apply: rename and move opt constants to apply.h
[git/git.git] / builtin / apply.c
1 /*
2 * apply.c
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This applies patches on top of some (arbitrary) version of the SCM.
7 *
8 */
9 #include "cache.h"
10 #include "lockfile.h"
11 #include "cache-tree.h"
12 #include "quote.h"
13 #include "blob.h"
14 #include "delta.h"
15 #include "builtin.h"
16 #include "string-list.h"
17 #include "dir.h"
18 #include "diff.h"
19 #include "parse-options.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "rerere.h"
23 #include "apply.h"
24
25 static const char * const apply_usage[] = {
26 N_("git apply [<options>] [<patch>...]"),
27 NULL
28 };
29
30 static void set_default_whitespace_mode(struct apply_state *state)
31 {
32 if (!state->whitespace_option && !apply_default_whitespace)
33 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error);
34 }
35
36 /*
37 * This represents one "hunk" from a patch, starting with
38 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The
39 * patch text is pointed at by patch, and its byte length
40 * is stored in size. leading and trailing are the number
41 * of context lines.
42 */
43 struct fragment {
44 unsigned long leading, trailing;
45 unsigned long oldpos, oldlines;
46 unsigned long newpos, newlines;
47 /*
48 * 'patch' is usually borrowed from buf in apply_patch(),
49 * but some codepaths store an allocated buffer.
50 */
51 const char *patch;
52 unsigned free_patch:1,
53 rejected:1;
54 int size;
55 int linenr;
56 struct fragment *next;
57 };
58
59 /*
60 * When dealing with a binary patch, we reuse "leading" field
61 * to store the type of the binary hunk, either deflated "delta"
62 * or deflated "literal".
63 */
64 #define binary_patch_method leading
65 #define BINARY_DELTA_DEFLATED 1
66 #define BINARY_LITERAL_DEFLATED 2
67
68 /*
69 * This represents a "patch" to a file, both metainfo changes
70 * such as creation/deletion, filemode and content changes represented
71 * as a series of fragments.
72 */
73 struct patch {
74 char *new_name, *old_name, *def_name;
75 unsigned int old_mode, new_mode;
76 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */
77 int rejected;
78 unsigned ws_rule;
79 int lines_added, lines_deleted;
80 int score;
81 unsigned int is_toplevel_relative:1;
82 unsigned int inaccurate_eof:1;
83 unsigned int is_binary:1;
84 unsigned int is_copy:1;
85 unsigned int is_rename:1;
86 unsigned int recount:1;
87 unsigned int conflicted_threeway:1;
88 unsigned int direct_to_threeway:1;
89 struct fragment *fragments;
90 char *result;
91 size_t resultsize;
92 char old_sha1_prefix[41];
93 char new_sha1_prefix[41];
94 struct patch *next;
95
96 /* three-way fallback result */
97 struct object_id threeway_stage[3];
98 };
99
100 static void free_fragment_list(struct fragment *list)
101 {
102 while (list) {
103 struct fragment *next = list->next;
104 if (list->free_patch)
105 free((char *)list->patch);
106 free(list);
107 list = next;
108 }
109 }
110
111 static void free_patch(struct patch *patch)
112 {
113 free_fragment_list(patch->fragments);
114 free(patch->def_name);
115 free(patch->old_name);
116 free(patch->new_name);
117 free(patch->result);
118 free(patch);
119 }
120
121 static void free_patch_list(struct patch *list)
122 {
123 while (list) {
124 struct patch *next = list->next;
125 free_patch(list);
126 list = next;
127 }
128 }
129
130 /*
131 * A line in a file, len-bytes long (includes the terminating LF,
132 * except for an incomplete line at the end if the file ends with
133 * one), and its contents hashes to 'hash'.
134 */
135 struct line {
136 size_t len;
137 unsigned hash : 24;
138 unsigned flag : 8;
139 #define LINE_COMMON 1
140 #define LINE_PATCHED 2
141 };
142
143 /*
144 * This represents a "file", which is an array of "lines".
145 */
146 struct image {
147 char *buf;
148 size_t len;
149 size_t nr;
150 size_t alloc;
151 struct line *line_allocated;
152 struct line *line;
153 };
154
155 static uint32_t hash_line(const char *cp, size_t len)
156 {
157 size_t i;
158 uint32_t h;
159 for (i = 0, h = 0; i < len; i++) {
160 if (!isspace(cp[i])) {
161 h = h * 3 + (cp[i] & 0xff);
162 }
163 }
164 return h;
165 }
166
167 /*
168 * Compare lines s1 of length n1 and s2 of length n2, ignoring
169 * whitespace difference. Returns 1 if they match, 0 otherwise
170 */
171 static int fuzzy_matchlines(const char *s1, size_t n1,
172 const char *s2, size_t n2)
173 {
174 const char *last1 = s1 + n1 - 1;
175 const char *last2 = s2 + n2 - 1;
176 int result = 0;
177
178 /* ignore line endings */
179 while ((*last1 == '\r') || (*last1 == '\n'))
180 last1--;
181 while ((*last2 == '\r') || (*last2 == '\n'))
182 last2--;
183
184 /* skip leading whitespaces, if both begin with whitespace */
185 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) {
186 while (isspace(*s1) && (s1 <= last1))
187 s1++;
188 while (isspace(*s2) && (s2 <= last2))
189 s2++;
190 }
191 /* early return if both lines are empty */
192 if ((s1 > last1) && (s2 > last2))
193 return 1;
194 while (!result) {
195 result = *s1++ - *s2++;
196 /*
197 * Skip whitespace inside. We check for whitespace on
198 * both buffers because we don't want "a b" to match
199 * "ab"
200 */
201 if (isspace(*s1) && isspace(*s2)) {
202 while (isspace(*s1) && s1 <= last1)
203 s1++;
204 while (isspace(*s2) && s2 <= last2)
205 s2++;
206 }
207 /*
208 * If we reached the end on one side only,
209 * lines don't match
210 */
211 if (
212 ((s2 > last2) && (s1 <= last1)) ||
213 ((s1 > last1) && (s2 <= last2)))
214 return 0;
215 if ((s1 > last1) && (s2 > last2))
216 break;
217 }
218
219 return !result;
220 }
221
222 static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag)
223 {
224 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc);
225 img->line_allocated[img->nr].len = len;
226 img->line_allocated[img->nr].hash = hash_line(bol, len);
227 img->line_allocated[img->nr].flag = flag;
228 img->nr++;
229 }
230
231 /*
232 * "buf" has the file contents to be patched (read from various sources).
233 * attach it to "image" and add line-based index to it.
234 * "image" now owns the "buf".
235 */
236 static void prepare_image(struct image *image, char *buf, size_t len,
237 int prepare_linetable)
238 {
239 const char *cp, *ep;
240
241 memset(image, 0, sizeof(*image));
242 image->buf = buf;
243 image->len = len;
244
245 if (!prepare_linetable)
246 return;
247
248 ep = image->buf + image->len;
249 cp = image->buf;
250 while (cp < ep) {
251 const char *next;
252 for (next = cp; next < ep && *next != '\n'; next++)
253 ;
254 if (next < ep)
255 next++;
256 add_line_info(image, cp, next - cp, 0);
257 cp = next;
258 }
259 image->line = image->line_allocated;
260 }
261
262 static void clear_image(struct image *image)
263 {
264 free(image->buf);
265 free(image->line_allocated);
266 memset(image, 0, sizeof(*image));
267 }
268
269 /* fmt must contain _one_ %s and no other substitution */
270 static void say_patch_name(FILE *output, const char *fmt, struct patch *patch)
271 {
272 struct strbuf sb = STRBUF_INIT;
273
274 if (patch->old_name && patch->new_name &&
275 strcmp(patch->old_name, patch->new_name)) {
276 quote_c_style(patch->old_name, &sb, NULL, 0);
277 strbuf_addstr(&sb, " => ");
278 quote_c_style(patch->new_name, &sb, NULL, 0);
279 } else {
280 const char *n = patch->new_name;
281 if (!n)
282 n = patch->old_name;
283 quote_c_style(n, &sb, NULL, 0);
284 }
285 fprintf(output, fmt, sb.buf);
286 fputc('\n', output);
287 strbuf_release(&sb);
288 }
289
290 #define SLOP (16)
291
292 static int read_patch_file(struct strbuf *sb, int fd)
293 {
294 if (strbuf_read(sb, fd, 0) < 0)
295 return error_errno("git apply: failed to read");
296
297 /*
298 * Make sure that we have some slop in the buffer
299 * so that we can do speculative "memcmp" etc, and
300 * see to it that it is NUL-filled.
