apply: avoid out-of-bounds access in fuzzy_matchlines()
[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 unsigned int is_toplevel_relative:1;
214 unsigned int inaccurate_eof:1;
215 unsigned int is_binary:1;
216 unsigned int is_copy:1;
217 unsigned int is_rename:1;
218 unsigned int recount:1;
219 unsigned int conflicted_threeway:1;
220 unsigned int direct_to_threeway:1;
221 struct fragment *fragments;
222 char *result;
223 size_t resultsize;
224 char old_sha1_prefix[41];
225 char new_sha1_prefix[41];
226 struct patch *next;
227
228 /* three-way fallback result */
229 struct object_id threeway_stage[3];
230 };
231
232 static void free_fragment_list(struct fragment *list)
233 {
234 while (list) {
235 struct fragment *next = list->next;
236 if (list->free_patch)
237 free((char *)list->patch);
238 free(list);
239 list = next;
240 }
241 }
242
243 static void free_patch(struct patch *patch)
244 {
245 free_fragment_list(patch->fragments);
246 free(patch->def_name);
247 free(patch->old_name);
248 free(patch->new_name);
249 free(patch->result);
250 free(patch);
251 }
252
253 static void free_patch_list(struct patch *list)
254 {
255 while (list) {
256 struct patch *next = list->next;
257 free_patch(list);
258 list = next;
259 }
260 }
261
262 /*
263 * A line in a file, len-bytes long (includes the terminating LF,
264 * except for an incomplete line at the end if the file ends with
265 * one), and its contents hashes to 'hash'.
266 */
267 struct line {
268 size_t len;
269 unsigned hash : 24;
270 unsigned flag : 8;
271 #define LINE_COMMON 1
272 #define LINE_PATCHED 2
273 };
274
275 /*
276 * This represents a "file", which is an array of "lines".
277 */
278 struct image {
279 char *buf;
280 size_t len;
281 size_t nr;
282 size_t alloc;
283 struct line *line_allocated;
284 struct line *line;
285 };
286
287 static uint32_t hash_line(const char *cp, size_t len)
288 {
289 size_t i;
290 uint32_t h;
291 for (i = 0, h = 0; i < len; i++) {
292 if (!isspace(cp[i])) {
293 h = h * 3 + (cp[i] & 0xff);
294 }
295 }
296 return h;
297 }
298
299 /*
300 * Compare lines s1 of length n1 and s2 of length n2, ignoring
301 * whitespace difference. Returns 1 if they match, 0 otherwise
302 */
303 static int fuzzy_matchlines(const char *s1, size_t n1,
304 const char *s2, size_t n2)
305 {
306 const char *end1 = s1 + n1;
307 const char *end2 = s2 + n2;
308
309 /* ignore line endings */
310 while (s1 < end1 && (end1[-1] == '\r' || end1[-1] == '\n'))
311 end1--;
312 while (s2 < end2 && (end2[-1] == '\r' || end2[-1] == '\n'))
313 end2--;
314
315 while (s1 < end1 && s2 < end2) {
316 if (isspace(*s1)) {
317 /*
318 * Skip whitespace. We check on both buffers
319 * because we don't want "a b" to match "ab".
320 */
321 if (!isspace(*s2))
322 return 0;
323 while (s1 < end1 && isspace(*s1))
324 s1++;
325 while (s2 < end2 && isspace(*s2))
326 s2++;
327 } else if (*s1++ != *s2++)
328 return 0;
329 }
330
331 /* If we reached the end on one side only, lines don't match. */
332 return s1 == end1 && s2 == end2;
333 }
334
335 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
336 {
337 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
338 img->line_allocated[img->nr].len = len;
339 img->line_allocated[img->nr].hash = hash_line(bol, len);
340 img->line_allocated[img->nr].flag = flag;
341 img->nr++;
342 }
343
344 /*
345 * "buf" has the file contents to be patched (read from various sources).
346 * attach it to "image" and add line-based index to it.
347 * "image" now owns the "buf".
348 */
349 static void prepare_image(struct image *image, char *buf, size_t len,
350 int prepare_linetable)
351 {
352 const char *cp, *ep;
353
354 memset(image, 0, sizeof(*image));
355 image->buf = buf;
356 image->len = len;
357
358 if (!prepare_linetable)
359 return;
360
361 ep = image->buf + image->len;
362 cp = image->buf;
363 while (cp < ep) {
364 const char *next;
365 for (next = cp; next < ep && *next != '\n'; next++)
366 ;
367 if (next < ep)
368 next++;
369 add_line_info(image, cp, next - cp, 0);
370 cp = next;
371 }
372 image->line = image->line_allocated;
373 }
374
375 static void clear_image(struct image *image)
376 {
377 free(image->buf);
378 free(image->line_allocated);
379 memset(image, 0, sizeof(*image));
380 }
381
382 /* fmt must contain _one_ %s and no other substitution */
383 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
384 {
385 struct strbuf sb = STRBUF_INIT;
386
387 if (patch->old_name && patch->new_name &&
388 strcmp(patch->old_name, patch->new_name)) {
389 quote_c_style(patch->old_name, &sb, NULL, 0);
390 strbuf_addstr(&sb, " => ");
391 quote_c_style(patch->new_name, &sb, NULL, 0);
392 } else {
393 const char *n = patch->new_name;
394 if (!n)
395 n = patch->old_name;
396 quote_c_style(n, &sb, NULL, 0);
397 }
398 fprintf(output, fmt, sb.buf);
399 fputc('\n', output);
400 strbuf_release(&sb);
401 }
402
403 #define SLOP (16)
404
405 static int read_patch_file(struct strbuf *sb, int fd)
406 {
407 if (strbuf_read(sb, fd, 0) < 0)
408 return error_errno("git apply: failed to read");
409
410 /*
411 * Make sure that we have some slop in the buffer
412 * so that we can do speculative "memcmp" etc, and
413 * see to it that it is NUL-filled.
414 */
415 strbuf_grow(sb, SLOP);
416 memset(sb->buf + sb->len, 0, SLOP);
417 return 0;
418 }
419
420 static unsigned long linelen(const char *buffer, unsigned long size)
421 {
422 unsigned long len = 0;
423 while (size--) {
424 len++;
425 if (*buffer++ == '\n')
426 break;
427 }
428 return len;
429 }
430
431 static int is_dev_null(const char *str)
432 {
433 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
434 }
435
436 #define TERM_SPACE 1
437 #define TERM_TAB 2
438
439 static int name_terminate(int c, int terminate)
440 {
441 if (c == ' ' && !(terminate & TERM_SPACE))
442 return 0;
443 if (c == '\t' && !(terminate & TERM_TAB))
444 return 0;
445
446 return 1;
447 }
448
449 /* remove double slashes to make --index work with such filenames */
450 static char *squash_slash(char *name)
451 {
452 int i = 0, j = 0;
453
454 if (!name)
455 return NULL;
456
457 while (name[i]) {
458 if ((name[j++] = name[i++]) == '/')
459 while (name[i] == '/')
460 i++;
461 }
462 name[j] = '\0';
463 return name;
464 }
465
466 static char *find_name_gnu(struct apply_state *state,
467 const char *line,
468 const char *def,
469 int p_value)
470 {
471 struct strbuf name = STRBUF_INIT;
472 char *cp;
473
474 /*
475 * Proposed "new-style" GNU patch/diff format; see
476 * http://marc.info/?l=git&m=112927316408690&w=2
477 */
478 if (unquote_c_style(&name, line, NULL)) {
479 strbuf_release(&name);
480 return NULL;
481 }
482
483 for (cp = name.buf; p_value; p_value--) {
484 cp = strchr(cp, '/');
485 if (!cp) {
486 strbuf_release(&name);
487 return NULL;
488 }
489 cp++;
490 }
491
492 strbuf_remove(&name, 0, cp - name.buf);
493 if (state->root.len)
494 strbuf_insert(&name, 0, state->root.buf, state->root.len);
495 return squash_slash(strbuf_detach(&name, NULL));
496 }
497
498 static size_t sane_tz_len(const char *line, size_t len)
499 {
500 const char *tz, *p;
501
502 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
503 return 0;
504 tz = line + len - strlen(" +0500");
505
506 if (tz[1] != '+' && tz[1] != '-')
507 return 0;
508
509 for (p = tz + 2; p != line + len; p++)
510 if (!isdigit(*p))
511 return 0;
512
513 return line + len - tz;
514 }
515
516 static size_t tz_with_colon_len(const char *line, size_t len)
517 {
518 const char *tz, *p;
519
520 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
521 return 0;
522 tz = line + len - strlen(" +08:00");
523
524 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
525 return 0;
526 p = tz + 2;
527 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
528 !isdigit(*p++) || !isdigit(*p++))
529 return 0;
530
531 return line + len - tz;
532 }
533
534 static size_t date_len(const char *line, size_t len)
535 {
536 const char *date, *p;
537
538 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
539 return 0;
540 p = date = line + len - strlen("72-02-05");
541
542 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
543 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
544 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
545 return 0;
546
547 if (date - line >= strlen("19") &&
548 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
549 date -= strlen("19");
550
551 return line + len - date;
552 }
553
554 static size_t short_time_len(const char *line, size_t len)
555 {
556 const char *time, *p;
557
558 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
559 return 0;
560 p = time = line + len - strlen(" 07:01:32");
561
562 /* Permit 1-digit hours? */
563 if (*p++ != ' ' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
565 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
566 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
567 return 0;
568
569 return line + len - time;
570 }
571
572 static size_t fractional_time_len(const char *line, size_t len)
573 {
574 const char *p;
575 size_t n;
576
577 /* Expected format: 19:41:17.620000023 */
578 if (!len || !isdigit(line[len - 1]))
579 return 0;
580 p = line + len - 1;
581
582 /* Fractional seconds. */
583 while (p > line && isdigit(*p))
584 p--;
585 if (*p != '.')
586 return 0;
587
588 /* Hours, minutes, and whole seconds. */
589 n = short_time_len(line, p - line);
590 if (!n)
591 return 0;
592
593 return line + len - p + n;
594 }
595
596 static size_t trailing_spaces_len(const char *line, size_t len)
597 {
598 const char *p;
599
600 /* Expected format: ' ' x (1 or more) */
601 if (!len || line[len - 1] != ' ')
602 return 0;
603
604 p = line + len;
605 while (p != line) {
606 p--;
607 if (*p != ' ')
608 return line + len - (p + 1);
609 }
610
611 /* All spaces! */
612 return len;
613 }
614
615 static size_t diff_timestamp_len(const char *line, size_t len)
616 {
617 const char *end = line + len;
618 size_t n;
619
620 /*
621 * Posix: 2010-07-05 19:41:17
622 * GNU: 2010-07-05 19:41:17.620000023 -0500
623 */
624
625 if (!isdigit(end[-1]))
626 return 0;
627
628 n = sane_tz_len(line, end - line);
629 if (!n)
630 n = tz_with_colon_len(line, end - line);
631 end -= n;
632
633 n = short_time_len(line, end - line);
634 if (!n)
635 n = fractional_time_len(line, end - line);
636 end -= n;
637
638 n = date_len(line, end - line);
639 if (!n) /* No date. Too bad. */
640 return 0;
641 end -= n;
642
643 if (end == line) /* No space before date. */
644 return 0;
645 if (end[-1] == '\t') { /* Success! */
646 end--;
647 return line + len - end;
648 }
649 if (end[-1] != ' ') /* No space before date. */
650 return 0;
651
652 /* Whitespace damage. */
653 end -= trailing_spaces_len(line, end - line);
654 return line + len - end;
655 }
656
657 static char *find_name_common(struct apply_state *state,
658 const char *line,
659 const char *def,
660 int p_value,
661 const char *end,
662 int terminate)
663 {
664 int len;
665 const char *start = NULL;
666
667 if (p_value == 0)
668 start = line;
669 while (line != end) {
670 char c = *line;
671
672 if (!end && isspace(c)) {
673 if (c == '\n')
674 break;
675 if (name_terminate(c, terminate))
676 break;
677 }
678 line++;
679 if (c == '/' && !--p_value)
680 start = line;
681 }
682 if (!start)
683 return squash_slash(xstrdup_or_null(def));
684 len = line - start;
685 if (!len)
686 return squash_slash(xstrdup_or_null(def));
687
688 /*
689 * Generally we prefer the shorter name, especially
690 * if the other one is just a variation of that with
691 * something else tacked on to the end (ie "file.orig"
692 * or "file~").
693 */
694 if (def) {
695 int deflen = strlen(def);
696 if (deflen < len && !strncmp(start, def, deflen))
697 return squash_slash(xstrdup(def));
698 }
699
700 if (state->root.len) {
701 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
702 return squash_slash(ret);
703 }
704
705 return squash_slash(xmemdupz(start, len));
706 }
707
708 static char *find_name(struct apply_state *state,
709 const char *line,
710 char *def,
711 int p_value,
712 int terminate)
713 {
714 if (*line == '"') {
715 char *name = find_name_gnu(state, line, def, p_value);
716 if (name)
717 return name;
718 }
719
720 return find_name_common(state, line, def, p_value, NULL, terminate);
721 }
722
723 static char *find_name_traditional(struct apply_state *state,
724 const char *line,
725 char *def,
726 int p_value)
727 {
728 size_t len;
729 size_t date_len;
730
731 if (*line == '"') {
732 char *name = find_name_gnu(state, line, def, p_value);
733 if (name)
734 return name;
735 }
736
737 len = strchrnul(line, '\n') - line;
738 date_len = diff_timestamp_len(line, len);
739 if (!date_len)
740 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
741 len -= date_len;
742
743 return find_name_common(state, line, def, p_value, line + len, 0);
744 }
745
746 static int count_slashes(const char *cp)
747 {
748 int cnt = 0;
749 char ch;
750
751 while ((ch = *cp++))
752 if (ch == '/')
753 cnt++;
754 return cnt;
755 }
756
757 /*
758 * Given the string after "--- " or "+++ ", guess the appropriate
759 * p_value for the given patch.
