in place, otherwise it's updated on a new buffer
2172 * of length postlen
2175 static void update_pre_post_images(struct image *preimage,
2176 struct image *postimage,
2177 char *buf,
2178 size_t len, size_t postlen)
2180 int i, ctx, reduced;
2181 char *new, *old, *fixed;
2182 struct image fixed_preimage;
2185 * Update the preimage with whitespace fixes. Note that we
2186 * are not losing preimage->buf -- apply_one_fragment() will
2187 * free "oldlines".
2189 prepare_image(&fixed_preimage, buf, len, 1);
2190 assert(postlen
2191 ? fixed_preimage.nr == preimage->nr
2192 : fixed_preimage.nr <= preimage->nr);
2193 for (i = 0; i < fixed_preimage.nr; i++)
2194 fixed_preimage.line[i].flag = preimage->line[i].flag;
2195 free(preimage->line_allocated);
2196 *preimage = fixed_preimage;
2199 * Adjust the common context lines in postimage. This can be
2200 * done in-place when we are shrinking it with whitespace
2201 * fixing, but needs a new buffer when ignoring whitespace or
2202 * expanding leading tabs to spaces.
2204 * We trust the caller to tell us if the update can be done
2205 * in place (postlen==0) or not.
2207 old = postimage->buf;
2208 if (postlen)
2209 new = postimage->buf = xmalloc(postlen);
2210 else
2211 new = old;
2212 fixed = preimage->buf;
2214 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2215 size_t l_len = postimage->line[i].len;
2216 if (!(postimage->line[i].flag & LINE_COMMON)) {
2217 /* an added line -- no counterparts in preimage */
2218 memmove(new, old, l_len);
2219 old += l_len;
2220 new += l_len;
2221 continue;
2224 /* a common context -- skip it in the original postimage */
2225 old += l_len;
2227 /* and find the corresponding one in the fixed preimage */
2228 while (ctx < preimage->nr &&
2229 !(preimage->line[ctx].flag & LINE_COMMON)) {
2230 fixed += preimage->line[ctx].len;
2231 ctx++;
2235 * preimage is expected to run out, if the caller
2236 * fixed addition of trailing blank lines.
2238 if (preimage->nr <= ctx) {
2239 reduced++;
2240 continue;
2243 /* and copy it in, while fixing the line length */
2244 l_len = preimage->line[ctx].len;
2245 memcpy(new, fixed, l_len);
2246 new += l_len;
2247 fixed += l_len;
2248 postimage->line[i].len = l_len;
2249 ctx++;
2252 if (postlen
2253 ? postlen < new - postimage->buf
2254 : postimage->len < new - postimage->buf)
2255 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2256 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2258 /* Fix the length of the whole thing */
2259 postimage->len = new - postimage->buf;
2260 postimage->nr -= reduced;
2263 static int line_by_line_fuzzy_match(struct image *img,
2264 struct image *preimage,
2265 struct image *postimage,
2266 unsigned long try,
2267 int try_lno,
2268 int preimage_limit)
2270 int i;
2271 size_t imgoff = 0;
2272 size_t preoff = 0;
2273 size_t postlen = postimage->len;
2274 size_t extra_chars;
2275 char *buf;
2276 char *preimage_eof;
2277 char *preimage_end;
2278 struct strbuf fixed;
2279 char *fixed_buf;
2280 size_t fixed_len;
2282 for (i = 0; i < preimage_limit; i++) {
2283 size_t prelen = preimage->line[i].len;
2284 size_t imglen = img->line[try_lno+i].len;
2286 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2287 preimage->buf + preoff, prelen))
2288 return 0;
2289 if (preimage->line[i].flag & LINE_COMMON)
2290 postlen += imglen - prelen;
2291 imgoff += imglen;
2292 preoff += prelen;
2296 * Ok, the preimage matches with whitespace fuzz.
2298 * imgoff now holds the true length of the target that
2299 * matches the preimage before the end of the file.
2301 * Count the number of characters in the preimage that fall
2302 * beyond the end of the file and make sure that all of them
2303 * are whitespace characters. (This can only happen if
2304 * we are removing blank lines at the end of the file.)
2306 buf = preimage_eof = preimage->buf + preoff;
2307 for ( ; i < preimage->nr; i++)
2308 preoff += preimage->line[i].len;
2309 preimage_end = preimage->buf + preoff;
2310 for ( ; buf < preimage_end; buf++)
2311 if (!isspace(*buf))
2312 return 0;
2315 * Update the preimage and the common postimage context
2316 * lines to use the same whitespace as the target.
2317 * If whitespace is missing in the target (i.e.
2318 * if the preimage extends beyond the end of the file),
2319 * use the whitespace from the preimage.
2321 extra_chars = preimage_end - preimage_eof;
2322 strbuf_init(&fixed, imgoff + extra_chars);
2323 strbuf_add(&fixed, img->buf + try, imgoff);
2324 strbuf_add(&fixed, preimage_eof, extra_chars);
2325 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2326 update_pre_post_images(preimage, postimage,
2327 fixed_buf, fixed_len, postlen);
2328 return 1;
2331 static int match_fragment(struct image *img,
2332 struct image *preimage,
2333 struct image *postimage,
2334 unsigned long try,
2335 int try_lno,
2336 unsigned ws_rule,
2337 int match_beginning, int match_end)
2339 int i;
2340 char *fixed_buf, *buf, *orig, *target;
2341 struct strbuf fixed;
2342 size_t fixed_len, postlen;
2343 int preimage_limit;
2345 if (preimage->nr + try_lno <= img->nr) {
2347 * The hunk falls within the boundaries of img.
2349 preimage_limit = preimage->nr;
2350 if (match_end && (preimage->nr + try_lno != img->nr))
2351 return 0;
2352 } else if (ws_error_action == correct_ws_error &&
2353 (ws_rule & WS_BLANK_AT_EOF)) {
2355 * This hunk extends beyond the end of img, and we are
2356 * removing blank lines at the end of the file. This
2357 * many lines from the beginning of the preimage must
2358 * match with img, and the remainder of the preimage
2359 * must be blank.
2361 preimage_limit = img->nr - try_lno;
2362 } else {
2364 * The hunk extends beyond the end of the img and
2365 * we are not removing blanks at the end, so we
2366 * should reject the hunk at this position.
2368 return 0;
2371 if (match_beginning && try_lno)
2372 return 0;
2374 /* Quick hash check */
2375 for (i = 0; i < preimage_limit; i++)
2376 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2377 (preimage->line[i].hash != img->line[try_lno + i].hash))
2378 return 0;
2380 if (preimage_limit == preimage->nr) {
2382 * Do we have an exact match? If we were told to match
2383 * at the end, size must be exactly at try+fragsize,
2384 * otherwise try+fragsize must be still within the preimage,
2385 * and either case, the old piece should match the preimage
2386 * exactly.
2388 if ((match_end
2389 ? (try + preimage->len == img->len)
2390 : (try + preimage->len <= img->len)) &&
2391 !memcmp(img->buf + try, preimage->buf, preimage->len))
2392 return 1;
2393 } else {
2395 * The preimage extends beyond the end of img, so
2396 * there cannot be an exact match.
2398 * There must be one non-blank context line that match
2399 * a line before the end of img.
2401 char *buf_end;
2403 buf = preimage->buf;
2404 buf_end = buf;
2405 for (i = 0; i < preimage_limit; i++)
2406 buf_end += preimage->line[i].len;
2408 for ( ; buf < buf_end; buf++)
2409 if (!isspace(*buf))
2410 break;
2411 if (buf == buf_end)
2412 return 0;
2416 * No exact match. If we are ignoring whitespace, run a line-by-line
2417 * fuzzy matching. We collect all the line length information because
2418 * we need it to adjust whitespace if we match.
2420 if (ws_ignore_action == ignore_ws_change)
2421 return line_by_line_fuzzy_match(img, preimage, postimage,
2422 try, try_lno, preimage_limit);
2424 if (ws_error_action != correct_ws_error)
2425 return 0;
2428 * The hunk does not apply byte-by-byte, but the hash says
2429 * it might with whitespace fuzz. We weren't asked to
2430 * ignore whitespace, we were asked to correct whitespace
2431 * errors, so let's try matching after whitespace correction.
2433 * While checking the preimage against the target, whitespace
2434 * errors in both fixed, we count how large the corresponding
2435 * postimage needs to be. The postimage prepared by
2436 * apply_one_fragment() has whitespace errors fixed on added
2437 * lines already, but the common lines were propagated as-is,
2438 * which may become longer when their whitespace errors are
2439 * fixed.
2442 /* First count added lines in postimage */
2443 postlen = 0;
2444 for (i = 0; i < postimage->nr; i++) {
2445 if (!(postimage->line[i].flag & LINE_COMMON))
2446 postlen += postimage->line[i].len;
2450 * The preimage may extend beyond the end of the file,
2451 * but in this loop we will only handle the part of the
2452 * preimage that falls within the file.
2454 strbuf_init(&fixed, preimage->len + 1);
2455 orig = preimage->buf;
2456 target = img->buf + try;
2457 for (i = 0; i < preimage_limit; i++) {
2458 size_t oldlen = preimage->line[i].len;
2459 size_t tgtlen = img->line[try_lno + i].len;
2460 size_t fixstart = fixed.len;
2461 struct strbuf tgtfix;
2462 int match;
2464 /* Try fixing the line in the preimage */
2465 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2467 /* Try fixing the line in the target */
2468 strbuf_init(&tgtfix, tgtlen);
2469 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2472 * If they match, either the preimage was based on
2473 * a version before our tree fixed whitespace breakage,
2474 * or we are lacking a whitespace-fix patch the tree
2475 * the preimage was based on already had (i.e. target
2476 * has whitespace breakage, the preimage doesn't).
