.mailmap: normalize name for René Scharfe
[git/git.git] / apply.c
1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9
10 #include "cache.h"
11 #include "config.h"
12 #include "blob.h"
13 #include "delta.h"
14 #include "diff.h"
15 #include "dir.h"
16 #include "xdiff-interface.h"
17 #include "ll-merge.h"
18 #include "lockfile.h"
19 #include "parse-options.h"
20 #include "quote.h"
21 #include "rerere.h"
22 #include "apply.h"
23
24 static void git_apply_config(void)
25 {
26 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);
27 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);
28 git_config(git_default_config, NULL);
29 }
30
31 static int parse_whitespace_option(struct apply_state *state, const char *option)
32 {
33 if (!option) {
34 state->ws_error_action = warn_on_ws_error;
35 return 0;
36 }
37 if (!strcmp(option, "warn")) {
38 state->ws_error_action = warn_on_ws_error;
39 return 0;
40 }
41 if (!strcmp(option, "nowarn")) {
42 state->ws_error_action = nowarn_ws_error;
43 return 0;
44 }
45 if (!strcmp(option, "error")) {
46 state->ws_error_action = die_on_ws_error;
47 return 0;
48 }
49 if (!strcmp(option, "error-all")) {
50 state->ws_error_action = die_on_ws_error;
51 state->squelch_whitespace_errors = 0;
52 return 0;
53 }
54 if (!strcmp(option, "strip") || !strcmp(option, "fix")) {
55 state->ws_error_action = correct_ws_error;
56 return 0;
57 }
58 return error(_("unrecognized whitespace option '%s'"), option);
59 }
60
61 static int parse_ignorewhitespace_option(struct apply_state *state,
62 const char *option)
63 {
64 if (!option || !strcmp(option, "no") ||
65 !strcmp(option, "false") || !strcmp(option, "never") ||
66 !strcmp(option, "none")) {
67 state->ws_ignore_action = ignore_ws_none;
68 return 0;
69 }
70 if (!strcmp(option, "change")) {
71 state->ws_ignore_action = ignore_ws_change;
72 return 0;
73 }
74 return error(_("unrecognized whitespace ignore option '%s'"), option);
75 }
76
77 int init_apply_state(struct apply_state *state,
78 const char *prefix,
79 struct lock_file *lock_file)
80 {
81 memset(state, 0, sizeof(*state));
82 state->prefix = prefix;
83 state->lock_file = lock_file;
84 state->newfd = -1;
85 state->apply = 1;
86 state->line_termination = '\n';
87 state->p_value = 1;
88 state->p_context = UINT_MAX;
89 state->squelch_whitespace_errors = 5;
90 state->ws_error_action = warn_on_ws_error;
91 state->ws_ignore_action = ignore_ws_none;
92 state->linenr = 1;
93 string_list_init(&state->fn_table, 0);
94 string_list_init(&state->limit_by_name, 0);
95 string_list_init(&state->symlink_changes, 0);
96 strbuf_init(&state->root, 0);
97
98 git_apply_config();
99 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
100 return -1;
101 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
102 return -1;
103 return 0;
104 }
105
106 void clear_apply_state(struct apply_state *state)
107 {
108 string_list_clear(&state->limit_by_name, 0);
109 string_list_clear(&state->symlink_changes, 0);
110 strbuf_release(&state->root);
111
112 /* &state->fn_table is cleared at the end of apply_patch() */
113 }
114
115 static void mute_routine(const char *msg, va_list params)
116 {
117 /* do nothing */
118 }
119
120 int check_apply_state(struct apply_state *state, int force_apply)
121 {
122 int is_not_gitdir = !startup_info->have_repository;
123
124 if (state->apply_with_reject && state->threeway)
125 return error(_("--reject and --3way cannot be used together."));
126 if (state->cached && state->threeway)
127 return error(_("--cached and --3way cannot be used together."));
128 if (state->threeway) {
129 if (is_not_gitdir)
130 return error(_("--3way outside a repository"));
131 state->check_index = 1;
132 }
133 if (state->apply_with_reject) {
134 state->apply = 1;
135 if (state->apply_verbosity == verbosity_normal)
136 state->apply_verbosity = verbosity_verbose;
137 }
138 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
139 state->apply = 0;
140 if (state->check_index && is_not_gitdir)
141 return error(_("--index outside a repository"));
142 if (state->cached) {
143 if (is_not_gitdir)
144 return error(_("--cached outside a repository"));
145 state->check_index = 1;
146 }
147 if (state->check_index)
148 state->unsafe_paths = 0;
149 if (!state->lock_file)
150 return error("BUG: state->lock_file should not be NULL");
151
152 if (state->apply_verbosity <= verbosity_silent) {
153 state->saved_error_routine = get_error_routine();
154 state->saved_warn_routine = get_warn_routine();
155 set_error_routine(mute_routine);
156 set_warn_routine(mute_routine);
157 }
158
159 return 0;
160 }
161
162 static void set_default_whitespace_mode(struct apply_state *state)
163 {
164 if (!state->whitespace_option && !apply_default_whitespace)
165 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
166 }
167
168 /*
169 * This represents one "hunk" from a patch, starting with
170 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
171 * patch text is pointed at by patch, and its byte length
172 * is stored in size. leading and trailing are the number
173 * of context lines.
174 */
175 struct fragment {
176 unsigned long leading, trailing;
177 unsigned long oldpos, oldlines;
178 unsigned long newpos, newlines;
179 /*
180 * 'patch' is usually borrowed from buf in apply_patch(),
181 * but some codepaths store an allocated buffer.
182 */
183 const char *patch;
184 unsigned free_patch:1,
185 rejected:1;
186 int size;
187 int linenr;
188 struct fragment *next;
189 };
190
191 /*
192 * When dealing with a binary patch, we reuse "leading" field
193 * to store the type of the binary hunk, either deflated "delta"
194 * or deflated "literal".
195 */
196 #define binary_patch_method leading
197 #define BINARY_DELTA_DEFLATED 1
198 #define BINARY_LITERAL_DEFLATED 2
199
200 /*
201 * This represents a "patch" to a file, both metainfo changes
202 * such as creation/deletion, filemode and content changes represented
203 * as a series of fragments.
204 */
205 struct patch {
206 char *new_name, *old_name, *def_name;
207 unsigned int old_mode, new_mode;
208 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
209 int rejected;
210 unsigned ws_rule;
211 int lines_added, lines_deleted;
212 int score;
213 int extension_linenr; /* first line specifying delete/new/rename/copy */
214 unsigned int is_toplevel_relative:1;
215 unsigned int inaccurate_eof:1;
216 unsigned int is_binary:1;
217 unsigned int is_copy:1;
218 unsigned int is_rename:1;
219 unsigned int recount:1;
220 unsigned int conflicted_threeway:1;
221 unsigned int direct_to_threeway:1;
222 unsigned int crlf_in_old:1;
223 struct fragment *fragments;
224 char *result;
225 size_t resultsize;
226 char old_sha1_prefix[41];
227 char new_sha1_prefix[41];
228 struct patch *next;
229
230 /* three-way fallback result */
231 struct object_id threeway_stage[3];
232 };
233
234 static void free_fragment_list(struct fragment *list)
235 {
236 while (list) {
237 struct fragment *next = list->next;
238 if (list->free_patch)
239 free((char *)list->patch);
240 free(list);
241 list = next;
242 }
243 }
244
245 static void free_patch(struct patch *patch)
246 {
247 free_fragment_list(patch->fragments);
248 free(patch->def_name);
249 free(patch->old_name);
250 free(patch->new_name);
251 free(patch->result);
252 free(patch);
253 }
254
255 static void free_patch_list(struct patch *list)
256 {
257 while (list) {
258 struct patch *next = list->next;
259 free_patch(list);
260 list = next;
261 }
262 }
263
264 /*
265 * A line in a file, len-bytes long (includes the terminating LF,
266 * except for an incomplete line at the end if the file ends with
267 * one), and its contents hashes to 'hash'.
268 */
269 struct line {
270 size_t len;
271 unsigned hash : 24;
272 unsigned flag : 8;
273 #define LINE_COMMON 1
274 #define LINE_PATCHED 2
275 };
276
277 /*
278 * This represents a "file", which is an array of "lines".
279 */
280 struct image {
281 char *buf;
282 size_t len;
283 size_t nr;
284 size_t alloc;
285 struct line *line_allocated;
286 struct line *line;
287 };
288
289 static uint32_t hash_line(const char *cp, size_t len)
290 {
291 size_t i;
292 uint32_t h;
293 for (i = 0, h = 0; i < len; i++) {
294 if (!isspace(cp[i])) {
295 h = h * 3 + (cp[i] & 0xff);
296 }
297 }
298 return h;
299 }
300
301 /*
302 * Compare lines s1 of length n1 and s2 of length n2, ignoring
303 * whitespace difference. Returns 1 if they match, 0 otherwise
304 */
305 static int fuzzy_matchlines(const char *s1, size_t n1,
306 const char *s2, size_t n2)
307 {
308 const char *last1 = s1 + n1 - 1;
309 const char *last2 = s2 + n2 - 1;
310 int result = 0;
311
312 /* ignore line endings */
313 while ((*last1 == '\r') || (*last1 == '\n'))
314 last1--;
315 while ((*last2 == '\r') || (*last2 == '\n'))
316 last2--;
317
318 /* skip leading whitespaces, if both begin with whitespace */
319 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
320 while (isspace(*s1) && (s1 <= last1))
321 s1++;
322 while (isspace(*s2) && (s2 <= last2))
323 s2++;
324 }
325 /* early return if both lines are empty */
326 if ((s1 > last1) && (s2 > last2))
327 return 1;
328 while (!result) {
329 result = *s1++ - *s2++;
330 /*
331 * Skip whitespace inside. We check for whitespace on
332 * both buffers because we don't want "a b" to match
333 * "ab"
334 */
335 if (isspace(*s1) && isspace(*s2)) {
336 while (isspace(*s1) && s1 <= last1)
337 s1++;
338 while (isspace(*s2) && s2 <= last2)
339 s2++;
340 }
341 /*
342 * If we reached the end on one side only,
343 * lines don't match
344 */
345 if (
346 ((s2 > last2) && (s1 <= last1)) ||
347 ((s1 > last1) && (s2 <= last2)))
348 return 0;
349 if ((s1 > last1) && (s2 > last2))
350 break;
351 }
352
353 return !result;
354 }
355
356 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
357 {
358 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
359 img->line_allocated[img->nr].len = len;
360 img->line_allocated[img->nr].hash = hash_line(bol, len);
361 img->line_allocated[img->nr].flag = flag;
362 img->nr++;
363 }
364
365 /*
366 * "buf" has the file contents to be patched (read from various sources).
367 * attach it to "image" and add line-based index to it.
368 * "image" now owns the "buf".
369 */
370 static void prepare_image(struct image *image, char *buf, size_t len,
371 int prepare_linetable)
372 {
373 const char *cp, *ep;
374
375 memset(image, 0, sizeof(*image));
376 image->buf = buf;
377 image->len = len;
378
379 if (!prepare_linetable)
380 return;
381
382 ep = image->buf + image->len;
383 cp = image->buf;
384 while (cp < ep) {
385 const char *next;
386 for (next = cp; next < ep && *next != '\n'; next++)
387 ;
388 if (next < ep)
389 next++;
390 add_line_info(image, cp, next - cp, 0);
391 cp = next;
392 }
393 image->line = image->line_allocated;
394 }
395
396 static void clear_image(struct image *image)
397 {
398 free(image->buf);
399 free(image->line_allocated);
400 memset(image, 0, sizeof(*image));
401 }
402
403 /* fmt must contain _one_ %s and no other substitution */
404 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
405 {
406 struct strbuf sb = STRBUF_INIT;
407
408 if (patch->old_name && patch->new_name &&
409 strcmp(patch->old_name, patch->new_name)) {
410 quote_c_style(patch->old_name, &sb, NULL, 0);
411 strbuf_addstr(&sb, " => ");
412 quote_c_style(patch->new_name, &sb, NULL, 0);
413 } else {
414 const char *n = patch->new_name;
415 if (!n)
416 n = patch->old_name;
417 quote_c_style(n, &sb, NULL, 0);
418 }
419 fprintf(output, fmt, sb.buf);
420 fputc('\n', output);
421 strbuf_release(&sb);
422 }
423
424 #define SLOP (16)
425
426 static int read_patch_file(struct strbuf *sb, int fd)
427 {
428 if (strbuf_read(sb, fd, 0) < 0)
429 return error_errno("git apply: failed to read");
430
431 /*
432 * Make sure that we have some slop in the buffer
433 * so that we can do speculative "memcmp" etc, and
434 * see to it that it is NUL-filled.