760 */
761 static int guess_p_value(struct apply_state *state, const char *nameline)
762 {
763 char *name, *cp;
764 int val = -1;
765
766 if (is_dev_null(nameline))
767 return -1;
768 name = find_name_traditional(state, nameline, NULL, 0);
769 if (!name)
770 return -1;
771 cp = strchr(name, '/');
772 if (!cp)
773 val = 0;
774 else if (state->prefix) {
775 /*
776 * Does it begin with "a/$our-prefix" and such? Then this is
777 * very likely to apply to our directory.
778 */
779 if (!strncmp(name, state->prefix, state->prefix_length))
780 val = count_slashes(state->prefix);
781 else {
782 cp++;
783 if (!strncmp(cp, state->prefix, state->prefix_length))
784 val = count_slashes(state->prefix) + 1;
785 }
786 }
787 free(name);
788 return val;
789 }
790
791 /*
792 * Does the ---/+++ line have the POSIX timestamp after the last HT?
793 * GNU diff puts epoch there to signal a creation/deletion event. Is
794 * this such a timestamp?
795 */
796 static int has_epoch_timestamp(const char *nameline)
797 {
798 /*
799 * We are only interested in epoch timestamp; any non-zero
800 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
801 * For the same reason, the date must be either 1969-12-31 or
802 * 1970-01-01, and the seconds part must be "00".
803 */
804 const char stamp_regexp[] =
805 "^(1969-12-31|1970-01-01)"
806 " "
807 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
808 " "
809 "([-+][0-2][0-9]:?[0-5][0-9])\n";
810 const char *timestamp = NULL, *cp, *colon;
811 static regex_t *stamp;
812 regmatch_t m[10];
813 int zoneoffset;
814 int hourminute;
815 int status;
816
817 for (cp = nameline; *cp != '\n'; cp++) {
818 if (*cp == '\t')
819 timestamp = cp + 1;
820 }
821 if (!timestamp)
822 return 0;
823 if (!stamp) {
824 stamp = xmalloc(sizeof(*stamp));
825 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
826 warning(_("Cannot prepare timestamp regexp %s"),
827 stamp_regexp);
828 return 0;
829 }
830 }
831
832 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
833 if (status) {
834 if (status != REG_NOMATCH)
835 warning(_("regexec returned %d for input: %s"),
836 status, timestamp);
837 return 0;
838 }
839
840 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
841 if (*colon == ':')
842 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
843 else
844 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
845 if (timestamp[m[3].rm_so] == '-')
846 zoneoffset = -zoneoffset;
847
848 /*
849 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
850 * (west of GMT) or 1970-01-01 (east of GMT)
851 */
852 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
853 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
854 return 0;
855
856 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
857 strtol(timestamp + 14, NULL, 10) -
858 zoneoffset);
859
860 return ((zoneoffset < 0 && hourminute == 1440) ||
861 (0 <= zoneoffset && !hourminute));
862 }
863
864 /*
865 * Get the name etc info from the ---/+++ lines of a traditional patch header
866 *
867 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
868 * files, we can happily check the index for a match, but for creating a
869 * new file we should try to match whatever "patch" does. I have no idea.
870 */
871 static int parse_traditional_patch(struct apply_state *state,
872 const char *first,
873 const char *second,
874 struct patch *patch)
875 {
876 char *name;
877
878 first += 4; /* skip "--- " */
879 second += 4; /* skip "+++ " */
880 if (!state->p_value_known) {
881 int p, q;
882 p = guess_p_value(state, first);
883 q = guess_p_value(state, second);
884 if (p < 0) p = q;
885 if (0 <= p && p == q) {
886 state->p_value = p;
887 state->p_value_known = 1;
888 }
889 }
890 if (is_dev_null(first)) {
891 patch->is_new = 1;
892 patch->is_delete = 0;
893 name = find_name_traditional(state, second, NULL, state->p_value);
894 patch->new_name = name;
895 } else if (is_dev_null(second)) {
896 patch->is_new = 0;
897 patch->is_delete = 1;
898 name = find_name_traditional(state, first, NULL, state->p_value);
899 patch->old_name = name;
900 } else {
901 char *first_name;
902 first_name = find_name_traditional(state, first, NULL, state->p_value);
903 name = find_name_traditional(state, second, first_name, state->p_value);
904 free(first_name);
905 if (has_epoch_timestamp(first)) {
906 patch->is_new = 1;
907 patch->is_delete = 0;
908 patch->new_name = name;
909 } else if (has_epoch_timestamp(second)) {
910 patch->is_new = 0;
911 patch->is_delete = 1;
912 patch->old_name = name;
913 } else {
914 patch->old_name = name;
915 patch->new_name = xstrdup_or_null(name);
916 }
917 }
918 if (!name)
919 return error(_("unable to find filename in patch at line %d"), state->linenr);
920
921 return 0;
922 }
923
924 static int gitdiff_hdrend(struct apply_state *state,
925 const char *line,
926 struct patch *patch)
927 {
928 return 1;
929 }
930
931 /*
932 * We're anal about diff header consistency, to make
933 * sure that we don't end up having strange ambiguous
934 * patches floating around.
935 *
936 * As a result, gitdiff_{old|new}name() will check
937 * their names against any previous information, just
938 * to make sure..
939 */
940 #define DIFF_OLD_NAME 0
941 #define DIFF_NEW_NAME 1
942
943 static int gitdiff_verify_name(struct apply_state *state,
944 const char *line,
945 int isnull,
946 char **name,
947 int side)
948 {
949 if (!*name && !isnull) {
950 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
951 return 0;
952 }
953
954 if (*name) {
955 int len = strlen(*name);
956 char *another;
957 if (isnull)
958 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
959 *name, state->linenr);
960 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
961 if (!another || memcmp(another, *name, len + 1)) {
962 free(another);
963 return error((side == DIFF_NEW_NAME) ?
964 _("git apply: bad git-diff - inconsistent new filename on line %d") :
965 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
966 }
967 free(another);
968 } else {
969 /* expect "/dev/null" */
970 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
971 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
972 }
973
974 return 0;
975 }
976
977 static int gitdiff_oldname(struct apply_state *state,
978 const char *line,
979 struct patch *patch)
980 {
981 return gitdiff_verify_name(state, line,
982 patch->is_new, &patch->old_name,
983 DIFF_OLD_NAME);
984 }
985
986 static int gitdiff_newname(struct apply_state *state,
987 const char *line,
988 struct patch *patch)
989 {
990 return gitdiff_verify_name(state, line,
991 patch->is_delete, &patch->new_name,
992 DIFF_NEW_NAME);
993 }
994
995 static int gitdiff_oldmode(struct apply_state *state,
996 const char *line,
997 struct patch *patch)
998 {
999 patch->old_mode = strtoul(line, NULL, 8);
1000 return 0;
1001 }
1002
1003 static int gitdiff_newmode(struct apply_state *state,
1004 const char *line,
1005 struct patch *patch)
1006 {
1007 patch->new_mode = strtoul(line, NULL, 8);
1008 return 0;
1009 }
1010
1011 static int gitdiff_delete(struct apply_state *state,
1012 const char *line,
1013 struct patch *patch)
1014 {
1015 patch->is_delete = 1;
1016 free(patch->old_name);
1017 patch->old_name = xstrdup_or_null(patch->def_name);
1018 return gitdiff_oldmode(state, line, patch);
1019 }
1020
1021 static int gitdiff_newfile(struct apply_state *state,
1022 const char *line,
1023 struct patch *patch)
1024 {
1025 patch->is_new = 1;
1026 free(patch->new_name);
1027 patch->new_name = xstrdup_or_null(patch->def_name);
1028 return gitdiff_newmode(state, line, patch);
1029 }
1030
1031 static int gitdiff_copysrc(struct apply_state *state,
1032 const char *line,
1033 struct patch *patch)
1034 {
1035 patch->is_copy = 1;
1036 free(patch->old_name);
1037 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1038 return 0;
1039 }
1040
1041 static int gitdiff_copydst(struct apply_state *state,
1042 const char *line,
1043 struct patch *patch)
1044 {
1045 patch->is_copy = 1;
1046 free(patch->new_name);
1047 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1048 return 0;
1049 }
1050
1051 static int gitdiff_renamesrc(struct apply_state *state,
1052 const char *line,
1053 struct patch *patch)
1054 {
1055 patch->is_rename = 1;
1056 free(patch->old_name);
1057 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1058 return 0;
1059 }
1060
1061 static int gitdiff_renamedst(struct apply_state *state,
1062 const char *line,
1063 struct patch *patch)
1064 {
1065 patch->is_rename = 1;
1066 free(patch->new_name);
1067 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1068 return 0;
1069 }
1070
1071 static int gitdiff_similarity(struct apply_state *state,
1072 const char *line,
1073 struct patch *patch)
1074 {
1075 unsigned long val = strtoul(line, NULL, 10);
1076 if (val <= 100)
1077 patch->score = val;
1078 return 0;
1079 }
1080
1081 static int gitdiff_dissimilarity(struct apply_state *state,
1082 const char *line,
1083 struct patch *patch)
1084 {
1085 unsigned long val = strtoul(line, NULL, 10);
1086 if (val <= 100)
1087 patch->score = val;
1088 return 0;
1089 }
1090
1091 static int gitdiff_index(struct apply_state *state,
1092 const char *line,
1093 struct patch *patch)
1094 {
1095 /*
1096 * index line is N hexadecimal, "..", N hexadecimal,
1097 * and optional space with octal mode.
1098 */
1099 const char *ptr, *eol;
1100 int len;
1101
1102 ptr = strchr(line, '.');
1103 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1104 return 0;
1105 len = ptr - line;
1106 memcpy(patch->old_sha1_prefix, line, len);
1107 patch->old_sha1_prefix[len] = 0;
1108
1109 line = ptr + 2;
1110 ptr = strchr(line, ' ');
1111 eol = strchrnul(line, '\n');
1112
1113 if (!ptr || eol < ptr)
1114 ptr = eol;
1115 len = ptr - line;
1116
1117 if (40 < len)
1118 return 0;
1119 memcpy(patch->new_sha1_prefix, line, len);
1120 patch->new_sha1_prefix[len] = 0;
1121 if (*ptr == ' ')
1122 patch->old_mode = strtoul(ptr+1, NULL, 8);
1123 return 0;
1124 }
1125
1126 /*
1127 * This is normal for a diff that doesn't change anything: we'll fall through
1128 * into the next diff. Tell the parser to break out.
1129 */
1130 static int gitdiff_unrecognized(struct apply_state *state,
1131 const char *line,
1132 struct patch *patch)
1133 {
1134 return 1;
1135 }
1136
1137 /*
1138 * Skip p_value leading components from "line"; as we do not accept
1139 * absolute paths, return NULL in that case.
1140 */
1141 static const char *skip_tree_prefix(struct apply_state *state,
1142 const char *line,
1143 int llen)
1144 {
1145 int nslash;
1146 int i;
1147
1148 if (!state->p_value)
1149 return (llen && line[0] == '/') ? NULL : line;
1150
1151 nslash = state->p_value;
1152 for (i = 0; i < llen; i++) {
1153 int ch = line[i];
1154 if (ch == '/' && --nslash <= 0)
1155 return (i == 0) ? NULL : &line[i + 1];
1156 }
1157 return NULL;
1158 }
1159
1160 /*
1161 * This is to extract the same name that appears on "diff --git"
1162 * line. We do not find and return anything if it is a rename
1163 * patch, and it is OK because we will find the name elsewhere.
1164 * We need to reliably find name only when it is mode-change only,
1165 * creation or deletion of an empty file. In any of these cases,
1166 * both sides are the same name under a/ and b/ respectively.
1167 */
1168 static char *git_header_name(struct apply_state *state,
1169 const char *line,
1170 int llen)
1171 {
1172 const char *name;
1173 const char *second = NULL;
1174 size_t len, line_len;
1175
1176 line += strlen("diff --git ");
1177 llen -= strlen("diff --git ");
1178
1179 if (*line == '"') {
1180 const char *cp;
1181 struct strbuf first = STRBUF_INIT;
1182 struct strbuf sp = STRBUF_INIT;
1183
1184 if (unquote_c_style(&first, line, &second))
1185 goto free_and_fail1;
1186
1187 /* strip the a/b prefix including trailing slash */
1188 cp = skip_tree_prefix(state, first.buf, first.len);
1189 if (!cp)
1190 goto free_and_fail1;
1191 strbuf_remove(&first, 0, cp - first.buf);
1192
1193 /*
1194 * second points at one past closing dq of name.
1195 * find the second name.
1196 */
1197 while ((second < line + llen) && isspace(*second))
1198 second++;
1199
1200 if (line + llen <= second)
1201 goto free_and_fail1;
1202 if (*second == '"') {
1203 if (unquote_c_style(&sp, second, NULL))
1204 goto free_and_fail1;
1205 cp = skip_tree_prefix(state, sp.buf, sp.len);
1206 if (!cp)
1207 goto free_and_fail1;
1208 /* They must match, otherwise ignore */
1209 if (strcmp(cp, first.buf))
1210 goto free_and_fail1;
1211 strbuf_release(&sp);
1212 return strbuf_detach(&first, NULL);
1213 }
1214
1215 /* unquoted second */
1216 cp = skip_tree_prefix(state, second, line + llen - second);
1217 if (!cp)
1218 goto free_and_fail1;
1219 if (line + llen - cp != first.len ||
1220 memcmp(first.buf, cp, first.len))
1221 goto free_and_fail1;
1222 return strbuf_detach(&first, NULL);
1223
1224 free_and_fail1:
1225 strbuf_release(&first);
1226 strbuf_release(&sp);
1227 return NULL;
1228 }
1229
1230 /* unquoted first name */
1231 name = skip_tree_prefix(state, line, llen);
1232 if (!name)
1233 return NULL;
1234
1235 /*
1236 * since the first name is unquoted, a dq if exists must be
1237 * the beginning of the second name.