2477 * In either case, we are fixing the whitespace breakages
2478 * so we might as well take the fix together with their
2479 * real change.
2481 match = (tgtfix.len == fixed.len - fixstart &&
2482 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2483 fixed.len - fixstart));
2485 /* Add the length if this is common with the postimage */
2486 if (preimage->line[i].flag & LINE_COMMON)
2487 postlen += tgtfix.len;
2489 strbuf_release(&tgtfix);
2490 if (!match)
2491 goto unmatch_exit;
2493 orig += oldlen;
2494 target += tgtlen;
2499 * Now handle the lines in the preimage that falls beyond the
2500 * end of the file (if any). They will only match if they are
2501 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2502 * false).
2504 for ( ; i < preimage->nr; i++) {
2505 size_t fixstart = fixed.len; /* start of the fixed preimage */
2506 size_t oldlen = preimage->line[i].len;
2507 int j;
2509 /* Try fixing the line in the preimage */
2510 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2512 for (j = fixstart; j < fixed.len; j++)
2513 if (!isspace(fixed.buf[j]))
2514 goto unmatch_exit;
2516 orig += oldlen;
2520 * Yes, the preimage is based on an older version that still
2521 * has whitespace breakages unfixed, and fixing them makes the
2522 * hunk match. Update the context lines in the postimage.
2524 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2525 if (postlen < postimage->len)
2526 postlen = 0;
2527 update_pre_post_images(preimage, postimage,
2528 fixed_buf, fixed_len, postlen);
2529 return 1;
2531 unmatch_exit:
2532 strbuf_release(&fixed);
2533 return 0;
2536 static int find_pos(struct image *img,
2537 struct image *preimage,
2538 struct image *postimage,
2539 int line,
2540 unsigned ws_rule,
2541 int match_beginning, int match_end)
2543 int i;
2544 unsigned long backwards, forwards, try;
2545 int backwards_lno, forwards_lno, try_lno;
2548 * If match_beginning or match_end is specified, there is no
2549 * point starting from a wrong line that will never match and
2550 * wander around and wait for a match at the specified end.
2552 if (match_beginning)
2553 line = 0;
2554 else if (match_end)
2555 line = img->nr - preimage->nr;
2558 * Because the comparison is unsigned, the following test
2559 * will also take care of a negative line number that can
2560 * result when match_end and preimage is larger than the target.
2562 if ((size_t) line > img->nr)
2563 line = img->nr;
2565 try = 0;
2566 for (i = 0; i < line; i++)
2567 try += img->line[i].len;
2570 * There's probably some smart way to do this, but I'll leave
2571 * that to the smart and beautiful people. I'm simple and stupid.
2573 backwards = try;
2574 backwards_lno = line;
2575 forwards = try;
2576 forwards_lno = line;
2577 try_lno = line;
2579 for (i = 0; ; i++) {
2580 if (match_fragment(img, preimage, postimage,
2581 try, try_lno, ws_rule,
2582 match_beginning, match_end))
2583 return try_lno;
2585 again:
2586 if (backwards_lno == 0 && forwards_lno == img->nr)
2587 break;
2589 if (i & 1) {
2590 if (backwards_lno == 0) {
2591 i++;
2592 goto again;
2594 backwards_lno--;
2595 backwards -= img->line[backwards_lno].len;
2596 try = backwards;
2597 try_lno = backwards_lno;
2598 } else {
2599 if (forwards_lno == img->nr) {
2600 i++;
2601 goto again;
2603 forwards += img->line[forwards_lno].len;
2604 forwards_lno++;
2605 try = forwards;
2606 try_lno = forwards_lno;
2610 return -1;
2613 static void remove_first_line(struct image *img)
2615 img->buf += img->line[0].len;
2616 img->len -= img->line[0].len;
2617 img->line++;
2618 img->nr--;
2621 static void remove_last_line(struct image *img)
2623 img->len -= img->line[--img->nr].len;
2627 * The change from "preimage" and "postimage" has been found to
2628 * apply at applied_pos (counts in line numbers) in "img".
2629 * Update "img" to remove "preimage" and replace it with "postimage".
2631 static void update_image(struct apply_state *state,
2632 struct image *img,
2633 int applied_pos,
2634 struct image *preimage,
2635 struct image *postimage)
2638 * remove the copy of preimage at offset in img
2639 * and replace it with postimage
2641 int i, nr;
2642 size_t remove_count, insert_count, applied_at = 0;
2643 char *result;
2644 int preimage_limit;
2647 * If we are removing blank lines at the end of img,
2648 * the preimage may extend beyond the end.
2649 * If that is the case, we must be careful only to
2650 * remove the part of the preimage that falls within
2651 * the boundaries of img. Initialize preimage_limit
2652 * to the number of lines in the preimage that falls
2653 * within the boundaries.
2655 preimage_limit = preimage->nr;
2656 if (preimage_limit > img->nr - applied_pos)
2657 preimage_limit = img->nr - applied_pos;
2659 for (i = 0; i < applied_pos; i++)
2660 applied_at += img->line[i].len;
2662 remove_count = 0;
2663 for (i = 0; i < preimage_limit; i++)
2664 remove_count += img->line[applied_pos + i].len;
2665 insert_count = postimage->len;
2667 /* Adjust the contents */
2668 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2669 memcpy(result, img->buf, applied_at);
2670 memcpy(result + applied_at, postimage->buf, postimage->len);
2671 memcpy(result + applied_at + postimage->len,
2672 img->buf + (applied_at + remove_count),
2673 img->len - (applied_at + remove_count));
2674 free(img->buf);
2675 img->buf = result;
2676 img->len += insert_count - remove_count;
2677 result[img->len] = '\0';
2679 /* Adjust the line table */
2680 nr = img->nr + postimage->nr - preimage_limit;
2681 if (preimage_limit < postimage->nr) {
2683 * NOTE: this knows that we never call remove_first_line()
2684 * on anything other than pre/post image.
2686 REALLOC_ARRAY(img->line, nr);
2687 img->line_allocated = img->line;
2689 if (preimage_limit != postimage->nr)
2690 memmove(img->line + applied_pos + postimage->nr,
2691 img->line + applied_pos + preimage_limit,
2692 (img->nr - (applied_pos + preimage_limit)) *
2693 sizeof(*img->line));
2694 memcpy(img->line + applied_pos,
2695 postimage->line,
2696 postimage->nr * sizeof(*img->line));
2697 if (!state->allow_overlap)
2698 for (i = 0; i < postimage->nr; i++)
2699 img->line[applied_pos + i].flag |= LINE_PATCHED;
2700 img->nr = nr;
2704 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2705 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2706 * replace the part of "img" with "postimage" text.
2708 static int apply_one_fragment(struct apply_state *state,
2709 struct image *img, struct fragment *frag,
2710 int inaccurate_eof, unsigned ws_rule,
2711 int nth_fragment)
2713 int match_beginning, match_end;
2714 const char *patch = frag->patch;
2715 int size = frag->size;
2716 char *old, *oldlines;
2717 struct strbuf newlines;
2718 int new_blank_lines_at_end = 0;
2719 int found_new_blank_lines_at_end = 0;
2720 int hunk_linenr = frag->linenr;
2721 unsigned long leading, trailing;
2722 int pos, applied_pos;
2723 struct image preimage;
2724 struct image postimage;
2726 memset(&preimage, 0, sizeof(preimage));
2727 memset(&postimage, 0, sizeof(postimage));
2728 oldlines = xmalloc(size);
2729 strbuf_init(&newlines, size);
2731 old = oldlines;
2732 while (size > 0) {
2733 char first;
2734 int len = linelen(patch, size);
2735 int plen;
2736 int added_blank_line = 0;
2737 int is_blank_context = 0;
2738 size_t start;
2740 if (!len)
2741 break;
2744 * "plen" is how much of the line we should use for
2745 * the actual patch data. Normally we just remove the
2746 * first character on the line, but if the line is
2747 * followed by "\ No newline", then we also remove the
2748 * last one (which is the newline, of course).
2750 plen = len - 1;
2751 if (len < size && patch[len] == '\\')
2752 plen--;
2753 first = *patch;
2754 if (state->apply_in_reverse) {
2755 if (first == '-')
2756 first = '+';
2757 else if (first == '+')
2758 first = '-';
2761 switch (first) {
2762 case '\n':
2763 /* Newer GNU diff, empty context line */
2764 if (plen < 0)
2765 /* ... followed by '\No newline'; nothing */
2766 break;
2767 *old++ = '\n';
2768 strbuf_addch(&newlines, '\n');
2769 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2770 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2771 is_blank_context = 1;
2772 break;
2773 case ' ':
2774 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2775 ws_blank_line(patch + 1, plen, ws_rule))
2776 is_blank_context = 1;
2777 case '-':
2778 memcpy(old, patch + 1, plen);
2779 add_line_info(&preimage, old, plen,
2780 (first == ' ' ? LINE_COMMON : 0));
2781 old += plen;
2782 if (first == '-')
2783 break;
2784 /* Fall-through for ' ' */
2785 case '+':
2786 /* --no-add does not add new lines */
2787 if (first == '+' && state->no_add)
2788 break;
2790 start = newlines.len;
2791 if (first != '+' ||
2792 !whitespace_error ||
2793 ws_error_action != correct_ws_error) {
2794 strbuf_add(&newlines, patch + 1, plen);
2796 else {
2797 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);
2799 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2800 (first == '+' ? 0 : LINE_COMMON));
2801 if (first == '+' &&
2802 (ws_rule & WS_BLANK_AT_EOF) &&
2803 ws_blank_line(patch + 1, plen, ws_rule))
2804 added_blank_line = 1;
2805 break;
2806 case '@': case '\\':
2807 /* Ignore it, we already handled it */
2808 break;
2809 default:
2810 if (state->apply_verbosely)
2811 error(_("invalid start of line: '%c'"), first);
2812 applied_pos = -1;
2813 goto out;
2815 if (added_blank_line) {
2816 if (!new_blank_lines_at_end)
2817 found_new_blank_lines_at_end = hunk_linenr;
2818 new_blank_lines_at_end++;
2820 else if (is_blank_context)
2822 else
2823 new_blank_lines_at_end = 0;
2824 patch += len;
2825 size -= len;
2826 hunk_linenr++;
2828 if (inaccurate_eof &&
2829 old > oldlines && old[-1] == '\n' &&
2830 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2831 old--;
2832 strbuf_setlen(&newlines, newlines.len - 1);
2835 leading = frag->leading;
2836 trailing = frag->trailing;
2839 * A hunk to change lines at the beginning would begin with
2840 * @@ -1,L +N,M @@
2841 * but we need to be careful. -U0 that inserts before the second
2842 * line also has this pattern.