435 */
436 strbuf_grow(sb, SLOP);
437 memset(sb->buf + sb->len, 0, SLOP);
438 return 0;
439 }
440
441 static unsigned long linelen(const char *buffer, unsigned long size)
442 {
443 unsigned long len = 0;
444 while (size--) {
445 len++;
446 if (*buffer++ == '\n')
447 break;
448 }
449 return len;
450 }
451
452 static int is_dev_null(const char *str)
453 {
454 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
455 }
456
457 #define TERM_SPACE 1
458 #define TERM_TAB 2
459
460 static int name_terminate(int c, int terminate)
461 {
462 if (c == ' ' && !(terminate & TERM_SPACE))
463 return 0;
464 if (c == '\t' && !(terminate & TERM_TAB))
465 return 0;
466
467 return 1;
468 }
469
470 /* remove double slashes to make --index work with such filenames */
471 static char *squash_slash(char *name)
472 {
473 int i = 0, j = 0;
474
475 if (!name)
476 return NULL;
477
478 while (name[i]) {
479 if ((name[j++] = name[i++]) == '/')
480 while (name[i] == '/')
481 i++;
482 }
483 name[j] = '\0';
484 return name;
485 }
486
487 static char *find_name_gnu(struct apply_state *state,
488 const char *line,
489 const char *def,
490 int p_value)
491 {
492 struct strbuf name = STRBUF_INIT;
493 char *cp;
494
495 /*
496 * Proposed "new-style" GNU patch/diff format; see
497 * http://marc.info/?l=git&m=112927316408690&w=2
498 */
499 if (unquote_c_style(&name, line, NULL)) {
500 strbuf_release(&name);
501 return NULL;
502 }
503
504 for (cp = name.buf; p_value; p_value--) {
505 cp = strchr(cp, '/');
506 if (!cp) {
507 strbuf_release(&name);
508 return NULL;
509 }
510 cp++;
511 }
512
513 strbuf_remove(&name, 0, cp - name.buf);
514 if (state->root.len)
515 strbuf_insert(&name, 0, state->root.buf, state->root.len);
516 return squash_slash(strbuf_detach(&name, NULL));
517 }
518
519 static size_t sane_tz_len(const char *line, size_t len)
520 {
521 const char *tz, *p;
522
523 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
524 return 0;
525 tz = line + len - strlen(" +0500");
526
527 if (tz[1] != '+' && tz[1] != '-')
528 return 0;
529
530 for (p = tz + 2; p != line + len; p++)
531 if (!isdigit(*p))
532 return 0;
533
534 return line + len - tz;
535 }
536
537 static size_t tz_with_colon_len(const char *line, size_t len)
538 {
539 const char *tz, *p;
540
541 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
542 return 0;
543 tz = line + len - strlen(" +08:00");
544
545 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
546 return 0;
547 p = tz + 2;
548 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
549 !isdigit(*p++) || !isdigit(*p++))
550 return 0;
551
552 return line + len - tz;
553 }
554
555 static size_t date_len(const char *line, size_t len)
556 {
557 const char *date, *p;
558
559 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
560 return 0;
561 p = date = line + len - strlen("72-02-05");
562
563 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
564 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
565 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
566 return 0;
567
568 if (date - line >= strlen("19") &&
569 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
570 date -= strlen("19");
571
572 return line + len - date;
573 }
574
575 static size_t short_time_len(const char *line, size_t len)
576 {
577 const char *time, *p;
578
579 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
580 return 0;
581 p = time = line + len - strlen(" 07:01:32");
582
583 /* Permit 1-digit hours? */
584 if (*p++ != ' ' ||
585 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
586 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
587 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
588 return 0;
589
590 return line + len - time;
591 }
592
593 static size_t fractional_time_len(const char *line, size_t len)
594 {
595 const char *p;
596 size_t n;
597
598 /* Expected format: 19:41:17.620000023 */
599 if (!len || !isdigit(line[len - 1]))
600 return 0;
601 p = line + len - 1;
602
603 /* Fractional seconds. */
604 while (p > line && isdigit(*p))
605 p--;
606 if (*p != '.')
607 return 0;
608
609 /* Hours, minutes, and whole seconds. */
610 n = short_time_len(line, p - line);
611 if (!n)
612 return 0;
613
614 return line + len - p + n;
615 }
616
617 static size_t trailing_spaces_len(const char *line, size_t len)
618 {
619 const char *p;
620
621 /* Expected format: ' ' x (1 or more) */
622 if (!len || line[len - 1] != ' ')
623 return 0;
624
625 p = line + len;
626 while (p != line) {
627 p--;
628 if (*p != ' ')
629 return line + len - (p + 1);
630 }
631
632 /* All spaces! */
633 return len;
634 }
635
636 static size_t diff_timestamp_len(const char *line, size_t len)
637 {
638 const char *end = line + len;
639 size_t n;
640
641 /*
642 * Posix: 2010-07-05 19:41:17
643 * GNU: 2010-07-05 19:41:17.620000023 -0500
644 */
645
646 if (!isdigit(end[-1]))
647 return 0;
648
649 n = sane_tz_len(line, end - line);
650 if (!n)
651 n = tz_with_colon_len(line, end - line);
652 end -= n;
653
654 n = short_time_len(line, end - line);
655 if (!n)
656 n = fractional_time_len(line, end - line);
657 end -= n;
658
659 n = date_len(line, end - line);
660 if (!n) /* No date. Too bad. */
661 return 0;
662 end -= n;
663
664 if (end == line) /* No space before date. */
665 return 0;
666 if (end[-1] == '\t') { /* Success! */
667 end--;
668 return line + len - end;
669 }
670 if (end[-1] != ' ') /* No space before date. */
671 return 0;
672
673 /* Whitespace damage. */
674 end -= trailing_spaces_len(line, end - line);
675 return line + len - end;
676 }
677
678 static char *find_name_common(struct apply_state *state,
679 const char *line,
680 const char *def,
681 int p_value,
682 const char *end,
683 int terminate)
684 {
685 int len;
686 const char *start = NULL;
687
688 if (p_value == 0)
689 start = line;
690 while (line != end) {
691 char c = *line;
692
693 if (!end && isspace(c)) {
694 if (c == '\n')
695 break;
696 if (name_terminate(c, terminate))
697 break;
698 }
699 line++;
700 if (c == '/' && !--p_value)
701 start = line;
702 }
703 if (!start)
704 return squash_slash(xstrdup_or_null(def));
705 len = line - start;
706 if (!len)
707 return squash_slash(xstrdup_or_null(def));
708
709 /*
710 * Generally we prefer the shorter name, especially
711 * if the other one is just a variation of that with
712 * something else tacked on to the end (ie "file.orig"
713 * or "file~").
714 */
715 if (def) {
716 int deflen = strlen(def);
717 if (deflen < len && !strncmp(start, def, deflen))
718 return squash_slash(xstrdup(def));
719 }
720
721 if (state->root.len) {
722 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
723 return squash_slash(ret);
724 }
725
726 return squash_slash(xmemdupz(start, len));
727 }
728
729 static char *find_name(struct apply_state *state,
730 const char *line,
731 char *def,
732 int p_value,
733 int terminate)
734 {
735 if (*line == '"') {
736 char *name = find_name_gnu(state, line, def, p_value);
737 if (name)
738 return name;
739 }
740
741 return find_name_common(state, line, def, p_value, NULL, terminate);
742 }
743
744 static char *find_name_traditional(struct apply_state *state,
745 const char *line,
746 char *def,
747 int p_value)
748 {
749 size_t len;
750 size_t date_len;
751
752 if (*line == '"') {
753 char *name = find_name_gnu(state, line, def, p_value);
754 if (name)
755 return name;
756 }
757
758 len = strchrnul(line, '\n') - line;
759 date_len = diff_timestamp_len(line, len);
760 if (!date_len)
761 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
762 len -= date_len;
763
764 return find_name_common(state, line, def, p_value, line + len, 0);
765 }
766
767 /*
768 * Given the string after "--- " or "+++ ", guess the appropriate
769 * p_value for the given patch.
770 */
771 static int guess_p_value(struct apply_state *state, const char *nameline)
772 {
773 char *name, *cp;
774 int val = -1;
775
776 if (is_dev_null(nameline))
777 return -1;
778 name = find_name_traditional(state, nameline, NULL, 0);
779 if (!name)
780 return -1;
781 cp = strchr(name, '/');
782 if (!cp)
783 val = 0;
784 else if (state->prefix) {
785 /*
786 * Does it begin with "a/$our-prefix" and such? Then this is
787 * very likely to apply to our directory.
788 */
789 if (starts_with(name, state->prefix))
790 val = count_slashes(state->prefix);
791 else {
792 cp++;
793 if (starts_with(cp, state->prefix))
794 val = count_slashes(state->prefix) + 1;
795 }
796 }
797 free(name);
798 return val;
799 }
800
801 /*
802 * Does the ---/+++ line have the POSIX timestamp after the last HT?
803 * GNU diff puts epoch there to signal a creation/deletion event. Is
804 * this such a timestamp?
805 */
806 static int has_epoch_timestamp(const char *nameline)
807 {
808 /*
809 * We are only interested in epoch timestamp; any non-zero
810 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
811 * For the same reason, the date must be either 1969-12-31 or
812 * 1970-01-01, and the seconds part must be "00".
813 */
814 const char stamp_regexp[] =
815 "^(1969-12-31|1970-01-01)"
816 " "
817 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
818 " "
819 "([-+][0-2][0-9]:?[0-5][0-9])\n";
820 const char *timestamp = NULL, *cp, *colon;
821 static regex_t *stamp;
822 regmatch_t m[10];
823 int zoneoffset;
824 int hourminute;
825 int status;
826
827 for (cp = nameline; *cp != '\n'; cp++) {
828 if (*cp == '\t')
829 timestamp = cp + 1;
830 }
831 if (!timestamp)
832 return 0;
833 if (!stamp) {
834 stamp = xmalloc(sizeof(*stamp));
835 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
836 warning(_("Cannot prepare timestamp regexp %s"),
837 stamp_regexp);
838 return 0;
839 }
840 }
841
842 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
843 if (status) {
844 if (status != REG_NOMATCH)
845 warning(_("regexec returned %d for input: %s"),
846 status, timestamp);
847 return 0;
848 }
849
850 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
851 if (*colon == ':')
852 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
853 else
854 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
855 if (timestamp[m[3].rm_so] == '-')
856 zoneoffset = -zoneoffset;
857
858 /*
859 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
860 * (west of GMT) or 1970-01-01 (east of GMT)
861 */
862 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
863 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
864 return 0;
865
866 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
867 strtol(timestamp + 14, NULL, 10) -
868 zoneoffset);
869
870 return ((zoneoffset < 0 && hourminute == 1440) ||
871 (0 <= zoneoffset && !hourminute));
872 }
873
874 /*
875 * Get the name etc info from the ---/+++ lines of a traditional patch header
876 *
877 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
878 * files, we can happily check the index for a match, but for creating a
879 * new file we should try to match whatever "patch" does. I have no idea.
880 */
881 static int parse_traditional_patch(struct apply_state *state,
882 const char *first,
883 const char *second,
884 struct patch *patch)
885 {
886 char *name;
887
888 first += 4; /* skip "--- " */
889 second += 4; /* skip "+++ " */
890 if (!state->p_value_known) {
891 int p, q;
892 p = guess_p_value(state, first);
893 q = guess_p_value(state, second);
894 if (p < 0) p = q;
895 if (0 <= p && p == q) {
896 state->p_value = p;
897 state->p_value_known = 1;
898 }
899 }
900 if (is_dev_null(first)) {
901 patch->is_new = 1;
902 patch->is_delete = 0;
903 name = find_name_traditional(state, second, NULL, state->p_value);
904 patch->new_name = name;
905 } else if (is_dev_null(second)) {
906 patch->is_new = 0;
907 patch->is_delete = 1;
908 name = find_name_traditional(state, first, NULL, state->p_value);
909 patch->old_name = name;
910 } else {
911 char *first_name;
912 first_name = find_name_traditional(state, first, NULL, state->p_value);
913 name = find_name_traditional(state, second, first_name, state->p_value);
914 free(first_name);
915 if (has_epoch_timestamp(first)) {
916 patch->is_new = 1;
917 patch->is_delete = 0;
918 patch->new_name = name;
919 } else if (has_epoch_timestamp(second)) {
920 patch->is_new = 0;
921 patch->is_delete = 1;
922 patch->old_name = name;
923 } else {
924 patch->old_name = name;
925 patch->new_name = xstrdup_or_null(name);
926 }
927 }
928 if (!name)
929 return error(_("unable to find filename in patch at line %d"), state->linenr);
930
931 return 0;
932 }
933
934 static int gitdiff_hdrend(struct apply_state *state,
935 const char *line,
936 struct patch *patch)
937 {
938 return 1;
939 }
940
941 /*
942 * We're anal about diff header consistency, to make
943 * sure that we don't end up having strange ambiguous
944 * patches floating around.
945 *
946 * As a result, gitdiff_{old|new}name() will check
947 * their names against any previous information, just
948 * to make sure..
949 */
950 #define DIFF_OLD_NAME 0
951 #define DIFF_NEW_NAME 1
952
953 static int gitdiff_verify_name(struct apply_state *state,
954 const char *line,
955 int isnull,
956 char **name,
957 int side)
958 {
959 if (!*name && !isnull) {
960 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
961 return 0;
962 }
963
964 if (*name) {
965 char *another;
966 if (isnull)
967 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
968 *name, state->linenr);
969 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
970 if (!another || strcmp(another, *name)) {
971 free(another);
972 return error((side == DIFF_NEW_NAME) ?
973 _("git apply: bad git-diff - inconsistent new filename on line %d") :
974 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
975 }
976 free(another);
977 } else {
978 if (!starts_with(line, "/dev/null\n"))
979 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
980 }
981
982 return 0;
983 }
984
985 static int gitdiff_oldname(struct apply_state *state,
986 const char *line,
987 struct patch *patch)
988 {
989 return gitdiff_verify_name(state, line,
990 patch->is_new, &patch->old_name,
991 DIFF_OLD_NAME);
992 }
993
994 static int gitdiff_newname(struct apply_state *state,
995 const char *line,
996 struct patch *patch)
997 {
998 return gitdiff_verify_name(state, line,
999 patch->is_delete, &patch->new_name,
1000 DIFF_NEW_NAME);
1001 }
1002
1003 static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
1004 {
1005 char *end;
1006 *mode = strtoul(line, &end, 8);
1007 if (end == line || !isspace(*end))
1008 return error(_("invalid mode on line %d: %s"), linenr, line);
1009 return 0;
1010 }
1011
1012 static int gitdiff_oldmode(struct apply_state *state,
1013 const char *line,
1014 struct patch *patch)
1015 {
1016 return parse_mode_line(line, state->linenr, &patch->old_mode);
1017 }
1018
1019 static int gitdiff_newmode(struct apply_state *state,
1020 const char *line,
1021 struct patch *patch)
1022 {
1023 return parse_mode_line(line, state->linenr, &patch->new_mode);
1024 }
1025
1026 static int gitdiff_delete(struct apply_state *state,
1027 const char *line,
1028 struct patch *patch)
1029 {
1030 patch->is_delete = 1;
1031 free(patch->old_name);
1032 patch->old_name = xstrdup_or_null(patch->def_name);
1033 return gitdiff_oldmode(state, line, patch);
1034 }
1035
1036 static int gitdiff_newfile(struct apply_state *state,
1037 const char *line,
1038 struct patch *patch)
1039 {
1040 patch->is_new = 1;
1041 free(patch->new_name);
1042 patch->new_name = xstrdup_or_null(patch->def_name);
1043 return gitdiff_newmode(state, line, patch);
1044 }
1045
1046 static int gitdiff_copysrc(struct apply_state *state,
1047 const char *line,
1048 struct patch *patch)
1049 {
1050 patch->is_copy = 1;
1051 free(patch->old_name);
1052 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1053 return 0;
1054 }
1055
1056 static int gitdiff_copydst(struct apply_state *state,
1057 const char *line,
1058 struct patch *patch)
1059 {
1060 patch->is_copy = 1;
1061 free(patch->new_name);
1062 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1063 return 0;
1064 }
1065
1066 static int gitdiff_renamesrc(struct apply_state *state,
1067 const char *line,
1068 struct patch *patch)
1069 {
1070 patch->is_rename = 1;
1071 free(patch->old_name);
1072 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1073 return 0;
1074 }
1075
1076 static int gitdiff_renamedst(struct apply_state *state,
1077 const char *line,
1078 struct patch *patch)
1079 {
1080 patch->is_rename = 1;
1081 free(patch->new_name);
1082 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1083 return 0;
1084 }
1085
1086 static int gitdiff_similarity(struct apply_state *state,
1087 const char *line,
1088 struct patch *patch)
1089 {
1090 unsigned long val = strtoul(line, NULL, 10);
1091 if (val <= 100)
1092 patch->score = val;
1093 return 0;
1094 }
1095
1096 static int gitdiff_dissimilarity(struct apply_state *state,
1097 const char *line,
1098 struct patch *patch)
1099 {
1100 unsigned long val = strtoul(line, NULL, 10);
1101 if (val <= 100)
1102 patch->score = val;
1103 return 0;
1104 }
1105
1106 static int gitdiff_index(struct apply_state *state,
1107 const char *line,
1108 struct patch *patch)
1109 {
1110 /*
1111 * index line is N hexadecimal, "..", N hexadecimal,
1112 * and optional space with octal mode.
