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