2844 * And a hunk to add to an empty file would begin with
2845 * @@ -0,0 +N,M @@
2847 * In other words, a hunk that is (frag->oldpos <= 1) with or
2848 * without leading context must match at the beginning.
2850 match_beginning = (!frag->oldpos ||
2851 (frag->oldpos == 1 && !state->unidiff_zero));
2854 * A hunk without trailing lines must match at the end.
2855 * However, we simply cannot tell if a hunk must match end
2856 * from the lack of trailing lines if the patch was generated
2857 * with unidiff without any context.
2859 match_end = !state->unidiff_zero && !trailing;
2861 pos = frag->newpos ? (frag->newpos - 1) : 0;
2862 preimage.buf = oldlines;
2863 preimage.len = old - oldlines;
2864 postimage.buf = newlines.buf;
2865 postimage.len = newlines.len;
2866 preimage.line = preimage.line_allocated;
2867 postimage.line = postimage.line_allocated;
2869 for (;;) {
2871 applied_pos = find_pos(img, &preimage, &postimage, pos,
2872 ws_rule, match_beginning, match_end);
2874 if (applied_pos >= 0)
2875 break;
2877 /* Am I at my context limits? */
2878 if ((leading <= p_context) && (trailing <= p_context))
2879 break;
2880 if (match_beginning || match_end) {
2881 match_beginning = match_end = 0;
2882 continue;
2886 * Reduce the number of context lines; reduce both
2887 * leading and trailing if they are equal otherwise
2888 * just reduce the larger context.
2890 if (leading >= trailing) {
2891 remove_first_line(&preimage);
2892 remove_first_line(&postimage);
2893 pos--;
2894 leading--;
2896 if (trailing > leading) {
2897 remove_last_line(&preimage);
2898 remove_last_line(&postimage);
2899 trailing--;
2903 if (applied_pos >= 0) {
2904 if (new_blank_lines_at_end &&
2905 preimage.nr + applied_pos >= img->nr &&
2906 (ws_rule & WS_BLANK_AT_EOF) &&
2907 ws_error_action != nowarn_ws_error) {
2908 record_ws_error(WS_BLANK_AT_EOF, "+", 1,
2909 found_new_blank_lines_at_end);
2910 if (ws_error_action == correct_ws_error) {
2911 while (new_blank_lines_at_end--)
2912 remove_last_line(&postimage);
2915 * We would want to prevent write_out_results()
2916 * from taking place in apply_patch() that follows
2917 * the callchain led us here, which is:
2918 * apply_patch->check_patch_list->check_patch->
2919 * apply_data->apply_fragments->apply_one_fragment
2921 if (ws_error_action == die_on_ws_error)
2922 apply = 0;
2925 if (state->apply_verbosely && applied_pos != pos) {
2926 int offset = applied_pos - pos;
2927 if (state->apply_in_reverse)
2928 offset = 0 - offset;
2929 fprintf_ln(stderr,
2930 Q_("Hunk #%d succeeded at %d (offset %d line).",
2931 "Hunk #%d succeeded at %d (offset %d lines).",
2932 offset),
2933 nth_fragment, applied_pos + 1, offset);
2937 * Warn if it was necessary to reduce the number
2938 * of context lines.
2940 if ((leading != frag->leading) ||
2941 (trailing != frag->trailing))
2942 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2943 " to apply fragment at %d"),
2944 leading, trailing, applied_pos+1);
2945 update_image(state, img, applied_pos, &preimage, &postimage);
2946 } else {
2947 if (state->apply_verbosely)
2948 error(_("while searching for:\n%.*s"),
2949 (int)(old - oldlines), oldlines);
2952 out:
2953 free(oldlines);
2954 strbuf_release(&newlines);
2955 free(preimage.line_allocated);
2956 free(postimage.line_allocated);
2958 return (applied_pos < 0);
2961 static int apply_binary_fragment(struct apply_state *state,
2962 struct image *img,
2963 struct patch *patch)
2965 struct fragment *fragment = patch->fragments;
2966 unsigned long len;
2967 void *dst;
2969 if (!fragment)
2970 return error(_("missing binary patch data for '%s'"),
2971 patch->new_name ?
2972 patch->new_name :
2973 patch->old_name);
2975 /* Binary patch is irreversible without the optional second hunk */
2976 if (state->apply_in_reverse) {
2977 if (!fragment->next)
2978 return error("cannot reverse-apply a binary patch "
2979 "without the reverse hunk to '%s'",
2980 patch->new_name
2981 ? patch->new_name : patch->old_name);
2982 fragment = fragment->next;
2984 switch (fragment->binary_patch_method) {
2985 case BINARY_DELTA_DEFLATED:
2986 dst = patch_delta(img->buf, img->len, fragment->patch,
2987 fragment->size, &len);
2988 if (!dst)
2989 return -1;
2990 clear_image(img);
2991 img->buf = dst;
2992 img->len = len;
2993 return 0;
2994 case BINARY_LITERAL_DEFLATED:
2995 clear_image(img);
2996 img->len = fragment->size;
2997 img->buf = xmemdupz(fragment->patch, img->len);
2998 return 0;
3000 return -1;
3004 * Replace "img" with the result of applying the binary patch.
3005 * The binary patch data itself in patch->fragment is still kept
3006 * but the preimage prepared by the caller in "img" is freed here
3007 * or in the helper function apply_binary_fragment() this calls.
3009 static int apply_binary(struct apply_state *state,
3010 struct image *img,
3011 struct patch *patch)
3013 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3014 unsigned char sha1[20];
3017 * For safety, we require patch index line to contain
3018 * full 40-byte textual SHA1 for old and new, at least for now.
3020 if (strlen(patch->old_sha1_prefix) != 40 ||
3021 strlen(patch->new_sha1_prefix) != 40 ||
3022 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
3023 get_sha1_hex(patch->new_sha1_prefix, sha1))
3024 return error("cannot apply binary patch to '%s' "
3025 "without full index line", name);
3027 if (patch->old_name) {
3029 * See if the old one matches what the patch
3030 * applies to.
3032 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3033 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
3034 return error("the patch applies to '%s' (%s), "
3035 "which does not match the "
3036 "current contents.",
3037 name, sha1_to_hex(sha1));
3039 else {
3040 /* Otherwise, the old one must be empty. */
3041 if (img->len)
3042 return error("the patch applies to an empty "
3043 "'%s' but it is not empty", name);
3046 get_sha1_hex(patch->new_sha1_prefix, sha1);
3047 if (is_null_sha1(sha1)) {
3048 clear_image(img);
3049 return 0; /* deletion patch */
3052 if (has_sha1_file(sha1)) {
3053 /* We already have the postimage */
3054 enum object_type type;
3055 unsigned long size;
3056 char *result;
3058 result = read_sha1_file(sha1, &type, &size);
3059 if (!result)
3060 return error("the necessary postimage %s for "
3061 "'%s' cannot be read",
3062 patch->new_sha1_prefix, name);
3063 clear_image(img);
3064 img->buf = result;
3065 img->len = size;
3066 } else {
3068 * We have verified buf matches the preimage;
3069 * apply the patch data to it, which is stored
3070 * in the patch->fragments->{patch,size}.
3072 if (apply_binary_fragment(state, img, patch))
3073 return error(_("binary patch does not apply to '%s'"),
3074 name);
3076 /* verify that the result matches */
3077 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3078 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3079 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3080 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3083 return 0;
3086 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3088 struct fragment *frag = patch->fragments;
3089 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3090 unsigned ws_rule = patch->ws_rule;
3091 unsigned inaccurate_eof = patch->inaccurate_eof;
3092 int nth = 0;
3094 if (patch->is_binary)
3095 return apply_binary(state, img, patch);
3097 while (frag) {
3098 nth++;
3099 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3100 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3101 if (!state->apply_with_reject)
3102 return -1;
3103 frag->rejected = 1;
3105 frag = frag->next;
3107 return 0;
3110 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3112 if (S_ISGITLINK(mode)) {
3113 strbuf_grow(buf, 100);
3114 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3115 } else {
3116 enum object_type type;
3117 unsigned long sz;
3118 char *result;
3120 result = read_sha1_file(sha1, &type, &sz);
3121 if (!result)
3122 return -1;
3123 /* XXX read_sha1_file NUL-terminates */
3124 strbuf_attach(buf, result, sz, sz + 1);
3126 return 0;
3129 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3131 if (!ce)
3132 return 0;
3133 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3136 static struct patch *in_fn_table(const char *name)
3138 struct string_list_item *item;
3140 if (name == NULL)
3141 return NULL;
3143 item = string_list_lookup(&fn_table, name);
3144 if (item != NULL)
3145 return (struct patch *)item->util;
3147 return NULL;
3151 * item->util in the filename table records the status of the path.