1113 */
1114 const char *ptr, *eol;
1115 int len;
1116
1117 ptr = strchr(line, '.');
1118 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1119 return 0;
1120 len = ptr - line;
1121 memcpy(patch->old_sha1_prefix, line, len);
1122 patch->old_sha1_prefix[len] = 0;
1123
1124 line = ptr + 2;
1125 ptr = strchr(line, ' ');
1126 eol = strchrnul(line, '\n');
1127
1128 if (!ptr || eol < ptr)
1129 ptr = eol;
1130 len = ptr - line;
1131
1132 if (40 < len)
1133 return 0;
1134 memcpy(patch->new_sha1_prefix, line, len);
1135 patch->new_sha1_prefix[len] = 0;
1136 if (*ptr == ' ')
1137 return gitdiff_oldmode(state, ptr + 1, patch);
1138 return 0;
1139 }
1140
1141 /*
1142 * This is normal for a diff that doesn't change anything: we'll fall through
1143 * into the next diff. Tell the parser to break out.
1144 */
1145 static int gitdiff_unrecognized(struct apply_state *state,
1146 const char *line,
1147 struct patch *patch)
1148 {
1149 return 1;
1150 }
1151
1152 /*
1153 * Skip p_value leading components from "line"; as we do not accept
1154 * absolute paths, return NULL in that case.
1155 */
1156 static const char *skip_tree_prefix(struct apply_state *state,
1157 const char *line,
1158 int llen)
1159 {
1160 int nslash;
1161 int i;
1162
1163 if (!state->p_value)
1164 return (llen && line[0] == '/') ? NULL : line;
1165
1166 nslash = state->p_value;
1167 for (i = 0; i < llen; i++) {
1168 int ch = line[i];
1169 if (ch == '/' && --nslash <= 0)
1170 return (i == 0) ? NULL : &line[i + 1];
1171 }
1172 return NULL;
1173 }
1174
1175 /*
1176 * This is to extract the same name that appears on "diff --git"
1177 * line. We do not find and return anything if it is a rename
1178 * patch, and it is OK because we will find the name elsewhere.
1179 * We need to reliably find name only when it is mode-change only,
1180 * creation or deletion of an empty file. In any of these cases,
1181 * both sides are the same name under a/ and b/ respectively.
1182 */
1183 static char *git_header_name(struct apply_state *state,
1184 const char *line,
1185 int llen)
1186 {
1187 const char *name;
1188 const char *second = NULL;
1189 size_t len, line_len;
1190
1191 line += strlen("diff --git ");
1192 llen -= strlen("diff --git ");
1193
1194 if (*line == '"') {
1195 const char *cp;
1196 struct strbuf first = STRBUF_INIT;
1197 struct strbuf sp = STRBUF_INIT;
1198
1199 if (unquote_c_style(&first, line, &second))
1200 goto free_and_fail1;
1201
1202 /* strip the a/b prefix including trailing slash */
1203 cp = skip_tree_prefix(state, first.buf, first.len);
1204 if (!cp)
1205 goto free_and_fail1;
1206 strbuf_remove(&first, 0, cp - first.buf);
1207
1208 /*
1209 * second points at one past closing dq of name.
1210 * find the second name.
1211 */
1212 while ((second < line + llen) && isspace(*second))
1213 second++;
1214
1215 if (line + llen <= second)
1216 goto free_and_fail1;
1217 if (*second == '"') {
1218 if (unquote_c_style(&sp, second, NULL))
1219 goto free_and_fail1;
1220 cp = skip_tree_prefix(state, sp.buf, sp.len);
1221 if (!cp)
1222 goto free_and_fail1;
1223 /* They must match, otherwise ignore */
1224 if (strcmp(cp, first.buf))
1225 goto free_and_fail1;
1226 strbuf_release(&sp);
1227 return strbuf_detach(&first, NULL);
1228 }
1229
1230 /* unquoted second */
1231 cp = skip_tree_prefix(state, second, line + llen - second);
1232 if (!cp)
1233 goto free_and_fail1;
1234 if (line + llen - cp != first.len ||
1235 memcmp(first.buf, cp, first.len))
1236 goto free_and_fail1;
1237 return strbuf_detach(&first, NULL);
1238
1239 free_and_fail1:
1240 strbuf_release(&first);
1241 strbuf_release(&sp);
1242 return NULL;
1243 }
1244
1245 /* unquoted first name */
1246 name = skip_tree_prefix(state, line, llen);
1247 if (!name)
1248 return NULL;
1249
1250 /*
1251 * since the first name is unquoted, a dq if exists must be
1252 * the beginning of the second name.
1253 */
1254 for (second = name; second < line + llen; second++) {
1255 if (*second == '"') {
1256 struct strbuf sp = STRBUF_INIT;
1257 const char *np;
1258
1259 if (unquote_c_style(&sp, second, NULL))
1260 goto free_and_fail2;
1261
1262 np = skip_tree_prefix(state, sp.buf, sp.len);
1263 if (!np)
1264 goto free_and_fail2;
1265
1266 len = sp.buf + sp.len - np;
1267 if (len < second - name &&
1268 !strncmp(np, name, len) &&
1269 isspace(name[len])) {
1270 /* Good */
1271 strbuf_remove(&sp, 0, np - sp.buf);
1272 return strbuf_detach(&sp, NULL);
1273 }
1274
1275 free_and_fail2:
1276 strbuf_release(&sp);
1277 return NULL;
1278 }
1279 }
1280
1281 /*
1282 * Accept a name only if it shows up twice, exactly the same
1283 * form.
1284 */
1285 second = strchr(name, '\n');
1286 if (!second)
1287 return NULL;
1288 line_len = second - name;
1289 for (len = 0 ; ; len++) {
1290 switch (name[len]) {
1291 default:
1292 continue;
1293 case '\n':
1294 return NULL;
1295 case '\t': case ' ':
1296 /*
1297 * Is this the separator between the preimage
1298 * and the postimage pathname? Again, we are
1299 * only interested in the case where there is
1300 * no rename, as this is only to set def_name
1301 * and a rename patch has the names elsewhere
1302 * in an unambiguous form.
1303 */
1304 if (!name[len + 1])
1305 return NULL; /* no postimage name */
1306 second = skip_tree_prefix(state, name + len + 1,
1307 line_len - (len + 1));
1308 if (!second)
1309 return NULL;
1310 /*
1311 * Does len bytes starting at "name" and "second"
1312 * (that are separated by one HT or SP we just
1313 * found) exactly match?
1314 */
1315 if (second[len] == '\n' && !strncmp(name, second, len))
1316 return xmemdupz(name, len);
1317 }
1318 }
1319 }
1320
1321 static int check_header_line(struct apply_state *state, struct patch *patch)
1322 {
1323 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1324 (patch->is_rename == 1) + (patch->is_copy == 1);
1325 if (extensions > 1)
1326 return error(_("inconsistent header lines %d and %d"),
1327 patch->extension_linenr, state->linenr);
1328 if (extensions && !patch->extension_linenr)
1329 patch->extension_linenr = state->linenr;
1330 return 0;
1331 }
1332
1333 /* Verify that we recognize the lines following a git header */
1334 static int parse_git_header(struct apply_state *state,
1335 const char *line,
1336 int len,
1337 unsigned int size,
1338 struct patch *patch)
1339 {
1340 unsigned long offset;
1341
1342 /* A git diff has explicit new/delete information, so we don't guess */
1343 patch->is_new = 0;
1344 patch->is_delete = 0;
1345
1346 /*
1347 * Some things may not have the old name in the
1348 * rest of the headers anywhere (pure mode changes,
1349 * or removing or adding empty files), so we get
1350 * the default name from the header.
1351 */
1352 patch->def_name = git_header_name(state, line, len);
1353 if (patch->def_name && state->root.len) {
1354 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1355 free(patch->def_name);
1356 patch->def_name = s;
1357 }
1358
1359 line += len;
1360 size -= len;
1361 state->linenr++;
1362 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1363 static const struct opentry {
1364 const char *str;
1365 int (*fn)(struct apply_state *, const char *, struct patch *);
1366 } optable[] = {
1367 { "@@ -", gitdiff_hdrend },
1368 { "--- ", gitdiff_oldname },
1369 { "+++ ", gitdiff_newname },
1370 { "old mode ", gitdiff_oldmode },
1371 { "new mode ", gitdiff_newmode },
1372 { "deleted file mode ", gitdiff_delete },
1373 { "new file mode ", gitdiff_newfile },
1374 { "copy from ", gitdiff_copysrc },
1375 { "copy to ", gitdiff_copydst },
1376 { "rename old ", gitdiff_renamesrc },
1377 { "rename new ", gitdiff_renamedst },
1378 { "rename from ", gitdiff_renamesrc },
1379 { "rename to ", gitdiff_renamedst },
1380 { "similarity index ", gitdiff_similarity },
1381 { "dissimilarity index ", gitdiff_dissimilarity },
1382 { "index ", gitdiff_index },
1383 { "", gitdiff_unrecognized },
1384 };
1385 int i;
1386
1387 len = linelen(line, size);
1388 if (!len || line[len-1] != '\n')
1389 break;
1390 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1391 const struct opentry *p = optable + i;
1392 int oplen = strlen(p->str);
1393 int res;
1394 if (len < oplen || memcmp(p->str, line, oplen))
1395 continue;
1396 res = p->fn(state, line + oplen, patch);
1397 if (res < 0)
1398 return -1;
1399 if (check_header_line(state, patch))
1400 return -1;
1401 if (res > 0)
1402 return offset;
1403 break;
1404 }
1405 }
1406
1407 return offset;
1408 }
1409
1410 static int parse_num(const char *line, unsigned long *p)
1411 {
1412 char *ptr;
1413
1414 if (!isdigit(*line))
1415 return 0;
1416 *p = strtoul(line, &ptr, 10);
1417 return ptr - line;
1418 }
1419
1420 static int parse_range(const char *line, int len, int offset, const char *expect,
1421 unsigned long *p1, unsigned long *p2)
1422 {
1423 int digits, ex;
1424
1425 if (offset < 0 || offset >= len)
1426 return -1;
1427 line += offset;
1428 len -= offset;
1429
1430 digits = parse_num(line, p1);
1431 if (!digits)
1432 return -1;
1433
1434 offset += digits;
1435 line += digits;
1436 len -= digits;
1437
1438 *p2 = 1;
1439 if (*line == ',') {
1440 digits = parse_num(line+1, p2);
1441 if (!digits)
1442 return -1;
1443
1444 offset += digits+1;
1445 line += digits+1;
1446 len -= digits+1;
1447 }
1448
1449 ex = strlen(expect);
1450 if (ex > len)
1451 return -1;
1452 if (memcmp(line, expect, ex))
1453 return -1;
1454
1455 return offset + ex;
1456 }
1457
1458 static void recount_diff(const char *line, int size, struct fragment *fragment)
1459 {
1460 int oldlines = 0, newlines = 0, ret = 0;
1461
1462 if (size < 1) {
1463 warning("recount: ignore empty hunk");
1464 return;
1465 }
1466
1467 for (;;) {
1468 int len = linelen(line, size);
1469 size -= len;
1470 line += len;
1471
1472 if (size < 1)
1473 break;
1474
1475 switch (*line) {
1476 case ' ': case '\n':
1477 newlines++;
1478 /* fall through */
1479 case '-':
1480 oldlines++;
1481 continue;
1482 case '+':
1483 newlines++;
1484 continue;
1485 case '\\':
1486 continue;
1487 case '@':
1488 ret = size < 3 || !starts_with(line, "@@ ");
1489 break;
1490 case 'd':
1491 ret = size < 5 || !starts_with(line, "diff ");
1492 break;
1493 default:
1494 ret = -1;
1495 break;
1496 }
1497 if (ret) {
1498 warning(_("recount: unexpected line: %.*s"),
1499 (int)linelen(line, size), line);
1500 return;
1501 }
1502 break;
1503 }
1504 fragment->oldlines = oldlines;
1505 fragment->newlines = newlines;
1506 }
1507
1508 /*
1509 * Parse a unified diff fragment header of the
1510 * form "@@ -a,b +c,d @@"
1511 */
1512 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1513 {
1514 int offset;
1515
1516 if (!len || line[len-1] != '\n')
1517 return -1;
1518
1519 /* Figure out the number of lines in a fragment */
1520 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1521 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1522
1523 return offset;
1524 }
1525
1526 /*
1527 * Find file diff header
1528 *
1529 * Returns:
1530 * -1 if no header was found
1531 * -128 in case of error
1532 * the size of the header in bytes (called "offset") otherwise
1533 */
1534 static int find_header(struct apply_state *state,
1535 const char *line,
1536 unsigned long size,
1537 int *hdrsize,
1538 struct patch *patch)
1539 {
1540 unsigned long offset, len;
1541
1542 patch->is_toplevel_relative = 0;
1543 patch->is_rename = patch->is_copy = 0;
1544 patch->is_new = patch->is_delete = -1;
1545 patch->old_mode = patch->new_mode = 0;
1546 patch->old_name = patch->new_name = NULL;
1547 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1548 unsigned long nextlen;
1549
1550 len = linelen(line, size);
1551 if (!len)
1552 break;
1553
1554 /* Testing this early allows us to take a few shortcuts.. */
1555 if (len < 6)
1556 continue;
1557
1558 /*
1559 * Make sure we don't find any unconnected patch fragments.
1560 * That's a sign that we didn't find a header, and that a
1561 * patch has become corrupted/broken up.