1238 */
1239 for (second = name; second < line + llen; second++) {
1240 if (*second == '"') {
1241 struct strbuf sp = STRBUF_INIT;
1242 const char *np;
1243
1244 if (unquote_c_style(&sp, second, NULL))
1245 goto free_and_fail2;
1246
1247 np = skip_tree_prefix(state, sp.buf, sp.len);
1248 if (!np)
1249 goto free_and_fail2;
1250
1251 len = sp.buf + sp.len - np;
1252 if (len < second - name &&
1253 !strncmp(np, name, len) &&
1254 isspace(name[len])) {
1255 /* Good */
1256 strbuf_remove(&sp, 0, np - sp.buf);
1257 return strbuf_detach(&sp, NULL);
1258 }
1259
1260 free_and_fail2:
1261 strbuf_release(&sp);
1262 return NULL;
1263 }
1264 }
1265
1266 /*
1267 * Accept a name only if it shows up twice, exactly the same
1268 * form.
1269 */
1270 second = strchr(name, '\n');
1271 if (!second)
1272 return NULL;
1273 line_len = second - name;
1274 for (len = 0 ; ; len++) {
1275 switch (name[len]) {
1276 default:
1277 continue;
1278 case '\n':
1279 return NULL;
1280 case '\t': case ' ':
1281 /*
1282 * Is this the separator between the preimage
1283 * and the postimage pathname? Again, we are
1284 * only interested in the case where there is
1285 * no rename, as this is only to set def_name
1286 * and a rename patch has the names elsewhere
1287 * in an unambiguous form.
1288 */
1289 if (!name[len + 1])
1290 return NULL; /* no postimage name */
1291 second = skip_tree_prefix(state, name + len + 1,
1292 line_len - (len + 1));
1293 if (!second)
1294 return NULL;
1295 /*
1296 * Does len bytes starting at "name" and "second"
1297 * (that are separated by one HT or SP we just
1298 * found) exactly match?
1299 */
1300 if (second[len] == '\n' && !strncmp(name, second, len))
1301 return xmemdupz(name, len);
1302 }
1303 }
1304 }
1305
1306 /* Verify that we recognize the lines following a git header */
1307 static int parse_git_header(struct apply_state *state,
1308 const char *line,
1309 int len,
1310 unsigned int size,
1311 struct patch *patch)
1312 {
1313 unsigned long offset;
1314
1315 /* A git diff has explicit new/delete information, so we don't guess */
1316 patch->is_new = 0;
1317 patch->is_delete = 0;
1318
1319 /*
1320 * Some things may not have the old name in the
1321 * rest of the headers anywhere (pure mode changes,
1322 * or removing or adding empty files), so we get
1323 * the default name from the header.
1324 */
1325 patch->def_name = git_header_name(state, line, len);
1326 if (patch->def_name && state->root.len) {
1327 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1328 free(patch->def_name);
1329 patch->def_name = s;
1330 }
1331
1332 line += len;
1333 size -= len;
1334 state->linenr++;
1335 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1336 static const struct opentry {
1337 const char *str;
1338 int (*fn)(struct apply_state *, const char *, struct patch *);
1339 } optable[] = {
1340 { "@@ -", gitdiff_hdrend },
1341 { "--- ", gitdiff_oldname },
1342 { "+++ ", gitdiff_newname },
1343 { "old mode ", gitdiff_oldmode },
1344 { "new mode ", gitdiff_newmode },
1345 { "deleted file mode ", gitdiff_delete },
1346 { "new file mode ", gitdiff_newfile },
1347 { "copy from ", gitdiff_copysrc },
1348 { "copy to ", gitdiff_copydst },
1349 { "rename old ", gitdiff_renamesrc },
1350 { "rename new ", gitdiff_renamedst },
1351 { "rename from ", gitdiff_renamesrc },
1352 { "rename to ", gitdiff_renamedst },
1353 { "similarity index ", gitdiff_similarity },
1354 { "dissimilarity index ", gitdiff_dissimilarity },
1355 { "index ", gitdiff_index },
1356 { "", gitdiff_unrecognized },
1357 };
1358 int i;
1359
1360 len = linelen(line, size);
1361 if (!len || line[len-1] != '\n')
1362 break;
1363 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1364 const struct opentry *p = optable + i;
1365 int oplen = strlen(p->str);
1366 int res;
1367 if (len < oplen || memcmp(p->str, line, oplen))
1368 continue;
1369 res = p->fn(state, line + oplen, patch);
1370 if (res < 0)
1371 return -1;
1372 if (res > 0)
1373 return offset;
1374 break;
1375 }
1376 }
1377
1378 return offset;
1379 }
1380
1381 static int parse_num(const char *line, unsigned long *p)
1382 {
1383 char *ptr;
1384
1385 if (!isdigit(*line))
1386 return 0;
1387 *p = strtoul(line, &ptr, 10);
1388 return ptr - line;
1389 }
1390
1391 static int parse_range(const char *line, int len, int offset, const char *expect,
1392 unsigned long *p1, unsigned long *p2)
1393 {
1394 int digits, ex;
1395
1396 if (offset < 0 || offset >= len)
1397 return -1;
1398 line += offset;
1399 len -= offset;
1400
1401 digits = parse_num(line, p1);
1402 if (!digits)
1403 return -1;
1404
1405 offset += digits;
1406 line += digits;
1407 len -= digits;
1408
1409 *p2 = 1;
1410 if (*line == ',') {
1411 digits = parse_num(line+1, p2);
1412 if (!digits)
1413 return -1;
1414
1415 offset += digits+1;
1416 line += digits+1;
1417 len -= digits+1;
1418 }
1419
1420 ex = strlen(expect);
1421 if (ex > len)
1422 return -1;
1423 if (memcmp(line, expect, ex))
1424 return -1;
1425
1426 return offset + ex;
1427 }
1428
1429 static void recount_diff(const char *line, int size, struct fragment *fragment)
1430 {
1431 int oldlines = 0, newlines = 0, ret = 0;
1432
1433 if (size < 1) {
1434 warning("recount: ignore empty hunk");
1435 return;
1436 }
1437
1438 for (;;) {
1439 int len = linelen(line, size);
1440 size -= len;
1441 line += len;
1442
1443 if (size < 1)
1444 break;
1445
1446 switch (*line) {
1447 case ' ': case '\n':
1448 newlines++;
1449 /* fall through */
1450 case '-':
1451 oldlines++;
1452 continue;
1453 case '+':
1454 newlines++;
1455 continue;
1456 case '\\':
1457 continue;
1458 case '@':
1459 ret = size < 3 || !starts_with(line, "@@ ");
1460 break;
1461 case 'd':
1462 ret = size < 5 || !starts_with(line, "diff ");
1463 break;
1464 default:
1465 ret = -1;
1466 break;
1467 }
1468 if (ret) {
1469 warning(_("recount: unexpected line: %.*s"),
1470 (int)linelen(line, size), line);
1471 return;
1472 }
1473 break;
1474 }
1475 fragment->oldlines = oldlines;
1476 fragment->newlines = newlines;
1477 }
1478
1479 /*
1480 * Parse a unified diff fragment header of the
1481 * form "@@ -a,b +c,d @@"
1482 */
1483 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1484 {
1485 int offset;
1486
1487 if (!len || line[len-1] != '\n')
1488 return -1;
1489
1490 /* Figure out the number of lines in a fragment */
1491 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1492 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1493
1494 return offset;
1495 }
1496
1497 /*
1498 * Find file diff header
1499 *
1500 * Returns:
1501 * -1 if no header was found
1502 * -128 in case of error
1503 * the size of the header in bytes (called "offset") otherwise
1504 */
1505 static int find_header(struct apply_state *state,
1506 const char *line,
1507 unsigned long size,
1508 int *hdrsize,
1509 struct patch *patch)
1510 {
1511 unsigned long offset, len;
1512
1513 patch->is_toplevel_relative = 0;
1514 patch->is_rename = patch->is_copy = 0;
1515 patch->is_new = patch->is_delete = -1;
1516 patch->old_mode = patch->new_mode = 0;
1517 patch->old_name = patch->new_name = NULL;
1518 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1519 unsigned long nextlen;
1520
1521 len = linelen(line, size);
1522 if (!len)
1523 break;
1524
1525 /* Testing this early allows us to take a few shortcuts.. */
1526 if (len < 6)
1527 continue;
1528
1529 /*
1530 * Make sure we don't find any unconnected patch fragments.
1531 * That's a sign that we didn't find a header, and that a
1532 * patch has become corrupted/broken up.
1533 */
1534 if (!memcmp("@@ -", line, 4)) {
1535 struct fragment dummy;
1536 if (parse_fragment_header(line, len, &dummy) < 0)
1537 continue;
1538 error(_("patch fragment without header at line %d: %.*s"),
1539 state->linenr, (int)len-1, line);
1540 return -128;
1541 }
1542
1543 if (size < len + 6)
1544 break;
1545
1546 /*
1547 * Git patch? It might not have a real patch, just a rename
1548 * or mode change, so we handle that specially
1549 */
1550 if (!memcmp("diff --git ", line, 11)) {
1551 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1552 if (git_hdr_len < 0)
1553 return -128;
1554 if (git_hdr_len <= len)
1555 continue;
1556 if (!patch->old_name && !patch->new_name) {
1557 if (!patch->def_name) {
1558 error(Q_("git diff header lacks filename information when removing "
1559 "%d leading pathname component (line %d)",
1560 "git diff header lacks filename information when removing "
1561 "%d leading pathname components (line %d)",
1562 state->p_value),
1563 state->p_value, state->linenr);
1564 return -128;
1565 }
1566 patch->old_name = xstrdup(patch->def_name);
1567 patch->new_name = xstrdup(patch->def_name);
1568 }
1569 if (!patch->is_delete && !patch->new_name) {
1570 error(_("git diff header lacks filename information "
1571 "(line %d)"), state->linenr);
1572 return -128;
1573 }
1574 patch->is_toplevel_relative = 1;
1575 *hdrsize = git_hdr_len;
1576 return offset;
1577 }
1578
1579 /* --- followed by +++ ? */
1580 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1581 continue;
1582
1583 /*
1584 * We only accept unified patches, so we want it to
1585 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1586 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1587 */
1588 nextlen = linelen(line + len, size - len);
1589 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1590 continue;
1591
1592 /* Ok, we'll consider it a patch */
1593 if (parse_traditional_patch(state, line, line+len, patch))
1594 return -128;
1595 *hdrsize = len + nextlen;
1596 state->linenr += 2;
1597 return offset;
1598 }
1599 return -1;
1600 }
1601
1602 static void record_ws_error(struct apply_state *state,
1603 unsigned result,
1604 const char *line,
1605 int len,
1606 int linenr)
1607 {
1608 char *err;
1609
1610 if (!result)
1611 return;
1612
1613 state->whitespace_error++;
1614 if (state->squelch_whitespace_errors &&
1615 state->squelch_whitespace_errors < state->whitespace_error)
1616 return;
1617
1618 err = whitespace_error_string(result);
1619 if (state->apply_verbosity > verbosity_silent)
1620 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1621 state->patch_input_file, linenr, err, len, line);
1622 free(err);
1623 }
1624
1625 static void check_whitespace(struct apply_state *state,
1626 const char *line,
1627 int len,
1628 unsigned ws_rule)
1629 {
1630 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1631
1632 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1633 }
1634
1635 /*
1636 * Parse a unified diff. Note that this really needs to parse each
1637 * fragment separately, since the only way to know the difference
1638 * between a "---" that is part of a patch, and a "---" that starts
1639 * the next patch is to look at the line counts..
1640 */
1641 static int parse_fragment(struct apply_state *state,
1642 const char *line,
1643 unsigned long size,
1644 struct patch *patch,
1645 struct fragment *fragment)
1646 {
1647 int added, deleted;
1648 int len = linelen(line, size), offset;
1649 unsigned long oldlines, newlines;
1650 unsigned long leading, trailing;
1651
1652 offset = parse_fragment_header(line, len, fragment);
1653 if (offset < 0)
1654 return -1;
1655 if (offset > 0 && patch->recount)
1656 recount_diff(line + offset, size - offset, fragment);
1657 oldlines = fragment->oldlines;
1658 newlines = fragment->newlines;
1659 leading = 0;
1660 trailing = 0;
1661
1662 /* Parse the thing.. */
1663 line += len;
1664 size -= len;
1665 state->linenr++;
1666 added = deleted = 0;
1667 for (offset = len;
1668 0 < size;
1669 offset += len, size -= len, line += len, state->linenr++) {
1670 if (!oldlines && !newlines)
1671 break;
1672 len = linelen(line, size);
1673 if (!len || line[len-1] != '\n')
1674 return -1;
1675 switch (*line) {
1676 default:
1677 return -1;
1678 case '\n': /* newer GNU diff, an empty context line */
1679 case ' ':
1680 oldlines--;
1681 newlines--;
1682 if (!deleted && !added)
1683 leading++;
1684 trailing++;
1685 if (!state->apply_in_reverse &&
1686 state->ws_error_action == correct_ws_error)
1687 check_whitespace(state, line, len, patch->ws_rule);
1688 break;
1689 case '-':
1690 if (state->apply_in_reverse &&
1691 state->ws_error_action != nowarn_ws_error)
1692 check_whitespace(state, line, len, patch->ws_rule);
1693 deleted++;
1694 oldlines--;
1695 trailing = 0;
1696 break;
1697 case '+':
1698 if (!state->apply_in_reverse &&
1699 state->ws_error_action != nowarn_ws_error)
1700 check_whitespace(state, line, len, patch->ws_rule);
1701 added++;
1702 newlines--;
1703 trailing = 0;
1704 break;
1705
1706 /*
1707 * We allow "\ No newline at end of file". Depending
1708 * on locale settings when the patch was produced we
1709 * don't know what this line looks like. The only
1710 * thing we do know is that it begins with "\ ".