3152 * Usually it points at a patch (whose result records the contents
3153 * of it after applying it), but it could be PATH_WAS_DELETED for a
3154 * path that a previously applied patch has already removed, or
3155 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3157 * The latter is needed to deal with a case where two paths A and B
3158 * are swapped by first renaming A to B and then renaming B to A;
3159 * moving A to B should not be prevented due to presence of B as we
3160 * will remove it in a later patch.
3162 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3163 #define PATH_WAS_DELETED ((struct patch *) -1)
3165 static int to_be_deleted(struct patch *patch)
3167 return patch == PATH_TO_BE_DELETED;
3170 static int was_deleted(struct patch *patch)
3172 return patch == PATH_WAS_DELETED;
3175 static void add_to_fn_table(struct patch *patch)
3177 struct string_list_item *item;
3180 * Always add new_name unless patch is a deletion
3181 * This should cover the cases for normal diffs,
3182 * file creations and copies
3184 if (patch->new_name != NULL) {
3185 item = string_list_insert(&fn_table, patch->new_name);
3186 item->util = patch;
3190 * store a failure on rename/deletion cases because
3191 * later chunks shouldn't patch old names
3193 if ((patch->new_name == NULL) || (patch->is_rename)) {
3194 item = string_list_insert(&fn_table, patch->old_name);
3195 item->util = PATH_WAS_DELETED;
3199 static void prepare_fn_table(struct patch *patch)
3202 * store information about incoming file deletion
3204 while (patch) {
3205 if ((patch->new_name == NULL) || (patch->is_rename)) {
3206 struct string_list_item *item;
3207 item = string_list_insert(&fn_table, patch->old_name);
3208 item->util = PATH_TO_BE_DELETED;
3210 patch = patch->next;
3214 static int checkout_target(struct index_state *istate,
3215 struct cache_entry *ce, struct stat *st)
3217 struct checkout costate;
3219 memset(&costate, 0, sizeof(costate));
3220 costate.base_dir = "";
3221 costate.refresh_cache = 1;
3222 costate.istate = istate;
3223 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3224 return error(_("cannot checkout %s"), ce->name);
3225 return 0;
3228 static struct patch *previous_patch(struct patch *patch, int *gone)
3230 struct patch *previous;
3232 *gone = 0;
3233 if (patch->is_copy || patch->is_rename)
3234 return NULL; /* "git" patches do not depend on the order */
3236 previous = in_fn_table(patch->old_name);
3237 if (!previous)
3238 return NULL;
3240 if (to_be_deleted(previous))
3241 return NULL; /* the deletion hasn't happened yet */
3243 if (was_deleted(previous))
3244 *gone = 1;
3246 return previous;
3249 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3251 if (S_ISGITLINK(ce->ce_mode)) {
3252 if (!S_ISDIR(st->st_mode))
3253 return -1;
3254 return 0;
3256 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3259 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3261 static int load_patch_target(struct apply_state *state,
3262 struct strbuf *buf,
3263 const struct cache_entry *ce,
3264 struct stat *st,
3265 const char *name,
3266 unsigned expected_mode)
3268 if (state->cached || state->check_index) {
3269 if (read_file_or_gitlink(ce, buf))
3270 return error(_("read of %s failed"), name);
3271 } else if (name) {
3272 if (S_ISGITLINK(expected_mode)) {
3273 if (ce)
3274 return read_file_or_gitlink(ce, buf);
3275 else
3276 return SUBMODULE_PATCH_WITHOUT_INDEX;
3277 } else if (has_symlink_leading_path(name, strlen(name))) {
3278 return error(_("reading from '%s' beyond a symbolic link"), name);
3279 } else {
3280 if (read_old_data(st, name, buf))
3281 return error(_("read of %s failed"), name);
3284 return 0;
3288 * We are about to apply "patch"; populate the "image" with the
3289 * current version we have, from the working tree or from the index,
3290 * depending on the situation e.g. --cached/--index. If we are
3291 * applying a non-git patch that incrementally updates the tree,
3292 * we read from the result of a previous diff.
3294 static int load_preimage(struct apply_state *state,
3295 struct image *image,
3296 struct patch *patch, struct stat *st,
3297 const struct cache_entry *ce)
3299 struct strbuf buf = STRBUF_INIT;
3300 size_t len;
3301 char *img;
3302 struct patch *previous;
3303 int status;
3305 previous = previous_patch(patch, &status);
3306 if (status)
3307 return error(_("path %s has been renamed/deleted"),
3308 patch->old_name);
3309 if (previous) {
3310 /* We have a patched copy in memory; use that. */
3311 strbuf_add(&buf, previous->result, previous->resultsize);
3312 } else {
3313 status = load_patch_target(state, &buf, ce, st,
3314 patch->old_name, patch->old_mode);
3315 if (status < 0)
3316 return status;
3317 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3319 * There is no way to apply subproject
3320 * patch without looking at the index.
3321 * NEEDSWORK: shouldn't this be flagged
3322 * as an error???
3324 free_fragment_list(patch->fragments);
3325 patch->fragments = NULL;
3326 } else if (status) {
3327 return error(_("read of %s failed"), patch->old_name);
3331 img = strbuf_detach(&buf, &len);
3332 prepare_image(image, img, len, !patch->is_binary);
3333 return 0;
3336 static int three_way_merge(struct image *image,
3337 char *path,
3338 const unsigned char *base,
3339 const unsigned char *ours,
3340 const unsigned char *theirs)
3342 mmfile_t base_file, our_file, their_file;
3343 mmbuffer_t result = { NULL };
3344 int status;
3346 read_mmblob(&base_file, base);
3347 read_mmblob(&our_file, ours);
3348 read_mmblob(&their_file, theirs);
3349 status = ll_merge(&result, path,
3350 &base_file, "base",
3351 &our_file, "ours",
3352 &their_file, "theirs", NULL);
3353 free(base_file.ptr);
3354 free(our_file.ptr);
3355 free(their_file.ptr);
3356 if (status < 0 || !result.ptr) {
3357 free(result.ptr);
3358 return -1;
3360 clear_image(image);
3361 image->buf = result.ptr;
3362 image->len = result.size;
3364 return status;
3368 * When directly falling back to add/add three-way merge, we read from
3369 * the current contents of the new_name. In no cases other than that
3370 * this function will be called.
3372 static int load_current(struct apply_state *state,
3373 struct image *image,
3374 struct patch *patch)
3376 struct strbuf buf = STRBUF_INIT;
3377 int status, pos;
3378 size_t len;
3379 char *img;
3380 struct stat st;
3381 struct cache_entry *ce;
3382 char *name = patch->new_name;
3383 unsigned mode = patch->new_mode;
3385 if (!patch->is_new)
3386 die("BUG: patch to %s is not a creation", patch->old_name);
3388 pos = cache_name_pos(name, strlen(name));
3389 if (pos < 0)
3390 return error(_("%s: does not exist in index"), name);
3391 ce = active_cache[pos];
3392 if (lstat(name, &st)) {
3393 if (errno != ENOENT)
3394 return error(_("%s: %s"), name, strerror(errno));
3395 if (checkout_target(&the_index, ce, &st))
3396 return -1;
3398 if (verify_index_match(ce, &st))
3399 return error(_("%s: does not match index"), name);
3401 status = load_patch_target(state, &buf, ce, &st, name, mode);
3402 if (status < 0)
3403 return status;
3404 else if (status)
3405 return -1;
3406 img = strbuf_detach(&buf, &len);
3407 prepare_image(image, img, len, !patch->is_binary);
3408 return 0;
3411 static int try_threeway(struct apply_state *state,
3412 struct image *image,
3413 struct patch *patch,
3414 struct stat *st,
3415 const struct cache_entry *ce)
3417 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3418 struct strbuf buf = STRBUF_INIT;
3419 size_t len;
3420 int status;
3421 char *img;
3422 struct image tmp_image;
3424 /* No point falling back to 3-way merge in these cases */
3425 if (patch->is_delete ||
3426 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3427 return -1;
3429 /* Preimage the patch was prepared for */
3430 if (patch->is_new)
3431 write_sha1_file("", 0, blob_type, pre_sha1);
3432 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3433 read_blob_object(&buf, pre_sha1, patch->old_mode))
3434 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3436 fprintf(stderr, "Falling back to three-way merge...\n");
3438 img = strbuf_detach(&buf, &len);
3439 prepare_image(&tmp_image, img, len, 1);
3440 /* Apply the patch to get the post image */
3441 if (apply_fragments(state, &tmp_image, patch) < 0) {
3442 clear_image(&tmp_image);
3443 return -1;
3445 /* post_sha1[] is theirs */
3446 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3447 clear_image(&tmp_image);
3449 /* our_sha1[] is ours */
3450 if (patch->is_new) {
3451 if (load_current(state, &tmp_image, patch))
3452 return error("cannot read the current contents of '%s'",
3453 patch->new_name);
3454 } else {
3455 if (load_preimage(state, &tmp_image, patch, st, ce))
3456 return error("cannot read the current contents of '%s'",
3457 patch->old_name);
3459 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3460 clear_image(&tmp_image);
3462 /* in-core three-way merge between post and our using pre as base */
3463 status = three_way_merge(image, patch->new_name,
3464 pre_sha1, our_sha1, post_sha1);
3465 if (status < 0) {
3466 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3467 return status;
3470 if (status) {
3471 patch->conflicted_threeway = 1;
3472 if (patch->is_new)
3473 oidclr(&patch->threeway_stage[0]);
3474 else
3475 hashcpy(patch->threeway_stage[0].hash, pre_sha1);
3476 hashcpy(patch->threeway_stage[1].hash, our_sha1);
3477 hashcpy(patch->threeway_stage[2].hash, post_sha1);
3478 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3479 } else {
3480 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3482 return 0;
3485 static int apply_data(struct apply_state *state, struct patch *patch,
3486 struct stat *st, const struct cache_entry *ce)
3488 struct image image;
3490 if (load_preimage(state, &image, patch, st, ce) < 0)
3491 return -1;
3493 if (patch->direct_to_threeway ||
3494 apply_fragments(state, &image, patch) < 0) {
3495 /* Note: with --reject, apply_fragments() returns 0 */
3496 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3497 return -1;
3499 patch->result = image.buf;
3500 patch->resultsize = image.len;
3501 add_to_fn_table(patch);
3502 free(image.line_allocated);
3504 if (0 < patch->is_delete && patch->resultsize)
3505 return error(_("removal patch leaves file contents"));
3507 return 0;
3511 * If "patch" that we are looking at modifies or deletes what we have,
3512 * we would want it not to lose any local modification we have, either
3513 * in the working tree or in the index.