1562 */
1563 if (!memcmp("@@ -", line, 4)) {
1564 struct fragment dummy;
1565 if (parse_fragment_header(line, len, &dummy) < 0)
1566 continue;
1567 error(_("patch fragment without header at line %d: %.*s"),
1568 state->linenr, (int)len-1, line);
1569 return -128;
1570 }
1571
1572 if (size < len + 6)
1573 break;
1574
1575 /*
1576 * Git patch? It might not have a real patch, just a rename
1577 * or mode change, so we handle that specially
1578 */
1579 if (!memcmp("diff --git ", line, 11)) {
1580 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1581 if (git_hdr_len < 0)
1582 return -128;
1583 if (git_hdr_len <= len)
1584 continue;
1585 if (!patch->old_name && !patch->new_name) {
1586 if (!patch->def_name) {
1587 error(Q_("git diff header lacks filename information when removing "
1588 "%d leading pathname component (line %d)",
1589 "git diff header lacks filename information when removing "
1590 "%d leading pathname components (line %d)",
1591 state->p_value),
1592 state->p_value, state->linenr);
1593 return -128;
1594 }
1595 patch->old_name = xstrdup(patch->def_name);
1596 patch->new_name = xstrdup(patch->def_name);
1597 }
1598 if ((!patch->new_name && !patch->is_delete) ||
1599 (!patch->old_name && !patch->is_new)) {
1600 error(_("git diff header lacks filename information "
1601 "(line %d)"), state->linenr);
1602 return -128;
1603 }
1604 patch->is_toplevel_relative = 1;
1605 *hdrsize = git_hdr_len;
1606 return offset;
1607 }
1608
1609 /* --- followed by +++ ? */
1610 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1611 continue;
1612
1613 /*
1614 * We only accept unified patches, so we want it to
1615 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1616 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1617 */
1618 nextlen = linelen(line + len, size - len);
1619 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1620 continue;
1621
1622 /* Ok, we'll consider it a patch */
1623 if (parse_traditional_patch(state, line, line+len, patch))
1624 return -128;
1625 *hdrsize = len + nextlen;
1626 state->linenr += 2;
1627 return offset;
1628 }
1629 return -1;
1630 }
1631
1632 static void record_ws_error(struct apply_state *state,
1633 unsigned result,
1634 const char *line,
1635 int len,
1636 int linenr)
1637 {
1638 char *err;
1639
1640 if (!result)
1641 return;
1642
1643 state->whitespace_error++;
1644 if (state->squelch_whitespace_errors &&
1645 state->squelch_whitespace_errors < state->whitespace_error)
1646 return;
1647
1648 err = whitespace_error_string(result);
1649 if (state->apply_verbosity > verbosity_silent)
1650 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1651 state->patch_input_file, linenr, err, len, line);
1652 free(err);
1653 }
1654
1655 static void check_whitespace(struct apply_state *state,
1656 const char *line,
1657 int len,
1658 unsigned ws_rule)
1659 {
1660 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1661
1662 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1663 }
1664
1665 /*
1666 * Check if the patch has context lines with CRLF or
1667 * the patch wants to remove lines with CRLF.
1668 */
1669 static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1670 {
1671 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1672 patch->ws_rule |= WS_CR_AT_EOL;
1673 patch->crlf_in_old = 1;
1674 }
1675 }
1676
1677
1678 /*
1679 * Parse a unified diff. Note that this really needs to parse each
1680 * fragment separately, since the only way to know the difference
1681 * between a "---" that is part of a patch, and a "---" that starts
1682 * the next patch is to look at the line counts..
1683 */
1684 static int parse_fragment(struct apply_state *state,
1685 const char *line,
1686 unsigned long size,
1687 struct patch *patch,
1688 struct fragment *fragment)
1689 {
1690 int added, deleted;
1691 int len = linelen(line, size), offset;
1692 unsigned long oldlines, newlines;
1693 unsigned long leading, trailing;
1694
1695 offset = parse_fragment_header(line, len, fragment);
1696 if (offset < 0)
1697 return -1;
1698 if (offset > 0 && patch->recount)
1699 recount_diff(line + offset, size - offset, fragment);
1700 oldlines = fragment->oldlines;
1701 newlines = fragment->newlines;
1702 leading = 0;
1703 trailing = 0;
1704
1705 /* Parse the thing.. */
1706 line += len;
1707 size -= len;
1708 state->linenr++;
1709 added = deleted = 0;
1710 for (offset = len;
1711 0 < size;
1712 offset += len, size -= len, line += len, state->linenr++) {
1713 if (!oldlines && !newlines)
1714 break;
1715 len = linelen(line, size);
1716 if (!len || line[len-1] != '\n')
1717 return -1;
1718 switch (*line) {
1719 default:
1720 return -1;
1721 case '\n': /* newer GNU diff, an empty context line */
1722 case ' ':
1723 oldlines--;
1724 newlines--;
1725 if (!deleted && !added)
1726 leading++;
1727 trailing++;
1728 check_old_for_crlf(patch, line, len);
1729 if (!state->apply_in_reverse &&
1730 state->ws_error_action == correct_ws_error)
1731 check_whitespace(state, line, len, patch->ws_rule);
1732 break;
1733 case '-':
1734 if (!state->apply_in_reverse)
1735 check_old_for_crlf(patch, line, len);
1736 if (state->apply_in_reverse &&
1737 state->ws_error_action != nowarn_ws_error)
1738 check_whitespace(state, line, len, patch->ws_rule);
1739 deleted++;
1740 oldlines--;
1741 trailing = 0;
1742 break;
1743 case '+':
1744 if (state->apply_in_reverse)
1745 check_old_for_crlf(patch, line, len);
1746 if (!state->apply_in_reverse &&
1747 state->ws_error_action != nowarn_ws_error)
1748 check_whitespace(state, line, len, patch->ws_rule);
1749 added++;
1750 newlines--;
1751 trailing = 0;
1752 break;
1753
1754 /*
1755 * We allow "\ No newline at end of file". Depending
1756 * on locale settings when the patch was produced we
1757 * don't know what this line looks like. The only
1758 * thing we do know is that it begins with "\ ".
1759 * Checking for 12 is just for sanity check -- any
1760 * l10n of "\ No newline..." is at least that long.
1761 */
1762 case '\\':
1763 if (len < 12 || memcmp(line, "\\ ", 2))
1764 return -1;
1765 break;
1766 }
1767 }
1768 if (oldlines || newlines)
1769 return -1;
1770 if (!deleted && !added)
1771 return -1;
1772
1773 fragment->leading = leading;
1774 fragment->trailing = trailing;
1775
1776 /*
1777 * If a fragment ends with an incomplete line, we failed to include
1778 * it in the above loop because we hit oldlines == newlines == 0
1779 * before seeing it.
1780 */
1781 if (12 < size && !memcmp(line, "\\ ", 2))
1782 offset += linelen(line, size);
1783
1784 patch->lines_added += added;
1785 patch->lines_deleted += deleted;
1786
1787 if (0 < patch->is_new && oldlines)
1788 return error(_("new file depends on old contents"));
1789 if (0 < patch->is_delete && newlines)
1790 return error(_("deleted file still has contents"));
1791 return offset;
1792 }
1793
1794 /*
1795 * We have seen "diff --git a/... b/..." header (or a traditional patch
1796 * header). Read hunks that belong to this patch into fragments and hang
1797 * them to the given patch structure.
1798 *
1799 * The (fragment->patch, fragment->size) pair points into the memory given
1800 * by the caller, not a copy, when we return.
1801 *
1802 * Returns:
1803 * -1 in case of error,
1804 * the number of bytes in the patch otherwise.
1805 */
1806 static int parse_single_patch(struct apply_state *state,
1807 const char *line,
1808 unsigned long size,
1809 struct patch *patch)
1810 {
1811 unsigned long offset = 0;
1812 unsigned long oldlines = 0, newlines = 0, context = 0;
1813 struct fragment **fragp = &patch->fragments;
1814
1815 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1816 struct fragment *fragment;
1817 int len;
1818
1819 fragment = xcalloc(1, sizeof(*fragment));
1820 fragment->linenr = state->linenr;
1821 len = parse_fragment(state, line, size, patch, fragment);
1822 if (len <= 0) {
1823 free(fragment);
1824 return error(_("corrupt patch at line %d"), state->linenr);
1825 }
1826 fragment->patch = line;
1827 fragment->size = len;
1828 oldlines += fragment->oldlines;
1829 newlines += fragment->newlines;
1830 context += fragment->leading + fragment->trailing;
1831
1832 *fragp = fragment;
1833 fragp = &fragment->next;
1834
1835 offset += len;
1836 line += len;
1837 size -= len;
1838 }
1839
1840 /*
1841 * If something was removed (i.e. we have old-lines) it cannot
1842 * be creation, and if something was added it cannot be
1843 * deletion. However, the reverse is not true; --unified=0
1844 * patches that only add are not necessarily creation even
1845 * though they do not have any old lines, and ones that only
1846 * delete are not necessarily deletion.
1847 *
1848 * Unfortunately, a real creation/deletion patch do _not_ have
1849 * any context line by definition, so we cannot safely tell it
1850 * apart with --unified=0 insanity. At least if the patch has
1851 * more than one hunk it is not creation or deletion.
1852 */
1853 if (patch->is_new < 0 &&
1854 (oldlines || (patch->fragments && patch->fragments->next)))
1855 patch->is_new = 0;
1856 if (patch->is_delete < 0 &&
1857 (newlines || (patch->fragments && patch->fragments->next)))
1858 patch->is_delete = 0;
1859
1860 if (0 < patch->is_new && oldlines)
1861 return error(_("new file %s depends on old contents"), patch->new_name);
1862 if (0 < patch->is_delete && newlines)
1863 return error(_("deleted file %s still has contents"), patch->old_name);
1864 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
1865 fprintf_ln(stderr,
1866 _("** warning: "
1867 "file %s becomes empty but is not deleted"),
1868 patch->new_name);
1869
1870 return offset;
1871 }
1872
1873 static inline int metadata_changes(struct patch *patch)
1874 {
1875 return patch->is_rename > 0 ||
1876 patch->is_copy > 0 ||
1877 patch->is_new > 0 ||
1878 patch->is_delete ||
1879 (patch->old_mode && patch->new_mode &&
1880 patch->old_mode != patch->new_mode);
1881 }
1882
1883 static char *inflate_it(const void *data, unsigned long size,
1884 unsigned long inflated_size)
1885 {
1886 git_zstream stream;
1887 void *out;
1888 int st;
1889
1890 memset(&stream, 0, sizeof(stream));
1891
1892 stream.next_in = (unsigned char *)data;
1893 stream.avail_in = size;
1894 stream.next_out = out = xmalloc(inflated_size);
1895 stream.avail_out = inflated_size;
1896 git_inflate_init(&stream);
1897 st = git_inflate(&stream, Z_FINISH);
1898 git_inflate_end(&stream);
1899 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1900 free(out);
1901 return NULL;
1902 }
1903 return out;
1904 }
1905
1906 /*
1907 * Read a binary hunk and return a new fragment; fragment->patch
1908 * points at an allocated memory that the caller must free, so
1909 * it is marked as "->free_patch = 1".
1910 */
1911 static struct fragment *parse_binary_hunk(struct apply_state *state,
1912 char **buf_p,
1913 unsigned long *sz_p,
1914 int *status_p,
1915 int *used_p)
1916 {
1917 /*
1918 * Expect a line that begins with binary patch method ("literal"
1919 * or "delta"), followed by the length of data before deflating.
1920 * a sequence of 'length-byte' followed by base-85 encoded data
1921 * should follow, terminated by a newline.
1922 *
1923 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1924 * and we would limit the patch line to 66 characters,
1925 * so one line can fit up to 13 groups that would decode
1926 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1927 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1928 */
1929 int llen, used;
1930 unsigned long size = *sz_p;
1931 char *buffer = *buf_p;
1932 int patch_method;
1933 unsigned long origlen;
1934 char *data = NULL;
1935 int hunk_size = 0;
1936 struct fragment *frag;
1937
1938 llen = linelen(buffer, size);
1939 used = llen;
1940
1941 *status_p = 0;
1942
1943 if (starts_with(buffer, "delta ")) {
1944 patch_method = BINARY_DELTA_DEFLATED;
1945 origlen = strtoul(buffer + 6, NULL, 10);
1946 }
1947 else if (starts_with(buffer, "literal ")) {
1948 patch_method = BINARY_LITERAL_DEFLATED;
1949 origlen = strtoul(buffer + 8, NULL, 10);
1950 }
1951 else
1952 return NULL;
1953
1954 state->linenr++;
1955 buffer += llen;
1956 while (1) {
1957 int byte_length, max_byte_length, newsize;
1958 llen = linelen(buffer, size);
1959 used += llen;
1960 state->linenr++;
1961 if (llen == 1) {
1962 /* consume the blank line */
1963 buffer++;
1964 size--;
1965 break;
1966 }
1967 /*
1968 * Minimum line is "A00000\n" which is 7-byte long,
1969 * and the line length must be multiple of 5 plus 2.
1970 */
1971 if ((llen < 7) || (llen-2) % 5)
1972 goto corrupt;
1973 max_byte_length = (llen - 2) / 5 * 4;
1974 byte_length = *buffer;
1975 if ('A' <= byte_length && byte_length <= 'Z')
1976 byte_length = byte_length - 'A' + 1;
1977 else if ('a' <= byte_length && byte_length <= 'z')
1978 byte_length = byte_length - 'a' + 27;
1979 else
1980 goto corrupt;
1981 /* if the input length was not multiple of 4, we would
1982 * have filler at the end but the filler should never
1983 * exceed 3 bytes
1984 */
1985 if (max_byte_length < byte_length ||
1986 byte_length <= max_byte_length - 4)
1987 goto corrupt;
1988 newsize = hunk_size + byte_length;
1989 data = xrealloc(data, newsize);
1990 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1991 goto corrupt;
1992 hunk_size = newsize;
1993 buffer += llen;
1994 size -= llen;
1995 }
1996
1997 frag = xcalloc(1, sizeof(*frag));
1998 frag->patch = inflate_it(data, hunk_size, origlen);
1999 frag->free_patch = 1;
2000 if (!frag->patch)
2001 goto corrupt;
2002 free(data);
2003 frag->size = origlen;
2004 *buf_p = buffer;
2005 *sz_p = size;
2006 *used_p = used;
2007 frag->binary_patch_method = patch_method;
2008 return frag;
2009
2010 corrupt:
2011 free(data);
2012 *status_p = -1;
2013 error(_("corrupt binary patch at line %d: %.*s"),
2014 state->linenr-1, llen-1, buffer);
2015 return NULL;
2016 }
2017
2018 /*
2019 * Returns:
2020 * -1 in case of error,
2021 * the length of the parsed binary patch otherwise
2022 */
2023 static int parse_binary(struct apply_state *state,
2024 char *buffer,
2025 unsigned long size,
2026 struct patch *patch)
2027 {
2028 /*
2029 * We have read "GIT binary patch\n"; what follows is a line
2030 * that says the patch method (currently, either "literal" or
2031 * "delta") and the length of data before deflating; a
2032 * sequence of 'length-byte' followed by base-85 encoded data
2033 * follows.
