Merge branch 'rs/apply-validate-input' into maint
[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 "blob.h"
12 #include "delta.h"
13 #include "diff.h"
14 #include "dir.h"
15 #include "xdiff-interface.h"
16 #include "ll-merge.h"
17 #include "lockfile.h"
18 #include "parse-options.h"
19 #include "quote.h"
20 #include "rerere.h"
21 #include "apply.h"
22
23 static void git_apply_config(void)
24 {
25 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
26 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
27 git_config(git_default_config, NULL);
28 }
29
30 static int parse_whitespace_option(struct apply_state *state, const char *option)
31 {
32 if (!option) {
33 state->ws_error_action = warn_on_ws_error;
34 return 0;
35 }
36 if (!strcmp(option, "warn")) {
37 state->ws_error_action = warn_on_ws_error;
38 return 0;
39 }
40 if (!strcmp(option, "nowarn")) {
41 state->ws_error_action = nowarn_ws_error;
42 return 0;
43 }
44 if (!strcmp(option, "error")) {
45 state->ws_error_action = die_on_ws_error;
46 return 0;
47 }
48 if (!strcmp(option, "error-all")) {
49 state->ws_error_action = die_on_ws_error;
50 state->squelch_whitespace_errors = 0;
51 return 0;
52 }
53 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
54 state->ws_error_action = correct_ws_error;
55 return 0;
56 }
57 return error(_("unrecognized whitespace option '%s'"), option);
58 }
59
60 static int parse_ignorewhitespace_option(struct apply_state *state,
61 const char *option)
62 {
63 if (!option || !strcmp(option, "no") ||
64 !strcmp(option, "false") || !strcmp(option, "never") ||
65 !strcmp(option, "none")) {
66 state->ws_ignore_action = ignore_ws_none;
67 return 0;
68 }
69 if (!strcmp(option, "change")) {
70 state->ws_ignore_action = ignore_ws_change;
71 return 0;
72 }
73 return error(_("unrecognized whitespace ignore option '%s'"), option);
74 }
75
76 int init_apply_state(struct apply_state *state,
77 const char *prefix,
78 struct lock_file *lock_file)
79 {
80 memset(state, 0, sizeof(*state));
81 state->prefix = prefix;
82 state->prefix_length = state->prefix ? strlen(state->prefix) : 0;
83 state->lock_file = lock_file;
84 state->newfd = -1;
85 state->apply = 1;
86 state->line_termination = '\n';
87 state->p_value = 1;
88 state->p_context = UINT_MAX;
89 state->squelch_whitespace_errors = 5;
90 state->ws_error_action = warn_on_ws_error;
91 state->ws_ignore_action = ignore_ws_none;
92 state->linenr = 1;
93 string_list_init(&state->fn_table, 0);
94 string_list_init(&state->limit_by_name, 0);
95 string_list_init(&state->symlink_changes, 0);
96 strbuf_init(&state->root, 0);
97
98 git_apply_config();
99 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
100 return -1;
101 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
102 return -1;
103 return 0;
104 }
105
106 void clear_apply_state(struct apply_state *state)
107 {
108 string_list_clear(&state->limit_by_name, 0);
109 string_list_clear(&state->symlink_changes, 0);
110 strbuf_release(&state->root);
111
112 /* &state->fn_table is cleared at the end of apply_patch() */
113 }
114
115 static void mute_routine(const char *msg, va_list params)
116 {
117 /* do nothing */
118 }
119
120 int check_apply_state(struct apply_state *state, int force_apply)
121 {
122 int is_not_gitdir = !startup_info->have_repository;
123
124 if (state->apply_with_reject && state->threeway)
125 return error(_("--reject and --3way cannot be used together."));
126 if (state->cached && state->threeway)
127 return error(_("--cached and --3way cannot be used together."));
128 if (state->threeway) {
129 if (is_not_gitdir)
130 return error(_("--3way outside a repository"));
131 state->check_index = 1;
132 }
133 if (state->apply_with_reject) {
134 state->apply = 1;
135 if (state->apply_verbosity == verbosity_normal)
136 state->apply_verbosity = verbosity_verbose;
137 }
138 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
139 state->apply = 0;
140 if (state->check_index && is_not_gitdir)
141 return error(_("--index outside a repository"));
142 if (state->cached) {
143 if (is_not_gitdir)
144 return error(_("--cached outside a repository"));
145 state->check_index = 1;
146 }
147 if (state->check_index)
148 state->unsafe_paths = 0;
149 if (!state->lock_file)
150 return error("BUG: state->lock_file should not be NULL");
151
152 if (state->apply_verbosity <= verbosity_silent) {
153 state->saved_error_routine = get_error_routine();
154 state->saved_warn_routine = get_warn_routine();
155 set_error_routine(mute_routine);
156 set_warn_routine(mute_routine);
157 }
158
159 return 0;
160 }
161
162 static void set_default_whitespace_mode(struct apply_state *state)
163 {
164 if (!state->whitespace_option && !apply_default_whitespace)
165 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
166 }
167
168 /*
169 * This represents one "hunk" from a patch, starting with
170 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
171 * patch text is pointed at by patch, and its byte length
172 * is stored in size. leading and trailing are the number
173 * of context lines.
174 */
175 struct fragment {
176 unsigned long leading, trailing;
177 unsigned long oldpos, oldlines;
178 unsigned long newpos, newlines;
179 /*
180 * 'patch' is usually borrowed from buf in apply_patch(),
181 * but some codepaths store an allocated buffer.
182 */
183 const char *patch;
184 unsigned free_patch:1,
185 rejected:1;
186 int size;
187 int linenr;
188 struct fragment *next;
189 };
190
191 /*
192 * When dealing with a binary patch, we reuse "leading" field
193 * to store the type of the binary hunk, either deflated "delta"
194 * or deflated "literal".
195 */
196 #define binary_patch_method leading
197 #define BINARY_DELTA_DEFLATED 1
198 #define BINARY_LITERAL_DEFLATED 2
199
200 /*
201 * This represents a "patch" to a file, both metainfo changes
202 * such as creation/deletion, filemode and content changes represented
203 * as a series of fragments.
204 */
205 struct patch {
206 char *new_name, *old_name, *def_name;
207 unsigned int old_mode, new_mode;
208 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
209 int rejected;
210 unsigned ws_rule;
211 int lines_added, lines_deleted;
212 int score;
213 int extension_linenr; /* first line specifying delete/new/rename/copy */
214 unsigned int is_toplevel_relative:1;
215 unsigned int inaccurate_eof:1;
216 unsigned int is_binary:1;
217 unsigned int is_copy:1;
218 unsigned int is_rename:1;
219 unsigned int recount:1;
220 unsigned int conflicted_threeway:1;
221 unsigned int direct_to_threeway:1;
222 struct fragment *fragments;
223 char *result;
224 size_t resultsize;
225 char old_sha1_prefix[41];
226 char new_sha1_prefix[41];
227 struct patch *next;
228
229 /* three-way fallback result */
230 struct object_id threeway_stage[3];
231 };
232
233 static void free_fragment_list(struct fragment *list)
234 {
235 while (list) {
236 struct fragment *next = list->next;
237 if (list->free_patch)
238 free((char *)list->patch);
239 free(list);
240 list = next;
241 }
242 }
243
244 static void free_patch(struct patch *patch)
245 {
246 free_fragment_list(patch->fragments);
247 free(patch->def_name);
248 free(patch->old_name);
249 free(patch->new_name);
250 free(patch->result);
251 free(patch);
252 }
253
254 static void free_patch_list(struct patch *list)
255 {
256 while (list) {
257 struct patch *next = list->next;
258 free_patch(list);
259 list = next;
260 }
261 }
262
263 /*
264 * A line in a file, len-bytes long (includes the terminating LF,
265 * except for an incomplete line at the end if the file ends with
266 * one), and its contents hashes to 'hash'.
267 */
268 struct line {
269 size_t len;
270 unsigned hash : 24;
271 unsigned flag : 8;
272 #define LINE_COMMON 1
273 #define LINE_PATCHED 2
274 };
275
276 /*
277 * This represents a "file", which is an array of "lines".
278 */
279 struct image {
280 char *buf;
281 size_t len;
282 size_t nr;
283 size_t alloc;
284 struct line *line_allocated;
285 struct line *line;
286 };
287
288 static uint32_t hash_line(const char *cp, size_t len)
289 {
290 size_t i;
291 uint32_t h;
292 for (i = 0, h = 0; i < len; i++) {
293 if (!isspace(cp[i])) {
294 h = h * 3 + (cp[i] & 0xff);
295 }
296 }
297 return h;
298 }
299
300 /*
301 * Compare lines s1 of length n1 and s2 of length n2, ignoring
302 * whitespace difference. Returns 1 if they match, 0 otherwise
303 */
304 static int fuzzy_matchlines(const char *s1, size_t n1,
305 const char *s2, size_t n2)
306 {
307 const char *last1 = s1 + n1 - 1;
308 const char *last2 = s2 + n2 - 1;
309 int result = 0;
310
311 /* ignore line endings */
312 while ((*last1 == '\r') || (*last1 == '\n'))
313 last1--;
314 while ((*last2 == '\r') || (*last2 == '\n'))
315 last2--;
316
317 /* skip leading whitespaces, if both begin with whitespace */
318 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
319 while (isspace(*s1) && (s1 <= last1))
320 s1++;
321 while (isspace(*s2) && (s2 <= last2))
322 s2++;
323 }
324 /* early return if both lines are empty */
325 if ((s1 > last1) && (s2 > last2))
326 return 1;
327 while (!result) {
328 result = *s1++ - *s2++;
329 /*
330 * Skip whitespace inside. We check for whitespace on
331 * both buffers because we don't want "a b" to match
332 * "ab"
333 */
334 if (isspace(*s1) && isspace(*s2)) {
335 while (isspace(*s1) && s1 <= last1)
336 s1++;
337 while (isspace(*s2) && s2 <= last2)
338 s2++;
339 }
340 /*
341 * If we reached the end on one side only,
342 * lines don't match
343 */
344 if (
345 ((s2 > last2) && (s1 <= last1)) ||
346 ((s1 > last1) && (s2 <= last2)))
347 return 0;
348 if ((s1 > last1) && (s2 > last2))
349 break;
350 }
351
352 return !result;
353 }
354
355 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
356 {
357 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
358 img->line_allocated[img->nr].len = len;
359 img->line_allocated[img->nr].hash = hash_line(bol, len);
360 img->line_allocated[img->nr].flag = flag;
361 img->nr++;
362 }
363
364 /*
365 * "buf" has the file contents to be patched (read from various sources).
366 * attach it to "image" and add line-based index to it.
367 * "image" now owns the "buf".
368 */
369 static void prepare_image(struct image *image, char *buf, size_t len,
370 int prepare_linetable)
371 {
372 const char *cp, *ep;
373
374 memset(image, 0, sizeof(*image));
375 image->buf = buf;
376 image->len = len;
377
378 if (!prepare_linetable)
379 return;
380
381 ep = image->buf + image->len;
382 cp = image->buf;
383 while (cp < ep) {
384 const char *next;
385 for (next = cp; next < ep && *next != '\n'; next++)
386 ;
387 if (next < ep)
388 next++;
389 add_line_info(image, cp, next - cp, 0);
390 cp = next;
391 }
392 image->line = image->line_allocated;
393 }
394
395 static void clear_image(struct image *image)
396 {
397 free(image->buf);
398 free(image->line_allocated);
399 memset(image, 0, sizeof(*image));
400 }
401
402 /* fmt must contain _one_ %s and no other substitution */
403 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
404 {
405 struct strbuf sb = STRBUF_INIT;
406
407 if (patch->old_name && patch->new_name &&
408 strcmp(patch->old_name, patch->new_name)) {
409 quote_c_style(patch->old_name, &sb, NULL, 0);
410 strbuf_addstr(&sb, " => ");
411 quote_c_style(patch->new_name, &sb, NULL, 0);
412 } else {
413 const char *n = patch->new_name;
414 if (!n)
415 n = patch->old_name;
416 quote_c_style(n, &sb, NULL, 0);
417 }
418 fprintf(output, fmt, sb.buf);
419 fputc('\n', output);
420 strbuf_release(&sb);
421 }
422
423 #define SLOP (16)
424
425 static int read_patch_file(struct strbuf *sb, int fd)
426 {
427 if (strbuf_read(sb, fd, 0) < 0)
428 return error_errno("git apply: failed to read");
429
430 /*
431 * Make sure that we have some slop in the buffer
432 * so that we can do speculative "memcmp" etc, and
433 * see to it that it is NUL-filled.