3515 * This also decides if a non-git patch is a creation patch or a
3516 * modification to an existing empty file. We do not check the state
3517 * of the current tree for a creation patch in this function; the caller
3518 * check_patch() separately makes sure (and errors out otherwise) that
3519 * the path the patch creates does not exist in the current tree.
3521 static int check_preimage(struct apply_state *state,
3522 struct patch *patch,
3523 struct cache_entry **ce,
3524 struct stat *st)
3526 const char *old_name = patch->old_name;
3527 struct patch *previous = NULL;
3528 int stat_ret = 0, status;
3529 unsigned st_mode = 0;
3531 if (!old_name)
3532 return 0;
3534 assert(patch->is_new <= 0);
3535 previous = previous_patch(patch, &status);
3537 if (status)
3538 return error(_("path %s has been renamed/deleted"), old_name);
3539 if (previous) {
3540 st_mode = previous->new_mode;
3541 } else if (!state->cached) {
3542 stat_ret = lstat(old_name, st);
3543 if (stat_ret && errno != ENOENT)
3544 return error(_("%s: %s"), old_name, strerror(errno));
3547 if (state->check_index && !previous) {
3548 int pos = cache_name_pos(old_name, strlen(old_name));
3549 if (pos < 0) {
3550 if (patch->is_new < 0)
3551 goto is_new;
3552 return error(_("%s: does not exist in index"), old_name);
3554 *ce = active_cache[pos];
3555 if (stat_ret < 0) {
3556 if (checkout_target(&the_index, *ce, st))
3557 return -1;
3559 if (!state->cached && verify_index_match(*ce, st))
3560 return error(_("%s: does not match index"), old_name);
3561 if (state->cached)
3562 st_mode = (*ce)->ce_mode;
3563 } else if (stat_ret < 0) {
3564 if (patch->is_new < 0)
3565 goto is_new;
3566 return error(_("%s: %s"), old_name, strerror(errno));
3569 if (!state->cached && !previous)
3570 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3572 if (patch->is_new < 0)
3573 patch->is_new = 0;
3574 if (!patch->old_mode)
3575 patch->old_mode = st_mode;
3576 if ((st_mode ^ patch->old_mode) & S_IFMT)
3577 return error(_("%s: wrong type"), old_name);
3578 if (st_mode != patch->old_mode)
3579 warning(_("%s has type %o, expected %o"),
3580 old_name, st_mode, patch->old_mode);
3581 if (!patch->new_mode && !patch->is_delete)
3582 patch->new_mode = st_mode;
3583 return 0;
3585 is_new:
3586 patch->is_new = 1;
3587 patch->is_delete = 0;
3588 free(patch->old_name);
3589 patch->old_name = NULL;
3590 return 0;
3594 #define EXISTS_IN_INDEX 1
3595 #define EXISTS_IN_WORKTREE 2
3597 static int check_to_create(struct apply_state *state,
3598 const char *new_name,
3599 int ok_if_exists)
3601 struct stat nst;
3603 if (state->check_index &&
3604 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3605 !ok_if_exists)
3606 return EXISTS_IN_INDEX;
3607 if (state->cached)
3608 return 0;
3610 if (!lstat(new_name, &nst)) {
3611 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3612 return 0;
3614 * A leading component of new_name might be a symlink
3615 * that is going to be removed with this patch, but
3616 * still pointing at somewhere that has the path.
3617 * In such a case, path "new_name" does not exist as
3618 * far as git is concerned.
3620 if (has_symlink_leading_path(new_name, strlen(new_name)))
3621 return 0;
3623 return EXISTS_IN_WORKTREE;
3624 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3625 return error("%s: %s", new_name, strerror(errno));
3627 return 0;
3631 * We need to keep track of how symlinks in the preimage are
3632 * manipulated by the patches. A patch to add a/b/c where a/b
3633 * is a symlink should not be allowed to affect the directory
3634 * the symlink points at, but if the same patch removes a/b,
3635 * it is perfectly fine, as the patch removes a/b to make room
3636 * to create a directory a/b so that a/b/c can be created.
3638 static struct string_list symlink_changes;
3639 #define SYMLINK_GOES_AWAY 01
3640 #define SYMLINK_IN_RESULT 02
3642 static uintptr_t register_symlink_changes(const char *path, uintptr_t what)
3644 struct string_list_item *ent;
3646 ent = string_list_lookup(&symlink_changes, path);
3647 if (!ent) {
3648 ent = string_list_insert(&symlink_changes, path);
3649 ent->util = (void *)0;
3651 ent->util = (void *)(what | ((uintptr_t)ent->util));
3652 return (uintptr_t)ent->util;
3655 static uintptr_t check_symlink_changes(const char *path)
3657 struct string_list_item *ent;
3659 ent = string_list_lookup(&symlink_changes, path);
3660 if (!ent)
3661 return 0;
3662 return (uintptr_t)ent->util;
3665 static void prepare_symlink_changes(struct patch *patch)
3667 for ( ; patch; patch = patch->next) {
3668 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3669 (patch->is_rename || patch->is_delete))
3670 /* the symlink at patch->old_name is removed */
3671 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);
3673 if (patch->new_name && S_ISLNK(patch->new_mode))
3674 /* the symlink at patch->new_name is created or remains */
3675 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);
3679 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3681 do {
3682 unsigned int change;
3684 while (--name->len && name->buf[name->len] != '/')
3685 ; /* scan backwards */
3686 if (!name->len)
3687 break;
3688 name->buf[name->len] = '\0';
3689 change = check_symlink_changes(name->buf);
3690 if (change & SYMLINK_IN_RESULT)
3691 return 1;
3692 if (change & SYMLINK_GOES_AWAY)
3694 * This cannot be "return 0", because we may
3695 * see a new one created at a higher level.
3697 continue;
3699 /* otherwise, check the preimage */
3700 if (state->check_index) {
3701 struct cache_entry *ce;
3703 ce = cache_file_exists(name->buf, name->len, ignore_case);
3704 if (ce && S_ISLNK(ce->ce_mode))
3705 return 1;
3706 } else {
3707 struct stat st;
3708 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3709 return 1;
3711 } while (1);
3712 return 0;
3715 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3717 int ret;
3718 struct strbuf name = STRBUF_INIT;
3720 assert(*name_ != '\0');
3721 strbuf_addstr(&name, name_);
3722 ret = path_is_beyond_symlink_1(state, &name);
3723 strbuf_release(&name);
3725 return ret;
3728 static void die_on_unsafe_path(struct patch *patch)
3730 const char *old_name = NULL;
3731 const char *new_name = NULL;
3732 if (patch->is_delete)
3733 old_name = patch->old_name;
3734 else if (!patch->is_new && !patch->is_copy)
3735 old_name = patch->old_name;
3736 if (!patch->is_delete)
3737 new_name = patch->new_name;
3739 if (old_name && !verify_path(old_name))
3740 die(_("invalid path '%s'"), old_name);
3741 if (new_name && !verify_path(new_name))
3742 die(_("invalid path '%s'"), new_name);
3746 * Check and apply the patch in-core; leave the result in patch->result
3747 * for the caller to write it out to the final destination.
3749 static int check_patch(struct apply_state *state, struct patch *patch)
3751 struct stat st;
3752 const char *old_name = patch->old_name;
3753 const char *new_name = patch->new_name;
3754 const char *name = old_name ? old_name : new_name;
3755 struct cache_entry *ce = NULL;
3756 struct patch *tpatch;
3757 int ok_if_exists;
3758 int status;
3760 patch->rejected = 1; /* we will drop this after we succeed */
3762 status = check_preimage(state, patch, &ce, &st);
3763 if (status)
3764 return status;
3765 old_name = patch->old_name;
3768 * A type-change diff is always split into a patch to delete
3769 * old, immediately followed by a patch to create new (see
3770 * diff.c::run_diff()); in such a case it is Ok that the entry
3771 * to be deleted by the previous patch is still in the working
3772 * tree and in the index.
3774 * A patch to swap-rename between A and B would first rename A
3775 * to B and then rename B to A. While applying the first one,
3776 * the presence of B should not stop A from getting renamed to
3777 * B; ask to_be_deleted() about the later rename. Removal of
3778 * B and rename from A to B is handled the same way by asking
3779 * was_deleted().