2034 *
2035 * When a binary patch is reversible, there is another binary
2036 * hunk in the same format, starting with patch method (either
2037 * "literal" or "delta") with the length of data, and a sequence
2038 * of length-byte + base-85 encoded data, terminated with another
2039 * empty line. This data, when applied to the postimage, produces
2040 * the preimage.
2041 */
2042 struct fragment *forward;
2043 struct fragment *reverse;
2044 int status;
2045 int used, used_1;
2046
2047 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2048 if (!forward && !status)
2049 /* there has to be one hunk (forward hunk) */
2050 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2051 if (status)
2052 /* otherwise we already gave an error message */
2053 return status;
2054
2055 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2056 if (reverse)
2057 used += used_1;
2058 else if (status) {
2059 /*
2060 * Not having reverse hunk is not an error, but having
2061 * a corrupt reverse hunk is.
2062 */
2063 free((void*) forward->patch);
2064 free(forward);
2065 return status;
2066 }
2067 forward->next = reverse;
2068 patch->fragments = forward;
2069 patch->is_binary = 1;
2070 return used;
2071 }
2072
2073 static void prefix_one(struct apply_state *state, char **name)
2074 {
2075 char *old_name = *name;
2076 if (!old_name)
2077 return;
2078 *name = prefix_filename(state->prefix, *name);
2079 free(old_name);
2080 }
2081
2082 static void prefix_patch(struct apply_state *state, struct patch *p)
2083 {
2084 if (!state->prefix || p->is_toplevel_relative)
2085 return;
2086 prefix_one(state, &p->new_name);
2087 prefix_one(state, &p->old_name);
2088 }
2089
2090 /*
2091 * include/exclude
2092 */
2093
2094 static void add_name_limit(struct apply_state *state,
2095 const char *name,
2096 int exclude)
2097 {
2098 struct string_list_item *it;
2099
2100 it = string_list_append(&state->limit_by_name, name);
2101 it->util = exclude ? NULL : (void *) 1;
2102 }
2103
2104 static int use_patch(struct apply_state *state, struct patch *p)
2105 {
2106 const char *pathname = p->new_name ? p->new_name : p->old_name;
2107 int i;
2108
2109 /* Paths outside are not touched regardless of "--include" */
2110 if (state->prefix && *state->prefix) {
2111 const char *rest;
2112 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
2113 return 0;
2114 }
2115
2116 /* See if it matches any of exclude/include rule */
2117 for (i = 0; i < state->limit_by_name.nr; i++) {
2118 struct string_list_item *it = &state->limit_by_name.items[i];
2119 if (!wildmatch(it->string, pathname, 0))
2120 return (it->util != NULL);
2121 }
2122
2123 /*
2124 * If we had any include, a path that does not match any rule is
2125 * not used. Otherwise, we saw bunch of exclude rules (or none)
2126 * and such a path is used.
2127 */
2128 return !state->has_include;
2129 }
2130
2131 /*
2132 * Read the patch text in "buffer" that extends for "size" bytes; stop
2133 * reading after seeing a single patch (i.e. changes to a single file).
2134 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2135 *
2136 * Returns:
2137 * -1 if no header was found or parse_binary() failed,
2138 * -128 on another error,
2139 * the number of bytes consumed otherwise,
2140 * so that the caller can call us again for the next patch.
2141 */
2142 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2143 {
2144 int hdrsize, patchsize;
2145 int offset = find_header(state, buffer, size, &hdrsize, patch);
2146
2147 if (offset < 0)
2148 return offset;
2149
2150 prefix_patch(state, patch);
2151
2152 if (!use_patch(state, patch))
2153 patch->ws_rule = 0;
2154 else
2155 patch->ws_rule = whitespace_rule(patch->new_name
2156 ? patch->new_name
2157 : patch->old_name);
2158
2159 patchsize = parse_single_patch(state,
2160 buffer + offset + hdrsize,
2161 size - offset - hdrsize,
2162 patch);
2163
2164 if (patchsize < 0)
2165 return -128;
2166
2167 if (!patchsize) {
2168 static const char git_binary[] = "GIT binary patch\n";
2169 int hd = hdrsize + offset;
2170 unsigned long llen = linelen(buffer + hd, size - hd);
2171
2172 if (llen == sizeof(git_binary) - 1 &&
2173 !memcmp(git_binary, buffer + hd, llen)) {
2174 int used;
2175 state->linenr++;
2176 used = parse_binary(state, buffer + hd + llen,
2177 size - hd - llen, patch);
2178 if (used < 0)
2179 return -1;
2180 if (used)
2181 patchsize = used + llen;
2182 else
2183 patchsize = 0;
2184 }
2185 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2186 static const char *binhdr[] = {
2187 "Binary files ",
2188 "Files ",
2189 NULL,
2190 };
2191 int i;
2192 for (i = 0; binhdr[i]; i++) {
2193 int len = strlen(binhdr[i]);
2194 if (len < size - hd &&
2195 !memcmp(binhdr[i], buffer + hd, len)) {
2196 state->linenr++;
2197 patch->is_binary = 1;
2198 patchsize = llen;
2199 break;
2200 }
2201 }
2202 }
2203
2204 /* Empty patch cannot be applied if it is a text patch
2205 * without metadata change. A binary patch appears
2206 * empty to us here.
2207 */
2208 if ((state->apply || state->check) &&
2209 (!patch->is_binary && !metadata_changes(patch))) {
2210 error(_("patch with only garbage at line %d"), state->linenr);
2211 return -128;
2212 }
2213 }
2214
2215 return offset + hdrsize + patchsize;
2216 }
2217
2218 static void reverse_patches(struct patch *p)
2219 {
2220 for (; p; p = p->next) {
2221 struct fragment *frag = p->fragments;
2222
2223 SWAP(p->new_name, p->old_name);
2224 SWAP(p->new_mode, p->old_mode);
2225 SWAP(p->is_new, p->is_delete);
2226 SWAP(p->lines_added, p->lines_deleted);
2227 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
2228
2229 for (; frag; frag = frag->next) {
2230 SWAP(frag->newpos, frag->oldpos);
2231 SWAP(frag->newlines, frag->oldlines);
2232 }
2233 }
2234 }
2235
2236 static const char pluses[] =
2237 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2238 static const char minuses[]=
2239 "----------------------------------------------------------------------";
2240
2241 static void show_stats(struct apply_state *state, struct patch *patch)
2242 {
2243 struct strbuf qname = STRBUF_INIT;
2244 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2245 int max, add, del;
2246
2247 quote_c_style(cp, &qname, NULL, 0);
2248
2249 /*
2250 * "scale" the filename
2251 */
2252 max = state->max_len;
2253 if (max > 50)
2254 max = 50;
2255
2256 if (qname.len > max) {
2257 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2258 if (!cp)
2259 cp = qname.buf + qname.len + 3 - max;
2260 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2261 }
2262
2263 if (patch->is_binary) {
2264 printf(" %-*s | Bin\n", max, qname.buf);
2265 strbuf_release(&qname);
2266 return;
2267 }
2268
2269 printf(" %-*s |", max, qname.buf);
2270 strbuf_release(&qname);
2271
2272 /*
2273 * scale the add/delete
2274 */
2275 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2276 add = patch->lines_added;
2277 del = patch->lines_deleted;
2278
2279 if (state->max_change > 0) {
2280 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2281 add = (add * max + state->max_change / 2) / state->max_change;
2282 del = total - add;
2283 }
2284 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2285 add, pluses, del, minuses);
2286 }
2287
2288 static int read_old_data(struct stat *st, struct patch *patch,
2289 const char *path, struct strbuf *buf)
2290 {
2291 enum safe_crlf safe_crlf = patch->crlf_in_old ?
2292 SAFE_CRLF_KEEP_CRLF : SAFE_CRLF_RENORMALIZE;
2293 switch (st->st_mode & S_IFMT) {
2294 case S_IFLNK:
2295 if (strbuf_readlink(buf, path, st->st_size) < 0)
2296 return error(_("unable to read symlink %s"), path);
2297 return 0;
2298 case S_IFREG:
2299 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2300 return error(_("unable to open or read %s"), path);
2301 /*
2302 * "git apply" without "--index/--cached" should never look
2303 * at the index; the target file may not have been added to
2304 * the index yet, and we may not even be in any Git repository.
2305 * Pass NULL to convert_to_git() to stress this; the function
2306 * should never look at the index when explicit crlf option
2307 * is given.
2308 */
2309 convert_to_git(NULL, path, buf->buf, buf->len, buf, safe_crlf);
2310 return 0;
2311 default:
2312 return -1;
2313 }
2314 }
2315
2316 /*
2317 * Update the preimage, and the common lines in postimage,
2318 * from buffer buf of length len. If postlen is 0 the postimage
2319 * is updated in place, otherwise it's updated on a new buffer
2320 * of length postlen
2321 */
2322
2323 static void update_pre_post_images(struct image *preimage,
2324 struct image *postimage,
2325 char *buf,
2326 size_t len, size_t postlen)
2327 {
2328 int i, ctx, reduced;
2329 char *new, *old, *fixed;
2330 struct image fixed_preimage;
2331
2332 /*
2333 * Update the preimage with whitespace fixes. Note that we
2334 * are not losing preimage->buf -- apply_one_fragment() will
2335 * free "oldlines".
2336 */
2337 prepare_image(&fixed_preimage, buf, len, 1);
2338 assert(postlen
2339 ? fixed_preimage.nr == preimage->nr
2340 : fixed_preimage.nr <= preimage->nr);
2341 for (i = 0; i < fixed_preimage.nr; i++)
2342 fixed_preimage.line[i].flag = preimage->line[i].flag;
2343 free(preimage->line_allocated);
2344 *preimage = fixed_preimage;
2345
2346 /*
2347 * Adjust the common context lines in postimage. This can be
2348 * done in-place when we are shrinking it with whitespace
2349 * fixing, but needs a new buffer when ignoring whitespace or
2350 * expanding leading tabs to spaces.
2351 *
2352 * We trust the caller to tell us if the update can be done
2353 * in place (postlen==0) or not.
2354 */
2355 old = postimage->buf;
2356 if (postlen)
2357 new = postimage->buf = xmalloc(postlen);
2358 else
2359 new = old;
2360 fixed = preimage->buf;
2361
2362 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2363 size_t l_len = postimage->line[i].len;
2364 if (!(postimage->line[i].flag & LINE_COMMON)) {
2365 /* an added line -- no counterparts in preimage */
2366 memmove(new, old, l_len);
2367 old += l_len;
2368 new += l_len;
2369 continue;
2370 }
2371
2372 /* a common context -- skip it in the original postimage */
2373 old += l_len;
2374
2375 /* and find the corresponding one in the fixed preimage */
2376 while (ctx < preimage->nr &&
2377 !(preimage->line[ctx].flag & LINE_COMMON)) {
2378 fixed += preimage->line[ctx].len;
2379 ctx++;
2380 }
2381
2382 /*
2383 * preimage is expected to run out, if the caller
2384 * fixed addition of trailing blank lines.
2385 */
2386 if (preimage->nr <= ctx) {
2387 reduced++;
2388 continue;
2389 }
2390
2391 /* and copy it in, while fixing the line length */
2392 l_len = preimage->line[ctx].len;
2393 memcpy(new, fixed, l_len);
2394 new += l_len;
2395 fixed += l_len;
2396 postimage->line[i].len = l_len;
2397 ctx++;
2398 }
2399
2400 if (postlen
2401 ? postlen < new - postimage->buf
2402 : postimage->len < new - postimage->buf)
2403 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2404 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2405
2406 /* Fix the length of the whole thing */
2407 postimage->len = new - postimage->buf;
2408 postimage->nr -= reduced;
2409 }
2410
2411 static int line_by_line_fuzzy_match(struct image *img,
2412 struct image *preimage,
2413 struct image *postimage,
2414 unsigned long try,
2415 int try_lno,
2416 int preimage_limit)
2417 {
2418 int i;
2419 size_t imgoff = 0;
2420 size_t preoff = 0;
2421 size_t postlen = postimage->len;
2422 size_t extra_chars;
2423 char *buf;
2424 char *preimage_eof;
2425 char *preimage_end;
2426 struct strbuf fixed;
2427 char *fixed_buf;
2428 size_t fixed_len;
2429
2430 for (i = 0; i < preimage_limit; i++) {
2431 size_t prelen = preimage->line[i].len;
2432 size_t imglen = img->line[try_lno+i].len;
2433
2434 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2435 preimage->buf + preoff, prelen))
2436 return 0;
2437 if (preimage->line[i].flag & LINE_COMMON)
2438 postlen += imglen - prelen;
2439 imgoff += imglen;
2440 preoff += prelen;
2441 }
2442
2443 /*
2444 * Ok, the preimage matches with whitespace fuzz.
2445 *
2446 * imgoff now holds the true length of the target that
2447 * matches the preimage before the end of the file.
2448 *
2449 * Count the number of characters in the preimage that fall
2450 * beyond the end of the file and make sure that all of them
2451 * are whitespace characters. (This can only happen if
2452 * we are removing blank lines at the end of the file.)
2453 */
2454 buf = preimage_eof = preimage->buf + preoff;
2455 for ( ; i < preimage->nr; i++)
2456 preoff += preimage->line[i].len;
2457 preimage_end = preimage->buf + preoff;
2458 for ( ; buf < preimage_end; buf++)
2459 if (!isspace(*buf))
2460 return 0;
2461
2462 /*
2463 * Update the preimage and the common postimage context
2464 * lines to use the same whitespace as the target.
2465 * If whitespace is missing in the target (i.e.
2466 * if the preimage extends beyond the end of the file),
2467 * use the whitespace from the preimage.
