factor out refresh_and_write_cache function
[git/git.git] / apply.c
1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9
10 #include "cache.h"
11 #include "config.h"
12 #include "object-store.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "diff.h"
16 #include "dir.h"
17 #include "xdiff-interface.h"
18 #include "ll-merge.h"
19 #include "lockfile.h"
20 #include "parse-options.h"
21 #include "quote.h"
22 #include "rerere.h"
23 #include "apply.h"
24
25 struct gitdiff_data {
26 struct strbuf *root;
27 int linenr;
28 int p_value;
29 };
30
31 static void git_apply_config(void)
32 {
33 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
34 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
35 git_config(git_default_config, NULL);
36 }
37
38 static int parse_whitespace_option(struct apply_state *state, const char *option)
39 {
40 if (!option) {
41 state->ws_error_action = warn_on_ws_error;
42 return 0;
43 }
44 if (!strcmp(option, "warn")) {
45 state->ws_error_action = warn_on_ws_error;
46 return 0;
47 }
48 if (!strcmp(option, "nowarn")) {
49 state->ws_error_action = nowarn_ws_error;
50 return 0;
51 }
52 if (!strcmp(option, "error")) {
53 state->ws_error_action = die_on_ws_error;
54 return 0;
55 }
56 if (!strcmp(option, "error-all")) {
57 state->ws_error_action = die_on_ws_error;
58 state->squelch_whitespace_errors = 0;
59 return 0;
60 }
61 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
62 state->ws_error_action = correct_ws_error;
63 return 0;
64 }
65 /*
66 * Please update $__git_whitespacelist in git-completion.bash
67 * when you add new options.
68 */
69 return error(_("unrecognized whitespace option '%s'"), option);
70 }
71
72 static int parse_ignorewhitespace_option(struct apply_state *state,
73 const char *option)
74 {
75 if (!option || !strcmp(option, "no") ||
76 !strcmp(option, "false") || !strcmp(option, "never") ||
77 !strcmp(option, "none")) {
78 state->ws_ignore_action = ignore_ws_none;
79 return 0;
80 }
81 if (!strcmp(option, "change")) {
82 state->ws_ignore_action = ignore_ws_change;
83 return 0;
84 }
85 return error(_("unrecognized whitespace ignore option '%s'"), option);
86 }
87
88 int init_apply_state(struct apply_state *state,
89 struct repository *repo,
90 const char *prefix)
91 {
92 memset(state, 0, sizeof(*state));
93 state->prefix = prefix;
94 state->repo = repo;
95 state->apply = 1;
96 state->line_termination = '\n';
97 state->p_value = 1;
98 state->p_context = UINT_MAX;
99 state->squelch_whitespace_errors = 5;
100 state->ws_error_action = warn_on_ws_error;
101 state->ws_ignore_action = ignore_ws_none;
102 state->linenr = 1;
103 string_list_init(&state->fn_table, 0);
104 string_list_init(&state->limit_by_name, 0);
105 string_list_init(&state->symlink_changes, 0);
106 strbuf_init(&state->root, 0);
107
108 git_apply_config();
109 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
110 return -1;
111 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
112 return -1;
113 return 0;
114 }
115
116 void clear_apply_state(struct apply_state *state)
117 {
118 string_list_clear(&state->limit_by_name, 0);
119 string_list_clear(&state->symlink_changes, 0);
120 strbuf_release(&state->root);
121
122 /* &state->fn_table is cleared at the end of apply_patch() */
123 }
124
125 static void mute_routine(const char *msg, va_list params)
126 {
127 /* do nothing */
128 }
129
130 int check_apply_state(struct apply_state *state, int force_apply)
131 {
132 int is_not_gitdir = !startup_info->have_repository;
133
134 if (state->apply_with_reject && state->threeway)
135 return error(_("--reject and --3way cannot be used together."));
136 if (state->cached && state->threeway)
137 return error(_("--cached and --3way cannot be used together."));
138 if (state->threeway) {
139 if (is_not_gitdir)
140 return error(_("--3way outside a repository"));
141 state->check_index = 1;
142 }
143 if (state->apply_with_reject) {
144 state->apply = 1;
145 if (state->apply_verbosity == verbosity_normal)
146 state->apply_verbosity = verbosity_verbose;
147 }
148 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
149 state->apply = 0;
150 if (state->check_index && is_not_gitdir)
151 return error(_("--index outside a repository"));
152 if (state->cached) {
153 if (is_not_gitdir)
154 return error(_("--cached outside a repository"));
155 state->check_index = 1;
156 }
157 if (state->ita_only && (state->check_index || is_not_gitdir))
158 state->ita_only = 0;
159 if (state->check_index)
160 state->unsafe_paths = 0;
161
162 if (state->apply_verbosity <= verbosity_silent) {
163 state->saved_error_routine = get_error_routine();
164 state->saved_warn_routine = get_warn_routine();
165 set_error_routine(mute_routine);
166 set_warn_routine(mute_routine);
167 }
168
169 return 0;
170 }
171
172 static void set_default_whitespace_mode(struct apply_state *state)
173 {
174 if (!state->whitespace_option && !apply_default_whitespace)
175 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
176 }
177
178 /*
179 * This represents one "hunk" from a patch, starting with
180 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
181 * patch text is pointed at by patch, and its byte length
182 * is stored in size. leading and trailing are the number
183 * of context lines.
184 */
185 struct fragment {
186 unsigned long leading, trailing;
187 unsigned long oldpos, oldlines;
188 unsigned long newpos, newlines;
189 /*
190 * 'patch' is usually borrowed from buf in apply_patch(),
191 * but some codepaths store an allocated buffer.
192 */
193 const char *patch;
194 unsigned free_patch:1,
195 rejected:1;
196 int size;
197 int linenr;
198 struct fragment *next;
199 };
200
201 /*
202 * When dealing with a binary patch, we reuse "leading" field
203 * to store the type of the binary hunk, either deflated "delta"
204 * or deflated "literal".
205 */
206 #define binary_patch_method leading
207 #define BINARY_DELTA_DEFLATED 1
208 #define BINARY_LITERAL_DEFLATED 2
209
210 static void free_fragment_list(struct fragment *list)
211 {
212 while (list) {
213 struct fragment *next = list->next;
214 if (list->free_patch)
215 free((char *)list->patch);
216 free(list);
217 list = next;
218 }
219 }
220
221 static void free_patch(struct patch *patch)
222 {
223 free_fragment_list(patch->fragments);
224 free(patch->def_name);
225 free(patch->old_name);
226 free(patch->new_name);
227 free(patch->result);
228 free(patch);
229 }
230
231 static void free_patch_list(struct patch *list)
232 {
233 while (list) {
234 struct patch *next = list->next;
235 free_patch(list);
236 list = next;
237 }
238 }
239
240 /*
241 * A line in a file, len-bytes long (includes the terminating LF,
242 * except for an incomplete line at the end if the file ends with
243 * one), and its contents hashes to 'hash'.
244 */
245 struct line {
246 size_t len;
247 unsigned hash : 24;
248 unsigned flag : 8;
249 #define LINE_COMMON 1
250 #define LINE_PATCHED 2
251 };
252
253 /*
254 * This represents a "file", which is an array of "lines".
255 */
256 struct image {
257 char *buf;
258 size_t len;
259 size_t nr;
260 size_t alloc;
261 struct line *line_allocated;
262 struct line *line;
263 };
264
265 static uint32_t hash_line(const char *cp, size_t len)
266 {
267 size_t i;
268 uint32_t h;
269 for (i = 0, h = 0; i < len; i++) {
270 if (!isspace(cp[i])) {
271 h = h * 3 + (cp[i] & 0xff);
272 }
273 }
274 return h;
275 }
276
277 /*
278 * Compare lines s1 of length n1 and s2 of length n2, ignoring
279 * whitespace difference. Returns 1 if they match, 0 otherwise
280 */
281 static int fuzzy_matchlines(const char *s1, size_t n1,
282 const char *s2, size_t n2)
283 {
284 const char *end1 = s1 + n1;
285 const char *end2 = s2 + n2;
286
287 /* ignore line endings */
288 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
289 end1--;
290 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
291 end2--;
292
293 while (s1 < end1 && s2 < end2) {
294 if (isspace(*s1)) {
295 /*
296 * Skip whitespace. We check on both buffers
297 * because we don't want "a b" to match "ab".
298 */
299 if (!isspace(*s2))
300 return 0;
301 while (s1 < end1 && isspace(*s1))
302 s1++;
303 while (s2 < end2 && isspace(*s2))
304 s2++;
305 } else if (*s1++ != *s2++)
306 return 0;
307 }
308
309 /* If we reached the end on one side only, lines don't match. */
310 return s1 == end1 && s2 == end2;
311 }
312
313 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
314 {
315 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
316 img->line_allocated[img->nr].len = len;
317 img->line_allocated[img->nr].hash = hash_line(bol, len);
318 img->line_allocated[img->nr].flag = flag;
319 img->nr++;
320 }
321
322 /*
323 * "buf" has the file contents to be patched (read from various sources).
324 * attach it to "image" and add line-based index to it.
325 * "image" now owns the "buf".
326 */
327 static void prepare_image(struct image *image, char *buf, size_t len,
328 int prepare_linetable)
329 {
330 const char *cp, *ep;
331
332 memset(image, 0, sizeof(*image));
333 image->buf = buf;
334 image->len = len;
335
336 if (!prepare_linetable)
337 return;
338
339 ep = image->buf + image->len;
340 cp = image->buf;
341 while (cp < ep) {
342 const char *next;
343 for (next = cp; next < ep && *next != '\n'; next++)
344 ;
345 if (next < ep)
346 next++;
347 add_line_info(image, cp, next - cp, 0);
348 cp = next;
349 }
350 image->line = image->line_allocated;
351 }
352
353 static void clear_image(struct image *image)
354 {
355 free(image->buf);
356 free(image->line_allocated);
357 memset(image, 0, sizeof(*image));
358 }
359
360 /* fmt must contain _one_ %s and no other substitution */
361 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
362 {
363 struct strbuf sb = STRBUF_INIT;
364
365 if (patch->old_name && patch->new_name &&
366 strcmp(patch->old_name, patch->new_name)) {
367 quote_c_style(patch->old_name, &sb, NULL, 0);
368 strbuf_addstr(&sb, " => ");
369 quote_c_style(patch->new_name, &sb, NULL, 0);
370 } else {
371 const char *n = patch->new_name;
372 if (!n)
373 n = patch->old_name;
374 quote_c_style(n, &sb, NULL, 0);
375 }
376 fprintf(output, fmt, sb.buf);
377 fputc('\n', output);
378 strbuf_release(&sb);
379 }
380
381 #define SLOP (16)
382
383 static int read_patch_file(struct strbuf *sb, int fd)
384 {
385 if (strbuf_read(sb, fd, 0) < 0)
386 return error_errno("git apply: failed to read");
387
388 /*
389 * Make sure that we have some slop in the buffer
390 * so that we can do speculative "memcmp" etc, and
391 * see to it that it is NUL-filled.
392 */
393 strbuf_grow(sb, SLOP);
394 memset(sb->buf + sb->len, 0, SLOP);
395 return 0;
396 }
397
398 static unsigned long linelen(const char *buffer, unsigned long size)
399 {
400 unsigned long len = 0;
401 while (size--) {
402 len++;
403 if (*buffer++ == '\n')
404 break;
405 }
406 return len;
407 }
408
409 static int is_dev_null(const char *str)
410 {
411 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
412 }
413
414 #define TERM_SPACE 1
415 #define TERM_TAB 2
416
417 static int name_terminate(int c, int terminate)
418 {
419 if (c == ' ' && !(terminate & TERM_SPACE))
420 return 0;
421 if (c == '\t' && !(terminate & TERM_TAB))
422 return 0;
423
424 return 1;
425 }
426
427 /* remove double slashes to make --index work with such filenames */
428 static char *squash_slash(char *name)
429 {
430 int i = 0, j = 0;
431
432 if (!name)
433 return NULL;
434
435 while (name[i]) {
436 if ((name[j++] = name[i++]) == '/')
437 while (name[i] == '/')
438 i++;
439 }
440 name[j] = '\0';
441 return name;
442 }
443
444 static char *find_name_gnu(struct strbuf *root,
445 const char *line,
446 int p_value)
447 {
448 struct strbuf name = STRBUF_INIT;
449 char *cp;
450
451 /*
452 * Proposed "new-style" GNU patch/diff format; see
453 * https://public-inbox.org/git/7vll0wvb2a.fsf@assigned-by-dhcp.cox.net/
454 */
455 if (unquote_c_style(&name, line, NULL)) {
456 strbuf_release(&name);
457 return NULL;
458 }
459
460 for (cp = name.buf; p_value; p_value--) {
461 cp = strchr(cp, '/');
462 if (!cp) {
463 strbuf_release(&name);
464 return NULL;
465 }
466 cp++;
467 }
468
469 strbuf_remove(&name, 0, cp - name.buf);
470 if (root->len)
471 strbuf_insert(&name, 0, root->buf, root->len);
472 return squash_slash(strbuf_detach(&name, NULL));
473 }
474
475 static size_t sane_tz_len(const char *line, size_t len)
476 {
477 const char *tz, *p;
478
479 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
480 return 0;
481 tz = line + len - strlen(" +0500");
482
483 if (tz[1] != '+' && tz[1] != '-')
484 return 0;
485
486 for (p = tz + 2; p != line + len; p++)
487 if (!isdigit(*p))
488 return 0;
489
490 return line + len - tz;
491 }
492
493 static size_t tz_with_colon_len(const char *line, size_t len)
494 {
495 const char *tz, *p;
496
497 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
498 return 0;
499 tz = line + len - strlen(" +08:00");
500
501 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
502 return 0;
503 p = tz + 2;
504 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
505 !isdigit(*p++) || !isdigit(*p++))
506 return 0;
507
508 return line + len - tz;
509 }
510
511 static size_t date_len(const char *line, size_t len)
512 {
513 const char *date, *p;
514
515 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
516 return 0;
517 p = date = line + len - strlen("72-02-05");
518
519 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
520 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
521 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
522 return 0;
523
524 if (date - line >= strlen("19") &&
525 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
526 date -= strlen("19");
527
528 return line + len - date;
529 }
530
531 static size_t short_time_len(const char *line, size_t len)
532 {
533 const char *time, *p;
534
535 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
536 return 0;
537 p = time = line + len - strlen(" 07:01:32");
538
539 /* Permit 1-digit hours? */
540 if (*p++ != ' ' ||
541 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
542 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
543 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
544 return 0;
545
546 return line + len - time;
547 }
548
549 static size_t fractional_time_len(const char *line, size_t len)
550 {
551 const char *p;
552 size_t n;
553
554 /* Expected format: 19:41:17.620000023 */
555 if (!len || !isdigit(line[len - 1]))
556 return 0;
557 p = line + len - 1;
558
559 /* Fractional seconds. */
560 while (p > line && isdigit(*p))
561 p--;
562 if (*p != '.')