1711 * Checking for 12 is just for sanity check -- any
1712 * l10n of "\ No newline..." is at least that long.
1713 */
1714 case '\\':
1715 if (len < 12 || memcmp(line, "\\ ", 2))
1716 return -1;
1717 break;
1718 }
1719 }
1720 if (oldlines || newlines)
1721 return -1;
1722 if (!deleted && !added)
1723 return -1;
1724
1725 fragment->leading = leading;
1726 fragment->trailing = trailing;
1727
1728 /*
1729 * If a fragment ends with an incomplete line, we failed to include
1730 * it in the above loop because we hit oldlines == newlines == 0
1731 * before seeing it.
1732 */
1733 if (12 < size && !memcmp(line, "\\ ", 2))
1734 offset += linelen(line, size);
1735
1736 patch->lines_added += added;
1737 patch->lines_deleted += deleted;
1738
1739 if (0 < patch->is_new && oldlines)
1740 return error(_("new file depends on old contents"));
1741 if (0 < patch->is_delete && newlines)
1742 return error(_("deleted file still has contents"));
1743 return offset;
1744 }
1745
1746 /*
1747 * We have seen "diff --git a/... b/..." header (or a traditional patch
1748 * header). Read hunks that belong to this patch into fragments and hang
1749 * them to the given patch structure.
1750 *
1751 * The (fragment->patch, fragment->size) pair points into the memory given
1752 * by the caller, not a copy, when we return.
1753 *
1754 * Returns:
1755 * -1 in case of error,
1756 * the number of bytes in the patch otherwise.
1757 */
1758 static int parse_single_patch(struct apply_state *state,
1759 const char *line,
1760 unsigned long size,
1761 struct patch *patch)
1762 {
1763 unsigned long offset = 0;
1764 unsigned long oldlines = 0, newlines = 0, context = 0;
1765 struct fragment **fragp = &patch->fragments;
1766
1767 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1768 struct fragment *fragment;
1769 int len;
1770
1771 fragment = xcalloc(1, sizeof(*fragment));
1772 fragment->linenr = state->linenr;
1773 len = parse_fragment(state, line, size, patch, fragment);
1774 if (len <= 0) {
1775 free(fragment);
1776 return error(_("corrupt patch at line %d"), state->linenr);
1777 }
1778 fragment->patch = line;
1779 fragment->size = len;
1780 oldlines += fragment->oldlines;
1781 newlines += fragment->newlines;
1782 context += fragment->leading + fragment->trailing;
1783
1784 *fragp = fragment;
1785 fragp = &fragment->next;
1786
1787 offset += len;
1788 line += len;
1789 size -= len;
1790 }
1791
1792 /*
1793 * If something was removed (i.e. we have old-lines) it cannot
1794 * be creation, and if something was added it cannot be
1795 * deletion. However, the reverse is not true; --unified=0
1796 * patches that only add are not necessarily creation even
1797 * though they do not have any old lines, and ones that only
1798 * delete are not necessarily deletion.
1799 *
1800 * Unfortunately, a real creation/deletion patch do _not_ have
1801 * any context line by definition, so we cannot safely tell it
1802 * apart with --unified=0 insanity. At least if the patch has
1803 * more than one hunk it is not creation or deletion.
1804 */
1805 if (patch->is_new < 0 &&
1806 (oldlines || (patch->fragments && patch->fragments->next)))
1807 patch->is_new = 0;
1808 if (patch->is_delete < 0 &&
1809 (newlines || (patch->fragments && patch->fragments->next)))
1810 patch->is_delete = 0;
1811
1812 if (0 < patch->is_new && oldlines)
1813 return error(_("new file %s depends on old contents"), patch->new_name);
1814 if (0 < patch->is_delete && newlines)
1815 return error(_("deleted file %s still has contents"), patch->old_name);
1816 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1817 fprintf_ln(stderr,
1818 _("** warning: "
1819 "file %s becomes empty but is not deleted"),
1820 patch->new_name);
1821
1822 return offset;
1823 }
1824
1825 static inline int metadata_changes(struct patch *patch)
1826 {
1827 return patch->is_rename > 0 ||
1828 patch->is_copy > 0 ||
1829 patch->is_new > 0 ||
1830 patch->is_delete ||
1831 (patch->old_mode && patch->new_mode &&
1832 patch->old_mode != patch->new_mode);
1833 }
1834
1835 static char *inflate_it(const void *data, unsigned long size,
1836 unsigned long inflated_size)
1837 {
1838 git_zstream stream;
1839 void *out;
1840 int st;
1841
1842 memset(&stream, 0, sizeof(stream));
1843
1844 stream.next_in = (unsigned char *)data;
1845 stream.avail_in = size;
1846 stream.next_out = out = xmalloc(inflated_size);
1847 stream.avail_out = inflated_size;
1848 git_inflate_init(&stream);
1849 st = git_inflate(&stream, Z_FINISH);
1850 git_inflate_end(&stream);
1851 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1852 free(out);
1853 return NULL;
1854 }
1855 return out;
1856 }
1857
1858 /*
1859 * Read a binary hunk and return a new fragment; fragment->patch
1860 * points at an allocated memory that the caller must free, so
1861 * it is marked as "->free_patch = 1".
1862 */
1863 static struct fragment *parse_binary_hunk(struct apply_state *state,
1864 char **buf_p,
1865 unsigned long *sz_p,
1866 int *status_p,
1867 int *used_p)
1868 {
1869 /*
1870 * Expect a line that begins with binary patch method ("literal"
1871 * or "delta"), followed by the length of data before deflating.
1872 * a sequence of 'length-byte' followed by base-85 encoded data
1873 * should follow, terminated by a newline.
1874 *
1875 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1876 * and we would limit the patch line to 66 characters,
1877 * so one line can fit up to 13 groups that would decode
1878 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1879 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1880 */
1881 int llen, used;
1882 unsigned long size = *sz_p;
1883 char *buffer = *buf_p;
1884 int patch_method;
1885 unsigned long origlen;
1886 char *data = NULL;
1887 int hunk_size = 0;
1888 struct fragment *frag;
1889
1890 llen = linelen(buffer, size);
1891 used = llen;
1892
1893 *status_p = 0;
1894
1895 if (starts_with(buffer, "delta ")) {
1896 patch_method = BINARY_DELTA_DEFLATED;
1897 origlen = strtoul(buffer + 6, NULL, 10);
1898 }
1899 else if (starts_with(buffer, "literal ")) {
1900 patch_method = BINARY_LITERAL_DEFLATED;
1901 origlen = strtoul(buffer + 8, NULL, 10);
1902 }
1903 else
1904 return NULL;
1905
1906 state->linenr++;
1907 buffer += llen;
1908 while (1) {
1909 int byte_length, max_byte_length, newsize;
1910 llen = linelen(buffer, size);
1911 used += llen;
1912 state->linenr++;
1913 if (llen == 1) {
1914 /* consume the blank line */
1915 buffer++;
1916 size--;
1917 break;
1918 }
1919 /*
1920 * Minimum line is "A00000\n" which is 7-byte long,
1921 * and the line length must be multiple of 5 plus 2.
1922 */
1923 if ((llen < 7) || (llen-2) % 5)
1924 goto corrupt;
1925 max_byte_length = (llen - 2) / 5 * 4;
1926 byte_length = *buffer;
1927 if ('A' <= byte_length && byte_length <= 'Z')
1928 byte_length = byte_length - 'A' + 1;
1929 else if ('a' <= byte_length && byte_length <= 'z')
1930 byte_length = byte_length - 'a' + 27;
1931 else
1932 goto corrupt;
1933 /* if the input length was not multiple of 4, we would
1934 * have filler at the end but the filler should never
1935 * exceed 3 bytes
1936 */
1937 if (max_byte_length < byte_length ||
1938 byte_length <= max_byte_length - 4)
1939 goto corrupt;
1940 newsize = hunk_size + byte_length;
1941 data = xrealloc(data, newsize);
1942 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1943 goto corrupt;
1944 hunk_size = newsize;
1945 buffer += llen;
1946 size -= llen;
1947 }
1948
1949 frag = xcalloc(1, sizeof(*frag));
1950 frag->patch = inflate_it(data, hunk_size, origlen);
1951 frag->free_patch = 1;
1952 if (!frag->patch)
1953 goto corrupt;
1954 free(data);
1955 frag->size = origlen;
1956 *buf_p = buffer;
1957 *sz_p = size;
1958 *used_p = used;
1959 frag->binary_patch_method = patch_method;
1960 return frag;
1961
1962 corrupt:
1963 free(data);
1964 *status_p = -1;
1965 error(_("corrupt binary patch at line %d: %.*s"),
1966 state->linenr-1, llen-1, buffer);
1967 return NULL;
1968 }
1969
1970 /*
1971 * Returns:
1972 * -1 in case of error,
1973 * the length of the parsed binary patch otherwise
1974 */
1975 static int parse_binary(struct apply_state *state,
1976 char *buffer,
1977 unsigned long size,
1978 struct patch *patch)
1979 {
1980 /*
1981 * We have read "GIT binary patch\n"; what follows is a line
1982 * that says the patch method (currently, either "literal" or
1983 * "delta") and the length of data before deflating; a
1984 * sequence of 'length-byte' followed by base-85 encoded data
1985 * follows.
1986 *
1987 * When a binary patch is reversible, there is another binary
1988 * hunk in the same format, starting with patch method (either
1989 * "literal" or "delta") with the length of data, and a sequence
1990 * of length-byte + base-85 encoded data, terminated with another
1991 * empty line. This data, when applied to the postimage, produces
1992 * the preimage.
1993 */
1994 struct fragment *forward;
1995 struct fragment *reverse;
1996 int status;
1997 int used, used_1;
1998
1999 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2000 if (!forward && !status)
2001 /* there has to be one hunk (forward hunk) */
2002 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2003 if (status)
2004 /* otherwise we already gave an error message */
2005 return status;
2006
2007 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2008 if (reverse)
2009 used += used_1;
2010 else if (status) {
2011 /*
2012 * Not having reverse hunk is not an error, but having
2013 * a corrupt reverse hunk is.
2014 */
2015 free((void*) forward->patch);
2016 free(forward);
2017 return status;
2018 }
2019 forward->next = reverse;
2020 patch->fragments = forward;
2021 patch->is_binary = 1;
2022 return used;
2023 }
2024
2025 static void prefix_one(struct apply_state *state, char **name)
2026 {
2027 char *old_name = *name;
2028 if (!old_name)
2029 return;
2030 *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));
2031 free(old_name);
2032 }
2033
2034 static void prefix_patch(struct apply_state *state, struct patch *p)
2035 {
2036 if (!state->prefix || p->is_toplevel_relative)
2037 return;
2038 prefix_one(state, &p->new_name);
2039 prefix_one(state, &p->old_name);
2040 }
2041
2042 /*
2043 * include/exclude
2044 */
2045
2046 static void add_name_limit(struct apply_state *state,
2047 const char *name,
2048 int exclude)
2049 {
2050 struct string_list_item *it;
2051
2052 it = string_list_append(&state->limit_by_name, name);
2053 it->util = exclude ? NULL : (void *) 1;
2054 }
2055
2056 static int use_patch(struct apply_state *state, struct patch *p)
2057 {
2058 const char *pathname = p->new_name ? p->new_name : p->old_name;
2059 int i;
2060
2061 /* Paths outside are not touched regardless of "--include" */
2062 if (0 < state->prefix_length) {
2063 int pathlen = strlen(pathname);
2064 if (pathlen <= state->prefix_length ||
2065 memcmp(state->prefix, pathname, state->prefix_length))
2066 return 0;
2067 }
2068
2069 /* See if it matches any of exclude/include rule */
2070 for (i = 0; i < state->limit_by_name.nr; i++) {
2071 struct string_list_item *it = &state->limit_by_name.items[i];
2072 if (!wildmatch(it->string, pathname, 0, NULL))
2073 return (it->util != NULL);
2074 }
2075
2076 /*
2077 * If we had any include, a path that does not match any rule is
2078 * not used. Otherwise, we saw bunch of exclude rules (or none)
2079 * and such a path is used.
2080 */
2081 return !state->has_include;
2082 }
2083
2084 /*
2085 * Read the patch text in "buffer" that extends for "size" bytes; stop
2086 * reading after seeing a single patch (i.e. changes to a single file).
2087 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2088 *
2089 * Returns:
2090 * -1 if no header was found or parse_binary() failed,
2091 * -128 on another error,
2092 * the number of bytes consumed otherwise,
2093 * so that the caller can call us again for the next patch.