3781 if ((tpatch = in_fn_table(new_name)) &&
3782 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3783 ok_if_exists = 1;
3784 else
3785 ok_if_exists = 0;
3787 if (new_name &&
3788 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3789 int err = check_to_create(state, new_name, ok_if_exists);
3791 if (err && state->threeway) {
3792 patch->direct_to_threeway = 1;
3793 } else switch (err) {
3794 case 0:
3795 break; /* happy */
3796 case EXISTS_IN_INDEX:
3797 return error(_("%s: already exists in index"), new_name);
3798 break;
3799 case EXISTS_IN_WORKTREE:
3800 return error(_("%s: already exists in working directory"),
3801 new_name);
3802 default:
3803 return err;
3806 if (!patch->new_mode) {
3807 if (0 < patch->is_new)
3808 patch->new_mode = S_IFREG | 0644;
3809 else
3810 patch->new_mode = patch->old_mode;
3814 if (new_name && old_name) {
3815 int same = !strcmp(old_name, new_name);
3816 if (!patch->new_mode)
3817 patch->new_mode = patch->old_mode;
3818 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3819 if (same)
3820 return error(_("new mode (%o) of %s does not "
3821 "match old mode (%o)"),
3822 patch->new_mode, new_name,
3823 patch->old_mode);
3824 else
3825 return error(_("new mode (%o) of %s does not "
3826 "match old mode (%o) of %s"),
3827 patch->new_mode, new_name,
3828 patch->old_mode, old_name);
3832 if (!state->unsafe_paths)
3833 die_on_unsafe_path(patch);
3836 * An attempt to read from or delete a path that is beyond a
3837 * symbolic link will be prevented by load_patch_target() that
3838 * is called at the beginning of apply_data() so we do not
3839 * have to worry about a patch marked with "is_delete" bit
3840 * here. We however need to make sure that the patch result
3841 * is not deposited to a path that is beyond a symbolic link
3842 * here.
3844 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3845 return error(_("affected file '%s' is beyond a symbolic link"),
3846 patch->new_name);
3848 if (apply_data(state, patch, &st, ce) < 0)
3849 return error(_("%s: patch does not apply"), name);
3850 patch->rejected = 0;
3851 return 0;
3854 static int check_patch_list(struct apply_state *state, struct patch *patch)
3856 int err = 0;
3858 prepare_symlink_changes(patch);
3859 prepare_fn_table(patch);
3860 while (patch) {
3861 if (state->apply_verbosely)
3862 say_patch_name(stderr,
3863 _("Checking patch %s..."), patch);
3864 err |= check_patch(state, patch);
3865 patch = patch->next;
3867 return err;
3870 /* This function tries to read the sha1 from the current index */
3871 static int get_current_sha1(const char *path, unsigned char *sha1)
3873 int pos;
3875 if (read_cache() < 0)
3876 return -1;
3877 pos = cache_name_pos(path, strlen(path));
3878 if (pos < 0)
3879 return -1;
3880 hashcpy(sha1, active_cache[pos]->sha1);
3881 return 0;
3884 static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])
3887 * A usable gitlink patch has only one fragment (hunk) that looks like:
3888 * @@ -1 +1 @@
3889 * -Subproject commit <old sha1>
3890 * +Subproject commit <new sha1>
3891 * or
3892 * @@ -1 +0,0 @@
3893 * -Subproject commit <old sha1>
3894 * for a removal patch.
3896 struct fragment *hunk = p->fragments;
3897 static const char heading[] = "-Subproject commit ";
3898 char *preimage;
3900 if (/* does the patch have only one hunk? */
3901 hunk && !hunk->next &&
3902 /* is its preimage one line? */
3903 hunk->oldpos == 1 && hunk->oldlines == 1 &&
3904 /* does preimage begin with the heading? */
3905 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
3906 starts_with(++preimage, heading) &&
3907 /* does it record full SHA-1? */
3908 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&
3909 preimage[sizeof(heading) + 40 - 1] == '\n' &&
3910 /* does the abbreviated name on the index line agree with it? */
3911 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
3912 return 0; /* it all looks fine */
3914 /* we may have full object name on the index line */
3915 return get_sha1_hex(p->old_sha1_prefix, sha1);
3918 /* Build an index that contains the just the files needed for a 3way merge */
3919 static void build_fake_ancestor(struct patch *list, const char *filename)
3921 struct patch *patch;
3922 struct index_state result = { NULL };
3923 static struct lock_file lock;
3925 /* Once we start supporting the reverse patch, it may be
3926 * worth showing the new sha1 prefix, but until then...
3928 for (patch = list; patch; patch = patch->next) {
3929 unsigned char sha1[20];
3930 struct cache_entry *ce;
3931 const char *name;
3933 name = patch->old_name ? patch->old_name : patch->new_name;
3934 if (0 < patch->is_new)
3935 continue;
3937 if (S_ISGITLINK(patch->old_mode)) {
3938 if (!preimage_sha1_in_gitlink_patch(patch, sha1))
3939 ; /* ok, the textual part looks sane */
3940 else
3941 die("sha1 information is lacking or useless for submodule %s",
3942 name);
3943 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {
3944 ; /* ok */
3945 } else if (!patch->lines_added && !patch->lines_deleted) {
3946 /* mode-only change: update the current */
3947 if (get_current_sha1(patch->old_name, sha1))
3948 die("mode change for %s, which is not "
3949 "in current HEAD", name);
3950 } else
3951 die("sha1 information is lacking or useless "
3952 "(%s).", name);
3954 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);
3955 if (!ce)
3956 die(_("make_cache_entry failed for path '%s'"), name);
3957 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))
3958 die ("Could not add %s to temporary index", name);
3961 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);
3962 if (write_locked_index(&result, &lock, COMMIT_LOCK))
3963 die ("Could not write temporary index to %s", filename);
3965 discard_index(&result);
3968 static void stat_patch_list(struct patch *patch)
3970 int files, adds, dels;
3972 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
3973 files++;
3974 adds += patch->lines_added;
3975 dels += patch->lines_deleted;
3976 show_stats(patch);
3979 print_stat_summary(stdout, files, adds, dels);
3982 static void numstat_patch_list(struct apply_state *state,
3983 struct patch *patch)
3985 for ( ; patch; patch = patch->next) {
3986 const char *name;
3987 name = patch->new_name ? patch->new_name : patch->old_name;
3988 if (patch->is_binary)
3989 printf("-\t-\t");
3990 else
3991 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
3992 write_name_quoted(name, stdout, state->line_termination);
3996 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
3998 if (mode)
3999 printf(" %s mode %06o %s\n", newdelete, mode, name);
4000 else
4001 printf(" %s %s\n", newdelete, name);
4004 static void show_mode_change(struct patch *p, int show_name)
4006 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4007 if (show_name)
4008 printf(" mode change %06o => %06o %s\n",
4009 p->old_mode, p->new_mode, p->new_name);
4010 else
4011 printf(" mode change %06o => %06o\n",
4012 p->old_mode, p->new_mode);
4016 static void show_rename_copy(struct patch *p)
4018 const char *renamecopy = p->is_rename ? "rename" : "copy";
4019 const char *old, *new;
4021 /* Find common prefix */
4022 old = p->old_name;
4023 new = p->new_name;
4024 while (1) {
4025 const char *slash_old, *slash_new;
4026 slash_old = strchr(old, '/');
4027 slash_new = strchr(new, '/');
4028 if (!slash_old ||
4029 !slash_new ||
4030 slash_old - old != slash_new - new ||
4031 memcmp(old, new, slash_new - new))
4032 break;
4033 old = slash_old + 1;
4034 new = slash_new + 1;
4036 /* p->old_name thru old is the common prefix, and old and new
4037 * through the end of names are renames
4039 if (old != p->old_name)
4040 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4041 (int)(old - p->old_name), p->old_name,
4042 old, new, p->score);
4043 else
4044 printf(" %s %s => %s (%d%%)\n", renamecopy,
4045 p->old_name, p->new_name, p->score);
4046 show_mode_change(p, 0);
4049 static void summary_patch_list(struct patch *patch)
4051 struct patch *p;
4053 for (p = patch; p; p = p->next) {
4054 if (p->is_new)
4055 show_file_mode_name("create", p->new_mode, p->new_name);
4056 else if (p->is_delete)
4057 show_file_mode_name("delete", p->old_mode, p->old_name);
4058 else {
4059 if (p->is_rename || p->is_copy)
4060 show_rename_copy(p);
4061 else {
4062 if (p->score) {
4063 printf(" rewrite %s (%d%%)\n",
4064 p->new_name, p->score);
4065 show_mode_change(p, 0);
4067 else
4068 show_mode_change(p, 1);
4074 static void patch_stats(struct patch *patch)
4076 int lines = patch->lines_added + patch->lines_deleted;
4078 if (lines > max_change)
4079 max_change = lines;
4080 if (patch->old_name) {
4081 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4082 if (!len)
4083 len = strlen(patch->old_name);
4084 if (len > max_len)
4085 max_len = len;
4087 if (patch->new_name) {
4088 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4089 if (!len)
4090 len = strlen(patch->new_name);
4091 if (len > max_len)
4092 max_len = len;
4096 static void remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4098 if (state->update_index) {
4099 if (remove_file_from_cache(patch->old_name) < 0)
4100 die(_("unable to remove %s from index"), patch->old_name);
4102 if (!state->cached) {
4103 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4104 remove_path(patch->old_name);
4109 static void add_index_file(struct apply_state *state,
4110 const char *path,
4111 unsigned mode,
4112 void *buf,
4113 unsigned long size)
4115 struct stat st;
4116 struct cache_entry *ce;
4117 int namelen = strlen(path);
4118 unsigned ce_size = cache_entry_size(namelen);
4120 if (!state->update_index)
4121 return;
4123 ce = xcalloc(1, ce_size);
4124 memcpy(ce->name, path, namelen);
4125 ce->ce_mode = create_ce_mode(mode);
4126 ce->ce_flags = create_ce_flags(0);
4127 ce->ce_namelen = namelen;
4128 if (S_ISGITLINK(mode)) {
4129 const char *s;
4131 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4132 get_sha1_hex(s, ce->sha1))
4133 die(_("corrupt patch for submodule %s"), path);
4134 } else {
4135 if (!state->cached) {
4136 if (lstat(path, &st) < 0)
4137 die_errno(_("unable to stat newly created file '%s'"),
4138 path);
4139 fill_stat_cache_info(ce, &st);
4141 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
4142 die(_("unable to create backing store for newly created file %s"), path);
4144 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4145 die(_("unable to add cache entry for %s"), path);
4148 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4150 int fd;
4151 struct strbuf nbuf = STRBUF_INIT;
4153 if (S_ISGITLINK(mode)) {
4154 struct stat st;
4155 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4156 return 0;
4157 return mkdir(path, 0777);
4160 if (has_symlinks && S_ISLNK(mode))
4161 /* Although buf:size is counted string, it also is NUL
4162 * terminated.