563 return 0;
564
565 /* Hours, minutes, and whole seconds. */
566 n = short_time_len(line, p - line);
567 if (!n)
568 return 0;
569
570 return line + len - p + n;
571 }
572
573 static size_t trailing_spaces_len(const char *line, size_t len)
574 {
575 const char *p;
576
577 /* Expected format: ' ' x (1 or more) */
578 if (!len || line[len - 1] != ' ')
579 return 0;
580
581 p = line + len;
582 while (p != line) {
583 p--;
584 if (*p != ' ')
585 return line + len - (p + 1);
586 }
587
588 /* All spaces! */
589 return len;
590 }
591
592 static size_t diff_timestamp_len(const char *line, size_t len)
593 {
594 const char *end = line + len;
595 size_t n;
596
597 /*
598 * Posix: 2010-07-05 19:41:17
599 * GNU: 2010-07-05 19:41:17.620000023 -0500
600 */
601
602 if (!isdigit(end[-1]))
603 return 0;
604
605 n = sane_tz_len(line, end - line);
606 if (!n)
607 n = tz_with_colon_len(line, end - line);
608 end -= n;
609
610 n = short_time_len(line, end - line);
611 if (!n)
612 n = fractional_time_len(line, end - line);
613 end -= n;
614
615 n = date_len(line, end - line);
616 if (!n) /* No date. Too bad. */
617 return 0;
618 end -= n;
619
620 if (end == line) /* No space before date. */
621 return 0;
622 if (end[-1] == '\t') { /* Success! */
623 end--;
624 return line + len - end;
625 }
626 if (end[-1] != ' ') /* No space before date. */
627 return 0;
628
629 /* Whitespace damage. */
630 end -= trailing_spaces_len(line, end - line);
631 return line + len - end;
632 }
633
634 static char *find_name_common(struct strbuf *root,
635 const char *line,
636 const char *def,
637 int p_value,
638 const char *end,
639 int terminate)
640 {
641 int len;
642 const char *start = NULL;
643
644 if (p_value == 0)
645 start = line;
646 while (line != end) {
647 char c = *line;
648
649 if (!end && isspace(c)) {
650 if (c == '\n')
651 break;
652 if (name_terminate(c, terminate))
653 break;
654 }
655 line++;
656 if (c == '/' && !--p_value)
657 start = line;
658 }
659 if (!start)
660 return squash_slash(xstrdup_or_null(def));
661 len = line - start;
662 if (!len)
663 return squash_slash(xstrdup_or_null(def));
664
665 /*
666 * Generally we prefer the shorter name, especially
667 * if the other one is just a variation of that with
668 * something else tacked on to the end (ie "file.orig"
669 * or "file~").
670 */
671 if (def) {
672 int deflen = strlen(def);
673 if (deflen < len && !strncmp(start, def, deflen))
674 return squash_slash(xstrdup(def));
675 }
676
677 if (root->len) {
678 char *ret = xstrfmt("%s%.*s", root->buf, len, start);
679 return squash_slash(ret);
680 }
681
682 return squash_slash(xmemdupz(start, len));
683 }
684
685 static char *find_name(struct strbuf *root,
686 const char *line,
687 char *def,
688 int p_value,
689 int terminate)
690 {
691 if (*line == '"') {
692 char *name = find_name_gnu(root, line, p_value);
693 if (name)
694 return name;
695 }
696
697 return find_name_common(root, line, def, p_value, NULL, terminate);
698 }
699
700 static char *find_name_traditional(struct strbuf *root,
701 const char *line,
702 char *def,
703 int p_value)
704 {
705 size_t len;
706 size_t date_len;
707
708 if (*line == '"') {
709 char *name = find_name_gnu(root, line, p_value);
710 if (name)
711 return name;
712 }
713
714 len = strchrnul(line, '\n') - line;
715 date_len = diff_timestamp_len(line, len);
716 if (!date_len)
717 return find_name_common(root, line, def, p_value, NULL, TERM_TAB);
718 len -= date_len;
719
720 return find_name_common(root, line, def, p_value, line + len, 0);
721 }
722
723 /*
724 * Given the string after "--- " or "+++ ", guess the appropriate
725 * p_value for the given patch.
726 */
727 static int guess_p_value(struct apply_state *state, const char *nameline)
728 {
729 char *name, *cp;
730 int val = -1;
731
732 if (is_dev_null(nameline))
733 return -1;
734 name = find_name_traditional(&state->root, nameline, NULL, 0);
735 if (!name)
736 return -1;
737 cp = strchr(name, '/');
738 if (!cp)
739 val = 0;
740 else if (state->prefix) {
741 /*
742 * Does it begin with "a/$our-prefix" and such? Then this is
743 * very likely to apply to our directory.
744 */
745 if (starts_with(name, state->prefix))
746 val = count_slashes(state->prefix);
747 else {
748 cp++;
749 if (starts_with(cp, state->prefix))
750 val = count_slashes(state->prefix) + 1;
751 }
752 }
753 free(name);
754 return val;
755 }
756
757 /*
758 * Does the ---/+++ line have the POSIX timestamp after the last HT?
759 * GNU diff puts epoch there to signal a creation/deletion event. Is
760 * this such a timestamp?
761 */
762 static int has_epoch_timestamp(const char *nameline)
763 {
764 /*
765 * We are only interested in epoch timestamp; any non-zero
766 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
767 * For the same reason, the date must be either 1969-12-31 or
768 * 1970-01-01, and the seconds part must be "00".
769 */
770 const char stamp_regexp[] =
771 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
772 " "
773 "([-+][0-2][0-9]:?[0-5][0-9])\n";
774 const char *timestamp = NULL, *cp, *colon;
775 static regex_t *stamp;
776 regmatch_t m[10];
777 int zoneoffset, epoch_hour, hour, minute;
778 int status;
779
780 for (cp = nameline; *cp != '\n'; cp++) {
781 if (*cp == '\t')
782 timestamp = cp + 1;
783 }
784 if (!timestamp)
785 return 0;
786
787 /*
788 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
789 * (west of GMT) or 1970-01-01 (east of GMT)
790 */
791 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
792 epoch_hour = 24;
793 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
794 epoch_hour = 0;
795 else
796 return 0;
797
798 if (!stamp) {
799 stamp = xmalloc(sizeof(*stamp));
800 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
801 warning(_("Cannot prepare timestamp regexp %s"),
802 stamp_regexp);
803 return 0;
804 }
805 }
806
807 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
808 if (status) {
809 if (status != REG_NOMATCH)
810 warning(_("regexec returned %d for input: %s"),
811 status, timestamp);
812 return 0;
813 }
814
815 hour = strtol(timestamp, NULL, 10);
816 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
817
818 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
819 if (*colon == ':')
820 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
821 else
822 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
823 if (timestamp[m[3].rm_so] == '-')
824 zoneoffset = -zoneoffset;
825
826 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
827 }
828
829 /*
830 * Get the name etc info from the ---/+++ lines of a traditional patch header
831 *
832 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
833 * files, we can happily check the index for a match, but for creating a
834 * new file we should try to match whatever "patch" does. I have no idea.
835 */
836 static int parse_traditional_patch(struct apply_state *state,
837 const char *first,
838 const char *second,
839 struct patch *patch)
840 {
841 char *name;
842
843 first += 4; /* skip "--- " */
844 second += 4; /* skip "+++ " */
845 if (!state->p_value_known) {
846 int p, q;
847 p = guess_p_value(state, first);
848 q = guess_p_value(state, second);
849 if (p < 0) p = q;
850 if (0 <= p && p == q) {
851 state->p_value = p;
852 state->p_value_known = 1;
853 }
854 }
855 if (is_dev_null(first)) {
856 patch->is_new = 1;
857 patch->is_delete = 0;
858 name = find_name_traditional(&state->root, second, NULL, state->p_value);
859 patch->new_name = name;
860 } else if (is_dev_null(second)) {
861 patch->is_new = 0;
862 patch->is_delete = 1;
863 name = find_name_traditional(&state->root, first, NULL, state->p_value);
864 patch->old_name = name;
865 } else {
866 char *first_name;
867 first_name = find_name_traditional(&state->root, first, NULL, state->p_value);
868 name = find_name_traditional(&state->root, second, first_name, state->p_value);
869 free(first_name);
870 if (has_epoch_timestamp(first)) {
871 patch->is_new = 1;
872 patch->is_delete = 0;
873 patch->new_name = name;
874 } else if (has_epoch_timestamp(second)) {
875 patch->is_new = 0;
876 patch->is_delete = 1;
877 patch->old_name = name;
878 } else {
879 patch->old_name = name;
880 patch->new_name = xstrdup_or_null(name);
881 }
882 }
883 if (!name)
884 return error(_("unable to find filename in patch at line %d"), state->linenr);
885
886 return 0;
887 }
888
889 static int gitdiff_hdrend(struct gitdiff_data *state,
890 const char *line,
891 struct patch *patch)
892 {
893 return 1;
894 }
895
896 /*
897 * We're anal about diff header consistency, to make
898 * sure that we don't end up having strange ambiguous
899 * patches floating around.
900 *
901 * As a result, gitdiff_{old|new}name() will check
902 * their names against any previous information, just
903 * to make sure..
904 */
905 #define DIFF_OLD_NAME 0
906 #define DIFF_NEW_NAME 1
907
908 static int gitdiff_verify_name(struct gitdiff_data *state,
909 const char *line,
910 int isnull,
911 char **name,
912 int side)
913 {
914 if (!*name && !isnull) {
915 *name = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
916 return 0;
917 }
918
919 if (*name) {
920 char *another;
921 if (isnull)
922 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
923 *name, state->linenr);
924 another = find_name(state->root, line, NULL, state->p_value, TERM_TAB);
925 if (!another || strcmp(another, *name)) {
926 free(another);
927 return error((side == DIFF_NEW_NAME) ?
928 _("git apply: bad git-diff - inconsistent new filename on line %d") :
929 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
930 }
931 free(another);
932 } else {
933 if (!is_dev_null(line))
934 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
935 }
936
937 return 0;
938 }
939
940 static int gitdiff_oldname(struct gitdiff_data *state,
941 const char *line,
942 struct patch *patch)
943 {
944 return gitdiff_verify_name(state, line,
945 patch->is_new, &patch->old_name,
946 DIFF_OLD_NAME);
947 }
948
949 static int gitdiff_newname(struct gitdiff_data *state,
950 const char *line,
951 struct patch *patch)
952 {
953 return gitdiff_verify_name(state, line,
954 patch->is_delete, &patch->new_name,
955 DIFF_NEW_NAME);
956 }
957
958 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
959 {
960 char *end;
961 *mode = strtoul(line, &end, 8);
962 if (end == line || !isspace(*end))
963 return error(_("invalid mode on line %d: %s"), linenr, line);
964 return 0;
965 }
966
967 static int gitdiff_oldmode(struct gitdiff_data *state,
968 const char *line,
969 struct patch *patch)
970 {
971 return parse_mode_line(line, state->linenr, &patch->old_mode);
972 }
973
974 static int gitdiff_newmode(struct gitdiff_data *state,
975 const char *line,
976 struct patch *patch)
977 {
978 return parse_mode_line(line, state->linenr, &patch->new_mode);
979 }
980
981 static int gitdiff_delete(struct gitdiff_data *state,
982 const char *line,
983 struct patch *patch)
984 {
985 patch->is_delete = 1;
986 free(patch->old_name);
987 patch->old_name = xstrdup_or_null(patch->def_name);
988 return gitdiff_oldmode(state, line, patch);
989 }
990
991 static int gitdiff_newfile(struct gitdiff_data *state,
992 const char *line,
993 struct patch *patch)
994 {
995 patch->is_new = 1;
996 free(patch->new_name);
997 patch->new_name = xstrdup_or_null(patch->def_name);
998 return gitdiff_newmode(state, line, patch);
999 }
1000
1001 static int gitdiff_copysrc(struct gitdiff_data *state,
1002 const char *line,
1003 struct patch *patch)
1004 {
1005 patch->is_copy = 1;
1006 free(patch->old_name);
1007 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1008 return 0;
1009 }
1010
1011 static int gitdiff_copydst(struct gitdiff_data *state,
1012 const char *line,
1013 struct patch *patch)
1014 {
1015 patch->is_copy = 1;
1016 free(patch->new_name);
1017 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1018 return 0;
1019 }
1020
1021 static int gitdiff_renamesrc(struct gitdiff_data *state,
1022 const char *line,
1023 struct patch *patch)
1024 {
1025 patch->is_rename = 1;
1026 free(patch->old_name);
1027 patch->old_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1028 return 0;
1029 }
1030
1031 static int gitdiff_renamedst(struct gitdiff_data *state,
1032 const char *line,
1033 struct patch *patch)
1034 {
1035 patch->is_rename = 1;
1036 free(patch->new_name);
1037 patch->new_name = find_name(state->root, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1038 return 0;
1039 }
1040
1041 static int gitdiff_similarity(struct gitdiff_data *state,
1042 const char *line,
1043 struct patch *patch)
1044 {
1045 unsigned long val = strtoul(line, NULL, 10);
1046 if (val <= 100)
1047 patch->score = val;
1048 return 0;
1049 }
1050
1051 static int gitdiff_dissimilarity(struct gitdiff_data *state,
1052 const char *line,
1053 struct patch *patch)
1054 {
1055 unsigned long val = strtoul(line, NULL, 10);
1056 if (val <= 100)
1057 patch->score = val;
1058 return 0;
1059 }
1060
1061 static int gitdiff_index(struct gitdiff_data *state,
1062 const char *line,
1063 struct patch *patch)
1064 {
1065 /*
1066 * index line is N hexadecimal, "..", N hexadecimal,
1067 * and optional space with octal mode.
1068 */
1069 const char *ptr, *eol;
1070 int len;
1071 const unsigned hexsz = the_hash_algo->hexsz;
1072
1073 ptr = strchr(line, '.');
1074 if (!ptr || ptr[1] != '.' || hexsz < ptr - line)
1075 return 0;
1076 len = ptr - line;
1077 memcpy(patch->old_oid_prefix, line, len);
1078 patch->old_oid_prefix[len] = 0;
1079
1080 line = ptr + 2;
1081 ptr = strchr(line, ' ');
1082 eol = strchrnul(line, '\n');
1083
1084 if (!ptr || eol < ptr)
1085 ptr = eol;
1086 len = ptr - line;
1087
1088 if (hexsz < len)
1089 return 0;
1090 memcpy(patch->new_oid_prefix, line, len);
1091 patch->new_oid_prefix[len] = 0;
1092 if (*ptr == ' ')
1093 return gitdiff_oldmode(state, ptr + 1, patch);
1094 return 0;
1095 }
1096
1097 /*
1098 * This is normal for a diff that doesn't change anything: we'll fall through
1099 * into the next diff. Tell the parser to break out.