2094 */
2095 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2096 {
2097 int hdrsize, patchsize;
2098 int offset = find_header(state, buffer, size, &hdrsize, patch);
2099
2100 if (offset < 0)
2101 return offset;
2102
2103 prefix_patch(state, patch);
2104
2105 if (!use_patch(state, patch))
2106 patch->ws_rule = 0;
2107 else
2108 patch->ws_rule = whitespace_rule(patch->new_name
2109 ? patch->new_name
2110 : patch->old_name);
2111
2112 patchsize = parse_single_patch(state,
2113 buffer + offset + hdrsize,
2114 size - offset - hdrsize,
2115 patch);
2116
2117 if (patchsize < 0)
2118 return -128;
2119
2120 if (!patchsize) {
2121 static const char git_binary[] = "GIT binary patch\n";
2122 int hd = hdrsize + offset;
2123 unsigned long llen = linelen(buffer + hd, size - hd);
2124
2125 if (llen == sizeof(git_binary) - 1 &&
2126 !memcmp(git_binary, buffer + hd, llen)) {
2127 int used;
2128 state->linenr++;
2129 used = parse_binary(state, buffer + hd + llen,
2130 size - hd - llen, patch);
2131 if (used < 0)
2132 return -1;
2133 if (used)
2134 patchsize = used + llen;
2135 else
2136 patchsize = 0;
2137 }
2138 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2139 static const char *binhdr[] = {
2140 "Binary files ",
2141 "Files ",
2142 NULL,
2143 };
2144 int i;
2145 for (i = 0; binhdr[i]; i++) {
2146 int len = strlen(binhdr[i]);
2147 if (len < size - hd &&
2148 !memcmp(binhdr[i], buffer + hd, len)) {
2149 state->linenr++;
2150 patch->is_binary = 1;
2151 patchsize = llen;
2152 break;
2153 }
2154 }
2155 }
2156
2157 /* Empty patch cannot be applied if it is a text patch
2158 * without metadata change. A binary patch appears
2159 * empty to us here.
2160 */
2161 if ((state->apply || state->check) &&
2162 (!patch->is_binary && !metadata_changes(patch))) {
2163 error(_("patch with only garbage at line %d"), state->linenr);
2164 return -128;
2165 }
2166 }
2167
2168 return offset + hdrsize + patchsize;
2169 }
2170
2171 #define swap(a,b) myswap((a),(b),sizeof(a))
2172
2173 #define myswap(a, b, size) do { \
2174 unsigned char mytmp[size]; \
2175 memcpy(mytmp, &a, size); \
2176 memcpy(&a, &b, size); \
2177 memcpy(&b, mytmp, size); \
2178 } while (0)
2179
2180 static void reverse_patches(struct patch *p)
2181 {
2182 for (; p; p = p->next) {
2183 struct fragment *frag = p->fragments;
2184
2185 swap(p->new_name, p->old_name);
2186 swap(p->new_mode, p->old_mode);
2187 swap(p->is_new, p->is_delete);
2188 swap(p->lines_added, p->lines_deleted);
2189 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2190
2191 for (; frag; frag = frag->next) {
2192 swap(frag->newpos, frag->oldpos);
2193 swap(frag->newlines, frag->oldlines);
2194 }
2195 }
2196 }
2197
2198 static const char pluses[] =
2199 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2200 static const char minuses[]=
2201 "----------------------------------------------------------------------";
2202
2203 static void show_stats(struct apply_state *state, struct patch *patch)
2204 {
2205 struct strbuf qname = STRBUF_INIT;
2206 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2207 int max, add, del;
2208
2209 quote_c_style(cp, &qname, NULL, 0);
2210
2211 /*
2212 * "scale" the filename
2213 */
2214 max = state->max_len;
2215 if (max > 50)
2216 max = 50;
2217
2218 if (qname.len > max) {
2219 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2220 if (!cp)
2221 cp = qname.buf + qname.len + 3 - max;
2222 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2223 }
2224
2225 if (patch->is_binary) {
2226 printf(" %-*s | Bin\n", max, qname.buf);
2227 strbuf_release(&qname);
2228 return;
2229 }
2230
2231 printf(" %-*s |", max, qname.buf);
2232 strbuf_release(&qname);
2233
2234 /*
2235 * scale the add/delete
2236 */
2237 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2238 add = patch->lines_added;
2239 del = patch->lines_deleted;
2240
2241 if (state->max_change > 0) {
2242 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2243 add = (add * max + state->max_change / 2) / state->max_change;
2244 del = total - add;
2245 }
2246 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2247 add, pluses, del, minuses);
2248 }
2249
2250 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2251 {
2252 switch (st->st_mode & S_IFMT) {
2253 case S_IFLNK:
2254 if (strbuf_readlink(buf, path, st->st_size) < 0)
2255 return error(_("unable to read symlink %s"), path);
2256 return 0;
2257 case S_IFREG:
2258 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2259 return error(_("unable to open or read %s"), path);
2260 convert_to_git(path, buf->buf, buf->len, buf, 0);
2261 return 0;
2262 default:
2263 return -1;
2264 }
2265 }
2266
2267 /*
2268 * Update the preimage, and the common lines in postimage,
2269 * from buffer buf of length len. If postlen is 0 the postimage
2270 * is updated in place, otherwise it's updated on a new buffer
2271 * of length postlen
2272 */
2273
2274 static void update_pre_post_images(struct image *preimage,
2275 struct image *postimage,
2276 char *buf,
2277 size_t len, size_t postlen)
2278 {
2279 int i, ctx, reduced;
2280 char *new, *old, *fixed;
2281 struct image fixed_preimage;
2282
2283 /*
2284 * Update the preimage with whitespace fixes. Note that we
2285 * are not losing preimage->buf -- apply_one_fragment() will
2286 * free "oldlines".
2287 */
2288 prepare_image(&fixed_preimage, buf, len, 1);
2289 assert(postlen
2290 ? fixed_preimage.nr == preimage->nr
2291 : fixed_preimage.nr <= preimage->nr);
2292 for (i = 0; i < fixed_preimage.nr; i++)
2293 fixed_preimage.line[i].flag = preimage->line[i].flag;
2294 free(preimage->line_allocated);
2295 *preimage = fixed_preimage;
2296
2297 /*
2298 * Adjust the common context lines in postimage. This can be
2299 * done in-place when we are shrinking it with whitespace
2300 * fixing, but needs a new buffer when ignoring whitespace or
2301 * expanding leading tabs to spaces.
2302 *
2303 * We trust the caller to tell us if the update can be done
2304 * in place (postlen==0) or not.
2305 */
2306 old = postimage->buf;
2307 if (postlen)
2308 new = postimage->buf = xmalloc(postlen);
2309 else
2310 new = old;
2311 fixed = preimage->buf;
2312
2313 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2314 size_t l_len = postimage->line[i].len;
2315 if (!(postimage->line[i].flag & LINE_COMMON)) {
2316 /* an added line -- no counterparts in preimage */
2317 memmove(new, old, l_len);
2318 old += l_len;
2319 new += l_len;
2320 continue;
2321 }
2322
2323 /* a common context -- skip it in the original postimage */
2324 old += l_len;
2325
2326 /* and find the corresponding one in the fixed preimage */
2327 while (ctx < preimage->nr &&
2328 !(preimage->line[ctx].flag & LINE_COMMON)) {
2329 fixed += preimage->line[ctx].len;
2330 ctx++;
2331 }
2332
2333 /*
2334 * preimage is expected to run out, if the caller
2335 * fixed addition of trailing blank lines.
2336 */
2337 if (preimage->nr <= ctx) {
2338 reduced++;
2339 continue;
2340 }
2341
2342 /* and copy it in, while fixing the line length */
2343 l_len = preimage->line[ctx].len;
2344 memcpy(new, fixed, l_len);
2345 new += l_len;
2346 fixed += l_len;
2347 postimage->line[i].len = l_len;
2348 ctx++;
2349 }
2350
2351 if (postlen
2352 ? postlen < new - postimage->buf
2353 : postimage->len < new - postimage->buf)
2354 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2355 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2356
2357 /* Fix the length of the whole thing */
2358 postimage->len = new - postimage->buf;
2359 postimage->nr -= reduced;
2360 }
2361
2362 static int line_by_line_fuzzy_match(struct image *img,
2363 struct image *preimage,
2364 struct image *postimage,
2365 unsigned long try,
2366 int try_lno,
2367 int preimage_limit)
2368 {
2369 int i;
2370 size_t imgoff = 0;
2371 size_t preoff = 0;
2372 size_t postlen = postimage->len;
2373 size_t extra_chars;
2374 char *buf;
2375 char *preimage_eof;
2376 char *preimage_end;
2377 struct strbuf fixed;
2378 char *fixed_buf;
2379 size_t fixed_len;
2380
2381 for (i = 0; i < preimage_limit; i++) {
2382 size_t prelen = preimage->line[i].len;
2383 size_t imglen = img->line[try_lno+i].len;
2384
2385 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2386 preimage->buf + preoff, prelen))
2387 return 0;
2388 if (preimage->line[i].flag & LINE_COMMON)
2389 postlen += imglen - prelen;
2390 imgoff += imglen;
2391 preoff += prelen;
2392 }
2393
2394 /*
2395 * Ok, the preimage matches with whitespace fuzz.
2396 *
2397 * imgoff now holds the true length of the target that
2398 * matches the preimage before the end of the file.
2399 *
2400 * Count the number of characters in the preimage that fall
2401 * beyond the end of the file and make sure that all of them
2402 * are whitespace characters. (This can only happen if
2403 * we are removing blank lines at the end of the file.)
2404 */
2405 buf = preimage_eof = preimage->buf + preoff;
2406 for ( ; i < preimage->nr; i++)
2407 preoff += preimage->line[i].len;
2408 preimage_end = preimage->buf + preoff;
2409 for ( ; buf < preimage_end; buf++)
2410 if (!isspace(*buf))
2411 return 0;
2412
2413 /*
2414 * Update the preimage and the common postimage context
2415 * lines to use the same whitespace as the target.
2416 * If whitespace is missing in the target (i.e.
2417 * if the preimage extends beyond the end of the file),
2418 * use the whitespace from the preimage.
2419 */
2420 extra_chars = preimage_end - preimage_eof;
2421 strbuf_init(&fixed, imgoff + extra_chars);
2422 strbuf_add(&fixed, img->buf + try, imgoff);
2423 strbuf_add(&fixed, preimage_eof, extra_chars);
2424 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2425 update_pre_post_images(preimage, postimage,
2426 fixed_buf, fixed_len, postlen);
2427 return 1;
2428 }
2429
2430 static int match_fragment(struct apply_state *state,
2431 struct image *img,
2432 struct image *preimage,
2433 struct image *postimage,
2434 unsigned long try,
2435 int try_lno,
2436 unsigned ws_rule,
2437 int match_beginning, int match_end)
2438 {
2439 int i;
2440 char *fixed_buf, *buf, *orig, *target;
2441 struct strbuf fixed;
2442 size_t fixed_len, postlen;
2443 int preimage_limit;
2444
2445 if (preimage->nr + try_lno <= img->nr) {
2446 /*
2447 * The hunk falls within the boundaries of img.
2448 */
2449 preimage_limit = preimage->nr;
2450 if (match_end && (preimage->nr + try_lno != img->nr))
2451 return 0;
2452 } else if (state->ws_error_action == correct_ws_error &&
2453 (ws_rule & WS_BLANK_AT_EOF)) {
2454 /*
2455 * This hunk extends beyond the end of img, and we are
2456 * removing blank lines at the end of the file. This
2457 * many lines from the beginning of the preimage must
2458 * match with img, and the remainder of the preimage
2459 * must be blank.
2460 */
2461 preimage_limit = img->nr - try_lno;
2462 } else {
2463 /*
2464 * The hunk extends beyond the end of the img and
2465 * we are not removing blanks at the end, so we
2466 * should reject the hunk at this position.
2467 */
2468 return 0;
2469 }
2470
2471 if (match_beginning && try_lno)
2472 return 0;
2473
2474 /* Quick hash check */
2475 for (i = 0; i < preimage_limit; i++)
2476 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2477 (preimage->line[i].hash != img->line[try_lno + i].hash))
2478 return 0;
2479
2480 if (preimage_limit == preimage->nr) {
2481 /*
2482 * Do we have an exact match? If we were told to match
2483 * at the end, size must be exactly at try+fragsize,
2484 * otherwise try+fragsize must be still within the preimage,
2485 * and either case, the old piece should match the preimage
2486 * exactly.
2487 */
2488 if ((match_end
2489 ? (try + preimage->len == img->len)
2490 : (try + preimage->len <= img->len)) &&
2491 !memcmp(img->buf + try, preimage->buf, preimage->len))
2492 return 1;
2493 } else {
2494 /*
2495 * The preimage extends beyond the end of img, so
2496 * there cannot be an exact match.
2497 *
2498 * There must be one non-blank context line that match
2499 * a line before the end of img.
2500 */
2501 char *buf_end;
2502
2503 buf = preimage->buf;
2504 buf_end = buf;
2505 for (i = 0; i < preimage_limit; i++)
2506 buf_end += preimage->line[i].len;
2507
2508 for ( ; buf < buf_end; buf++)
2509 if (!isspace(*buf))
2510 break;
2511 if (buf == buf_end)
2512 return 0;
2513 }
2514
2515 /*
2516 * No exact match. If we are ignoring whitespace, run a line-by-line
2517 * fuzzy matching. We collect all the line length information because
2518 * we need it to adjust whitespace if we match.
2519 */
2520 if (state->ws_ignore_action == ignore_ws_change)
2521 return line_by_line_fuzzy_match(img, preimage, postimage,
2522 try, try_lno, preimage_limit);
2523
2524 if (state->ws_error_action != correct_ws_error)
2525 return 0;
2526
2527 /*
2528 * The hunk does not apply byte-by-byte, but the hash says
2529 * it might with whitespace fuzz. We weren't asked to
2530 * ignore whitespace, we were asked to correct whitespace
2531 * errors, so let's try matching after whitespace correction.
2532 *
2533 * While checking the preimage against the target, whitespace
2534 * errors in both fixed, we count how large the corresponding
2535 * postimage needs to be. The postimage prepared by
2536 * apply_one_fragment() has whitespace errors fixed on added
2537 * lines already, but the common lines were propagated as-is,
2538 * which may become longer when their whitespace errors are
2539 * fixed.