2468 */
2469 extra_chars = preimage_end - preimage_eof;
2470 strbuf_init(&fixed, imgoff + extra_chars);
2471 strbuf_add(&fixed, img->buf + try, imgoff);
2472 strbuf_add(&fixed, preimage_eof, extra_chars);
2473 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2474 update_pre_post_images(preimage, postimage,
2475 fixed_buf, fixed_len, postlen);
2476 return 1;
2477 }
2478
2479 static int match_fragment(struct apply_state *state,
2480 struct image *img,
2481 struct image *preimage,
2482 struct image *postimage,
2483 unsigned long try,
2484 int try_lno,
2485 unsigned ws_rule,
2486 int match_beginning, int match_end)
2487 {
2488 int i;
2489 char *fixed_buf, *buf, *orig, *target;
2490 struct strbuf fixed;
2491 size_t fixed_len, postlen;
2492 int preimage_limit;
2493
2494 if (preimage->nr + try_lno <= img->nr) {
2495 /*
2496 * The hunk falls within the boundaries of img.
2497 */
2498 preimage_limit = preimage->nr;
2499 if (match_end && (preimage->nr + try_lno != img->nr))
2500 return 0;
2501 } else if (state->ws_error_action == correct_ws_error &&
2502 (ws_rule & WS_BLANK_AT_EOF)) {
2503 /*
2504 * This hunk extends beyond the end of img, and we are
2505 * removing blank lines at the end of the file. This
2506 * many lines from the beginning of the preimage must
2507 * match with img, and the remainder of the preimage
2508 * must be blank.
2509 */
2510 preimage_limit = img->nr - try_lno;
2511 } else {
2512 /*
2513 * The hunk extends beyond the end of the img and
2514 * we are not removing blanks at the end, so we
2515 * should reject the hunk at this position.
2516 */
2517 return 0;
2518 }
2519
2520 if (match_beginning && try_lno)
2521 return 0;
2522
2523 /* Quick hash check */
2524 for (i = 0; i < preimage_limit; i++)
2525 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2526 (preimage->line[i].hash != img->line[try_lno + i].hash))
2527 return 0;
2528
2529 if (preimage_limit == preimage->nr) {
2530 /*
2531 * Do we have an exact match? If we were told to match
2532 * at the end, size must be exactly at try+fragsize,
2533 * otherwise try+fragsize must be still within the preimage,
2534 * and either case, the old piece should match the preimage
2535 * exactly.
2536 */
2537 if ((match_end
2538 ? (try + preimage->len == img->len)
2539 : (try + preimage->len <= img->len)) &&
2540 !memcmp(img->buf + try, preimage->buf, preimage->len))
2541 return 1;
2542 } else {
2543 /*
2544 * The preimage extends beyond the end of img, so
2545 * there cannot be an exact match.
2546 *
2547 * There must be one non-blank context line that match
2548 * a line before the end of img.
2549 */
2550 char *buf_end;
2551
2552 buf = preimage->buf;
2553 buf_end = buf;
2554 for (i = 0; i < preimage_limit; i++)
2555 buf_end += preimage->line[i].len;
2556
2557 for ( ; buf < buf_end; buf++)
2558 if (!isspace(*buf))
2559 break;
2560 if (buf == buf_end)
2561 return 0;
2562 }
2563
2564 /*
2565 * No exact match. If we are ignoring whitespace, run a line-by-line
2566 * fuzzy matching. We collect all the line length information because
2567 * we need it to adjust whitespace if we match.
2568 */
2569 if (state->ws_ignore_action == ignore_ws_change)
2570 return line_by_line_fuzzy_match(img, preimage, postimage,
2571 try, try_lno, preimage_limit);
2572
2573 if (state->ws_error_action != correct_ws_error)
2574 return 0;
2575
2576 /*
2577 * The hunk does not apply byte-by-byte, but the hash says
2578 * it might with whitespace fuzz. We weren't asked to
2579 * ignore whitespace, we were asked to correct whitespace
2580 * errors, so let's try matching after whitespace correction.
2581 *
2582 * While checking the preimage against the target, whitespace
2583 * errors in both fixed, we count how large the corresponding
2584 * postimage needs to be. The postimage prepared by
2585 * apply_one_fragment() has whitespace errors fixed on added
2586 * lines already, but the common lines were propagated as-is,
2587 * which may become longer when their whitespace errors are
2588 * fixed.
2589 */
2590
2591 /* First count added lines in postimage */
2592 postlen = 0;
2593 for (i = 0; i < postimage->nr; i++) {
2594 if (!(postimage->line[i].flag & LINE_COMMON))
2595 postlen += postimage->line[i].len;
2596 }
2597
2598 /*
2599 * The preimage may extend beyond the end of the file,
2600 * but in this loop we will only handle the part of the
2601 * preimage that falls within the file.
2602 */
2603 strbuf_init(&fixed, preimage->len + 1);
2604 orig = preimage->buf;
2605 target = img->buf + try;
2606 for (i = 0; i < preimage_limit; i++) {
2607 size_t oldlen = preimage->line[i].len;
2608 size_t tgtlen = img->line[try_lno + i].len;
2609 size_t fixstart = fixed.len;
2610 struct strbuf tgtfix;
2611 int match;
2612
2613 /* Try fixing the line in the preimage */
2614 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2615
2616 /* Try fixing the line in the target */
2617 strbuf_init(&tgtfix, tgtlen);
2618 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2619
2620 /*
2621 * If they match, either the preimage was based on
2622 * a version before our tree fixed whitespace breakage,
2623 * or we are lacking a whitespace-fix patch the tree
2624 * the preimage was based on already had (i.e. target
2625 * has whitespace breakage, the preimage doesn't).
2626 * In either case, we are fixing the whitespace breakages
2627 * so we might as well take the fix together with their
2628 * real change.
2629 */
2630 match = (tgtfix.len == fixed.len - fixstart &&
2631 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2632 fixed.len - fixstart));
2633
2634 /* Add the length if this is common with the postimage */
2635 if (preimage->line[i].flag & LINE_COMMON)
2636 postlen += tgtfix.len;
2637
2638 strbuf_release(&tgtfix);
2639 if (!match)
2640 goto unmatch_exit;
2641
2642 orig += oldlen;
2643 target += tgtlen;
2644 }
2645
2646
2647 /*
2648 * Now handle the lines in the preimage that falls beyond the
2649 * end of the file (if any). They will only match if they are
2650 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2651 * false).
2652 */
2653 for ( ; i < preimage->nr; i++) {
2654 size_t fixstart = fixed.len; /* start of the fixed preimage */
2655 size_t oldlen = preimage->line[i].len;
2656 int j;
2657
2658 /* Try fixing the line in the preimage */
2659 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2660
2661 for (j = fixstart; j < fixed.len; j++)
2662 if (!isspace(fixed.buf[j]))
2663 goto unmatch_exit;
2664
2665 orig += oldlen;
2666 }
2667
2668 /*
2669 * Yes, the preimage is based on an older version that still
2670 * has whitespace breakages unfixed, and fixing them makes the
2671 * hunk match. Update the context lines in the postimage.
2672 */
2673 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2674 if (postlen < postimage->len)
2675 postlen = 0;
2676 update_pre_post_images(preimage, postimage,
2677 fixed_buf, fixed_len, postlen);
2678 return 1;
2679
2680 unmatch_exit:
2681 strbuf_release(&fixed);
2682 return 0;
2683 }
2684
2685 static int find_pos(struct apply_state *state,
2686 struct image *img,
2687 struct image *preimage,
2688 struct image *postimage,
2689 int line,
2690 unsigned ws_rule,
2691 int match_beginning, int match_end)
2692 {
2693 int i;
2694 unsigned long backwards, forwards, try;
2695 int backwards_lno, forwards_lno, try_lno;
2696
2697 /*
2698 * If match_beginning or match_end is specified, there is no
2699 * point starting from a wrong line that will never match and
2700 * wander around and wait for a match at the specified end.
2701 */
2702 if (match_beginning)
2703 line = 0;
2704 else if (match_end)
2705 line = img->nr - preimage->nr;
2706
2707 /*
2708 * Because the comparison is unsigned, the following test
2709 * will also take care of a negative line number that can
2710 * result when match_end and preimage is larger than the target.
2711 */
2712 if ((size_t) line > img->nr)
2713 line = img->nr;
2714
2715 try = 0;
2716 for (i = 0; i < line; i++)
2717 try += img->line[i].len;
2718
2719 /*
2720 * There's probably some smart way to do this, but I'll leave
2721 * that to the smart and beautiful people. I'm simple and stupid.
2722 */
2723 backwards = try;
2724 backwards_lno = line;
2725 forwards = try;
2726 forwards_lno = line;
2727 try_lno = line;
2728
2729 for (i = 0; ; i++) {
2730 if (match_fragment(state, img, preimage, postimage,
2731 try, try_lno, ws_rule,
2732 match_beginning, match_end))
2733 return try_lno;
2734
2735 again:
2736 if (backwards_lno == 0 && forwards_lno == img->nr)
2737 break;
2738
2739 if (i & 1) {
2740 if (backwards_lno == 0) {
2741 i++;
2742 goto again;
2743 }
2744 backwards_lno--;
2745 backwards -= img->line[backwards_lno].len;
2746 try = backwards;
2747 try_lno = backwards_lno;
2748 } else {
2749 if (forwards_lno == img->nr) {
2750 i++;
2751 goto again;
2752 }
2753 forwards += img->line[forwards_lno].len;
2754 forwards_lno++;
2755 try = forwards;
2756 try_lno = forwards_lno;
2757 }
2758
2759 }
2760 return -1;
2761 }
2762
2763 static void remove_first_line(struct image *img)
2764 {
2765 img->buf += img->line[0].len;
2766 img->len -= img->line[0].len;
2767 img->line++;
2768 img->nr--;
2769 }
2770
2771 static void remove_last_line(struct image *img)
2772 {
2773 img->len -= img->line[--img->nr].len;
2774 }
2775
2776 /*
2777 * The change from "preimage" and "postimage" has been found to
2778 * apply at applied_pos (counts in line numbers) in "img".
2779 * Update "img" to remove "preimage" and replace it with "postimage".
2780 */
2781 static void update_image(struct apply_state *state,
2782 struct image *img,
2783 int applied_pos,
2784 struct image *preimage,
2785 struct image *postimage)
2786 {
2787 /*
2788 * remove the copy of preimage at offset in img
2789 * and replace it with postimage
2790 */
2791 int i, nr;
2792 size_t remove_count, insert_count, applied_at = 0;
2793 char *result;
2794 int preimage_limit;
2795
2796 /*
2797 * If we are removing blank lines at the end of img,
2798 * the preimage may extend beyond the end.
2799 * If that is the case, we must be careful only to
2800 * remove the part of the preimage that falls within
2801 * the boundaries of img. Initialize preimage_limit
2802 * to the number of lines in the preimage that falls
2803 * within the boundaries.
2804 */
2805 preimage_limit = preimage->nr;
2806 if (preimage_limit > img->nr - applied_pos)
2807 preimage_limit = img->nr - applied_pos;
2808
2809 for (i = 0; i < applied_pos; i++)
2810 applied_at += img->line[i].len;
2811
2812 remove_count = 0;
2813 for (i = 0; i < preimage_limit; i++)
2814 remove_count += img->line[applied_pos + i].len;
2815 insert_count = postimage->len;
2816
2817 /* Adjust the contents */
2818 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2819 memcpy(result, img->buf, applied_at);
2820 memcpy(result + applied_at, postimage->buf, postimage->len);
2821 memcpy(result + applied_at + postimage->len,
2822 img->buf + (applied_at + remove_count),
2823 img->len - (applied_at + remove_count));
2824 free(img->buf);
2825 img->buf = result;
2826 img->len += insert_count - remove_count;
2827 result[img->len] = '\0';
2828
2829 /* Adjust the line table */
2830 nr = img->nr + postimage->nr - preimage_limit;
2831 if (preimage_limit < postimage->nr) {
2832 /*
2833 * NOTE: this knows that we never call remove_first_line()
2834 * on anything other than pre/post image.
2835 */
2836 REALLOC_ARRAY(img->line, nr);
2837 img->line_allocated = img->line;
2838 }
2839 if (preimage_limit != postimage->nr)
2840 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2841 img->line + applied_pos + preimage_limit,
2842 img->nr - (applied_pos + preimage_limit));
2843 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
2844 if (!state->allow_overlap)
2845 for (i = 0; i < postimage->nr; i++)
2846 img->line[applied_pos + i].flag |= LINE_PATCHED;
2847 img->nr = nr;
2848 }
2849
2850 /*
2851 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2852 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2853 * replace the part of "img" with "postimage" text.
2854 */
2855 static int apply_one_fragment(struct apply_state *state,
2856 struct image *img, struct fragment *frag,
2857 int inaccurate_eof, unsigned ws_rule,
2858 int nth_fragment)
2859 {
2860 int match_beginning, match_end;
2861 const char *patch = frag->patch;
2862 int size = frag->size;
2863 char *old, *oldlines;
2864 struct strbuf newlines;
2865 int new_blank_lines_at_end = 0;
2866 int found_new_blank_lines_at_end = 0;
2867 int hunk_linenr = frag->linenr;
2868 unsigned long leading, trailing;
2869 int pos, applied_pos;
2870 struct image preimage;
2871 struct image postimage;
2872
2873 memset(&preimage, 0, sizeof(preimage));
2874 memset(&postimage, 0, sizeof(postimage));
2875 oldlines = xmalloc(size);
2876 strbuf_init(&newlines, size);
2877
2878 old = oldlines;
2879 while (size > 0) {
2880 char first;
2881 int len = linelen(patch, size);
2882 int plen;
2883 int added_blank_line = 0;
2884 int is_blank_context = 0;
2885 size_t start;
2886
2887 if (!len)
2888 break;
2889
2890 /*
2891 * "plen" is how much of the line we should use for
2892 * the actual patch data. Normally we just remove the
2893 * first character on the line, but if the line is
2894 * followed by "\ No newline", then we also remove the
2895 * last one (which is the newline, of course).