1100 */
1101 static int gitdiff_unrecognized(struct gitdiff_data *state,
1102 const char *line,
1103 struct patch *patch)
1104 {
1105 return 1;
1106 }
1107
1108 /*
1109 * Skip p_value leading components from "line"; as we do not accept
1110 * absolute paths, return NULL in that case.
1111 */
1112 static const char *skip_tree_prefix(int p_value,
1113 const char *line,
1114 int llen)
1115 {
1116 int nslash;
1117 int i;
1118
1119 if (!p_value)
1120 return (llen && line[0] == '/') ? NULL : line;
1121
1122 nslash = p_value;
1123 for (i = 0; i < llen; i++) {
1124 int ch = line[i];
1125 if (ch == '/' && --nslash <= 0)
1126 return (i == 0) ? NULL : &line[i + 1];
1127 }
1128 return NULL;
1129 }
1130
1131 /*
1132 * This is to extract the same name that appears on "diff --git"
1133 * line. We do not find and return anything if it is a rename
1134 * patch, and it is OK because we will find the name elsewhere.
1135 * We need to reliably find name only when it is mode-change only,
1136 * creation or deletion of an empty file. In any of these cases,
1137 * both sides are the same name under a/ and b/ respectively.
1138 */
1139 static char *git_header_name(int p_value,
1140 const char *line,
1141 int llen)
1142 {
1143 const char *name;
1144 const char *second = NULL;
1145 size_t len, line_len;
1146
1147 line += strlen("diff --git ");
1148 llen -= strlen("diff --git ");
1149
1150 if (*line == '"') {
1151 const char *cp;
1152 struct strbuf first = STRBUF_INIT;
1153 struct strbuf sp = STRBUF_INIT;
1154
1155 if (unquote_c_style(&first, line, &second))
1156 goto free_and_fail1;
1157
1158 /* strip the a/b prefix including trailing slash */
1159 cp = skip_tree_prefix(p_value, first.buf, first.len);
1160 if (!cp)
1161 goto free_and_fail1;
1162 strbuf_remove(&first, 0, cp - first.buf);
1163
1164 /*
1165 * second points at one past closing dq of name.
1166 * find the second name.
1167 */
1168 while ((second < line + llen) && isspace(*second))
1169 second++;
1170
1171 if (line + llen <= second)
1172 goto free_and_fail1;
1173 if (*second == '"') {
1174 if (unquote_c_style(&sp, second, NULL))
1175 goto free_and_fail1;
1176 cp = skip_tree_prefix(p_value, sp.buf, sp.len);
1177 if (!cp)
1178 goto free_and_fail1;
1179 /* They must match, otherwise ignore */
1180 if (strcmp(cp, first.buf))
1181 goto free_and_fail1;
1182 strbuf_release(&sp);
1183 return strbuf_detach(&first, NULL);
1184 }
1185
1186 /* unquoted second */
1187 cp = skip_tree_prefix(p_value, second, line + llen - second);
1188 if (!cp)
1189 goto free_and_fail1;
1190 if (line + llen - cp != first.len ||
1191 memcmp(first.buf, cp, first.len))
1192 goto free_and_fail1;
1193 return strbuf_detach(&first, NULL);
1194
1195 free_and_fail1:
1196 strbuf_release(&first);
1197 strbuf_release(&sp);
1198 return NULL;
1199 }
1200
1201 /* unquoted first name */
1202 name = skip_tree_prefix(p_value, line, llen);
1203 if (!name)
1204 return NULL;
1205
1206 /*
1207 * since the first name is unquoted, a dq if exists must be
1208 * the beginning of the second name.
1209 */
1210 for (second = name; second < line + llen; second++) {
1211 if (*second == '"') {
1212 struct strbuf sp = STRBUF_INIT;
1213 const char *np;
1214
1215 if (unquote_c_style(&sp, second, NULL))
1216 goto free_and_fail2;
1217
1218 np = skip_tree_prefix(p_value, sp.buf, sp.len);
1219 if (!np)
1220 goto free_and_fail2;
1221
1222 len = sp.buf + sp.len - np;
1223 if (len < second - name &&
1224 !strncmp(np, name, len) &&
1225 isspace(name[len])) {
1226 /* Good */
1227 strbuf_remove(&sp, 0, np - sp.buf);
1228 return strbuf_detach(&sp, NULL);
1229 }
1230
1231 free_and_fail2:
1232 strbuf_release(&sp);
1233 return NULL;
1234 }
1235 }
1236
1237 /*
1238 * Accept a name only if it shows up twice, exactly the same
1239 * form.
1240 */
1241 second = strchr(name, '\n');
1242 if (!second)
1243 return NULL;
1244 line_len = second - name;
1245 for (len = 0 ; ; len++) {
1246 switch (name[len]) {
1247 default:
1248 continue;
1249 case '\n':
1250 return NULL;
1251 case '\t': case ' ':
1252 /*
1253 * Is this the separator between the preimage
1254 * and the postimage pathname? Again, we are
1255 * only interested in the case where there is
1256 * no rename, as this is only to set def_name
1257 * and a rename patch has the names elsewhere
1258 * in an unambiguous form.
1259 */
1260 if (!name[len + 1])
1261 return NULL; /* no postimage name */
1262 second = skip_tree_prefix(p_value, name + len + 1,
1263 line_len - (len + 1));
1264 if (!second)
1265 return NULL;
1266 /*
1267 * Does len bytes starting at "name" and "second"
1268 * (that are separated by one HT or SP we just
1269 * found) exactly match?
1270 */
1271 if (second[len] == '\n' && !strncmp(name, second, len))
1272 return xmemdupz(name, len);
1273 }
1274 }
1275 }
1276
1277 static int check_header_line(int linenr, struct patch *patch)
1278 {
1279 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1280 (patch->is_rename == 1) + (patch->is_copy == 1);
1281 if (extensions > 1)
1282 return error(_("inconsistent header lines %d and %d"),
1283 patch->extension_linenr, linenr);
1284 if (extensions && !patch->extension_linenr)
1285 patch->extension_linenr = linenr;
1286 return 0;
1287 }
1288
1289 int parse_git_diff_header(struct strbuf *root,
1290 int *linenr,
1291 int p_value,
1292 const char *line,
1293 int len,
1294 unsigned int size,
1295 struct patch *patch)
1296 {
1297 unsigned long offset;
1298 struct gitdiff_data parse_hdr_state;
1299
1300 /* A git diff has explicit new/delete information, so we don't guess */
1301 patch->is_new = 0;
1302 patch->is_delete = 0;
1303
1304 /*
1305 * Some things may not have the old name in the
1306 * rest of the headers anywhere (pure mode changes,
1307 * or removing or adding empty files), so we get
1308 * the default name from the header.
1309 */
1310 patch->def_name = git_header_name(p_value, line, len);
1311 if (patch->def_name && root->len) {
1312 char *s = xstrfmt("%s%s", root->buf, patch->def_name);
1313 free(patch->def_name);
1314 patch->def_name = s;
1315 }
1316
1317 line += len;
1318 size -= len;
1319 (*linenr)++;
1320 parse_hdr_state.root = root;
1321 parse_hdr_state.linenr = *linenr;
1322 parse_hdr_state.p_value = p_value;
1323
1324 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, (*linenr)++) {
1325 static const struct opentry {
1326 const char *str;
1327 int (*fn)(struct gitdiff_data *, const char *, struct patch *);
1328 } optable[] = {
1329 { "@@ -", gitdiff_hdrend },
1330 { "--- ", gitdiff_oldname },
1331 { "+++ ", gitdiff_newname },
1332 { "old mode ", gitdiff_oldmode },
1333 { "new mode ", gitdiff_newmode },
1334 { "deleted file mode ", gitdiff_delete },
1335 { "new file mode ", gitdiff_newfile },
1336 { "copy from ", gitdiff_copysrc },
1337 { "copy to ", gitdiff_copydst },
1338 { "rename old ", gitdiff_renamesrc },
1339 { "rename new ", gitdiff_renamedst },
1340 { "rename from ", gitdiff_renamesrc },
1341 { "rename to ", gitdiff_renamedst },
1342 { "similarity index ", gitdiff_similarity },
1343 { "dissimilarity index ", gitdiff_dissimilarity },
1344 { "index ", gitdiff_index },
1345 { "", gitdiff_unrecognized },
1346 };
1347 int i;
1348
1349 len = linelen(line, size);
1350 if (!len || line[len-1] != '\n')
1351 break;
1352 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1353 const struct opentry *p = optable + i;
1354 int oplen = strlen(p->str);
1355 int res;
1356 if (len < oplen || memcmp(p->str, line, oplen))
1357 continue;
1358 res = p->fn(&parse_hdr_state, line + oplen, patch);
1359 if (res < 0)
1360 return -1;
1361 if (check_header_line(*linenr, patch))
1362 return -1;
1363 if (res > 0)
1364 return offset;
1365 break;
1366 }
1367 }
1368
1369 return offset;
1370 }
1371
1372 static int parse_num(const char *line, unsigned long *p)
1373 {
1374 char *ptr;
1375
1376 if (!isdigit(*line))
1377 return 0;
1378 *p = strtoul(line, &ptr, 10);
1379 return ptr - line;
1380 }
1381
1382 static int parse_range(const char *line, int len, int offset, const char *expect,
1383 unsigned long *p1, unsigned long *p2)
1384 {
1385 int digits, ex;
1386
1387 if (offset < 0 || offset >= len)
1388 return -1;
1389 line += offset;
1390 len -= offset;
1391
1392 digits = parse_num(line, p1);
1393 if (!digits)
1394 return -1;
1395
1396 offset += digits;
1397 line += digits;
1398 len -= digits;
1399
1400 *p2 = 1;
1401 if (*line == ',') {
1402 digits = parse_num(line+1, p2);
1403 if (!digits)
1404 return -1;
1405
1406 offset += digits+1;
1407 line += digits+1;
1408 len -= digits+1;
1409 }
1410
1411 ex = strlen(expect);
1412 if (ex > len)
1413 return -1;
1414 if (memcmp(line, expect, ex))
1415 return -1;
1416
1417 return offset + ex;
1418 }
1419
1420 static void recount_diff(const char *line, int size, struct fragment *fragment)
1421 {
1422 int oldlines = 0, newlines = 0, ret = 0;
1423
1424 if (size < 1) {
1425 warning("recount: ignore empty hunk");
1426 return;
1427 }
1428
1429 for (;;) {
1430 int len = linelen(line, size);
1431 size -= len;
1432 line += len;
1433
1434 if (size < 1)
1435 break;
1436
1437 switch (*line) {
1438 case ' ': case '\n':
1439 newlines++;
1440 /* fall through */
1441 case '-':
1442 oldlines++;
1443 continue;
1444 case '+':
1445 newlines++;
1446 continue;
1447 case '\\':
1448 continue;
1449 case '@':
1450 ret = size < 3 || !starts_with(line, "@@ ");
1451 break;
1452 case 'd':
1453 ret = size < 5 || !starts_with(line, "diff ");
1454 break;
1455 default:
1456 ret = -1;
1457 break;
1458 }
1459 if (ret) {
1460 warning(_("recount: unexpected line: %.*s"),
1461 (int)linelen(line, size), line);
1462 return;
1463 }
1464 break;
1465 }
1466 fragment->oldlines = oldlines;
1467 fragment->newlines = newlines;
1468 }
1469
1470 /*
1471 * Parse a unified diff fragment header of the
1472 * form "@@ -a,b +c,d @@"
1473 */
1474 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1475 {
1476 int offset;
1477
1478 if (!len || line[len-1] != '\n')
1479 return -1;
1480
1481 /* Figure out the number of lines in a fragment */
1482 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1483 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1484
1485 return offset;
1486 }
1487
1488 /*
1489 * Find file diff header
1490 *
1491 * Returns:
1492 * -1 if no header was found
1493 * -128 in case of error
1494 * the size of the header in bytes (called "offset") otherwise
1495 */
1496 static int find_header(struct apply_state *state,
1497 const char *line,
1498 unsigned long size,
1499 int *hdrsize,
1500 struct patch *patch)
1501 {
1502 unsigned long offset, len;
1503
1504 patch->is_toplevel_relative = 0;
1505 patch->is_rename = patch->is_copy = 0;
1506 patch->is_new = patch->is_delete = -1;
1507 patch->old_mode = patch->new_mode = 0;
1508 patch->old_name = patch->new_name = NULL;
1509 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1510 unsigned long nextlen;
1511
1512 len = linelen(line, size);
1513 if (!len)
1514 break;
1515
1516 /* Testing this early allows us to take a few shortcuts.. */
1517 if (len < 6)
1518 continue;
1519
1520 /*
1521 * Make sure we don't find any unconnected patch fragments.
1522 * That's a sign that we didn't find a header, and that a
1523 * patch has become corrupted/broken up.
1524 */
1525 if (!memcmp("@@ -", line, 4)) {
1526 struct fragment dummy;
1527 if (parse_fragment_header(line, len, &dummy) < 0)
1528 continue;
1529 error(_("patch fragment without header at line %d: %.*s"),
1530 state->linenr, (int)len-1, line);
1531 return -128;
1532 }
1533
1534 if (size < len + 6)
1535 break;
1536
1537 /*
1538 * Git patch? It might not have a real patch, just a rename
1539 * or mode change, so we handle that specially
1540 */
1541 if (!memcmp("diff --git ", line, 11)) {
1542 int git_hdr_len = parse_git_diff_header(&state->root, &state->linenr,
1543 state->p_value, line, len,
1544 size, patch);
1545 if (git_hdr_len < 0)
1546 return -128;
1547 if (git_hdr_len <= len)
1548 continue;
1549 if (!patch->old_name && !patch->new_name) {
1550 if (!patch->def_name) {
1551 error(Q_("git diff header lacks filename information when removing "
1552 "%d leading pathname component (line %d)",
1553 "git diff header lacks filename information when removing "
1554 "%d leading pathname components (line %d)",
1555 state->p_value),
1556 state->p_value, state->linenr);
1557 return -128;
1558 }
1559 patch->old_name = xstrdup(patch->def_name);
1560 patch->new_name = xstrdup(patch->def_name);
1561 }
1562 if ((!patch->new_name && !patch->is_delete) ||
1563 (!patch->old_name && !patch->is_new)) {
1564 error(_("git diff header lacks filename information "
1565 "(line %d)"), state->linenr);
1566 return -128;
1567 }
1568 patch->is_toplevel_relative = 1;
1569 *hdrsize = git_hdr_len;
1570 return offset;
1571 }
1572
1573 /* --- followed by +++ ? */
1574 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1575 continue;
1576
1577 /*
1578 * We only accept unified patches, so we want it to
1579 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1580 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1581 */
1582 nextlen = linelen(line + len, size - len);
1583 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1584 continue;
1585
1586 /* Ok, we'll consider it a patch */
1587 if (parse_traditional_patch(state, line, line+len, patch))
1588 return -128;
1589 *hdrsize = len + nextlen;
1590 state->linenr += 2;
1591 return offset;
1592 }
1593 return -1;
1594 }
1595
1596 static void record_ws_error(struct apply_state *state,
1597 unsigned result,
1598 const char *line,
1599 int len,
1600 int linenr)
1601 {
1602 char *err;
1603
1604 if (!result)
1605 return;
1606
1607 state->whitespace_error++;
1608 if (state->squelch_whitespace_errors &&
1609 state->squelch_whitespace_errors < state->whitespace_error)
1610 return;
1611
1612 err = whitespace_error_string(result);
1613 if (state->apply_verbosity > verbosity_silent)
1614 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1615 state->patch_input_file, linenr, err, len, line);
1616 free(err);
1617 }
1618
1619 static void check_whitespace(struct apply_state *state,
1620 const char *line,
1621 int len,
1622 unsigned ws_rule)
1623 {
1624 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1625
1626 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1627 }
1628
1629 /*
1630 * Check if the patch has context lines with CRLF or
1631 * the patch wants to remove lines with CRLF.
