]> git.ipfire.org Git - thirdparty/git.git/blame - apply.c
apply: move lockfile into `apply_state`
[thirdparty/git.git] / apply.c
CommitLineData
13b5af22
CC
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
bb493a5c 10#include "cache.h"
b2141fc1 11#include "config.h"
13b5af22
CC
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"
bb493a5c 18#include "lockfile.h"
13b5af22
CC
19#include "parse-options.h"
20#include "quote.h"
21#include "rerere.h"
bb493a5c
CC
22#include "apply.h"
23
24static 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
9123d5dd 31static int parse_whitespace_option(struct apply_state *state, const char *option)
bb493a5c
CC
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
9123d5dd
CC
61static int parse_ignorewhitespace_option(struct apply_state *state,
62 const char *option)
bb493a5c
CC
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
2f5a6d12 77int init_apply_state(struct apply_state *state,
6d058c88 78 const char *prefix)
bb493a5c
CC
79{
80 memset(state, 0, sizeof(*state));
81 state->prefix = prefix;
bb493a5c
CC
82 state->newfd = -1;
83 state->apply = 1;
84 state->line_termination = '\n';
85 state->p_value = 1;
86 state->p_context = UINT_MAX;
87 state->squelch_whitespace_errors = 5;
88 state->ws_error_action = warn_on_ws_error;
89 state->ws_ignore_action = ignore_ws_none;
90 state->linenr = 1;
91 string_list_init(&state->fn_table, 0);
92 string_list_init(&state->limit_by_name, 0);
93 string_list_init(&state->symlink_changes, 0);
94 strbuf_init(&state->root, 0);
95
96 git_apply_config();
97 if (apply_default_whitespace && parse_whitespace_option(state, apply_default_whitespace))
2f5a6d12 98 return -1;
bb493a5c 99 if (apply_default_ignorewhitespace && parse_ignorewhitespace_option(state, apply_default_ignorewhitespace))
2f5a6d12
CC
100 return -1;
101 return 0;
bb493a5c
CC
102}
103
104void clear_apply_state(struct apply_state *state)
105{
106 string_list_clear(&state->limit_by_name, 0);
107 string_list_clear(&state->symlink_changes, 0);
108 strbuf_release(&state->root);
109
110 /* &state->fn_table is cleared at the end of apply_patch() */
111}
b6446d54 112
45b78d8b
CC
113static void mute_routine(const char *msg, va_list params)
114{
115 /* do nothing */
116}
117
b6446d54
CC
118int check_apply_state(struct apply_state *state, int force_apply)
119{
120 int is_not_gitdir = !startup_info->have_repository;
121
122 if (state->apply_with_reject && state->threeway)
d1d42bf5 123 return error(_("--reject and --3way cannot be used together."));
b6446d54 124 if (state->cached && state->threeway)
d1d42bf5 125 return error(_("--cached and --3way cannot be used together."));
b6446d54
CC
126 if (state->threeway) {
127 if (is_not_gitdir)
128 return error(_("--3way outside a repository"));
129 state->check_index = 1;
130 }
a46160d2
CC
131 if (state->apply_with_reject) {
132 state->apply = 1;
133 if (state->apply_verbosity == verbosity_normal)
134 state->apply_verbosity = verbosity_verbose;
135 }
b6446d54
CC
136 if (!force_apply && (state->diffstat || state->numstat || state->summary || state->check || state->fake_ancestor))
137 state->apply = 0;
138 if (state->check_index && is_not_gitdir)
139 return error(_("--index outside a repository"));
140 if (state->cached) {
141 if (is_not_gitdir)
142 return error(_("--cached outside a repository"));
143 state->check_index = 1;
144 }
145 if (state->check_index)
146 state->unsafe_paths = 0;
b6446d54 147
45b78d8b
CC
148 if (state->apply_verbosity <= verbosity_silent) {
149 state->saved_error_routine = get_error_routine();
150 state->saved_warn_routine = get_warn_routine();
151 set_error_routine(mute_routine);
152 set_warn_routine(mute_routine);
153 }
154
b6446d54
CC
155 return 0;
156}
13b5af22
CC
157
158static void set_default_whitespace_mode(struct apply_state *state)
159{
160 if (!state->whitespace_option && !apply_default_whitespace)
161 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
162}
163
164/*
165 * This represents one "hunk" from a patch, starting with
166 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
167 * patch text is pointed at by patch, and its byte length
168 * is stored in size. leading and trailing are the number
169 * of context lines.
170 */
171struct fragment {
172 unsigned long leading, trailing;
173 unsigned long oldpos, oldlines;
174 unsigned long newpos, newlines;
175 /*
176 * 'patch' is usually borrowed from buf in apply_patch(),
177 * but some codepaths store an allocated buffer.
178 */
179 const char *patch;
180 unsigned free_patch:1,
181 rejected:1;
182 int size;
183 int linenr;
184 struct fragment *next;
185};
186
187/*
188 * When dealing with a binary patch, we reuse "leading" field
189 * to store the type of the binary hunk, either deflated "delta"
190 * or deflated "literal".
191 */
192#define binary_patch_method leading
193#define BINARY_DELTA_DEFLATED 1
194#define BINARY_LITERAL_DEFLATED 2
195
196/*
197 * This represents a "patch" to a file, both metainfo changes
198 * such as creation/deletion, filemode and content changes represented
199 * as a series of fragments.
200 */
201struct patch {
202 char *new_name, *old_name, *def_name;
203 unsigned int old_mode, new_mode;
204 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
205 int rejected;
206 unsigned ws_rule;
207 int lines_added, lines_deleted;
208 int score;
d70e9c5c 209 int extension_linenr; /* first line specifying delete/new/rename/copy */
13b5af22
CC
210 unsigned int is_toplevel_relative:1;
211 unsigned int inaccurate_eof:1;
212 unsigned int is_binary:1;
213 unsigned int is_copy:1;
214 unsigned int is_rename:1;
215 unsigned int recount:1;
216 unsigned int conflicted_threeway:1;
217 unsigned int direct_to_threeway:1;
c24f3aba 218 unsigned int crlf_in_old:1;
13b5af22
CC
219 struct fragment *fragments;
220 char *result;
221 size_t resultsize;
222 char old_sha1_prefix[41];
223 char new_sha1_prefix[41];
224 struct patch *next;
225
226 /* three-way fallback result */
227 struct object_id threeway_stage[3];
228};
229
230static void free_fragment_list(struct fragment *list)
231{
232 while (list) {
233 struct fragment *next = list->next;
234 if (list->free_patch)
235 free((char *)list->patch);
236 free(list);
237 list = next;
238 }
239}
240
241static void free_patch(struct patch *patch)
242{
243 free_fragment_list(patch->fragments);
244 free(patch->def_name);
245 free(patch->old_name);
246 free(patch->new_name);
247 free(patch->result);
248 free(patch);
249}
250
251static void free_patch_list(struct patch *list)
252{
253 while (list) {
254 struct patch *next = list->next;
255 free_patch(list);
256 list = next;
257 }
258}
259
260/*
261 * A line in a file, len-bytes long (includes the terminating LF,
262 * except for an incomplete line at the end if the file ends with
263 * one), and its contents hashes to 'hash'.
264 */
265struct line {
266 size_t len;
267 unsigned hash : 24;
268 unsigned flag : 8;
269#define LINE_COMMON 1
270#define LINE_PATCHED 2
271};
272
273/*
274 * This represents a "file", which is an array of "lines".
275 */
276struct image {
277 char *buf;
278 size_t len;
279 size_t nr;
280 size_t alloc;
281 struct line *line_allocated;
282 struct line *line;
283};
284
285static uint32_t hash_line(const char *cp, size_t len)
286{
287 size_t i;
288 uint32_t h;
289 for (i = 0, h = 0; i < len; i++) {
290 if (!isspace(cp[i])) {
291 h = h * 3 + (cp[i] & 0xff);
292 }
293 }
294 return h;
295}
296
297/*
298 * Compare lines s1 of length n1 and s2 of length n2, ignoring
299 * whitespace difference. Returns 1 if they match, 0 otherwise
300 */
301static int fuzzy_matchlines(const char *s1, size_t n1,
302 const char *s2, size_t n2)
303{
304 const char *last1 = s1 + n1 - 1;
305 const char *last2 = s2 + n2 - 1;
306 int result = 0;
307
308 /* ignore line endings */
309 while ((*last1 == '\r') || (*last1 == '\n'))
310 last1--;
311 while ((*last2 == '\r') || (*last2 == '\n'))
312 last2--;
313
314 /* skip leading whitespaces, if both begin with whitespace */
315 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
316 while (isspace(*s1) && (s1 <= last1))
317 s1++;
318 while (isspace(*s2) && (s2 <= last2))
319 s2++;
320 }
321 /* early return if both lines are empty */
322 if ((s1 > last1) && (s2 > last2))
323 return 1;
324 while (!result) {
325 result = *s1++ - *s2++;
326 /*
327 * Skip whitespace inside. We check for whitespace on
328 * both buffers because we don't want "a b" to match
329 * "ab"
330 */
331 if (isspace(*s1) && isspace(*s2)) {
332 while (isspace(*s1) && s1 <= last1)
333 s1++;
334 while (isspace(*s2) && s2 <= last2)
335 s2++;
336 }
337 /*
338 * If we reached the end on one side only,
339 * lines don't match
340 */
341 if (
342 ((s2 > last2) && (s1 <= last1)) ||
343 ((s1 > last1) && (s2 <= last2)))
344 return 0;
345 if ((s1 > last1) && (s2 > last2))
346 break;
347 }
348
349 return !result;
350}
351
352static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
353{
354 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
355 img->line_allocated[img->nr].len = len;
356 img->line_allocated[img->nr].hash = hash_line(bol, len);
357 img->line_allocated[img->nr].flag = flag;
358 img->nr++;
359}
360
361/*
362 * "buf" has the file contents to be patched (read from various sources).
363 * attach it to "image" and add line-based index to it.
364 * "image" now owns the "buf".
365 */
366static void prepare_image(struct image *image, char *buf, size_t len,
367 int prepare_linetable)
368{
369 const char *cp, *ep;
370
371 memset(image, 0, sizeof(*image));
372 image->buf = buf;
373 image->len = len;
374
375 if (!prepare_linetable)
376 return;
377
378 ep = image->buf + image->len;
379 cp = image->buf;
380 while (cp < ep) {
381 const char *next;
382 for (next = cp; next < ep && *next != '\n'; next++)
383 ;
384 if (next < ep)
385 next++;
386 add_line_info(image, cp, next - cp, 0);
387 cp = next;
388 }
389 image->line = image->line_allocated;
390}
391
392static void clear_image(struct image *image)
393{
394 free(image->buf);
395 free(image->line_allocated);
396 memset(image, 0, sizeof(*image));
397}
398
399/* fmt must contain _one_ %s and no other substitution */
400static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
401{
402 struct strbuf sb = STRBUF_INIT;
403
404 if (patch->old_name && patch->new_name &&
405 strcmp(patch->old_name, patch->new_name)) {
406 quote_c_style(patch->old_name, &sb, NULL, 0);
407 strbuf_addstr(&sb, " => ");
408 quote_c_style(patch->new_name, &sb, NULL, 0);
409 } else {
410 const char *n = patch->new_name;
411 if (!n)
412 n = patch->old_name;
413 quote_c_style(n, &sb, NULL, 0);
414 }
415 fprintf(output, fmt, sb.buf);
416 fputc('\n', output);
417 strbuf_release(&sb);
418}
419
420#define SLOP (16)
421
422static int read_patch_file(struct strbuf *sb, int fd)
423{
424 if (strbuf_read(sb, fd, 0) < 0)
425 return error_errno("git apply: failed to read");
426
427 /*
428 * Make sure that we have some slop in the buffer
429 * so that we can do speculative "memcmp" etc, and
430 * see to it that it is NUL-filled.
431 */
432 strbuf_grow(sb, SLOP);
433 memset(sb->buf + sb->len, 0, SLOP);
434 return 0;
435}
436
437static unsigned long linelen(const char *buffer, unsigned long size)
438{
439 unsigned long len = 0;
440 while (size--) {
441 len++;
442 if (*buffer++ == '\n')
443 break;
444 }
445 return len;
446}
447
448static int is_dev_null(const char *str)
449{
450 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
451}
452
453#define TERM_SPACE 1
454#define TERM_TAB 2
455
456static int name_terminate(int c, int terminate)
457{
458 if (c == ' ' && !(terminate & TERM_SPACE))
459 return 0;
460 if (c == '\t' && !(terminate & TERM_TAB))
461 return 0;
462
463 return 1;
464}
465
466/* remove double slashes to make --index work with such filenames */
467static char *squash_slash(char *name)
468{
469 int i = 0, j = 0;
470
471 if (!name)
472 return NULL;
473
474 while (name[i]) {
475 if ((name[j++] = name[i++]) == '/')
476 while (name[i] == '/')
477 i++;
478 }
479 name[j] = '\0';
480 return name;
481}
482
483static char *find_name_gnu(struct apply_state *state,
484 const char *line,
485 const char *def,
486 int p_value)
487{
488 struct strbuf name = STRBUF_INIT;
489 char *cp;
490
491 /*
492 * Proposed "new-style" GNU patch/diff format; see
493 * http://marc.info/?l=git&m=112927316408690&w=2
494 */
495 if (unquote_c_style(&name, line, NULL)) {
496 strbuf_release(&name);
497 return NULL;
498 }
499
500 for (cp = name.buf; p_value; p_value--) {
501 cp = strchr(cp, '/');
502 if (!cp) {
503 strbuf_release(&name);
504 return NULL;
505 }
506 cp++;
507 }
508
509 strbuf_remove(&name, 0, cp - name.buf);
510 if (state->root.len)
511 strbuf_insert(&name, 0, state->root.buf, state->root.len);
512 return squash_slash(strbuf_detach(&name, NULL));
513}
514
515static size_t sane_tz_len(const char *line, size_t len)
516{
517 const char *tz, *p;
518
519 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
520 return 0;
521 tz = line + len - strlen(" +0500");
522
523 if (tz[1] != '+' && tz[1] != '-')
524 return 0;
525
526 for (p = tz + 2; p != line + len; p++)
527 if (!isdigit(*p))
528 return 0;
529
530 return line + len - tz;
531}
532
533static size_t tz_with_colon_len(const char *line, size_t len)
534{
535 const char *tz, *p;
536
537 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
538 return 0;
539 tz = line + len - strlen(" +08:00");
540
541 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
542 return 0;
543 p = tz + 2;
544 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
545 !isdigit(*p++) || !isdigit(*p++))
546 return 0;
547
548 return line + len - tz;
549}
550
551static size_t date_len(const char *line, size_t len)
552{
553 const char *date, *p;
554
555 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
556 return 0;
557 p = date = line + len - strlen("72-02-05");
558
559 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
560 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
561 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
562 return 0;
563
564 if (date - line >= strlen("19") &&
565 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
566 date -= strlen("19");
567
568 return line + len - date;
569}
570
571static size_t short_time_len(const char *line, size_t len)
572{
573 const char *time, *p;
574
575 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
576 return 0;
577 p = time = line + len - strlen(" 07:01:32");
578
579 /* Permit 1-digit hours? */
580 if (*p++ != ' ' ||
581 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
582 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
583 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
584 return 0;
585
586 return line + len - time;
587}
588
589static size_t fractional_time_len(const char *line, size_t len)
590{
591 const char *p;
592 size_t n;
593
594 /* Expected format: 19:41:17.620000023 */
595 if (!len || !isdigit(line[len - 1]))
596 return 0;
597 p = line + len - 1;
598
599 /* Fractional seconds. */
600 while (p > line && isdigit(*p))
601 p--;
602 if (*p != '.')
603 return 0;
604
605 /* Hours, minutes, and whole seconds. */
606 n = short_time_len(line, p - line);
607 if (!n)
608 return 0;
609
610 return line + len - p + n;
611}
612
613static size_t trailing_spaces_len(const char *line, size_t len)
614{
615 const char *p;
616
617 /* Expected format: ' ' x (1 or more) */
618 if (!len || line[len - 1] != ' ')
619 return 0;
620
621 p = line + len;
622 while (p != line) {
623 p--;
624 if (*p != ' ')
625 return line + len - (p + 1);
626 }
627
628 /* All spaces! */
629 return len;
630}
631
632static size_t diff_timestamp_len(const char *line, size_t len)
633{
634 const char *end = line + len;
635 size_t n;
636
637 /*
638 * Posix: 2010-07-05 19:41:17
639 * GNU: 2010-07-05 19:41:17.620000023 -0500
640 */
641
642 if (!isdigit(end[-1]))
643 return 0;
644
645 n = sane_tz_len(line, end - line);
646 if (!n)
647 n = tz_with_colon_len(line, end - line);
648 end -= n;
649
650 n = short_time_len(line, end - line);
651 if (!n)
652 n = fractional_time_len(line, end - line);
653 end -= n;
654
655 n = date_len(line, end - line);
656 if (!n) /* No date. Too bad. */
657 return 0;
658 end -= n;
659
660 if (end == line) /* No space before date. */
661 return 0;
662 if (end[-1] == '\t') { /* Success! */
663 end--;
664 return line + len - end;
665 }
666 if (end[-1] != ' ') /* No space before date. */
667 return 0;
668
669 /* Whitespace damage. */
670 end -= trailing_spaces_len(line, end - line);
671 return line + len - end;
672}
673
674static char *find_name_common(struct apply_state *state,
675 const char *line,
676 const char *def,
677 int p_value,
678 const char *end,
679 int terminate)
680{
681 int len;
682 const char *start = NULL;
683
684 if (p_value == 0)
685 start = line;
686 while (line != end) {
687 char c = *line;
688
689 if (!end && isspace(c)) {
690 if (c == '\n')
691 break;
692 if (name_terminate(c, terminate))
693 break;
694 }
695 line++;
696 if (c == '/' && !--p_value)
697 start = line;
698 }
699 if (!start)
700 return squash_slash(xstrdup_or_null(def));
701 len = line - start;
702 if (!len)
703 return squash_slash(xstrdup_or_null(def));
704
705 /*
706 * Generally we prefer the shorter name, especially
707 * if the other one is just a variation of that with
708 * something else tacked on to the end (ie "file.orig"
709 * or "file~").
710 */
711 if (def) {
712 int deflen = strlen(def);
713 if (deflen < len && !strncmp(start, def, deflen))
714 return squash_slash(xstrdup(def));
715 }
716
717 if (state->root.len) {
718 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
719 return squash_slash(ret);
720 }
721
722 return squash_slash(xmemdupz(start, len));
723}
724
725static char *find_name(struct apply_state *state,
726 const char *line,
727 char *def,
728 int p_value,
729 int terminate)
730{
731 if (*line == '"') {
732 char *name = find_name_gnu(state, line, def, p_value);
733 if (name)
734 return name;
735 }
736
737 return find_name_common(state, line, def, p_value, NULL, terminate);
738}
739
740static char *find_name_traditional(struct apply_state *state,
741 const char *line,
742 char *def,
743 int p_value)
744{
745 size_t len;
746 size_t date_len;
747
748 if (*line == '"') {
749 char *name = find_name_gnu(state, line, def, p_value);
750 if (name)
751 return name;
752 }
753
754 len = strchrnul(line, '\n') - line;
755 date_len = diff_timestamp_len(line, len);
756 if (!date_len)
757 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
758 len -= date_len;
759
760 return find_name_common(state, line, def, p_value, line + len, 0);
761}
762
13b5af22
CC
763/*
764 * Given the string after "--- " or "+++ ", guess the appropriate
765 * p_value for the given patch.
766 */
767static int guess_p_value(struct apply_state *state, const char *nameline)
768{
769 char *name, *cp;
770 int val = -1;
771
772 if (is_dev_null(nameline))
773 return -1;
774 name = find_name_traditional(state, nameline, NULL, 0);
775 if (!name)
776 return -1;
777 cp = strchr(name, '/');
778 if (!cp)
779 val = 0;
780 else if (state->prefix) {
781 /*
782 * Does it begin with "a/$our-prefix" and such? Then this is
783 * very likely to apply to our directory.