301 */
302 strbuf_grow(sb, SLOP);
303 memset(sb->buf + sb->len, 0, SLOP);
304 return 0;
305 }
306
307 static unsigned long linelen(const char *buffer, unsigned long size)
308 {
309 unsigned long len = 0;
310 while (size--) {
311 len++;
312 if (*buffer++ == '\n')
313 break;
314 }
315 return len;
316 }
317
318 static int is_dev_null(const char *str)
319 {
320 return skip_prefix(str, "/dev/null", &str) && isspace(*str);
321 }
322
323 #define TERM_SPACE 1
324 #define TERM_TAB 2
325
326 static int name_terminate(int c, int terminate)
327 {
328 if (c == ' ' && !(terminate & TERM_SPACE))
329 return 0;
330 if (c == '\t' && !(terminate & TERM_TAB))
331 return 0;
332
333 return 1;
334 }
335
336 /* remove double slashes to make --index work with such filenames */
337 static char *squash_slash(char *name)
338 {
339 int i = 0, j = 0;
340
341 if (!name)
342 return NULL;
343
344 while (name[i]) {
345 if ((name[j++] = name[i++]) == '/')
346 while (name[i] == '/')
347 i++;
348 }
349 name[j] = '\0';
350 return name;
351 }
352
353 static char *find_name_gnu(struct apply_state *state,
354 const char *line,
355 const char *def,
356 int p_value)
357 {
358 struct strbuf name = STRBUF_INIT;
359 char *cp;
360
361 /*
362 * Proposed "new-style" GNU patch/diff format; see
363 * http://marc.info/?l=git&m=112927316408690&w=2
364 */
365 if (unquote_c_style(&name, line, NULL)) {
366 strbuf_release(&name);
367 return NULL;
368 }
369
370 for (cp = name.buf; p_value; p_value--) {
371 cp = strchr(cp, '/');
372 if (!cp) {
373 strbuf_release(&name);
374 return NULL;
375 }
376 cp++;
377 }
378
379 strbuf_remove(&name, 0, cp - name.buf);
380 if (state->root.len)
381 strbuf_insert(&name, 0, state->root.buf, state->root.len);
382 return squash_slash(strbuf_detach(&name, NULL));
383 }
384
385 static size_t sane_tz_len(const char *line, size_t len)
386 {
387 const char *tz, *p;
388
389 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ')
390 return 0;
391 tz = line + len - strlen(" +0500");
392
393 if (tz[1] != '+' && tz[1] != '-')
394 return 0;
395
396 for (p = tz + 2; p != line + len; p++)
397 if (!isdigit(*p))
398 return 0;
399
400 return line + len - tz;
401 }
402
403 static size_t tz_with_colon_len(const char *line, size_t len)
404 {
405 const char *tz, *p;
406
407 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':')
408 return 0;
409 tz = line + len - strlen(" +08:00");
410
411 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-'))
412 return 0;
413 p = tz + 2;
414 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
415 !isdigit(*p++) || !isdigit(*p++))
416 return 0;
417
418 return line + len - tz;
419 }
420
421 static size_t date_len(const char *line, size_t len)
422 {
423 const char *date, *p;
424
425 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-')
426 return 0;
427 p = date = line + len - strlen("72-02-05");
428
429 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
430 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' ||
431 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */
432 return 0;
433
434 if (date - line >= strlen("19") &&
435 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */
436 date -= strlen("19");
437
438 return line + len - date;
439 }
440
441 static size_t short_time_len(const char *line, size_t len)
442 {
443 const char *time, *p;
444
445 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':')
446 return 0;
447 p = time = line + len - strlen(" 07:01:32");
448
449 /* Permit 1-digit hours? */
450 if (*p++ != ' ' ||
451 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
452 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' ||
453 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */
454 return 0;
455
456 return line + len - time;
457 }
458
459 static size_t fractional_time_len(const char *line, size_t len)
460 {
461 const char *p;
462 size_t n;
463
464 /* Expected format: 19:41:17.620000023 */
465 if (!len || !isdigit(line[len - 1]))
466 return 0;
467 p = line + len - 1;
468
469 /* Fractional seconds. */
470 while (p > line && isdigit(*p))
471 p--;
472 if (*p != '.')
473 return 0;
474
475 /* Hours, minutes, and whole seconds. */
476 n = short_time_len(line, p - line);
477 if (!n)
478 return 0;
479
480 return line + len - p + n;
481 }
482
483 static size_t trailing_spaces_len(const char *line, size_t len)
484 {
485 const char *p;
486
487 /* Expected format: ' ' x (1 or more) */
488 if (!len || line[len - 1] != ' ')
489 return 0;
490
491 p = line + len;
492 while (p != line) {
493 p--;
494 if (*p != ' ')
495 return line + len - (p + 1);
496 }
497
498 /* All spaces! */
499 return len;
500 }
501
502 static size_t diff_timestamp_len(const char *line, size_t len)
503 {
504 const char *end = line + len;
505 size_t n;
506
507 /*
508 * Posix: 2010-07-05 19:41:17
509 * GNU: 2010-07-05 19:41:17.620000023 -0500
510 */
511
512 if (!isdigit(end[-1]))
513 return 0;
514
515 n = sane_tz_len(line, end - line);
516 if (!n)
517 n = tz_with_colon_len(line, end - line);
518 end -= n;
519
520 n = short_time_len(line, end - line);
521 if (!n)
522 n = fractional_time_len(line, end - line);
523 end -= n;
524
525 n = date_len(line, end - line);
526 if (!n) /* No date. Too bad. */
527 return 0;
528 end -= n;
529
530 if (end == line) /* No space before date. */
531 return 0;
532 if (end[-1] == '\t') { /* Success! */
533 end--;
534 return line + len - end;
535 }
536 if (end[-1] != ' ') /* No space before date. */
537 return 0;
538
539 /* Whitespace damage. */
540 end -= trailing_spaces_len(line, end - line);
541 return line + len - end;
542 }
543
544 static char *find_name_common(struct apply_state *state,
545 const char *line,
546 const char *def,
547 int p_value,
548 const char *end,
549 int terminate)
550 {
551 int len;
552 const char *start = NULL;
553
554 if (p_value == 0)
555 start = line;
556 while (line != end) {
557 char c = *line;
558
559 if (!end && isspace(c)) {
560 if (c == '\n')
561 break;
562 if (name_terminate(c, terminate))
563 break;
564 }
565 line++;
566 if (c == '/' && !--p_value)
567 start = line;
568 }
569 if (!start)
570 return squash_slash(xstrdup_or_null(def));
571 len = line - start;
572 if (!len)
573 return squash_slash(xstrdup_or_null(def));
574
575 /*
576 * Generally we prefer the shorter name, especially
577 * if the other one is just a variation of that with
578 * something else tacked on to the end (ie "file.orig"
579 * or "file~").
580 */
581 if (def) {
582 int deflen = strlen(def);
583 if (deflen < len && !strncmp(start, def, deflen))
584 return squash_slash(xstrdup(def));
585 }
586
587 if (state->root.len) {
588 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start);
589 return squash_slash(ret);
590 }
591
592 return squash_slash(xmemdupz(start, len));
593 }
594
595 static char *find_name(struct apply_state *state,
596 const char *line,
597 char *def,
598 int p_value,
599 int terminate)
600 {
601 if (*line == '"') {
602 char *name = find_name_gnu(state, line, def, p_value);
603 if (name)
604 return name;
605 }
606
607 return find_name_common(state, line, def, p_value, NULL, terminate);
608 }
609
610 static char *find_name_traditional(struct apply_state *state,
611 const char *line,
612 char *def,
613 int p_value)
614 {
615 size_t len;
616 size_t date_len;
617
618 if (*line == '"') {
619 char *name = find_name_gnu(state, line, def, p_value);
620 if (name)
621 return name;
622 }
623
624 len = strchrnul(line, '\n') - line;
625 date_len = diff_timestamp_len(line, len);
626 if (!date_len)
627 return find_name_common(state, line, def, p_value, NULL, TERM_TAB);
628 len -= date_len;
629
630 return find_name_common(state, line, def, p_value, line + len, 0);
631 }
632
633 static int count_slashes(const char *cp)
634 {
635 int cnt = 0;
636 char ch;
637
638 while ((ch = *cp++))
639 if (ch == '/')
640 cnt++;
641 return cnt;
642 }
643
644 /*
645 * Given the string after "--- " or "+++ ", guess the appropriate
646 * p_value for the given patch.