2896 */
2897 plen = len - 1;
2898 if (len < size && patch[len] == '\\')
2899 plen--;
2900 first = *patch;
2901 if (state->apply_in_reverse) {
2902 if (first == '-')
2903 first = '+';
2904 else if (first == '+')
2905 first = '-';
2906 }
2907
2908 switch (first) {
2909 case '\n':
2910 /* Newer GNU diff, empty context line */
2911 if (plen < 0)
2912 /* ... followed by '\No newline'; nothing */
2913 break;
2914 *old++ = '\n';
2915 strbuf_addch(&newlines, '\n');
2916 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2917 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2918 is_blank_context = 1;
2919 break;
2920 case ' ':
2921 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2922 ws_blank_line(patch + 1, plen, ws_rule))
2923 is_blank_context = 1;
2924 case '-':
2925 memcpy(old, patch + 1, plen);
2926 add_line_info(&preimage, old, plen,
2927 (first == ' ' ? LINE_COMMON : 0));
2928 old += plen;
2929 if (first == '-')
2930 break;
2931 /* Fall-through for ' ' */
2932 case '+':
2933 /* --no-add does not add new lines */
2934 if (first == '+' && state->no_add)
2935 break;
2936
2937 start = newlines.len;
2938 if (first != '+' ||
2939 !state->whitespace_error ||
2940 state->ws_error_action != correct_ws_error) {
2941 strbuf_add(&newlines, patch + 1, plen);
2942 }
2943 else {
2944 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2945 }
2946 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2947 (first == '+' ? 0 : LINE_COMMON));
2948 if (first == '+' &&
2949 (ws_rule & WS_BLANK_AT_EOF) &&
2950 ws_blank_line(patch + 1, plen, ws_rule))
2951 added_blank_line = 1;
2952 break;
2953 case '@': case '\\':
2954 /* Ignore it, we already handled it */
2955 break;
2956 default:
2957 if (state->apply_verbosity > verbosity_normal)
2958 error(_("invalid start of line: '%c'"), first);
2959 applied_pos = -1;
2960 goto out;
2961 }
2962 if (added_blank_line) {
2963 if (!new_blank_lines_at_end)
2964 found_new_blank_lines_at_end = hunk_linenr;
2965 new_blank_lines_at_end++;
2966 }
2967 else if (is_blank_context)
2968 ;
2969 else
2970 new_blank_lines_at_end = 0;
2971 patch += len;
2972 size -= len;
2973 hunk_linenr++;
2974 }
2975 if (inaccurate_eof &&
2976 old > oldlines && old[-1] == '\n' &&
2977 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2978 old--;
2979 strbuf_setlen(&newlines, newlines.len - 1);
2980 }
2981
2982 leading = frag->leading;
2983 trailing = frag->trailing;
2984
2985 /*
2986 * A hunk to change lines at the beginning would begin with
2987 * @@ -1,L +N,M @@
2988 * but we need to be careful. -U0 that inserts before the second
2989 * line also has this pattern.
2990 *
2991 * And a hunk to add to an empty file would begin with
2992 * @@ -0,0 +N,M @@
2993 *
2994 * In other words, a hunk that is (frag->oldpos <= 1) with or
2995 * without leading context must match at the beginning.
2996 */
2997 match_beginning = (!frag->oldpos ||
2998 (frag->oldpos == 1 && !state->unidiff_zero));
2999
3000 /*
3001 * A hunk without trailing lines must match at the end.
3002 * However, we simply cannot tell if a hunk must match end
3003 * from the lack of trailing lines if the patch was generated
3004 * with unidiff without any context.
3005 */
3006 match_end = !state->unidiff_zero && !trailing;
3007
3008 pos = frag->newpos ? (frag->newpos - 1) : 0;
3009 preimage.buf = oldlines;
3010 preimage.len = old - oldlines;
3011 postimage.buf = newlines.buf;
3012 postimage.len = newlines.len;
3013 preimage.line = preimage.line_allocated;
3014 postimage.line = postimage.line_allocated;
3015
3016 for (;;) {
3017
3018 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3019 ws_rule, match_beginning, match_end);
3020
3021 if (applied_pos >= 0)
3022 break;
3023
3024 /* Am I at my context limits? */
3025 if ((leading <= state->p_context) && (trailing <= state->p_context))
3026 break;
3027 if (match_beginning || match_end) {
3028 match_beginning = match_end = 0;
3029 continue;
3030 }
3031
3032 /*
3033 * Reduce the number of context lines; reduce both
3034 * leading and trailing if they are equal otherwise
3035 * just reduce the larger context.
3036 */
3037 if (leading >= trailing) {
3038 remove_first_line(&preimage);
3039 remove_first_line(&postimage);
3040 pos--;
3041 leading--;
3042 }
3043 if (trailing > leading) {
3044 remove_last_line(&preimage);
3045 remove_last_line(&postimage);
3046 trailing--;
3047 }
3048 }
3049
3050 if (applied_pos >= 0) {
3051 if (new_blank_lines_at_end &&
3052 preimage.nr + applied_pos >= img->nr &&
3053 (ws_rule & WS_BLANK_AT_EOF) &&
3054 state->ws_error_action != nowarn_ws_error) {
3055 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3056 found_new_blank_lines_at_end);
3057 if (state->ws_error_action == correct_ws_error) {
3058 while (new_blank_lines_at_end--)
3059 remove_last_line(&postimage);
3060 }
3061 /*
3062 * We would want to prevent write_out_results()
3063 * from taking place in apply_patch() that follows
3064 * the callchain led us here, which is:
3065 * apply_patch->check_patch_list->check_patch->
3066 * apply_data->apply_fragments->apply_one_fragment
3067 */
3068 if (state->ws_error_action == die_on_ws_error)
3069 state->apply = 0;
3070 }
3071
3072 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
3073 int offset = applied_pos - pos;
3074 if (state->apply_in_reverse)
3075 offset = 0 - offset;
3076 fprintf_ln(stderr,
3077 Q_("Hunk #%d succeeded at %d (offset %d line).",
3078 "Hunk #%d succeeded at %d (offset %d lines).",
3079 offset),
3080 nth_fragment, applied_pos + 1, offset);
3081 }
3082
3083 /*
3084 * Warn if it was necessary to reduce the number
3085 * of context lines.
3086 */
3087 if ((leading != frag->leading ||
3088 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
3089 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3090 " to apply fragment at %d"),
3091 leading, trailing, applied_pos+1);
3092 update_image(state, img, applied_pos, &preimage, &postimage);
3093 } else {
3094 if (state->apply_verbosity > verbosity_normal)
3095 error(_("while searching for:\n%.*s"),
3096 (int)(old - oldlines), oldlines);
3097 }
3098
3099 out:
3100 free(oldlines);
3101 strbuf_release(&newlines);
3102 free(preimage.line_allocated);
3103 free(postimage.line_allocated);
3104
3105 return (applied_pos < 0);
3106 }
3107
3108 static int apply_binary_fragment(struct apply_state *state,
3109 struct image *img,
3110 struct patch *patch)
3111 {
3112 struct fragment *fragment = patch->fragments;
3113 unsigned long len;
3114 void *dst;
3115
3116 if (!fragment)
3117 return error(_("missing binary patch data for '%s'"),
3118 patch->new_name ?
3119 patch->new_name :
3120 patch->old_name);
3121
3122 /* Binary patch is irreversible without the optional second hunk */
3123 if (state->apply_in_reverse) {
3124 if (!fragment->next)
3125 return error(_("cannot reverse-apply a binary patch "
3126 "without the reverse hunk to '%s'"),
3127 patch->new_name
3128 ? patch->new_name : patch->old_name);
3129 fragment = fragment->next;
3130 }
3131 switch (fragment->binary_patch_method) {
3132 case BINARY_DELTA_DEFLATED:
3133 dst = patch_delta(img->buf, img->len, fragment->patch,
3134 fragment->size, &len);
3135 if (!dst)
3136 return -1;
3137 clear_image(img);
3138 img->buf = dst;
3139 img->len = len;
3140 return 0;
3141 case BINARY_LITERAL_DEFLATED:
3142 clear_image(img);
3143 img->len = fragment->size;
3144 img->buf = xmemdupz(fragment->patch, img->len);
3145 return 0;
3146 }
3147 return -1;
3148 }
3149
3150 /*
3151 * Replace "img" with the result of applying the binary patch.
3152 * The binary patch data itself in patch->fragment is still kept
3153 * but the preimage prepared by the caller in "img" is freed here
3154 * or in the helper function apply_binary_fragment() this calls.
3155 */
3156 static int apply_binary(struct apply_state *state,
3157 struct image *img,
3158 struct patch *patch)
3159 {
3160 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3161 struct object_id oid;
3162
3163 /*
3164 * For safety, we require patch index line to contain
3165 * full 40-byte textual SHA1 for old and new, at least for now.
3166 */
3167 if (strlen(patch->old_sha1_prefix) != 40 ||
3168 strlen(patch->new_sha1_prefix) != 40 ||
3169 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3170 get_oid_hex(patch->new_sha1_prefix, &oid))
3171 return error(_("cannot apply binary patch to '%s' "
3172 "without full index line"), name);
3173
3174 if (patch->old_name) {
3175 /*
3176 * See if the old one matches what the patch
3177 * applies to.
3178 */
3179 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3180 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
3181 return error(_("the patch applies to '%s' (%s), "
3182 "which does not match the "
3183 "current contents."),
3184 name, oid_to_hex(&oid));
3185 }
3186 else {
3187 /* Otherwise, the old one must be empty. */
3188 if (img->len)
3189 return error(_("the patch applies to an empty "
3190 "'%s' but it is not empty"), name);
3191 }
3192
3193 get_oid_hex(patch->new_sha1_prefix, &oid);
3194 if (is_null_oid(&oid)) {
3195 clear_image(img);
3196 return 0; /* deletion patch */
3197 }
3198
3199 if (has_sha1_file(oid.hash)) {
3200 /* We already have the postimage */
3201 enum object_type type;
3202 unsigned long size;
3203 char *result;
3204
3205 result = read_sha1_file(oid.hash, &type, &size);
3206 if (!result)
3207 return error(_("the necessary postimage %s for "
3208 "'%s' cannot be read"),
3209 patch->new_sha1_prefix, name);
3210 clear_image(img);
3211 img->buf = result;
3212 img->len = size;
3213 } else {
3214 /*
3215 * We have verified buf matches the preimage;
3216 * apply the patch data to it, which is stored
3217 * in the patch->fragments->{patch,size}.
3218 */
3219 if (apply_binary_fragment(state, img, patch))
3220 return error(_("binary patch does not apply to '%s'"),
3221 name);
3222
3223 /* verify that the result matches */
3224 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3225 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
3226 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3227 name, patch->new_sha1_prefix, oid_to_hex(&oid));
3228 }
3229
3230 return 0;
3231 }
3232
3233 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3234 {
3235 struct fragment *frag = patch->fragments;
3236 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3237 unsigned ws_rule = patch->ws_rule;
3238 unsigned inaccurate_eof = patch->inaccurate_eof;
3239 int nth = 0;
3240
3241 if (patch->is_binary)
3242 return apply_binary(state, img, patch);
3243
3244 while (frag) {
3245 nth++;
3246 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3247 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3248 if (!state->apply_with_reject)
3249 return -1;
3250 frag->rejected = 1;
3251 }
3252 frag = frag->next;
3253 }
3254 return 0;
3255 }
3256
3257 static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
3258 {
3259 if (S_ISGITLINK(mode)) {
3260 strbuf_grow(buf, 100);
3261 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
3262 } else {
3263 enum object_type type;
3264 unsigned long sz;
3265 char *result;
3266
3267 result = read_sha1_file(oid->hash, &type, &sz);
3268 if (!result)
3269 return -1;
3270 /* XXX read_sha1_file NUL-terminates */
3271 strbuf_attach(buf, result, sz, sz + 1);
3272 }
3273 return 0;
3274 }
3275
3276 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3277 {
3278 if (!ce)
3279 return 0;
3280 return read_blob_object(buf, &ce->oid, ce->ce_mode);
3281 }
3282
3283 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3284 {
3285 struct string_list_item *item;
3286
3287 if (name == NULL)
3288 return NULL;
3289
3290 item = string_list_lookup(&state->fn_table, name);
3291 if (item != NULL)
3292 return (struct patch *)item->util;
3293
3294 return NULL;
3295 }
3296
3297 /*
3298 * item->util in the filename table records the status of the path.
3299 * Usually it points at a patch (whose result records the contents
3300 * of it after applying it), but it could be PATH_WAS_DELETED for a
3301 * path that a previously applied patch has already removed, or
3302 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3303 *
3304 * The latter is needed to deal with a case where two paths A and B
3305 * are swapped by first renaming A to B and then renaming B to A;
3306 * moving A to B should not be prevented due to presence of B as we
3307 * will remove it in a later patch.
3308 */
3309 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3310 #define PATH_WAS_DELETED ((struct patch *) -1)
3311
3312 static int to_be_deleted(struct patch *patch)
3313 {
3314 return patch == PATH_TO_BE_DELETED;
3315 }
3316
3317 static int was_deleted(struct patch *patch)
3318 {
3319 return patch == PATH_WAS_DELETED;
3320 }
3321
3322 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3323 {
3324 struct string_list_item *item;
3325
3326 /*
3327 * Always add new_name unless patch is a deletion
3328 * This should cover the cases for normal diffs,
3329 * file creations and copies
3330 */
3331 if (patch->new_name != NULL) {
3332 item = string_list_insert(&state->fn_table, patch->new_name);
3333 item->util = patch;
3334 }
3335
3336 /*
3337 * store a failure on rename/deletion cases because
3338 * later chunks shouldn't patch old names
3339 */
3340 if ((patch->new_name == NULL) || (patch->is_rename)) {
3341 item = string_list_insert(&state->fn_table, patch->old_name);
3342 item->util = PATH_WAS_DELETED;
3343 }
3344 }
3345
3346 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3347 {
3348 /*
3349 * store information about incoming file deletion
3350 */
3351 while (patch) {
3352 if ((patch->new_name == NULL) || (patch->is_rename)) {
3353 struct string_list_item *item;
3354 item = string_list_insert(&state->fn_table, patch->old_name);
3355 item->util = PATH_TO_BE_DELETED;
3356 }
3357 patch = patch->next;
3358 }
3359 }
3360
3361 static int checkout_target(struct index_state *istate,
3362 struct cache_entry *ce, struct stat *st)
3363 {
3364 struct checkout costate = CHECKOUT_INIT;
3365
3366 costate.refresh_cache = 1;
3367 costate.istate = istate;
3368 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3369 return error(_("cannot checkout %s"), ce->name);
3370 return 0;
3371 }
3372
3373 static struct patch *previous_patch(struct apply_state *state,
3374 struct patch *patch,
3375 int *gone)
3376 {
3377 struct patch *previous;
3378
3379 *gone = 0;
3380 if (patch->is_copy || patch->is_rename)
3381 return NULL; /* "git" patches do not depend on the order */
3382
3383 previous = in_fn_table(state, patch->old_name);
3384 if (!previous)
3385 return NULL;
3386
3387 if (to_be_deleted(previous))
3388 return NULL; /* the deletion hasn't happened yet */
3389
3390 if (was_deleted(previous))
3391 *gone = 1;
3392
3393 return previous;
3394 }
3395
3396 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3397 {
3398 if (S_ISGITLINK(ce->ce_mode)) {
3399 if (!S_ISDIR(st->st_mode))
3400 return -1;
3401 return 0;
3402 }
3403 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3404 }
3405
3406 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3407
3408 static int load_patch_target(struct apply_state *state,
3409 struct strbuf *buf,
3410 const struct cache_entry *ce,
3411 struct stat *st,
3412 struct patch *patch,
3413 const char *name,
3414 unsigned expected_mode)
3415 {
3416 if (state->cached || state->check_index) {
3417 if (read_file_or_gitlink(ce, buf))
3418 return error(_("failed to read %s"), name);
3419 } else if (name) {
3420 if (S_ISGITLINK(expected_mode)) {
3421 if (ce)
3422 return read_file_or_gitlink(ce, buf);
3423 else
3424 return SUBMODULE_PATCH_WITHOUT_INDEX;
3425 } else if (has_symlink_leading_path(name, strlen(name))) {
3426 return error(_("reading from '%s' beyond a symbolic link"), name);
3427 } else {
3428 if (read_old_data(st, patch, name, buf))
3429 return error(_("failed to read %s"), name);
3430 }
3431 }
3432 return 0;
3433 }
3434
3435 /*
3436 * We are about to apply "patch"; populate the "image" with the
3437 * current version we have, from the working tree or from the index,
3438 * depending on the situation e.g. --cached/--index. If we are
3439 * applying a non-git patch that incrementally updates the tree,
3440 * we read from the result of a previous diff.