4164 return symlink(buf, path);
4166 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4167 if (fd < 0)
4168 return -1;
4170 if (convert_to_working_tree(path, buf, size, &nbuf)) {
4171 size = nbuf.len;
4172 buf = nbuf.buf;
4174 write_or_die(fd, buf, size);
4175 strbuf_release(&nbuf);
4177 if (close(fd) < 0)
4178 die_errno(_("closing file '%s'"), path);
4179 return 0;
4183 * We optimistically assume that the directories exist,
4184 * which is true 99% of the time anyway. If they don't,
4185 * we create them and try again.
4187 static void create_one_file(struct apply_state *state,
4188 char *path,
4189 unsigned mode,
4190 const char *buf,
4191 unsigned long size)
4193 if (state->cached)
4194 return;
4195 if (!try_create_file(path, mode, buf, size))
4196 return;
4198 if (errno == ENOENT) {
4199 if (safe_create_leading_directories(path))
4200 return;
4201 if (!try_create_file(path, mode, buf, size))
4202 return;
4205 if (errno == EEXIST || errno == EACCES) {
4206 /* We may be trying to create a file where a directory
4207 * used to be.
4209 struct stat st;
4210 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4211 errno = EEXIST;
4214 if (errno == EEXIST) {
4215 unsigned int nr = getpid();
4217 for (;;) {
4218 char newpath[PATH_MAX];
4219 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4220 if (!try_create_file(newpath, mode, buf, size)) {
4221 if (!rename(newpath, path))
4222 return;
4223 unlink_or_warn(newpath);
4224 break;
4226 if (errno != EEXIST)
4227 break;
4228 ++nr;
4231 die_errno(_("unable to write file '%s' mode %o"), path, mode);
4234 static void add_conflicted_stages_file(struct apply_state *state,
4235 struct patch *patch)
4237 int stage, namelen;
4238 unsigned ce_size, mode;
4239 struct cache_entry *ce;
4241 if (!state->update_index)
4242 return;
4243 namelen = strlen(patch->new_name);
4244 ce_size = cache_entry_size(namelen);
4245 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4247 remove_file_from_cache(patch->new_name);
4248 for (stage = 1; stage < 4; stage++) {
4249 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4250 continue;
4251 ce = xcalloc(1, ce_size);
4252 memcpy(ce->name, patch->new_name, namelen);
4253 ce->ce_mode = create_ce_mode(mode);
4254 ce->ce_flags = create_ce_flags(stage);
4255 ce->ce_namelen = namelen;
4256 hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);
4257 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
4258 die(_("unable to add cache entry for %s"), patch->new_name);
4262 static void create_file(struct apply_state *state, struct patch *patch)
4264 char *path = patch->new_name;
4265 unsigned mode = patch->new_mode;
4266 unsigned long size = patch->resultsize;
4267 char *buf = patch->result;
4269 if (!mode)
4270 mode = S_IFREG | 0644;
4271 create_one_file(state, path, mode, buf, size);
4273 if (patch->conflicted_threeway)
4274 add_conflicted_stages_file(state, patch);
4275 else
4276 add_index_file(state, path, mode, buf, size);
4279 /* phase zero is to remove, phase one is to create */
4280 static void write_out_one_result(struct apply_state *state,
4281 struct patch *patch,
4282 int phase)
4284 if (patch->is_delete > 0) {
4285 if (phase == 0)
4286 remove_file(state, patch, 1);
4287 return;
4289 if (patch->is_new > 0 || patch->is_copy) {
4290 if (phase == 1)
4291 create_file(state, patch);
4292 return;
4295 * Rename or modification boils down to the same
4296 * thing: remove the old, write the new
4298 if (phase == 0)
4299 remove_file(state, patch, patch->is_rename);
4300 if (phase == 1)
4301 create_file(state, patch);
4304 static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4306 FILE *rej;
4307 char namebuf[PATH_MAX];
4308 struct fragment *frag;
4309 int cnt = 0;
4310 struct strbuf sb = STRBUF_INIT;
4312 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4313 if (!frag->rejected)
4314 continue;
4315 cnt++;
4318 if (!cnt) {
4319 if (state->apply_verbosely)
4320 say_patch_name(stderr,
4321 _("Applied patch %s cleanly."), patch);
4322 return 0;
4325 /* This should not happen, because a removal patch that leaves
4326 * contents are marked "rejected" at the patch level.
4328 if (!patch->new_name)
4329 die(_("internal error"));
4331 /* Say this even without --verbose */
4332 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4333 "Applying patch %%s with %d rejects...",
4334 cnt),
4335 cnt);
4336 say_patch_name(stderr, sb.buf, patch);
4337 strbuf_release(&sb);
4339 cnt = strlen(patch->new_name);
4340 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4341 cnt = ARRAY_SIZE(namebuf) - 5;
4342 warning(_("truncating .rej filename to %.*s.rej"),
4343 cnt - 1, patch->new_name);
4345 memcpy(namebuf, patch->new_name, cnt);
4346 memcpy(namebuf + cnt, ".rej", 5);
4348 rej = fopen(namebuf, "w");
4349 if (!rej)
4350 return error(_("cannot open %s: %s"), namebuf, strerror(errno));
4352 /* Normal git tools never deal with .rej, so do not pretend
4353 * this is a git patch by saying --git or giving extended
4354 * headers. While at it, maybe please "kompare" that wants
4355 * the trailing TAB and some garbage at the end of line ;-).
4357 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4358 patch->new_name, patch->new_name);
4359 for (cnt = 1, frag = patch->fragments;
4360 frag;
4361 cnt++, frag = frag->next) {
4362 if (!frag->rejected) {
4363 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
4364 continue;
4366 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
4367 fprintf(rej, "%.*s", frag->size, frag->patch);
4368 if (frag->patch[frag->size-1] != '\n')
4369 fputc('\n', rej);
4371 fclose(rej);
4372 return -1;
4375 static int write_out_results(struct apply_state *state, struct patch *list)
4377 int phase;
4378 int errs = 0;
4379 struct patch *l;
4380 struct string_list cpath = STRING_LIST_INIT_DUP;
4382 for (phase = 0; phase < 2; phase++) {
4383 l = list;
4384 while (l) {
4385 if (l->rejected)
4386 errs = 1;
4387 else {
4388 write_out_one_result(state, l, phase);
4389 if (phase == 1) {
4390 if (write_out_one_reject(state, l))
4391 errs = 1;
4392 if (l->conflicted_threeway) {
4393 string_list_append(&cpath, l->new_name);
4394 errs = 1;
4398 l = l->next;
4402 if (cpath.nr) {
4403 struct string_list_item *item;
4405 string_list_sort(&cpath);
4406 for_each_string_list_item(item, &cpath)
4407 fprintf(stderr, "U %s\n", item->string);
4408 string_list_clear(&cpath, 0);
4410 rerere(0);
4413 return errs;
4416 static struct lock_file lock_file;
4418 #define INACCURATE_EOF (1<<0)
4419 #define RECOUNT (1<<1)
4421 static int apply_patch(struct apply_state *state,
4422 int fd,
4423 const char *filename,
4424 int options)
4426 size_t offset;
4427 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4428 struct patch *list = NULL, **listp = &list;
4429 int skipped_patch = 0;
4431 patch_input_file = filename;
4432 read_patch_file(&buf, fd);
4433 offset = 0;
4434 while (offset < buf.len) {
4435 struct patch *patch;
4436 int nr;
4438 patch = xcalloc(1, sizeof(*patch));
4439 patch->inaccurate_eof = !!(options & INACCURATE_EOF);
4440 patch->recount = !!(options & RECOUNT);
4441 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4442 if (nr < 0) {
4443 free_patch(patch);
4444 break;
4446 if (state->apply_in_reverse)
4447 reverse_patches(patch);
4448 if (use_patch(state, patch)) {
4449 patch_stats(patch);
4450 *listp = patch;
4451 listp = &patch->next;
4453 else {
4454 if (state->apply_verbosely)
4455 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4456 free_patch(patch);
4457 skipped_patch++;
4459 offset += nr;
4462 if (!list && !skipped_patch)
4463 die(_("unrecognized input"));
4465 if (whitespace_error && (ws_error_action == die_on_ws_error))
4466 apply = 0;
4468 state->update_index = state->check_index && apply;
4469 if (state->update_index && newfd < 0)
4470 newfd = hold_locked_index(&lock_file, 1);
4472 if (state->check_index) {
4473 if (read_cache() < 0)
4474 die(_("unable to read index file"));
4477 if ((state->check || apply) &&
4478 check_patch_list(state, list) < 0 &&
4479 !state->apply_with_reject)
4480 exit(1);
4482 if (apply && write_out_results(state, list)) {
4483 if (state->apply_with_reject)
4484 exit(1);
4485 /* with --3way, we still need to write the index out */
4486 return 1;
4489 if (fake_ancestor)
4490 build_fake_ancestor(list, fake_ancestor);
4492 if (state->diffstat)
4493 stat_patch_list(list);
4495 if (state->numstat)
4496 numstat_patch_list(state, list);
4498 if (state->summary)
4499 summary_patch_list(list);