647 */
648 static int guess_p_value(struct apply_state *state, const char *nameline)
649 {
650 char *name, *cp;
651 int val = -1;
652
653 if (is_dev_null(nameline))
654 return -1;
655 name = find_name_traditional(state, nameline, NULL, 0);
656 if (!name)
657 return -1;
658 cp = strchr(name, '/');
659 if (!cp)
660 val = 0;
661 else if (state->prefix) {
662 /*
663 * Does it begin with "a/$our-prefix" and such? Then this is
664 * very likely to apply to our directory.
665 */
666 if (!strncmp(name, state->prefix, state->prefix_length))
667 val = count_slashes(state->prefix);
668 else {
669 cp++;
670 if (!strncmp(cp, state->prefix, state->prefix_length))
671 val = count_slashes(state->prefix) + 1;
672 }
673 }
674 free(name);
675 return val;
676 }
677
678 /*
679 * Does the ---/+++ line have the POSIX timestamp after the last HT?
680 * GNU diff puts epoch there to signal a creation/deletion event. Is
681 * this such a timestamp?
682 */
683 static int has_epoch_timestamp(const char *nameline)
684 {
685 /*
686 * We are only interested in epoch timestamp; any non-zero
687 * fraction cannot be one, hence "(\.0+)?" in the regexp below.
688 * For the same reason, the date must be either 1969-12-31 or
689 * 1970-01-01, and the seconds part must be "00".
690 */
691 const char stamp_regexp[] =
692 "^(1969-12-31|1970-01-01)"
693 " "
694 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?"
695 " "
696 "([-+][0-2][0-9]:?[0-5][0-9])\n";
697 const char *timestamp = NULL, *cp, *colon;
698 static regex_t *stamp;
699 regmatch_t m[10];
700 int zoneoffset;
701 int hourminute;
702 int status;
703
704 for (cp = nameline; *cp != '\n'; cp++) {
705 if (*cp == '\t')
706 timestamp = cp + 1;
707 }
708 if (!timestamp)
709 return 0;
710 if (!stamp) {
711 stamp = xmalloc(sizeof(*stamp));
712 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) {
713 warning(_("Cannot prepare timestamp regexp %s"),
714 stamp_regexp);
715 return 0;
716 }
717 }
718
719 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0);
720 if (status) {
721 if (status != REG_NOMATCH)
722 warning(_("regexec returned %d for input: %s"),
723 status, timestamp);
724 return 0;
725 }
726
727 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10);
728 if (*colon == ':')
729 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10);
730 else
731 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100);
732 if (timestamp[m[3].rm_so] == '-')
733 zoneoffset = -zoneoffset;
734
735 /*
736 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31
737 * (west of GMT) or 1970-01-01 (east of GMT)
738 */
739 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) ||
740 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10)))
741 return 0;
742
743 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 +
744 strtol(timestamp + 14, NULL, 10) -
745 zoneoffset);
746
747 return ((zoneoffset < 0 && hourminute == 1440) ||
748 (0 <= zoneoffset && !hourminute));
749 }
750
751 /*
752 * Get the name etc info from the ---/+++ lines of a traditional patch header
753 *
754 * FIXME! The end-of-filename heuristics are kind of screwy. For existing
755 * files, we can happily check the index for a match, but for creating a
756 * new file we should try to match whatever "patch" does. I have no idea.
757 */
758 static int parse_traditional_patch(struct apply_state *state,
759 const char *first,
760 const char *second,
761 struct patch *patch)
762 {
763 char *name;
764
765 first += 4; /* skip "--- " */
766 second += 4; /* skip "+++ " */
767 if (!state->p_value_known) {
768 int p, q;
769 p = guess_p_value(state, first);
770 q = guess_p_value(state, second);
771 if (p < 0) p = q;
772 if (0 <= p && p == q) {
773 state->p_value = p;
774 state->p_value_known = 1;
775 }
776 }
777 if (is_dev_null(first)) {
778 patch->is_new = 1;
779 patch->is_delete = 0;
780 name = find_name_traditional(state, second, NULL, state->p_value);
781 patch->new_name = name;
782 } else if (is_dev_null(second)) {
783 patch->is_new = 0;
784 patch->is_delete = 1;
785 name = find_name_traditional(state, first, NULL, state->p_value);
786 patch->old_name = name;
787 } else {
788 char *first_name;
789 first_name = find_name_traditional(state, first, NULL, state->p_value);
790 name = find_name_traditional(state, second, first_name, state->p_value);
791 free(first_name);
792 if (has_epoch_timestamp(first)) {
793 patch->is_new = 1;
794 patch->is_delete = 0;
795 patch->new_name = name;
796 } else if (has_epoch_timestamp(second)) {
797 patch->is_new = 0;
798 patch->is_delete = 1;
799 patch->old_name = name;
800 } else {
801 patch->old_name = name;
802 patch->new_name = xstrdup_or_null(name);
803 }
804 }
805 if (!name)
806 return error(_("unable to find filename in patch at line %d"), state->linenr);
807
808 return 0;
809 }
810
811 static int gitdiff_hdrend(struct apply_state *state,
812 const char *line,
813 struct patch *patch)
814 {
815 return 1;
816 }
817
818 /*
819 * We're anal about diff header consistency, to make
820 * sure that we don't end up having strange ambiguous
821 * patches floating around.
822 *
823 * As a result, gitdiff_{old|new}name() will check
824 * their names against any previous information, just
825 * to make sure..
826 */
827 #define DIFF_OLD_NAME 0
828 #define DIFF_NEW_NAME 1
829
830 static int gitdiff_verify_name(struct apply_state *state,
831 const char *line,
832 int isnull,
833 char **name,
834 int side)
835 {
836 if (!*name && !isnull) {
837 *name = find_name(state, line, NULL, state->p_value, TERM_TAB);
838 return 0;
839 }
840
841 if (*name) {
842 int len = strlen(*name);
843 char *another;
844 if (isnull)
845 return error(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"),
846 *name, state->linenr);
847 another = find_name(state, line, NULL, state->p_value, TERM_TAB);
848 if (!another || memcmp(another, *name, len + 1)) {
849 free(another);
850 return error((side == DIFF_NEW_NAME) ?
851 _("git apply: bad git-diff - inconsistent new filename on line %d") :
852 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr);
853 }
854 free(another);
855 } else {
856 /* expect "/dev/null" */
857 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
858 return error(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr);
859 }
860
861 return 0;
862 }
863
864 static int gitdiff_oldname(struct apply_state *state,
865 const char *line,
866 struct patch *patch)
867 {
868 return gitdiff_verify_name(state, line,
869 patch->is_new, &patch->old_name,
870 DIFF_OLD_NAME);
871 }
872
873 static int gitdiff_newname(struct apply_state *state,
874 const char *line,
875 struct patch *patch)
876 {
877 return gitdiff_verify_name(state, line,
878 patch->is_delete, &patch->new_name,
879 DIFF_NEW_NAME);
880 }
881
882 static int gitdiff_oldmode(struct apply_state *state,
883 const char *line,
884 struct patch *patch)
885 {
886 patch->old_mode = strtoul(line, NULL, 8);
887 return 0;
888 }
889
890 static int gitdiff_newmode(struct apply_state *state,
891 const char *line,
892 struct patch *patch)
893 {
894 patch->new_mode = strtoul(line, NULL, 8);
895 return 0;
896 }
897
898 static int gitdiff_delete(struct apply_state *state,
899 const char *line,
900 struct patch *patch)
901 {
902 patch->is_delete = 1;
903 free(patch->old_name);
904 patch->old_name = xstrdup_or_null(patch->def_name);
905 return gitdiff_oldmode(state, line, patch);
906 }
907
908 static int gitdiff_newfile(struct apply_state *state,
909 const char *line,
910 struct patch *patch)
911 {
912 patch->is_new = 1;
913 free(patch->new_name);
914 patch->new_name = xstrdup_or_null(patch->def_name);
915 return gitdiff_newmode(state, line, patch);
916 }
917
918 static int gitdiff_copysrc(struct apply_state *state,
919 const char *line,
920 struct patch *patch)
921 {
922 patch->is_copy = 1;
923 free(patch->old_name);
924 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
925 return 0;
926 }
927
928 static int gitdiff_copydst(struct apply_state *state,
929 const char *line,
930 struct patch *patch)
931 {
932 patch->is_copy = 1;
933 free(patch->new_name);
934 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
935 return 0;
936 }
937
938 static int gitdiff_renamesrc(struct apply_state *state,
939 const char *line,
940 struct patch *patch)
941 {
942 patch->is_rename = 1;
943 free(patch->old_name);
944 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
945 return 0;
946 }
947
948 static int gitdiff_renamedst(struct apply_state *state,
949 const char *line,
950 struct patch *patch)
951 {
952 patch->is_rename = 1;
953 free(patch->new_name);
954 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0);
955 return 0;
956 }
957
958 static int gitdiff_similarity(struct apply_state *state,
959 const char *line,
960 struct patch *patch)
961 {
962 unsigned long val = strtoul(line, NULL, 10);
963 if (val <= 100)
964 patch->score = val;
965 return 0;
966 }
967
968 static int gitdiff_dissimilarity(struct apply_state *state,
969 const char *line,
970 struct patch *patch)
971 {
972 unsigned long val = strtoul(line, NULL, 10);
973 if (val <= 100)
974 patch->score = val;
975 return 0;
976 }
977
978 static int gitdiff_index(struct apply_state *state,
979 const char *line,
980 struct patch *patch)
981 {
982 /*
983 * index line is N hexadecimal, "..", N hexadecimal,
984 * and optional space with octal mode.