784 */
881529c8 785 if (starts_with(name, state->prefix))
13b5af22
CC
786 val = count_slashes(state->prefix);
787 else {
788 cp++;
881529c8 789 if (starts_with(cp, state->prefix))
13b5af22
CC
790 val = count_slashes(state->prefix) + 1;
791 }
792 }
793 free(name);
794 return val;
795}
796
797/*
798 * Does the ---/+++ line have the POSIX timestamp after the last HT?
799 * GNU diff puts epoch there to signal a creation/deletion event. Is
800 * this such a timestamp?
801 */
802static int has_epoch_timestamp(const char *nameline)
803{
804 /*
805 * We are only interested in epoch timestamp; any non-zero
806 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
807 * For the same reason, the date must be either 1969-12-31 or
808 * 1970-01-01, and the seconds part must be "00".
809 */
810 const char stamp_regexp[] =
0db3dc75 811 "^[0-2][0-9]:([0-5][0-9]):00(\\.0+)?"
13b5af22
CC
812 " "
813 "([-+][0-2][0-9]:?[0-5][0-9])\n";
814 const char *timestamp = NULL, *cp, *colon;
815 static regex_t *stamp;
816 regmatch_t m[10];
e4905019 817 int zoneoffset, epoch_hour, hour, minute;
13b5af22
CC
818 int status;
819
820 for (cp = nameline; *cp != '\n'; cp++) {
821 if (*cp == '\t')
822 timestamp = cp + 1;
823 }
824 if (!timestamp)
825 return 0;
e4905019
RS
826
827 /*
828 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
829 * (west of GMT) or 1970-01-01 (east of GMT)
830 */
0db3dc75 831 if (skip_prefix(timestamp, "1969-12-31 ", &timestamp))
e4905019 832 epoch_hour = 24;
0db3dc75 833 else if (skip_prefix(timestamp, "1970-01-01 ", &timestamp))
e4905019
RS
834 epoch_hour = 0;
835 else
836 return 0;
837
13b5af22
CC
838 if (!stamp) {
839 stamp = xmalloc(sizeof(*stamp));
840 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
841 warning(_("Cannot prepare timestamp regexp %s"),
842 stamp_regexp);
843 return 0;
844 }
845 }
846
847 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
848 if (status) {
849 if (status != REG_NOMATCH)
850 warning(_("regexec returned %d for input: %s"),
851 status, timestamp);
852 return 0;
853 }
854
0db3dc75
RS
855 hour = strtol(timestamp, NULL, 10);
856 minute = strtol(timestamp + m[1].rm_so, NULL, 10);
e4905019 857
13b5af22
CC
858 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
859 if (*colon == ':')
860 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
861 else
862 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
863 if (timestamp[m[3].rm_so] == '-')
864 zoneoffset = -zoneoffset;
865
e4905019 866 return hour * 60 + minute - zoneoffset == epoch_hour * 60;
13b5af22
CC
867}
868
869/*
870 * Get the name etc info from the ---/+++ lines of a traditional patch header
871 *
872 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
873 * files, we can happily check the index for a match, but for creating a
874 * new file we should try to match whatever "patch" does. I have no idea.
875 */
876static int parse_traditional_patch(struct apply_state *state,
877 const char *first,
878 const char *second,
879 struct patch *patch)
880{
881 char *name;
882
883 first += 4; /* skip "--- " */
884 second += 4; /* skip "+++ " */
885 if (!state->p_value_known) {
886 int p, q;
887 p = guess_p_value(state, first);
888 q = guess_p_value(state, second);
889 if (p < 0) p = q;
890 if (0 <= p && p == q) {
891 state->p_value = p;
892 state->p_value_known = 1;
893 }
894 }
895 if (is_dev_null(first)) {
896 patch->is_new = 1;
897 patch->is_delete = 0;
898 name = find_name_traditional(state, second, NULL, state->p_value);
899 patch->new_name = name;
900 } else if (is_dev_null(second)) {
901 patch->is_new = 0;
902 patch->is_delete = 1;
903 name = find_name_traditional(state, first, NULL, state->p_value);
904 patch->old_name = name;
905 } else {
906 char *first_name;
907 first_name = find_name_traditional(state, first, NULL, state->p_value);
908 name = find_name_traditional(state, second, first_name, state->p_value);
909 free(first_name);
910 if (has_epoch_timestamp(first)) {
911 patch->is_new = 1;
912 patch->is_delete = 0;
913 patch->new_name = name;
914 } else if (has_epoch_timestamp(second)) {
915 patch->is_new = 0;
916 patch->is_delete = 1;
917 patch->old_name = name;
918 } else {
919 patch->old_name = name;
920 patch->new_name = xstrdup_or_null(name);
921 }
922 }
923 if (!name)
924 return error(_("unable to find filename in patch at line %d"), state->linenr);
925
926 return 0;
927}
928
929static int gitdiff_hdrend(struct apply_state *state,
930 const char *line,
931 struct patch *patch)
932{
933 return 1;
934}
935
936/*
937 * We're anal about diff header consistency, to make
938 * sure that we don't end up having strange ambiguous
939 * patches floating around.
940 *
941 * As a result, gitdiff_{old|new}name() will check
942 * their names against any previous information, just
943 * to make sure..
944 */
945#define DIFF_OLD_NAME 0
946#define DIFF_NEW_NAME 1
947
948static int gitdiff_verify_name(struct apply_state *state,
949 const char *line,
950 int isnull,
951 char **name,
952 int side)
953{
954 if (!*name && !isnull) {
955 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
956 return 0;
957 }
958
959 if (*name) {
13b5af22
CC
960 char *another;
961 if (isnull)
962 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
963 *name, state->linenr);
964 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
2d105451 965 if (!another || strcmp(another, *name)) {
13b5af22
CC
966 free(another);
967 return error((side == DIFF_NEW_NAME) ?
968 _("git apply: bad git-diff - inconsistent new filename on line %d") :
969 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
970 }
971 free(another);
972 } else {
8bc172e5 973 if (!starts_with(line, "/dev/null\n"))
13b5af22
CC
974 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
975 }
976
977 return 0;
978}
979
980static int gitdiff_oldname(struct apply_state *state,
981 const char *line,
982 struct patch *patch)
983{
984 return gitdiff_verify_name(state, line,
985 patch->is_new, &patch->old_name,
986 DIFF_OLD_NAME);
987}
988
989static int gitdiff_newname(struct apply_state *state,
990 const char *line,
991 struct patch *patch)
992{
993 return gitdiff_verify_name(state, line,
994 patch->is_delete, &patch->new_name,
995 DIFF_NEW_NAME);
996}
997
44e5471a
RS
998static int parse_mode_line(const char *line, int linenr, unsigned int *mode)
999{
1000 char *end;
1001 *mode = strtoul(line, &end, 8);
1002 if (end == line || !isspace(*end))
1003 return error(_("invalid mode on line %d: %s"), linenr, line);
1004 return 0;
1005}
1006
13b5af22
CC
1007static int gitdiff_oldmode(struct apply_state *state,
1008 const char *line,
1009 struct patch *patch)
1010{
44e5471a 1011 return parse_mode_line(line, state->linenr, &patch->old_mode);
13b5af22
CC
1012}
1013
1014static int gitdiff_newmode(struct apply_state *state,
1015 const char *line,
1016 struct patch *patch)
1017{
44e5471a 1018 return parse_mode_line(line, state->linenr, &patch->new_mode);
13b5af22
CC
1019}
1020
1021static int gitdiff_delete(struct apply_state *state,
1022 const char *line,
1023 struct patch *patch)
1024{
1025 patch->is_delete = 1;
1026 free(patch->old_name);
1027 patch->old_name = xstrdup_or_null(patch->def_name);
1028 return gitdiff_oldmode(state, line, patch);
1029}
1030
1031static int gitdiff_newfile(struct apply_state *state,
1032 const char *line,
1033 struct patch *patch)
1034{
1035 patch->is_new = 1;
1036 free(patch->new_name);
1037 patch->new_name = xstrdup_or_null(patch->def_name);
1038 return gitdiff_newmode(state, line, patch);
1039}
1040
1041static int gitdiff_copysrc(struct apply_state *state,
1042 const char *line,
1043 struct patch *patch)
1044{
1045 patch->is_copy = 1;
1046 free(patch->old_name);
1047 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1048 return 0;
1049}
1050
1051static int gitdiff_copydst(struct apply_state *state,
1052 const char *line,
1053 struct patch *patch)
1054{
1055 patch->is_copy = 1;
1056 free(patch->new_name);
1057 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1058 return 0;
1059}
1060
1061static int gitdiff_renamesrc(struct apply_state *state,
1062 const char *line,
1063 struct patch *patch)
1064{
1065 patch->is_rename = 1;
1066 free(patch->old_name);
1067 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1068 return 0;
1069}
1070
1071static int gitdiff_renamedst(struct apply_state *state,
1072 const char *line,
1073 struct patch *patch)
1074{
1075 patch->is_rename = 1;
1076 free(patch->new_name);
1077 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
1078 return 0;
1079}
1080
1081static int gitdiff_similarity(struct apply_state *state,
1082 const char *line,
1083 struct patch *patch)
1084{
1085 unsigned long val = strtoul(line, NULL, 10);
1086 if (val <= 100)
1087 patch->score = val;
1088 return 0;
1089}
1090
1091static int gitdiff_dissimilarity(struct apply_state *state,
1092 const char *line,
1093 struct patch *patch)
1094{
1095 unsigned long val = strtoul(line, NULL, 10);
1096 if (val <= 100)
1097 patch->score = val;
1098 return 0;
1099}
1100
1101static int gitdiff_index(struct apply_state *state,
1102 const char *line,
1103 struct patch *patch)
1104{
1105 /*
1106 * index line is N hexadecimal, "..", N hexadecimal,
1107 * and optional space with octal mode.
1108 */
1109 const char *ptr, *eol;
1110 int len;
1111
1112 ptr = strchr(line, '.');
1113 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
1114 return 0;
1115 len = ptr - line;
1116 memcpy(patch->old_sha1_prefix, line, len);
1117 patch->old_sha1_prefix[len] = 0;
1118
1119 line = ptr + 2;
1120 ptr = strchr(line, ' ');
1121 eol = strchrnul(line, '\n');
1122
1123 if (!ptr || eol < ptr)
1124 ptr = eol;
1125 len = ptr - line;
1126
1127 if (40 < len)
1128 return 0;
1129 memcpy(patch->new_sha1_prefix, line, len);
1130 patch->new_sha1_prefix[len] = 0;
1131 if (*ptr == ' ')
44e5471a 1132 return gitdiff_oldmode(state, ptr + 1, patch);
13b5af22
CC
1133 return 0;
1134}
1135
1136/*
1137 * This is normal for a diff that doesn't change anything: we'll fall through
1138 * into the next diff. Tell the parser to break out.
1139 */
1140static int gitdiff_unrecognized(struct apply_state *state,
1141 const char *line,
1142 struct patch *patch)
1143{
1144 return 1;
1145}
1146
1147/*
1148 * Skip p_value leading components from "line"; as we do not accept
1149 * absolute paths, return NULL in that case.
1150 */
1151static const char *skip_tree_prefix(struct apply_state *state,
1152 const char *line,
1153 int llen)
1154{
1155 int nslash;
1156 int i;
1157
1158 if (!state->p_value)
1159 return (llen && line[0] == '/') ? NULL : line;
1160
1161 nslash = state->p_value;
1162 for (i = 0; i < llen; i++) {
1163 int ch = line[i];
1164 if (ch == '/' && --nslash <= 0)
1165 return (i == 0) ? NULL : &line[i + 1];
1166 }
1167 return NULL;
1168}
1169
1170/*
1171 * This is to extract the same name that appears on "diff --git"
1172 * line. We do not find and return anything if it is a rename
1173 * patch, and it is OK because we will find the name elsewhere.
1174 * We need to reliably find name only when it is mode-change only,
1175 * creation or deletion of an empty file. In any of these cases,
1176 * both sides are the same name under a/ and b/ respectively.
1177 */
1178static char *git_header_name(struct apply_state *state,
1179 const char *line,
1180 int llen)
1181{
1182 const char *name;
1183 const char *second = NULL;
1184 size_t len, line_len;
1185
1186 line += strlen("diff --git ");
1187 llen -= strlen("diff --git ");
1188
1189 if (*line == '"') {
1190 const char *cp;
1191 struct strbuf first = STRBUF_INIT;
1192 struct strbuf sp = STRBUF_INIT;
1193
1194 if (unquote_c_style(&first, line, &second))
1195 goto free_and_fail1;
1196
1197 /* strip the a/b prefix including trailing slash */
1198 cp = skip_tree_prefix(state, first.buf, first.len);
1199 if (!cp)
1200 goto free_and_fail1;
1201 strbuf_remove(&first, 0, cp - first.buf);
1202
1203 /*
1204 * second points at one past closing dq of name.
1205 * find the second name.
1206 */
1207 while ((second < line + llen) && isspace(*second))
1208 second++;
1209
1210 if (line + llen <= second)
1211 goto free_and_fail1;
1212 if (*second == '"') {
1213 if (unquote_c_style(&sp, second, NULL))
1214 goto free_and_fail1;
1215 cp = skip_tree_prefix(state, sp.buf, sp.len);
1216 if (!cp)
1217 goto free_and_fail1;
1218 /* They must match, otherwise ignore */
1219 if (strcmp(cp, first.buf))
1220 goto free_and_fail1;
1221 strbuf_release(&sp);
1222 return strbuf_detach(&first, NULL);
1223 }
1224
1225 /* unquoted second */
1226 cp = skip_tree_prefix(state, second, line + llen - second);
1227 if (!cp)
1228 goto free_and_fail1;
1229 if (line + llen - cp != first.len ||
1230 memcmp(first.buf, cp, first.len))
1231 goto free_and_fail1;
1232 return strbuf_detach(&first, NULL);
1233
1234 free_and_fail1:
1235 strbuf_release(&first);
1236 strbuf_release(&sp);
1237 return NULL;
1238 }
1239
1240 /* unquoted first name */
1241 name = skip_tree_prefix(state, line, llen);
1242 if (!name)
1243 return NULL;
1244
1245 /*
1246 * since the first name is unquoted, a dq if exists must be
1247 * the beginning of the second name.
1248 */
1249 for (second = name; second < line + llen; second++) {
1250 if (*second == '"') {
1251 struct strbuf sp = STRBUF_INIT;
1252 const char *np;
1253
1254 if (unquote_c_style(&sp, second, NULL))
1255 goto free_and_fail2;
1256
1257 np = skip_tree_prefix(state, sp.buf, sp.len);
1258 if (!np)
1259 goto free_and_fail2;
1260
1261 len = sp.buf + sp.len - np;
1262 if (len < second - name &&
1263 !strncmp(np, name, len) &&
1264 isspace(name[len])) {
1265 /* Good */
1266 strbuf_remove(&sp, 0, np - sp.buf);
1267 return strbuf_detach(&sp, NULL);
1268 }
1269
1270 free_and_fail2:
1271 strbuf_release(&sp);
1272 return NULL;
1273 }
1274 }
1275
1276 /*
1277 * Accept a name only if it shows up twice, exactly the same
1278 * form.
1279 */
1280 second = strchr(name, '\n');
1281 if (!second)
1282 return NULL;
1283 line_len = second - name;
1284 for (len = 0 ; ; len++) {
1285 switch (name[len]) {
1286 default:
1287 continue;
1288 case '\n':
1289 return NULL;
1290 case '\t': case ' ':
1291 /*
1292 * Is this the separator between the preimage
1293 * and the postimage pathname? Again, we are
1294 * only interested in the case where there is
1295 * no rename, as this is only to set def_name
1296 * and a rename patch has the names elsewhere
1297 * in an unambiguous form.
1298 */
1299 if (!name[len + 1])
1300 return NULL; /* no postimage name */
1301 second = skip_tree_prefix(state, name + len + 1,
1302 line_len - (len + 1));
1303 if (!second)
1304 return NULL;
1305 /*
1306 * Does len bytes starting at "name" and "second"
1307 * (that are separated by one HT or SP we just
1308 * found) exactly match?
1309 */
1310 if (second[len] == '\n' && !strncmp(name, second, len))
1311 return xmemdupz(name, len);
1312 }
1313 }
1314}
1315
d70e9c5c
RS
1316static int check_header_line(struct apply_state *state, struct patch *patch)
1317{
1318 int extensions = (patch->is_delete == 1) + (patch->is_new == 1) +
1319 (patch->is_rename == 1) + (patch->is_copy == 1);
1320 if (extensions > 1)
1321 return error(_("inconsistent header lines %d and %d"),
1322 patch->extension_linenr, state->linenr);
1323 if (extensions && !patch->extension_linenr)
1324 patch->extension_linenr = state->linenr;
1325 return 0;
1326}
1327
13b5af22
CC
1328/* Verify that we recognize the lines following a git header */
1329static int parse_git_header(struct apply_state *state,
1330 const char *line,
1331 int len,
1332 unsigned int size,
1333 struct patch *patch)
1334{
1335 unsigned long offset;
1336
1337 /* A git diff has explicit new/delete information, so we don't guess */
1338 patch->is_new = 0;
1339 patch->is_delete = 0;
1340
1341 /*
1342 * Some things may not have the old name in the
1343 * rest of the headers anywhere (pure mode changes,
1344 * or removing or adding empty files), so we get
1345 * the default name from the header.