1632 */
1633 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1634 {
1635 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1636 patch->ws_rule |= WS_CR_AT_EOL;
1637 patch->crlf_in_old = 1;
1638 }
1639 }
1640
1641
1642 /*
1643 * Parse a unified diff. Note that this really needs to parse each
1644 * fragment separately, since the only way to know the difference
1645 * between a "---" that is part of a patch, and a "---" that starts
1646 * the next patch is to look at the line counts..
1647 */
1648 static int parse_fragment(struct apply_state *state,
1649 const char *line,
1650 unsigned long size,
1651 struct patch *patch,
1652 struct fragment *fragment)
1653 {
1654 int added, deleted;
1655 int len = linelen(line, size), offset;
1656 unsigned long oldlines, newlines;
1657 unsigned long leading, trailing;
1658
1659 offset = parse_fragment_header(line, len, fragment);
1660 if (offset < 0)
1661 return -1;
1662 if (offset > 0 && patch->recount)
1663 recount_diff(line + offset, size - offset, fragment);
1664 oldlines = fragment->oldlines;
1665 newlines = fragment->newlines;
1666 leading = 0;
1667 trailing = 0;
1668
1669 /* Parse the thing.. */
1670 line += len;
1671 size -= len;
1672 state->linenr++;
1673 added = deleted = 0;
1674 for (offset = len;
1675 0 < size;
1676 offset += len, size -= len, line += len, state->linenr++) {
1677 if (!oldlines && !newlines)
1678 break;
1679 len = linelen(line, size);
1680 if (!len || line[len-1] != '\n')
1681 return -1;
1682 switch (*line) {
1683 default:
1684 return -1;
1685 case '\n': /* newer GNU diff, an empty context line */
1686 case ' ':
1687 oldlines--;
1688 newlines--;
1689 if (!deleted && !added)
1690 leading++;
1691 trailing++;
1692 check_old_for_crlf(patch, line, len);
1693 if (!state->apply_in_reverse &&
1694 state->ws_error_action == correct_ws_error)
1695 check_whitespace(state, line, len, patch->ws_rule);
1696 break;
1697 case '-':
1698 if (!state->apply_in_reverse)
1699 check_old_for_crlf(patch, line, len);
1700 if (state->apply_in_reverse &&
1701 state->ws_error_action != nowarn_ws_error)
1702 check_whitespace(state, line, len, patch->ws_rule);
1703 deleted++;
1704 oldlines--;
1705 trailing = 0;
1706 break;
1707 case '+':
1708 if (state->apply_in_reverse)
1709 check_old_for_crlf(patch, line, len);
1710 if (!state->apply_in_reverse &&
1711 state->ws_error_action != nowarn_ws_error)
1712 check_whitespace(state, line, len, patch->ws_rule);
1713 added++;
1714 newlines--;
1715 trailing = 0;
1716 break;
1717
1718 /*
1719 * We allow "\ No newline at end of file". Depending
1720 * on locale settings when the patch was produced we
1721 * don't know what this line looks like. The only
1722 * thing we do know is that it begins with "\ ".
1723 * Checking for 12 is just for sanity check -- any
1724 * l10n of "\ No newline..." is at least that long.
1725 */
1726 case '\\':
1727 if (len < 12 || memcmp(line, "\\ ", 2))
1728 return -1;
1729 break;
1730 }
1731 }
1732 if (oldlines || newlines)
1733 return -1;
1734 if (!patch->recount && !deleted && !added)
1735 return -1;
1736
1737 fragment->leading = leading;
1738 fragment->trailing = trailing;
1739
1740 /*
1741 * If a fragment ends with an incomplete line, we failed to include
1742 * it in the above loop because we hit oldlines == newlines == 0
1743 * before seeing it.
1744 */
1745 if (12 < size && !memcmp(line, "\\ ", 2))
1746 offset += linelen(line, size);
1747
1748 patch->lines_added += added;
1749 patch->lines_deleted += deleted;
1750
1751 if (0 < patch->is_new && oldlines)
1752 return error(_("new file depends on old contents"));
1753 if (0 < patch->is_delete && newlines)
1754 return error(_("deleted file still has contents"));
1755 return offset;
1756 }
1757
1758 /*
1759 * We have seen "diff --git a/... b/..." header (or a traditional patch
1760 * header). Read hunks that belong to this patch into fragments and hang
1761 * them to the given patch structure.
1762 *
1763 * The (fragment->patch, fragment->size) pair points into the memory given
1764 * by the caller, not a copy, when we return.
1765 *
1766 * Returns:
1767 * -1 in case of error,
1768 * the number of bytes in the patch otherwise.
1769 */
1770 static int parse_single_patch(struct apply_state *state,
1771 const char *line,
1772 unsigned long size,
1773 struct patch *patch)
1774 {
1775 unsigned long offset = 0;
1776 unsigned long oldlines = 0, newlines = 0, context = 0;
1777 struct fragment **fragp = &patch->fragments;
1778
1779 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1780 struct fragment *fragment;
1781 int len;
1782
1783 fragment = xcalloc(1, sizeof(*fragment));
1784 fragment->linenr = state->linenr;
1785 len = parse_fragment(state, line, size, patch, fragment);
1786 if (len <= 0) {
1787 free(fragment);
1788 return error(_("corrupt patch at line %d"), state->linenr);
1789 }
1790 fragment->patch = line;
1791 fragment->size = len;
1792 oldlines += fragment->oldlines;
1793 newlines += fragment->newlines;
1794 context += fragment->leading + fragment->trailing;
1795
1796 *fragp = fragment;
1797 fragp = &fragment->next;
1798
1799 offset += len;
1800 line += len;
1801 size -= len;
1802 }
1803
1804 /*
1805 * If something was removed (i.e. we have old-lines) it cannot
1806 * be creation, and if something was added it cannot be
1807 * deletion. However, the reverse is not true; --unified=0
1808 * patches that only add are not necessarily creation even
1809 * though they do not have any old lines, and ones that only
1810 * delete are not necessarily deletion.
1811 *
1812 * Unfortunately, a real creation/deletion patch do _not_ have
1813 * any context line by definition, so we cannot safely tell it
1814 * apart with --unified=0 insanity. At least if the patch has
1815 * more than one hunk it is not creation or deletion.
1816 */
1817 if (patch->is_new < 0 &&
1818 (oldlines || (patch->fragments && patch->fragments->next)))
1819 patch->is_new = 0;
1820 if (patch->is_delete < 0 &&
1821 (newlines || (patch->fragments && patch->fragments->next)))
1822 patch->is_delete = 0;
1823
1824 if (0 < patch->is_new && oldlines)
1825 return error(_("new file %s depends on old contents"), patch->new_name);
1826 if (0 < patch->is_delete && newlines)
1827 return error(_("deleted file %s still has contents"), patch->old_name);
1828 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1829 fprintf_ln(stderr,
1830 _("** warning: "
1831 "file %s becomes empty but is not deleted"),
1832 patch->new_name);
1833
1834 return offset;
1835 }
1836
1837 static inline int metadata_changes(struct patch *patch)
1838 {
1839 return patch->is_rename > 0 ||
1840 patch->is_copy > 0 ||
1841 patch->is_new > 0 ||
1842 patch->is_delete ||
1843 (patch->old_mode && patch->new_mode &&
1844 patch->old_mode != patch->new_mode);
1845 }
1846
1847 static char *inflate_it(const void *data, unsigned long size,
1848 unsigned long inflated_size)
1849 {
1850 git_zstream stream;
1851 void *out;
1852 int st;
1853
1854 memset(&stream, 0, sizeof(stream));
1855
1856 stream.next_in = (unsigned char *)data;
1857 stream.avail_in = size;
1858 stream.next_out = out = xmalloc(inflated_size);
1859 stream.avail_out = inflated_size;
1860 git_inflate_init(&stream);
1861 st = git_inflate(&stream, Z_FINISH);
1862 git_inflate_end(&stream);
1863 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1864 free(out);
1865 return NULL;
1866 }
1867 return out;
1868 }
1869
1870 /*
1871 * Read a binary hunk and return a new fragment; fragment->patch
1872 * points at an allocated memory that the caller must free, so
1873 * it is marked as "->free_patch = 1".
1874 */
1875 static struct fragment *parse_binary_hunk(struct apply_state *state,
1876 char **buf_p,
1877 unsigned long *sz_p,
1878 int *status_p,
1879 int *used_p)
1880 {
1881 /*
1882 * Expect a line that begins with binary patch method ("literal"
1883 * or "delta"), followed by the length of data before deflating.
1884 * a sequence of 'length-byte' followed by base-85 encoded data
1885 * should follow, terminated by a newline.
1886 *
1887 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1888 * and we would limit the patch line to 66 characters,
1889 * so one line can fit up to 13 groups that would decode
1890 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1891 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1892 */
1893 int llen, used;
1894 unsigned long size = *sz_p;
1895 char *buffer = *buf_p;
1896 int patch_method;
1897 unsigned long origlen;
1898 char *data = NULL;
1899 int hunk_size = 0;
1900 struct fragment *frag;
1901
1902 llen = linelen(buffer, size);
1903 used = llen;
1904
1905 *status_p = 0;
1906
1907 if (starts_with(buffer, "delta ")) {
1908 patch_method = BINARY_DELTA_DEFLATED;
1909 origlen = strtoul(buffer + 6, NULL, 10);
1910 }
1911 else if (starts_with(buffer, "literal ")) {
1912 patch_method = BINARY_LITERAL_DEFLATED;
1913 origlen = strtoul(buffer + 8, NULL, 10);
1914 }
1915 else
1916 return NULL;
1917
1918 state->linenr++;
1919 buffer += llen;
1920 while (1) {
1921 int byte_length, max_byte_length, newsize;
1922 llen = linelen(buffer, size);
1923 used += llen;
1924 state->linenr++;
1925 if (llen == 1) {
1926 /* consume the blank line */
1927 buffer++;
1928 size--;
1929 break;
1930 }
1931 /*
1932 * Minimum line is "A00000\n" which is 7-byte long,
1933 * and the line length must be multiple of 5 plus 2.
1934 */
1935 if ((llen < 7) || (llen-2) % 5)
1936 goto corrupt;
1937 max_byte_length = (llen - 2) / 5 * 4;
1938 byte_length = *buffer;
1939 if ('A' <= byte_length && byte_length <= 'Z')
1940 byte_length = byte_length - 'A' + 1;
1941 else if ('a' <= byte_length && byte_length <= 'z')
1942 byte_length = byte_length - 'a' + 27;
1943 else
1944 goto corrupt;
1945 /* if the input length was not multiple of 4, we would
1946 * have filler at the end but the filler should never
1947 * exceed 3 bytes
1948 */
1949 if (max_byte_length < byte_length ||
1950 byte_length <= max_byte_length - 4)
1951 goto corrupt;
1952 newsize = hunk_size + byte_length;
1953 data = xrealloc(data, newsize);
1954 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1955 goto corrupt;
1956 hunk_size = newsize;
1957 buffer += llen;
1958 size -= llen;
1959 }
1960
1961 frag = xcalloc(1, sizeof(*frag));
1962 frag->patch = inflate_it(data, hunk_size, origlen);
1963 frag->free_patch = 1;
1964 if (!frag->patch)
1965 goto corrupt;
1966 free(data);
1967 frag->size = origlen;
1968 *buf_p = buffer;
1969 *sz_p = size;
1970 *used_p = used;
1971 frag->binary_patch_method = patch_method;
1972 return frag;
1973
1974 corrupt:
1975 free(data);
1976 *status_p = -1;
1977 error(_("corrupt binary patch at line %d: %.*s"),
1978 state->linenr-1, llen-1, buffer);
1979 return NULL;
1980 }
1981
1982 /*
1983 * Returns:
1984 * -1 in case of error,
1985 * the length of the parsed binary patch otherwise
1986 */
1987 static int parse_binary(struct apply_state *state,
1988 char *buffer,
1989 unsigned long size,
1990 struct patch *patch)
1991 {
1992 /*
1993 * We have read "GIT binary patch\n"; what follows is a line
1994 * that says the patch method (currently, either "literal" or
1995 * "delta") and the length of data before deflating; a
1996 * sequence of 'length-byte' followed by base-85 encoded data
1997 * follows.
1998 *
1999 * When a binary patch is reversible, there is another binary
2000 * hunk in the same format, starting with patch method (either
2001 * "literal" or "delta") with the length of data, and a sequence
2002 * of length-byte + base-85 encoded data, terminated with another
2003 * empty line. This data, when applied to the postimage, produces
2004 * the preimage.
2005 */
2006 struct fragment *forward;
2007 struct fragment *reverse;
2008 int status;
2009 int used, used_1;
2010
2011 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2012 if (!forward && !status)
2013 /* there has to be one hunk (forward hunk) */
2014 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2015 if (status)
2016 /* otherwise we already gave an error message */
2017 return status;
2018
2019 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2020 if (reverse)
2021 used += used_1;
2022 else if (status) {
2023 /*
2024 * Not having reverse hunk is not an error, but having
2025 * a corrupt reverse hunk is.