434 */
435 strbuf_grow(sb, SLOP);
436 memset(sb->buf + sb->len, 0, SLOP);
437 return 0;
438 }
439
440 static unsigned long linelen(const char *buffer, unsigned long size)
441 {
442 unsigned long len = 0;
443 while (size--) {
444 len++;
445 if (*buffer++ == '\n')
446 break;
447 }
448 return len;
449 }
450
451 static int is_dev_null(const char *str)
452 {
453 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
454 }
455
456 #define TERM_SPACE 1
457 #define TERM_TAB 2
458
459 static int name_terminate(int c, int terminate)
460 {
461 if (c == ' ' && !(terminate & TERM_SPACE))
462 return 0;
463 if (c == '\t' && !(terminate & TERM_TAB))
464 return 0;
465
466 return 1;
467 }
468
469 /* remove double slashes to make --index work with such filenames */
470 static char *squash_slash(char *name)
471 {
472 int i = 0, j = 0;
473
474 if (!name)
475 return NULL;
476
477 while (name[i]) {
478 if ((name[j++] = name[i++]) == '/')
479 while (name[i] == '/')
480 i++;
481 }
482 name[j] = '\0';
483 return name;
484 }
485
486 static char *find_name_gnu(struct apply_state *state,
487 const char *line,
488 const char *def,
489 int p_value)
490 {
491 struct strbuf name = STRBUF_INIT;
492 char *cp;
493
494 /*
495 * Proposed "new-style" GNU patch/diff format; see
496 * http://marc.info/?l=git&m=112927316408690&w=2
497 */
498 if (unquote_c_style(&name, line, NULL)) {
499 strbuf_release(&name);
500 return NULL;
501 }
502
503 for (cp = name.buf; p_value; p_value--) {
504 cp = strchr(cp, '/');
505 if (!cp) {
506 strbuf_release(&name);
507 return NULL;
508 }
509 cp++;
510 }
511
512 strbuf_remove(&name, 0, cp - name.buf);
513 if (state->root.len)
514 strbuf_insert(&name, 0, state->root.buf, state->root.len);
515 return squash_slash(strbuf_detach(&name, NULL));
516 }
517
518 static size_t sane_tz_len(const char *line, size_t len)
519 {
520 const char *tz, *p;
521
522 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
523 return 0;
524 tz = line + len - strlen(" +0500");
525
526 if (tz[1] != '+' && tz[1] != '-')
527 return 0;
528
529 for (p = tz + 2; p != line + len; p++)
530 if (!isdigit(*p))
531 return 0;
532
533 return line + len - tz;
534 }
535
536 static size_t tz_with_colon_len(const char *line, size_t len)
537 {
538 const char *tz, *p;
539
540 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
541 return 0;
542 tz = line + len - strlen(" +08:00");
543
544 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
545 return 0;
546 p = tz + 2;
547 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
548 !isdigit(*p++) || !isdigit(*p++))
549 return 0;
550
551 return line + len - tz;
552 }
553
554 static size_t date_len(const char *line, size_t len)
555 {
556 const char *date, *p;
557
558 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
559 return 0;
560 p = date = line + len - strlen("72-02-05");
561
562 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
563 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
564 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
565 return 0;
566
567 if (date - line >= strlen("19") &&
568 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
569 date -= strlen("19");
570
571 return line + len - date;
572 }
573
574 static size_t short_time_len(const char *line, size_t len)
575 {
576 const char *time, *p;
577
578 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
579 return 0;
580 p = time = line + len - strlen(" 07:01:32");
581
582 /* Permit 1-digit hours? */
583 if (*p++ != ' ' ||
584 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
585 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
586 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
587 return 0;
588
589 return line + len - time;
590 }
591
592 static size_t fractional_time_len(const char *line, size_t len)
593 {
594 const char *p;
595 size_t n;
596
597 /* Expected format: 19:41:17.620000023 */
598 if (!len || !isdigit(line[len - 1]))
599 return 0;
600 p = line + len - 1;
601
602 /* Fractional seconds. */
603 while (p > line && isdigit(*p))
604 p--;
605 if (*p != '.')
606 return 0;
607
608 /* Hours, minutes, and whole seconds. */
609 n = short_time_len(line, p - line);
610 if (!n)
611 return 0;
612
613 return line + len - p + n;
614 }
615
616 static size_t trailing_spaces_len(const char *line, size_t len)
617 {
618 const char *p;
619
620 /* Expected format: ' ' x (1 or more) */
621 if (!len || line[len - 1] != ' ')
622 return 0;
623
624 p = line + len;
625 while (p != line) {
626 p--;
627 if (*p != ' ')
628 return line + len - (p + 1);
629 }
630
631 /* All spaces! */
632 return len;
633 }
634
635 static size_t diff_timestamp_len(const char *line, size_t len)
636 {
637 const char *end = line + len;
638 size_t n;
639
640 /*
641 * Posix: 2010-07-05 19:41:17
642 * GNU: 2010-07-05 19:41:17.620000023 -0500
643 */
644
645 if (!isdigit(end[-1]))
646 return 0;
647
648 n = sane_tz_len(line, end - line);
649 if (!n)
650 n = tz_with_colon_len(line, end - line);
651 end -= n;
652
653 n = short_time_len(line, end - line);
654 if (!n)
655 n = fractional_time_len(line, end - line);
656 end -= n;
657
658 n = date_len(line, end - line);
659 if (!n) /* No date. Too bad. */
660 return 0;
661 end -= n;
662
663 if (end == line) /* No space before date. */
664 return 0;
665 if (end[-1] == '\t') { /* Success! */
666 end--;
667 return line + len - end;
668 }
669 if (end[-1] != ' ') /* No space before date. */
670 return 0;
671
672 /* Whitespace damage. */
673 end -= trailing_spaces_len(line, end - line);
674 return line + len - end;
675 }
676
677 static char *find_name_common(struct apply_state *state,
678 const char *line,
679 const char *def,
680 int p_value,
681 const char *end,
682 int terminate)
683 {
684 int len;
685 const char *start = NULL;
686
687 if (p_value == 0)
688 start = line;
689 while (line != end) {
690 char c = *line;
691
692 if (!end && isspace(c)) {
693 if (c == '\n')
694 break;
695 if (name_terminate(c, terminate))
696 break;
697 }
698 line++;
699 if (c == '/' && !--p_value)
700 start = line;
701 }
702 if (!start)
703 return squash_slash(xstrdup_or_null(def));
704 len = line - start;
705 if (!len)
706 return squash_slash(xstrdup_or_null(def));
707
708 /*
709 * Generally we prefer the shorter name, especially
710 * if the other one is just a variation of that with
711 * something else tacked on to the end (ie "file.orig"
712 * or "file~").
713 */
714 if (def) {
715 int deflen = strlen(def);
716 if (deflen < len && !strncmp(start, def, deflen))
717 return squash_slash(xstrdup(def));
718 }
719
720 if (state->root.len) {
721 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
722 return squash_slash(ret);
723 }
724
725 return squash_slash(xmemdupz(start, len));
726 }
727
728 static char *find_name(struct apply_state *state,
729 const char *line,
730 char *def,
731 int p_value,
732 int terminate)
733 {
734 if (*line == '"') {
735 char *name = find_name_gnu(state, line, def, p_value);
736 if (name)
737 return name;
738 }
739
740 return find_name_common(state, line, def, p_value, NULL, terminate);
741 }
742
743 static char *find_name_traditional(struct apply_state *state,
744 const char *line,
745 char *def,
746 int p_value)
747 {
748 size_t len;
749 size_t date_len;
750
751 if (*line == '"') {
752 char *name = find_name_gnu(state, line, def, p_value);
753 if (name)
754 return name;
755 }
756
757 len = strchrnul(line, '\n') - line;
758 date_len = diff_timestamp_len(line, len);
759 if (!date_len)
760 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
761 len -= date_len;
762
763 return find_name_common(state, line, def, p_value, line + len, 0);
764 }
765
766 static int count_slashes(const char *cp)
767 {
768 int cnt = 0;
769 char ch;
770
771 while ((ch = *cp++))
772 if (ch == '/')
773 cnt++;
774 return cnt;
775 }
776
777 /*
778 * Given the string after "--- " or "+++ ", guess the appropriate
779 * p_value for the given patch.
780 */
781 static int guess_p_value(struct apply_state *state, const char *nameline)
782 {
783 char *name, *cp;
784 int val = -1;
785
786 if (is_dev_null(nameline))
787 return -1;
788 name = find_name_traditional(state, nameline, NULL, 0);
789 if (!name)
790 return -1;
791 cp = strchr(name, '/');
792 if (!cp)
793 val = 0;
794 else if (state->prefix) {
795 /*
796 * Does it begin with "a/$our-prefix" and such? Then this is
797 * very likely to apply to our directory.
798 */
799 if (!strncmp(name, state->prefix, state->prefix_length))
800 val = count_slashes(state->prefix);
801 else {
802 cp++;
803 if (!strncmp(cp, state->prefix, state->prefix_length))
804 val = count_slashes(state->prefix) + 1;
805 }
806 }
807 free(name);
808 return val;
809 }
810
811 /*
812 * Does the ---/+++ line have the POSIX timestamp after the last HT?
813 * GNU diff puts epoch there to signal a creation/deletion event. Is
814 * this such a timestamp?
815 */
816 static int has_epoch_timestamp(const char *nameline)
817 {
818 /*
819 * We are only interested in epoch timestamp; any non-zero
820 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
821 * For the same reason, the date must be either 1969-12-31 or
822 * 1970-01-01, and the seconds part must be "00".
823 */
824 const char stamp_regexp[] =
825 "^(1969-12-31|1970-01-01)"
826 " "
827 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
828 " "
829 "([-+][0-2][0-9]:?[0-5][0-9])\n";
830 const char *timestamp = NULL, *cp, *colon;
831 static regex_t *stamp;
832 regmatch_t m[10];
833 int zoneoffset;
834 int hourminute;
835 int status;
836
837 for (cp = nameline; *cp != '\n'; cp++) {
838 if (*cp == '\t')
839 timestamp = cp + 1;
840 }
841 if (!timestamp)
842 return 0;
843 if (!stamp) {
844 stamp = xmalloc(sizeof(*stamp));
845 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
846 warning(_("Cannot prepare timestamp regexp %s"),
847 stamp_regexp);
848 return 0;
849 }
850 }
851
852 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
853 if (status) {
854 if (status != REG_NOMATCH)
855 warning(_("regexec returned %d for input: %s"),
856 status, timestamp);
857 return 0;
858 }
859
860 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
861 if (*colon == ':')
862 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
863 else
864 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
865 if (timestamp[m[3].rm_so] == '-')
866 zoneoffset = -zoneoffset;
867
868 /*
869 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
870 * (west of GMT) or 1970-01-01 (east of GMT)
871 */
872 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
873 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
874 return 0;
875
876 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
877 strtol(timestamp + 14, NULL, 10) -
878 zoneoffset);
879
880 return ((zoneoffset < 0 && hourminute == 1440) ||
881 (0 <= zoneoffset && !hourminute));
882 }
883
884 /*
885 * Get the name etc info from the ---/+++ lines of a traditional patch header
886 *
887 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
888 * files, we can happily check the index for a match, but for creating a
889 * new file we should try to match whatever "patch" does. I have no idea.
890 */
891 static int parse_traditional_patch(struct apply_state *state,
892 const char *first,
893 const char *second,
894 struct patch *patch)
895 {
896 char *name;
897
898 first += 4; /* skip "--- " */
899 second += 4; /* skip "+++ " */
900 if (!state->p_value_known) {
901 int p, q;
902 p = guess_p_value(state, first);
903 q = guess_p_value(state, second);
904 if (p < 0) p = q;
905 if (0 <= p && p == q) {
906 state->p_value = p;
907 state->p_value_known = 1;
908 }
909 }
910 if (is_dev_null(first)) {
911 patch->is_new = 1;
912 patch->is_delete = 0;
913 name = find_name_traditional(state, second, NULL, state->p_value);
914 patch->new_name = name;
915 } else if (is_dev_null(second)) {
916 patch->is_new = 0;
917 patch->is_delete = 1;
918 name = find_name_traditional(state, first, NULL, state->p_value);
919 patch->old_name = name;
920 } else {
921 char *first_name;
922 first_name = find_name_traditional(state, first, NULL, state->p_value);
923 name = find_name_traditional(state, second, first_name, state->p_value);
924 free(first_name);
925 if (has_epoch_timestamp(first)) {
926 patch->is_new = 1;
927 patch->is_delete = 0;
928 patch->new_name = name;
929 } else if (has_epoch_timestamp(second)) {
930 patch->is_new = 0;
931 patch->is_delete = 1;
932 patch->old_name = name;
933 } else {
934 patch->old_name = name;
935 patch->new_name = xstrdup_or_null(name);
936 }
937 }
938 if (!name)
939 return error(_("unable to find filename in patch at line %d"), state->linenr);
940
941 return 0;
942 }
943
944 static int gitdiff_hdrend(struct apply_state *state,
945 const char *line,
946 struct patch *patch)
947 {
948 return 1;
949 }
950
951 /*
952 * We're anal about diff header consistency, to make
953 * sure that we don't end up having strange ambiguous
954 * patches floating around.
955 *
956 * As a result, gitdiff_{old|new}name() will check
957 * their names against any previous information, just
958 * to make sure..
959 */
960 #define DIFF_OLD_NAME 0
961 #define DIFF_NEW_NAME 1
962
963 static int gitdiff_verify_name(struct apply_state *state,
964 const char *line,
965 int isnull,
966 char **name,
967 int side)
968 {
969 if (!*name && !isnull) {
970 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
971 return 0;
972 }
973
974 if (*name) {
975 int len = strlen(*name);
976 char *another;
977 if (isnull)
978 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
979 *name, state->linenr);
980 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
981 if (!another || memcmp(another, *name, len + 1)) {
982 free(another);
983 return error((side == DIFF_NEW_NAME) ?
984 _("git apply: bad git-diff - inconsistent new filename on line %d") :
985 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
986 }
987 free(another);
988 } else {
989 /* expect "/dev/null" */
990 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
991 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
992 }
993
994 return 0;
995 }
996
997 static int gitdiff_oldname(struct apply_state *state,
998 const char *line,
999 struct patch *patch)
1000 {
1001 return gitdiff_verify_name(state, line,
1002 patch->is_new, &patch->old_name,
1003 DIFF_OLD_NAME);
1004 }
1005
1006 static int gitdiff_newname(struct apply_state *state,
1007 const char *line,
1008 struct patch *patch)
1009 {
1010 return gitdiff_verify_name(state, line,
1011 patch->is_delete, &patch->new_name,
1012 DIFF_NEW_NAME);
1013 }
1014
1015 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
1016 {
1017 char *end;
1018 *mode = strtoul(line, &end, 8);
1019 if (end == line || !isspace(*end))
1020 return error(_("invalid mode on line %d: %s"), linenr, line);
1021 return 0;
1022 }
1023
1024 static int gitdiff_oldmode(struct apply_state *state,
1025 const char *line,
1026 struct patch *patch)
1027 {
1028 return parse_mode_line(line, state->linenr, &patch->old_mode);
1029 }
1030
1031 static int gitdiff_newmode(struct apply_state *state,
1032 const char *line,
1033 struct patch *patch)
1034 {
1035 return parse_mode_line(line, state->linenr, &patch->new_mode);
1036 }
1037
1038 static int gitdiff_delete(struct apply_state *state,
1039 const char *line,
1040 struct patch *patch)
1041 {
1042 patch->is_delete = 1;
1043 free(patch->old_name);
1044 patch->old_name = xstrdup_or_null(patch->def_name);
1045 return gitdiff_oldmode(state, line, patch);
1046 }
1047
1048 static int gitdiff_newfile(struct apply_state *state,
1049 const char *line,
1050 struct patch *patch)
1051 {
1052 patch->is_new = 1;
1053 free(patch->new_name);
1054 patch->new_name = xstrdup_or_null(patch->def_name);
1055 return gitdiff_newmode(state, line, patch);
1056 }
1057
1058 static int gitdiff_copysrc(struct apply_state *state,
1059 const char *line,
1060 struct patch *patch)
1061 {
1062 patch->is_copy = 1;
1063 free(patch->old_name);
1064 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1065 return 0;
1066 }
1067
1068 static int gitdiff_copydst(struct apply_state *state,
1069 const char *line,
1070 struct patch *patch)
1071 {
1072 patch->is_copy = 1;
1073 free(patch->new_name);
1074 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1075 return 0;
1076 }
1077
1078 static int gitdiff_renamesrc(struct apply_state *state,
1079 const char *line,
1080 struct patch *patch)
1081 {
1082 patch->is_rename = 1;
1083 free(patch->old_name);
1084 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1085 return 0;
1086 }
1087
1088 static int gitdiff_renamedst(struct apply_state *state,
1089 const char *line,
1090 struct patch *patch)
1091 {
1092 patch->is_rename = 1;
1093 free(patch->new_name);
1094 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1095 return 0;
1096 }
1097
1098 static int gitdiff_similarity(struct apply_state *state,
1099 const char *line,
1100 struct patch *patch)
1101 {
1102 unsigned long val = strtoul(line, NULL, 10);
1103 if (val <= 100)
1104 patch->score = val;
1105 return 0;
1106 }
1107
1108 static int gitdiff_dissimilarity(struct apply_state *state,
1109 const char *line,
1110 struct patch *patch)
1111 {
1112 unsigned long val = strtoul(line, NULL, 10);
1113 if (val <= 100)
1114 patch->score = val;
1115 return 0;
1116 }
1117
1118 static int gitdiff_index(struct apply_state *state,
1119 const char *line,
1120 struct patch *patch)
1121 {
1122 /*
1123 * index line is N hexadecimal, "..", N hexadecimal,
1124 * and optional space with octal mode.