1346 */
1347 patch->def_name = git_header_name(state, line, len);
1348 if (patch->def_name && state->root.len) {
1349 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1350 free(patch->def_name);
1351 patch->def_name = s;
1352 }
1353
1354 line += len;
1355 size -= len;
1356 state->linenr++;
1357 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1358 static const struct opentry {
1359 const char *str;
1360 int (*fn)(struct apply_state *, const char *, struct patch *);
1361 } optable[] = {
1362 { "@@ -", gitdiff_hdrend },
1363 { "--- ", gitdiff_oldname },
1364 { "+++ ", gitdiff_newname },
1365 { "old mode ", gitdiff_oldmode },
1366 { "new mode ", gitdiff_newmode },
1367 { "deleted file mode ", gitdiff_delete },
1368 { "new file mode ", gitdiff_newfile },
1369 { "copy from ", gitdiff_copysrc },
1370 { "copy to ", gitdiff_copydst },
1371 { "rename old ", gitdiff_renamesrc },
1372 { "rename new ", gitdiff_renamedst },
1373 { "rename from ", gitdiff_renamesrc },
1374 { "rename to ", gitdiff_renamedst },
1375 { "similarity index ", gitdiff_similarity },
1376 { "dissimilarity index ", gitdiff_dissimilarity },
1377 { "index ", gitdiff_index },
1378 { "", gitdiff_unrecognized },
1379 };
1380 int i;
1381
1382 len = linelen(line, size);
1383 if (!len || line[len-1] != '\n')
1384 break;
1385 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1386 const struct opentry *p = optable + i;
1387 int oplen = strlen(p->str);
1388 int res;
1389 if (len < oplen || memcmp(p->str, line, oplen))
1390 continue;
1391 res = p->fn(state, line + oplen, patch);
1392 if (res < 0)
1393 return -1;
d70e9c5c
RS
1394 if (check_header_line(state, patch))
1395 return -1;
13b5af22
CC
1396 if (res > 0)
1397 return offset;
1398 break;
1399 }
1400 }
1401
1402 return offset;
1403}
1404
1405static int parse_num(const char *line, unsigned long *p)
1406{
1407 char *ptr;
1408
1409 if (!isdigit(*line))
1410 return 0;
1411 *p = strtoul(line, &ptr, 10);
1412 return ptr - line;
1413}
1414
1415static int parse_range(const char *line, int len, int offset, const char *expect,
1416 unsigned long *p1, unsigned long *p2)
1417{
1418 int digits, ex;
1419
1420 if (offset < 0 || offset >= len)
1421 return -1;
1422 line += offset;
1423 len -= offset;
1424
1425 digits = parse_num(line, p1);
1426 if (!digits)
1427 return -1;
1428
1429 offset += digits;
1430 line += digits;
1431 len -= digits;
1432
1433 *p2 = 1;
1434 if (*line == ',') {
1435 digits = parse_num(line+1, p2);
1436 if (!digits)
1437 return -1;
1438
1439 offset += digits+1;
1440 line += digits+1;
1441 len -= digits+1;
1442 }
1443
1444 ex = strlen(expect);
1445 if (ex > len)
1446 return -1;
1447 if (memcmp(line, expect, ex))
1448 return -1;
1449
1450 return offset + ex;
1451}
1452
1453static void recount_diff(const char *line, int size, struct fragment *fragment)
1454{
1455 int oldlines = 0, newlines = 0, ret = 0;
1456
1457 if (size < 1) {
1458 warning("recount: ignore empty hunk");
1459 return;
1460 }
1461
1462 for (;;) {
1463 int len = linelen(line, size);
1464 size -= len;
1465 line += len;
1466
1467 if (size < 1)
1468 break;
1469
1470 switch (*line) {
1471 case ' ': case '\n':
1472 newlines++;
1473 /* fall through */
1474 case '-':
1475 oldlines++;
1476 continue;
1477 case '+':
1478 newlines++;
1479 continue;
1480 case '\\':
1481 continue;
1482 case '@':
1483 ret = size < 3 || !starts_with(line, "@@ ");
1484 break;
1485 case 'd':
1486 ret = size < 5 || !starts_with(line, "diff ");
1487 break;
1488 default:
1489 ret = -1;
1490 break;
1491 }
1492 if (ret) {
1493 warning(_("recount: unexpected line: %.*s"),
1494 (int)linelen(line, size), line);
1495 return;
1496 }
1497 break;
1498 }
1499 fragment->oldlines = oldlines;
1500 fragment->newlines = newlines;
1501}
1502
1503/*
1504 * Parse a unified diff fragment header of the
1505 * form "@@ -a,b +c,d @@"
1506 */
1507static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1508{
1509 int offset;
1510
1511 if (!len || line[len-1] != '\n')
1512 return -1;
1513
1514 /* Figure out the number of lines in a fragment */
1515 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1516 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1517
1518 return offset;
1519}
1520
1521/*
1522 * Find file diff header
1523 *
1524 * Returns:
1525 * -1 if no header was found
1526 * -128 in case of error
1527 * the size of the header in bytes (called "offset") otherwise
1528 */
1529static int find_header(struct apply_state *state,
1530 const char *line,
1531 unsigned long size,
1532 int *hdrsize,
1533 struct patch *patch)
1534{
1535 unsigned long offset, len;
1536
1537 patch->is_toplevel_relative = 0;
1538 patch->is_rename = patch->is_copy = 0;
1539 patch->is_new = patch->is_delete = -1;
1540 patch->old_mode = patch->new_mode = 0;
1541 patch->old_name = patch->new_name = NULL;
1542 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1543 unsigned long nextlen;
1544
1545 len = linelen(line, size);
1546 if (!len)
1547 break;
1548
1549 /* Testing this early allows us to take a few shortcuts.. */
1550 if (len < 6)
1551 continue;
1552
1553 /*
1554 * Make sure we don't find any unconnected patch fragments.
1555 * That's a sign that we didn't find a header, and that a
1556 * patch has become corrupted/broken up.
1557 */
1558 if (!memcmp("@@ -", line, 4)) {
1559 struct fragment dummy;
1560 if (parse_fragment_header(line, len, &dummy) < 0)
1561 continue;
1562 error(_("patch fragment without header at line %d: %.*s"),
1563 state->linenr, (int)len-1, line);
1564 return -128;
1565 }
1566
1567 if (size < len + 6)
1568 break;
1569
1570 /*
1571 * Git patch? It might not have a real patch, just a rename
1572 * or mode change, so we handle that specially
1573 */
1574 if (!memcmp("diff --git ", line, 11)) {
1575 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1576 if (git_hdr_len < 0)
1577 return -128;
1578 if (git_hdr_len <= len)
1579 continue;
1580 if (!patch->old_name && !patch->new_name) {
1581 if (!patch->def_name) {
1582 error(Q_("git diff header lacks filename information when removing "
1583 "%d leading pathname component (line %d)",
1584 "git diff header lacks filename information when removing "
1585 "%d leading pathname components (line %d)",
1586 state->p_value),
1587 state->p_value, state->linenr);
1588 return -128;
1589 }
1590 patch->old_name = xstrdup(patch->def_name);
1591 patch->new_name = xstrdup(patch->def_name);
1592 }
42699741
RS
1593 if ((!patch->new_name && !patch->is_delete) ||
1594 (!patch->old_name && !patch->is_new)) {
f25dfb5e
VA
1595 error(_("git diff header lacks filename information "
1596 "(line %d)"), state->linenr);
13b5af22
CC
1597 return -128;
1598 }
1599 patch->is_toplevel_relative = 1;
1600 *hdrsize = git_hdr_len;
1601 return offset;
1602 }
1603
1604 /* --- followed by +++ ? */
1605 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1606 continue;
1607
1608 /*
1609 * We only accept unified patches, so we want it to
1610 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1611 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1612 */
1613 nextlen = linelen(line + len, size - len);
1614 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1615 continue;
1616
1617 /* Ok, we'll consider it a patch */
1618 if (parse_traditional_patch(state, line, line+len, patch))
1619 return -128;
1620 *hdrsize = len + nextlen;
1621 state->linenr += 2;
1622 return offset;
1623 }
1624 return -1;
1625}
1626
1627static void record_ws_error(struct apply_state *state,
1628 unsigned result,
1629 const char *line,
1630 int len,
1631 int linenr)
1632{
1633 char *err;
1634
1635 if (!result)
1636 return;
1637
1638 state->whitespace_error++;
1639 if (state->squelch_whitespace_errors &&
1640 state->squelch_whitespace_errors < state->whitespace_error)
1641 return;
1642
1643 err = whitespace_error_string(result);
a46160d2
CC
1644 if (state->apply_verbosity > verbosity_silent)
1645 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1646 state->patch_input_file, linenr, err, len, line);
13b5af22
CC
1647 free(err);
1648}
1649
1650static void check_whitespace(struct apply_state *state,
1651 const char *line,
1652 int len,
1653 unsigned ws_rule)
1654{
1655 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1656
1657 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1658}
1659
c24f3aba
TB
1660/*
1661 * Check if the patch has context lines with CRLF or
1662 * the patch wants to remove lines with CRLF.
1663 */
1664static void check_old_for_crlf(struct patch *patch, const char *line, int len)
1665{
1666 if (len >= 2 && line[len-1] == '\n' && line[len-2] == '\r') {
1667 patch->ws_rule |= WS_CR_AT_EOL;
1668 patch->crlf_in_old = 1;
1669 }
1670}
1671
1672
13b5af22
CC
1673/*
1674 * Parse a unified diff. Note that this really needs to parse each
1675 * fragment separately, since the only way to know the difference
1676 * between a "---" that is part of a patch, and a "---" that starts
1677 * the next patch is to look at the line counts..
1678 */
1679static int parse_fragment(struct apply_state *state,
1680 const char *line,
1681 unsigned long size,
1682 struct patch *patch,
1683 struct fragment *fragment)
1684{
1685 int added, deleted;
1686 int len = linelen(line, size), offset;
1687 unsigned long oldlines, newlines;
1688 unsigned long leading, trailing;
1689
1690 offset = parse_fragment_header(line, len, fragment);
1691 if (offset < 0)
1692 return -1;
1693 if (offset > 0 && patch->recount)
1694 recount_diff(line + offset, size - offset, fragment);
1695 oldlines = fragment->oldlines;
1696 newlines = fragment->newlines;
1697 leading = 0;
1698 trailing = 0;
1699
1700 /* Parse the thing.. */
1701 line += len;
1702 size -= len;
1703 state->linenr++;
1704 added = deleted = 0;
1705 for (offset = len;
1706 0 < size;
1707 offset += len, size -= len, line += len, state->linenr++) {
1708 if (!oldlines && !newlines)
1709 break;
1710 len = linelen(line, size);
1711 if (!len || line[len-1] != '\n')
1712 return -1;
1713 switch (*line) {
1714 default:
1715 return -1;
1716 case '\n': /* newer GNU diff, an empty context line */
1717 case ' ':
1718 oldlines--;
1719 newlines--;
1720 if (!deleted && !added)
1721 leading++;
1722 trailing++;
c24f3aba 1723 check_old_for_crlf(patch, line, len);
13b5af22
CC
1724 if (!state->apply_in_reverse &&
1725 state->ws_error_action == correct_ws_error)
1726 check_whitespace(state, line, len, patch->ws_rule);
1727 break;
1728 case '-':
c24f3aba
TB
1729 if (!state->apply_in_reverse)
1730 check_old_for_crlf(patch, line, len);
13b5af22
CC
1731 if (state->apply_in_reverse &&
1732 state->ws_error_action != nowarn_ws_error)
1733 check_whitespace(state, line, len, patch->ws_rule);
1734 deleted++;
1735 oldlines--;
1736 trailing = 0;
1737 break;
1738 case '+':
c24f3aba
TB
1739 if (state->apply_in_reverse)
1740 check_old_for_crlf(patch, line, len);
13b5af22
CC
1741 if (!state->apply_in_reverse &&
1742 state->ws_error_action != nowarn_ws_error)
1743 check_whitespace(state, line, len, patch->ws_rule);
1744 added++;
1745 newlines--;
1746 trailing = 0;
1747 break;
1748
1749 /*
1750 * We allow "\ No newline at end of file". Depending
1751 * on locale settings when the patch was produced we
1752 * don't know what this line looks like. The only
1753 * thing we do know is that it begins with "\ ".
1754 * Checking for 12 is just for sanity check -- any
1755 * l10n of "\ No newline..." is at least that long.
1756 */
1757 case '\\':
1758 if (len < 12 || memcmp(line, "\\ ", 2))
1759 return -1;
1760 break;
1761 }
1762 }
1763 if (oldlines || newlines)
1764 return -1;
1765 if (!deleted && !added)
1766 return -1;
1767
1768 fragment->leading = leading;
1769 fragment->trailing = trailing;
1770
1771 /*
1772 * If a fragment ends with an incomplete line, we failed to include
1773 * it in the above loop because we hit oldlines == newlines == 0
1774 * before seeing it.
1775 */
1776 if (12 < size && !memcmp(line, "\\ ", 2))
1777 offset += linelen(line, size);
1778
1779 patch->lines_added += added;
1780 patch->lines_deleted += deleted;
1781
1782 if (0 < patch->is_new && oldlines)
1783 return error(_("new file depends on old contents"));
1784 if (0 < patch->is_delete && newlines)
1785 return error(_("deleted file still has contents"));
1786 return offset;
1787}
1788
1789/*
1790 * We have seen "diff --git a/... b/..." header (or a traditional patch
1791 * header). Read hunks that belong to this patch into fragments and hang
1792 * them to the given patch structure.
1793 *
1794 * The (fragment->patch, fragment->size) pair points into the memory given
1795 * by the caller, not a copy, when we return.
1796 *
1797 * Returns:
1798 * -1 in case of error,
1799 * the number of bytes in the patch otherwise.
1800 */
1801static int parse_single_patch(struct apply_state *state,
1802 const char *line,
1803 unsigned long size,
1804 struct patch *patch)
1805{
1806 unsigned long offset = 0;
1807 unsigned long oldlines = 0, newlines = 0, context = 0;
1808 struct fragment **fragp = &patch->fragments;
1809
1810 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1811 struct fragment *fragment;
1812 int len;
1813
1814 fragment = xcalloc(1, sizeof(*fragment));
1815 fragment->linenr = state->linenr;
1816 len = parse_fragment(state, line, size, patch, fragment);
1817 if (len <= 0) {
1818 free(fragment);
1819 return error(_("corrupt patch at line %d"), state->linenr);
1820 }
1821 fragment->patch = line;
1822 fragment->size = len;
1823 oldlines += fragment->oldlines;
1824 newlines += fragment->newlines;
1825 context += fragment->leading + fragment->trailing;
1826
1827 *fragp = fragment;
1828 fragp = &fragment->next;
1829
1830 offset += len;
1831 line += len;
1832 size -= len;
1833 }
1834
1835 /*
1836 * If something was removed (i.e. we have old-lines) it cannot
1837 * be creation, and if something was added it cannot be
1838 * deletion. However, the reverse is not true; --unified=0
1839 * patches that only add are not necessarily creation even
1840 * though they do not have any old lines, and ones that only
1841 * delete are not necessarily deletion.
1842 *
1843 * Unfortunately, a real creation/deletion patch do _not_ have
1844 * any context line by definition, so we cannot safely tell it
1845 * apart with --unified=0 insanity. At least if the patch has
1846 * more than one hunk it is not creation or deletion.
1847 */
1848 if (patch->is_new < 0 &&
1849 (oldlines || (patch->fragments && patch->fragments->next)))
1850 patch->is_new = 0;
1851 if (patch->is_delete < 0 &&
1852 (newlines || (patch->fragments && patch->fragments->next)))
1853 patch->is_delete = 0;
1854
1855 if (0 < patch->is_new && oldlines)
1856 return error(_("new file %s depends on old contents"), patch->new_name);
1857 if (0 < patch->is_delete && newlines)
1858 return error(_("deleted file %s still has contents"), patch->old_name);
a46160d2 1859 if (!patch->is_delete && !newlines && context && state->apply_verbosity > verbosity_silent)
13b5af22
CC
1860 fprintf_ln(stderr,
1861 _("** warning: "
1862 "file %s becomes empty but is not deleted"),
1863 patch->new_name);
1864
1865 return offset;
1866}
1867
1868static inline int metadata_changes(struct patch *patch)
1869{
1870 return patch->is_rename > 0 ||
1871 patch->is_copy > 0 ||
1872 patch->is_new > 0 ||
1873 patch->is_delete ||
1874 (patch->old_mode && patch->new_mode &&
1875 patch->old_mode != patch->new_mode);
1876}
1877
1878static char *inflate_it(const void *data, unsigned long size,
1879 unsigned long inflated_size)
1880{
1881 git_zstream stream;
1882 void *out;
1883 int st;
1884
1885 memset(&stream, 0, sizeof(stream));
1886
1887 stream.next_in = (unsigned char *)data;
1888 stream.avail_in = size;
1889 stream.next_out = out = xmalloc(inflated_size);
1890 stream.avail_out = inflated_size;
1891 git_inflate_init(&stream);
1892 st = git_inflate(&stream, Z_FINISH);
1893 git_inflate_end(&stream);
1894 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1895 free(out);
1896 return NULL;
1897 }
1898 return out;
1899}
1900
1901/*
1902 * Read a binary hunk and return a new fragment; fragment->patch
1903 * points at an allocated memory that the caller must free, so
1904 * it is marked as "->free_patch = 1".
1905 */
1906static struct fragment *parse_binary_hunk(struct apply_state *state,
1907 char **buf_p,
1908 unsigned long *sz_p,
1909 int *status_p,
1910 int *used_p)
1911{
1912 /*
1913 * Expect a line that begins with binary patch method ("literal"
1914 * or "delta"), followed by the length of data before deflating.
1915 * a sequence of 'length-byte' followed by base-85 encoded data
1916 * should follow, terminated by a newline.
1917 *
1918 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1919 * and we would limit the patch line to 66 characters,
1920 * so one line can fit up to 13 groups that would decode
1921 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1922 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1923 */
1924 int llen, used;
1925 unsigned long size = *sz_p;
1926 char *buffer = *buf_p;
1927 int patch_method;
1928 unsigned long origlen;
1929 char *data = NULL;
1930 int hunk_size = 0;
1931 struct fragment *frag;
1932
1933 llen = linelen(buffer, size);
1934 used = llen;
1935
1936 *status_p = 0;
1937
1938 if (starts_with(buffer, "delta ")) {
1939 patch_method = BINARY_DELTA_DEFLATED;
1940 origlen = strtoul(buffer + 6, NULL, 10);
1941 }
1942 else if (starts_with(buffer, "literal ")) {
1943 patch_method = BINARY_LITERAL_DEFLATED;
1944 origlen = strtoul(buffer + 8, NULL, 10);
1945 }
1946 else
1947 return NULL;
1948
1949 state->linenr++;
1950 buffer += llen;
1951 while (1) {
1952 int byte_length, max_byte_length, newsize;
1953 llen = linelen(buffer, size);
1954 used += llen;
1955 state->linenr++;
1956 if (llen == 1) {
1957 /* consume the blank line */
1958 buffer++;
1959 size--;
1960 break;
1961 }
1962 /*
1963 * Minimum line is "A00000\n" which is 7-byte long,
1964 * and the line length must be multiple of 5 plus 2.
1965 */
1966 if ((llen < 7) || (llen-2) % 5)
1967 goto corrupt;
1968 max_byte_length = (llen - 2) / 5 * 4;
1969 byte_length = *buffer;
1970 if ('A' <= byte_length && byte_length <= 'Z')
1971 byte_length = byte_length - 'A' + 1;
1972 else if ('a' <= byte_length && byte_length <= 'z')
1973 byte_length = byte_length - 'a' + 27;
1974 else
1975 goto corrupt;
1976 /* if the input length was not multiple of 4, we would
1977 * have filler at the end but the filler should never
1978 * exceed 3 bytes
1979 */
1980 if (max_byte_length < byte_length ||
1981 byte_length <= max_byte_length - 4)
1982 goto corrupt;
1983 newsize = hunk_size + byte_length;
1984 data = xrealloc(data, newsize);
1985 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1986 goto corrupt;
1987 hunk_size = newsize;
1988 buffer += llen;
1989 size -= llen;
1990 }
1991
1992 frag = xcalloc(1, sizeof(*frag));
1993 frag->patch = inflate_it(data, hunk_size, origlen);
1994 frag->free_patch = 1;
1995 if (!frag->patch)
1996 goto corrupt;
1997 free(data);
1998 frag->size = origlen;
1999 *buf_p = buffer;
2000 *sz_p = size;
2001 *used_p = used;
2002 frag->binary_patch_method = patch_method;
2003 return frag;
2004
2005 corrupt:
2006 free(data);
2007 *status_p = -1;
2008 error(_("corrupt binary patch at line %d: %.*s"),
2009 state->linenr-1, llen-1, buffer);
2010 return NULL;
2011}
2012
2013/*
2014 * Returns:
2015 * -1 in case of error,
2016 * the length of the parsed binary patch otherwise
2017 */
2018static int parse_binary(struct apply_state *state,
2019 char *buffer,
2020 unsigned long size,
2021 struct patch *patch)
2022{
2023 /*
2024 * We have read "GIT binary patch\n"; what follows is a line
2025 * that says the patch method (currently, either "literal" or
2026 * "delta") and the length of data before deflating; a
2027 * sequence of 'length-byte' followed by base-85 encoded data
2028 * follows.
2029 *
2030 * When a binary patch is reversible, there is another binary
2031 * hunk in the same format, starting with patch method (either
2032 * "literal" or "delta") with the length of data, and a sequence
2033 * of length-byte + base-85 encoded data, terminated with another
2034 * empty line. This data, when applied to the postimage, produces
2035 * the preimage.
2036 */
2037 struct fragment *forward;
2038 struct fragment *reverse;
2039 int status;
2040 int used, used_1;
2041
2042 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
2043 if (!forward && !status)
2044 /* there has to be one hunk (forward hunk) */
2045 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
2046 if (status)
2047 /* otherwise we already gave an error message */
2048 return status;
2049
2050 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
2051 if (reverse)
2052 used += used_1;
2053 else if (status) {
2054 /*
2055 * Not having reverse hunk is not an error, but having
2056 * a corrupt reverse hunk is.