985 */
986 const char *ptr, *eol;
987 int len;
988
989 ptr = strchr(line, '.');
990 if (!ptr || ptr[1] != '.' || 40 < ptr - line)
991 return 0;
992 len = ptr - line;
993 memcpy(patch->old_sha1_prefix, line, len);
994 patch->old_sha1_prefix[len] = 0;
995
996 line = ptr + 2;
997 ptr = strchr(line, ' ');
998 eol = strchrnul(line, '\n');
999
1000 if (!ptr || eol < ptr)
1001 ptr = eol;
1002 len = ptr - line;
1003
1004 if (40 < len)
1005 return 0;
1006 memcpy(patch->new_sha1_prefix, line, len);
1007 patch->new_sha1_prefix[len] = 0;
1008 if (*ptr == ' ')
1009 patch->old_mode = strtoul(ptr+1, NULL, 8);
1010 return 0;
1011 }
1012
1013 /*
1014 * This is normal for a diff that doesn't change anything: we'll fall through
1015 * into the next diff. Tell the parser to break out.
1016 */
1017 static int gitdiff_unrecognized(struct apply_state *state,
1018 const char *line,
1019 struct patch *patch)
1020 {
1021 return 1;
1022 }
1023
1024 /*
1025 * Skip p_value leading components from "line"; as we do not accept
1026 * absolute paths, return NULL in that case.
1027 */
1028 static const char *skip_tree_prefix(struct apply_state *state,
1029 const char *line,
1030 int llen)
1031 {
1032 int nslash;
1033 int i;
1034
1035 if (!state->p_value)
1036 return (llen && line[0] == '/') ? NULL : line;
1037
1038 nslash = state->p_value;
1039 for (i = 0; i < llen; i++) {
1040 int ch = line[i];
1041 if (ch == '/' && --nslash <= 0)
1042 return (i == 0) ? NULL : &line[i + 1];
1043 }
1044 return NULL;
1045 }
1046
1047 /*
1048 * This is to extract the same name that appears on "diff --git"
1049 * line. We do not find and return anything if it is a rename
1050 * patch, and it is OK because we will find the name elsewhere.
1051 * We need to reliably find name only when it is mode-change only,
1052 * creation or deletion of an empty file. In any of these cases,
1053 * both sides are the same name under a/ and b/ respectively.
1054 */
1055 static char *git_header_name(struct apply_state *state,
1056 const char *line,
1057 int llen)
1058 {
1059 const char *name;
1060 const char *second = NULL;
1061 size_t len, line_len;
1062
1063 line += strlen("diff --git ");
1064 llen -= strlen("diff --git ");
1065
1066 if (*line == '"') {
1067 const char *cp;
1068 struct strbuf first = STRBUF_INIT;
1069 struct strbuf sp = STRBUF_INIT;
1070
1071 if (unquote_c_style(&first, line, &second))
1072 goto free_and_fail1;
1073
1074 /* strip the a/b prefix including trailing slash */
1075 cp = skip_tree_prefix(state, first.buf, first.len);
1076 if (!cp)
1077 goto free_and_fail1;
1078 strbuf_remove(&first, 0, cp - first.buf);
1079
1080 /*
1081 * second points at one past closing dq of name.
1082 * find the second name.
1083 */
1084 while ((second < line + llen) && isspace(*second))
1085 second++;
1086
1087 if (line + llen <= second)
1088 goto free_and_fail1;
1089 if (*second == '"') {
1090 if (unquote_c_style(&sp, second, NULL))
1091 goto free_and_fail1;
1092 cp = skip_tree_prefix(state, sp.buf, sp.len);
1093 if (!cp)
1094 goto free_and_fail1;
1095 /* They must match, otherwise ignore */
1096 if (strcmp(cp, first.buf))
1097 goto free_and_fail1;
1098 strbuf_release(&sp);
1099 return strbuf_detach(&first, NULL);
1100 }
1101
1102 /* unquoted second */
1103 cp = skip_tree_prefix(state, second, line + llen - second);
1104 if (!cp)
1105 goto free_and_fail1;
1106 if (line + llen - cp != first.len ||
1107 memcmp(first.buf, cp, first.len))
1108 goto free_and_fail1;
1109 return strbuf_detach(&first, NULL);
1110
1111 free_and_fail1:
1112 strbuf_release(&first);
1113 strbuf_release(&sp);
1114 return NULL;
1115 }
1116
1117 /* unquoted first name */
1118 name = skip_tree_prefix(state, line, llen);
1119 if (!name)
1120 return NULL;
1121
1122 /*
1123 * since the first name is unquoted, a dq if exists must be
1124 * the beginning of the second name.
1125 */
1126 for (second = name; second < line + llen; second++) {
1127 if (*second == '"') {
1128 struct strbuf sp = STRBUF_INIT;
1129 const char *np;
1130
1131 if (unquote_c_style(&sp, second, NULL))
1132 goto free_and_fail2;
1133
1134 np = skip_tree_prefix(state, sp.buf, sp.len);
1135 if (!np)
1136 goto free_and_fail2;
1137
1138 len = sp.buf + sp.len - np;
1139 if (len < second - name &&
1140 !strncmp(np, name, len) &&
1141 isspace(name[len])) {
1142 /* Good */
1143 strbuf_remove(&sp, 0, np - sp.buf);
1144 return strbuf_detach(&sp, NULL);
1145 }
1146
1147 free_and_fail2:
1148 strbuf_release(&sp);
1149 return NULL;
1150 }
1151 }
1152
1153 /*
1154 * Accept a name only if it shows up twice, exactly the same
1155 * form.
1156 */
1157 second = strchr(name, '\n');
1158 if (!second)
1159 return NULL;
1160 line_len = second - name;
1161 for (len = 0 ; ; len++) {
1162 switch (name[len]) {
1163 default:
1164 continue;
1165 case '\n':
1166 return NULL;
1167 case '\t': case ' ':
1168 /*
1169 * Is this the separator between the preimage
1170 * and the postimage pathname? Again, we are
1171 * only interested in the case where there is
1172 * no rename, as this is only to set def_name
1173 * and a rename patch has the names elsewhere
1174 * in an unambiguous form.
1175 */
1176 if (!name[len + 1])
1177 return NULL; /* no postimage name */
1178 second = skip_tree_prefix(state, name + len + 1,
1179 line_len - (len + 1));
1180 if (!second)
1181 return NULL;
1182 /*
1183 * Does len bytes starting at "name" and "second"
1184 * (that are separated by one HT or SP we just
1185 * found) exactly match?
1186 */
1187 if (second[len] == '\n' && !strncmp(name, second, len))
1188 return xmemdupz(name, len);
1189 }
1190 }
1191 }
1192
1193 /* Verify that we recognize the lines following a git header */
1194 static int parse_git_header(struct apply_state *state,
1195 const char *line,
1196 int len,
1197 unsigned int size,
1198 struct patch *patch)
1199 {
1200 unsigned long offset;
1201
1202 /* A git diff has explicit new/delete information, so we don't guess */
1203 patch->is_new = 0;
1204 patch->is_delete = 0;
1205
1206 /*
1207 * Some things may not have the old name in the
1208 * rest of the headers anywhere (pure mode changes,
1209 * or removing or adding empty files), so we get
1210 * the default name from the header.