4501 free_patch_list(list);
4502 strbuf_release(&buf);
4503 string_list_clear(&fn_table, 0);
4504 return 0;
4507 static void git_apply_config(void)
4509 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
4510 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
4511 git_config(git_default_config, NULL);
4514 static int option_parse_exclude(const struct option *opt,
4515 const char *arg, int unset)
4517 add_name_limit(arg, 1);
4518 return 0;
4521 static int option_parse_include(const struct option *opt,
4522 const char *arg, int unset)
4524 add_name_limit(arg, 0);
4525 has_include = 1;
4526 return 0;
4529 static int option_parse_p(const struct option *opt,
4530 const char *arg, int unset)
4532 state_p_value = atoi(arg);
4533 p_value_known = 1;
4534 return 0;
4537 static int option_parse_space_change(const struct option *opt,
4538 const char *arg, int unset)
4540 if (unset)
4541 ws_ignore_action = ignore_ws_none;
4542 else
4543 ws_ignore_action = ignore_ws_change;
4544 return 0;
4547 static int option_parse_whitespace(const struct option *opt,
4548 const char *arg, int unset)
4550 const char **whitespace_option = opt->value;
4552 *whitespace_option = arg;
4553 parse_whitespace_option(arg);
4554 return 0;
4557 static int option_parse_directory(const struct option *opt,
4558 const char *arg, int unset)
4560 strbuf_reset(&root);
4561 strbuf_addstr(&root, arg);
4562 strbuf_complete(&root, '/');
4563 return 0;
4566 static void init_apply_state(struct apply_state *state, const char *prefix)
4568 memset(state, 0, sizeof(*state));
4569 state->prefix = prefix;
4570 state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
4571 state->line_termination = '\n';
4573 git_apply_config();
4574 if (apply_default_whitespace)
4575 parse_whitespace_option(apply_default_whitespace);
4576 if (apply_default_ignorewhitespace)
4577 parse_ignorewhitespace_option(apply_default_ignorewhitespace);
4580 static void clear_apply_state(struct apply_state *state)
4582 /* empty for now */
4585 int cmd_apply(int argc, const char **argv, const char *prefix)
4587 int i;
4588 int errs = 0;
4589 int is_not_gitdir = !startup_info->have_repository;
4590 int force_apply = 0;
4591 int options = 0;
4592 int read_stdin = 1;
4593 struct apply_state state;
4595 const char *whitespace_option = NULL;
4597 struct option builtin_apply_options[] = {
4598 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),
4599 N_("don't apply changes matching the given path"),
4600 0, option_parse_exclude },
4601 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),
4602 N_("apply changes matching the given path"),
4603 0, option_parse_include },
4604 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),
4605 N_("remove <num> leading slashes from traditional diff paths"),
4606 0, option_parse_p },
4607 OPT_BOOL(0, "no-add", &state.no_add,
4608 N_("ignore additions made by the patch")),
4609 OPT_BOOL(0, "stat", &state.diffstat,
4610 N_("instead of applying the patch, output diffstat for the input")),
4611 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4612 OPT_NOOP_NOARG(0, "binary"),
4613 OPT_BOOL(0, "numstat", &state.numstat,
4614 N_("show number of added and deleted lines in decimal notation")),
4615 OPT_BOOL(0, "summary", &state.summary,
4616 N_("instead of applying the patch, output a summary for the input")),
4617 OPT_BOOL(0, "check", &state.check,
4618 N_("instead of applying the patch, see if the patch is applicable")),
4619 OPT_BOOL(0, "index", &state.check_index,
4620 N_("make sure the patch is applicable to the current index")),
4621 OPT_BOOL(0, "cached", &state.cached,
4622 N_("apply a patch without touching the working tree")),
4623 OPT_BOOL(0, "unsafe-paths", &state.unsafe_paths,
4624 N_("accept a patch that touches outside the working area")),
4625 OPT_BOOL(0, "apply", &force_apply,
4626 N_("also apply the patch (use with --stat/--summary/--check)")),
4627 OPT_BOOL('3', "3way", &state.threeway,
4628 N_( "attempt three-way merge if a patch does not apply")),
4629 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,
4630 N_("build a temporary index based on embedded index information")),
4631 /* Think twice before adding "--nul" synonym to this */
4632 OPT_SET_INT('z', NULL, &state.line_termination,
4633 N_("paths are separated with NUL character"), '\0'),
4634 OPT_INTEGER('C', NULL, &p_context,
4635 N_("ensure at least <n> lines of context match")),
4636 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),
4637 N_("detect new or modified lines that have whitespace errors"),
4638 0, option_parse_whitespace },
4639 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,
4640 N_("ignore changes in whitespace when finding context"),
4641 PARSE_OPT_NOARG, option_parse_space_change },
4642 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,
4643 N_("ignore changes in whitespace when finding context"),
4644 PARSE_OPT_NOARG, option_parse_space_change },
4645 OPT_BOOL('R', "reverse", &state.apply_in_reverse,
4646 N_("apply the patch in reverse")),
4647 OPT_BOOL(0, "unidiff-zero", &state.unidiff_zero,
4648 N_("don't expect at least one line of context")),
4649 OPT_BOOL(0, "reject", &state.apply_with_reject,
4650 N_("leave the rejected hunks in corresponding *.rej files")),
4651 OPT_BOOL(0, "allow-overlap", &state.allow_overlap,
4652 N_("allow overlapping hunks")),
4653 OPT__VERBOSE(&state.apply_verbosely, N_("be verbose")),
4654 OPT_BIT(0, "inaccurate-eof", &options,
4655 N_("tolerate incorrectly detected missing new-line at the end of file"),
4656 INACCURATE_EOF),
4657 OPT_BIT(0, "recount", &options,
4658 N_("do not trust the line counts in the hunk headers"),
4659 RECOUNT),
4660 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),
4661 N_("prepend <root> to all filenames"),
4662 0, option_parse_directory },
4663 OPT_END()
4666 init_apply_state(&state, prefix);
4668 argc = parse_options(argc, argv, state.prefix, builtin_apply_options,
4669 apply_usage, 0);
4671 if (state.apply_with_reject && state.threeway)
4672 die("--reject and --3way cannot be used together.");
4673 if (state.cached && state.threeway)
4674 die("--cached and --3way cannot be used together.");
4675 if (state.threeway) {
4676 if (is_not_gitdir)
4677 die(_("--3way outside a repository"));
4678 state.check_index = 1;
4680 if (state.apply_with_reject)
4681 apply = state.apply_verbosely = 1;
4682 if (!force_apply && (state.diffstat || state.numstat || state.summary || state.check || fake_ancestor))
4683 apply = 0;
4684 if (state.check_index && is_not_gitdir)
4685 die(_("--index outside a repository"));
4686 if (state.cached) {
4687 if (is_not_gitdir)
4688 die(_("--cached outside a repository"));
4689 state.check_index = 1;
4691 if (state.check_index)
4692 state.unsafe_paths = 0;
4694 for (i = 0; i < argc; i++) {
4695 const char *arg = argv[i];
4696 int fd;
4698 if (!strcmp(arg, "-")) {
4699 errs |= apply_patch(&state, 0, "<stdin>", options);
4700 read_stdin = 0;
4701 continue;
4702 } else if (0 < state.prefix_length)
4703 arg = prefix_filename(state.prefix,
4704 state.prefix_length,
4705 arg);
4707 fd = open(arg, O_RDONLY);
4708 if (fd < 0)
4709 die_errno(_("can't open patch '%s'"), arg);
4710 read_stdin = 0;
4711 set_default_whitespace_mode(whitespace_option);
4712 errs |= apply_patch(&state, fd, arg, options);
4713 close(fd);
4715 set_default_whitespace_mode(whitespace_option);
4716 if (read_stdin)
4717 errs |= apply_patch(&state, 0, "<stdin>", options);
4718 if (whitespace_error) {
4719 if (squelch_whitespace_errors &&
4720 squelch_whitespace_errors < whitespace_error) {
4721 int squelched =
4722 whitespace_error - squelch_whitespace_errors;
4723 warning(Q_("squelched %d whitespace error",
4724 "squelched %d whitespace errors",
4725 squelched),
4726 squelched);
4728 if (ws_error_action == die_on_ws_error)
4729 die(Q_("%d line adds whitespace errors.",
4730 "%d lines add whitespace errors.",
4731 whitespace_error),
4732 whitespace_error);
4733 if (applied_after_fixing_ws && apply)
4734 warning("%d line%s applied after"
4735 " fixing whitespace errors.",
4736 applied_after_fixing_ws,
4737 applied_after_fixing_ws == 1 ? "" : "s");
4738 else if (whitespace_error)
4739 warning(Q_("%d line adds whitespace errors.",
4740 "%d lines add whitespace errors.",
4741 whitespace_error),
4742 whitespace_error);
4745 if (state.update_index) {
4746 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))
4747 die(_("Unable to write new index file"));
4750 clear_apply_state(&state);
4752 return !!errs;