2057 */
2058 free((void*) forward->patch);
2059 free(forward);
2060 return status;
2061 }
2062 forward->next = reverse;
2063 patch->fragments = forward;
2064 patch->is_binary = 1;
2065 return used;
2066}
2067
2068static void prefix_one(struct apply_state *state, char **name)
2069{
2070 char *old_name = *name;
2071 if (!old_name)
2072 return;
e4da43b1 2073 *name = prefix_filename(state->prefix, *name);
13b5af22
CC
2074 free(old_name);
2075}
2076
2077static void prefix_patch(struct apply_state *state, struct patch *p)
2078{
2079 if (!state->prefix || p->is_toplevel_relative)
2080 return;
2081 prefix_one(state, &p->new_name);
2082 prefix_one(state, &p->old_name);
2083}
2084
2085/*
2086 * include/exclude
2087 */
2088
2089static void add_name_limit(struct apply_state *state,
2090 const char *name,
2091 int exclude)
2092{
2093 struct string_list_item *it;
2094
2095 it = string_list_append(&state->limit_by_name, name);
2096 it->util = exclude ? NULL : (void *) 1;
2097}
2098
2099static int use_patch(struct apply_state *state, struct patch *p)
2100{
2101 const char *pathname = p->new_name ? p->new_name : p->old_name;
2102 int i;
2103
2104 /* Paths outside are not touched regardless of "--include" */
881529c8
RS
2105 if (state->prefix && *state->prefix) {
2106 const char *rest;
2107 if (!skip_prefix(pathname, state->prefix, &rest) || !*rest)
13b5af22
CC
2108 return 0;
2109 }
2110
2111 /* See if it matches any of exclude/include rule */
2112 for (i = 0; i < state->limit_by_name.nr; i++) {
2113 struct string_list_item *it = &state->limit_by_name.items[i];
55d34269 2114 if (!wildmatch(it->string, pathname, 0))
13b5af22
CC
2115 return (it->util != NULL);
2116 }
2117
2118 /*
2119 * If we had any include, a path that does not match any rule is
2120 * not used. Otherwise, we saw bunch of exclude rules (or none)
2121 * and such a path is used.
2122 */
2123 return !state->has_include;
2124}
2125
2126/*
2127 * Read the patch text in "buffer" that extends for "size" bytes; stop
2128 * reading after seeing a single patch (i.e. changes to a single file).
2129 * Create fragments (i.e. patch hunks) and hang them to the given patch.
2130 *
2131 * Returns:
2132 * -1 if no header was found or parse_binary() failed,
2133 * -128 on another error,
2134 * the number of bytes consumed otherwise,
2135 * so that the caller can call us again for the next patch.
2136 */
2137static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
2138{
2139 int hdrsize, patchsize;
2140 int offset = find_header(state, buffer, size, &hdrsize, patch);
2141
2142 if (offset < 0)
2143 return offset;
2144
2145 prefix_patch(state, patch);
2146
2147 if (!use_patch(state, patch))
2148 patch->ws_rule = 0;
2149 else
2150 patch->ws_rule = whitespace_rule(patch->new_name
2151 ? patch->new_name
2152 : patch->old_name);
2153
2154 patchsize = parse_single_patch(state,
2155 buffer + offset + hdrsize,
2156 size - offset - hdrsize,
2157 patch);
2158
2159 if (patchsize < 0)
2160 return -128;
2161
2162 if (!patchsize) {
2163 static const char git_binary[] = "GIT binary patch\n";
2164 int hd = hdrsize + offset;
2165 unsigned long llen = linelen(buffer + hd, size - hd);
2166
2167 if (llen == sizeof(git_binary) - 1 &&
2168 !memcmp(git_binary, buffer + hd, llen)) {
2169 int used;
2170 state->linenr++;
2171 used = parse_binary(state, buffer + hd + llen,
2172 size - hd - llen, patch);
2173 if (used < 0)
2174 return -1;
2175 if (used)
2176 patchsize = used + llen;
2177 else
2178 patchsize = 0;
2179 }
2180 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2181 static const char *binhdr[] = {
2182 "Binary files ",
2183 "Files ",
2184 NULL,
2185 };
2186 int i;
2187 for (i = 0; binhdr[i]; i++) {
2188 int len = strlen(binhdr[i]);
2189 if (len < size - hd &&
2190 !memcmp(binhdr[i], buffer + hd, len)) {
2191 state->linenr++;
2192 patch->is_binary = 1;
2193 patchsize = llen;
2194 break;
2195 }
2196 }
2197 }
2198
2199 /* Empty patch cannot be applied if it is a text patch
2200 * without metadata change. A binary patch appears
2201 * empty to us here.
2202 */
2203 if ((state->apply || state->check) &&
2204 (!patch->is_binary && !metadata_changes(patch))) {
2205 error(_("patch with only garbage at line %d"), state->linenr);
2206 return -128;
2207 }
2208 }
2209
2210 return offset + hdrsize + patchsize;
2211}
2212
13b5af22
CC
2213static void reverse_patches(struct patch *p)
2214{
2215 for (; p; p = p->next) {
2216 struct fragment *frag = p->fragments;
2217
db101991
RS
2218 SWAP(p->new_name, p->old_name);
2219 SWAP(p->new_mode, p->old_mode);
2220 SWAP(p->is_new, p->is_delete);
2221 SWAP(p->lines_added, p->lines_deleted);
2222 SWAP(p->old_sha1_prefix, p->new_sha1_prefix);
13b5af22
CC
2223
2224 for (; frag; frag = frag->next) {
db101991
RS
2225 SWAP(frag->newpos, frag->oldpos);
2226 SWAP(frag->newlines, frag->oldlines);
13b5af22
CC
2227 }
2228 }
2229}
2230
2231static const char pluses[] =
2232"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2233static const char minuses[]=
2234"----------------------------------------------------------------------";
2235
2236static void show_stats(struct apply_state *state, struct patch *patch)
2237{
2238 struct strbuf qname = STRBUF_INIT;
2239 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2240 int max, add, del;
2241
2242 quote_c_style(cp, &qname, NULL, 0);
2243
2244 /*
2245 * "scale" the filename
2246 */
2247 max = state->max_len;
2248 if (max > 50)
2249 max = 50;
2250
2251 if (qname.len > max) {
2252 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2253 if (!cp)
2254 cp = qname.buf + qname.len + 3 - max;
2255 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2256 }
2257
2258 if (patch->is_binary) {
2259 printf(" %-*s | Bin\n", max, qname.buf);
2260 strbuf_release(&qname);
2261 return;
2262 }
2263
2264 printf(" %-*s |", max, qname.buf);
2265 strbuf_release(&qname);
2266
2267 /*
2268 * scale the add/delete
2269 */
2270 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2271 add = patch->lines_added;
2272 del = patch->lines_deleted;
2273
2274 if (state->max_change > 0) {
2275 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2276 add = (add * max + state->max_change / 2) / state->max_change;
2277 del = total - add;
2278 }
2279 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2280 add, pluses, del, minuses);
2281}
2282
c24f3aba
TB
2283static int read_old_data(struct stat *st, struct patch *patch,
2284 const char *path, struct strbuf *buf)
13b5af22 2285{
c24f3aba
TB
2286 enum safe_crlf safe_crlf = patch->crlf_in_old ?
2287 SAFE_CRLF_KEEP_CRLF : SAFE_CRLF_RENORMALIZE;
13b5af22
CC
2288 switch (st->st_mode & S_IFMT) {
2289 case S_IFLNK:
2290 if (strbuf_readlink(buf, path, st->st_size) < 0)
2291 return error(_("unable to read symlink %s"), path);
2292 return 0;
2293 case S_IFREG:
2294 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2295 return error(_("unable to open or read %s"), path);
c24f3aba
TB
2296 /*
2297 * "git apply" without "--index/--cached" should never look
2298 * at the index; the target file may not have been added to
2299 * the index yet, and we may not even be in any Git repository.
2300 * Pass NULL to convert_to_git() to stress this; the function
2301 * should never look at the index when explicit crlf option
2302 * is given.
2303 */
2304 convert_to_git(NULL, path, buf->buf, buf->len, buf, safe_crlf);
13b5af22
CC
2305 return 0;
2306 default:
2307 return -1;
2308 }
2309}
2310
2311/*
2312 * Update the preimage, and the common lines in postimage,
2313 * from buffer buf of length len. If postlen is 0 the postimage
2314 * is updated in place, otherwise it's updated on a new buffer
2315 * of length postlen
2316 */
2317
2318static void update_pre_post_images(struct image *preimage,
2319 struct image *postimage,
2320 char *buf,
2321 size_t len, size_t postlen)
2322{
2323 int i, ctx, reduced;
2324 char *new, *old, *fixed;
2325 struct image fixed_preimage;
2326
2327 /*
2328 * Update the preimage with whitespace fixes. Note that we
2329 * are not losing preimage->buf -- apply_one_fragment() will
2330 * free "oldlines".
2331 */
2332 prepare_image(&fixed_preimage, buf, len, 1);
2333 assert(postlen
2334 ? fixed_preimage.nr == preimage->nr
2335 : fixed_preimage.nr <= preimage->nr);
2336 for (i = 0; i < fixed_preimage.nr; i++)
2337 fixed_preimage.line[i].flag = preimage->line[i].flag;
2338 free(preimage->line_allocated);
2339 *preimage = fixed_preimage;
2340
2341 /*
2342 * Adjust the common context lines in postimage. This can be
2343 * done in-place when we are shrinking it with whitespace
2344 * fixing, but needs a new buffer when ignoring whitespace or
2345 * expanding leading tabs to spaces.
2346 *
2347 * We trust the caller to tell us if the update can be done
2348 * in place (postlen==0) or not.
2349 */
2350 old = postimage->buf;
2351 if (postlen)
2352 new = postimage->buf = xmalloc(postlen);
2353 else
2354 new = old;
2355 fixed = preimage->buf;
2356
2357 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2358 size_t l_len = postimage->line[i].len;
2359 if (!(postimage->line[i].flag & LINE_COMMON)) {
2360 /* an added line -- no counterparts in preimage */
2361 memmove(new, old, l_len);
2362 old += l_len;
2363 new += l_len;
2364 continue;
2365 }
2366
2367 /* a common context -- skip it in the original postimage */
2368 old += l_len;
2369
2370 /* and find the corresponding one in the fixed preimage */
2371 while (ctx < preimage->nr &&
2372 !(preimage->line[ctx].flag & LINE_COMMON)) {
2373 fixed += preimage->line[ctx].len;
2374 ctx++;
2375 }
2376
2377 /*
2378 * preimage is expected to run out, if the caller
2379 * fixed addition of trailing blank lines.
2380 */
2381 if (preimage->nr <= ctx) {
2382 reduced++;
2383 continue;
2384 }
2385
2386 /* and copy it in, while fixing the line length */
2387 l_len = preimage->line[ctx].len;
2388 memcpy(new, fixed, l_len);
2389 new += l_len;
2390 fixed += l_len;
2391 postimage->line[i].len = l_len;
2392 ctx++;
2393 }
2394
2395 if (postlen
2396 ? postlen < new - postimage->buf
2397 : postimage->len < new - postimage->buf)
2398 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2399 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2400
2401 /* Fix the length of the whole thing */
2402 postimage->len = new - postimage->buf;
2403 postimage->nr -= reduced;
2404}
2405
2406static int line_by_line_fuzzy_match(struct image *img,
2407 struct image *preimage,
2408 struct image *postimage,
2409 unsigned long try,
2410 int try_lno,
2411 int preimage_limit)
2412{
2413 int i;
2414 size_t imgoff = 0;
2415 size_t preoff = 0;
2416 size_t postlen = postimage->len;
2417 size_t extra_chars;
2418 char *buf;
2419 char *preimage_eof;
2420 char *preimage_end;
2421 struct strbuf fixed;
2422 char *fixed_buf;
2423 size_t fixed_len;
2424
2425 for (i = 0; i < preimage_limit; i++) {
2426 size_t prelen = preimage->line[i].len;
2427 size_t imglen = img->line[try_lno+i].len;
2428
2429 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2430 preimage->buf + preoff, prelen))
2431 return 0;
2432 if (preimage->line[i].flag & LINE_COMMON)
2433 postlen += imglen - prelen;
2434 imgoff += imglen;
2435 preoff += prelen;
2436 }
2437
2438 /*
2439 * Ok, the preimage matches with whitespace fuzz.
2440 *
2441 * imgoff now holds the true length of the target that
2442 * matches the preimage before the end of the file.
2443 *
2444 * Count the number of characters in the preimage that fall
2445 * beyond the end of the file and make sure that all of them
2446 * are whitespace characters. (This can only happen if
2447 * we are removing blank lines at the end of the file.)
2448 */
2449 buf = preimage_eof = preimage->buf + preoff;
2450 for ( ; i < preimage->nr; i++)
2451 preoff += preimage->line[i].len;
2452 preimage_end = preimage->buf + preoff;
2453 for ( ; buf < preimage_end; buf++)
2454 if (!isspace(*buf))
2455 return 0;
2456
2457 /*
2458 * Update the preimage and the common postimage context
2459 * lines to use the same whitespace as the target.
2460 * If whitespace is missing in the target (i.e.
2461 * if the preimage extends beyond the end of the file),
2462 * use the whitespace from the preimage.
2463 */
2464 extra_chars = preimage_end - preimage_eof;
2465 strbuf_init(&fixed, imgoff + extra_chars);
2466 strbuf_add(&fixed, img->buf + try, imgoff);
2467 strbuf_add(&fixed, preimage_eof, extra_chars);
2468 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2469 update_pre_post_images(preimage, postimage,
2470 fixed_buf, fixed_len, postlen);
2471 return 1;
2472}
2473
2474static int match_fragment(struct apply_state *state,
2475 struct image *img,
2476 struct image *preimage,
2477 struct image *postimage,
2478 unsigned long try,
2479 int try_lno,
2480 unsigned ws_rule,
2481 int match_beginning, int match_end)
2482{
2483 int i;
2484 char *fixed_buf, *buf, *orig, *target;
2485 struct strbuf fixed;
2486 size_t fixed_len, postlen;
2487 int preimage_limit;
2488
2489 if (preimage->nr + try_lno <= img->nr) {
2490 /*
2491 * The hunk falls within the boundaries of img.
2492 */
2493 preimage_limit = preimage->nr;
2494 if (match_end && (preimage->nr + try_lno != img->nr))
2495 return 0;
2496 } else if (state->ws_error_action == correct_ws_error &&
2497 (ws_rule & WS_BLANK_AT_EOF)) {
2498 /*
2499 * This hunk extends beyond the end of img, and we are
2500 * removing blank lines at the end of the file. This
2501 * many lines from the beginning of the preimage must
2502 * match with img, and the remainder of the preimage
2503 * must be blank.
2504 */
2505 preimage_limit = img->nr - try_lno;
2506 } else {
2507 /*
2508 * The hunk extends beyond the end of the img and
2509 * we are not removing blanks at the end, so we
2510 * should reject the hunk at this position.
2511 */
2512 return 0;
2513 }
2514
2515 if (match_beginning && try_lno)
2516 return 0;
2517
2518 /* Quick hash check */
2519 for (i = 0; i < preimage_limit; i++)
2520 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2521 (preimage->line[i].hash != img->line[try_lno + i].hash))
2522 return 0;
2523
2524 if (preimage_limit == preimage->nr) {
2525 /*
2526 * Do we have an exact match? If we were told to match
2527 * at the end, size must be exactly at try+fragsize,
2528 * otherwise try+fragsize must be still within the preimage,
2529 * and either case, the old piece should match the preimage
2530 * exactly.
2531 */
2532 if ((match_end
2533 ? (try + preimage->len == img->len)
2534 : (try + preimage->len <= img->len)) &&
2535 !memcmp(img->buf + try, preimage->buf, preimage->len))
2536 return 1;
2537 } else {
2538 /*
2539 * The preimage extends beyond the end of img, so
2540 * there cannot be an exact match.
2541 *
2542 * There must be one non-blank context line that match
2543 * a line before the end of img.
2544 */
2545 char *buf_end;
2546
2547 buf = preimage->buf;
2548 buf_end = buf;
2549 for (i = 0; i < preimage_limit; i++)
2550 buf_end += preimage->line[i].len;
2551
2552 for ( ; buf < buf_end; buf++)
2553 if (!isspace(*buf))
2554 break;
2555 if (buf == buf_end)
2556 return 0;
2557 }
2558
2559 /*
2560 * No exact match. If we are ignoring whitespace, run a line-by-line
2561 * fuzzy matching. We collect all the line length information because
2562 * we need it to adjust whitespace if we match.
2563 */
2564 if (state->ws_ignore_action == ignore_ws_change)
2565 return line_by_line_fuzzy_match(img, preimage, postimage,
2566 try, try_lno, preimage_limit);
2567
2568 if (state->ws_error_action != correct_ws_error)
2569 return 0;
2570
2571 /*
2572 * The hunk does not apply byte-by-byte, but the hash says
2573 * it might with whitespace fuzz. We weren't asked to
2574 * ignore whitespace, we were asked to correct whitespace
2575 * errors, so let's try matching after whitespace correction.
2576 *
2577 * While checking the preimage against the target, whitespace
2578 * errors in both fixed, we count how large the corresponding
2579 * postimage needs to be. The postimage prepared by
2580 * apply_one_fragment() has whitespace errors fixed on added
2581 * lines already, but the common lines were propagated as-is,
2582 * which may become longer when their whitespace errors are
2583 * fixed.
2584 */
2585
2586 /* First count added lines in postimage */
2587 postlen = 0;
2588 for (i = 0; i < postimage->nr; i++) {
2589 if (!(postimage->line[i].flag & LINE_COMMON))
2590 postlen += postimage->line[i].len;
2591 }
2592
2593 /*
2594 * The preimage may extend beyond the end of the file,
2595 * but in this loop we will only handle the part of the
2596 * preimage that falls within the file.
2597 */
2598 strbuf_init(&fixed, preimage->len + 1);
2599 orig = preimage->buf;
2600 target = img->buf + try;
2601 for (i = 0; i < preimage_limit; i++) {
2602 size_t oldlen = preimage->line[i].len;
2603 size_t tgtlen = img->line[try_lno + i].len;
2604 size_t fixstart = fixed.len;
2605 struct strbuf tgtfix;
2606 int match;
2607
2608 /* Try fixing the line in the preimage */
2609 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2610
2611 /* Try fixing the line in the target */
2612 strbuf_init(&tgtfix, tgtlen);
2613 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2614
2615 /*
2616 * If they match, either the preimage was based on
2617 * a version before our tree fixed whitespace breakage,
2618 * or we are lacking a whitespace-fix patch the tree
2619 * the preimage was based on already had (i.e. target
2620 * has whitespace breakage, the preimage doesn't).
2621 * In either case, we are fixing the whitespace breakages
2622 * so we might as well take the fix together with their
2623 * real change.
2624 */
2625 match = (tgtfix.len == fixed.len - fixstart &&
2626 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2627 fixed.len - fixstart));
2628
2629 /* Add the length if this is common with the postimage */
2630 if (preimage->line[i].flag & LINE_COMMON)
2631 postlen += tgtfix.len;
2632
2633 strbuf_release(&tgtfix);
2634 if (!match)
2635 goto unmatch_exit;
2636
2637 orig += oldlen;
2638 target += tgtlen;
2639 }
2640
2641
2642 /*
2643 * Now handle the lines in the preimage that falls beyond the
2644 * end of the file (if any). They will only match if they are
2645 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2646 * false).
2647 */
2648 for ( ; i < preimage->nr; i++) {
2649 size_t fixstart = fixed.len; /* start of the fixed preimage */
2650 size_t oldlen = preimage->line[i].len;
2651 int j;
2652
2653 /* Try fixing the line in the preimage */
2654 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2655
2656 for (j = fixstart; j < fixed.len; j++)
2657 if (!isspace(fixed.buf[j]))
2658 goto unmatch_exit;
2659
2660 orig += oldlen;
2661 }
2662
2663 /*
2664 * Yes, the preimage is based on an older version that still
2665 * has whitespace breakages unfixed, and fixing them makes the
2666 * hunk match. Update the context lines in the postimage.
2667 */
2668 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2669 if (postlen < postimage->len)
2670 postlen = 0;
2671 update_pre_post_images(preimage, postimage,
2672 fixed_buf, fixed_len, postlen);
2673 return 1;
2674
2675 unmatch_exit:
2676 strbuf_release(&fixed);
2677 return 0;
2678}
2679
2680static int find_pos(struct apply_state *state,
2681 struct image *img,
2682 struct image *preimage,
2683 struct image *postimage,
2684 int line,
2685 unsigned ws_rule,
2686 int match_beginning, int match_end)
2687{
2688 int i;
2689 unsigned long backwards, forwards, try;
2690 int backwards_lno, forwards_lno, try_lno;
2691
2692 /*
2693 * If match_beginning or match_end is specified, there is no
2694 * point starting from a wrong line that will never match and
2695 * wander around and wait for a match at the specified end.