1211 */
1212 patch->def_name = git_header_name(state, line, len);
1213 if (patch->def_name && state->root.len) {
1214 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);
1215 free(patch->def_name);
1216 patch->def_name = s;
1217 }
1218
1219 line += len;
1220 size -= len;
1221 state->linenr++;
1222 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {
1223 static const struct opentry {
1224 const char *str;
1225 int (*fn)(struct apply_state *, const char *, struct patch *);
1226 } optable[] = {
1227 { "@@ -", gitdiff_hdrend },
1228 { "--- ", gitdiff_oldname },
1229 { "+++ ", gitdiff_newname },
1230 { "old mode ", gitdiff_oldmode },
1231 { "new mode ", gitdiff_newmode },
1232 { "deleted file mode ", gitdiff_delete },
1233 { "new file mode ", gitdiff_newfile },
1234 { "copy from ", gitdiff_copysrc },
1235 { "copy to ", gitdiff_copydst },
1236 { "rename old ", gitdiff_renamesrc },
1237 { "rename new ", gitdiff_renamedst },
1238 { "rename from ", gitdiff_renamesrc },
1239 { "rename to ", gitdiff_renamedst },
1240 { "similarity index ", gitdiff_similarity },
1241 { "dissimilarity index ", gitdiff_dissimilarity },
1242 { "index ", gitdiff_index },
1243 { "", gitdiff_unrecognized },
1244 };
1245 int i;
1246
1247 len = linelen(line, size);
1248 if (!len || line[len-1] != '\n')
1249 break;
1250 for (i = 0; i < ARRAY_SIZE(optable); i++) {
1251 const struct opentry *p = optable + i;
1252 int oplen = strlen(p->str);
1253 int res;
1254 if (len < oplen || memcmp(p->str, line, oplen))
1255 continue;
1256 res = p->fn(state, line + oplen, patch);
1257 if (res < 0)
1258 return -1;
1259 if (res > 0)
1260 return offset;
1261 break;
1262 }
1263 }
1264
1265 return offset;
1266 }
1267
1268 static int parse_num(const char *line, unsigned long *p)
1269 {
1270 char *ptr;
1271
1272 if (!isdigit(*line))
1273 return 0;
1274 *p = strtoul(line, &ptr, 10);
1275 return ptr - line;
1276 }
1277
1278 static int parse_range(const char *line, int len, int offset, const char *expect,
1279 unsigned long *p1, unsigned long *p2)
1280 {
1281 int digits, ex;
1282
1283 if (offset < 0 || offset >= len)
1284 return -1;
1285 line += offset;
1286 len -= offset;
1287
1288 digits = parse_num(line, p1);
1289 if (!digits)
1290 return -1;
1291
1292 offset += digits;
1293 line += digits;
1294 len -= digits;
1295
1296 *p2 = 1;
1297 if (*line == ',') {
1298 digits = parse_num(line+1, p2);
1299 if (!digits)
1300 return -1;
1301
1302 offset += digits+1;
1303 line += digits+1;
1304 len -= digits+1;
1305 }
1306
1307 ex = strlen(expect);
1308 if (ex > len)
1309 return -1;
1310 if (memcmp(line, expect, ex))
1311 return -1;
1312
1313 return offset + ex;
1314 }
1315
1316 static void recount_diff(const char *line, int size, struct fragment *fragment)
1317 {
1318 int oldlines = 0, newlines = 0, ret = 0;
1319
1320 if (size < 1) {
1321 warning("recount: ignore empty hunk");
1322 return;
1323 }
1324
1325 for (;;) {
1326 int len = linelen(line, size);
1327 size -= len;
1328 line += len;
1329
1330 if (size < 1)
1331 break;
1332
1333 switch (*line) {
1334 case ' ': case '\n':
1335 newlines++;
1336 /* fall through */
1337 case '-':
1338 oldlines++;
1339 continue;
1340 case '+':
1341 newlines++;
1342 continue;
1343 case '\\':
1344 continue;
1345 case '@':
1346 ret = size < 3 || !starts_with(line, "@@ ");
1347 break;
1348 case 'd':
1349 ret = size < 5 || !starts_with(line, "diff ");
1350 break;
1351 default:
1352 ret = -1;
1353 break;
1354 }
1355 if (ret) {
1356 warning(_("recount: unexpected line: %.*s"),
1357 (int)linelen(line, size), line);
1358 return;
1359 }
1360 break;
1361 }
1362 fragment->oldlines = oldlines;
1363 fragment->newlines = newlines;
1364 }
1365
1366 /*
1367 * Parse a unified diff fragment header of the
1368 * form "@@ -a,b +c,d @@"
1369 */
1370 static int parse_fragment_header(const char *line, int len, struct fragment *fragment)
1371 {
1372 int offset;
1373
1374 if (!len || line[len-1] != '\n')
1375 return -1;
1376
1377 /* Figure out the number of lines in a fragment */
1378 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
1379 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
1380
1381 return offset;
1382 }
1383
1384 /*
1385 * Find file diff header
1386 *
1387 * Returns:
1388 * -1 if no header was found
1389 * -128 in case of error
1390 * the size of the header in bytes (called "offset") otherwise
1391 */
1392 static int find_header(struct apply_state *state,
1393 const char *line,
1394 unsigned long size,
1395 int *hdrsize,
1396 struct patch *patch)
1397 {
1398 unsigned long offset, len;
1399
1400 patch->is_toplevel_relative = 0;
1401 patch->is_rename = patch->is_copy = 0;
1402 patch->is_new = patch->is_delete = -1;
1403 patch->old_mode = patch->new_mode = 0;
1404 patch->old_name = patch->new_name = NULL;
1405 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {
1406 unsigned long nextlen;
1407
1408 len = linelen(line, size);
1409 if (!len)
1410 break;
1411
1412 /* Testing this early allows us to take a few shortcuts.. */
1413 if (len < 6)
1414 continue;
1415
1416 /*
1417 * Make sure we don't find any unconnected patch fragments.
1418 * That's a sign that we didn't find a header, and that a
1419 * patch has become corrupted/broken up.
1420 */
1421 if (!memcmp("@@ -", line, 4)) {
1422 struct fragment dummy;
1423 if (parse_fragment_header(line, len, &dummy) < 0)
1424 continue;
1425 error(_("patch fragment without header at line %d: %.*s"),
1426 state->linenr, (int)len-1, line);
1427 return -128;
1428 }
1429
1430 if (size < len + 6)
1431 break;
1432
1433 /*
1434 * Git patch? It might not have a real patch, just a rename
1435 * or mode change, so we handle that specially
1436 */
1437 if (!memcmp("diff --git ", line, 11)) {
1438 int git_hdr_len = parse_git_header(state, line, len, size, patch);
1439 if (git_hdr_len < 0)
1440 return -128;
1441 if (git_hdr_len <= len)
1442 continue;
1443 if (!patch->old_name && !patch->new_name) {
1444 if (!patch->def_name) {
1445 error(Q_("git diff header lacks filename information when removing "
1446 "%d leading pathname component (line %d)",
1447 "git diff header lacks filename information when removing "
1448 "%d leading pathname components (line %d)",
1449 state->p_value),
1450 state->p_value, state->linenr);
1451 return -128;
1452 }
1453 patch->old_name = xstrdup(patch->def_name);
1454 patch->new_name = xstrdup(patch->def_name);
1455 }
1456 if (!patch->is_delete && !patch->new_name) {
1457 error("git diff header lacks filename information "
1458 "(line %d)", state->linenr);
1459 return -128;
1460 }
1461 patch->is_toplevel_relative = 1;
1462 *hdrsize = git_hdr_len;
1463 return offset;
1464 }
1465
1466 /* --- followed by +++ ? */
1467 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))
1468 continue;
1469
1470 /*
1471 * We only accept unified patches, so we want it to
1472 * at least have "@@ -a,b +c,d @@\n", which is 14 chars
1473 * minimum ("@@ -0,0 +1 @@\n" is the shortest).
1474 */
1475 nextlen = linelen(line + len, size - len);
1476 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
1477 continue;
1478
1479 /* Ok, we'll consider it a patch */
1480 if (parse_traditional_patch(state, line, line+len, patch))
1481 return -128;
1482 *hdrsize = len + nextlen;
1483 state->linenr += 2;
1484 return offset;
1485 }
1486 return -1;
1487 }
1488
1489 static void record_ws_error(struct apply_state *state,
1490 unsigned result,
1491 const char *line,
1492 int len,
1493 int linenr)
1494 {
1495 char *err;
1496
1497 if (!result)
1498 return;
1499
1500 state->whitespace_error++;
1501 if (state->squelch_whitespace_errors &&
1502 state->squelch_whitespace_errors < state->whitespace_error)
1503 return;
1504
1505 err = whitespace_error_string(result);
1506 fprintf(stderr, "%s:%d: %s.\n%.*s\n",
1507 state->patch_input_file, linenr, err, len, line);
1508 free(err);
1509 }
1510
1511 static void check_whitespace(struct apply_state *state,
1512 const char *line,
1513 int len,
1514 unsigned ws_rule)
1515 {
1516 unsigned result = ws_check(line + 1, len - 1, ws_rule);
1517
1518 record_ws_error(state, result, line + 1, len - 2, state->linenr);
1519 }
1520
1521 /*
1522 * Parse a unified diff. Note that this really needs to parse each
1523 * fragment separately, since the only way to know the difference
1524 * between a "---" that is part of a patch, and a "---" that starts
1525 * the next patch is to look at the line counts..