3441 */
3442 static int load_preimage(struct apply_state *state,
3443 struct image *image,
3444 struct patch *patch, struct stat *st,
3445 const struct cache_entry *ce)
3446 {
3447 struct strbuf buf = STRBUF_INIT;
3448 size_t len;
3449 char *img;
3450 struct patch *previous;
3451 int status;
3452
3453 previous = previous_patch(state, patch, &status);
3454 if (status)
3455 return error(_("path %s has been renamed/deleted"),
3456 patch->old_name);
3457 if (previous) {
3458 /* We have a patched copy in memory; use that. */
3459 strbuf_add(&buf, previous->result, previous->resultsize);
3460 } else {
3461 status = load_patch_target(state, &buf, ce, st, patch,
3462 patch->old_name, patch->old_mode);
3463 if (status < 0)
3464 return status;
3465 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3466 /*
3467 * There is no way to apply subproject
3468 * patch without looking at the index.
3469 * NEEDSWORK: shouldn't this be flagged
3470 * as an error???
3471 */
3472 free_fragment_list(patch->fragments);
3473 patch->fragments = NULL;
3474 } else if (status) {
3475 return error(_("failed to read %s"), patch->old_name);
3476 }
3477 }
3478
3479 img = strbuf_detach(&buf, &len);
3480 prepare_image(image, img, len, !patch->is_binary);
3481 return 0;
3482 }
3483
3484 static int three_way_merge(struct image *image,
3485 char *path,
3486 const struct object_id *base,
3487 const struct object_id *ours,
3488 const struct object_id *theirs)
3489 {
3490 mmfile_t base_file, our_file, their_file;
3491 mmbuffer_t result = { NULL };
3492 int status;
3493
3494 read_mmblob(&base_file, base);
3495 read_mmblob(&our_file, ours);
3496 read_mmblob(&their_file, theirs);
3497 status = ll_merge(&result, path,
3498 &base_file, "base",
3499 &our_file, "ours",
3500 &their_file, "theirs", NULL);
3501 free(base_file.ptr);
3502 free(our_file.ptr);
3503 free(their_file.ptr);
3504 if (status < 0 || !result.ptr) {
3505 free(result.ptr);
3506 return -1;
3507 }
3508 clear_image(image);
3509 image->buf = result.ptr;
3510 image->len = result.size;
3511
3512 return status;
3513 }
3514
3515 /*
3516 * When directly falling back to add/add three-way merge, we read from
3517 * the current contents of the new_name. In no cases other than that
3518 * this function will be called.
3519 */
3520 static int load_current(struct apply_state *state,
3521 struct image *image,
3522 struct patch *patch)
3523 {
3524 struct strbuf buf = STRBUF_INIT;
3525 int status, pos;
3526 size_t len;
3527 char *img;
3528 struct stat st;
3529 struct cache_entry *ce;
3530 char *name = patch->new_name;
3531 unsigned mode = patch->new_mode;
3532
3533 if (!patch->is_new)
3534 die("BUG: patch to %s is not a creation", patch->old_name);
3535
3536 pos = cache_name_pos(name, strlen(name));
3537 if (pos < 0)
3538 return error(_("%s: does not exist in index"), name);
3539 ce = active_cache[pos];
3540 if (lstat(name, &st)) {
3541 if (errno != ENOENT)
3542 return error_errno("%s", name);
3543 if (checkout_target(&the_index, ce, &st))
3544 return -1;
3545 }
3546 if (verify_index_match(ce, &st))
3547 return error(_("%s: does not match index"), name);
3548
3549 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
3550 if (status < 0)
3551 return status;
3552 else if (status)
3553 return -1;
3554 img = strbuf_detach(&buf, &len);
3555 prepare_image(image, img, len, !patch->is_binary);
3556 return 0;
3557 }
3558
3559 static int try_threeway(struct apply_state *state,
3560 struct image *image,
3561 struct patch *patch,
3562 struct stat *st,
3563 const struct cache_entry *ce)
3564 {
3565 struct object_id pre_oid, post_oid, our_oid;
3566 struct strbuf buf = STRBUF_INIT;
3567 size_t len;
3568 int status;
3569 char *img;
3570 struct image tmp_image;
3571
3572 /* No point falling back to 3-way merge in these cases */
3573 if (patch->is_delete ||
3574 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3575 return -1;
3576
3577 /* Preimage the patch was prepared for */
3578 if (patch->is_new)
3579 write_sha1_file("", 0, blob_type, pre_oid.hash);
3580 else if (get_sha1(patch->old_sha1_prefix, pre_oid.hash) ||
3581 read_blob_object(&buf, &pre_oid, patch->old_mode))
3582 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
3583
3584 if (state->apply_verbosity > verbosity_silent)
3585 fprintf(stderr, _("Falling back to three-way merge...\n"));
3586
3587 img = strbuf_detach(&buf, &len);
3588 prepare_image(&tmp_image, img, len, 1);
3589 /* Apply the patch to get the post image */
3590 if (apply_fragments(state, &tmp_image, patch) < 0) {
3591 clear_image(&tmp_image);
3592 return -1;
3593 }
3594 /* post_oid is theirs */
3595 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
3596 clear_image(&tmp_image);
3597
3598 /* our_oid is ours */
3599 if (patch->is_new) {
3600 if (load_current(state, &tmp_image, patch))
3601 return error(_("cannot read the current contents of '%s'"),
3602 patch->new_name);
3603 } else {
3604 if (load_preimage(state, &tmp_image, patch, st, ce))
3605 return error(_("cannot read the current contents of '%s'"),
3606 patch->old_name);
3607 }
3608 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
3609 clear_image(&tmp_image);
3610
3611 /* in-core three-way merge between post and our using pre as base */
3612 status = three_way_merge(image, patch->new_name,
3613 &pre_oid, &our_oid, &post_oid);
3614 if (status < 0) {
3615 if (state->apply_verbosity > verbosity_silent)
3616 fprintf(stderr,
3617 _("Failed to fall back on three-way merge...\n"));
3618 return status;
3619 }
3620
3621 if (status) {
3622 patch->conflicted_threeway = 1;
3623 if (patch->is_new)
3624 oidclr(&patch->threeway_stage[0]);
3625 else
3626 oidcpy(&patch->threeway_stage[0], &pre_oid);
3627 oidcpy(&patch->threeway_stage[1], &our_oid);
3628 oidcpy(&patch->threeway_stage[2], &post_oid);
3629 if (state->apply_verbosity > verbosity_silent)
3630 fprintf(stderr,
3631 _("Applied patch to '%s' with conflicts.\n"),
3632 patch->new_name);
3633 } else {
3634 if (state->apply_verbosity > verbosity_silent)
3635 fprintf(stderr,
3636 _("Applied patch to '%s' cleanly.\n"),
3637 patch->new_name);
3638 }
3639 return 0;
3640 }
3641
3642 static int apply_data(struct apply_state *state, struct patch *patch,
3643 struct stat *st, const struct cache_entry *ce)
3644 {
3645 struct image image;
3646
3647 if (load_preimage(state, &image, patch, st, ce) < 0)
3648 return -1;
3649
3650 if (patch->direct_to_threeway ||
3651 apply_fragments(state, &image, patch) < 0) {
3652 /* Note: with --reject, apply_fragments() returns 0 */
3653 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3654 return -1;
3655 }
3656 patch->result = image.buf;
3657 patch->resultsize = image.len;
3658 add_to_fn_table(state, patch);
3659 free(image.line_allocated);
3660
3661 if (0 < patch->is_delete && patch->resultsize)
3662 return error(_("removal patch leaves file contents"));
3663
3664 return 0;
3665 }
3666
3667 /*
3668 * If "patch" that we are looking at modifies or deletes what we have,
3669 * we would want it not to lose any local modification we have, either
3670 * in the working tree or in the index.
3671 *
3672 * This also decides if a non-git patch is a creation patch or a
3673 * modification to an existing empty file. We do not check the state
3674 * of the current tree for a creation patch in this function; the caller
3675 * check_patch() separately makes sure (and errors out otherwise) that
3676 * the path the patch creates does not exist in the current tree.
3677 */
3678 static int check_preimage(struct apply_state *state,
3679 struct patch *patch,
3680 struct cache_entry **ce,
3681 struct stat *st)
3682 {
3683 const char *old_name = patch->old_name;
3684 struct patch *previous = NULL;
3685 int stat_ret = 0, status;
3686 unsigned st_mode = 0;
3687
3688 if (!old_name)
3689 return 0;
3690
3691 assert(patch->is_new <= 0);
3692 previous = previous_patch(state, patch, &status);
3693
3694 if (status)
3695 return error(_("path %s has been renamed/deleted"), old_name);
3696 if (previous) {
3697 st_mode = previous->new_mode;
3698 } else if (!state->cached) {
3699 stat_ret = lstat(old_name, st);
3700 if (stat_ret && errno != ENOENT)
3701 return error_errno("%s", old_name);
3702 }
3703
3704 if (state->check_index && !previous) {
3705 int pos = cache_name_pos(old_name, strlen(old_name));
3706 if (pos < 0) {
3707 if (patch->is_new < 0)
3708 goto is_new;
3709 return error(_("%s: does not exist in index"), old_name);
3710 }
3711 *ce = active_cache[pos];
3712 if (stat_ret < 0) {
3713 if (checkout_target(&the_index, *ce, st))
3714 return -1;
3715 }
3716 if (!state->cached && verify_index_match(*ce, st))
3717 return error(_("%s: does not match index"), old_name);
3718 if (state->cached)
3719 st_mode = (*ce)->ce_mode;
3720 } else if (stat_ret < 0) {
3721 if (patch->is_new < 0)
3722 goto is_new;
3723 return error_errno("%s", old_name);
3724 }
3725
3726 if (!state->cached && !previous)
3727 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3728
3729 if (patch->is_new < 0)
3730 patch->is_new = 0;
3731 if (!patch->old_mode)
3732 patch->old_mode = st_mode;
3733 if ((st_mode ^ patch->old_mode) & S_IFMT)
3734 return error(_("%s: wrong type"), old_name);
3735 if (st_mode != patch->old_mode)
3736 warning(_("%s has type %o, expected %o"),
3737 old_name, st_mode, patch->old_mode);
3738 if (!patch->new_mode && !patch->is_delete)
3739 patch->new_mode = st_mode;
3740 return 0;
3741
3742 is_new:
3743 patch->is_new = 1;
3744 patch->is_delete = 0;
3745 FREE_AND_NULL(patch->old_name);
3746 return 0;
3747 }
3748
3749
3750 #define EXISTS_IN_INDEX 1
3751 #define EXISTS_IN_WORKTREE 2
3752
3753 static int check_to_create(struct apply_state *state,
3754 const char *new_name,
3755 int ok_if_exists)
3756 {
3757 struct stat nst;
3758
3759 if (state->check_index &&
3760 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3761 !ok_if_exists)
3762 return EXISTS_IN_INDEX;
3763 if (state->cached)
3764 return 0;
3765
3766 if (!lstat(new_name, &nst)) {
3767 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3768 return 0;
3769 /*
3770 * A leading component of new_name might be a symlink
3771 * that is going to be removed with this patch, but
3772 * still pointing at somewhere that has the path.
3773 * In such a case, path "new_name" does not exist as
3774 * far as git is concerned.
3775 */
3776 if (has_symlink_leading_path(new_name, strlen(new_name)))
3777 return 0;
3778
3779 return EXISTS_IN_WORKTREE;
3780 } else if (!is_missing_file_error(errno)) {
3781 return error_errno("%s", new_name);
3782 }
3783 return 0;
3784 }
3785
3786 static uintptr_t register_symlink_changes(struct apply_state *state,
3787 const char *path,
3788 uintptr_t what)
3789 {
3790 struct string_list_item *ent;
3791
3792 ent = string_list_lookup(&state->symlink_changes, path);
3793 if (!ent) {
3794 ent = string_list_insert(&state->symlink_changes, path);
3795 ent->util = (void *)0;
3796 }
3797 ent->util = (void *)(what | ((uintptr_t)ent->util));
3798 return (uintptr_t)ent->util;
3799 }
3800
3801 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3802 {
3803 struct string_list_item *ent;
3804
3805 ent = string_list_lookup(&state->symlink_changes, path);
3806 if (!ent)
3807 return 0;
3808 return (uintptr_t)ent->util;
3809 }
3810
3811 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3812 {
3813 for ( ; patch; patch = patch->next) {
3814 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3815 (patch->is_rename || patch->is_delete))
3816 /* the symlink at patch->old_name is removed */
3817 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3818
3819 if (patch->new_name && S_ISLNK(patch->new_mode))
3820 /* the symlink at patch->new_name is created or remains */
3821 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3822 }
3823 }
3824
3825 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3826 {
3827 do {
3828 unsigned int change;
3829
3830 while (--name->len && name->buf[name->len] != '/')
3831 ; /* scan backwards */
3832 if (!name->len)
3833 break;
3834 name->buf[name->len] = '\0';
3835 change = check_symlink_changes(state, name->buf);
3836 if (change & APPLY_SYMLINK_IN_RESULT)
3837 return 1;
3838 if (change & APPLY_SYMLINK_GOES_AWAY)
3839 /*
3840 * This cannot be "return 0", because we may
3841 * see a new one created at a higher level.
3842 */
3843 continue;
3844