2696 */
2697 if (match_beginning)
2698 line = 0;
2699 else if (match_end)
2700 line = img->nr - preimage->nr;
2701
2702 /*
2703 * Because the comparison is unsigned, the following test
2704 * will also take care of a negative line number that can
2705 * result when match_end and preimage is larger than the target.
2706 */
2707 if ((size_t) line > img->nr)
2708 line = img->nr;
2709
2710 try = 0;
2711 for (i = 0; i < line; i++)
2712 try += img->line[i].len;
2713
2714 /*
2715 * There's probably some smart way to do this, but I'll leave
2716 * that to the smart and beautiful people. I'm simple and stupid.
2717 */
2718 backwards = try;
2719 backwards_lno = line;
2720 forwards = try;
2721 forwards_lno = line;
2722 try_lno = line;
2723
2724 for (i = 0; ; i++) {
2725 if (match_fragment(state, img, preimage, postimage,
2726 try, try_lno, ws_rule,
2727 match_beginning, match_end))
2728 return try_lno;
2729
2730 again:
2731 if (backwards_lno == 0 && forwards_lno == img->nr)
2732 break;
2733
2734 if (i & 1) {
2735 if (backwards_lno == 0) {
2736 i++;
2737 goto again;
2738 }
2739 backwards_lno--;
2740 backwards -= img->line[backwards_lno].len;
2741 try = backwards;
2742 try_lno = backwards_lno;
2743 } else {
2744 if (forwards_lno == img->nr) {
2745 i++;
2746 goto again;
2747 }
2748 forwards += img->line[forwards_lno].len;
2749 forwards_lno++;
2750 try = forwards;
2751 try_lno = forwards_lno;
2752 }
2753
2754 }
2755 return -1;
2756}
2757
2758static void remove_first_line(struct image *img)
2759{
2760 img->buf += img->line[0].len;
2761 img->len -= img->line[0].len;
2762 img->line++;
2763 img->nr--;
2764}
2765
2766static void remove_last_line(struct image *img)
2767{
2768 img->len -= img->line[--img->nr].len;
2769}
2770
2771/*
2772 * The change from "preimage" and "postimage" has been found to
2773 * apply at applied_pos (counts in line numbers) in "img".
2774 * Update "img" to remove "preimage" and replace it with "postimage".
2775 */
2776static void update_image(struct apply_state *state,
2777 struct image *img,
2778 int applied_pos,
2779 struct image *preimage,
2780 struct image *postimage)
2781{
2782 /*
2783 * remove the copy of preimage at offset in img
2784 * and replace it with postimage
2785 */
2786 int i, nr;
2787 size_t remove_count, insert_count, applied_at = 0;
2788 char *result;
2789 int preimage_limit;
2790
2791 /*
2792 * If we are removing blank lines at the end of img,
2793 * the preimage may extend beyond the end.
2794 * If that is the case, we must be careful only to
2795 * remove the part of the preimage that falls within
2796 * the boundaries of img. Initialize preimage_limit
2797 * to the number of lines in the preimage that falls
2798 * within the boundaries.
2799 */
2800 preimage_limit = preimage->nr;
2801 if (preimage_limit > img->nr - applied_pos)
2802 preimage_limit = img->nr - applied_pos;
2803
2804 for (i = 0; i < applied_pos; i++)
2805 applied_at += img->line[i].len;
2806
2807 remove_count = 0;
2808 for (i = 0; i < preimage_limit; i++)
2809 remove_count += img->line[applied_pos + i].len;
2810 insert_count = postimage->len;
2811
2812 /* Adjust the contents */
2813 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2814 memcpy(result, img->buf, applied_at);
2815 memcpy(result + applied_at, postimage->buf, postimage->len);
2816 memcpy(result + applied_at + postimage->len,
2817 img->buf + (applied_at + remove_count),
2818 img->len - (applied_at + remove_count));
2819 free(img->buf);
2820 img->buf = result;
2821 img->len += insert_count - remove_count;
2822 result[img->len] = '\0';
2823
2824 /* Adjust the line table */
2825 nr = img->nr + postimage->nr - preimage_limit;
2826 if (preimage_limit < postimage->nr) {
2827 /*
2828 * NOTE: this knows that we never call remove_first_line()
2829 * on anything other than pre/post image.
2830 */
2831 REALLOC_ARRAY(img->line, nr);
2832 img->line_allocated = img->line;
2833 }
2834 if (preimage_limit != postimage->nr)
17736641
RS
2835 MOVE_ARRAY(img->line + applied_pos + postimage->nr,
2836 img->line + applied_pos + preimage_limit,
2837 img->nr - (applied_pos + preimage_limit));
2838 COPY_ARRAY(img->line + applied_pos, postimage->line, postimage->nr);
13b5af22
CC
2839 if (!state->allow_overlap)
2840 for (i = 0; i < postimage->nr; i++)
2841 img->line[applied_pos + i].flag |= LINE_PATCHED;
2842 img->nr = nr;
2843}
2844
2845/*
2846 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2847 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2848 * replace the part of "img" with "postimage" text.
2849 */
2850static int apply_one_fragment(struct apply_state *state,
2851 struct image *img, struct fragment *frag,
2852 int inaccurate_eof, unsigned ws_rule,
2853 int nth_fragment)
2854{
2855 int match_beginning, match_end;
2856 const char *patch = frag->patch;
2857 int size = frag->size;
2858 char *old, *oldlines;
2859 struct strbuf newlines;
2860 int new_blank_lines_at_end = 0;
2861 int found_new_blank_lines_at_end = 0;
2862 int hunk_linenr = frag->linenr;
2863 unsigned long leading, trailing;
2864 int pos, applied_pos;
2865 struct image preimage;
2866 struct image postimage;
2867
2868 memset(&preimage, 0, sizeof(preimage));
2869 memset(&postimage, 0, sizeof(postimage));
2870 oldlines = xmalloc(size);
2871 strbuf_init(&newlines, size);
2872
2873 old = oldlines;
2874 while (size > 0) {
2875 char first;
2876 int len = linelen(patch, size);
2877 int plen;
2878 int added_blank_line = 0;
2879 int is_blank_context = 0;
2880 size_t start;
2881
2882 if (!len)
2883 break;
2884
2885 /*
2886 * "plen" is how much of the line we should use for
2887 * the actual patch data. Normally we just remove the
2888 * first character on the line, but if the line is
2889 * followed by "\ No newline", then we also remove the
2890 * last one (which is the newline, of course).
2891 */
2892 plen = len - 1;
2893 if (len < size && patch[len] == '\\')
2894 plen--;
2895 first = *patch;
2896 if (state->apply_in_reverse) {
2897 if (first == '-')
2898 first = '+';
2899 else if (first == '+')
2900 first = '-';
2901 }
2902
2903 switch (first) {
2904 case '\n':
2905 /* Newer GNU diff, empty context line */
2906 if (plen < 0)
2907 /* ... followed by '\No newline'; nothing */
2908 break;
2909 *old++ = '\n';
2910 strbuf_addch(&newlines, '\n');
2911 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2912 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2913 is_blank_context = 1;
2914 break;
2915 case ' ':
2916 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2917 ws_blank_line(patch + 1, plen, ws_rule))
2918 is_blank_context = 1;
1cf01a34 2919 /* fallthrough */
13b5af22
CC
2920 case '-':
2921 memcpy(old, patch + 1, plen);
2922 add_line_info(&preimage, old, plen,
2923 (first == ' ' ? LINE_COMMON : 0));
2924 old += plen;
2925 if (first == '-')
2926 break;
1cf01a34 2927 /* fallthrough */
13b5af22
CC
2928 case '+':
2929 /* --no-add does not add new lines */
2930 if (first == '+' && state->no_add)
2931 break;
2932
2933 start = newlines.len;
2934 if (first != '+' ||
2935 !state->whitespace_error ||
2936 state->ws_error_action != correct_ws_error) {
2937 strbuf_add(&newlines, patch + 1, plen);
2938 }
2939 else {
2940 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2941 }
2942 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2943 (first == '+' ? 0 : LINE_COMMON));
2944 if (first == '+' &&
2945 (ws_rule & WS_BLANK_AT_EOF) &&
2946 ws_blank_line(patch + 1, plen, ws_rule))
2947 added_blank_line = 1;
2948 break;
2949 case '@': case '\\':
2950 /* Ignore it, we already handled it */
2951 break;
2952 default:
a46160d2 2953 if (state->apply_verbosity > verbosity_normal)
13b5af22
CC
2954 error(_("invalid start of line: '%c'"), first);
2955 applied_pos = -1;
2956 goto out;
2957 }
2958 if (added_blank_line) {
2959 if (!new_blank_lines_at_end)
2960 found_new_blank_lines_at_end = hunk_linenr;
2961 new_blank_lines_at_end++;
2962 }
2963 else if (is_blank_context)
2964 ;
2965 else
2966 new_blank_lines_at_end = 0;
2967 patch += len;
2968 size -= len;
2969 hunk_linenr++;
2970 }
2971 if (inaccurate_eof &&
2972 old > oldlines && old[-1] == '\n' &&
2973 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2974 old--;
2975 strbuf_setlen(&newlines, newlines.len - 1);
2976 }
2977
2978 leading = frag->leading;
2979 trailing = frag->trailing;
2980
2981 /*
2982 * A hunk to change lines at the beginning would begin with
2983 * @@ -1,L +N,M @@
2984 * but we need to be careful. -U0 that inserts before the second
2985 * line also has this pattern.
2986 *
2987 * And a hunk to add to an empty file would begin with
2988 * @@ -0,0 +N,M @@
2989 *
2990 * In other words, a hunk that is (frag->oldpos <= 1) with or
2991 * without leading context must match at the beginning.
2992 */
2993 match_beginning = (!frag->oldpos ||
2994 (frag->oldpos == 1 && !state->unidiff_zero));
2995
2996 /*
2997 * A hunk without trailing lines must match at the end.
2998 * However, we simply cannot tell if a hunk must match end
2999 * from the lack of trailing lines if the patch was generated
3000 * with unidiff without any context.
3001 */
3002 match_end = !state->unidiff_zero && !trailing;
3003
3004 pos = frag->newpos ? (frag->newpos - 1) : 0;
3005 preimage.buf = oldlines;
3006 preimage.len = old - oldlines;
3007 postimage.buf = newlines.buf;
3008 postimage.len = newlines.len;
3009 preimage.line = preimage.line_allocated;
3010 postimage.line = postimage.line_allocated;
3011
3012 for (;;) {
3013
3014 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
3015 ws_rule, match_beginning, match_end);
3016
3017 if (applied_pos >= 0)
3018 break;
3019
3020 /* Am I at my context limits? */
3021 if ((leading <= state->p_context) && (trailing <= state->p_context))
3022 break;
3023 if (match_beginning || match_end) {
3024 match_beginning = match_end = 0;
3025 continue;
3026 }
3027
3028 /*
3029 * Reduce the number of context lines; reduce both
3030 * leading and trailing if they are equal otherwise
3031 * just reduce the larger context.
3032 */
3033 if (leading >= trailing) {
3034 remove_first_line(&preimage);
3035 remove_first_line(&postimage);
3036 pos--;
3037 leading--;
3038 }
3039 if (trailing > leading) {
3040 remove_last_line(&preimage);
3041 remove_last_line(&postimage);
3042 trailing--;
3043 }
3044 }
3045
3046 if (applied_pos >= 0) {
3047 if (new_blank_lines_at_end &&
3048 preimage.nr + applied_pos >= img->nr &&
3049 (ws_rule & WS_BLANK_AT_EOF) &&
3050 state->ws_error_action != nowarn_ws_error) {
3051 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
3052 found_new_blank_lines_at_end);
3053 if (state->ws_error_action == correct_ws_error) {
3054 while (new_blank_lines_at_end--)
3055 remove_last_line(&postimage);
3056 }
3057 /*
3058 * We would want to prevent write_out_results()
3059 * from taking place in apply_patch() that follows
3060 * the callchain led us here, which is:
3061 * apply_patch->check_patch_list->check_patch->
3062 * apply_data->apply_fragments->apply_one_fragment
3063 */
3064 if (state->ws_error_action == die_on_ws_error)
3065 state->apply = 0;
3066 }
3067
a46160d2 3068 if (state->apply_verbosity > verbosity_normal && applied_pos != pos) {
13b5af22
CC
3069 int offset = applied_pos - pos;
3070 if (state->apply_in_reverse)
3071 offset = 0 - offset;
3072 fprintf_ln(stderr,
3073 Q_("Hunk #%d succeeded at %d (offset %d line).",
3074 "Hunk #%d succeeded at %d (offset %d lines).",
3075 offset),
3076 nth_fragment, applied_pos + 1, offset);
3077 }
3078
3079 /*
3080 * Warn if it was necessary to reduce the number
3081 * of context lines.
3082 */
a46160d2
CC
3083 if ((leading != frag->leading ||
3084 trailing != frag->trailing) && state->apply_verbosity > verbosity_silent)
13b5af22
CC
3085 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
3086 " to apply fragment at %d"),
3087 leading, trailing, applied_pos+1);
3088 update_image(state, img, applied_pos, &preimage, &postimage);
3089 } else {
a46160d2 3090 if (state->apply_verbosity > verbosity_normal)
13b5af22
CC
3091 error(_("while searching for:\n%.*s"),
3092 (int)(old - oldlines), oldlines);
3093 }
3094
3095out:
3096 free(oldlines);
3097 strbuf_release(&newlines);
3098 free(preimage.line_allocated);
3099 free(postimage.line_allocated);
3100
3101 return (applied_pos < 0);
3102}
3103
3104static int apply_binary_fragment(struct apply_state *state,
3105 struct image *img,
3106 struct patch *patch)
3107{
3108 struct fragment *fragment = patch->fragments;
3109 unsigned long len;
3110 void *dst;
3111
3112 if (!fragment)
3113 return error(_("missing binary patch data for '%s'"),
3114 patch->new_name ?
3115 patch->new_name :
3116 patch->old_name);
3117
3118 /* Binary patch is irreversible without the optional second hunk */
3119 if (state->apply_in_reverse) {
3120 if (!fragment->next)
d1d42bf5
VA
3121 return error(_("cannot reverse-apply a binary patch "
3122 "without the reverse hunk to '%s'"),
13b5af22
CC
3123 patch->new_name
3124 ? patch->new_name : patch->old_name);
3125 fragment = fragment->next;
3126 }
3127 switch (fragment->binary_patch_method) {
3128 case BINARY_DELTA_DEFLATED:
3129 dst = patch_delta(img->buf, img->len, fragment->patch,
3130 fragment->size, &len);
3131 if (!dst)
3132 return -1;
3133 clear_image(img);
3134 img->buf = dst;
3135 img->len = len;
3136 return 0;
3137 case BINARY_LITERAL_DEFLATED:
3138 clear_image(img);
3139 img->len = fragment->size;
3140 img->buf = xmemdupz(fragment->patch, img->len);
3141 return 0;
3142 }
3143 return -1;
3144}
3145
3146/*
3147 * Replace "img" with the result of applying the binary patch.
3148 * The binary patch data itself in patch->fragment is still kept
3149 * but the preimage prepared by the caller in "img" is freed here
3150 * or in the helper function apply_binary_fragment() this calls.
3151 */
3152static int apply_binary(struct apply_state *state,
3153 struct image *img,
3154 struct patch *patch)
3155{
3156 const char *name = patch->old_name ? patch->old_name : patch->new_name;
4af9a7d3 3157 struct object_id oid;
13b5af22
CC
3158
3159 /*
3160 * For safety, we require patch index line to contain
3161 * full 40-byte textual SHA1 for old and new, at least for now.
3162 */
3163 if (strlen(patch->old_sha1_prefix) != 40 ||
3164 strlen(patch->new_sha1_prefix) != 40 ||
4af9a7d3
JH
3165 get_oid_hex(patch->old_sha1_prefix, &oid) ||
3166 get_oid_hex(patch->new_sha1_prefix, &oid))
d1d42bf5
VA
3167 return error(_("cannot apply binary patch to '%s' "
3168 "without full index line"), name);
13b5af22
CC
3169
3170 if (patch->old_name) {
3171 /*
3172 * See if the old one matches what the patch
3173 * applies to.
3174 */
4af9a7d3
JH
3175 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3176 if (strcmp(oid_to_hex(&oid), patch->old_sha1_prefix))
d1d42bf5
VA
3177 return error(_("the patch applies to '%s' (%s), "
3178 "which does not match the "
3179 "current contents."),
4af9a7d3 3180 name, oid_to_hex(&oid));
13b5af22
CC
3181 }
3182 else {
3183 /* Otherwise, the old one must be empty. */
3184 if (img->len)
d1d42bf5
VA
3185 return error(_("the patch applies to an empty "
3186 "'%s' but it is not empty"), name);
13b5af22
CC
3187 }
3188
4af9a7d3
JH
3189 get_oid_hex(patch->new_sha1_prefix, &oid);
3190 if (is_null_oid(&oid)) {
13b5af22
CC
3191 clear_image(img);
3192 return 0; /* deletion patch */
3193 }
3194
4af9a7d3 3195 if (has_sha1_file(oid.hash)) {
13b5af22
CC
3196 /* We already have the postimage */
3197 enum object_type type;
3198 unsigned long size;
3199 char *result;
3200
4af9a7d3 3201 result = read_sha1_file(oid.hash, &type, &size);
13b5af22 3202 if (!result)
d1d42bf5
VA
3203 return error(_("the necessary postimage %s for "
3204 "'%s' cannot be read"),
13b5af22
CC
3205 patch->new_sha1_prefix, name);
3206 clear_image(img);
3207 img->buf = result;
3208 img->len = size;
3209 } else {
3210 /*
3211 * We have verified buf matches the preimage;
3212 * apply the patch data to it, which is stored
3213 * in the patch->fragments->{patch,size}.
3214 */
3215 if (apply_binary_fragment(state, img, patch))
3216 return error(_("binary patch does not apply to '%s'"),
3217 name);
3218
3219 /* verify that the result matches */
4af9a7d3
JH
3220 hash_sha1_file(img->buf, img->len, blob_type, oid.hash);
3221 if (strcmp(oid_to_hex(&oid), patch->new_sha1_prefix))
13b5af22 3222 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
4af9a7d3 3223 name, patch->new_sha1_prefix, oid_to_hex(&oid));
13b5af22
CC
3224 }
3225
3226 return 0;
3227}
3228
3229static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3230{
3231 struct fragment *frag = patch->fragments;
3232 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3233 unsigned ws_rule = patch->ws_rule;
3234 unsigned inaccurate_eof = patch->inaccurate_eof;
3235 int nth = 0;
3236
3237 if (patch->is_binary)
3238 return apply_binary(state, img, patch);
3239
3240 while (frag) {
3241 nth++;
3242 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3243 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3244 if (!state->apply_with_reject)
3245 return -1;
3246 frag->rejected = 1;
3247 }
3248 frag = frag->next;
3249 }
3250 return 0;
3251}
3252
4af9a7d3 3253static int read_blob_object(struct strbuf *buf, const struct object_id *oid, unsigned mode)
13b5af22
CC
3254{
3255 if (S_ISGITLINK(mode)) {
3256 strbuf_grow(buf, 100);
4af9a7d3 3257 strbuf_addf(buf, "Subproject commit %s\n", oid_to_hex(oid));
13b5af22
CC
3258 } else {
3259 enum object_type type;
3260 unsigned long sz;
3261 char *result;
3262
4af9a7d3 3263 result = read_sha1_file(oid->hash, &type, &sz);
13b5af22
CC
3264 if (!result)
3265 return -1;
3266 /* XXX read_sha1_file NUL-terminates */
3267 strbuf_attach(buf, result, sz, sz + 1);
3268 }
3269 return 0;
3270}
3271
3272static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3273{
3274 if (!ce)
3275 return 0;
4af9a7d3 3276 return read_blob_object(buf, &ce->oid, ce->ce_mode);
13b5af22
CC
3277}
3278
3279static struct patch *in_fn_table(struct apply_state *state, const char *name)
3280{
3281 struct string_list_item *item;
3282
3283 if (name == NULL)
3284 return NULL;
3285
3286 item = string_list_lookup(&state->fn_table, name);
3287 if (item != NULL)
3288 return (struct patch *)item->util;
3289
3290 return NULL;
3291}
3292
3293/*
3294 * item->util in the filename table records the status of the path.