2540 */
2541
2542 /* First count added lines in postimage */
2543 postlen = 0;
2544 for (i = 0; i < postimage->nr; i++) {
2545 if (!(postimage->line[i].flag & LINE_COMMON))
2546 postlen += postimage->line[i].len;
2547 }
2548
2549 /*
2550 * The preimage may extend beyond the end of the file,
2551 * but in this loop we will only handle the part of the
2552 * preimage that falls within the file.
2553 */
2554 strbuf_init(&fixed, preimage->len + 1);
2555 orig = preimage->buf;
2556 target = img->buf + try;
2557 for (i = 0; i < preimage_limit; i++) {
2558 size_t oldlen = preimage->line[i].len;
2559 size_t tgtlen = img->line[try_lno + i].len;
2560 size_t fixstart = fixed.len;
2561 struct strbuf tgtfix;
2562 int match;
2563
2564 /* Try fixing the line in the preimage */
2565 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2566
2567 /* Try fixing the line in the target */
2568 strbuf_init(&tgtfix, tgtlen);
2569 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2570
2571 /*
2572 * If they match, either the preimage was based on
2573 * a version before our tree fixed whitespace breakage,
2574 * or we are lacking a whitespace-fix patch the tree
2575 * the preimage was based on already had (i.e. target
2576 * has whitespace breakage, the preimage doesn't).
2577 * In either case, we are fixing the whitespace breakages
2578 * so we might as well take the fix together with their
2579 * real change.
2580 */
2581 match = (tgtfix.len == fixed.len - fixstart &&
2582 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2583 fixed.len - fixstart));
2584
2585 /* Add the length if this is common with the postimage */
2586 if (preimage->line[i].flag & LINE_COMMON)
2587 postlen += tgtfix.len;
2588
2589 strbuf_release(&tgtfix);
2590 if (!match)
2591 goto unmatch_exit;
2592
2593 orig += oldlen;
2594 target += tgtlen;
2595 }
2596
2597
2598 /*
2599 * Now handle the lines in the preimage that falls beyond the
2600 * end of the file (if any). They will only match if they are
2601 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2602 * false).
2603 */
2604 for ( ; i < preimage->nr; i++) {
2605 size_t fixstart = fixed.len; /* start of the fixed preimage */
2606 size_t oldlen = preimage->line[i].len;
2607 int j;
2608
2609 /* Try fixing the line in the preimage */
2610 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2611
2612 for (j = fixstart; j < fixed.len; j++)
2613 if (!isspace(fixed.buf[j]))
2614 goto unmatch_exit;
2615
2616 orig += oldlen;
2617 }
2618
2619 /*
2620 * Yes, the preimage is based on an older version that still
2621 * has whitespace breakages unfixed, and fixing them makes the
2622 * hunk match. Update the context lines in the postimage.
2623 */
2624 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2625 if (postlen < postimage->len)
2626 postlen = 0;
2627 update_pre_post_images(preimage, postimage,
2628 fixed_buf, fixed_len, postlen);
2629 return 1;
2630
2631 unmatch_exit:
2632 strbuf_release(&fixed);
2633 return 0;
2634 }
2635
2636 static int find_pos(struct apply_state *state,
2637 struct image *img,
2638 struct image *preimage,
2639 struct image *postimage,
2640 int line,
2641 unsigned ws_rule,
2642 int match_beginning, int match_end)
2643 {
2644 int i;
2645 unsigned long backwards, forwards, try;
2646 int backwards_lno, forwards_lno, try_lno;
2647
2648 /*
2649 * If match_beginning or match_end is specified, there is no
2650 * point starting from a wrong line that will never match and
2651 * wander around and wait for a match at the specified end.
2652 */
2653 if (match_beginning)
2654 line = 0;
2655 else if (match_end)
2656 line = img->nr - preimage->nr;
2657
2658 /*
2659 * Because the comparison is unsigned, the following test
2660 * will also take care of a negative line number that can
2661 * result when match_end and preimage is larger than the target.
2662 */
2663 if ((size_t) line > img->nr)
2664 line = img->nr;
2665
2666 try = 0;
2667 for (i = 0; i < line; i++)
2668 try += img->line[i].len;
2669
2670 /*
2671 * There's probably some smart way to do this, but I'll leave
2672 * that to the smart and beautiful people. I'm simple and stupid.
2673 */
2674 backwards = try;
2675 backwards_lno = line;
2676 forwards = try;
2677 forwards_lno = line;
2678 try_lno = line;
2679
2680 for (i = 0; ; i++) {
2681 if (match_fragment(state, img, preimage, postimage,
2682 try, try_lno, ws_rule,
2683 match_beginning, match_end))
2684 return try_lno;
2685
2686 again:
2687 if (backwards_lno == 0 && forwards_lno == img->nr)
2688 break;
2689
2690 if (i & 1) {
2691 if (backwards_lno == 0) {
2692 i++;
2693 goto again;
2694 }
2695 backwards_lno--;
2696 backwards -= img->line[backwards_lno].len;
2697 try = backwards;
2698 try_lno = backwards_lno;
2699 } else {
2700 if (forwards_lno == img->nr) {
2701 i++;
2702 goto again;
2703 }
2704 forwards += img->line[forwards_lno].len;
2705 forwards_lno++;
2706 try = forwards;
2707 try_lno = forwards_lno;
2708 }
2709
2710 }
2711 return -1;
2712 }
2713
2714 static void remove_first_line(struct image *img)
2715 {
2716 img->buf += img->line[0].len;
2717 img->len -= img->line[0].len;
2718 img->line++;
2719 img->nr--;
2720 }
2721
2722 static void remove_last_line(struct image *img)
2723 {
2724 img->len -= img->line[--img->nr].len;
2725 }
2726
2727 /*
2728 * The change from "preimage" and "postimage" has been found to
2729 * apply at applied_pos (counts in line numbers) in "img".
2730 * Update "img" to remove "preimage" and replace it with "postimage".
2731 */
2732 static void update_image(struct apply_state *state,
2733 struct image *img,
2734 int applied_pos,
2735 struct image *preimage,
2736 struct image *postimage)
2737 {
2738 /*
2739 * remove the copy of preimage at offset in img
2740 * and replace it with postimage
2741 */
2742 int i, nr;
2743 size_t remove_count, insert_count, applied_at = 0;
2744 char *result;
2745 int preimage_limit;
2746
2747 /*
2748 * If we are removing blank lines at the end of img,
2749 * the preimage may extend beyond the end.
2750 * If that is the case, we must be careful only to
2751 * remove the part of the preimage that falls within
2752 * the boundaries of img. Initialize preimage_limit
2753 * to the number of lines in the preimage that falls
2754 * within the boundaries.
2755 */
2756 preimage_limit = preimage->nr;
2757 if (preimage_limit > img->nr - applied_pos)
2758 preimage_limit = img->nr - applied_pos;
2759
2760 for (i = 0; i < applied_pos; i++)
2761 applied_at += img->line[i].len;
2762
2763 remove_count = 0;
2764 for (i = 0; i < preimage_limit; i++)
2765 remove_count += img->line[applied_pos + i].len;
2766 insert_count = postimage->len;
2767
2768 /* Adjust the contents */
2769 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2770 memcpy(result, img->buf, applied_at);
2771 memcpy(result + applied_at, postimage->buf, postimage->len);
2772 memcpy(result + applied_at + postimage->len,
2773 img->buf + (applied_at + remove_count),
2774 img->len - (applied_at + remove_count));
2775 free(img->buf);
2776 img->buf = result;
2777 img->len += insert_count - remove_count;
2778 result[img->len] = '\0';
2779
2780 /* Adjust the line table */
2781 nr = img->nr + postimage->nr - preimage_limit;
2782 if (preimage_limit < postimage->nr) {
2783 /*
2784 * NOTE: this knows that we never call remove_first_line()
2785 * on anything other than pre/post image.
2786 */
2787 REALLOC_ARRAY(img->line, nr);
2788 img->line_allocated = img->line;
2789 }
2790 if (preimage_limit != postimage->nr)
2791 memmove(img->line + applied_pos + postimage->nr,
2792 img->line + applied_pos + preimage_limit,
2793 (img->nr - (applied_pos + preimage_limit)) *
2794 sizeof(*img->line));
2795 memcpy(img->line + applied_pos,
2796 postimage->line,
2797 postimage->nr * sizeof(*img->line));
2798 if (!state->allow_overlap)
2799 for (i = 0; i < postimage->nr; i++)
2800 img->line[applied_pos + i].flag |= LINE_PATCHED;
2801 img->nr = nr;
2802 }
2803
2804 /*
2805 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2806 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2807 * replace the part of "img" with "postimage" text.
2808 */
2809 static int apply_one_fragment(struct apply_state *state,
2810 struct image *img, struct fragment *frag,
2811 int inaccurate_eof, unsigned ws_rule,
2812 int nth_fragment)
2813 {
2814 int match_beginning, match_end;
2815 const char *patch = frag->patch;
2816 int size = frag->size;
2817 char *old, *oldlines;
2818 struct strbuf newlines;
2819 int new_blank_lines_at_end = 0;
2820 int found_new_blank_lines_at_end = 0;
2821 int hunk_linenr = frag->linenr;
2822 unsigned long leading, trailing;
2823 int pos, applied_pos;
2824 struct image preimage;
2825 struct image postimage;
2826
2827 memset(&preimage, 0, sizeof(preimage));
2828 memset(&postimage, 0, sizeof(postimage));
2829 oldlines = xmalloc(size);
2830 strbuf_init(&newlines, size);
2831
2832 old = oldlines;
2833 while (size > 0) {
2834 char first;
2835 int len = linelen(patch, size);
2836 int plen;
2837 int added_blank_line = 0;
2838 int is_blank_context = 0;
2839 size_t start;
2840
2841 if (!len)
2842 break;
2843
2844 /*
2845 * "plen" is how much of the line we should use for
2846 * the actual patch data. Normally we just remove the
2847 * first character on the line, but if the line is
2848 * followed by "\ No newline", then we also remove the
2849 * last one (which is the newline, of course).
2850 */
2851 plen = len - 1;
2852 if (len < size && patch[len] == '\\')
2853 plen--;
2854 first = *patch;
2855 if (state->apply_in_reverse) {
2856 if (first == '-')
2857 first = '+';
2858 else if (first == '+')
2859 first = '-';
2860 }
2861
2862 switch (first) {
2863 case '\n':
2864 /* Newer GNU diff, empty context line */
2865 if (plen < 0)
2866 /* ... followed by '\No newline'; nothing */
2867 break;
2868 *old++ = '\n';
2869 strbuf_addch(&newlines, '\n');
2870 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2871 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2872 is_blank_context = 1;
2873 break;
2874 case ' ':
2875 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2876 ws_blank_line(patch + 1, plen, ws_rule))
2877 is_blank_context = 1;
2878 case '-':
2879 memcpy(old, patch + 1, plen);
2880 add_line_info(&preimage, old, plen,
2881 (first == ' ' ? LINE_COMMON : 0));
2882 old += plen;
2883 if (first == '-')
2884 break;
2885 /* Fall-through for ' ' */
2886 case '+':
2887 /* --no-add does not add new lines */
2888 if (first == '+' && state->no_add)
2889 break;
2890
2891 start = newlines.len;
2892 if (first != '+' ||
2893 !state->whitespace_error ||
2894 state->ws_error_action != correct_ws_error) {
2895 strbuf_add(&newlines, patch + 1, plen);
2896 }
2897 else {
2898 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2899 }
2900 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2901 (first == '+' ? 0 : LINE_COMMON));
2902 if (first == '+' &&
2903 (ws_rule & WS_BLANK_AT_EOF) &&
2904 ws_blank_line(patch + 1, plen, ws_rule))
2905 added_blank_line = 1;
2906 break;
2907 case '@': case '\\':
2908 /* Ignore it, we already handled it */
2909 break;
2910 default:
2911 if (state->apply_verbosity > verbosity_normal)
2912 error(_("invalid start of line: '%c'"), first);
2913 applied_pos = -1;
2914 goto out;
2915 }
2916 if (added_blank_line) {
2917 if (!new_blank_lines_at_end)
2918 found_new_blank_lines_at_end = hunk_linenr;
2919 new_blank_lines_at_end++;
2920 }
2921 else if (is_blank_context)
2922 ;
2923 else
2924 new_blank_lines_at_end = 0;
2925 patch += len;
2926 size -= len;
2927 hunk_linenr++;
2928 }
2929 if (inaccurate_eof &&
2930 old > oldlines && old[-1] == '\n' &&
2931 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2932 old--;
2933 strbuf_setlen(&newlines, newlines.len - 1);
2934 }
2935
2936 leading = frag->leading;
2937 trailing = frag->trailing;
2938
2939 /*
2940 * A hunk to change lines at the beginning would begin with
2941 * @@ -1,L +N,M @@
2942 * but we need to be careful. -U0 that inserts before the second
2943 * line also has this pattern.
2944 *
2945 * And a hunk to add to an empty file would begin with
2946 * @@ -0,0 +N,M @@
2947 *
2948 * In other words, a hunk that is (frag->oldpos <= 1) with or
2949 * without leading context must match at the beginning.
2950 */
2951 match_beginning = (!frag->oldpos ||
2952 (frag->oldpos == 1 && !state->unidiff_zero));
2953
2954 /*
2955 * A hunk without trailing lines must match at the end.
2956 * However, we simply cannot tell if a hunk must match end
2957 * from the lack of trailing lines if the patch was generated
2958 * with unidiff without any context.