1125 */
1126 const char *ptr, *eol;
1127 int len;
1128
1129 ptr = strchr(line, '.');
1130 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1131 return 0;
1132 len = ptr - line;
1133 memcpy(patch->old_sha1_prefix, line, len);
1134 patch->old_sha1_prefix[len] = 0;
1135
1136 line = ptr + 2;
1137 ptr = strchr(line, ' ');
1138 eol = strchrnul(line, '\n');
1139
1140 if (!ptr || eol < ptr)
1141 ptr = eol;
1142 len = ptr - line;
1143
1144 if (40 < len)
1145 return 0;
1146 memcpy(patch->new_sha1_prefix, line, len);
1147 patch->new_sha1_prefix[len] = 0;
1148 if (*ptr == ' ')
1149 return gitdiff_oldmode(state, ptr + 1, patch);
1150 return 0;
1151 }
1152
1153 /*
1154 * This is normal for a diff that doesn't change anything: we'll fall through
1155 * into the next diff. Tell the parser to break out.
1156 */
1157 static int gitdiff_unrecognized(struct apply_state *state,
1158 const char *line,
1159 struct patch *patch)
1160 {
1161 return 1;
1162 }
1163
1164 /*
1165 * Skip p_value leading components from "line"; as we do not accept
1166 * absolute paths, return NULL in that case.
1167 */
1168 static const char *skip_tree_prefix(struct apply_state *state,
1169 const char *line,
1170 int llen)
1171 {
1172 int nslash;
1173 int i;
1174
1175 if (!state->p_value)
1176 return (llen && line[0] == '/') ? NULL : line;
1177
1178 nslash = state->p_value;
1179 for (i = 0; i < llen; i++) {
1180 int ch = line[i];
1181 if (ch == '/' && --nslash <= 0)
1182 return (i == 0) ? NULL : &line[i + 1];
1183 }
1184 return NULL;
1185 }
1186
1187 /*
1188 * This is to extract the same name that appears on "diff --git"
1189 * line. We do not find and return anything if it is a rename
1190 * patch, and it is OK because we will find the name elsewhere.
1191 * We need to reliably find name only when it is mode-change only,
1192 * creation or deletion of an empty file. In any of these cases,
1193 * both sides are the same name under a/ and b/ respectively.
1194 */
1195 static char *git_header_name(struct apply_state *state,
1196 const char *line,
1197 int llen)
1198 {
1199 const char *name;
1200 const char *second = NULL;
1201 size_t len, line_len;
1202
1203 line += strlen("diff --git ");
1204 llen -= strlen("diff --git ");
1205
1206 if (*line == '"') {
1207 const char *cp;
1208 struct strbuf first = STRBUF_INIT;
1209 struct strbuf sp = STRBUF_INIT;
1210
1211 if (unquote_c_style(&first, line, &second))
1212 goto free_and_fail1;
1213
1214 /* strip the a/b prefix including trailing slash */
1215 cp = skip_tree_prefix(state, first.buf, first.len);
1216 if (!cp)
1217 goto free_and_fail1;
1218 strbuf_remove(&first, 0, cp - first.buf);
1219
1220 /*
1221 * second points at one past closing dq of name.
1222 * find the second name.
1223 */
1224 while ((second < line + llen) && isspace(*second))
1225 second++;
1226
1227 if (line + llen <= second)
1228 goto free_and_fail1;
1229 if (*second == '"') {
1230 if (unquote_c_style(&sp, second, NULL))
1231 goto free_and_fail1;
1232 cp = skip_tree_prefix(state, sp.buf, sp.len);
1233 if (!cp)
1234 goto free_and_fail1;
1235 /* They must match, otherwise ignore */
1236 if (strcmp(cp, first.buf))
1237 goto free_and_fail1;
1238 strbuf_release(&sp);
1239 return strbuf_detach(&first, NULL);
1240 }
1241
1242 /* unquoted second */
1243 cp = skip_tree_prefix(state, second, line + llen - second);
1244 if (!cp)
1245 goto free_and_fail1;
1246 if (line + llen - cp != first.len ||
1247 memcmp(first.buf, cp, first.len))
1248 goto free_and_fail1;
1249 return strbuf_detach(&first, NULL);
1250
1251 free_and_fail1:
1252 strbuf_release(&first);
1253 strbuf_release(&sp);
1254 return NULL;
1255 }
1256
1257 /* unquoted first name */
1258 name = skip_tree_prefix(state, line, llen);
1259 if (!name)
1260 return NULL;
1261
1262 /*
1263 * since the first name is unquoted, a dq if exists must be
1264 * the beginning of the second name.
1265 */
1266 for (second = name; second < line + llen; second++) {
1267 if (*second == '"') {
1268 struct strbuf sp = STRBUF_INIT;
1269 const char *np;
1270
1271 if (unquote_c_style(&sp, second, NULL))
1272 goto free_and_fail2;
1273
1274 np = skip_tree_prefix(state, sp.buf, sp.len);
1275 if (!np)
1276 goto free_and_fail2;
1277
1278 len = sp.buf + sp.len - np;
1279 if (len < second - name &&
1280 !strncmp(np, name, len) &&
1281 isspace(name[len])) {
1282 /* Good */
1283 strbuf_remove(&sp, 0, np - sp.buf);
1284 return strbuf_detach(&sp, NULL);
1285 }
1286
1287 free_and_fail2:
1288 strbuf_release(&sp);
1289 return NULL;
1290 }
1291 }
1292
1293 /*
1294 * Accept a name only if it shows up twice, exactly the same
1295 * form.
1296 */
1297 second = strchr(name, '\n');
1298 if (!second)
1299 return NULL;
1300 line_len = second - name;
1301 for (len = 0 ; ; len++) {
1302 switch (name[len]) {
1303 default:
1304 continue;
1305 case '\n':
1306 return NULL;
1307 case '\t': case ' ':
1308 /*
1309 * Is this the separator between the preimage
1310 * and the postimage pathname? Again, we are
1311 * only interested in the case where there is
1312 * no rename, as this is only to set def_name
1313 * and a rename patch has the names elsewhere
1314 * in an unambiguous form.
1315 */
1316 if (!name[len + 1])
1317 return NULL; /* no postimage name */
1318 second = skip_tree_prefix(state, name + len + 1,
1319 line_len - (len + 1));
1320 if (!second)
1321 return NULL;
1322 /*
1323 * Does len bytes starting at "name" and "second"
1324 * (that are separated by one HT or SP we just
1325 * found) exactly match?
1326 */
1327 if (second[len] == '\n' && !strncmp(name, second, len))
1328 return xmemdupz(name, len);
1329 }
1330 }
1331 }
1332
1333 static int check_header_line(struct apply_state *state, struct patch *patch)
1334 {
1335 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1336 (patch->is_rename == 1) + (patch->is_copy == 1);
1337 if (extensions > 1)
1338 return error(_("inconsistent header lines %d and %d"),
1339 patch->extension_linenr, state->linenr);
1340 if (extensions && !patch->extension_linenr)
1341 patch->extension_linenr = state->linenr;
1342 return 0;
1343 }
1344
1345 /* Verify that we recognize the lines following a git header */
1346 static int parse_git_header(struct apply_state *state,
1347 const char *line,
1348 int len,
1349 unsigned int size,
1350 struct patch *patch)
1351 {
1352 unsigned long offset;
1353
1354 /* A git diff has explicit new/delete information, so we don't guess */
1355 patch->is_new = 0;
1356 patch->is_delete = 0;
1357
1358 /*
1359 * Some things may not have the old name in the
1360 * rest of the headers anywhere (pure mode changes,
1361 * or removing or adding empty files), so we get
1362 * the default name from the header.
1363 */
1364 patch->def_name = git_header_name(state, line, len);
1365 if (patch->def_name && state->root.len) {
1366 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1367 free(patch->def_name);
1368 patch->def_name = s;
1369 }
1370
1371 line += len;
1372 size -= len;
1373 state->linenr++;
1374 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1375 static const struct opentry {
1376 const char *str;
1377 int (*fn)(struct apply_state *, const char *, struct patch *);
1378 } optable[] = {
1379 { "@@ -", gitdiff_hdrend },
1380 { "--- ", gitdiff_oldname },
1381 { "+++ ", gitdiff_newname },
1382 { "old mode ", gitdiff_oldmode },
1383 { "new mode ", gitdiff_newmode },
1384 { "deleted file mode ", gitdiff_delete },
1385 { "new file mode ", gitdiff_newfile },
1386 { "copy from ", gitdiff_copysrc },
1387 { "copy to ", gitdiff_copydst },
1388 { "rename old ", gitdiff_renamesrc },
1389 { "rename new ", gitdiff_renamedst },
1390 { "rename from ", gitdiff_renamesrc },
1391 { "rename to ", gitdiff_renamedst },
1392 { "similarity index ", gitdiff_similarity },
1393 { "dissimilarity index ", gitdiff_dissimilarity },
1394 { "index ", gitdiff_index },
1395 { "", gitdiff_unrecognized },
1396 };
1397 int i;
1398
1399 len = linelen(line, size);
1400 if (!len || line[len-1] != '\n')
1401 break;
1402 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1403 const struct opentry *p = optable + i;
1404 int oplen = strlen(p->str);
1405 int res;
1406 if (len < oplen || memcmp(p->str, line, oplen))
1407 continue;
1408 res = p->fn(state, line + oplen, patch);
1409 if (res < 0)
1410 return -1;
1411 if (check_header_line(state, patch))
1412 return -1;
1413 if (res > 0)
1414 return offset;
1415 break;
1416 }
1417 }
1418
1419 return offset;
1420 }
1421
1422 static int parse_num(const char *line, unsigned long *p)
1423 {
1424 char *ptr;
1425
1426 if (!isdigit(*line))
1427 return 0;
1428 *p = strtoul(line, &ptr, 10);
1429 return ptr - line;
1430 }
1431
1432 static int parse_range(const char *line, int len, int offset, const char *expect,
1433 unsigned long *p1, unsigned long *p2)
1434 {
1435 int digits, ex;
1436
1437 if (offset < 0 || offset >= len)
1438 return -1;
1439 line += offset;
1440 len -= offset;
1441
1442 digits = parse_num(line, p1);
1443 if (!digits)
1444 return -1;
1445
1446 offset += digits;
1447 line += digits;
1448 len -= digits;
1449
1450 *p2 = 1;
1451 if (*line == ',') {
1452 digits = parse_num(line+1, p2);
1453 if (!digits)
1454 return -1;
1455
1456 offset += digits+1;
1457 line += digits+1;
1458 len -= digits+1;
1459 }
1460
1461 ex = strlen(expect);
1462 if (ex > len)
1463 return -1;
1464 if (memcmp(line, expect, ex))
1465 return -1;
1466
1467 return offset + ex;
1468 }
1469
1470 static void recount_diff(const char *line, int size, struct fragment *fragment)
1471 {
1472 int oldlines = 0, newlines = 0, ret = 0;
1473
1474 if (size < 1) {
1475 warning("recount: ignore empty hunk");
1476 return;
1477 }
1478
1479 for (;;) {
1480 int len = linelen(line, size);
1481 size -= len;
1482 line += len;
1483
1484 if (size < 1)
1485 break;
1486
1487 switch (*line) {
1488 case ' ': case '\n':
1489 newlines++;
1490 /* fall through */
1491 case '-':
1492 oldlines++;
1493 continue;
1494 case '+':
1495 newlines++;
1496 continue;
1497 case '\\':
1498 continue;
1499 case '@':
1500 ret = size < 3 || !starts_with(line, "@@ ");
1501 break;
1502 case 'd':
1503 ret = size < 5 || !starts_with(line, "diff ");
1504 break;
1505 default:
1506 ret = -1;
1507 break;
1508 }
1509 if (ret) {
1510 warning(_("recount: unexpected line: %.*s"),
1511 (int)linelen(line, size), line);
1512 return;
1513 }
1514 break;
1515 }
1516 fragment->oldlines = oldlines;
1517 fragment->newlines = newlines;
1518 }
1519
1520 /*
1521 * Parse a unified diff fragment header of the
1522 * form "@@ -a,b +c,d @@"
1523 */
1524 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1525 {
1526 int offset;
1527
1528 if (!len || line[len-1] != '\n')
1529 return -1;
1530
1531 /* Figure out the number of lines in a fragment */
1532 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1533 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1534
1535 return offset;
1536 }
1537
1538 /*
1539 * Find file diff header
1540 *
1541 * Returns:
1542 * -1 if no header was found
1543 * -128 in case of error
1544 * the size of the header in bytes (called "offset") otherwise
1545 */
1546 static int find_header(struct apply_state *state,
1547 const char *line,
1548 unsigned long size,
1549 int *hdrsize,
1550 struct patch *patch)
1551 {
1552 unsigned long offset, len;
1553
1554 patch->is_toplevel_relative = 0;
1555 patch->is_rename = patch->is_copy = 0;
1556 patch->is_new = patch->is_delete = -1;
1557 patch->old_mode = patch->new_mode = 0;
1558 patch->old_name = patch->new_name = NULL;
1559 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1560 unsigned long nextlen;
1561
1562 len = linelen(line, size);
1563 if (!len)
1564 break;
1565
1566 /* Testing this early allows us to take a few shortcuts.. */
1567 if (len < 6)
1568 continue;
1569
1570 /*
1571 * Make sure we don't find any unconnected patch fragments.