3295 * Usually it points at a patch (whose result records the contents
3296 * of it after applying it), but it could be PATH_WAS_DELETED for a
3297 * path that a previously applied patch has already removed, or
3298 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3299 *
3300 * The latter is needed to deal with a case where two paths A and B
3301 * are swapped by first renaming A to B and then renaming B to A;
3302 * moving A to B should not be prevented due to presence of B as we
3303 * will remove it in a later patch.
3304 */
3305#define PATH_TO_BE_DELETED ((struct patch *) -2)
3306#define PATH_WAS_DELETED ((struct patch *) -1)
3307
3308static int to_be_deleted(struct patch *patch)
3309{
3310 return patch == PATH_TO_BE_DELETED;
3311}
3312
3313static int was_deleted(struct patch *patch)
3314{
3315 return patch == PATH_WAS_DELETED;
3316}
3317
3318static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3319{
3320 struct string_list_item *item;
3321
3322 /*
3323 * Always add new_name unless patch is a deletion
3324 * This should cover the cases for normal diffs,
3325 * file creations and copies
3326 */
3327 if (patch->new_name != NULL) {
3328 item = string_list_insert(&state->fn_table, patch->new_name);
3329 item->util = patch;
3330 }
3331
3332 /*
3333 * store a failure on rename/deletion cases because
3334 * later chunks shouldn't patch old names
3335 */
3336 if ((patch->new_name == NULL) || (patch->is_rename)) {
3337 item = string_list_insert(&state->fn_table, patch->old_name);
3338 item->util = PATH_WAS_DELETED;
3339 }
3340}
3341
3342static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3343{
3344 /*
3345 * store information about incoming file deletion
3346 */
3347 while (patch) {
3348 if ((patch->new_name == NULL) || (patch->is_rename)) {
3349 struct string_list_item *item;
3350 item = string_list_insert(&state->fn_table, patch->old_name);
3351 item->util = PATH_TO_BE_DELETED;
3352 }
3353 patch = patch->next;
3354 }
3355}
3356
3357static int checkout_target(struct index_state *istate,
3358 struct cache_entry *ce, struct stat *st)
3359{
68e3d629 3360 struct checkout costate = CHECKOUT_INIT;
13b5af22 3361
13b5af22
CC
3362 costate.refresh_cache = 1;
3363 costate.istate = istate;
3364 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3365 return error(_("cannot checkout %s"), ce->name);
3366 return 0;
3367}
3368
3369static struct patch *previous_patch(struct apply_state *state,
3370 struct patch *patch,
3371 int *gone)
3372{
3373 struct patch *previous;
3374
3375 *gone = 0;
3376 if (patch->is_copy || patch->is_rename)
3377 return NULL; /* "git" patches do not depend on the order */
3378
3379 previous = in_fn_table(state, patch->old_name);
3380 if (!previous)
3381 return NULL;
3382
3383 if (to_be_deleted(previous))
3384 return NULL; /* the deletion hasn't happened yet */
3385
3386 if (was_deleted(previous))
3387 *gone = 1;
3388
3389 return previous;
3390}
3391
3392static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3393{
3394 if (S_ISGITLINK(ce->ce_mode)) {
3395 if (!S_ISDIR(st->st_mode))
3396 return -1;
3397 return 0;
3398 }
3399 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3400}
3401
3402#define SUBMODULE_PATCH_WITHOUT_INDEX 1
3403
3404static int load_patch_target(struct apply_state *state,
3405 struct strbuf *buf,
3406 const struct cache_entry *ce,
3407 struct stat *st,
c24f3aba 3408 struct patch *patch,
13b5af22
CC
3409 const char *name,
3410 unsigned expected_mode)
3411{
3412 if (state->cached || state->check_index) {
3413 if (read_file_or_gitlink(ce, buf))
3414 return error(_("failed to read %s"), name);
3415 } else if (name) {
3416 if (S_ISGITLINK(expected_mode)) {
3417 if (ce)
3418 return read_file_or_gitlink(ce, buf);
3419 else
3420 return SUBMODULE_PATCH_WITHOUT_INDEX;
3421 } else if (has_symlink_leading_path(name, strlen(name))) {
3422 return error(_("reading from '%s' beyond a symbolic link"), name);
3423 } else {
c24f3aba 3424 if (read_old_data(st, patch, name, buf))
13b5af22
CC
3425 return error(_("failed to read %s"), name);
3426 }
3427 }
3428 return 0;
3429}
3430
3431/*
3432 * We are about to apply "patch"; populate the "image" with the
3433 * current version we have, from the working tree or from the index,
3434 * depending on the situation e.g. --cached/--index. If we are
3435 * applying a non-git patch that incrementally updates the tree,
3436 * we read from the result of a previous diff.
3437 */
3438static int load_preimage(struct apply_state *state,
3439 struct image *image,
3440 struct patch *patch, struct stat *st,
3441 const struct cache_entry *ce)
3442{
3443 struct strbuf buf = STRBUF_INIT;
3444 size_t len;
3445 char *img;
3446 struct patch *previous;
3447 int status;
3448
3449 previous = previous_patch(state, patch, &status);
3450 if (status)
3451 return error(_("path %s has been renamed/deleted"),
3452 patch->old_name);
3453 if (previous) {
3454 /* We have a patched copy in memory; use that. */
3455 strbuf_add(&buf, previous->result, previous->resultsize);
3456 } else {
c24f3aba 3457 status = load_patch_target(state, &buf, ce, st, patch,
13b5af22
CC
3458 patch->old_name, patch->old_mode);
3459 if (status < 0)
3460 return status;
3461 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3462 /*
3463 * There is no way to apply subproject
3464 * patch without looking at the index.
3465 * NEEDSWORK: shouldn't this be flagged
3466 * as an error???
3467 */
3468 free_fragment_list(patch->fragments);
3469 patch->fragments = NULL;
3470 } else if (status) {
3471 return error(_("failed to read %s"), patch->old_name);
3472 }
3473 }
3474
3475 img = strbuf_detach(&buf, &len);
3476 prepare_image(image, img, len, !patch->is_binary);
3477 return 0;
3478}
3479
3480static int three_way_merge(struct image *image,
3481 char *path,
4af9a7d3
JH
3482 const struct object_id *base,
3483 const struct object_id *ours,
3484 const struct object_id *theirs)
13b5af22
CC
3485{
3486 mmfile_t base_file, our_file, their_file;
3487 mmbuffer_t result = { NULL };
3488 int status;
3489
3490 read_mmblob(&base_file, base);
3491 read_mmblob(&our_file, ours);
3492 read_mmblob(&their_file, theirs);
3493 status = ll_merge(&result, path,
3494 &base_file, "base",
3495 &our_file, "ours",
3496 &their_file, "theirs", NULL);
3497 free(base_file.ptr);
3498 free(our_file.ptr);
3499 free(their_file.ptr);
3500 if (status < 0 || !result.ptr) {
3501 free(result.ptr);
3502 return -1;
3503 }
3504 clear_image(image);
3505 image->buf = result.ptr;
3506 image->len = result.size;
3507
3508 return status;
3509}
3510
3511/*
3512 * When directly falling back to add/add three-way merge, we read from
3513 * the current contents of the new_name. In no cases other than that
3514 * this function will be called.
3515 */
3516static int load_current(struct apply_state *state,
3517 struct image *image,
3518 struct patch *patch)
3519{
3520 struct strbuf buf = STRBUF_INIT;
3521 int status, pos;
3522 size_t len;
3523 char *img;
3524 struct stat st;
3525 struct cache_entry *ce;
3526 char *name = patch->new_name;
3527 unsigned mode = patch->new_mode;
3528
3529 if (!patch->is_new)
3530 die("BUG: patch to %s is not a creation", patch->old_name);
3531
3532 pos = cache_name_pos(name, strlen(name));
3533 if (pos < 0)
3534 return error(_("%s: does not exist in index"), name);
3535 ce = active_cache[pos];
3536 if (lstat(name, &st)) {
3537 if (errno != ENOENT)
90875eca 3538 return error_errno("%s", name);
13b5af22
CC
3539 if (checkout_target(&the_index, ce, &st))
3540 return -1;
3541 }
3542 if (verify_index_match(ce, &st))
3543 return error(_("%s: does not match index"), name);
3544
c24f3aba 3545 status = load_patch_target(state, &buf, ce, &st, patch, name, mode);
13b5af22
CC
3546 if (status < 0)
3547 return status;
3548 else if (status)
3549 return -1;
3550 img = strbuf_detach(&buf, &len);
3551 prepare_image(image, img, len, !patch->is_binary);
3552 return 0;
3553}
3554
3555static int try_threeway(struct apply_state *state,
3556 struct image *image,
3557 struct patch *patch,
3558 struct stat *st,
3559 const struct cache_entry *ce)
3560{
4af9a7d3 3561 struct object_id pre_oid, post_oid, our_oid;
13b5af22
CC
3562 struct strbuf buf = STRBUF_INIT;
3563 size_t len;
3564 int status;
3565 char *img;
3566 struct image tmp_image;
3567
3568 /* No point falling back to 3-way merge in these cases */
3569 if (patch->is_delete ||
3570 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3571 return -1;
3572
3573 /* Preimage the patch was prepared for */
3574 if (patch->is_new)
4af9a7d3 3575 write_sha1_file("", 0, blob_type, pre_oid.hash);
e82caf38 3576 else if (get_oid(patch->old_sha1_prefix, &pre_oid) ||
4af9a7d3 3577 read_blob_object(&buf, &pre_oid, patch->old_mode))
d1d42bf5 3578 return error(_("repository lacks the necessary blob to fall back on 3-way merge."));
13b5af22 3579
a46160d2 3580 if (state->apply_verbosity > verbosity_silent)
5886637a 3581 fprintf(stderr, _("Falling back to three-way merge...\n"));
13b5af22
CC
3582
3583 img = strbuf_detach(&buf, &len);
3584 prepare_image(&tmp_image, img, len, 1);
3585 /* Apply the patch to get the post image */
3586 if (apply_fragments(state, &tmp_image, patch) < 0) {
3587 clear_image(&tmp_image);
3588 return -1;
3589 }
4af9a7d3
JH
3590 /* post_oid is theirs */
3591 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_oid.hash);
13b5af22
CC
3592 clear_image(&tmp_image);
3593
4af9a7d3 3594 /* our_oid is ours */
13b5af22
CC
3595 if (patch->is_new) {
3596 if (load_current(state, &tmp_image, patch))
d1d42bf5 3597 return error(_("cannot read the current contents of '%s'"),
13b5af22
CC
3598 patch->new_name);
3599 } else {
3600 if (load_preimage(state, &tmp_image, patch, st, ce))
d1d42bf5 3601 return error(_("cannot read the current contents of '%s'"),
13b5af22
CC
3602 patch->old_name);
3603 }
4af9a7d3 3604 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_oid.hash);
13b5af22
CC
3605 clear_image(&tmp_image);
3606
3607 /* in-core three-way merge between post and our using pre as base */
3608 status = three_way_merge(image, patch->new_name,
4af9a7d3 3609 &pre_oid, &our_oid, &post_oid);
13b5af22 3610 if (status < 0) {
a46160d2
CC
3611 if (state->apply_verbosity > verbosity_silent)
3612 fprintf(stderr,
5886637a 3613 _("Failed to fall back on three-way merge...\n"));
13b5af22
CC
3614 return status;
3615 }
3616
3617 if (status) {
3618 patch->conflicted_threeway = 1;
3619 if (patch->is_new)
3620 oidclr(&patch->threeway_stage[0]);
3621 else
4af9a7d3
JH
3622 oidcpy(&patch->threeway_stage[0], &pre_oid);
3623 oidcpy(&patch->threeway_stage[1], &our_oid);
3624 oidcpy(&patch->threeway_stage[2], &post_oid);
a46160d2
CC
3625 if (state->apply_verbosity > verbosity_silent)
3626 fprintf(stderr,
5886637a 3627 _("Applied patch to '%s' with conflicts.\n"),
a46160d2 3628 patch->new_name);
13b5af22 3629 } else {
a46160d2
CC
3630 if (state->apply_verbosity > verbosity_silent)
3631 fprintf(stderr,
5886637a 3632 _("Applied patch to '%s' cleanly.\n"),
a46160d2 3633 patch->new_name);
13b5af22
CC
3634 }
3635 return 0;
3636}
3637
3638static int apply_data(struct apply_state *state, struct patch *patch,
3639 struct stat *st, const struct cache_entry *ce)
3640{
3641 struct image image;
3642
3643 if (load_preimage(state, &image, patch, st, ce) < 0)
3644 return -1;
3645
3646 if (patch->direct_to_threeway ||
3647 apply_fragments(state, &image, patch) < 0) {
3648 /* Note: with --reject, apply_fragments() returns 0 */
3649 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3650 return -1;
3651 }
3652 patch->result = image.buf;
3653 patch->resultsize = image.len;
3654 add_to_fn_table(state, patch);
3655 free(image.line_allocated);
3656
3657 if (0 < patch->is_delete && patch->resultsize)
3658 return error(_("removal patch leaves file contents"));
3659
3660 return 0;
3661}
3662
3663/*
3664 * If "patch" that we are looking at modifies or deletes what we have,
3665 * we would want it not to lose any local modification we have, either
3666 * in the working tree or in the index.
3667 *
3668 * This also decides if a non-git patch is a creation patch or a
3669 * modification to an existing empty file. We do not check the state
3670 * of the current tree for a creation patch in this function; the caller
3671 * check_patch() separately makes sure (and errors out otherwise) that
3672 * the path the patch creates does not exist in the current tree.
3673 */
3674static int check_preimage(struct apply_state *state,
3675 struct patch *patch,
3676 struct cache_entry **ce,
3677 struct stat *st)
3678{
3679 const char *old_name = patch->old_name;
3680 struct patch *previous = NULL;
3681 int stat_ret = 0, status;
3682 unsigned st_mode = 0;
3683
3684 if (!old_name)
3685 return 0;
3686
3687 assert(patch->is_new <= 0);
3688 previous = previous_patch(state, patch, &status);
3689
3690 if (status)
3691 return error(_("path %s has been renamed/deleted"), old_name);
3692 if (previous) {
3693 st_mode = previous->new_mode;
3694 } else if (!state->cached) {
3695 stat_ret = lstat(old_name, st);
3696 if (stat_ret && errno != ENOENT)
90875eca 3697 return error_errno("%s", old_name);
13b5af22
CC
3698 }
3699
3700 if (state->check_index && !previous) {
3701 int pos = cache_name_pos(old_name, strlen(old_name));
3702 if (pos < 0) {
3703 if (patch->is_new < 0)
3704 goto is_new;
3705 return error(_("%s: does not exist in index"), old_name);
3706 }
3707 *ce = active_cache[pos];
3708 if (stat_ret < 0) {
3709 if (checkout_target(&the_index, *ce, st))
3710 return -1;
3711 }
3712 if (!state->cached && verify_index_match(*ce, st))
3713 return error(_("%s: does not match index"), old_name);
3714 if (state->cached)
3715 st_mode = (*ce)->ce_mode;
3716 } else if (stat_ret < 0) {
3717 if (patch->is_new < 0)
3718 goto is_new;
90875eca 3719 return error_errno("%s", old_name);
13b5af22
CC
3720 }
3721
3722 if (!state->cached && !previous)
3723 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3724
3725 if (patch->is_new < 0)
3726 patch->is_new = 0;
3727 if (!patch->old_mode)
3728 patch->old_mode = st_mode;
3729 if ((st_mode ^ patch->old_mode) & S_IFMT)
3730 return error(_("%s: wrong type"), old_name);
3731 if (st_mode != patch->old_mode)
3732 warning(_("%s has type %o, expected %o"),
3733 old_name, st_mode, patch->old_mode);
3734 if (!patch->new_mode && !patch->is_delete)
3735 patch->new_mode = st_mode;
3736 return 0;
3737
3738 is_new:
3739 patch->is_new = 1;
3740 patch->is_delete = 0;
6a83d902 3741 FREE_AND_NULL(patch->old_name);
13b5af22
CC
3742 return 0;
3743}
3744
3745
3746#define EXISTS_IN_INDEX 1
3747#define EXISTS_IN_WORKTREE 2
3748
3749static int check_to_create(struct apply_state *state,
3750 const char *new_name,
3751 int ok_if_exists)
3752{
3753 struct stat nst;
3754
3755 if (state->check_index &&
3756 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3757 !ok_if_exists)
3758 return EXISTS_IN_INDEX;
3759 if (state->cached)
3760 return 0;
3761
3762 if (!lstat(new_name, &nst)) {
3763 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3764 return 0;
3765 /*
3766 * A leading component of new_name might be a symlink
3767 * that is going to be removed with this patch, but
3768 * still pointing at somewhere that has the path.
3769 * In such a case, path "new_name" does not exist as
3770 * far as git is concerned.
3771 */
3772 if (has_symlink_leading_path(new_name, strlen(new_name)))
3773 return 0;
3774
3775 return EXISTS_IN_WORKTREE;
c7054209 3776 } else if (!is_missing_file_error(errno)) {
90875eca 3777 return error_errno("%s", new_name);
13b5af22
CC
3778 }
3779 return 0;
3780}
3781
3782static uintptr_t register_symlink_changes(struct apply_state *state,
3783 const char *path,
3784 uintptr_t what)
3785{
3786 struct string_list_item *ent;
3787
3788 ent = string_list_lookup(&state->symlink_changes, path);
3789 if (!ent) {
3790 ent = string_list_insert(&state->symlink_changes, path);
3791 ent->util = (void *)0;
3792 }
3793 ent->util = (void *)(what | ((uintptr_t)ent->util));
3794 return (uintptr_t)ent->util;
3795}
3796
3797static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3798{
3799 struct string_list_item *ent;
3800
3801 ent = string_list_lookup(&state->symlink_changes, path);
3802 if (!ent)
3803 return 0;
3804 return (uintptr_t)ent->util;
3805}
3806
3807static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3808{
3809 for ( ; patch; patch = patch->next) {
3810 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3811 (patch->is_rename || patch->is_delete))
3812 /* the symlink at patch->old_name is removed */
3813 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3814
3815 if (patch->new_name && S_ISLNK(patch->new_mode))
3816 /* the symlink at patch->new_name is created or remains */
3817 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3818 }
3819}
3820
3821static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3822{
3823 do {
3824 unsigned int change;
3825
3826 while (--name->len && name->buf[name->len] != '/')
3827 ; /* scan backwards */
3828 if (!name->len)
3829 break;
3830 name->buf[name->len] = '\0';
3831 change = check_symlink_changes(state, name->buf);
3832 if (change & APPLY_SYMLINK_IN_RESULT)
3833 return 1;
3834 if (change & APPLY_SYMLINK_GOES_AWAY)
3835 /*
3836 * This cannot be "return 0", because we may
3837 * see a new one created at a higher level.