2026 */
2027 free((void*) forward->patch);
2028 free(forward);
2029 return status;
2030 }
2031 forward->next = reverse;
2032 patch->fragments = forward;
2033 patch->is_binary = 1;
2034 return used;
2035 }
2036
2037 static void prefix_one(struct apply_state *state, char **name)
2038 {
2039 char *old_name = *name;
2040 if (!old_name)
2041 return;
2042 *name = prefix_filename(state->prefix, *name);
2043 free(old_name);
2044 }
2045
2046 static void prefix_patch(struct apply_state *state, struct patch *p)
2047 {
2048 if (!state->prefix || p->is_toplevel_relative)
2049 return;
2050 prefix_one(state, &p->new_name);
2051 prefix_one(state, &p->old_name);
2052 }
2053
2054 /*
2055 * include/exclude
2056 */
2057
2058 static void add_name_limit(struct apply_state *state,
2059 const char *name,
2060 int exclude)
2061 {
2062 struct string_list_item *it;
2063
2064 it = string_list_append(&state->limit_by_name, name);
2065 it->util = exclude ? NULL : (void *) 1;
2066 }
2067
2068 static int use_patch(struct apply_state *state, struct patch *p)
2069 {
2070 const char *pathname = p->new_name ? p->new_name : p->old_name;
2071 int i;
2072
2073 /* Paths outside are not touched regardless of "--include" */
2074 if (state->prefix && *state->prefix) {
2075 const char *rest;
2076 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2077 return 0;
2078 }
2079
2080 /* See if it matches any of exclude/include rule */
2081 for (i = 0; i < state->limit_by_name.nr; i++) {
2082 struct string_list_item *it = &state->limit_by_name.items[i];
2083 if (!wildmatch(it->string, pathname, 0))
2084 return (it->util != NULL);
2085 }
2086
2087 /*
2088 * If we had any include, a path that does not match any rule is
2089 * not used. Otherwise, we saw bunch of exclude rules (or none)
2090 * and such a path is used.
2091 */
2092 return !state->has_include;
2093 }
2094
2095 /*
2096 * Read the patch text in "buffer" that extends for "size" bytes; stop
2097 * reading after seeing a single patch (i.e. changes to a single file).
2098 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2099 *
2100 * Returns:
2101 * -1 if no header was found or parse_binary() failed,
2102 * -128 on another error,
2103 * the number of bytes consumed otherwise,
2104 * so that the caller can call us again for the next patch.
2105 */
2106 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2107 {
2108 int hdrsize, patchsize;
2109 int offset = find_header(state, buffer, size, &hdrsize, patch);
2110
2111 if (offset < 0)
2112 return offset;
2113
2114 prefix_patch(state, patch);
2115
2116 if (!use_patch(state, patch))
2117 patch->ws_rule = 0;
2118 else if (patch->new_name)
2119 patch->ws_rule = whitespace_rule(state->repo->index,
2120 patch->new_name);
2121 else
2122 patch->ws_rule = whitespace_rule(state->repo->index,
2123 patch->old_name);
2124
2125 patchsize = parse_single_patch(state,
2126 buffer + offset + hdrsize,
2127 size - offset - hdrsize,
2128 patch);
2129
2130 if (patchsize < 0)
2131 return -128;
2132
2133 if (!patchsize) {
2134 static const char git_binary[] = "GIT binary patch\n";
2135 int hd = hdrsize + offset;
2136 unsigned long llen = linelen(buffer + hd, size - hd);
2137
2138 if (llen == sizeof(git_binary) - 1 &&
2139 !memcmp(git_binary, buffer + hd, llen)) {
2140 int used;
2141 state->linenr++;
2142 used = parse_binary(state, buffer + hd + llen,
2143 size - hd - llen, patch);
2144 if (used < 0)
2145 return -1;
2146 if (used)
2147 patchsize = used + llen;
2148 else
2149 patchsize = 0;
2150 }
2151 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2152 static const char *binhdr[] = {
2153 "Binary files ",
2154 "Files ",
2155 NULL,
2156 };
2157 int i;
2158 for (i = 0; binhdr[i]; i++) {
2159 int len = strlen(binhdr[i]);
2160 if (len < size - hd &&
2161 !memcmp(binhdr[i], buffer + hd, len)) {
2162 state->linenr++;
2163 patch->is_binary = 1;
2164 patchsize = llen;
2165 break;
2166 }
2167 }
2168 }
2169
2170 /* Empty patch cannot be applied if it is a text patch
2171 * without metadata change. A binary patch appears
2172 * empty to us here.
2173 */
2174 if ((state->apply || state->check) &&
2175 (!patch->is_binary && !metadata_changes(patch))) {
2176 error(_("patch with only garbage at line %d"), state->linenr);
2177 return -128;
2178 }
2179 }
2180
2181 return offset + hdrsize + patchsize;
2182 }
2183
2184 static void reverse_patches(struct patch *p)
2185 {
2186 for (; p; p = p->next) {
2187 struct fragment *frag = p->fragments;
2188
2189 SWAP(p->new_name, p->old_name);
2190 SWAP(p->new_mode, p->old_mode);
2191 SWAP(p->is_new, p->is_delete);
2192 SWAP(p->lines_added, p->lines_deleted);
2193 SWAP(p->old_oid_prefix, p->new_oid_prefix);
2194
2195 for (; frag; frag = frag->next) {
2196 SWAP(frag->newpos, frag->oldpos);
2197 SWAP(frag->newlines, frag->oldlines);
2198 }
2199 }
2200 }
2201
2202 static const char pluses[] =
2203 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2204 static const char minuses[]=
2205 "----------------------------------------------------------------------";
2206
2207 static void show_stats(struct apply_state *state, struct patch *patch)
2208 {
2209 struct strbuf qname = STRBUF_INIT;
2210 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2211 int max, add, del;
2212
2213 quote_c_style(cp, &qname, NULL, 0);
2214
2215 /*
2216 * "scale" the filename
2217 */
2218 max = state->max_len;
2219 if (max > 50)
2220 max = 50;
2221
2222 if (qname.len > max) {
2223 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2224 if (!cp)
2225 cp = qname.buf + qname.len + 3 - max;
2226 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2227 }
2228
2229 if (patch->is_binary) {
2230 printf(" %-*s | Bin\n", max, qname.buf);
2231 strbuf_release(&qname);
2232 return;
2233 }
2234
2235 printf(" %-*s |", max, qname.buf);
2236 strbuf_release(&qname);
2237
2238 /*
2239 * scale the add/delete
2240 */
2241 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2242 add = patch->lines_added;
2243 del = patch->lines_deleted;
2244
2245 if (state->max_change > 0) {
2246 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2247 add = (add * max + state->max_change / 2) / state->max_change;
2248 del = total - add;
2249 }
2250 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2251 add, pluses, del, minuses);
2252 }
2253
2254 static int read_old_data(struct stat *st, struct patch *patch,
2255 const char *path, struct strbuf *buf)
2256 {
2257 int conv_flags = patch->crlf_in_old ?
2258 CONV_EOL_KEEP_CRLF : CONV_EOL_RENORMALIZE;
2259 switch (st->st_mode & S_IFMT) {
2260 case S_IFLNK:
2261 if (strbuf_readlink(buf, path, st->st_size) < 0)
2262 return error(_("unable to read symlink %s"), path);
2263 return 0;
2264 case S_IFREG:
2265 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2266 return error(_("unable to open or read %s"), path);
2267 /*
2268 * "git apply" without "--index/--cached" should never look
2269 * at the index; the target file may not have been added to
2270 * the index yet, and we may not even be in any Git repository.
2271 * Pass NULL to convert_to_git() to stress this; the function
2272 * should never look at the index when explicit crlf option
2273 * is given.
2274 */
2275 convert_to_git(NULL, path, buf->buf, buf->len, buf, conv_flags);
2276 return 0;
2277 default:
2278 return -1;
2279 }
2280 }
2281
2282 /*
2283 * Update the preimage, and the common lines in postimage,
2284 * from buffer buf of length len. If postlen is 0 the postimage
2285 * is updated in place, otherwise it's updated on a new buffer
2286 * of length postlen
2287 */
2288
2289 static void update_pre_post_images(struct image *preimage,
2290 struct image *postimage,
2291 char *buf,
2292 size_t len, size_t postlen)
2293 {
2294 int i, ctx, reduced;
2295 char *new_buf, *old_buf, *fixed;
2296 struct image fixed_preimage;
2297
2298 /*
2299 * Update the preimage with whitespace fixes. Note that we
2300 * are not losing preimage->buf -- apply_one_fragment() will
2301 * free "oldlines".
2302 */
2303 prepare_image(&fixed_preimage, buf, len, 1);
2304 assert(postlen
2305 ? fixed_preimage.nr == preimage->nr
2306 : fixed_preimage.nr <= preimage->nr);
2307 for (i = 0; i < fixed_preimage.nr; i++)
2308 fixed_preimage.line[i].flag = preimage->line[i].flag;
2309 free(preimage->line_allocated);
2310 *preimage = fixed_preimage;
2311
2312 /*
2313 * Adjust the common context lines in postimage. This can be
2314 * done in-place when we are shrinking it with whitespace
2315 * fixing, but needs a new buffer when ignoring whitespace or
2316 * expanding leading tabs to spaces.
2317 *
2318 * We trust the caller to tell us if the update can be done
2319 * in place (postlen==0) or not.
2320 */
2321 old_buf = postimage->buf;
2322 if (postlen)
2323 new_buf = postimage->buf = xmalloc(postlen);
2324 else
2325 new_buf = old_buf;
2326 fixed = preimage->buf;
2327
2328 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2329 size_t l_len = postimage->line[i].len;
2330 if (!(postimage->line[i].flag & LINE_COMMON)) {
2331 /* an added line -- no counterparts in preimage */
2332 memmove(new_buf, old_buf, l_len);
2333 old_buf += l_len;
2334 new_buf += l_len;
2335 continue;
2336 }
2337
2338 /* a common context -- skip it in the original postimage */
2339 old_buf += l_len;
2340
2341 /* and find the corresponding one in the fixed preimage */
2342 while (ctx < preimage->nr &&
2343 !(preimage->line[ctx].flag & LINE_COMMON)) {
2344 fixed += preimage->line[ctx].len;
2345 ctx++;
2346 }
2347
2348 /*
2349 * preimage is expected to run out, if the caller
2350 * fixed addition of trailing blank lines.
2351 */
2352 if (preimage->nr <= ctx) {
2353 reduced++;
2354 continue;
2355 }
2356
2357 /* and copy it in, while fixing the line length */
2358 l_len = preimage->line[ctx].len;
2359 memcpy(new_buf, fixed, l_len);
2360 new_buf += l_len;
2361 fixed += l_len;
2362 postimage->line[i].len = l_len;
2363 ctx++;
2364 }
2365
2366 if (postlen
2367 ? postlen < new_buf - postimage->buf
2368 : postimage->len < new_buf - postimage->buf)
2369 BUG("caller miscounted postlen: asked %d, orig = %d, used = %d",
2370 (int)postlen, (int) postimage->len, (int)(new_buf - postimage->buf));
2371
2372 /* Fix the length of the whole thing */
2373 postimage->len = new_buf - postimage->buf;
2374 postimage->nr -= reduced;
2375 }
2376
2377 static int line_by_line_fuzzy_match(struct image *img,
2378 struct image *preimage,
2379 struct image *postimage,
2380 unsigned long current,
2381 int current_lno,
2382 int preimage_limit)
2383 {
2384 int i;
2385 size_t imgoff = 0;
2386 size_t preoff = 0;
2387 size_t postlen = postimage->len;
2388 size_t extra_chars;
2389 char *buf;
2390 char *preimage_eof;
2391 char *preimage_end;
2392 struct strbuf fixed;
2393 char *fixed_buf;
2394 size_t fixed_len;
2395
2396 for (i = 0; i < preimage_limit; i++) {
2397 size_t prelen = preimage->line[i].len;
2398 size_t imglen = img->line[current_lno+i].len;
2399
2400 if (!fuzzy_matchlines(img->buf + current + imgoff, imglen,
2401 preimage->buf + preoff, prelen))
2402 return 0;
2403 if (preimage->line[i].flag & LINE_COMMON)
2404 postlen += imglen - prelen;
2405 imgoff += imglen;
2406 preoff += prelen;
2407 }
2408
2409 /*
2410 * Ok, the preimage matches with whitespace fuzz.
2411 *
2412 * imgoff now holds the true length of the target that
2413 * matches the preimage before the end of the file.
2414 *
2415 * Count the number of characters in the preimage that fall
2416 * beyond the end of the file and make sure that all of them
2417 * are whitespace characters. (This can only happen if
2418 * we are removing blank lines at the end of the file.)
2419 */
2420 buf = preimage_eof = preimage->buf + preoff;
2421 for ( ; i < preimage->nr; i++)
2422 preoff += preimage->line[i].len;
2423 preimage_end = preimage->buf + preoff;
2424 for ( ; buf < preimage_end; buf++)
2425 if (!isspace(*buf))
2426 return 0;
2427
2428 /*
2429 * Update the preimage and the common postimage context
2430 * lines to use the same whitespace as the target.
2431 * If whitespace is missing in the target (i.e.
2432 * if the preimage extends beyond the end of the file),
2433 * use the whitespace from the preimage.
2434 */
2435 extra_chars = preimage_end - preimage_eof;
2436 strbuf_init(&fixed, imgoff + extra_chars);
2437 strbuf_add(&fixed, img->buf + current, imgoff);
2438 strbuf_add(&fixed, preimage_eof, extra_chars);
2439 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2440 update_pre_post_images(preimage, postimage,
2441 fixed_buf, fixed_len, postlen);
2442 return 1;
2443 }
2444
2445 static int match_fragment(struct apply_state *state,
2446 struct image *img,
2447 struct image *preimage,
2448 struct image *postimage,
2449 unsigned long current,
2450 int current_lno,
2451 unsigned ws_rule,
2452 int match_beginning, int match_end)
2453 {
2454 int i;
2455 char *fixed_buf, *buf, *orig, *target;
2456 struct strbuf fixed;
2457 size_t fixed_len, postlen;
2458 int preimage_limit;
2459
2460 if (preimage->nr + current_lno <= img->nr) {
2461 /*
2462 * The hunk falls within the boundaries of img.
2463 */
2464 preimage_limit = preimage->nr;
2465 if (match_end && (preimage->nr + current_lno != img->nr))
2466 return 0;
2467 } else if (state->ws_error_action == correct_ws_error &&
2468 (ws_rule & WS_BLANK_AT_EOF)) {
2469 /*
2470 * This hunk extends beyond the end of img, and we are
2471 * removing blank lines at the end of the file. This
2472 * many lines from the beginning of the preimage must
2473 * match with img, and the remainder of the preimage
2474 * must be blank.
2475 */
2476 preimage_limit = img->nr - current_lno;
2477 } else {
2478 /*
2479 * The hunk extends beyond the end of the img and
2480 * we are not removing blanks at the end, so we
2481 * should reject the hunk at this position.