1572 * That's a sign that we didn't find a header, and that a
1573 * patch has become corrupted/broken up.
1574 */
1575 if (!memcmp("@@ -", line, 4)) {
1576 struct fragment dummy;
1577 if (parse_fragment_header(line, len, &dummy) < 0)
1578 continue;
1579 error(_("patch fragment without header at line %d: %.*s"),
1580 state->linenr, (int)len-1, line);
1581 return -128;
1582 }
1583
1584 if (size < len + 6)
1585 break;
1586
1587 /*
1588 * Git patch? It might not have a real patch, just a rename
1589 * or mode change, so we handle that specially
1590 */
1591 if (!memcmp("diff --git ", line, 11)) {
1592 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1593 if (git_hdr_len < 0)
1594 return -128;
1595 if (git_hdr_len <= len)
1596 continue;
1597 if (!patch->old_name && !patch->new_name) {
1598 if (!patch->def_name) {
1599 error(Q_("git diff header lacks filename information when removing "
1600 "%d leading pathname component (line %d)",
1601 "git diff header lacks filename information when removing "
1602 "%d leading pathname components (line %d)",
1603 state->p_value),
1604 state->p_value, state->linenr);
1605 return -128;
1606 }
1607 patch->old_name = xstrdup(patch->def_name);
1608 patch->new_name = xstrdup(patch->def_name);
1609 }
1610 if ((!patch->new_name && !patch->is_delete) ||
1611 (!patch->old_name && !patch->is_new)) {
1612 error(_("git diff header lacks filename information "
1613 "(line %d)"), state->linenr);
1614 return -128;
1615 }
1616 patch->is_toplevel_relative = 1;
1617 *hdrsize = git_hdr_len;
1618 return offset;
1619 }
1620
1621 /* --- followed by +++ ? */
1622 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1623 continue;
1624
1625 /*
1626 * We only accept unified patches, so we want it to
1627 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1628 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1629 */
1630 nextlen = linelen(line + len, size - len);
1631 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1632 continue;
1633
1634 /* Ok, we'll consider it a patch */
1635 if (parse_traditional_patch(state, line, line+len, patch))
1636 return -128;
1637 *hdrsize = len + nextlen;
1638 state->linenr += 2;
1639 return offset;
1640 }
1641 return -1;
1642 }
1643
1644 static void record_ws_error(struct apply_state *state,
1645 unsigned result,
1646 const char *line,
1647 int len,
1648 int linenr)
1649 {
1650 char *err;
1651
1652 if (!result)
1653 return;
1654
1655 state->whitespace_error++;
1656 if (state->squelch_whitespace_errors &&
1657 state->squelch_whitespace_errors < state->whitespace_error)
1658 return;
1659
1660 err = whitespace_error_string(result);
1661 if (state->apply_verbosity > verbosity_silent)
1662 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1663 state->patch_input_file, linenr, err, len, line);
1664 free(err);
1665 }
1666
1667 static void check_whitespace(struct apply_state *state,
1668 const char *line,
1669 int len,
1670 unsigned ws_rule)
1671 {
1672 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1673
1674 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1675 }
1676
1677 /*
1678 * Parse a unified diff. Note that this really needs to parse each
1679 * fragment separately, since the only way to know the difference
1680 * between a "---" that is part of a patch, and a "---" that starts
1681 * the next patch is to look at the line counts..
1682 */
1683 static int parse_fragment(struct apply_state *state,
1684 const char *line,
1685 unsigned long size,
1686 struct patch *patch,
1687 struct fragment *fragment)
1688 {
1689 int added, deleted;
1690 int len = linelen(line, size), offset;
1691 unsigned long oldlines, newlines;
1692 unsigned long leading, trailing;
1693
1694 offset = parse_fragment_header(line, len, fragment);
1695 if (offset < 0)
1696 return -1;
1697 if (offset > 0 && patch->recount)
1698 recount_diff(line + offset, size - offset, fragment);
1699 oldlines = fragment->oldlines;
1700 newlines = fragment->newlines;
1701 leading = 0;
1702 trailing = 0;
1703
1704 /* Parse the thing.. */
1705 line += len;
1706 size -= len;
1707 state->linenr++;
1708 added = deleted = 0;
1709 for (offset = len;
1710 0 < size;
1711 offset += len, size -= len, line += len, state->linenr++) {
1712 if (!oldlines && !newlines)
1713 break;
1714 len = linelen(line, size);
1715 if (!len || line[len-1] != '\n')
1716 return -1;
1717 switch (*line) {
1718 default:
1719 return -1;
1720 case '\n': /* newer GNU diff, an empty context line */
1721 case ' ':
1722 oldlines--;
1723 newlines--;
1724 if (!deleted && !added)
1725 leading++;
1726 trailing++;
1727 if (!state->apply_in_reverse &&
1728 state->ws_error_action == correct_ws_error)
1729 check_whitespace(state, line, len, patch->ws_rule);
1730 break;
1731 case '-':
1732 if (state->apply_in_reverse &&
1733 state->ws_error_action != nowarn_ws_error)
1734 check_whitespace(state, line, len, patch->ws_rule);
1735 deleted++;
1736 oldlines--;
1737 trailing = 0;
1738 break;
1739 case '+':
1740 if (!state->apply_in_reverse &&
1741 state->ws_error_action != nowarn_ws_error)
1742 check_whitespace(state, line, len, patch->ws_rule);
1743 added++;
1744 newlines--;
1745 trailing = 0;
1746 break;
1747
1748 /*
1749 * We allow "\ No newline at end of file". Depending
1750 * on locale settings when the patch was produced we
1751 * don't know what this line looks like. The only
1752 * thing we do know is that it begins with "\ ".
1753 * Checking for 12 is just for sanity check -- any
1754 * l10n of "\ No newline..." is at least that long.
1755 */
1756 case '\\':
1757 if (len < 12 || memcmp(line, "\\ ", 2))
1758 return -1;
1759 break;
1760 }
1761 }
1762 if (oldlines || newlines)
1763 return -1;
1764 if (!deleted && !added)
1765 return -1;
1766
1767 fragment->leading = leading;
1768 fragment->trailing = trailing;
1769
1770 /*
1771 * If a fragment ends with an incomplete line, we failed to include
1772 * it in the above loop because we hit oldlines == newlines == 0
1773 * before seeing it.
1774 */
1775 if (12 < size && !memcmp(line, "\\ ", 2))
1776 offset += linelen(line, size);
1777
1778 patch->lines_added += added;
1779 patch->lines_deleted += deleted;
1780
1781 if (0 < patch->is_new && oldlines)
1782 return error(_("new file depends on old contents"));
1783 if (0 < patch->is_delete && newlines)
1784 return error(_("deleted file still has contents"));
1785 return offset;
1786 }
1787
1788 /*
1789 * We have seen "diff --git a/... b/..." header (or a traditional patch
1790 * header). Read hunks that belong to this patch into fragments and hang
1791 * them to the given patch structure.
1792 *
1793 * The (fragment->patch, fragment->size) pair points into the memory given
1794 * by the caller, not a copy, when we return.
1795 *
1796 * Returns:
1797 * -1 in case of error,
1798 * the number of bytes in the patch otherwise.
1799 */
1800 static int parse_single_patch(struct apply_state *state,
1801 const char *line,
1802 unsigned long size,
1803 struct patch *patch)
1804 {
1805 unsigned long offset = 0;
1806 unsigned long oldlines = 0, newlines = 0, context = 0;
1807 struct fragment **fragp = &patch->fragments;
1808
1809 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1810 struct fragment *fragment;
1811 int len;
1812
1813 fragment = xcalloc(1, sizeof(*fragment));
1814 fragment->linenr = state->linenr;
1815 len = parse_fragment(state, line, size, patch, fragment);
1816 if (len <= 0) {
1817 free(fragment);
1818 return error(_("corrupt patch at line %d"), state->linenr);
1819 }
1820 fragment->patch = line;
1821 fragment->size = len;
1822 oldlines += fragment->oldlines;
1823 newlines += fragment->newlines;
1824 context += fragment->leading + fragment->trailing;
1825
1826 *fragp = fragment;
1827 fragp = &fragment->next;
1828
1829 offset += len;
1830 line += len;
1831 size -= len;
1832 }
1833
1834 /*
1835 * If something was removed (i.e. we have old-lines) it cannot
1836 * be creation, and if something was added it cannot be
1837 * deletion. However, the reverse is not true; --unified=0
1838 * patches that only add are not necessarily creation even
1839 * though they do not have any old lines, and ones that only
1840 * delete are not necessarily deletion.
1841 *
1842 * Unfortunately, a real creation/deletion patch do _not_ have
1843 * any context line by definition, so we cannot safely tell it
1844 * apart with --unified=0 insanity. At least if the patch has
1845 * more than one hunk it is not creation or deletion.
1846 */
1847 if (patch->is_new < 0 &&
1848 (oldlines || (patch->fragments && patch->fragments->next)))
1849 patch->is_new = 0;
1850 if (patch->is_delete < 0 &&
1851 (newlines || (patch->fragments && patch->fragments->next)))
1852 patch->is_delete = 0;
1853
1854 if (0 < patch->is_new && oldlines)
1855 return error(_("new file %s depends on old contents"), patch->new_name);
1856 if (0 < patch->is_delete && newlines)
1857 return error(_("deleted file %s still has contents"), patch->old_name);
1858 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1859 fprintf_ln(stderr,
1860 _("** warning: "
1861 "file %s becomes empty but is not deleted"),
1862 patch->new_name);
1863
1864 return offset;
1865 }
1866
1867 static inline int metadata_changes(struct patch *patch)
1868 {
1869 return patch->is_rename > 0 ||
1870 patch->is_copy > 0 ||
1871 patch->is_new > 0 ||
1872 patch->is_delete ||
1873 (patch->old_mode && patch->new_mode &&
1874 patch->old_mode != patch->new_mode);
1875 }
1876
1877 static char *inflate_it(const void *data, unsigned long size,
1878 unsigned long inflated_size)
1879 {
1880 git_zstream stream;
1881 void *out;
1882 int st;
1883
1884 memset(&stream, 0, sizeof(stream));
1885
1886 stream.next_in = (unsigned char *)data;
1887 stream.avail_in = size;
1888 stream.next_out = out = xmalloc(inflated_size);
1889 stream.avail_out = inflated_size;
1890 git_inflate_init(&stream);
1891 st = git_inflate(&stream, Z_FINISH);
1892 git_inflate_end(&stream);
1893 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1894 free(out);
1895 return NULL;
1896 }
1897 return out;
1898 }
1899
1900 /*
1901 * Read a binary hunk and return a new fragment; fragment->patch
1902 * points at an allocated memory that the caller must free, so
1903 * it is marked as "->free_patch = 1".
1904 */
1905 static struct fragment *parse_binary_hunk(struct apply_state *state,
1906 char **buf_p,
1907 unsigned long *sz_p,
1908 int *status_p,
1909 int *used_p)
1910 {
1911 /*
1912 * Expect a line that begins with binary patch method ("literal"
1913 * or "delta"), followed by the length of data before deflating.
1914 * a sequence of 'length-byte' followed by base-85 encoded data
1915 * should follow, terminated by a newline.
1916 *
1917 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1918 * and we would limit the patch line to 66 characters,
1919 * so one line can fit up to 13 groups that would decode
1920 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1921 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1922 */
1923 int llen, used;
1924 unsigned long size = *sz_p;
1925 char *buffer = *buf_p;
1926 int patch_method;
1927 unsigned long origlen;
1928 char *data = NULL;
1929 int hunk_size = 0;
1930 struct fragment *frag;
1931
1932 llen = linelen(buffer, size);
1933 used = llen;
1934
1935 *status_p = 0;
1936
1937 if (starts_with(buffer, "delta ")) {
1938 patch_method = BINARY_DELTA_DEFLATED;
1939 origlen = strtoul(buffer + 6, NULL, 10);
1940 }
1941 else if (starts_with(buffer, "literal ")) {
1942 patch_method = BINARY_LITERAL_DEFLATED;
1943 origlen = strtoul(buffer + 8, NULL, 10);
1944 }
1945 else
1946 return NULL;
1947
1948 state->linenr++;
1949 buffer += llen;
1950 while (1) {
1951 int byte_length, max_byte_length, newsize;
1952 llen = linelen(buffer, size);
1953 used += llen;
1954 state->linenr++;
1955 if (llen == 1) {
1956 /* consume the blank line */
1957 buffer++;
1958 size--;
1959 break;
1960 }
1961 /*
1962 * Minimum line is "A00000\n" which is 7-byte long,
1963 * and the line length must be multiple of 5 plus 2.
1964 */
1965 if ((llen < 7) || (llen-2) % 5)
1966 goto corrupt;
1967 max_byte_length = (llen - 2) / 5 * 4;
1968 byte_length = *buffer;
1969 if ('A' <= byte_length && byte_length <= 'Z')
1970 byte_length = byte_length - 'A' + 1;
1971 else if ('a' <= byte_length && byte_length <= 'z')
1972 byte_length = byte_length - 'a' + 27;
1973 else
1974 goto corrupt;
1975 /* if the input length was not multiple of 4, we would
1976 * have filler at the end but the filler should never
1977 * exceed 3 bytes
1978 */
1979 if (max_byte_length < byte_length ||
1980 byte_length <= max_byte_length - 4)
1981 goto corrupt;
1982 newsize = hunk_size + byte_length;
1983 data = xrealloc(data, newsize);
1984 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1985 goto corrupt;
1986 hunk_size = newsize;
1987 buffer += llen;
1988 size -= llen;
1989 }
1990
1991 frag = xcalloc(1, sizeof(*frag));
1992 frag->patch = inflate_it(data, hunk_size, origlen);
1993 frag->free_patch = 1;
1994 if (!frag->patch)
1995 goto corrupt;
1996 free(data);
1997 frag->size = origlen;
1998 *buf_p = buffer;
1999 *sz_p = size;
2000 *used_p = used;
2001 frag->binary_patch_method = patch_method;
2002 return frag;
2003
2004 corrupt:
2005 free(data);
2006 *status_p = -1;
2007 error(_("corrupt binary patch at line %d: %.*s"),
2008 state->linenr-1, llen-1, buffer);
2009 return NULL;
2010 }
2011
2012 /*
2013 * Returns:
2014 * -1 in case of error,
2015 * the length of the parsed binary patch otherwise
2016 */
2017 static int parse_binary(struct apply_state *state,
2018 char *buffer,
2019 unsigned long size,
2020 struct patch *patch)
2021 {
2022 /*
2023 * We have read "GIT binary patch\n"; what follows is a line
2024 * that says the patch method (currently, either "literal" or
2025 * "delta") and the length of data before deflating; a
2026 * sequence of 'length-byte' followed by base-85 encoded data
2027 * follows.