3838 */
3839 continue;
3840
3841 /* otherwise, check the preimage */
3842 if (state->check_index) {
3843 struct cache_entry *ce;
3844
3845 ce = cache_file_exists(name->buf, name->len, ignore_case);
3846 if (ce && S_ISLNK(ce->ce_mode))
3847 return 1;
3848 } else {
3849 struct stat st;
3850 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3851 return 1;
3852 }
3853 } while (1);
3854 return 0;
3855}
3856
3857static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3858{
3859 int ret;
3860 struct strbuf name = STRBUF_INIT;
3861
3862 assert(*name_ != '\0');
3863 strbuf_addstr(&name, name_);
3864 ret = path_is_beyond_symlink_1(state, &name);
3865 strbuf_release(&name);
3866
3867 return ret;
3868}
3869
3870static int check_unsafe_path(struct patch *patch)
3871{
3872 const char *old_name = NULL;
3873 const char *new_name = NULL;
3874 if (patch->is_delete)
3875 old_name = patch->old_name;
3876 else if (!patch->is_new && !patch->is_copy)
3877 old_name = patch->old_name;
3878 if (!patch->is_delete)
3879 new_name = patch->new_name;
3880
3881 if (old_name && !verify_path(old_name))
3882 return error(_("invalid path '%s'"), old_name);
3883 if (new_name && !verify_path(new_name))
3884 return error(_("invalid path '%s'"), new_name);
3885 return 0;
3886}
3887
3888/*
3889 * Check and apply the patch in-core; leave the result in patch->result
3890 * for the caller to write it out to the final destination.
3891 */
3892static int check_patch(struct apply_state *state, struct patch *patch)
3893{
3894 struct stat st;
3895 const char *old_name = patch->old_name;
3896 const char *new_name = patch->new_name;
3897 const char *name = old_name ? old_name : new_name;
3898 struct cache_entry *ce = NULL;
3899 struct patch *tpatch;
3900 int ok_if_exists;
3901 int status;
3902
3903 patch->rejected = 1; /* we will drop this after we succeed */
3904
3905 status = check_preimage(state, patch, &ce, &st);
3906 if (status)
3907 return status;
3908 old_name = patch->old_name;
3909
3910 /*
3911 * A type-change diff is always split into a patch to delete
3912 * old, immediately followed by a patch to create new (see
3913 * diff.c::run_diff()); in such a case it is Ok that the entry
3914 * to be deleted by the previous patch is still in the working
3915 * tree and in the index.
3916 *
3917 * A patch to swap-rename between A and B would first rename A
3918 * to B and then rename B to A. While applying the first one,
3919 * the presence of B should not stop A from getting renamed to
3920 * B; ask to_be_deleted() about the later rename. Removal of
3921 * B and rename from A to B is handled the same way by asking
3922 * was_deleted().
3923 */
3924 if ((tpatch = in_fn_table(state, new_name)) &&
3925 (was_deleted(tpatch) || to_be_deleted(tpatch)))
3926 ok_if_exists = 1;
3927 else
3928 ok_if_exists = 0;
3929
3930 if (new_name &&
3931 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {
3932 int err = check_to_create(state, new_name, ok_if_exists);
3933
3934 if (err && state->threeway) {
3935 patch->direct_to_threeway = 1;
3936 } else switch (err) {
3937 case 0:
3938 break; /* happy */
3939 case EXISTS_IN_INDEX:
3940 return error(_("%s: already exists in index"), new_name);
3941 break;
3942 case EXISTS_IN_WORKTREE:
3943 return error(_("%s: already exists in working directory"),
3944 new_name);
3945 default:
3946 return err;
3947 }
3948
3949 if (!patch->new_mode) {
3950 if (0 < patch->is_new)
3951 patch->new_mode = S_IFREG | 0644;
3952 else
3953 patch->new_mode = patch->old_mode;
3954 }
3955 }
3956
3957 if (new_name && old_name) {
3958 int same = !strcmp(old_name, new_name);
3959 if (!patch->new_mode)
3960 patch->new_mode = patch->old_mode;
3961 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {
3962 if (same)
3963 return error(_("new mode (%o) of %s does not "
3964 "match old mode (%o)"),
3965 patch->new_mode, new_name,
3966 patch->old_mode);
3967 else
3968 return error(_("new mode (%o) of %s does not "
3969 "match old mode (%o) of %s"),
3970 patch->new_mode, new_name,
3971 patch->old_mode, old_name);
3972 }
3973 }
3974
3975 if (!state->unsafe_paths && check_unsafe_path(patch))
3976 return -128;
3977
3978 /*
3979 * An attempt to read from or delete a path that is beyond a
3980 * symbolic link will be prevented by load_patch_target() that
3981 * is called at the beginning of apply_data() so we do not
3982 * have to worry about a patch marked with "is_delete" bit
3983 * here. We however need to make sure that the patch result
3984 * is not deposited to a path that is beyond a symbolic link
3985 * here.
3986 */
3987 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))
3988 return error(_("affected file '%s' is beyond a symbolic link"),
3989 patch->new_name);
3990
3991 if (apply_data(state, patch, &st, ce) < 0)
3992 return error(_("%s: patch does not apply"), name);
3993 patch->rejected = 0;
3994 return 0;
3995}
3996
3997static int check_patch_list(struct apply_state *state, struct patch *patch)
3998{
3999 int err = 0;
4000
4001 prepare_symlink_changes(state, patch);
4002 prepare_fn_table(state, patch);
4003 while (patch) {
4004 int res;
a46160d2 4005 if (state->apply_verbosity > verbosity_normal)
13b5af22
CC
4006 say_patch_name(stderr,
4007 _("Checking patch %s..."), patch);
4008 res = check_patch(state, patch);
4009 if (res == -128)
4010 return -128;
4011 err |= res;
4012 patch = patch->next;
4013 }
4014 return err;
4015}
4016
5b0b57fd
CC
4017static int read_apply_cache(struct apply_state *state)
4018{
4019 if (state->index_file)
4020 return read_cache_from(state->index_file);
4021 else
4022 return read_cache();
4023}
4024
4af9a7d3
JH
4025/* This function tries to read the object name from the current index */
4026static int get_current_oid(struct apply_state *state, const char *path,
4027 struct object_id *oid)
13b5af22
CC
4028{
4029 int pos;
4030
5b0b57fd 4031 if (read_apply_cache(state) < 0)
13b5af22
CC
4032 return -1;
4033 pos = cache_name_pos(path, strlen(path));
4034 if (pos < 0)
4035 return -1;
4af9a7d3 4036 oidcpy(oid, &active_cache[pos]->oid);
13b5af22
CC
4037 return 0;
4038}
4039
4af9a7d3 4040static int preimage_oid_in_gitlink_patch(struct patch *p, struct object_id *oid)
13b5af22
CC
4041{
4042 /*
4043 * A usable gitlink patch has only one fragment (hunk) that looks like:
4044 * @@ -1 +1 @@
4045 * -Subproject commit <old sha1>
4046 * +Subproject commit <new sha1>
4047 * or
4048 * @@ -1 +0,0 @@
4049 * -Subproject commit <old sha1>
4050 * for a removal patch.
4051 */
4052 struct fragment *hunk = p->fragments;
4053 static const char heading[] = "-Subproject commit ";
4054 char *preimage;
4055
4056 if (/* does the patch have only one hunk? */
4057 hunk && !hunk->next &&
4058 /* is its preimage one line? */
4059 hunk->oldpos == 1 && hunk->oldlines == 1 &&
4060 /* does preimage begin with the heading? */
4061 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&
4062 starts_with(++preimage, heading) &&
4063 /* does it record full SHA-1? */
4af9a7d3
JH
4064 !get_oid_hex(preimage + sizeof(heading) - 1, oid) &&
4065 preimage[sizeof(heading) + GIT_SHA1_HEXSZ - 1] == '\n' &&
13b5af22
CC
4066 /* does the abbreviated name on the index line agree with it? */
4067 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))
4068 return 0; /* it all looks fine */
4069
4070 /* we may have full object name on the index line */
4af9a7d3 4071 return get_oid_hex(p->old_sha1_prefix, oid);
13b5af22
CC
4072}
4073
4074/* Build an index that contains the just the files needed for a 3way merge */
b4290342 4075static int build_fake_ancestor(struct apply_state *state, struct patch *list)
13b5af22
CC
4076{
4077 struct patch *patch;
4078 struct index_state result = { NULL };
4079 static struct lock_file lock;
4080 int res;
4081
4082 /* Once we start supporting the reverse patch, it may be
4083 * worth showing the new sha1 prefix, but until then...
4084 */
4085 for (patch = list; patch; patch = patch->next) {
4af9a7d3 4086 struct object_id oid;
13b5af22
CC
4087 struct cache_entry *ce;
4088 const char *name;
4089
4090 name = patch->old_name ? patch->old_name : patch->new_name;
4091 if (0 < patch->is_new)
4092 continue;
4093
4094 if (S_ISGITLINK(patch->old_mode)) {
4af9a7d3 4095 if (!preimage_oid_in_gitlink_patch(patch, &oid))
13b5af22
CC
4096 ; /* ok, the textual part looks sane */
4097 else
d1d42bf5
VA
4098 return error(_("sha1 information is lacking or "
4099 "useless for submodule %s"), name);
e82caf38 4100 } else if (!get_oid_blob(patch->old_sha1_prefix, &oid)) {
13b5af22
CC
4101 ; /* ok */
4102 } else if (!patch->lines_added && !patch->lines_deleted) {
4103 /* mode-only change: update the current */
4af9a7d3 4104 if (get_current_oid(state, patch->old_name, &oid))
d1d42bf5
VA
4105 return error(_("mode change for %s, which is not "
4106 "in current HEAD"), name);
13b5af22 4107 } else
d1d42bf5
VA
4108 return error(_("sha1 information is lacking or useless "
4109 "(%s)."), name);
13b5af22 4110
4af9a7d3 4111 ce = make_cache_entry(patch->old_mode, oid.hash, name, 0, 0);
13b5af22
CC
4112 if (!ce)
4113 return error(_("make_cache_entry failed for path '%s'"),
4114 name);
4115 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD)) {
4116 free(ce);
d1d42bf5 4117 return error(_("could not add %s to temporary index"),
13b5af22
CC
4118 name);
4119 }
4120 }
4121
b4290342 4122 hold_lock_file_for_update(&lock, state->fake_ancestor, LOCK_DIE_ON_ERROR);
13b5af22
CC
4123 res = write_locked_index(&result, &lock, COMMIT_LOCK);
4124 discard_index(&result);
4125
e294e895
JH
4126 if (res)
4127 return error(_("could not write temporary index to %s"),
4128 state->fake_ancestor);
13b5af22 4129
e294e895
JH
4130 return 0;
4131}
13b5af22 4132
e294e895
JH
4133static void stat_patch_list(struct apply_state *state, struct patch *patch)
4134{
4135 int files, adds, dels;
13b5af22 4136
e294e895
JH
4137 for (files = adds = dels = 0 ; patch ; patch = patch->next) {
4138 files++;
4139 adds += patch->lines_added;
4140 dels += patch->lines_deleted;
4141 show_stats(state, patch);
4142 }
13b5af22 4143
e294e895
JH
4144 print_stat_summary(stdout, files, adds, dels);
4145}
13b5af22 4146
e294e895
JH
4147static void numstat_patch_list(struct apply_state *state,
4148 struct patch *patch)
4149{
4150 for ( ; patch; patch = patch->next) {
4151 const char *name;
4152 name = patch->new_name ? patch->new_name : patch->old_name;
4153 if (patch->is_binary)
4154 printf("-\t-\t");
4155 else
4156 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
4157 write_name_quoted(name, stdout, state->line_termination);
4158 }
4159}
4160
4161static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
4162{
4163 if (mode)
4164 printf(" %s mode %06o %s\n", newdelete, mode, name);
4165 else
4166 printf(" %s %s\n", newdelete, name);
4167}
4168
4169static void show_mode_change(struct patch *p, int show_name)
4170{
4171 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
4172 if (show_name)
4173 printf(" mode change %06o => %06o %s\n",
4174 p->old_mode, p->new_mode, p->new_name);
4175 else
4176 printf(" mode change %06o => %06o\n",
4177 p->old_mode, p->new_mode);
4178 }
4179}
4180
4181static void show_rename_copy(struct patch *p)
4182{
4183 const char *renamecopy = p->is_rename ? "rename" : "copy";
4184 const char *old, *new;
4185
4186 /* Find common prefix */
4187 old = p->old_name;
4188 new = p->new_name;
4189 while (1) {
4190 const char *slash_old, *slash_new;
4191 slash_old = strchr(old, '/');
4192 slash_new = strchr(new, '/');
4193 if (!slash_old ||
4194 !slash_new ||
4195 slash_old - old != slash_new - new ||
4196 memcmp(old, new, slash_new - new))
4197 break;
4198 old = slash_old + 1;
4199 new = slash_new + 1;
4200 }
4201 /* p->old_name thru old is the common prefix, and old and new
4202 * through the end of names are renames
4203 */
4204 if (old != p->old_name)
4205 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
4206 (int)(old - p->old_name), p->old_name,
4207 old, new, p->score);
4208 else
4209 printf(" %s %s => %s (%d%%)\n", renamecopy,
4210 p->old_name, p->new_name, p->score);
4211 show_mode_change(p, 0);
4212}
4213
4214static void summary_patch_list(struct patch *patch)
4215{
4216 struct patch *p;
4217
4218 for (p = patch; p; p = p->next) {
4219 if (p->is_new)
4220 show_file_mode_name("create", p->new_mode, p->new_name);
4221 else if (p->is_delete)
4222 show_file_mode_name("delete", p->old_mode, p->old_name);
4223 else {
4224 if (p->is_rename || p->is_copy)
4225 show_rename_copy(p);
4226 else {
4227 if (p->score) {
4228 printf(" rewrite %s (%d%%)\n",
4229 p->new_name, p->score);
4230 show_mode_change(p, 0);
4231 }
4232 else
4233 show_mode_change(p, 1);
4234 }
4235 }
4236 }
4237}
4238
4239static void patch_stats(struct apply_state *state, struct patch *patch)
4240{
4241 int lines = patch->lines_added + patch->lines_deleted;
4242
4243 if (lines > state->max_change)
4244 state->max_change = lines;
4245 if (patch->old_name) {
4246 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
4247 if (!len)
4248 len = strlen(patch->old_name);
4249 if (len > state->max_len)
4250 state->max_len = len;
4251 }
4252 if (patch->new_name) {
4253 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
4254 if (!len)
4255 len = strlen(patch->new_name);
4256 if (len > state->max_len)
4257 state->max_len = len;
4258 }
4259}
4260
4261static int remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)
4262{
4263 if (state->update_index) {
4264 if (remove_file_from_cache(patch->old_name) < 0)
4265 return error(_("unable to remove %s from index"), patch->old_name);
4266 }
4267 if (!state->cached) {
4268 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {
4269 remove_path(patch->old_name);
4270 }
4271 }
4272 return 0;
4273}
4274
4275static int add_index_file(struct apply_state *state,
4276 const char *path,
4277 unsigned mode,
4278 void *buf,
4279 unsigned long size)
4280{
4281 struct stat st;
4282 struct cache_entry *ce;
4283 int namelen = strlen(path);
4284 unsigned ce_size = cache_entry_size(namelen);
4285
4286 if (!state->update_index)
4287 return 0;
4288
4289 ce = xcalloc(1, ce_size);
4290 memcpy(ce->name, path, namelen);
4291 ce->ce_mode = create_ce_mode(mode);
4292 ce->ce_flags = create_ce_flags(0);
4293 ce->ce_namelen = namelen;
4294 if (S_ISGITLINK(mode)) {
4295 const char *s;
4296
4297 if (!skip_prefix(buf, "Subproject commit ", &s) ||
4298 get_oid_hex(s, &ce->oid)) {
13b5af22 4299 free(ce);
e294e895 4300 return error(_("corrupt patch for submodule %s"), path);
13b5af22
CC
4301 }
4302 } else {
4303 if (!state->cached) {
4304 if (lstat(path, &st) < 0) {
4305 free(ce);
90875eca
CC
4306 return error_errno(_("unable to stat newly "
4307 "created file '%s'"),
4308 path);
13b5af22
CC
4309 }
4310 fill_stat_cache_info(ce, &st);
4311 }
4af9a7d3 4312 if (write_sha1_file(buf, size, blob_type, ce->oid.hash) < 0) {
13b5af22
CC
4313 free(ce);
4314 return error(_("unable to create backing store "
4315 "for newly created file %s"), path);
4316 }
4317 }
4318 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4319 free(ce);
4320 return error(_("unable to add cache entry for %s"), path);
4321 }
4322
4323 return 0;
4324}
4325
4326/*
4327 * Returns:
4328 * -1 if an unrecoverable error happened
4329 * 0 if everything went well
4330 * 1 if a recoverable error happened
4331 */
4332static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
4333{
4334 int fd, res;
4335 struct strbuf nbuf = STRBUF_INIT;
4336
4337 if (S_ISGITLINK(mode)) {
4338 struct stat st;
4339 if (!lstat(path, &st) && S_ISDIR(st.st_mode))
4340 return 0;
4341 return !!mkdir(path, 0777);
4342 }
4343
4344 if (has_symlinks && S_ISLNK(mode))
4345 /* Although buf:size is counted string, it also is NUL
4346 * terminated.
4347 */
4348 return !!symlink(buf, path);
4349
4350 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
4351 if (fd < 0)
4352 return 1;
4353
4354 if (convert_to_working_tree(path, buf, size, &nbuf)) {
4355 size = nbuf.len;
4356 buf = nbuf.buf;
4357 }
4358
4359 res = write_in_full(fd, buf, size) < 0;
4360 if (res)
4361 error_errno(_("failed to write to '%s'"), path);
4362 strbuf_release(&nbuf);
4363
4364 if (close(fd) < 0 && !res)
4365 return error_errno(_("closing file '%s'"), path);
4366
4367 return res ? -1 : 0;
4368}
4369
4370/*
4371 * We optimistically assume that the directories exist,
4372 * which is true 99% of the time anyway. If they don't,
4373 * we create them and try again.
4374 *
4375 * Returns:
4376 * -1 on error
4377 * 0 otherwise
4378 */
4379static int create_one_file(struct apply_state *state,
4380 char *path,
4381 unsigned mode,
4382 const char *buf,
4383 unsigned long size)
4384{
4385 int res;
4386
4387 if (state->cached)
4388 return 0;
4389
4390 res = try_create_file(path, mode, buf, size);
4391 if (res < 0)
4392 return -1;
4393 if (!res)
4394 return 0;
4395
4396 if (errno == ENOENT) {
4397 if (safe_create_leading_directories(path))
4398 return 0;
4399 res = try_create_file(path, mode, buf, size);
4400 if (res < 0)
4401 return -1;
4402 if (!res)
4403 return 0;
4404 }
4405
4406 if (errno == EEXIST || errno == EACCES) {
4407 /* We may be trying to create a file where a directory
4408 * used to be.