2482 */
2483 return 0;
2484 }
2485
2486 if (match_beginning && current_lno)
2487 return 0;
2488
2489 /* Quick hash check */
2490 for (i = 0; i < preimage_limit; i++)
2491 if ((img->line[current_lno + i].flag & LINE_PATCHED) ||
2492 (preimage->line[i].hash != img->line[current_lno + i].hash))
2493 return 0;
2494
2495 if (preimage_limit == preimage->nr) {
2496 /*
2497 * Do we have an exact match? If we were told to match
2498 * at the end, size must be exactly at current+fragsize,
2499 * otherwise current+fragsize must be still within the preimage,
2500 * and either case, the old piece should match the preimage
2501 * exactly.
2502 */
2503 if ((match_end
2504 ? (current + preimage->len == img->len)
2505 : (current + preimage->len <= img->len)) &&
2506 !memcmp(img->buf + current, preimage->buf, preimage->len))
2507 return 1;
2508 } else {
2509 /*
2510 * The preimage extends beyond the end of img, so
2511 * there cannot be an exact match.
2512 *
2513 * There must be one non-blank context line that match
2514 * a line before the end of img.
2515 */
2516 char *buf_end;
2517
2518 buf = preimage->buf;
2519 buf_end = buf;
2520 for (i = 0; i < preimage_limit; i++)
2521 buf_end += preimage->line[i].len;
2522
2523 for ( ; buf < buf_end; buf++)
2524 if (!isspace(*buf))
2525 break;
2526 if (buf == buf_end)
2527 return 0;
2528 }
2529
2530 /*
2531 * No exact match. If we are ignoring whitespace, run a line-by-line
2532 * fuzzy matching. We collect all the line length information because
2533 * we need it to adjust whitespace if we match.
2534 */
2535 if (state->ws_ignore_action == ignore_ws_change)
2536 return line_by_line_fuzzy_match(img, preimage, postimage,
2537 current, current_lno, preimage_limit);
2538
2539 if (state->ws_error_action != correct_ws_error)
2540 return 0;
2541
2542 /*
2543 * The hunk does not apply byte-by-byte, but the hash says
2544 * it might with whitespace fuzz. We weren't asked to
2545 * ignore whitespace, we were asked to correct whitespace
2546 * errors, so let's try matching after whitespace correction.
2547 *
2548 * While checking the preimage against the target, whitespace
2549 * errors in both fixed, we count how large the corresponding
2550 * postimage needs to be. The postimage prepared by
2551 * apply_one_fragment() has whitespace errors fixed on added
2552 * lines already, but the common lines were propagated as-is,
2553 * which may become longer when their whitespace errors are
2554 * fixed.
2555 */
2556
2557 /* First count added lines in postimage */
2558 postlen = 0;
2559 for (i = 0; i < postimage->nr; i++) {
2560 if (!(postimage->line[i].flag & LINE_COMMON))
2561 postlen += postimage->line[i].len;
2562 }
2563
2564 /*
2565 * The preimage may extend beyond the end of the file,
2566 * but in this loop we will only handle the part of the
2567 * preimage that falls within the file.
2568 */
2569 strbuf_init(&fixed, preimage->len + 1);
2570 orig = preimage->buf;
2571 target = img->buf + current;
2572 for (i = 0; i < preimage_limit; i++) {
2573 size_t oldlen = preimage->line[i].len;
2574 size_t tgtlen = img->line[current_lno + i].len;
2575 size_t fixstart = fixed.len;
2576 struct strbuf tgtfix;
2577 int match;
2578
2579 /* Try fixing the line in the preimage */
2580 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2581
2582 /* Try fixing the line in the target */
2583 strbuf_init(&tgtfix, tgtlen);
2584 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2585
2586 /*
2587 * If they match, either the preimage was based on
2588 * a version before our tree fixed whitespace breakage,
2589 * or we are lacking a whitespace-fix patch the tree
2590 * the preimage was based on already had (i.e. target
2591 * has whitespace breakage, the preimage doesn't).
2592 * In either case, we are fixing the whitespace breakages
2593 * so we might as well take the fix together with their
2594 * real change.
2595 */
2596 match = (tgtfix.len == fixed.len - fixstart &&
2597 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2598 fixed.len - fixstart));
2599
2600 /* Add the length if this is common with the postimage */
2601 if (preimage->line[i].flag & LINE_COMMON)
2602 postlen += tgtfix.len;
2603
2604 strbuf_release(&tgtfix);
2605 if (!match)
2606 goto unmatch_exit;
2607
2608 orig += oldlen;
2609 target += tgtlen;
2610 }
2611
2612
2613 /*
2614 * Now handle the lines in the preimage that falls beyond the
2615 * end of the file (if any). They will only match if they are
2616 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2617 * false).
2618 */
2619 for ( ; i < preimage->nr; i++) {
2620 size_t fixstart = fixed.len; /* start of the fixed preimage */
2621 size_t oldlen = preimage->line[i].len;
2622 int j;
2623
2624 /* Try fixing the line in the preimage */
2625 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2626
2627 for (j = fixstart; j < fixed.len; j++)
2628 if (!isspace(fixed.buf[j]))
2629 goto unmatch_exit;
2630
2631 orig += oldlen;
2632 }
2633
2634 /*
2635 * Yes, the preimage is based on an older version that still
2636 * has whitespace breakages unfixed, and fixing them makes the
2637 * hunk match. Update the context lines in the postimage.
2638 */
2639 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2640 if (postlen < postimage->len)
2641 postlen = 0;
2642 update_pre_post_images(preimage, postimage,
2643 fixed_buf, fixed_len, postlen);
2644 return 1;
2645
2646 unmatch_exit:
2647 strbuf_release(&fixed);
2648 return 0;
2649 }
2650
2651 static int find_pos(struct apply_state *state,
2652 struct image *img,
2653 struct image *preimage,
2654 struct image *postimage,
2655 int line,
2656 unsigned ws_rule,
2657 int match_beginning, int match_end)
2658 {
2659 int i;
2660 unsigned long backwards, forwards, current;
2661 int backwards_lno, forwards_lno, current_lno;
2662
2663 /*
2664 * If match_beginning or match_end is specified, there is no
2665 * point starting from a wrong line that will never match and
2666 * wander around and wait for a match at the specified end.
2667 */
2668 if (match_beginning)
2669 line = 0;
2670 else if (match_end)
2671 line = img->nr - preimage->nr;
2672
2673 /*
2674 * Because the comparison is unsigned, the following test
2675 * will also take care of a negative line number that can
2676 * result when match_end and preimage is larger than the target.
2677 */
2678 if ((size_t) line > img->nr)
2679 line = img->nr;
2680
2681 current = 0;
2682 for (i = 0; i < line; i++)
2683 current += img->line[i].len;
2684
2685 /*
2686 * There's probably some smart way to do this, but I'll leave
2687 * that to the smart and beautiful people. I'm simple and stupid.
2688 */
2689 backwards = current;
2690 backwards_lno = line;
2691 forwards = current;
2692 forwards_lno = line;
2693 current_lno = line;
2694
2695 for (i = 0; ; i++) {
2696 if (match_fragment(state, img, preimage, postimage,
2697 current, current_lno, ws_rule,
2698 match_beginning, match_end))
2699 return current_lno;
2700
2701 again:
2702 if (backwards_lno == 0 && forwards_lno == img->nr)
2703 break;
2704
2705 if (i & 1) {
2706 if (backwards_lno == 0) {
2707 i++;
2708 goto again;
2709 }
2710 backwards_lno--;
2711 backwards -= img->line[backwards_lno].len;
2712 current = backwards;
2713 current_lno = backwards_lno;
2714 } else {
2715 if (forwards_lno == img->nr) {
2716 i++;
2717 goto again;
2718 }
2719 forwards += img->line[forwards_lno].len;
2720 forwards_lno++;
2721 current = forwards;
2722 current_lno = forwards_lno;
2723 }
2724
2725 }
2726 return -1;
2727 }
2728
2729 static void remove_first_line(struct image *img)
2730 {
2731 img->buf += img->line[0].len;
2732 img->len -= img->line[0].len;
2733 img->line++;
2734 img->nr--;
2735 }
2736
2737 static void remove_last_line(struct image *img)
2738 {
2739 img->len -= img->line[--img->nr].len;
2740 }
2741
2742 /*
2743 * The change from "preimage" and "postimage" has been found to
2744 * apply at applied_pos (counts in line numbers) in "img".
2745 * Update "img" to remove "preimage" and replace it with "postimage".
2746 */
2747 static void update_image(struct apply_state *state,
2748 struct image *img,
2749 int applied_pos,
2750 struct image *preimage,
2751 struct image *postimage)
2752 {
2753 /*
2754 * remove the copy of preimage at offset in img
2755 * and replace it with postimage
2756 */
2757 int i, nr;
2758 size_t remove_count, insert_count, applied_at = 0;
2759 char *result;
2760 int preimage_limit;
2761
2762 /*
2763 * If we are removing blank lines at the end of img,
2764 * the preimage may extend beyond the end.
2765 * If that is the case, we must be careful only to
2766 * remove the part of the preimage that falls within
2767 * the boundaries of img. Initialize preimage_limit
2768 * to the number of lines in the preimage that falls
2769 * within the boundaries.
2770 */
2771 preimage_limit = preimage->nr;
2772 if (preimage_limit > img->nr - applied_pos)
2773 preimage_limit = img->nr - applied_pos;
2774
2775 for (i = 0; i < applied_pos; i++)
2776 applied_at += img->line[i].len;
2777
2778 remove_count = 0;
2779 for (i = 0; i < preimage_limit; i++)
2780 remove_count += img->line[applied_pos + i].len;
2781 insert_count = postimage->len;
2782
2783 /* Adjust the contents */
2784 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2785 memcpy(result, img->buf, applied_at);
2786 memcpy(result + applied_at, postimage->buf, postimage->len);
2787 memcpy(result + applied_at + postimage->len,
2788 img->buf + (applied_at + remove_count),
2789 img->len - (applied_at + remove_count));
2790 free(img->buf);
2791 img->buf = result;
2792 img->len += insert_count - remove_count;
2793 result[img->len] = '\0';
2794
2795 /* Adjust the line table */
2796 nr = img->nr + postimage->nr - preimage_limit;
2797 if (preimage_limit < postimage->nr) {
2798 /*
2799 * NOTE: this knows that we never call remove_first_line()
2800 * on anything other than pre/post image.
2801 */
2802 REALLOC_ARRAY(img->line, nr);
2803 img->line_allocated = img->line;
2804 }
2805 if (preimage_limit != postimage->nr)
2806 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2807 img->line + applied_pos + preimage_limit,
2808 img->nr - (applied_pos + preimage_limit));
2809 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2810 if (!state->allow_overlap)
2811 for (i = 0; i < postimage->nr; i++)
2812 img->line[applied_pos + i].flag |= LINE_PATCHED;
2813 img->nr = nr;
2814 }
2815
2816 /*
2817 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2818 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2819 * replace the part of "img" with "postimage" text.
2820 */
2821 static int apply_one_fragment(struct apply_state *state,
2822 struct image *img, struct fragment *frag,
2823 int inaccurate_eof, unsigned ws_rule,
2824 int nth_fragment)
2825 {
2826 int match_beginning, match_end;
2827 const char *patch = frag->patch;
2828 int size = frag->size;
2829 char *old, *oldlines;
2830 struct strbuf newlines;
2831 int new_blank_lines_at_end = 0;
2832 int found_new_blank_lines_at_end = 0;
2833 int hunk_linenr = frag->linenr;
2834 unsigned long leading, trailing;
2835 int pos, applied_pos;
2836 struct image preimage;
2837 struct image postimage;
2838
2839 memset(&preimage, 0, sizeof(preimage));
2840 memset(&postimage, 0, sizeof(postimage));
2841 oldlines = xmalloc(size);
2842 strbuf_init(&newlines, size);
2843
2844 old = oldlines;
2845 while (size > 0) {
2846 char first;
2847 int len = linelen(patch, size);
2848 int plen;
2849 int added_blank_line = 0;
2850 int is_blank_context = 0;
2851 size_t start;
2852
2853 if (!len)
2854 break;
2855
2856 /*
2857 * "plen" is how much of the line we should use for
2858 * the actual patch data. Normally we just remove the
2859 * first character on the line, but if the line is
2860 * followed by "\ No newline", then we also remove the
2861 * last one (which is the newline, of course).
2862 */
2863 plen = len - 1;
2864 if (len < size && patch[len] == '\\')
2865 plen--;
2866 first = *patch;
2867 if (state->apply_in_reverse) {
2868 if (first == '-')
2869 first = '+';
2870 else if (first == '+')
2871 first = '-';
2872 }
2873
2874 switch (first) {
2875 case '\n':
2876 /* Newer GNU diff, empty context line */
2877 if (plen < 0)
2878 /* ... followed by '\No newline'; nothing */
2879 break;
2880 *old++ = '\n';
2881 strbuf_addch(&newlines, '\n');
2882 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2883 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2884 is_blank_context = 1;
2885 break;
2886 case ' ':
2887 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2888 ws_blank_line(patch + 1, plen, ws_rule))
2889 is_blank_context = 1;
2890 /* fallthrough */
2891 case '-':
2892 memcpy(old, patch + 1, plen);
2893 add_line_info(&preimage, old, plen,
2894 (first == ' ' ? LINE_COMMON : 0));
2895 old += plen;
2896 if (first == '-')
2897 break;
2898 /* fallthrough */
2899 case '+':
2900 /* --no-add does not add new lines */
2901 if (first == '+' && state->no_add)
2902 break;
2903
2904 start = newlines.len;
2905 if (first != '+' ||
2906 !state->whitespace_error ||
2907 state->ws_error_action != correct_ws_error) {
2908 strbuf_add(&newlines, patch + 1, plen);
2909 }
2910 else {
2911 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2912 }
2913 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2914 (first == '+' ? 0 : LINE_COMMON));
2915 if (first == '+' &&
2916 (ws_rule & WS_BLANK_AT_EOF) &&
2917 ws_blank_line(patch + 1, plen, ws_rule))
2918 added_blank_line = 1;
2919 break;
2920 case '@': case '\\':
2921 /* Ignore it, we already handled it */
2922 break;
2923 default:
2924 if (state->apply_verbosity > verbosity_normal)
2925 error(_("invalid start of line: '%c'"), first);
2926 applied_pos = -1;
2927 goto out;
2928 }
2929 if (added_blank_line) {
2930 if (!new_blank_lines_at_end)
2931 found_new_blank_lines_at_end = hunk_linenr;
2932 new_blank_lines_at_end++;
2933 }
2934 else if (is_blank_context)
2935 ;
2936 else
2937 new_blank_lines_at_end = 0;
2938 patch += len;
2939 size -= len;
2940 hunk_linenr++;
2941 }
2942 if (inaccurate_eof &&
2943 old > oldlines && old[-1] == '\n' &&
2944 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2945 old--;
2946 strbuf_setlen(&newlines, newlines.len - 1);
2947 preimage.line_allocated[preimage.nr - 1].len--;
2948 postimage.line_allocated[postimage.nr - 1].len--;
2949 }
2950
2951 leading = frag->leading;
2952 trailing = frag->trailing;
2953
2954 /*
2955 * A hunk to change lines at the beginning would begin with
2956 * @@ -1,L +N,M @@
2957 * but we need to be careful. -U0 that inserts before the second
2958 * line also has this pattern.