2959 */
2960 match_end = !state->unidiff_zero && !trailing;
2961
2962 pos = frag->newpos ? (frag->newpos - 1) : 0;
2963 preimage.buf = oldlines;
2964 preimage.len = old - oldlines;
2965 postimage.buf = newlines.buf;
2966 postimage.len = newlines.len;
2967 preimage.line = preimage.line_allocated;
2968 postimage.line = postimage.line_allocated;
2969
2970 for (;;) {
2971
2972 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2973 ws_rule, match_beginning, match_end);
2974
2975 if (applied_pos >= 0)
2976 break;
2977
2978 /* Am I at my context limits? */
2979 if ((leading <= state->p_context) && (trailing <= state->p_context))
2980 break;
2981 if (match_beginning || match_end) {
2982 match_beginning = match_end = 0;
2983 continue;
2984 }
2985
2986 /*
2987 * Reduce the number of context lines; reduce both
2988 * leading and trailing if they are equal otherwise
2989 * just reduce the larger context.
2990 */
2991 if (leading >= trailing) {
2992 remove_first_line(&preimage);
2993 remove_first_line(&postimage);
2994 pos--;
2995 leading--;
2996 }
2997 if (trailing > leading) {
2998 remove_last_line(&preimage);
2999 remove_last_line(&postimage);
3000 trailing--;
3001 }
3002 }
3003
3004 if (applied_pos >= 0) {
3005 if (new_blank_lines_at_end &&
3006 preimage.nr + applied_pos >= img->nr &&
3007 (ws_rule & WS_BLANK_AT_EOF) &&
3008 state->ws_error_action != nowarn_ws_error) {
3009 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3010 found_new_blank_lines_at_end);
3011 if (state->ws_error_action == correct_ws_error) {
3012 while (new_blank_lines_at_end--)
3013 remove_last_line(&postimage);
3014 }
3015 /*
3016 * We would want to prevent write_out_results()
3017 * from taking place in apply_patch() that follows
3018 * the callchain led us here, which is:
3019 * apply_patch->check_patch_list->check_patch->
3020 * apply_data->apply_fragments->apply_one_fragment
3021 */
3022 if (state->ws_error_action == die_on_ws_error)
3023 state->apply = 0;
3024 }
3025
3026 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3027 int offset = applied_pos - pos;
3028 if (state->apply_in_reverse)
3029 offset = 0 - offset;
3030 fprintf_ln(stderr,
3031 Q_("Hunk #%d succeeded at %d (offset %d line).",
3032 "Hunk #%d succeeded at %d (offset %d lines).",
3033 offset),
3034 nth_fragment, applied_pos + 1, offset);
3035 }
3036
3037 /*
3038 * Warn if it was necessary to reduce the number
3039 * of context lines.
3040 */
3041 if ((leading != frag->leading ||
3042 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3043 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3044 " to apply fragment at %d"),
3045 leading, trailing, applied_pos+1);
3046 update_image(state, img, applied_pos, &preimage, &postimage);
3047 } else {
3048 if (state->apply_verbosity > verbosity_normal)
3049 error(_("while searching for:\n%.*s"),
3050 (int)(old - oldlines), oldlines);
3051 }
3052
3053 out:
3054 free(oldlines);
3055 strbuf_release(&newlines);
3056 free(preimage.line_allocated);
3057 free(postimage.line_allocated);
3058
3059 return (applied_pos < 0);
3060 }
3061
3062 static int apply_binary_fragment(struct apply_state *state,
3063 struct image *img,
3064 struct patch *patch)
3065 {
3066 struct fragment *fragment = patch->fragments;
3067 unsigned long len;
3068 void *dst;
3069
3070 if (!fragment)
3071 return error(_("missing binary patch data for '%s'"),
3072 patch->new_name ?
3073 patch->new_name :
3074 patch->old_name);
3075
3076 /* Binary patch is irreversible without the optional second hunk */
3077 if (state->apply_in_reverse) {
3078 if (!fragment->next)
3079 return error(_("cannot reverse-apply a binary patch "
3080 "without the reverse hunk to '%s'"),
3081 patch->new_name
3082 ? patch->new_name : patch->old_name);
3083 fragment = fragment->next;
3084 }
3085 switch (fragment->binary_patch_method) {
3086 case BINARY_DELTA_DEFLATED:
3087 dst = patch_delta(img->buf, img->len, fragment->patch,
3088 fragment->size, &len);
3089 if (!dst)
3090 return -1;
3091 clear_image(img);
3092 img->buf = dst;
3093 img->len = len;
3094 return 0;
3095 case BINARY_LITERAL_DEFLATED:
3096 clear_image(img);
3097 img->len = fragment->size;
3098 img->buf = xmemdupz(fragment->patch, img->len);
3099 return 0;
3100 }
3101 return -1;
3102 }
3103
3104 /*
3105 * Replace "img" with the result of applying the binary patch.
3106 * The binary patch data itself in patch->fragment is still kept
3107 * but the preimage prepared by the caller in "img" is freed here
3108 * or in the helper function apply_binary_fragment() this calls.
3109 */
3110 static int apply_binary(struct apply_state *state,
3111 struct image *img,
3112 struct patch *patch)
3113 {
3114 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3115 struct object_id oid;
3116
3117 /*
3118 * For safety, we require patch index line to contain
3119 * full 40-byte textual SHA1 for old and new, at least for now.
3120 */
3121 if (strlen(patch->old_sha1_prefix) != 40 ||
3122 strlen(patch->new_sha1_prefix) != 40 ||
3123 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3124 get_oid_hex(patch->new_sha1_prefix, &oid))
3125 return error(_("cannot apply binary patch to '%s' "
3126 "without full index line"), name);
3127
3128 if (patch->old_name) {
3129 /*
3130 * See if the old one matches what the patch
3131 * applies to.
3132 */
3133 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3134 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3135 return error(_("the patch applies to '%s' (%s), "
3136 "which does not match the "
3137 "current contents."),
3138 name, oid_to_hex(&oid));
3139 }
3140 else {
3141 /* Otherwise, the old one must be empty. */
3142 if (img->len)
3143 return error(_("the patch applies to an empty "
3144 "'%s' but it is not empty"), name);
3145 }
3146
3147 get_oid_hex(patch->new_sha1_prefix, &oid);
3148 if (is_null_oid(&oid)) {
3149 clear_image(img);
3150 return 0; /* deletion patch */
3151 }
3152
3153 if (has_sha1_file(oid.hash)) {
3154 /* We already have the postimage */
3155 enum object_type type;
3156 unsigned long size;
3157 char *result;
3158
3159 result = read_sha1_file(oid.hash, &type, &size);
3160 if (!result)
3161 return error(_("the necessary postimage %s for "
3162 "'%s' cannot be read"),
3163 patch->new_sha1_prefix, name);
3164 clear_image(img);
3165 img->buf = result;
3166 img->len = size;
3167 } else {
3168 /*
3169 * We have verified buf matches the preimage;
3170 * apply the patch data to it, which is stored
3171 * in the patch->fragments->{patch,size}.
3172 */
3173 if (apply_binary_fragment(state, img, patch))
3174 return error(_("binary patch does not apply to '%s'"),
3175 name);
3176
3177 /* verify that the result matches */
3178 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3179 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3180 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3181 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3182 }
3183
3184 return 0;
3185 }
3186
3187 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3188 {
3189 struct fragment *frag = patch->fragments;
3190 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3191 unsigned ws_rule = patch->ws_rule;
3192 unsigned inaccurate_eof = patch->inaccurate_eof;
3193 int nth = 0;
3194
3195 if (patch->is_binary)
3196 return apply_binary(state, img, patch);
3197
3198 while (frag) {
3199 nth++;
3200 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3201 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3202 if (!state->apply_with_reject)
3203 return -1;
3204 frag->rejected = 1;
3205 }
3206 frag = frag->next;
3207 }
3208 return 0;
3209 }
3210
3211 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3212 {
3213 if (S_ISGITLINK(mode)) {
3214 strbuf_grow(buf, 100);
3215 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3216 } else {
3217 enum object_type type;
3218 unsigned long sz;
3219 char *result;
3220
3221 result = read_sha1_file(oid->hash, &type, &sz);
3222 if (!result)
3223 return -1;
3224 /* XXX read_sha1_file NUL-terminates */
3225 strbuf_attach(buf, result, sz, sz + 1);
3226 }
3227 return 0;
3228 }
3229
3230 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3231 {
3232 if (!ce)
3233 return 0;
3234 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3235 }
3236
3237 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3238 {
3239 struct string_list_item *item;
3240
3241 if (name == NULL)
3242 return NULL;
3243
3244 item = string_list_lookup(&state->fn_table, name);
3245 if (item != NULL)
3246 return (struct patch *)item->util;
3247
3248 return NULL;
3249 }
3250
3251 /*
3252 * item->util in the filename table records the status of the path.
3253 * Usually it points at a patch (whose result records the contents
3254 * of it after applying it), but it could be PATH_WAS_DELETED for a
3255 * path that a previously applied patch has already removed, or
3256 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3257 *
3258 * The latter is needed to deal with a case where two paths A and B
3259 * are swapped by first renaming A to B and then renaming B to A;
3260 * moving A to B should not be prevented due to presence of B as we
3261 * will remove it in a later patch.
3262 */
3263 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3264 #define PATH_WAS_DELETED ((struct patch *) -1)
3265
3266 static int to_be_deleted(struct patch *patch)
3267 {
3268 return patch == PATH_TO_BE_DELETED;
3269 }
3270
3271 static int was_deleted(struct patch *patch)
3272 {
3273 return patch == PATH_WAS_DELETED;
3274 }
3275
3276 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3277 {
3278 struct string_list_item *item;
3279
3280 /*
3281 * Always add new_name unless patch is a deletion
3282 * This should cover the cases for normal diffs,
3283 * file creations and copies
3284 */
3285 if (patch->new_name != NULL) {
3286 item = string_list_insert(&state->fn_table, patch->new_name);
3287 item->util = patch;
3288 }
3289
3290 /*
3291 * store a failure on rename/deletion cases because
3292 * later chunks shouldn't patch old names
3293 */
3294 if ((patch->new_name == NULL) || (patch->is_rename)) {
3295 item = string_list_insert(&state->fn_table, patch->old_name);
3296 item->util = PATH_WAS_DELETED;
3297 }
3298 }
3299
3300 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3301 {
3302 /*
3303 * store information about incoming file deletion
3304 */
3305 while (patch) {
3306 if ((patch->new_name == NULL) || (patch->is_rename)) {
3307 struct string_list_item *item;
3308 item = string_list_insert(&state->fn_table, patch->old_name);
3309 item->util = PATH_TO_BE_DELETED;
3310 }
3311 patch = patch->next;
3312 }
3313 }
3314
3315 static int checkout_target(struct index_state *istate,
3316 struct cache_entry *ce, struct stat *st)
3317 {
3318 struct checkout costate = CHECKOUT_INIT;
3319
3320 costate.refresh_cache = 1;
3321 costate.istate = istate;
3322 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3323 return error(_("cannot checkout %s"), ce->name);
3324 return 0;
3325 }
3326
3327 static struct patch *previous_patch(struct apply_state *state,
3328 struct patch *patch,
3329 int *gone)
3330 {
3331 struct patch *previous;
3332
3333 *gone = 0;
3334 if (patch->is_copy || patch->is_rename)
3335 return NULL; /* "git" patches do not depend on the order */
3336
3337 previous = in_fn_table(state, patch->old_name);
3338 if (!previous)
3339 return NULL;
3340
3341 if (to_be_deleted(previous))
3342 return NULL; /* the deletion hasn't happened yet */
3343
3344 if (was_deleted(previous))
3345 *gone = 1;
3346
3347 return previous;
3348 }
3349
3350 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3351 {
3352 if (S_ISGITLINK(ce->ce_mode)) {
3353 if (!S_ISDIR(st->st_mode))
3354 return -1;
3355 return 0;
3356 }
3357 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3358 }
3359
3360 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3361
3362 static int load_patch_target(struct apply_state *state,
3363 struct strbuf *buf,
3364 const struct cache_entry *ce,
3365 struct stat *st,
3366 const char *name,
3367 unsigned expected_mode)
3368 {
3369 if (state->cached || state->check_index) {
3370 if (read_file_or_gitlink(ce, buf))
3371 return error(_("failed to read %s"), name);
3372 } else if (name) {
3373 if (S_ISGITLINK(expected_mode)) {
3374 if (ce)
3375 return read_file_or_gitlink(ce, buf);
3376 else
3377 return SUBMODULE_PATCH_WITHOUT_INDEX;
3378 } else if (has_symlink_leading_path(name, strlen(name))) {
3379 return error(_("reading from '%s' beyond a symbolic link"), name);
3380 } else {
3381 if (read_old_data(st, name, buf))
3382 return error(_("failed to read %s"), name);
3383 }
3384 }
3385 return 0;
3386 }
3387
3388 /*
3389 * We are about to apply "patch"; populate the "image" with the
3390 * current version we have, from the working tree or from the index,
3391 * depending on the situation e.g. --cached/--index. If we are
3392 * applying a non-git patch that incrementally updates the tree,
3393 * we read from the result of a previous diff.
3394 */
3395 static int load_preimage(struct apply_state *state,
3396 struct image *image,
3397 struct patch *patch, struct stat *st,
3398 const struct cache_entry *ce)
3399 {
3400 struct strbuf buf = STRBUF_INIT;
3401 size_t len;
3402 char *img;
3403 struct patch *previous;
3404 int status;
3405
3406 previous = previous_patch(state, patch, &status);
3407 if (status)
3408 return error(_("path %s has been renamed/deleted"),
3409 patch->old_name);
3410 if (previous) {
3411 /* We have a patched copy in memory; use that. */
3412 strbuf_add(&buf, previous->result, previous->resultsize);
3413 } else {
3414 status = load_patch_target(state, &buf, ce, st,
3415 patch->old_name, patch->old_mode);
3416 if (status < 0)
3417 return status;
3418 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3419 /*
3420 * There is no way to apply subproject
3421 * patch without looking at the index.
3422 * NEEDSWORK: shouldn't this be flagged
3423 * as an error???