1526 */
1527 static int parse_fragment(struct apply_state *state,
1528 const char *line,
1529 unsigned long size,
1530 struct patch *patch,
1531 struct fragment *fragment)
1532 {
1533 int added, deleted;
1534 int len = linelen(line, size), offset;
1535 unsigned long oldlines, newlines;
1536 unsigned long leading, trailing;
1537
1538 offset = parse_fragment_header(line, len, fragment);
1539 if (offset < 0)
1540 return -1;
1541 if (offset > 0 && patch->recount)
1542 recount_diff(line + offset, size - offset, fragment);
1543 oldlines = fragment->oldlines;
1544 newlines = fragment->newlines;
1545 leading = 0;
1546 trailing = 0;
1547
1548 /* Parse the thing.. */
1549 line += len;
1550 size -= len;
1551 state->linenr++;
1552 added = deleted = 0;
1553 for (offset = len;
1554 0 < size;
1555 offset += len, size -= len, line += len, state->linenr++) {
1556 if (!oldlines && !newlines)
1557 break;
1558 len = linelen(line, size);
1559 if (!len || line[len-1] != '\n')
1560 return -1;
1561 switch (*line) {
1562 default:
1563 return -1;
1564 case '\n': /* newer GNU diff, an empty context line */
1565 case ' ':
1566 oldlines--;
1567 newlines--;
1568 if (!deleted && !added)
1569 leading++;
1570 trailing++;
1571 if (!state->apply_in_reverse &&
1572 state->ws_error_action == correct_ws_error)
1573 check_whitespace(state, line, len, patch->ws_rule);
1574 break;
1575 case '-':
1576 if (state->apply_in_reverse &&
1577 state->ws_error_action != nowarn_ws_error)
1578 check_whitespace(state, line, len, patch->ws_rule);
1579 deleted++;
1580 oldlines--;
1581 trailing = 0;
1582 break;
1583 case '+':
1584 if (!state->apply_in_reverse &&
1585 state->ws_error_action != nowarn_ws_error)
1586 check_whitespace(state, line, len, patch->ws_rule);
1587 added++;
1588 newlines--;
1589 trailing = 0;
1590 break;
1591
1592 /*
1593 * We allow "\ No newline at end of file". Depending
1594 * on locale settings when the patch was produced we
1595 * don't know what this line looks like. The only
1596 * thing we do know is that it begins with "\ ".
1597 * Checking for 12 is just for sanity check -- any
1598 * l10n of "\ No newline..." is at least that long.
1599 */
1600 case '\\':
1601 if (len < 12 || memcmp(line, "\\ ", 2))
1602 return -1;
1603 break;
1604 }
1605 }
1606 if (oldlines || newlines)
1607 return -1;
1608 if (!deleted && !added)
1609 return -1;
1610
1611 fragment->leading = leading;
1612 fragment->trailing = trailing;
1613
1614 /*
1615 * If a fragment ends with an incomplete line, we failed to include
1616 * it in the above loop because we hit oldlines == newlines == 0
1617 * before seeing it.
1618 */
1619 if (12 < size && !memcmp(line, "\\ ", 2))
1620 offset += linelen(line, size);
1621
1622 patch->lines_added += added;
1623 patch->lines_deleted += deleted;
1624
1625 if (0 < patch->is_new && oldlines)
1626 return error(_("new file depends on old contents"));
1627 if (0 < patch->is_delete && newlines)
1628 return error(_("deleted file still has contents"));
1629 return offset;
1630 }
1631
1632 /*
1633 * We have seen "diff --git a/... b/..." header (or a traditional patch
1634 * header). Read hunks that belong to this patch into fragments and hang
1635 * them to the given patch structure.
1636 *
1637 * The (fragment->patch, fragment->size) pair points into the memory given
1638 * by the caller, not a copy, when we return.
1639 *
1640 * Returns:
1641 * -1 in case of error,
1642 * the number of bytes in the patch otherwise.
1643 */
1644 static int parse_single_patch(struct apply_state *state,
1645 const char *line,
1646 unsigned long size,
1647 struct patch *patch)
1648 {
1649 unsigned long offset = 0;
1650 unsigned long oldlines = 0, newlines = 0, context = 0;
1651 struct fragment **fragp = &patch->fragments;
1652
1653 while (size > 4 && !memcmp(line, "@@ -", 4)) {
1654 struct fragment *fragment;
1655 int len;
1656
1657 fragment = xcalloc(1, sizeof(*fragment));
1658 fragment->linenr = state->linenr;
1659 len = parse_fragment(state, line, size, patch, fragment);
1660 if (len <= 0) {
1661 free(fragment);
1662 return error(_("corrupt patch at line %d"), state->linenr);
1663 }
1664 fragment->patch = line;
1665 fragment->size = len;
1666 oldlines += fragment->oldlines;
1667 newlines += fragment->newlines;
1668 context += fragment->leading + fragment->trailing;
1669
1670 *fragp = fragment;
1671 fragp = &fragment->next;
1672
1673 offset += len;
1674 line += len;
1675 size -= len;
1676 }
1677
1678 /*
1679 * If something was removed (i.e. we have old-lines) it cannot
1680 * be creation, and if something was added it cannot be
1681 * deletion. However, the reverse is not true; --unified=0
1682 * patches that only add are not necessarily creation even
1683 * though they do not have any old lines, and ones that only
1684 * delete are not necessarily deletion.
1685 *
1686 * Unfortunately, a real creation/deletion patch do _not_ have
1687 * any context line by definition, so we cannot safely tell it
1688 * apart with --unified=0 insanity. At least if the patch has
1689 * more than one hunk it is not creation or deletion.
1690 */
1691 if (patch->is_new < 0 &&
1692 (oldlines || (patch->fragments && patch->fragments->next)))
1693 patch->is_new = 0;
1694 if (patch->is_delete < 0 &&
1695 (newlines || (patch->fragments && patch->fragments->next)))
1696 patch->is_delete = 0;
1697
1698 if (0 < patch->is_new && oldlines)
1699 return error(_("new file %s depends on old contents"), patch->new_name);
1700 if (0 < patch->is_delete && newlines)
1701 return error(_("deleted file %s still has contents"), patch->old_name);
1702 if (!patch->is_delete && !newlines && context)
1703 fprintf_ln(stderr,
1704 _("** warning: "
1705 "file %s becomes empty but is not deleted"),
1706 patch->new_name);
1707
1708 return offset;
1709 }
1710
1711 static inline int metadata_changes(struct patch *patch)
1712 {
1713 return patch->is_rename > 0 ||
1714 patch->is_copy > 0 ||
1715 patch->is_new > 0 ||
1716 patch->is_delete ||
1717 (patch->old_mode && patch->new_mode &&
1718 patch->old_mode != patch->new_mode);
1719 }
1720
1721 static char *inflate_it(const void *data, unsigned long size,
1722 unsigned long inflated_size)
1723 {
1724 git_zstream stream;
1725 void *out;
1726 int st;
1727
1728 memset(&stream, 0, sizeof(stream));
1729
1730 stream.next_in = (unsigned char *)data;
1731 stream.avail_in = size;
1732 stream.next_out = out = xmalloc(inflated_size);
1733 stream.avail_out = inflated_size;
1734 git_inflate_init(&stream);
1735 st = git_inflate(&stream, Z_FINISH);
1736 git_inflate_end(&stream);
1737 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1738 free(out);
1739 return NULL;
1740 }
1741 return out;
1742 }
1743
1744 /*
1745 * Read a binary hunk and return a new fragment; fragment->patch
1746 * points at an allocated memory that the caller must free, so
1747 * it is marked as "->free_patch = 1".
1748 */
1749 static struct fragment *parse_binary_hunk(struct apply_state *state,
1750 char **buf_p,
1751 unsigned long *sz_p,
1752 int *status_p,
1753 int *used_p)
1754 {
1755 /*
1756 * Expect a line that begins with binary patch method ("literal"
1757 * or "delta"), followed by the length of data before deflating.
1758 * a sequence of 'length-byte' followed by base-85 encoded data
1759 * should follow, terminated by a newline.