4409 */
4410 struct stat st;
4411 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))
4412 errno = EEXIST;
4413 }
4414
4415 if (errno == EEXIST) {
4416 unsigned int nr = getpid();
4417
4418 for (;;) {
4419 char newpath[PATH_MAX];
4420 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);
4421 res = try_create_file(newpath, mode, buf, size);
4422 if (res < 0)
4423 return -1;
4424 if (!res) {
4425 if (!rename(newpath, path))
4426 return 0;
4427 unlink_or_warn(newpath);
4428 break;
4429 }
4430 if (errno != EEXIST)
4431 break;
4432 ++nr;
4433 }
4434 }
4435 return error_errno(_("unable to write file '%s' mode %o"),
4436 path, mode);
4437}
4438
4439static int add_conflicted_stages_file(struct apply_state *state,
4440 struct patch *patch)
4441{
4442 int stage, namelen;
4443 unsigned ce_size, mode;
4444 struct cache_entry *ce;
4445
4446 if (!state->update_index)
4447 return 0;
4448 namelen = strlen(patch->new_name);
4449 ce_size = cache_entry_size(namelen);
4450 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);
4451
4452 remove_file_from_cache(patch->new_name);
4453 for (stage = 1; stage < 4; stage++) {
4454 if (is_null_oid(&patch->threeway_stage[stage - 1]))
4455 continue;
4456 ce = xcalloc(1, ce_size);
4457 memcpy(ce->name, patch->new_name, namelen);
4458 ce->ce_mode = create_ce_mode(mode);
4459 ce->ce_flags = create_ce_flags(stage);
4460 ce->ce_namelen = namelen;
4af9a7d3 4461 oidcpy(&ce->oid, &patch->threeway_stage[stage - 1]);
13b5af22
CC
4462 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0) {
4463 free(ce);
4464 return error(_("unable to add cache entry for %s"),
4465 patch->new_name);
4466 }
4467 }
4468
4469 return 0;
4470}
4471
4472static int create_file(struct apply_state *state, struct patch *patch)
4473{
4474 char *path = patch->new_name;
4475 unsigned mode = patch->new_mode;
4476 unsigned long size = patch->resultsize;
4477 char *buf = patch->result;
4478
4479 if (!mode)
4480 mode = S_IFREG | 0644;
4481 if (create_one_file(state, path, mode, buf, size))
4482 return -1;
4483
4484 if (patch->conflicted_threeway)
4485 return add_conflicted_stages_file(state, patch);
4486 else
4487 return add_index_file(state, path, mode, buf, size);
4488}
4489
4490/* phase zero is to remove, phase one is to create */
4491static int write_out_one_result(struct apply_state *state,
4492 struct patch *patch,
4493 int phase)
4494{
4495 if (patch->is_delete > 0) {
4496 if (phase == 0)
4497 return remove_file(state, patch, 1);
4498 return 0;
4499 }
4500 if (patch->is_new > 0 || patch->is_copy) {
4501 if (phase == 1)
4502 return create_file(state, patch);
4503 return 0;
4504 }
4505 /*
4506 * Rename or modification boils down to the same
4507 * thing: remove the old, write the new
4508 */
4509 if (phase == 0)
4510 return remove_file(state, patch, patch->is_rename);
4511 if (phase == 1)
4512 return create_file(state, patch);
4513 return 0;
4514}
4515
4516static int write_out_one_reject(struct apply_state *state, struct patch *patch)
4517{
4518 FILE *rej;
4519 char namebuf[PATH_MAX];
4520 struct fragment *frag;
4521 int cnt = 0;
4522 struct strbuf sb = STRBUF_INIT;
4523
4524 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
4525 if (!frag->rejected)
4526 continue;
4527 cnt++;
4528 }
4529
4530 if (!cnt) {
a46160d2 4531 if (state->apply_verbosity > verbosity_normal)
13b5af22
CC
4532 say_patch_name(stderr,
4533 _("Applied patch %s cleanly."), patch);
4534 return 0;
4535 }
4536
4537 /* This should not happen, because a removal patch that leaves
4538 * contents are marked "rejected" at the patch level.
4539 */
4540 if (!patch->new_name)
4541 die(_("internal error"));
4542
4543 /* Say this even without --verbose */
4544 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",
4545 "Applying patch %%s with %d rejects...",
4546 cnt),
4547 cnt);
a46160d2
CC
4548 if (state->apply_verbosity > verbosity_silent)
4549 say_patch_name(stderr, sb.buf, patch);
13b5af22
CC
4550 strbuf_release(&sb);
4551
4552 cnt = strlen(patch->new_name);
4553 if (ARRAY_SIZE(namebuf) <= cnt + 5) {
4554 cnt = ARRAY_SIZE(namebuf) - 5;
4555 warning(_("truncating .rej filename to %.*s.rej"),
4556 cnt - 1, patch->new_name);
4557 }
4558 memcpy(namebuf, patch->new_name, cnt);
4559 memcpy(namebuf + cnt, ".rej", 5);
4560
4561 rej = fopen(namebuf, "w");
4562 if (!rej)
90875eca 4563 return error_errno(_("cannot open %s"), namebuf);
13b5af22
CC
4564
4565 /* Normal git tools never deal with .rej, so do not pretend
4566 * this is a git patch by saying --git or giving extended
4567 * headers. While at it, maybe please "kompare" that wants
4568 * the trailing TAB and some garbage at the end of line ;-).
4569 */
4570 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
4571 patch->new_name, patch->new_name);
4572 for (cnt = 1, frag = patch->fragments;
4573 frag;
4574 cnt++, frag = frag->next) {
4575 if (!frag->rejected) {
a46160d2
CC
4576 if (state->apply_verbosity > verbosity_silent)
4577 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);
13b5af22
CC
4578 continue;
4579 }
a46160d2
CC
4580 if (state->apply_verbosity > verbosity_silent)
4581 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);
13b5af22
CC
4582 fprintf(rej, "%.*s", frag->size, frag->patch);
4583 if (frag->patch[frag->size-1] != '\n')
4584 fputc('\n', rej);
4585 }
4586 fclose(rej);
4587 return -1;
4588}
4589
4590/*
4591 * Returns:
4592 * -1 if an error happened
4593 * 0 if the patch applied cleanly
4594 * 1 if the patch did not apply cleanly
4595 */
4596static int write_out_results(struct apply_state *state, struct patch *list)
4597{
4598 int phase;
4599 int errs = 0;
4600 struct patch *l;
4601 struct string_list cpath = STRING_LIST_INIT_DUP;
4602
4603 for (phase = 0; phase < 2; phase++) {
4604 l = list;
4605 while (l) {
4606 if (l->rejected)
4607 errs = 1;
4608 else {
4609 if (write_out_one_result(state, l, phase)) {
4610 string_list_clear(&cpath, 0);
4611 return -1;
4612 }
4613 if (phase == 1) {
4614 if (write_out_one_reject(state, l))
4615 errs = 1;
4616 if (l->conflicted_threeway) {
4617 string_list_append(&cpath, l->new_name);
4618 errs = 1;
4619 }
4620 }
4621 }
4622 l = l->next;
4623 }
4624 }
4625
4626 if (cpath.nr) {
4627 struct string_list_item *item;
4628
4629 string_list_sort(&cpath);
a46160d2
CC
4630 if (state->apply_verbosity > verbosity_silent) {
4631 for_each_string_list_item(item, &cpath)
4632 fprintf(stderr, "U %s\n", item->string);
4633 }
13b5af22
CC
4634 string_list_clear(&cpath, 0);
4635
4636 rerere(0);
4637 }
4638
4639 return errs;
4640}
4641
4642/*
4643 * Try to apply a patch.
4644 *
4645 * Returns:
4646 * -128 if a bad error happened (like patch unreadable)
4647 * -1 if patch did not apply and user cannot deal with it
4648 * 0 if the patch applied
4649 * 1 if the patch did not apply but user might fix it
4650 */
4651static int apply_patch(struct apply_state *state,
4652 int fd,
4653 const char *filename,
4654 int options)
4655{
4656 size_t offset;
4657 struct strbuf buf = STRBUF_INIT; /* owns the patch text */
4658 struct patch *list = NULL, **listp = &list;
4659 int skipped_patch = 0;
4660 int res = 0;
4661
4662 state->patch_input_file = filename;
4663 if (read_patch_file(&buf, fd) < 0)
4664 return -128;
4665 offset = 0;
4666 while (offset < buf.len) {
4667 struct patch *patch;
4668 int nr;
4669
4670 patch = xcalloc(1, sizeof(*patch));
4671 patch->inaccurate_eof = !!(options & APPLY_OPT_INACCURATE_EOF);
4672 patch->recount = !!(options & APPLY_OPT_RECOUNT);
4673 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);
4674 if (nr < 0) {
4675 free_patch(patch);
4676 if (nr == -128) {
4677 res = -128;
4678 goto end;
4679 }
4680 break;
4681 }
4682 if (state->apply_in_reverse)
4683 reverse_patches(patch);
4684 if (use_patch(state, patch)) {
4685 patch_stats(state, patch);
4686 *listp = patch;
4687 listp = &patch->next;
4688 }
4689 else {
a46160d2 4690 if (state->apply_verbosity > verbosity_normal)
13b5af22
CC
4691 say_patch_name(stderr, _("Skipped patch '%s'."), patch);
4692 free_patch(patch);
4693 skipped_patch++;
4694 }
4695 offset += nr;
4696 }
4697
4698 if (!list && !skipped_patch) {
4699 error(_("unrecognized input"));
4700 res = -128;
4701 goto end;
4702 }
4703
4704 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))
4705 state->apply = 0;
4706
4707 state->update_index = state->check_index && state->apply;
5b0b57fd
CC
4708 if (state->update_index && state->newfd < 0) {
4709 if (state->index_file)
6d058c88 4710 state->newfd = hold_lock_file_for_update(&state->lock_file,
5b0b57fd
CC
4711 state->index_file,
4712 LOCK_DIE_ON_ERROR);
4713 else
6d058c88 4714 state->newfd = hold_locked_index(&state->lock_file, LOCK_DIE_ON_ERROR);
5b0b57fd 4715 }
13b5af22 4716
5b0b57fd 4717 if (state->check_index && read_apply_cache(state) < 0) {
13b5af22
CC
4718 error(_("unable to read index file"));
4719 res = -128;
4720 goto end;
4721 }
4722
4723 if (state->check || state->apply) {
4724 int r = check_patch_list(state, list);
4725 if (r == -128) {
4726 res = -128;
4727 goto end;
4728 }
4729 if (r < 0 && !state->apply_with_reject) {
4730 res = -1;
4731 goto end;
4732 }
4733 }
4734
4735 if (state->apply) {
4736 int write_res = write_out_results(state, list);
4737 if (write_res < 0) {
4738 res = -128;
4739 goto end;
4740 }
4741 if (write_res > 0) {
4742 /* with --3way, we still need to write the index out */
4743 res = state->apply_with_reject ? -1 : 1;
4744 goto end;
4745 }
4746 }
4747
4748 if (state->fake_ancestor &&
b4290342 4749 build_fake_ancestor(state, list)) {
13b5af22
CC
4750 res = -128;
4751 goto end;
4752 }
4753
487beee0 4754 if (state->diffstat && state->apply_verbosity > verbosity_silent)
13b5af22
CC
4755 stat_patch_list(state, list);
4756
487beee0 4757 if (state->numstat && state->apply_verbosity > verbosity_silent)
13b5af22
CC
4758 numstat_patch_list(state, list);
4759
487beee0 4760 if (state->summary && state->apply_verbosity > verbosity_silent)
13b5af22
CC
4761 summary_patch_list(list);
4762
4763end:
4764 free_patch_list(list);
4765 strbuf_release(&buf);
4766 string_list_clear(&state->fn_table, 0);
4767 return res;
4768}
4769
7e1bad24
CC
4770static int apply_option_parse_exclude(const struct option *opt,
4771 const char *arg, int unset)
13b5af22
CC
4772{
4773 struct apply_state *state = opt->value;
4774 add_name_limit(state, arg, 1);
4775 return 0;
4776}
4777
7e1bad24
CC
4778static int apply_option_parse_include(const struct option *opt,
4779 const char *arg, int unset)
13b5af22
CC
4780{
4781 struct apply_state *state = opt->value;
4782 add_name_limit(state, arg, 0);
4783 state->has_include = 1;
4784 return 0;
4785}
4786
7e1bad24
CC
4787static int apply_option_parse_p(const struct option *opt,
4788 const char *arg,
4789 int unset)
13b5af22
CC
4790{
4791 struct apply_state *state = opt->value;
4792 state->p_value = atoi(arg);
4793 state->p_value_known = 1;
4794 return 0;
4795}
4796
7e1bad24
CC
4797static int apply_option_parse_space_change(const struct option *opt,
4798 const char *arg, int unset)
13b5af22
CC
4799{
4800 struct apply_state *state = opt->value;
4801 if (unset)
4802 state->ws_ignore_action = ignore_ws_none;
4803 else
4804 state->ws_ignore_action = ignore_ws_change;
4805 return 0;
4806}
4807
7e1bad24
CC
4808static int apply_option_parse_whitespace(const struct option *opt,
4809 const char *arg, int unset)
13b5af22
CC
4810{
4811 struct apply_state *state = opt->value;
4812 state->whitespace_option = arg;
4813 if (parse_whitespace_option(state, arg))
4814 exit(1);
4815 return 0;
4816}
4817
7e1bad24
CC
4818static int apply_option_parse_directory(const struct option *opt,
4819 const char *arg, int unset)
13b5af22
CC
4820{
4821 struct apply_state *state = opt->value;
4822 strbuf_reset(&state->root);
4823 strbuf_addstr(&state->root, arg);
4824 strbuf_complete(&state->root, '/');
4825 return 0;
4826}
4827
4828int apply_all_patches(struct apply_state *state,
4829 int argc,
4830 const char **argv,
4831 int options)
4832{
4833 int i;
4834 int res;
4835 int errs = 0;
4836 int read_stdin = 1;
4837
4838 for (i = 0; i < argc; i++) {
4839 const char *arg = argv[i];
e4da43b1 4840 char *to_free = NULL;
13b5af22
CC
4841 int fd;
4842
4843 if (!strcmp(arg, "-")) {
4844 res = apply_patch(state, 0, "<stdin>", options);
4845 if (res < 0)
4846 goto end;
4847 errs |= res;
4848 read_stdin = 0;
4849 continue;
e4da43b1
JK
4850 } else
4851 arg = to_free = prefix_filename(state->prefix, arg);
13b5af22
CC
4852
4853 fd = open(arg, O_RDONLY);
4854 if (fd < 0) {
4855 error(_("can't open patch '%s': %s"), arg, strerror(errno));
4856 res = -128;
e4da43b1 4857 free(to_free);
13b5af22
CC
4858 goto end;
4859 }
4860 read_stdin = 0;
4861 set_default_whitespace_mode(state);
4862 res = apply_patch(state, fd, arg, options);
4863 close(fd);
e4da43b1 4864 free(to_free);
13b5af22
CC
4865 if (res < 0)
4866 goto end;
4867 errs |= res;
4868 }
4869 set_default_whitespace_mode(state);
4870 if (read_stdin) {
4871 res = apply_patch(state, 0, "<stdin>", options);
4872 if (res < 0)
4873 goto end;
4874 errs |= res;
4875 }
4876
4877 if (state->whitespace_error) {
4878 if (state->squelch_whitespace_errors &&
4879 state->squelch_whitespace_errors < state->whitespace_error) {
4880 int squelched =
4881 state->whitespace_error - state->squelch_whitespace_errors;
4882 warning(Q_("squelched %d whitespace error",
4883 "squelched %d whitespace errors",
4884 squelched),
4885 squelched);
4886 }
4887 if (state->ws_error_action == die_on_ws_error) {
4888 error(Q_("%d line adds whitespace errors.",
4889 "%d lines add whitespace errors.",
4890 state->whitespace_error),
4891 state->whitespace_error);
4892 res = -128;
4893 goto end;
4894 }
4895 if (state->applied_after_fixing_ws && state->apply)
965d5c85
VA
4896 warning(Q_("%d line applied after"
4897 " fixing whitespace errors.",
4898 "%d lines applied after"
4899 " fixing whitespace errors.",
4900 state->applied_after_fixing_ws),
4901 state->applied_after_fixing_ws);
13b5af22
CC
4902 else if (state->whitespace_error)
4903 warning(Q_("%d line adds whitespace errors.",
4904 "%d lines add whitespace errors.",
4905 state->whitespace_error),
4906 state->whitespace_error);
4907 }
4908
4909 if (state->update_index) {
6d058c88 4910 res = write_locked_index(&the_index, &state->lock_file, COMMIT_LOCK);
13b5af22
CC
4911 if (res) {
4912 error(_("Unable to write new index file"));
4913 res = -128;
4914 goto end;
4915 }
4916 state->newfd = -1;
4917 }
4918
45b78d8b 4919 res = !!errs;
13b5af22
CC
4920
4921end:
4922 if (state->newfd >= 0) {
6d058c88 4923 rollback_lock_file(&state->lock_file);
13b5af22
CC
4924 state->newfd = -1;
4925 }
4926
45b78d8b
CC
4927 if (state->apply_verbosity <= verbosity_silent) {
4928 set_error_routine(state->saved_error_routine);
4929 set_warn_routine(state->saved_warn_routine);
4930 }
4931
4932 if (res > -1)
4933 return res;
13b5af22
CC
4934 return (res == -1 ? 1 : 128);
4935}
7e1bad24
CC
4936
4937int apply_parse_options(int argc, const char **argv,
4938 struct apply_state *state,
4939 int *force_apply, int *options,
4940 const char * const *apply_usage)
4941{
4942 struct option builtin_apply_options[] = {
4943 { OPTION_CALLBACK, 0, "exclude", state, N_("path"),
4944 N_("don't apply changes matching the given path"),
4945 0, apply_option_parse_exclude },
4946 { OPTION_CALLBACK, 0, "include", state, N_("path"),
4947 N_("apply changes matching the given path"),
4948 0, apply_option_parse_include },
4949 { OPTION_CALLBACK, 'p', NULL, state, N_("num"),
4950 N_("remove <num> leading slashes from traditional diff paths"),
4951 0, apply_option_parse_p },
4952 OPT_BOOL(0, "no-add", &state->no_add,
4953 N_("ignore additions made by the patch")),
4954 OPT_BOOL(0, "stat", &state->diffstat,
4955 N_("instead of applying the patch, output diffstat for the input")),
4956 OPT_NOOP_NOARG(0, "allow-binary-replacement"),
4957 OPT_NOOP_NOARG(0, "binary"),
4958 OPT_BOOL(0, "numstat", &state->numstat,
4959 N_("show number of added and deleted lines in decimal notation")),
4960 OPT_BOOL(0, "summary", &state->summary,
4961 N_("instead of applying the patch, output a summary for the input")),
4962 OPT_BOOL(0, "check", &state->check,
4963 N_("instead of applying the patch, see if the patch is applicable")),
4964 OPT_BOOL(0, "index", &state->check_index,
4965 N_("make sure the patch is applicable to the current index")),
4966 OPT_BOOL(0, "cached", &state->cached,
4967 N_("apply a patch without touching the working tree")),
4968 OPT_BOOL(0, "unsafe-paths", &state->unsafe_paths,
4969 N_("accept a patch that touches outside the working area")),
4970 OPT_BOOL(0, "apply", force_apply,
4971 N_("also apply the patch (use with --stat/--summary/--check)")),
4972 OPT_BOOL('3', "3way", &state->threeway,
4973 N_( "attempt three-way merge if a patch does not apply")),
4974 OPT_FILENAME(0, "build-fake-ancestor", &state->fake_ancestor,
4975 N_("build a temporary index based on embedded index information")),
4976 /* Think twice before adding "--nul" synonym to this */
4977 OPT_SET_INT('z', NULL, &state->line_termination,
4978 N_("paths are separated with NUL character"), '\0'),
4979 OPT_INTEGER('C', NULL, &state->p_context,
4980 N_("ensure at least <n> lines of context match")),
4981 { OPTION_CALLBACK, 0, "whitespace", state, N_("action"),
4982 N_("detect new or modified lines that have whitespace errors"),
4983 0, apply_option_parse_whitespace },
4984 { OPTION_CALLBACK, 0, "ignore-space-change", state, NULL,
4985 N_("ignore changes in whitespace when finding context"),
4986 PARSE_OPT_NOARG, apply_option_parse_space_change },
4987 { OPTION_CALLBACK, 0, "ignore-whitespace", state, NULL,
4988 N_("ignore changes in whitespace when finding context"),
4989 PARSE_OPT_NOARG, apply_option_parse_space_change },
4990 OPT_BOOL('R', "reverse", &state->apply_in_reverse,
4991 N_("apply the patch in reverse")),
4992 OPT_BOOL(0, "unidiff-zero", &state->unidiff_zero,
4993 N_("don't expect at least one line of context")),
4994 OPT_BOOL(0, "reject", &state->apply_with_reject,
4995 N_("leave the rejected hunks in corresponding *.rej files")),
4996 OPT_BOOL(0, "allow-overlap", &state->allow_overlap,
4997 N_("allow overlapping hunks")),
4998 OPT__VERBOSE(&state->apply_verbosity, N_("be verbose")),
4999 OPT_BIT(0, "inaccurate-eof", options,
5000 N_("tolerate incorrectly detected missing new-line at the end of file"),
5001 APPLY_OPT_INACCURATE_EOF),
5002 OPT_BIT(0, "recount", options,
5003 N_("do not trust the line counts in the hunk headers"),
5004 APPLY_OPT_RECOUNT),
5005 { OPTION_CALLBACK, 0, "directory", state, N_("root"),
5006 N_("prepend <root> to all filenames"),
5007 0, apply_option_parse_directory },
5008 OPT_END()
5009 };
5010
5011 return parse_options(argc, argv, state->prefix, builtin_apply_options, apply_usage, 0);
5012}