2959 *
2960 * And a hunk to add to an empty file would begin with
2961 * @@ -0,0 +N,M @@
2962 *
2963 * In other words, a hunk that is (frag->oldpos <= 1) with or
2964 * without leading context must match at the beginning.
2965 */
2966 match_beginning = (!frag->oldpos ||
2967 (frag->oldpos == 1 && !state->unidiff_zero));
2968
2969 /*
2970 * A hunk without trailing lines must match at the end.
2971 * However, we simply cannot tell if a hunk must match end
2972 * from the lack of trailing lines if the patch was generated
2973 * with unidiff without any context.
2974 */
2975 match_end = !state->unidiff_zero && !trailing;
2976
2977 pos = frag->newpos ? (frag->newpos - 1) : 0;
2978 preimage.buf = oldlines;
2979 preimage.len = old - oldlines;
2980 postimage.buf = newlines.buf;
2981 postimage.len = newlines.len;
2982 preimage.line = preimage.line_allocated;
2983 postimage.line = postimage.line_allocated;
2984
2985 for (;;) {
2986
2987 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2988 ws_rule, match_beginning, match_end);
2989
2990 if (applied_pos >= 0)
2991 break;
2992
2993 /* Am I at my context limits? */
2994 if ((leading <= state->p_context) && (trailing <= state->p_context))
2995 break;
2996 if (match_beginning || match_end) {
2997 match_beginning = match_end = 0;
2998 continue;
2999 }
3000
3001 /*
3002 * Reduce the number of context lines; reduce both
3003 * leading and trailing if they are equal otherwise
3004 * just reduce the larger context.
3005 */
3006 if (leading >= trailing) {
3007 remove_first_line(&preimage);
3008 remove_first_line(&postimage);
3009 pos--;
3010 leading--;
3011 }
3012 if (trailing > leading) {
3013 remove_last_line(&preimage);
3014 remove_last_line(&postimage);
3015 trailing--;
3016 }
3017 }
3018
3019 if (applied_pos >= 0) {
3020 if (new_blank_lines_at_end &&
3021 preimage.nr + applied_pos >= img->nr &&
3022 (ws_rule & WS_BLANK_AT_EOF) &&
3023 state->ws_error_action != nowarn_ws_error) {
3024 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3025 found_new_blank_lines_at_end);
3026 if (state->ws_error_action == correct_ws_error) {
3027 while (new_blank_lines_at_end--)
3028 remove_last_line(&postimage);
3029 }
3030 /*
3031 * We would want to prevent write_out_results()
3032 * from taking place in apply_patch() that follows
3033 * the callchain led us here, which is:
3034 * apply_patch->check_patch_list->check_patch->
3035 * apply_data->apply_fragments->apply_one_fragment
3036 */
3037 if (state->ws_error_action == die_on_ws_error)
3038 state->apply = 0;
3039 }
3040
3041 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3042 int offset = applied_pos - pos;
3043 if (state->apply_in_reverse)
3044 offset = 0 - offset;
3045 fprintf_ln(stderr,
3046 Q_("Hunk #%d succeeded at %d (offset %d line).",
3047 "Hunk #%d succeeded at %d (offset %d lines).",
3048 offset),
3049 nth_fragment, applied_pos + 1, offset);
3050 }
3051
3052 /*
3053 * Warn if it was necessary to reduce the number
3054 * of context lines.
3055 */
3056 if ((leading != frag->leading ||
3057 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3058 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3059 " to apply fragment at %d"),
3060 leading, trailing, applied_pos+1);
3061 update_image(state, img, applied_pos, &preimage, &postimage);
3062 } else {
3063 if (state->apply_verbosity > verbosity_normal)
3064 error(_("while searching for:\n%.*s"),
3065 (int)(old - oldlines), oldlines);
3066 }
3067
3068 out:
3069 free(oldlines);
3070 strbuf_release(&newlines);
3071 free(preimage.line_allocated);
3072 free(postimage.line_allocated);
3073
3074 return (applied_pos < 0);
3075 }
3076
3077 static int apply_binary_fragment(struct apply_state *state,
3078 struct image *img,
3079 struct patch *patch)
3080 {
3081 struct fragment *fragment = patch->fragments;
3082 unsigned long len;
3083 void *dst;
3084
3085 if (!fragment)
3086 return error(_("missing binary patch data for '%s'"),
3087 patch->new_name ?
3088 patch->new_name :
3089 patch->old_name);
3090
3091 /* Binary patch is irreversible without the optional second hunk */
3092 if (state->apply_in_reverse) {
3093 if (!fragment->next)
3094 return error(_("cannot reverse-apply a binary patch "
3095 "without the reverse hunk to '%s'"),
3096 patch->new_name
3097 ? patch->new_name : patch->old_name);
3098 fragment = fragment->next;
3099 }
3100 switch (fragment->binary_patch_method) {
3101 case BINARY_DELTA_DEFLATED:
3102 dst = patch_delta(img->buf, img->len, fragment->patch,
3103 fragment->size, &len);
3104 if (!dst)
3105 return -1;
3106 clear_image(img);
3107 img->buf = dst;
3108 img->len = len;
3109 return 0;
3110 case BINARY_LITERAL_DEFLATED:
3111 clear_image(img);
3112 img->len = fragment->size;
3113 img->buf = xmemdupz(fragment->patch, img->len);
3114 return 0;
3115 }
3116 return -1;
3117 }
3118
3119 /*
3120 * Replace "img" with the result of applying the binary patch.
3121 * The binary patch data itself in patch->fragment is still kept
3122 * but the preimage prepared by the caller in "img" is freed here
3123 * or in the helper function apply_binary_fragment() this calls.
3124 */
3125 static int apply_binary(struct apply_state *state,
3126 struct image *img,
3127 struct patch *patch)
3128 {
3129 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3130 struct object_id oid;
3131 const unsigned hexsz = the_hash_algo->hexsz;
3132
3133 /*
3134 * For safety, we require patch index line to contain
3135 * full hex textual object ID for old and new, at least for now.
3136 */
3137 if (strlen(patch->old_oid_prefix) != hexsz ||
3138 strlen(patch->new_oid_prefix) != hexsz ||
3139 get_oid_hex(patch->old_oid_prefix, &oid) ||
3140 get_oid_hex(patch->new_oid_prefix, &oid))
3141 return error(_("cannot apply binary patch to '%s' "
3142 "without full index line"), name);
3143
3144 if (patch->old_name) {
3145 /*
3146 * See if the old one matches what the patch
3147 * applies to.
3148 */
3149 hash_object_file(img->buf, img->len, blob_type, &oid);
3150 if (strcmp(oid_to_hex(&oid), patch->old_oid_prefix))
3151 return error(_("the patch applies to '%s' (%s), "
3152 "which does not match the "
3153 "current contents."),
3154 name, oid_to_hex(&oid));
3155 }
3156 else {
3157 /* Otherwise, the old one must be empty. */
3158 if (img->len)
3159 return error(_("the patch applies to an empty "
3160 "'%s' but it is not empty"), name);
3161 }
3162
3163 get_oid_hex(patch->new_oid_prefix, &oid);
3164 if (is_null_oid(&oid)) {
3165 clear_image(img);
3166 return 0; /* deletion patch */
3167 }
3168
3169 if (has_object_file(&oid)) {
3170 /* We already have the postimage */
3171 enum object_type type;
3172 unsigned long size;
3173 char *result;
3174
3175 result = read_object_file(&oid, &type, &size);
3176 if (!result)
3177 return error(_("the necessary postimage %s for "
3178 "'%s' cannot be read"),
3179 patch->new_oid_prefix, name);
3180 clear_image(img);
3181 img->buf = result;
3182 img->len = size;
3183 } else {
3184 /*
3185 * We have verified buf matches the preimage;
3186 * apply the patch data to it, which is stored
3187 * in the patch->fragments->{patch,size}.
3188 */
3189 if (apply_binary_fragment(state, img, patch))
3190 return error(_("binary patch does not apply to '%s'"),
3191 name);
3192
3193 /* verify that the result matches */
3194 hash_object_file(img->buf, img->len, blob_type, &oid);
3195 if (strcmp(oid_to_hex(&oid), patch->new_oid_prefix))
3196 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3197 name, patch->new_oid_prefix, oid_to_hex(&oid));
3198 }
3199
3200 return 0;
3201 }
3202
3203 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3204 {
3205 struct fragment *frag = patch->fragments;
3206 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3207 unsigned ws_rule = patch->ws_rule;
3208 unsigned inaccurate_eof = patch->inaccurate_eof;
3209 int nth = 0;
3210
3211 if (patch->is_binary)
3212 return apply_binary(state, img, patch);
3213
3214 while (frag) {
3215 nth++;
3216 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3217 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3218 if (!state->apply_with_reject)
3219 return -1;
3220 frag->rejected = 1;
3221 }
3222 frag = frag->next;
3223 }
3224 return 0;
3225 }
3226
3227 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3228 {
3229 if (S_ISGITLINK(mode)) {
3230 strbuf_grow(buf, 100);
3231 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3232 } else {
3233 enum object_type type;
3234 unsigned long sz;
3235 char *result;
3236
3237 result = read_object_file(oid, &type, &sz);
3238 if (!result)
3239 return -1;
3240 /* XXX read_sha1_file NUL-terminates */
3241 strbuf_attach(buf, result, sz, sz + 1);
3242 }
3243 return 0;
3244 }
3245
3246 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3247 {
3248 if (!ce)
3249 return 0;
3250 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3251 }
3252
3253 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3254 {
3255 struct string_list_item *item;
3256
3257 if (name == NULL)
3258 return NULL;
3259
3260 item = string_list_lookup(&state->fn_table, name);
3261 if (item != NULL)
3262 return (struct patch *)item->util;
3263
3264 return NULL;
3265 }
3266
3267 /*
3268 * item->util in the filename table records the status of the path.
3269 * Usually it points at a patch (whose result records the contents
3270 * of it after applying it), but it could be PATH_WAS_DELETED for a
3271 * path that a previously applied patch has already removed, or
3272 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3273 *
3274 * The latter is needed to deal with a case where two paths A and B
3275 * are swapped by first renaming A to B and then renaming B to A;
3276 * moving A to B should not be prevented due to presence of B as we
3277 * will remove it in a later patch.
3278 */
3279 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3280 #define PATH_WAS_DELETED ((struct patch *) -1)
3281
3282 static int to_be_deleted(struct patch *patch)
3283 {
3284 return patch == PATH_TO_BE_DELETED;
3285 }
3286
3287 static int was_deleted(struct patch *patch)
3288 {
3289 return patch == PATH_WAS_DELETED;
3290 }
3291
3292 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3293 {
3294 struct string_list_item *item;
3295
3296 /*
3297 * Always add new_name unless patch is a deletion
3298 * This should cover the cases for normal diffs,
3299 * file creations and copies
3300 */
3301 if (patch->new_name != NULL) {
3302 item = string_list_insert(&state->fn_table, patch->new_name);
3303 item->util = patch;
3304 }
3305
3306 /*
3307 * store a failure on rename/deletion cases because
3308 * later chunks shouldn't patch old names
3309 */
3310 if ((patch->new_name == NULL) || (patch->is_rename)) {
3311 item = string_list_insert(&state->fn_table, patch->old_name);
3312 item->util = PATH_WAS_DELETED;
3313 }
3314 }
3315
3316 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3317 {
3318 /*
3319 * store information about incoming file deletion
3320 */
3321 while (patch) {
3322 if ((patch->new_name == NULL) || (patch->is_rename)) {
3323 struct string_list_item *item;
3324 item = string_list_insert(&state->fn_table, patch->old_name);
3325 item->util = PATH_TO_BE_DELETED;
3326 }
3327 patch = patch->next;
3328 }
3329 }
3330
3331 static int checkout_target(struct index_state *istate,
3332 struct cache_entry *ce, struct stat *st)
3333 {
3334 struct checkout costate = CHECKOUT_INIT;
3335
3336 costate.refresh_cache = 1;
3337 costate.istate = istate;
3338 if (checkout_entry(ce, &costate, NULL, NULL) ||
3339 lstat(ce->name, st))
3340 return error(_("cannot checkout %s"), ce->name);
3341 return 0;
3342 }
3343
3344 static struct patch *previous_patch(struct apply_state *state,
3345 struct patch *patch,
3346 int *gone)
3347 {
3348 struct patch *previous;
3349
3350 *gone = 0;
3351 if (patch->is_copy || patch->is_rename)
3352 return NULL; /* "git" patches do not depend on the order */
3353
3354 previous = in_fn_table(state, patch->old_name);
3355 if (!previous)
3356 return NULL;
3357
3358 if (to_be_deleted(previous))
3359 return NULL; /* the deletion hasn't happened yet */
3360
3361 if (was_deleted(previous))
3362 *gone = 1;
3363
3364 return previous;
3365 }
3366
3367 static int verify_index_match(struct apply_state *state,
3368 const struct cache_entry *ce,
3369 struct stat *st)
3370 {
3371 if (S_ISGITLINK(ce->ce_mode)) {
3372 if (!S_ISDIR(st->st_mode))
3373 return -1;
3374 return 0;
3375 }
3376 return ie_match_stat(state->repo->index, ce, st,
3377 CE_MATCH_IGNORE_VALID | CE_MATCH_IGNORE_SKIP_WORKTREE);
3378 }
3379
3380 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3381
3382 static int load_patch_target(struct apply_state *state,
3383 struct strbuf *buf,
3384 const struct cache_entry *ce,
3385 struct stat *st,
3386 struct patch *patch,
3387 const char *name,
3388 unsigned expected_mode)
3389 {
3390 if (state->cached || state->check_index) {
3391 if (read_file_or_gitlink(ce, buf))
3392 return error(_("failed to read %s"), name);
3393 } else if (name) {
3394 if (S_ISGITLINK(expected_mode)) {
3395 if (ce)
3396 return read_file_or_gitlink(ce, buf);
3397 else
3398 return SUBMODULE_PATCH_WITHOUT_INDEX;
3399 } else if (has_symlink_leading_path(name, strlen(name))) {
3400 return error(_("reading from '%s' beyond a symbolic link"), name);
3401 } else {
3402 if (read_old_data(st, patch, name, buf))
3403 return error(_("failed to read %s"), name);
3404 }
3405 }
3406 return 0;
3407 }
3408
3409 /*
3410 * We are about to apply "patch"; populate the "image" with the
3411 * current version we have, from the working tree or from the index,
3412 * depending on the situation e.g. --cached/--index. If we are
3413 * applying a non-git patch that incrementally updates the tree,
3414 * we read from the result of a previous diff.