1760 *
1761 * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1762 * and we would limit the patch line to 66 characters,
1763 * so one line can fit up to 13 groups that would decode
1764 * to 52 bytes max. The length byte 'A'-'Z' corresponds
1765 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1766 */
1767 int llen, used;
1768 unsigned long size = *sz_p;
1769 char *buffer = *buf_p;
1770 int patch_method;
1771 unsigned long origlen;
1772 char *data = NULL;
1773 int hunk_size = 0;
1774 struct fragment *frag;
1775
1776 llen = linelen(buffer, size);
1777 used = llen;
1778
1779 *status_p = 0;
1780
1781 if (starts_with(buffer, "delta ")) {
1782 patch_method = BINARY_DELTA_DEFLATED;
1783 origlen = strtoul(buffer + 6, NULL, 10);
1784 }
1785 else if (starts_with(buffer, "literal ")) {
1786 patch_method = BINARY_LITERAL_DEFLATED;
1787 origlen = strtoul(buffer + 8, NULL, 10);
1788 }
1789 else
1790 return NULL;
1791
1792 state->linenr++;
1793 buffer += llen;
1794 while (1) {
1795 int byte_length, max_byte_length, newsize;
1796 llen = linelen(buffer, size);
1797 used += llen;
1798 state->linenr++;
1799 if (llen == 1) {
1800 /* consume the blank line */
1801 buffer++;
1802 size--;
1803 break;
1804 }
1805 /*
1806 * Minimum line is "A00000\n" which is 7-byte long,
1807 * and the line length must be multiple of 5 plus 2.
1808 */
1809 if ((llen < 7) || (llen-2) % 5)
1810 goto corrupt;
1811 max_byte_length = (llen - 2) / 5 * 4;
1812 byte_length = *buffer;
1813 if ('A' <= byte_length && byte_length <= 'Z')
1814 byte_length = byte_length - 'A' + 1;
1815 else if ('a' <= byte_length && byte_length <= 'z')
1816 byte_length = byte_length - 'a' + 27;
1817 else
1818 goto corrupt;
1819 /* if the input length was not multiple of 4, we would
1820 * have filler at the end but the filler should never
1821 * exceed 3 bytes
1822 */
1823 if (max_byte_length < byte_length ||
1824 byte_length <= max_byte_length - 4)
1825 goto corrupt;
1826 newsize = hunk_size + byte_length;
1827 data = xrealloc(data, newsize);
1828 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1829 goto corrupt;
1830 hunk_size = newsize;
1831 buffer += llen;
1832 size -= llen;
1833 }
1834
1835 frag = xcalloc(1, sizeof(*frag));
1836 frag->patch = inflate_it(data, hunk_size, origlen);
1837 frag->free_patch = 1;
1838 if (!frag->patch)
1839 goto corrupt;
1840 free(data);
1841 frag->size = origlen;
1842 *buf_p = buffer;
1843 *sz_p = size;
1844 *used_p = used;
1845 frag->binary_patch_method = patch_method;
1846 return frag;
1847
1848 corrupt:
1849 free(data);
1850 *status_p = -1;
1851 error(_("corrupt binary patch at line %d: %.*s"),
1852 state->linenr-1, llen-1, buffer);
1853 return NULL;
1854 }
1855
1856 /*
1857 * Returns:
1858 * -1 in case of error,
1859 * the length of the parsed binary patch otherwise
1860 */
1861 static int parse_binary(struct apply_state *state,
1862 char *buffer,
1863 unsigned long size,
1864 struct patch *patch)
1865 {
1866 /*
1867 * We have read "GIT binary patch\n"; what follows is a line
1868 * that says the patch method (currently, either "literal" or
1869 * "delta") and the length of data before deflating; a
1870 * sequence of 'length-byte' followed by base-85 encoded data
1871 * follows.
1872 *
1873 * When a binary patch is reversible, there is another binary
1874 * hunk in the same format, starting with patch method (either
1875 * "literal" or "delta") with the length of data, and a sequence
1876 * of length-byte + base-85 encoded data, terminated with another
1877 * empty line. This data, when applied to the postimage, produces
1878 * the preimage.
1879 */
1880 struct fragment *forward;
1881 struct fragment *reverse;
1882 int status;
1883 int used, used_1;
1884
1885 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);
1886 if (!forward && !status)
1887 /* there has to be one hunk (forward hunk) */
1888 return error(_("unrecognized binary patch at line %d"), state->linenr-1);
1889 if (status)
1890 /* otherwise we already gave an error message */
1891 return status;
1892
1893 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);
1894 if (reverse)
1895 used += used_1;
1896 else if (status) {
1897 /*
1898 * Not having reverse hunk is not an error, but having
1899 * a corrupt reverse hunk is.
1900 */
1901 free((void*) forward->patch);
1902 free(forward);
1903 return status;
1904 }
1905 forward->next = reverse;
1906 patch->fragments = forward;
1907 patch->is_binary = 1;
1908 return used;
1909 }
1910
1911 static void prefix_one(struct apply_state *state, char **name)
1912 {
1913 char *old_name = *name;
1914 if (!old_name)
1915 return;
1916 *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));
1917 free(old_name);
1918 }
1919
1920 static void prefix_patch(struct apply_state *state, struct patch *p)
1921 {
1922 if (!state->prefix || p->is_toplevel_relative)
1923 return;
1924 prefix_one(state, &p->new_name);
1925 prefix_one(state, &p->old_name);
1926 }
1927
1928 /*
1929 * include/exclude
1930 */
1931
1932 static void add_name_limit(struct apply_state *state,
1933 const char *name,
1934 int exclude)
1935 {
1936 struct string_list_item *it;
1937
1938 it = string_list_append(&state->limit_by_name, name);
1939 it->util = exclude ? NULL : (void *) 1;
1940 }
1941
1942 static int use_patch(struct apply_state *state, struct patch *p)
1943 {
1944 const char *pathname = p->new_name ? p->new_name : p->old_name;
1945 int i;
1946
1947 /* Paths outside are not touched regardless of "--include" */
1948 if (0 < state->prefix_length) {
1949 int pathlen = strlen(pathname);
1950 if (pathlen <= state->prefix_length ||
1951 memcmp(state->prefix, pathname, state->prefix_length))
1952 return 0;
1953 }
1954
1955 /* See if it matches any of exclude/include rule */
1956 for (i = 0; i < state->limit_by_name.nr; i++) {
1957 struct string_list_item *it = &state->limit_by_name.items[i];
1958 if (!wildmatch(it->string, pathname, 0, NULL))
1959 return (it->util != NULL);
1960 }
1961
1962 /*
1963 * If we had any include, a path that does not match any rule is
1964 * not used. Otherwise, we saw bunch of exclude rules (or none)
1965 * and such a path is used.
1966 */
1967 return !state->has_include;
1968 }
1969
1970 /*
1971 * Read the patch text in "buffer" that extends for "size" bytes; stop
1972 * reading after seeing a single patch (i.e. changes to a single file).
1973 * Create fragments (i.e. patch hunks) and hang them to the given patch.
1974 *
1975 * Returns:
1976 * -1 if no header was found or parse_binary() failed,
1977 * -128 on another error,
1978 * the number of bytes consumed otherwise,
1979 * so that the caller can call us again for the next patch.
1980 */
1981 static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)
1982 {
1983 int hdrsize, patchsize;
1984 int offset = find_header(state, buffer, size, &hdrsize, patch);
1985
1986 if (offset < 0)
1987 return offset;
1988
1989 prefix_patch(state, patch);
1990
1991 if (!use_patch(state, patch))
1992 patch->ws_rule = 0;
1993 else
1994 patch->ws_rule = whitespace_rule(patch->new_name
1995 ? patch->new_name
1996 : patch->old_name);
1997
1998 patchsize = parse_single_patch(state,
1999 buffer + offset + hdrsize,
2000 size - offset - hdrsize,
2001 patch);
2002
2003 if (patchsize < 0)
2004 return -128;
2005
2006 if (!patchsize) {
2007 static const char git_binary[] = "GIT binary patch\n";
2008 int hd = hdrsize + offset;
2009 unsigned long llen = linelen(buffer + hd, size - hd);
2010
2011 if (llen == sizeof(git_binary) - 1 &&
2012 !memcmp(git_binary, buffer + hd, llen)) {
2013 int used;
2014 state->linenr++;
2015 used = parse_binary(state, buffer + hd + llen,
2016 size - hd - llen, patch);
2017 if (used < 0)
2018 return -1;
2019 if (used)
2020 patchsize = used + llen;
2021 else
2022 patchsize = 0;
2023 }
2024 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
2025 static const char *binhdr[] = {
2026 "Binary files ",
2027 "Files ",
2028 NULL,
2029 };
2030 int i;
2031 for (i = 0; binhdr[i]; i++) {
2032 int len = strlen(binhdr[i]);
2033 if (len < size - hd &&
2034 !memcmp(binhdr[i], buffer + hd, len)) {
2035 state->linenr++;
2036 patch->is_binary = 1;
2037 patchsize = llen;
2038 break;
2039 }
2040 }
2041 }
2042
2043 /* Empty patch cannot be applied if it is a text patch
2044 * without metadata change. A binary patch appears
2045 * empty to us here.