3424 */
3425 free_fragment_list(patch->fragments);
3426 patch->fragments = NULL;
3427 } else if (status) {
3428 return error(_("failed to read %s"), patch->old_name);
3429 }
3430 }
3431
3432 img = strbuf_detach(&buf, &len);
3433 prepare_image(image, img, len, !patch->is_binary);
3434 return 0;
3435 }
3436
3437 static int three_way_merge(struct image *image,
3438 char *path,
3439 const struct object_id *base,
3440 const struct object_id *ours,
3441 const struct object_id *theirs)
3442 {
3443 mmfile_t base_file, our_file, their_file;
3444 mmbuffer_t result = { NULL };
3445 int status;
3446
3447 read_mmblob(&base_file, base);
3448 read_mmblob(&our_file, ours);
3449 read_mmblob(&their_file, theirs);
3450 status = ll_merge(&result, path,
3451 &base_file, "base",
3452 &our_file, "ours",
3453 &their_file, "theirs", NULL);
3454 free(base_file.ptr);
3455 free(our_file.ptr);
3456 free(their_file.ptr);
3457 if (status < 0 || !result.ptr) {
3458 free(result.ptr);
3459 return -1;
3460 }
3461 clear_image(image);
3462 image->buf = result.ptr;
3463 image->len = result.size;
3464
3465 return status;
3466 }
3467
3468 /*
3469 * When directly falling back to add/add three-way merge, we read from
3470 * the current contents of the new_name. In no cases other than that
3471 * this function will be called.
3472 */
3473 static int load_current(struct apply_state *state,
3474 struct image *image,
3475 struct patch *patch)
3476 {
3477 struct strbuf buf = STRBUF_INIT;
3478 int status, pos;
3479 size_t len;
3480 char *img;
3481 struct stat st;
3482 struct cache_entry *ce;
3483 char *name = patch->new_name;
3484 unsigned mode = patch->new_mode;
3485
3486 if (!patch->is_new)
3487 die("BUG: patch to %s is not a creation", patch->old_name);
3488
3489 pos = cache_name_pos(name, strlen(name));
3490 if (pos < 0)
3491 return error(_("%s: does not exist in index"), name);
3492 ce = active_cache[pos];
3493 if (lstat(name, &st)) {
3494 if (errno != ENOENT)
3495 return error_errno("%s", name);
3496 if (checkout_target(&the_index, ce, &st))
3497 return -1;
3498 }
3499 if (verify_index_match(ce, &st))
3500 return error(_("%s: does not match index"), name);
3501
3502 status = load_patch_target(state, &buf, ce, &st, name, mode);
3503 if (status < 0)
3504 return status;
3505 else if (status)
3506 return -1;
3507 img = strbuf_detach(&buf, &len);
3508 prepare_image(image, img, len, !patch->is_binary);
3509 return 0;
3510 }
3511
3512 static int try_threeway(struct apply_state *state,
3513 struct image *image,
3514 struct patch *patch,
3515 struct stat *st,
3516 const struct cache_entry *ce)
3517 {
3518 struct object_id pre_oid, post_oid, our_oid;
3519 struct strbuf buf = STRBUF_INIT;
3520 size_t len;
3521 int status;
3522 char *img;
3523 struct image tmp_image;
3524
3525 /* No point falling back to 3-way merge in these cases */
3526 if (patch->is_delete ||
3527 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3528 return -1;
3529
3530 /* Preimage the patch was prepared for */
3531 if (patch->is_new)
3532 write_sha1_file("", 0, blob_type, pre_oid.hash);
3533 else if (get_sha1(patch->old_sha1_prefix, pre_oid.hash) ||
3534 read_blob_object(&buf, &pre_oid, patch->old_mode))
3535 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3536
3537 if (state->apply_verbosity > verbosity_silent)
3538 fprintf(stderr, _("Falling back to three-way merge...\n"));
3539
3540 img = strbuf_detach(&buf, &len);
3541 prepare_image(&tmp_image, img, len, 1);
3542 /* Apply the patch to get the post image */
3543 if (apply_fragments(state, &tmp_image, patch) < 0) {
3544 clear_image(&tmp_image);
3545 return -1;
3546 }
3547 /* post_oid is theirs */
3548 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3549 clear_image(&tmp_image);
3550
3551 /* our_oid is ours */
3552 if (patch->is_new) {
3553 if (load_current(state, &tmp_image, patch))
3554 return error(_("cannot read the current contents of '%s'"),
3555 patch->new_name);
3556 } else {
3557 if (load_preimage(state, &tmp_image, patch, st, ce))
3558 return error(_("cannot read the current contents of '%s'"),
3559 patch->old_name);
3560 }
3561 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3562 clear_image(&tmp_image);
3563
3564 /* in-core three-way merge between post and our using pre as base */
3565 status = three_way_merge(image, patch->new_name,
3566 &pre_oid, &our_oid, &post_oid);
3567 if (status < 0) {
3568 if (state->apply_verbosity > verbosity_silent)
3569 fprintf(stderr,
3570 _("Failed to fall back on three-way merge...\n"));
3571 return status;
3572 }
3573
3574 if (status) {
3575 patch->conflicted_threeway = 1;
3576 if (patch->is_new)
3577 oidclr(&patch->threeway_stage[0]);
3578 else
3579 oidcpy(&patch->threeway_stage[0], &pre_oid);
3580 oidcpy(&patch->threeway_stage[1], &our_oid);
3581 oidcpy(&patch->threeway_stage[2], &post_oid);
3582 if (state->apply_verbosity > verbosity_silent)
3583 fprintf(stderr,
3584 _("Applied patch to '%s' with conflicts.\n"),
3585 patch->new_name);
3586 } else {
3587 if (state->apply_verbosity > verbosity_silent)
3588 fprintf(stderr,
3589 _("Applied patch to '%s' cleanly.\n"),
3590 patch->new_name);
3591 }
3592 return 0;
3593 }
3594
3595 static int apply_data(struct apply_state *state, struct patch *patch,
3596 struct stat *st, const struct cache_entry *ce)
3597 {
3598 struct image image;
3599
3600 if (load_preimage(state, &image, patch, st, ce) < 0)
3601 return -1;
3602
3603 if (patch->direct_to_threeway ||
3604 apply_fragments(state, &image, patch) < 0) {
3605 /* Note: with --reject, apply_fragments() returns 0 */
3606 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3607 return -1;
3608 }
3609 patch->result = image.buf;
3610 patch->resultsize = image.len;
3611 add_to_fn_table(state, patch);
3612 free(image.line_allocated);
3613
3614 if (0 < patch->is_delete && patch->resultsize)
3615 return error(_("removal patch leaves file contents"));
3616
3617 return 0;
3618 }
3619
3620 /*
3621 * If "patch" that we are looking at modifies or deletes what we have,
3622 * we would want it not to lose any local modification we have, either
3623 * in the working tree or in the index.
3624 *
3625 * This also decides if a non-git patch is a creation patch or a
3626 * modification to an existing empty file. We do not check the state
3627 * of the current tree for a creation patch in this function; the caller
3628 * check_patch() separately makes sure (and errors out otherwise) that
3629 * the path the patch creates does not exist in the current tree.
3630 */
3631 static int check_preimage(struct apply_state *state,
3632 struct patch *patch,
3633 struct cache_entry **ce,
3634 struct stat *st)
3635 {
3636 const char *old_name = patch->old_name;
3637 struct patch *previous = NULL;
3638 int stat_ret = 0, status;
3639 unsigned st_mode = 0;
3640
3641 if (!old_name)
3642 return 0;
3643
3644 assert(patch->is_new <= 0);
3645 previous = previous_patch(state, patch, &status);
3646
3647 if (status)
3648 return error(_("path %s has been renamed/deleted"), old_name);
3649 if (previous) {
3650 st_mode = previous->new_mode;
3651 } else if (!state->cached) {
3652 stat_ret = lstat(old_name, st);
3653 if (stat_ret && errno != ENOENT)
3654 return error_errno("%s", old_name);
3655 }
3656
3657 if (state->check_index && !previous) {
3658 int pos = cache_name_pos(old_name, strlen(old_name));
3659 if (pos < 0) {
3660 if (patch->is_new < 0)
3661 goto is_new;
3662 return error(_("%s: does not exist in index"), old_name);
3663 }
3664 *ce = active_cache[pos];
3665 if (stat_ret < 0) {
3666 if (checkout_target(&the_index, *ce, st))
3667 return -1;
3668 }
3669 if (!state->cached && verify_index_match(*ce, st))
3670 return error(_("%s: does not match index"), old_name);
3671 if (state->cached)
3672 st_mode = (*ce)->ce_mode;
3673 } else if (stat_ret < 0) {
3674 if (patch->is_new < 0)
3675 goto is_new;
3676 return error_errno("%s", old_name);
3677 }
3678
3679 if (!state->cached && !previous)
3680 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3681
3682 if (patch->is_new < 0)
3683 patch->is_new = 0;
3684 if (!patch->old_mode)
3685 patch->old_mode = st_mode;
3686 if ((st_mode ^ patch->old_mode) & S_IFMT)
3687 return error(_("%s: wrong type"), old_name);
3688 if (st_mode != patch->old_mode)
3689 warning(_("%s has type %o, expected %o"),
3690 old_name, st_mode, patch->old_mode);
3691 if (!patch->new_mode && !patch->is_delete)
3692 patch->new_mode = st_mode;
3693 return 0;
3694
3695 is_new:
3696 patch->is_new = 1;
3697 patch->is_delete = 0;
3698 free(patch->old_name);
3699 patch->old_name = NULL;
3700 return 0;
3701 }
3702
3703
3704 #define EXISTS_IN_INDEX 1
3705 #define EXISTS_IN_WORKTREE 2
3706
3707 static int check_to_create(struct apply_state *state,
3708 const char *new_name,
3709 int ok_if_exists)
3710 {
3711 struct stat nst;
3712
3713 if (state->check_index &&
3714 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3715 !ok_if_exists)
3716 return EXISTS_IN_INDEX;
3717 if (state->cached)
3718 return 0;
3719
3720 if (!lstat(new_name, &nst)) {
3721 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3722 return 0;
3723 /*
3724 * A leading component of new_name might be a symlink
3725 * that is going to be removed with this patch, but
3726 * still pointing at somewhere that has the path.
3727 * In such a case, path "new_name" does not exist as
3728 * far as git is concerned.
3729 */
3730 if (has_symlink_leading_path(new_name, strlen(new_name)))
3731 return 0;
3732
3733 return EXISTS_IN_WORKTREE;
3734 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3735 return error_errno("%s", new_name);
3736 }
3737 return 0;
3738 }
3739
3740 static uintptr_t register_symlink_changes(struct apply_state *state,
3741 const char *path,
3742 uintptr_t what)
3743 {
3744 struct string_list_item *ent;
3745
3746 ent = string_list_lookup(&state->symlink_changes, path);
3747 if (!ent) {
3748 ent = string_list_insert(&state->symlink_changes, path);
3749 ent->util = (void *)0;
3750 }
3751 ent->util = (void *)(what | ((uintptr_t)ent->util));
3752 return (uintptr_t)ent->util;
3753 }
3754
3755 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3756 {
3757 struct string_list_item *ent;
3758
3759 ent = string_list_lookup(&state->symlink_changes, path);
3760 if (!ent)
3761 return 0;
3762 return (uintptr_t)ent->util;
3763 }
3764
3765 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3766 {
3767 for ( ; patch; patch = patch->next) {
3768 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3769 (patch->is_rename || patch->is_delete))
3770 /* the symlink at patch->old_name is removed */
3771 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3772
3773 if (patch->new_name && S_ISLNK(patch->new_mode))
3774 /* the symlink at patch->new_name is created or remains */
3775 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3776 }
3777 }
3778
3779 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3780 {
3781 do {
3782 unsigned int change;
3783
3784 while (--name->len && name->buf[name->len] != '/')
3785 ; /* scan backwards */
3786 if (!name->len)
3787 break;
3788 name->buf[name->len] = '\0';
3789 change = check_symlink_changes(state, name->buf);
3790 if (change & APPLY_SYMLINK_IN_RESULT)
3791 return 1;
3792 if (change & APPLY_SYMLINK_GOES_AWAY)
3793 /*
3794 * This cannot be "return 0", because we may
3795 * see a new one created at a higher level.
3796 */
3797 continue;
3798
3799 /* otherwise, check the preimage */
3800 if (state->check_index) {
3801 struct cache_entry *ce;
3802
3803 ce = cache_file_exists(name->buf, name->len, ignore_case);
3804 if (ce && S_ISLNK(ce->ce_mode))
3805 return 1;
3806 } else {
3807 struct stat st;
3808 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3809 return 1;
3810 }
3811 } while (1);
3812 return 0;
3813 }
3814
3815 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3816 {
3817 int ret;
3818 struct strbuf name = STRBUF_INIT;
3819
3820 assert(*name_ != '\0');
3821 strbuf_addstr(&name, name_);
3822 ret = path_is_beyond_symlink_1(state, &name);
3823 strbuf_release(&name);
3824
3825 return ret;
3826 }
3827
3828 static int check_unsafe_path(struct patch *patch)
3829 {
3830 const char *old_name = NULL;
3831 const char *new_name = NULL;
3832 if (patch->is_delete)
3833 old_name = patch->old_name;
3834 else if (!patch->is_new && !patch->is_copy)
3835 old_name = patch->old_name;
3836 if (!patch->is_delete)
3837 new_name = patch->new_name;
3838
3839 if (old_name && !verify_path(old_name))
3840 return error(_("invalid path '%s'"), old_name);
3841 if (new_name && !verify_path(new_name))
3842 return error(_("invalid path '%s'"), new_name);
3843 return 0;
3844 }
3845
3846 /*
3847 * Check and apply the patch in-core; leave the result in patch->result
3848 * for the caller to write it out to the final destination.
3849 */
3850