2028 *
2029 * When a binary patch is reversible, there is another binary
2030 * hunk in the same format, starting with patch method (either
2031 * "literal" or "delta") with the length of data, and a sequence
2032 * of length-byte + base-85 encoded data, terminated with another
2033 * empty line. This data, when applied to the postimage, produces
2034 * the preimage.
2035 */
2036 struct fragment *forward;
2037 struct fragment *reverse;
2038 int status;
2039 int used, used_1;
2040
2041 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2042 if (!forward && !status)
2043 /* there has to be one hunk (forward hunk) */
2044 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2045 if (status)
2046 /* otherwise we already gave an error message */
2047 return status;
2048
2049 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2050 if (reverse)
2051 used += used_1;
2052 else if (status) {
2053 /*
2054 * Not having reverse hunk is not an error, but having
2055 * a corrupt reverse hunk is.
2056 */
2057 free((void*) forward->patch);
2058 free(forward);
2059 return status;
2060 }
2061 forward->next = reverse;
2062 patch->fragments = forward;
2063 patch->is_binary = 1;
2064 return used;
2065 }
2066
2067 static void prefix_one(struct apply_state *state, char **name)
2068 {
2069 char *old_name = *name;
2070 if (!old_name)
2071 return;
2072 *name = prefix_filename(state->prefix, *name);
2073 free(old_name);
2074 }
2075
2076 static void prefix_patch(struct apply_state *state, struct patch *p)
2077 {
2078 if (!state->prefix || p->is_toplevel_relative)
2079 return;
2080 prefix_one(state, &p->new_name);
2081 prefix_one(state, &p->old_name);
2082 }
2083
2084 /*
2085 * include/exclude
2086 */
2087
2088 static void add_name_limit(struct apply_state *state,
2089 const char *name,
2090 int exclude)
2091 {
2092 struct string_list_item *it;
2093
2094 it = string_list_append(&state->limit_by_name, name);
2095 it->util = exclude ? NULL : (void *) 1;
2096 }
2097
2098 static int use_patch(struct apply_state *state, struct patch *p)
2099 {
2100 const char *pathname = p->new_name ? p->new_name : p->old_name;
2101 int i;
2102
2103 /* Paths outside are not touched regardless of "--include" */
2104 if (0 < state->prefix_length) {
2105 int pathlen = strlen(pathname);
2106 if (pathlen <= state->prefix_length ||
2107 memcmp(state->prefix, pathname, state->prefix_length))
2108 return 0;
2109 }
2110
2111 /* See if it matches any of exclude/include rule */
2112 for (i = 0; i < state->limit_by_name.nr; i++) {
2113 struct string_list_item *it = &state->limit_by_name.items[i];
2114 if (!wildmatch(it->string, pathname, 0, NULL))
2115 return (it->util != NULL);
2116 }
2117
2118 /*
2119 * If we had any include, a path that does not match any rule is
2120 * not used. Otherwise, we saw bunch of exclude rules (or none)
2121 * and such a path is used.
2122 */
2123 return !state->has_include;
2124 }
2125
2126 /*
2127 * Read the patch text in "buffer" that extends for "size" bytes; stop
2128 * reading after seeing a single patch (i.e. changes to a single file).
2129 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2130 *
2131 * Returns:
2132 * -1 if no header was found or parse_binary() failed,
2133 * -128 on another error,
2134 * the number of bytes consumed otherwise,
2135 * so that the caller can call us again for the next patch.
2136 */
2137 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2138 {
2139 int hdrsize, patchsize;
2140 int offset = find_header(state, buffer, size, &hdrsize, patch);
2141
2142 if (offset < 0)
2143 return offset;
2144
2145 prefix_patch(state, patch);
2146
2147 if (!use_patch(state, patch))
2148 patch->ws_rule = 0;
2149 else
2150 patch->ws_rule = whitespace_rule(patch->new_name
2151 ? patch->new_name
2152 : patch->old_name);
2153
2154 patchsize = parse_single_patch(state,
2155 buffer + offset + hdrsize,
2156 size - offset - hdrsize,
2157 patch);
2158
2159 if (patchsize < 0)
2160 return -128;
2161
2162 if (!patchsize) {
2163 static const char git_binary[] = "GIT binary patch\n";
2164 int hd = hdrsize + offset;
2165 unsigned long llen = linelen(buffer + hd, size - hd);
2166
2167 if (llen == sizeof(git_binary) - 1 &&
2168 !memcmp(git_binary, buffer + hd, llen)) {
2169 int used;
2170 state->linenr++;
2171 used = parse_binary(state, buffer + hd + llen,
2172 size - hd - llen, patch);
2173 if (used < 0)
2174 return -1;
2175 if (used)
2176 patchsize = used + llen;
2177 else
2178 patchsize = 0;
2179 }
2180 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2181 static const char *binhdr[] = {
2182 "Binary files ",
2183 "Files ",
2184 NULL,
2185 };
2186 int i;
2187 for (i = 0; binhdr[i]; i++) {
2188 int len = strlen(binhdr[i]);
2189 if (len < size - hd &&
2190 !memcmp(binhdr[i], buffer + hd, len)) {
2191 state->linenr++;
2192 patch->is_binary = 1;
2193 patchsize = llen;
2194 break;
2195 }
2196 }
2197 }
2198
2199 /* Empty patch cannot be applied if it is a text patch
2200 * without metadata change. A binary patch appears
2201 * empty to us here.
2202 */
2203 if ((state->apply || state->check) &&
2204 (!patch->is_binary && !metadata_changes(patch))) {
2205 error(_("patch with only garbage at line %d"), state->linenr);
2206 return -128;
2207 }
2208 }
2209
2210 return offset + hdrsize + patchsize;
2211 }
2212
2213 static void reverse_patches(struct patch *p)
2214 {
2215 for (; p; p = p->next) {
2216 struct fragment *frag = p->fragments;
2217
2218 SWAP(p->new_name, p->old_name);
2219 SWAP(p->new_mode, p->old_mode);
2220 SWAP(p->is_new, p->is_delete);
2221 SWAP(p->lines_added, p->lines_deleted);
2222 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2223
2224 for (; frag; frag = frag->next) {
2225 SWAP(frag->newpos, frag->oldpos);
2226 SWAP(frag->newlines, frag->oldlines);
2227 }
2228 }
2229 }
2230
2231 static const char pluses[] =
2232 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2233 static const char minuses[]=
2234 "----------------------------------------------------------------------";
2235
2236 static void show_stats(struct apply_state *state, struct patch *patch)
2237 {
2238 struct strbuf qname = STRBUF_INIT;
2239 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2240 int max, add, del;
2241
2242 quote_c_style(cp, &qname, NULL, 0);
2243
2244 /*
2245 * "scale" the filename
2246 */
2247 max = state->max_len;
2248 if (max > 50)
2249 max = 50;
2250
2251 if (qname.len > max) {
2252 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2253 if (!cp)
2254 cp = qname.buf + qname.len + 3 - max;
2255 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2256 }
2257
2258 if (patch->is_binary) {
2259 printf(" %-*s | Bin\n", max, qname.buf);
2260 strbuf_release(&qname);
2261 return;
2262 }
2263
2264 printf(" %-*s |", max, qname.buf);
2265 strbuf_release(&qname);
2266
2267 /*
2268 * scale the add/delete
2269 */
2270 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2271 add = patch->lines_added;
2272 del = patch->lines_deleted;
2273
2274 if (state->max_change > 0) {
2275 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2276 add = (add * max + state->max_change / 2) / state->max_change;
2277 del = total - add;
2278 }
2279 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2280 add, pluses, del, minuses);
2281 }
2282
2283 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2284 {
2285 switch (st->st_mode & S_IFMT) {
2286 case S_IFLNK:
2287 if (strbuf_readlink(buf, path, st->st_size) < 0)
2288 return error(_("unable to read symlink %s"), path);
2289 return 0;
2290 case S_IFREG:
2291 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2292 return error(_("unable to open or read %s"), path);
2293 convert_to_git(path, buf->buf, buf->len, buf, 0);
2294 return 0;
2295 default:
2296 return -1;
2297 }
2298 }
2299
2300 /*
2301 * Update the preimage, and the common lines in postimage,
2302 * from buffer buf of length len. If postlen is 0 the postimage
2303 * is updated in place, otherwise it's updated on a new buffer
2304 * of length postlen
2305 */
2306
2307 static void update_pre_post_images(struct image *preimage,
2308 struct image *postimage,
2309 char *buf,
2310 size_t len, size_t postlen)
2311 {
2312 int i, ctx, reduced;
2313 char *new, *old, *fixed;
2314 struct image fixed_preimage;
2315
2316 /*
2317 * Update the preimage with whitespace fixes. Note that we
2318 * are not losing preimage->buf -- apply_one_fragment() will
2319 * free "oldlines".
2320 */
2321 prepare_image(&fixed_preimage, buf, len, 1);
2322 assert(postlen
2323 ? fixed_preimage.nr == preimage->nr
2324 : fixed_preimage.nr <= preimage->nr);
2325 for (i = 0; i < fixed_preimage.nr; i++)
2326 fixed_preimage.line[i].flag = preimage->line[i].flag;
2327 free(preimage->line_allocated);
2328 *preimage = fixed_preimage;
2329
2330 /*
2331 * Adjust the common context lines in postimage. This can be
2332 * done in-place when we are shrinking it with whitespace
2333 * fixing, but needs a new buffer when ignoring whitespace or
2334 * expanding leading tabs to spaces.
2335 *
2336 * We trust the caller to tell us if the update can be done
2337 * in place (postlen==0) or not.
2338 */
2339 old = postimage->buf;
2340 if (postlen)
2341 new = postimage->buf = xmalloc(postlen);
2342 else
2343 new = old;
2344 fixed = preimage->buf;
2345
2346 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2347 size_t l_len = postimage->line[i].len;
2348 if (!(postimage->line[i].flag & LINE_COMMON)) {
2349 /* an added line -- no counterparts in preimage */
2350 memmove(new, old, l_len);
2351 old += l_len;
2352 new += l_len;
2353 continue;
2354 }
2355
2356 /* a common context -- skip it in the original postimage */
2357 old += l_len;
2358
2359 /* and find the corresponding one in the fixed preimage */
2360 while (ctx < preimage->nr &&
2361 !(preimage->line[ctx].flag & LINE_COMMON)) {
2362 fixed += preimage->line[ctx].len;
2363 ctx++;
2364 }
2365
2366 /*
2367 * preimage is expected to run out, if the caller
2368 * fixed addition of trailing blank lines.
2369 */
2370 if (preimage->nr <= ctx) {
2371 reduced++;
2372 continue;
2373 }
2374
2375 /* and copy it in, while fixing the line length */
2376 l_len = preimage->line[ctx].len;
2377 memcpy(new, fixed, l_len);
2378 new += l_len;
2379 fixed += l_len;
2380 postimage->line[i].len = l_len;
2381 ctx++;
2382 }
2383
2384 if (postlen
2385 ? postlen < new - postimage->buf
2386 : postimage->len < new - postimage->buf)
2387 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2388 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2389
2390 /* Fix the length of the whole thing */
2391 postimage->len = new - postimage->buf;
2392 postimage->nr -= reduced;
2393 }
2394
2395 static int line_by_line_fuzzy_match(struct image *img,
2396 struct image *preimage,
2397 struct image *postimage,
2398 unsigned long try,
2399 int try_lno,
2400 int preimage_limit)
2401 {
2402 int i;
2403 size_t imgoff = 0;
2404 size_t preoff = 0;
2405 size_t postlen = postimage->len;
2406 size_t extra_chars;
2407 char *buf;
2408 char *preimage_eof;
2409 char *preimage_end;
2410 struct strbuf fixed;
2411 char *fixed_buf;
2412 size_t fixed_len;
2413
2414 for (i = 0; i < preimage_limit; i++) {
2415 size_t prelen = preimage->line[i].len;
2416 size_t imglen = img->line[try_lno+i].len;
2417
2418 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2419 preimage->buf + preoff, prelen))
2420 return 0;
2421 if (preimage->line[i].flag & LINE_COMMON)
2422 postlen += imglen - prelen;
2423 imgoff += imglen;
2424 preoff += prelen;
2425 }
2426
2427 /*
2428 * Ok, the preimage matches with whitespace fuzz.
2429 *
2430 * imgoff now holds the true length of the target that
2431 * matches the preimage before the end of the file.
2432 *
2433 * Count the number of characters in the preimage that fall
2434 * beyond the end of the file and make sure that all of them
2435 * are whitespace characters. (This can only happen if
2436 * we are removing blank lines at the end of the file.)
2437 */
2438 buf = preimage_eof = preimage->buf + preoff;
2439 for ( ; i < preimage->nr; i++)
2440 preoff += preimage->line[i].len;
2441 preimage_end = preimage->buf + preoff;
2442 for ( ; buf < preimage_end; buf++)
2443 if (!isspace(*buf))
2444 return 0;
2445
2446 /*
2447 * Update the preimage and the common postimage context
2448 * lines to use the same whitespace as the target.
2449 * If whitespace is missing in the target (i.e.
2450 * if the preimage extends beyond the end of the file),
2451 * use the whitespace from the preimage.
2452 */
2453 extra_chars = preimage_end - preimage_eof;
2454 strbuf_init(&fixed, imgoff + extra_chars);
2455 strbuf_add(&fixed, img->buf + try, imgoff);
2456 strbuf_add(&fixed, preimage_eof, extra_chars);
2457 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2458 update_pre_post_images(preimage, postimage,
2459 fixed_buf, fixed_len, postlen);
2460 return 1;
2461 }
2462
2463 static int match_fragment(struct apply_state *state,
2464 struct image *img,
2465 struct image *preimage,
2466 struct image *postimage,
2467 unsigned long try,
2468 int try_lno,
2469 unsigned ws_rule,
2470 int match_beginning, int match_end)
2471 {
2472 int i;
2473 char *fixed_buf, *buf, *orig, *target;
2474 struct strbuf fixed;
2475 size_t fixed_len, postlen;
2476 int preimage_limit;
2477
2478 if (preimage->nr + try_lno <= img->nr) {
2479 /*
2480 * The hunk falls within the boundaries of img.