3415 */
3416 static int load_preimage(struct apply_state *state,
3417 struct image *image,
3418 struct patch *patch, struct stat *st,
3419 const struct cache_entry *ce)
3420 {
3421 struct strbuf buf = STRBUF_INIT;
3422 size_t len;
3423 char *img;
3424 struct patch *previous;
3425 int status;
3426
3427 previous = previous_patch(state, patch, &status);
3428 if (status)
3429 return error(_("path %s has been renamed/deleted"),
3430 patch->old_name);
3431 if (previous) {
3432 /* We have a patched copy in memory; use that. */
3433 strbuf_add(&buf, previous->result, previous->resultsize);
3434 } else {
3435 status = load_patch_target(state, &buf, ce, st, patch,
3436 patch->old_name, patch->old_mode);
3437 if (status < 0)
3438 return status;
3439 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3440 /*
3441 * There is no way to apply subproject
3442 * patch without looking at the index.
3443 * NEEDSWORK: shouldn't this be flagged
3444 * as an error???
3445 */
3446 free_fragment_list(patch->fragments);
3447 patch->fragments = NULL;
3448 } else if (status) {
3449 return error(_("failed to read %s"), patch->old_name);
3450 }
3451 }
3452
3453 img = strbuf_detach(&buf, &len);
3454 prepare_image(image, img, len, !patch->is_binary);
3455 return 0;
3456 }
3457
3458 static int three_way_merge(struct apply_state *state,
3459 struct image *image,
3460 char *path,
3461 const struct object_id *base,
3462 const struct object_id *ours,
3463 const struct object_id *theirs)
3464 {
3465 mmfile_t base_file, our_file, their_file;
3466 mmbuffer_t result = { NULL };
3467 int status;
3468
3469 read_mmblob(&base_file, base);
3470 read_mmblob(&our_file, ours);
3471 read_mmblob(&their_file, theirs);
3472 status = ll_merge(&result, path,
3473 &base_file, "base",
3474 &our_file, "ours",
3475 &their_file, "theirs",
3476 state->repo->index,
3477 NULL);
3478 free(base_file.ptr);
3479 free(our_file.ptr);
3480 free(their_file.ptr);
3481 if (status < 0 || !result.ptr) {
3482 free(result.ptr);
3483 return -1;
3484 }
3485 clear_image(image);
3486 image->buf = result.ptr;
3487 image->len = result.size;
3488
3489 return status;
3490 }
3491
3492 /*
3493 * When directly falling back to add/add three-way merge, we read from
3494 * the current contents of the new_name. In no cases other than that
3495 * this function will be called.
3496 */
3497 static int load_current(struct apply_state *state,
3498 struct image *image,
3499 struct patch *patch)
3500 {
3501 struct strbuf buf = STRBUF_INIT;
3502 int status, pos;
3503 size_t len;
3504 char *img;
3505 struct stat st;
3506 struct cache_entry *ce;
3507 char *name = patch->new_name;
3508 unsigned mode = patch->new_mode;
3509
3510 if (!patch->is_new)
3511 BUG("patch to %s is not a creation", patch->old_name);
3512
3513 pos = index_name_pos(state->repo->index, name, strlen(name));
3514 if (pos < 0)
3515 return error(_("%s: does not exist in index"), name);
3516 ce = state->repo->index->cache[pos];
3517 if (lstat(name, &st)) {
3518 if (errno != ENOENT)
3519 return error_errno("%s", name);
3520 if (checkout_target(state->repo->index, ce, &st))
3521 return -1;
3522 }
3523 if (verify_index_match(state, ce, &st))
3524 return error(_("%s: does not match index"), name);
3525
3526 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3527 if (status < 0)
3528 return status;
3529 else if (status)
3530 return -1;
3531 img = strbuf_detach(&buf, &len);
3532 prepare_image(image, img, len, !patch->is_binary);
3533 return 0;
3534 }
3535
3536 static int try_threeway(struct apply_state *state,
3537 struct image *image,
3538 struct patch *patch,
3539 struct stat *st,
3540 const struct cache_entry *ce)
3541 {
3542 struct object_id pre_oid, post_oid, our_oid;
3543 struct strbuf buf = STRBUF_INIT;
3544 size_t len;
3545 int status;
3546 char *img;
3547 struct image tmp_image;
3548
3549 /* No point falling back to 3-way merge in these cases */
3550 if (patch->is_delete ||
3551 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3552 return -1;
3553
3554 /* Preimage the patch was prepared for */
3555 if (patch->is_new)
3556 write_object_file("", 0, blob_type, &pre_oid);
3557 else if (get_oid(patch->old_oid_prefix, &pre_oid) ||
3558 read_blob_object(&buf, &pre_oid, patch->old_mode))
3559 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3560
3561 if (state->apply_verbosity > verbosity_silent)
3562 fprintf(stderr, _("Falling back to three-way merge...\n"));
3563
3564 img = strbuf_detach(&buf, &len);
3565 prepare_image(&tmp_image, img, len, 1);
3566 /* Apply the patch to get the post image */
3567 if (apply_fragments(state, &tmp_image, patch) < 0) {
3568 clear_image(&tmp_image);
3569 return -1;
3570 }
3571 /* post_oid is theirs */
3572 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &post_oid);
3573 clear_image(&tmp_image);
3574
3575 /* our_oid is ours */
3576 if (patch->is_new) {
3577 if (load_current(state, &tmp_image, patch))
3578 return error(_("cannot read the current contents of '%s'"),
3579 patch->new_name);
3580 } else {
3581 if (load_preimage(state, &tmp_image, patch, st, ce))
3582 return error(_("cannot read the current contents of '%s'"),
3583 patch->old_name);
3584 }
3585 write_object_file(tmp_image.buf, tmp_image.len, blob_type, &our_oid);
3586 clear_image(&tmp_image);
3587
3588 /* in-core three-way merge between post and our using pre as base */
3589 status = three_way_merge(state, image, patch->new_name,
3590 &pre_oid, &our_oid, &post_oid);
3591 if (status < 0) {
3592 if (state->apply_verbosity > verbosity_silent)
3593 fprintf(stderr,
3594 _("Failed to fall back on three-way merge...\n"));
3595 return status;
3596 }
3597
3598 if (status) {
3599 patch->conflicted_threeway = 1;
3600 if (patch->is_new)
3601 oidclr(&patch->threeway_stage[0]);
3602 else
3603 oidcpy(&patch->threeway_stage[0], &pre_oid);
3604 oidcpy(&patch->threeway_stage[1], &our_oid);
3605 oidcpy(&patch->threeway_stage[2], &post_oid);
3606 if (state->apply_verbosity > verbosity_silent)
3607 fprintf(stderr,
3608 _("Applied patch to '%s' with conflicts.\n"),
3609 patch->new_name);
3610 } else {
3611 if (state->apply_verbosity > verbosity_silent)
3612 fprintf(stderr,
3613 _("Applied patch to '%s' cleanly.\n"),
3614 patch->new_name);
3615 }
3616 return 0;
3617 }
3618
3619 static int apply_data(struct apply_state *state, struct patch *patch,
3620 struct stat *st, const struct cache_entry *ce)
3621 {
3622 struct image image;
3623
3624 if (load_preimage(state, &image, patch, st, ce) < 0)
3625 return -1;
3626
3627 if (patch->direct_to_threeway ||
3628 apply_fragments(state, &image, patch) < 0) {
3629 /* Note: with --reject, apply_fragments() returns 0 */
3630 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3631 return -1;
3632 }
3633 patch->result = image.buf;
3634 patch->resultsize = image.len;
3635 add_to_fn_table(state, patch);
3636 free(image.line_allocated);
3637
3638 if (0 < patch->is_delete && patch->resultsize)
3639 return error(_("removal patch leaves file contents"));
3640
3641 return 0;
3642 }
3643
3644 /*
3645 * If "patch" that we are looking at modifies or deletes what we have,
3646 * we would want it not to lose any local modification we have, either
3647 * in the working tree or in the index.
3648 *
3649 * This also decides if a non-git patch is a creation patch or a
3650 * modification to an existing empty file. We do not check the state
3651 * of the current tree for a creation patch in this function; the caller
3652 * check_patch() separately makes sure (and errors out otherwise) that
3653 * the path the patch creates does not exist in the current tree.
3654 */
3655 static int check_preimage(struct apply_state *state,
3656 struct patch *patch,
3657 struct cache_entry **ce,
3658 struct stat *st)
3659 {
3660 const char *old_name = patch->old_name;
3661 struct patch *previous = NULL;
3662 int stat_ret = 0, status;
3663 unsigned st_mode = 0;
3664
3665 if (!old_name)
3666 return 0;
3667
3668 assert(patch->is_new <= 0);
3669 previous = previous_patch(state, patch, &status);
3670
3671 if (status)
3672 return error(_("path %s has been renamed/deleted"), old_name);
3673 if (previous) {
3674 st_mode = previous->new_mode;
3675 } else if (!state->cached) {
3676 stat_ret = lstat(old_name, st);
3677 if (stat_ret && errno != ENOENT)
3678 return error_errno("%s", old_name);
3679 }
3680
3681 if (state->check_index && !previous) {
3682 int pos = index_name_pos(state->repo->index, old_name,
3683 strlen(old_name));
3684 if (pos < 0) {
3685 if (patch->is_new < 0)
3686 goto is_new;
3687 return error(_("%s: does not exist in index"), old_name);
3688 }
3689 *ce = state->repo->index->cache[pos];
3690 if (stat_ret < 0) {
3691 if (checkout_target(state->repo->index, *ce, st))
3692 return -1;
3693 }
3694 if (!state->cached && verify_index_match(state, *ce, st))
3695 return error(_("%s: does not match index"), old_name);
3696 if (state->cached)
3697 st_mode = (*ce)->ce_mode;
3698 } else if (stat_ret < 0) {
3699 if (patch->is_new < 0)
3700 goto is_new;
3701 return error_errno("%s", old_name);
3702 }
3703
3704 if (!state->cached && !previous)
3705 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3706
3707 if (patch->is_new < 0)
3708 patch->is_new = 0;
3709 if (!patch->old_mode)
3710 patch->old_mode = st_mode;
3711 if ((st_mode ^ patch->old_mode) & S_IFMT)
3712 return error(_("%s: wrong type"), old_name);
3713 if (st_mode != patch->old_mode)
3714 warning(_("%s has type %o, expected %o"),
3715 old_name, st_mode, patch->old_mode);
3716 if (!patch->new_mode && !patch->is_delete)
3717 patch->new_mode = st_mode;
3718 return 0;
3719
3720 is_new:
3721 patch->is_new = 1;
3722 patch->is_delete = 0;
3723 FREE_AND_NULL(patch->old_name);
3724 return 0;
3725 }
3726
3727
3728 #define EXISTS_IN_INDEX 1
3729 #define EXISTS_IN_WORKTREE 2
3730
3731 static int check_to_create(struct apply_state *state,
3732 const char *new_name,
3733 int ok_if_exists)
3734 {
3735 struct stat nst;
3736
3737 if (state->check_index &&
3738 index_name_pos(state->repo->index, new_name, strlen(new_name)) >= 0 &&
3739 !ok_if_exists)
3740 return EXISTS_IN_INDEX;
3741 if (state->cached)
3742 return 0;
3743
3744 if (!lstat(new_name, &nst)) {
3745 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3746 return 0;
3747 /*
3748 * A leading component of new_name might be a symlink
3749 * that is going to be removed with this patch, but
3750 * still pointing at somewhere that has the path.
3751 * In such a case, path "new_name" does not exist as
3752 * far as git is concerned.
3753 */
3754 if (has_symlink_leading_path(new_name, strlen(new_name)))
3755 return 0;
3756
3757 return EXISTS_IN_WORKTREE;
3758 } else if (!is_missing_file_error(errno)) {
3759 return error_errno("%s", new_name);
3760 }
3761 return 0;
3762 }
3763
3764 static uintptr_t register_symlink_changes(struct apply_state *state,
3765 const char *path,
3766 uintptr_t what)
3767 {
3768 struct string_list_item *ent;
3769
3770 ent = string_list_lookup(&state->symlink_changes, path);
3771 if (!ent) {
3772 ent = string_list_insert(&state->symlink_changes, path);
3773 ent->util = (void *)0;
3774 }
3775 ent->util = (void *)(what | ((uintptr_t)ent->util));
3776 return (uintptr_t)ent->util;
3777 }
3778
3779 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3780 {
3781 struct string_list_item *ent;
3782
3783 ent = string_list_lookup(&state->symlink_changes, path);
3784 if (!ent)
3785 return 0;
3786 return (uintptr_t)ent->util;
3787 }
3788
3789 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3790 {
3791 for ( ; patch; patch = patch->next) {
3792 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3793 (patch->is_rename || patch->is_delete))
3794 /* the symlink at patch->old_name is removed */
3795 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3796
3797 if (patch->new_name && S_ISLNK(patch->new_mode))
3798 /* the symlink at patch->new_name is created or remains */
3799 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3800 }
3801 }
3802
3803 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3804 {
3805 do {
3806 unsigned int change;
3807
3808 while (--name->len && name->buf[name->len] != '/')
3809 ; /* scan backwards */
3810 if (!name->len)
3811 break;
3812 name->buf[name->len] = '\0';
3813 change = check_symlink_changes(state, name->buf);
3814 if (change & APPLY_SYMLINK_IN_RESULT)
3815 return 1;
3816 if (change & APPLY_SYMLINK_GOES_AWAY)
3817 /*
3818 * This cannot be "return 0", because we may
3819 * see a new one created at a higher level.
3820 */
3821 continue;
3822
3823 /* otherwise, check the preimage */
3824 if (state->check_index) {
3825 struct cache_entry *ce;
3826
3827 ce = index_file_exists(state->repo->index, name->buf,
3828 name->len, ignore_case);
3829 if (ce && S_ISLNK(ce->ce_mode))
3830 return 1;
3831 } else {
3832 struct stat st;
3833 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3834 return 1;
3835 }
3836 } while (1);
3837 return 0;
3838 }
3839
3840 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3841 {
3842 int ret;
3843 struct strbuf name = STRBUF_INIT;
3844
3845