2481 */
2482 preimage_limit = preimage->nr;
2483 if (match_end && (preimage->nr + try_lno != img->nr))
2484 return 0;
2485 } else if (state->ws_error_action == correct_ws_error &&
2486 (ws_rule & WS_BLANK_AT_EOF)) {
2487 /*
2488 * This hunk extends beyond the end of img, and we are
2489 * removing blank lines at the end of the file. This
2490 * many lines from the beginning of the preimage must
2491 * match with img, and the remainder of the preimage
2492 * must be blank.
2493 */
2494 preimage_limit = img->nr - try_lno;
2495 } else {
2496 /*
2497 * The hunk extends beyond the end of the img and
2498 * we are not removing blanks at the end, so we
2499 * should reject the hunk at this position.
2500 */
2501 return 0;
2502 }
2503
2504 if (match_beginning && try_lno)
2505 return 0;
2506
2507 /* Quick hash check */
2508 for (i = 0; i < preimage_limit; i++)
2509 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2510 (preimage->line[i].hash != img->line[try_lno + i].hash))
2511 return 0;
2512
2513 if (preimage_limit == preimage->nr) {
2514 /*
2515 * Do we have an exact match? If we were told to match
2516 * at the end, size must be exactly at try+fragsize,
2517 * otherwise try+fragsize must be still within the preimage,
2518 * and either case, the old piece should match the preimage
2519 * exactly.
2520 */
2521 if ((match_end
2522 ? (try + preimage->len == img->len)
2523 : (try + preimage->len <= img->len)) &&
2524 !memcmp(img->buf + try, preimage->buf, preimage->len))
2525 return 1;
2526 } else {
2527 /*
2528 * The preimage extends beyond the end of img, so
2529 * there cannot be an exact match.
2530 *
2531 * There must be one non-blank context line that match
2532 * a line before the end of img.
2533 */
2534 char *buf_end;
2535
2536 buf = preimage->buf;
2537 buf_end = buf;
2538 for (i = 0; i < preimage_limit; i++)
2539 buf_end += preimage->line[i].len;
2540
2541 for ( ; buf < buf_end; buf++)
2542 if (!isspace(*buf))
2543 break;
2544 if (buf == buf_end)
2545 return 0;
2546 }
2547
2548 /*
2549 * No exact match. If we are ignoring whitespace, run a line-by-line
2550 * fuzzy matching. We collect all the line length information because
2551 * we need it to adjust whitespace if we match.
2552 */
2553 if (state->ws_ignore_action == ignore_ws_change)
2554 return line_by_line_fuzzy_match(img, preimage, postimage,
2555 try, try_lno, preimage_limit);
2556
2557 if (state->ws_error_action != correct_ws_error)
2558 return 0;
2559
2560 /*
2561 * The hunk does not apply byte-by-byte, but the hash says
2562 * it might with whitespace fuzz. We weren't asked to
2563 * ignore whitespace, we were asked to correct whitespace
2564 * errors, so let's try matching after whitespace correction.
2565 *
2566 * While checking the preimage against the target, whitespace
2567 * errors in both fixed, we count how large the corresponding
2568 * postimage needs to be. The postimage prepared by
2569 * apply_one_fragment() has whitespace errors fixed on added
2570 * lines already, but the common lines were propagated as-is,
2571 * which may become longer when their whitespace errors are
2572 * fixed.
2573 */
2574
2575 /* First count added lines in postimage */
2576 postlen = 0;
2577 for (i = 0; i < postimage->nr; i++) {
2578 if (!(postimage->line[i].flag & LINE_COMMON))
2579 postlen += postimage->line[i].len;
2580 }
2581
2582 /*
2583 * The preimage may extend beyond the end of the file,
2584 * but in this loop we will only handle the part of the
2585 * preimage that falls within the file.
2586 */
2587 strbuf_init(&fixed, preimage->len + 1);
2588 orig = preimage->buf;
2589 target = img->buf + try;
2590 for (i = 0; i < preimage_limit; i++) {
2591 size_t oldlen = preimage->line[i].len;
2592 size_t tgtlen = img->line[try_lno + i].len;
2593 size_t fixstart = fixed.len;
2594 struct strbuf tgtfix;
2595 int match;
2596
2597 /* Try fixing the line in the preimage */
2598 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2599
2600 /* Try fixing the line in the target */
2601 strbuf_init(&tgtfix, tgtlen);
2602 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2603
2604 /*
2605 * If they match, either the preimage was based on
2606 * a version before our tree fixed whitespace breakage,
2607 * or we are lacking a whitespace-fix patch the tree
2608 * the preimage was based on already had (i.e. target
2609 * has whitespace breakage, the preimage doesn't).
2610 * In either case, we are fixing the whitespace breakages
2611 * so we might as well take the fix together with their
2612 * real change.
2613 */
2614 match = (tgtfix.len == fixed.len - fixstart &&
2615 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2616 fixed.len - fixstart));
2617
2618 /* Add the length if this is common with the postimage */
2619 if (preimage->line[i].flag & LINE_COMMON)
2620 postlen += tgtfix.len;
2621
2622 strbuf_release(&tgtfix);
2623 if (!match)
2624 goto unmatch_exit;
2625
2626 orig += oldlen;
2627 target += tgtlen;
2628 }
2629
2630
2631 /*
2632 * Now handle the lines in the preimage that falls beyond the
2633 * end of the file (if any). They will only match if they are
2634 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2635 * false).
2636 */
2637 for ( ; i < preimage->nr; i++) {
2638 size_t fixstart = fixed.len; /* start of the fixed preimage */
2639 size_t oldlen = preimage->line[i].len;
2640 int j;
2641
2642 /* Try fixing the line in the preimage */
2643 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2644
2645 for (j = fixstart; j < fixed.len; j++)
2646 if (!isspace(fixed.buf[j]))
2647 goto unmatch_exit;
2648
2649 orig += oldlen;
2650 }
2651
2652 /*
2653 * Yes, the preimage is based on an older version that still
2654 * has whitespace breakages unfixed, and fixing them makes the
2655 * hunk match. Update the context lines in the postimage.
2656 */
2657 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2658 if (postlen < postimage->len)
2659 postlen = 0;
2660 update_pre_post_images(preimage, postimage,
2661 fixed_buf, fixed_len, postlen);
2662 return 1;
2663
2664 unmatch_exit:
2665 strbuf_release(&fixed);
2666 return 0;
2667 }
2668
2669 static int find_pos(struct apply_state *state,
2670 struct image *img,
2671 struct image *preimage,
2672 struct image *postimage,
2673 int line,
2674 unsigned ws_rule,
2675 int match_beginning, int match_end)
2676 {
2677 int i;
2678 unsigned long backwards, forwards, try;
2679 int backwards_lno, forwards_lno, try_lno;
2680
2681 /*
2682 * If match_beginning or match_end is specified, there is no
2683 * point starting from a wrong line that will never match and
2684 * wander around and wait for a match at the specified end.
2685 */
2686 if (match_beginning)
2687 line = 0;
2688 else if (match_end)
2689 line = img->nr - preimage->nr;
2690
2691 /*
2692 * Because the comparison is unsigned, the following test
2693 * will also take care of a negative line number that can
2694 * result when match_end and preimage is larger than the target.
2695 */
2696 if ((size_t) line > img->nr)
2697 line = img->nr;
2698
2699 try = 0;
2700 for (i = 0; i < line; i++)
2701 try += img->line[i].len;
2702
2703 /*
2704 * There's probably some smart way to do this, but I'll leave
2705 * that to the smart and beautiful people. I'm simple and stupid.
2706 */
2707 backwards = try;
2708 backwards_lno = line;
2709 forwards = try;
2710 forwards_lno = line;
2711 try_lno = line;
2712
2713 for (i = 0; ; i++) {
2714 if (match_fragment(state, img, preimage, postimage,
2715 try, try_lno, ws_rule,
2716 match_beginning, match_end))
2717 return try_lno;
2718
2719 again:
2720 if (backwards_lno == 0 && forwards_lno == img->nr)
2721 break;
2722
2723 if (i & 1) {
2724 if (backwards_lno == 0) {
2725 i++;
2726 goto again;
2727 }
2728 backwards_lno--;
2729 backwards -= img->line[backwards_lno].len;
2730 try = backwards;
2731 try_lno = backwards_lno;
2732 } else {
2733 if (forwards_lno == img->nr) {
2734 i++;
2735 goto again;
2736 }
2737 forwards += img->line[forwards_lno].len;
2738 forwards_lno++;
2739 try = forwards;
2740 try_lno = forwards_lno;
2741 }
2742
2743 }
2744 return -1;
2745 }
2746
2747 static void remove_first_line(struct image *img)
2748 {
2749 img->buf += img->line[0].len;
2750 img->len -= img->line[0].len;
2751 img->line++;
2752 img->nr--;
2753 }
2754
2755 static void remove_last_line(struct image *img)
2756 {
2757 img->len -= img->line[--img->nr].len;
2758 }
2759
2760 /*
2761 * The change from "preimage" and "postimage" has been found to
2762 * apply at applied_pos (counts in line numbers) in "img".
2763 * Update "img" to remove "preimage" and replace it with "postimage".
2764 */
2765 static void update_image(struct apply_state *state,
2766 struct image *img,
2767 int applied_pos,
2768 struct image *preimage,
2769 struct image *postimage)
2770 {
2771 /*
2772 * remove the copy of preimage at offset in img
2773 * and replace it with postimage
2774 */
2775 int i, nr;
2776 size_t remove_count, insert_count, applied_at = 0;
2777 char *result;
2778 int preimage_limit;
2779
2780 /*
2781 * If we are removing blank lines at the end of img,
2782 * the preimage may extend beyond the end.
2783 * If that is the case, we must be careful only to
2784 * remove the part of the preimage that falls within
2785 * the boundaries of img. Initialize preimage_limit
2786 * to the number of lines in the preimage that falls
2787 * within the boundaries.
2788 */
2789 preimage_limit = preimage->nr;
2790 if (preimage_limit > img->nr - applied_pos)
2791 preimage_limit = img->nr - applied_pos;
2792
2793 for (i = 0; i < applied_pos; i++)
2794 applied_at += img->line[i].len;
2795
2796 remove_count = 0;
2797 for (i = 0; i < preimage_limit; i++)
2798 remove_count += img->line[applied_pos + i].len;
2799 insert_count = postimage->len;
2800
2801 /* Adjust the contents */
2802 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2803 memcpy(result, img->buf, applied_at);
2804 memcpy(result + applied_at, postimage->buf, postimage->len);
2805 memcpy(result + applied_at + postimage->len,
2806 img->buf + (applied_at + remove_count),
2807 img->len - (applied_at + remove_count));
2808 free(img->buf);
2809 img->buf = result;
2810 img->len += insert_count - remove_count;
2811 result[img->len] = '\0';
2812
2813 /* Adjust the line table */
2814 nr = img->nr + postimage->nr - preimage_limit;
2815 if (preimage_limit < postimage->nr) {
2816 /*
2817 * NOTE: this knows that we never call remove_first_line()
2818 * on anything other than pre/post image.
2819 */
2820 REALLOC_ARRAY(img->line, nr);
2821 img->line_allocated = img->line;
2822 }
2823 if (preimage_limit != postimage->nr)
2824 memmove(img->line + applied_pos + postimage->nr,
2825 img->line + applied_pos + preimage_limit,
2826 (img->nr - (applied_pos + preimage_limit)) *
2827 sizeof(*img->line));
2828 memcpy(img->line + applied_pos,
2829 postimage->line,
2830 postimage->nr * sizeof(*img->line));
2831 if (!state->allow_overlap)
2832 for (i = 0; i < postimage->nr; i++)
2833 img->line[applied_pos + i].flag |= LINE_PATCHED;
2834 img->nr = nr;
2835 }
2836
2837 /*
2838 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2839 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2840 * replace the part of "img" with "postimage" text.
2841 */
2842 static int apply_one_fragment(struct apply_state *state,
2843 struct image *img, struct fragment *frag,
2844 int inaccurate_eof, unsigned ws_rule,
2845 int nth_fragment)
2846 {
2847 int match_beginning, match_end;
2848 const char *patch = frag->patch;
2849 int size = frag->size;
2850 char *old, *oldlines;
2851 struct strbuf newlines;
2852 int new_blank_lines_at_end = 0;
2853 int found_new_blank_lines_at_end = 0;
2854 int hunk_linenr = frag->linenr;
2855 unsigned long leading, trailing;
2856 int pos, applied_pos;
2857 struct image preimage;
2858 struct image postimage;
2859
2860 memset(&preimage, 0, sizeof(preimage));
2861 memset(&postimage, 0, sizeof(postimage));
2862 oldlines = xmalloc(size);
2863 strbuf_init(&newlines, size);
2864
2865 old = oldlines;
2866 while (size > 0) {
2867 char first;
2868 int len = linelen(patch, size);
2869 int plen;
2870 int added_blank_line = 0;
2871 int is_blank_context = 0;
2872 size_t start;
2873
2874 if (!len)
2875 break;
2876
2877 /*
2878 * "plen" is how much of the line we should use for
2879 * the actual patch data. Normally we just remove the
2880 * first character on the line, but if the line is
2881 * followed by "\ No newline", then we also remove the
2882 * last one (which is the newline, of course).
2883 */
2884 plen = len - 1;
2885 if (len < size && patch[len] == '\\')
2886 plen--;
2887 first = *patch;
2888 if (state->apply_in_reverse) {
2889 if (first == '-')
2890 first = '+';
2891 else if (first == '+')
2892 first = '-';
2893 }
2894
2895 switch (first) {
2896 case '\n':
2897 /* Newer GNU diff, empty context line */
2898 if (plen < 0)
2899 /* ... followed by '\No newline'; nothing */
2900 break;
2901 *old++ = '\n';
2902 strbuf_addch(&newlines, '\n');
2903 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2904 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2905 is_blank_context = 1;
2906 break;
2907 case ' ':
2908 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2909 ws_blank_line(patch + 1, plen, ws_rule))
2910 is_blank_context = 1;
2911 case '-':
2912 memcpy(old, patch + 1, plen);
2913 add_line_info(&preimage, old, plen,
2914 (first == ' ' ? LINE_COMMON : 0));
2915 old += plen;
2916 if (first == '-')
2917 break;
2918 /* Fall-through for ' ' */
2919 case '+':
2920 /* --no-add does not add new lines */
2921 if (first == '+' && state->no_add)
2922 break;
2923
2924 start = newlines.len;
2925 if (first != '+' ||
2926 !state->whitespace_error ||
2927 state->ws_error_action != correct_ws_error) {
2928 strbuf_add(&newlines, patch + 1, plen);
2929 }
2930 else {
2931 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2932 }
2933 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2934 (first == '+' ? 0 : LINE_COMMON));
2935 if (first == '+' &&
2936 (ws_rule & WS_BLANK_AT_EOF) &&
2937 ws_blank_line(patch + 1, plen, ws_rule))
2938 added_blank_line = 1;
2939 break;
2940 case '@': case '\\':
2941 /* Ignore it, we already handled it */
2942 break;
2943 default:
2944 if (state->apply_verbosity > verbosity_normal)
2945 error(_("invalid start of line: '%c'"), first);
2946 applied_pos = -1;
2947 goto out;
2948 }
2949 if (added_blank_line) {
2950 if (!new_blank_lines_at_end)
2951 found_new_blank_lines_at_end = hunk_linenr;
2952 new_blank_lines_at_end++;
2953 }
2954 else if (is_blank_context)
2955 ;
2956 else
2957 new_blank_lines_at_end = 0;
2958 patch += len;
2959 size -= len;
2960 hunk_linenr++;
2961 }
2962 if (inaccurate_eof &&
2963 old > oldlines && old[-1] == '\n' &&
2964 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2965 old--;
2966 strbuf_setlen(&newlines, newlines.len - 1);
2967 }
2968
2969 leading = frag->leading;
2970 trailing = frag->trailing;
2971
2972 /*
2973 * A hunk to change lines at the beginning would begin with
2974 * @@ -1,L +N,M @@
2975 * but we need to be careful. -U0 that inserts before the second
2976 * line also has this pattern.
2977 *
2978 * And a hunk to add to an empty file would begin with
2979 * @@ -0,0 +N,M @@
2980 *
2981 * In other words, a hunk that is (frag->oldpos <= 1) with or
2982 * without leading context must match at the beginning.
2983 */
2984 match_beginning = (!frag->oldpos ||
2985 (frag->oldpos == 1 && !state->unidiff_zero));
2986
2987 /*
2988 * A hunk without trailing lines must match at the end.
2989 * However, we simply cannot tell if a hunk must match end
2990 * from the lack of trailing lines if the patch was generated
2991 * with unidiff without any context.
2992 */
2993 match_end = !state->unidiff_zero && !trailing;
2994
2995 pos = frag->newpos ? (frag->newpos - 1) : 0;
2996 preimage.buf = oldlines;
2997 preimage.len = old - oldlines;
2998 postimage.buf = newlines.buf;
2999 postimage.len = newlines.len;
3000 preimage.line = preimage.line_allocated;
3001 postimage.line = postimage.line_allocated;
3002
3003 for (;;) {
3004
3005 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3006 ws_rule, match_beginning, match_end);
3007
3008 if (applied_pos >= 0)
3009 break;
3010
3011 /* Am I at my context limits? */
3012 if ((leading <= state->p_context) && (trailing <= state->p_context))
3013 break;
3014 if (match_beginning || match_end) {
3015 match_beginning = match_end = 0;
3016 continue;
3017 }
3018
3019 /*
3020 * Reduce the number of context lines; reduce both
3021 * leading and trailing if they are equal otherwise
3022 * just reduce the larger context.
3023 */
3024 if (leading >= trailing) {
3025 remove_first_line(&preimage);
3026 remove_first_line(&postimage);
3027 pos--;
3028 leading--;
3029 }
3030 if (trailing > leading) {
3031 remove_last_line(&preimage);
3032 remove_last_line(&postimage);
3033 trailing--;
3034 }
3035 }
3036
3037 if (applied_pos >= 0) {
3038 if (new_blank_lines_at_end &&
3039 preimage.nr + applied_pos >= img->nr &&
3040 (ws_rule & WS_BLANK_AT_EOF) &&
3041 state->ws_error_action != nowarn_ws_error) {
3042 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3043 found_new_blank_lines_at_end);
3044 if (state->ws_error_action == correct_ws_error) {
3045 while (new_blank_lines_at_end--)
3046 remove_last_line(&postimage);
3047 }
3048 /*
3049 * We would want to prevent write_out_results()
3050 * from taking place in apply_patch() that follows
3051 * the callchain led us here, which is:
3052 * apply_patch->check_patch_list->check_patch->
3053 * apply_data->apply_fragments->apply_one_fragment
3054 */
3055 if (state->ws_error_action == die_on_ws_error)
3056 state->apply = 0;
3057 }
3058
3059 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3060 int offset = applied_pos - pos;
3061 if (state->apply_in_reverse)
3062 offset = 0 - offset;
3063 fprintf_ln(stderr,
3064 Q_("Hunk #%d succeeded at %d (offset %d line).",
3065 "Hunk #%d succeeded at %d (offset %d lines).",
3066 offset),
3067 nth_fragment, applied_pos + 1, offset);
3068 }
3069
3070 /*
3071 * Warn if it was necessary to reduce the number
3072 * of context lines.
3073 */
3074 if ((leading != frag->leading ||
3075 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3076 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3077 " to apply fragment at %d"),
3078 leading, trailing, applied_pos+1);
3079 update_image(state, img, applied_pos, &preimage, &postimage);
3080 } else {
3081 if (state->apply_verbosity > verbosity_normal)
3082 error(_("while searching for:\n%.*s"),
3083 (int)(old - oldlines), oldlines);
3084 }
3085
3086 out:
3087 free(oldlines);
3088 strbuf_release(&newlines);
3089 free(preimage.line_allocated);
3090 free(postimage.line_allocated);
3091
3092 return (applied_pos < 0);
3093 }
3094
3095 static int apply_binary_fragment(struct apply_state *state,
3096 struct image *img,
3097 struct patch *patch)
3098 {
3099 struct fragment *fragment = patch->fragments;
3100 unsigned long len;
3101 void *dst;
3102
3103 if (!fragment)
3104 return error(_("missing binary patch data for '%s'"),
3105 patch->new_name ?
3106 patch->new_name :
3107 patch->old_name);
3108
3109 /* Binary patch is irreversible without the optional second hunk */
3110 if (state->apply_in_reverse) {
3111 if (!fragment->next)
3112 return error(_("cannot reverse-apply a binary patch "
3113 "without the reverse hunk to '%s'"),
3114 patch->new_name
3115 ? patch->new_name : patch->old_name);
3116 fragment = fragment->next;
3117 }
3118 switch (fragment->binary_patch_method) {
3119 case BINARY_DELTA_DEFLATED:
3120 dst = patch_delta(img->buf, img->len, fragment->patch,
3121 fragment->size, &len);
3122 if (!dst)
3123 return -1;
3124 clear_image(img);
3125 img->buf = dst;
3126 img->len = len;
3127 return 0;
3128 case BINARY_LITERAL_DEFLATED:
3129 clear_image(img);
3130 img->len = fragment->size;
3131 img->buf = xmemdupz(fragment->patch, img->len);
3132 return 0;
3133 }
3134 return -1;
3135 }
3136
3137 /*
3138 * Replace "img" with the result of applying the binary patch.
3139 * The binary patch data itself in patch->fragment is still kept
3140 * but the preimage prepared by the caller in "img" is freed here
3141 * or in the helper function apply_binary_fragment() this calls.
3142 */
3143 static int apply_binary(struct apply_state *state,
3144 struct image *img,
3145 struct patch *patch)
3146 {
3147 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3148 struct object_id oid;
3149
3150 /*
3151 * For safety, we require patch index line to contain
3152 * full 40-byte textual SHA1 for old and new, at least for now.
3153 */
3154 if (strlen(patch->old_sha1_prefix) != 40 ||
3155 strlen(patch->new_sha1_prefix) != 40 ||
3156 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3157 get_oid_hex(patch->new_sha1_prefix, &oid))
3158 return error(_("cannot apply binary patch to '%s' "
3159 "without full index line"), name);
3160
3161 if (patch->old_name) {
3162 /*
3163 * See if the old one matches what the patch
3164 * applies to.
3165 */
3166 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3167 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3168 return error(_("the patch applies to '%s' (%s), "
3169 "which does not match the "
3170 "current contents."),
3171 name, oid_to_hex(&oid));
3172 }
3173 else {
3174 /* Otherwise, the old one must be empty. */
3175 if (img->len)
3176 return error(_("the patch applies to an empty "
3177 "'%s' but it is not empty"), name);
3178 }
3179
3180 get_oid_hex(patch->new_sha1_prefix, &oid);
3181 if (is_null_oid(&oid)) {
3182 clear_image(img);
3183 return 0; /* deletion patch */
3184 }
3185
3186 if (has_sha1_file(oid.hash)) {
3187 /* We already have the postimage */
3188 enum object_type type;
3189 unsigned long size;
3190 char *result;
3191
3192 result = read_sha1_file(oid.hash, &type, &size);
3193 if (!result)
3194 return error(_("the necessary postimage %s for "
3195 "'%s' cannot be read"),
3196 patch->new_sha1_prefix, name);
3197 clear_image(img);
3198 img->buf = result;
3199 img->len = size;
3200 } else {
3201 /*
3202 * We have verified buf matches the preimage;
3203 * apply the patch data to it, which is stored
3204 * in the patch->fragments->{patch,size}.
3205 */
3206 if (apply_binary_fragment(state, img, patch))
3207 return error(_("binary patch does not apply to '%s'"),
3208 name);
3209
3210 /* verify that the result matches */
3211 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3212 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3213 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3214 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3215 }
3216
3217 return 0;
3218 }
3219
3220 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3221 {
3222 struct fragment *frag = patch->fragments;
3223 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3224 unsigned ws_rule = patch->ws_rule;
3225 unsigned inaccurate_eof = patch->inaccurate_eof;
3226 int nth = 0;
3227
3228 if (patch->is_binary)
3229 return apply_binary(state, img, patch);
3230
3231 while (frag) {
3232 nth++;
3233 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3234 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3235 if (!state->apply_with_reject)
3236 return -1;
3237 frag->rejected = 1;
3238 }
3239 frag = frag->next;
3240 }
3241 return 0;
3242 }
3243
3244 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3245 {
3246 if (S_ISGITLINK(mode)) {
3247 strbuf_grow(buf, 100);
3248 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3249 } else {
3250 enum object_type type;
3251 unsigned long sz;
3252 char *result;
3253
3254 result = read_sha1_file(oid->hash, &type, &sz);
3255 if (!result)
3256 return -1;
3257 /* XXX read_sha1_file NUL-terminates */
3258 strbuf_attach(buf, result, sz, sz + 1);
3259 }
3260 return 0;
3261 }
3262
3263 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3264 {
3265 if (!ce)
3266 return 0;
3267 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3268 }
3269
3270 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3271 {
3272 struct string_list_item *item;
3273
3274 if (name == NULL)
3275 return NULL;
3276
3277 item = string_list_lookup(&state->fn_table, name);
3278 if (item != NULL)
3279 return (struct patch *)item->util;
3280
3281 return NULL;
3282 }
3283
3284 /*
3285 * item->util in the filename table records the status of the path.
3286 * Usually it points at a patch (whose result records the contents
3287 * of it after applying it), but it could be PATH_WAS_DELETED for a
3288 * path that a previously applied patch has already removed, or
3289 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3290 *
3291 * The latter is needed to deal with a case where two paths A and B
3292 * are swapped by first renaming A to B and then renaming B to A;
3293 * moving A to B should not be prevented due to presence of B as we
3294 * will remove it in a later patch.
3295 */
3296 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3297 #define PATH_WAS_DELETED ((struct patch *) -1)
3298
3299 static int to_be_deleted(struct patch *patch)
3300 {
3301 return patch == PATH_TO_BE_DELETED;
3302 }
3303
3304 static int was_deleted(struct patch *patch)
3305 {
3306 return patch == PATH_WAS_DELETED;
3307 }
3308
3309 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3310 {
3311 struct string_list_item *item;
3312
3313 /*
3314 * Always add new_name unless patch is a deletion
3315 * This should cover the cases for normal diffs,
3316 * file creations and copies
3317 */
3318 if (patch->new_name != NULL) {
3319 item = string_list_insert(&state->fn_table, patch->new_name);
3320 item->util = patch;
3321 }
3322
3323 /*
3324 * store a failure on rename/deletion cases because
3325 * later chunks shouldn't patch old names
3326 */
3327 if ((patch->new_name == NULL) || (patch->is_rename)) {
3328 item = string_list_insert(&state->fn_table, patch->old_name);
3329 item->util = PATH_WAS_DELETED;
3330 }
3331 }
3332
3333 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3334 {
3335 /*
3336 * store information about incoming file deletion
3337 */
3338 while (patch) {
3339 if ((patch->new_name == NULL) || (patch->is_rename)) {
3340 struct string_list_item *item;
3341 item = string_list_insert(&state->fn_table, patch->old_name);
3342 item->util = PATH_TO_BE_DELETED;
3343 }
3344 patch = patch->next;
3345 }
3346 }
3347
3348 static int checkout_target(struct index_state *istate,
3349 struct cache_entry *ce, struct stat *st)
3350 {
3351 struct checkout costate = CHECKOUT_INIT;
3352
3353 costate.refresh_cache = 1;
3354 costate.istate = istate;
3355 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3356 return error(_("cannot checkout %s"), ce->name);
3357 return 0;
3358 }
3359
3360 static struct patch *previous_patch(struct apply_state *state,
3361 struct patch *patch,
3362 int *gone)
3363 {
3364 struct patch *previous;
3365
3366 *gone = 0;
3367 if (patch->is_copy || patch->is_rename)
3368 return NULL; /* "git" patches do not depend on the order */
3369
3370 previous = in_fn_table(state, patch->old_name);
3371 if (!previous)
3372 return NULL;
3373
3374 if (to_be_deleted(previous))
3375 return NULL; /* the deletion hasn't happened yet */
3376
3377 if (was_deleted(previous))
3378 *gone = 1;
3379
3380 return previous;
3381 }
3382
3383 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3384 {
3385 if (S_ISGITLINK(ce->ce_mode)) {
3386 if (!S_ISDIR(st->st_mode))
3387 return -1;
3388 return 0;
3389 }
3390 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3391 }
3392
3393 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3394
3395 static int load_patch_target(struct apply_state *state,
3396 struct strbuf *buf,
3397 const struct cache_entry *ce,
3398 struct stat *st,
3399 const char *name,
3400 unsigned expected_mode)
3401 {
3402 if (state->cached || state->check_index) {
3403 if (read_file_or_gitlink(ce, buf))
3404 return error(_("failed to read %s"), name);
3405 } else if (name) {
3406 if (S_ISGITLINK(expected_mode)) {
3407 if (ce)
3408 return read_file_or_gitlink(ce, buf);
3409 else
3410 return SUBMODULE_PATCH_WITHOUT_INDEX;
3411 } else if (has_symlink_leading_path(name, strlen(name))) {
3412 return error(_("reading from '%s' beyond a symbolic link"), name);
3413 } else {
3414 if (read_old_data(st, name, buf))
3415 return error(_("failed to read %s"), name);
3416 }
3417 }
3418 return 0;
3419 }
3420
3421 /*
3422 * We are about to apply "patch"; populate the "image" with the
3423 * current version we have, from the working tree or from the index,
3424 * depending on the situation e.g. --cached/--index. If we are
3425 * applying a non-git patch that incrementally updates the tree,
3426 * we read from the result of a previous diff.
3427 */
3428 static int load_preimage(struct apply_state *state,
3429 struct image *image,
3430 struct patch *patch, struct stat *st,
3431 const struct cache_entry *ce)
3432 {
3433 struct strbuf buf = STRBUF_INIT;
3434 size_t len;
3435 char *img;
3436 struct patch *previous;
3437 int status;
3438
3439 previous = previous_patch(state, patch, &status);
3440 if (status)
3441 return error(_("path %s has been renamed/deleted"),
3442 patch->old_name);
3443 if (previous) {
3444 /* We have a patched copy in memory; use that. */
3445 strbuf_add(&buf, previous->result, previous->resultsize);
3446 } else {
3447 status = load_patch_target(state, &buf, ce, st,
3448 patch->old_name, patch->old_mode);
3449 if (status < 0)
3450 return status;
3451 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3452 /*
3453 * There is no way to apply subproject
3454 * patch without looking at the index.
3455 * NEEDSWORK: shouldn't this be flagged
3456 * as an error???
3457 */
3458 free_fragment_list(patch->fragments);
3459 patch->fragments = NULL;
3460 } else if (status) {
3461 return error(_("failed to read %s"), patch->old_name);
3462 }
3463 }
3464
3465 img = strbuf_detach(&buf, &len);
3466 prepare_image(image, img, len, !patch->is_binary);
3467 return 0;
3468 }
3469
3470 static int three_way_merge(struct image *image,
3471 char *path,
3472 const struct object_id *base,
3473 const struct object_id *ours,
3474 const struct object_id *theirs)
3475 {
3476 mmfile_t base_file, our_file, their_file;
3477 mmbuffer_t result = { NULL };
3478 int status;
3479
3480 read_mmblob(&base_file, base);
3481 read_mmblob(&our_file, ours);
3482 read_mmblob(&their_file, theirs);
3483 status = ll_merge(&result, path,
3484 &base_file, "base",
3485 &our_file, "ours",
3486 &their_file, "theirs", NULL);
3487 free(base_file.ptr);
3488 free(our_file.ptr);
3489 free(their_file.ptr);
3490 if (status < 0 || !result.ptr) {
3491 free(result.ptr);
3492 return -1;
3493 }
3494 clear_image(image);
3495 image->buf = result.ptr;
3496 image->len = result.size;
3497
3498 return status;
3499 }
3500
3501 /*
3502 * When directly falling back to add/add three-way merge, we read from
3503 * the current contents of the new_name. In no cases other than that
3504 * this function will be called.
3505 */
3506 static int load_current(struct apply_state *state,
3507 struct image *image,
3508 struct patch *patch)
3509 {
3510 struct strbuf buf = STRBUF_INIT;
3511 int status, pos;
3512 size_t len;
3513 char *img;
3514 struct stat st;
3515 struct cache_entry *ce;
3516 char *name = patch->new_name;
3517 unsigned mode = patch->new_mode;
3518
3519 if (!patch->is_new)
3520 die("BUG: patch to %s is not a creation", patch->old_name);
3521
3522 pos = cache_name_pos(name, strlen(name));
3523 if (pos < 0)
3524 return error(_("%s: does not exist in index"), name);
3525 ce = active_cache[pos];
3526 if (lstat(name, &st)) {
3527 if (errno != ENOENT)
3528 return error_errno("%s", name);
3529 if (checkout_target(&the_index, ce, &st))
3530 return -1;
3531 }
3532 if (verify_index_match(ce, &st))
3533 return error(_("%s: does not match index"), name);
3534
3535 status = load_patch_target(state, &buf, ce, &st, name, mode);
3536 if (status < 0)
3537 return status;
3538 else if (status)
3539 return -1;
3540 img = strbuf_detach(&buf, &len);
3541 prepare_image(image, img, len, !patch->is_binary);
3542 return 0;
3543 }
3544
3545 static int try_threeway(struct apply_state *state,
3546 struct image *image,
3547 struct patch *patch,
3548 struct stat *st,
3549 const struct cache_entry *ce)
3550 {
3551 struct object_id pre_oid, post_oid, our_oid;
3552 struct strbuf buf = STRBUF_INIT;
3553 size_t len;
3554 int status;
3555 char *img;
3556 struct image tmp_image;
3557
3558 /* No point falling back to 3-way merge in these cases */
3559 if (patch->is_delete ||
3560 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3561 return -1;
3562
3563 /* Preimage the patch was prepared for */
3564 if (patch->is_new)
3565 write_sha1_file("", 0, blob_type, pre_oid.hash);
3566 else if (get_sha1(patch->old_sha1_prefix, pre_oid.hash) ||
3567 read_blob_object(&buf, &pre_oid, patch->old_mode))
3568 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3569
3570 if (state->apply_verbosity > verbosity_silent)
3571 fprintf(stderr, _("Falling back to three-way merge...\n"));
3572
3573 img = strbuf_detach(&buf, &len);
3574 prepare_image(&tmp_image, img, len, 1);
3575 /* Apply the patch to get the post image */
3576 if (apply_fragments(state, &tmp_image, patch) < 0) {
3577 clear_image(&tmp_image);
3578 return -1;
3579 }
3580 /* post_oid is theirs */
3581 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3582 clear_image(&tmp_image);
3583
3584 /* our_oid is ours */
3585 if (patch->is_new) {
3586 if (load_current(state, &tmp_image, patch))
3587 return error(_("cannot read the current contents of '%s'"),
3588 patch->new_name);
3589 } else {
3590 if (load_preimage(state, &tmp_image, patch, st, ce))
3591 return error(_("cannot read the current contents of '%s'"),
3592 patch->old_name);
3593 }
3594 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3595 clear_image(&tmp_image);
3596
3597 /* in-core three-way merge between post and our using pre as base */
3598 status = three_way_merge(image, patch->new_name,
3599 &pre_oid, &our_oid, &post_oid);
3600 if (status < 0) {
3601 if (state->apply_verbosity > verbosity_silent)
3602 fprintf(stderr,
3603 _("Failed to fall back on three-way merge...\n"));
3604 return status;
3605 }
3606
3607 if (status) {
3608 patch->conflicted_threeway = 1;
3609 if (patch->is_new)
3610 oidclr(&patch->threeway_stage[0]);
3611 else
3612 oidcpy(&patch->threeway_stage[0], &pre_oid);
3613 oidcpy(&patch->threeway_stage[1], &our_oid);
3614 oidcpy(&patch->threeway_stage[2], &post_oid);
3615 if (state->apply_verbosity > verbosity_silent)
3616 fprintf(stderr,
3617 _("Applied patch to '%s' with conflicts.\n"),
3618 patch->new_name);
3619 } else {
3620 if (state->apply_verbosity > verbosity_silent)
3621 fprintf(stderr,
3622 _("Applied patch to '%s' cleanly.\n"),
3623 patch->new_name);
3624 }
3625 return 0;
3626 }
3627
3628 static int apply_data(struct apply_state *state, struct patch *patch,
3629 struct stat *st, const struct cache_entry *ce)
3630 {
3631 struct image image;
3632
3633 if (load_preimage(state, &image, patch, st, ce) < 0)
3634 return -1;
3635
3636 if (patch->direct_to_threeway ||
3637 apply_fragments(state, &image, patch) < 0) {
3638 /* Note: with --reject, apply_fragments() returns 0 */
3639 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3640 return -1;
3641 }
3642 patch->result = image.buf;
3643 patch->resultsize = image.len;
3644 add_to_fn_table(state, patch);
3645 free(image.line_allocated);
3646
3647 if (0 < patch->is_delete && patch->resultsize)
3648 return error(_("removal patch leaves file contents"));
3649
3650 return 0;
3651 }
3652
3653 /*
3654 * If "patch" that we are looking at modifies or deletes what we have,
3655 * we would want it not to lose any local modification we have, either
3656 * in the working tree or in the index.
3657 *
3658 * This also decides if a non-git patch is a creation patch or a
3659 * modification to an existing empty file. We do not check the state
3660 * of the current tree for a creation patch in this function; the caller
3661 * check_patch() separately makes sure (and errors out otherwise) that
3662 * the path the patch creates does not exist in the current tree.
3663 */
3664 static int check_preimage(struct apply_state *state,
3665 struct patch *patch,
3666 struct cache_entry **ce,
3667 struct stat *st)
3668 {
3669 const char *old_name = patch->old_name;
3670 struct patch *previous = NULL;
3671 int stat_ret = 0, status;
3672 unsigned st_mode = 0;
3673
3674 if (!old_name)
3675 return 0;
3676
3677 assert(patch->is_new <= 0);
3678 previous = previous_patch(state, patch, &status);
3679
3680 if (status)
3681 return error(_("path %s has been renamed/deleted"), old_name);
3682 if (previous) {
3683 st_mode = previous->new_mode;
3684 } else if (!state->cached) {
3685 stat_ret = lstat(old_name, st);
3686 if (stat_ret && errno != ENOENT)
3687 return error_errno("%s", old_name);
3688 }
3689
3690 if (state->check_index && !previous) {
3691 int pos = cache_name_pos(old_name, strlen(old_name));
3692 if (pos < 0) {
3693 if (patch->is_new < 0)
3694 goto is_new;
3695 return error(_("%s: does not exist in index"), old_name);
3696 }
3697 *ce = active_cache[pos];
3698 if (stat_ret < 0) {
3699 if (checkout_target(&the_index, *ce, st))
3700 return -1;
3701 }
3702 if (!state->cached && verify_index_match(*ce, st))
3703 return error(_("%s: does not match index"), old_name);
3704 if (state->cached)
3705 st_mode = (*ce)->ce_mode;
3706 } else if (stat_ret < 0) {
3707 if (patch->is_new < 0)
3708 goto is_new;
3709 return error_errno("%s", old_name);
3710 }
3711
3712 if (!state->cached && !previous)
3713 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3714
3715 if (patch->is_new < 0)
3716 patch->is_new = 0;
3717 if (!patch->old_mode)
3718 patch->old_mode = st_mode;
3719 if ((st_mode ^ patch->old_mode) & S_IFMT)
3720 return error(_("%s: wrong type"), old_name);
3721 if (st_mode != patch->old_mode)
3722 warning(_("%s has type %o, expected %o"),
3723 old_name, st_mode, patch->old_mode);
3724 if (!patch->new_mode && !patch->is_delete)
3725 patch->new_mode = st_mode;
3726 return 0;
3727
3728 is_new:
3729 patch->is_new = 1;
3730 patch->is_delete = 0;
3731 free(patch->old_name);
3732 patch->old_name = NULL;
3733 return 0;
3734 }
3735
3736
3737 #define EXISTS_IN_INDEX 1
3738 #define EXISTS_IN_WORKTREE 2
3739
3740 static int check_to_create(struct apply_state *state,
3741 const char *new_name,
3742 int ok_if_exists)
3743 {
3744 struct stat nst;
3745
3746 if (state->check_index &&
3747 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3748 !ok_if_exists)
3749 return EXISTS_IN_INDEX;
3750 if (state->cached)
3751 return 0;
3752
3753 if (!lstat(new_name, &nst)) {
3754 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3755 return 0;
3756 /*
3757 * A leading component of new_name might be a symlink
3758 * that is going to be removed with this patch, but
3759 * still pointing at somewhere that has the path.
3760 * In such a case, path "new_name" does not exist as
3761 * far as git is concerned.
3762 */
3763 if (has_symlink_leading_path(new_name, strlen(new_name)))
3764 return 0;
3765
3766 return EXISTS_IN_WORKTREE;
3767 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3768 return error_errno("%s", new_name);
3769 }
3770 return 0;
3771 }
3772
3773 static uintptr_t register_symlink_changes(struct apply_state *state,
3774 const char *path,
3775 uintptr_t what)
3776 {
3777 struct string_list_item *ent;
3778
3779 ent = string_list_lookup(&state->symlink_changes, path);
3780 if (!ent) {
3781 ent = string_list_insert(&state->symlink_changes, path);
3782 ent->util = (void *)0;
3783 }
3784 ent->util = (void *)(what | ((uintptr_t)ent->util));
3785 return (uintptr_t)ent->util;
3786 }
3787
3788 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3789 {
3790 struct string_list_item *ent;
3791
3792 ent = string_list_lookup(&state->symlink_changes, path);
3793 if (!ent)
3794 return 0;
3795 return (uintptr_t)ent->util;
3796 }
3797
3798 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3799 {
3800 for ( ; patch; patch = patch->next) {
3801 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3802 (patch->is_rename || patch->is_delete))
3803 /* the symlink at patch->old_name is removed */
3804 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3805
3806 if (patch->new_name && S_ISLNK(patch->new_mode))
3807 /* the symlink at patch->new_name is created or remains */
3808 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3809 }
3810 }
3811
3812 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3813 {
3814 do {
3815 unsigned int change;
3816
3817 while (--name->len && name->buf[name->len] != '/')
3818 ; /* scan backwards */
3819 if (!name->len)
3820 break;
3821 name->buf[name->len] = '\0';
3822 change = check_symlink_changes(state, name->buf);
3823 if (change & APPLY_SYMLINK_IN_RESULT)
3824 return 1;
3825 if (change & APPLY_SYMLINK_GOES_AWAY)
3826 /*
3827 * This cannot be "return 0", because we may
3828 * see a new one created at a higher level.
3829 */
3830 continue;
3831
3832 /* otherwise, check the preimage */
3833 if (state->check_index) {
3834 struct cache_entry *ce;
3835
3836 ce = cache_file_exists(name->buf, name->len, ignore_case);
3837 if (ce && S_ISLNK(ce->ce_mode))
3838 return 1;
3839 } else {
3840 struct stat st;
3841 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3842 return 1;