2046 */
2047 if ((state->apply || state->check) &&
2048 (!patch->is_binary && !metadata_changes(patch))) {
2049 error(_("patch with only garbage at line %d"), state->linenr);
2050 return -128;
2051 }
2052 }
2053
2054 return offset + hdrsize + patchsize;
2055 }
2056
2057 #define swap(a,b) myswap((a),(b),sizeof(a))
2058
2059 #define myswap(a, b, size) do { \
2060 unsigned char mytmp[size]; \
2061 memcpy(mytmp, &a, size); \
2062 memcpy(&a, &b, size); \
2063 memcpy(&b, mytmp, size); \
2064 } while (0)
2065
2066 static void reverse_patches(struct patch *p)
2067 {
2068 for (; p; p = p->next) {
2069 struct fragment *frag = p->fragments;
2070
2071 swap(p->new_name, p->old_name);
2072 swap(p->new_mode, p->old_mode);
2073 swap(p->is_new, p->is_delete);
2074 swap(p->lines_added, p->lines_deleted);
2075 swap(p->old_sha1_prefix, p->new_sha1_prefix);
2076
2077 for (; frag; frag = frag->next) {
2078 swap(frag->newpos, frag->oldpos);
2079 swap(frag->newlines, frag->oldlines);
2080 }
2081 }
2082 }
2083
2084 static const char pluses[] =
2085 "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
2086 static const char minuses[]=
2087 "----------------------------------------------------------------------";
2088
2089 static void show_stats(struct apply_state *state, struct patch *patch)
2090 {
2091 struct strbuf qname = STRBUF_INIT;
2092 char *cp = patch->new_name ? patch->new_name : patch->old_name;
2093 int max, add, del;
2094
2095 quote_c_style(cp, &qname, NULL, 0);
2096
2097 /*
2098 * "scale" the filename
2099 */
2100 max = state->max_len;
2101 if (max > 50)
2102 max = 50;
2103
2104 if (qname.len > max) {
2105 cp = strchr(qname.buf + qname.len + 3 - max, '/');
2106 if (!cp)
2107 cp = qname.buf + qname.len + 3 - max;
2108 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);
2109 }
2110
2111 if (patch->is_binary) {
2112 printf(" %-*s | Bin\n", max, qname.buf);
2113 strbuf_release(&qname);
2114 return;
2115 }
2116
2117 printf(" %-*s |", max, qname.buf);
2118 strbuf_release(&qname);
2119
2120 /*
2121 * scale the add/delete
2122 */
2123 max = max + state->max_change > 70 ? 70 - max : state->max_change;
2124 add = patch->lines_added;
2125 del = patch->lines_deleted;
2126
2127 if (state->max_change > 0) {
2128 int total = ((add + del) * max + state->max_change / 2) / state->max_change;
2129 add = (add * max + state->max_change / 2) / state->max_change;
2130 del = total - add;
2131 }
2132 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,
2133 add, pluses, del, minuses);
2134 }
2135
2136 static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)
2137 {
2138 switch (st->st_mode & S_IFMT) {
2139 case S_IFLNK:
2140 if (strbuf_readlink(buf, path, st->st_size) < 0)
2141 return error(_("unable to read symlink %s"), path);
2142 return 0;
2143 case S_IFREG:
2144 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)
2145 return error(_("unable to open or read %s"), path);
2146 convert_to_git(path, buf->buf, buf->len, buf, 0);
2147 return 0;
2148 default:
2149 return -1;
2150 }
2151 }
2152
2153 /*
2154 * Update the preimage, and the common lines in postimage,
2155 * from buffer buf of length len. If postlen is 0 the postimage
2156 * is updated in place, otherwise it's updated on a new buffer
2157 * of length postlen
2158 */
2159
2160 static void update_pre_post_images(struct image *preimage,
2161 struct image *postimage,
2162 char *buf,
2163 size_t len, size_t postlen)
2164 {
2165 int i, ctx, reduced;
2166 char *new, *old, *fixed;
2167 struct image fixed_preimage;
2168
2169 /*
2170 * Update the preimage with whitespace fixes. Note that we
2171 * are not losing preimage->buf -- apply_one_fragment() will
2172 * free "oldlines".
2173 */
2174 prepare_image(&fixed_preimage, buf, len, 1);
2175 assert(postlen
2176 ? fixed_preimage.nr == preimage->nr
2177 : fixed_preimage.nr <= preimage->nr);
2178 for (i = 0; i < fixed_preimage.nr; i++)
2179 fixed_preimage.line[i].flag = preimage->line[i].flag;
2180 free(preimage->line_allocated);
2181 *preimage = fixed_preimage;
2182
2183 /*
2184 * Adjust the common context lines in postimage. This can be
2185 * done in-place when we are shrinking it with whitespace
2186 * fixing, but needs a new buffer when ignoring whitespace or
2187 * expanding leading tabs to spaces.
2188 *
2189 * We trust the caller to tell us if the update can be done
2190 * in place (postlen==0) or not.
2191 */
2192 old = postimage->buf;
2193 if (postlen)
2194 new = postimage->buf = xmalloc(postlen);
2195 else
2196 new = old;
2197 fixed = preimage->buf;
2198
2199 for (i = reduced = ctx = 0; i < postimage->nr; i++) {
2200 size_t l_len = postimage->line[i].len;
2201 if (!(postimage->line[i].flag & LINE_COMMON)) {
2202 /* an added line -- no counterparts in preimage */
2203 memmove(new, old, l_len);
2204 old += l_len;
2205 new += l_len;
2206 continue;
2207 }
2208
2209 /* a common context -- skip it in the original postimage */
2210 old += l_len;
2211
2212 /* and find the corresponding one in the fixed preimage */
2213 while (ctx < preimage->nr &&
2214 !(preimage->line[ctx].flag & LINE_COMMON)) {
2215 fixed += preimage->line[ctx].len;
2216 ctx++;
2217 }
2218
2219 /*
2220 * preimage is expected to run out, if the caller
2221 * fixed addition of trailing blank lines.
2222 */
2223 if (preimage->nr <= ctx) {
2224 reduced++;
2225 continue;
2226 }
2227
2228 /* and copy it in, while fixing the line length */
2229 l_len = preimage->line[ctx].len;
2230 memcpy(new, fixed, l_len);
2231 new += l_len;
2232 fixed += l_len;
2233 postimage->line[i].len = l_len;
2234 ctx++;
2235 }
2236
2237 if (postlen
2238 ? postlen < new - postimage->buf
2239 : postimage->len < new - postimage->buf)
2240 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",
2241 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));
2242
2243 /* Fix the length of the whole thing */
2244 postimage->len = new - postimage->buf;
2245 postimage->nr -= reduced;
2246 }
2247
2248 static int line_by_line_fuzzy_match(struct image *img,
2249 struct image *preimage,
2250 struct image *postimage,
2251 unsigned long try,
2252 int try_lno,
2253 int preimage_limit)
2254 {
2255 int i;
2256 size_t imgoff = 0;
2257 size_t preoff = 0;
2258 size_t postlen = postimage->len;
2259 size_t extra_chars;
2260 char *buf;
2261 char *preimage_eof;
2262 char *preimage_end;
2263 struct strbuf fixed;
2264 char *fixed_buf;
2265 size_t fixed_len;
2266
2267 for (i = 0; i < preimage_limit; i++) {
2268 size_t prelen = preimage->line[i].len;
2269 size_t imglen = img->line[try_lno+i].len;
2270
2271 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,
2272 preimage->buf + preoff, prelen))
2273 return 0;
2274 if (preimage->line[i].flag & LINE_COMMON)
2275 postlen += imglen - prelen;
2276 imgoff += imglen;
2277 preoff += prelen;
2278 }
2279
2280 /*
2281 * Ok, the preimage matches with whitespace fuzz.
2282 *
2283 * imgoff now holds the true length of the target that
2284 * matches the preimage before the end of the file.
2285 *
2286 * Count the number of characters in the preimage that fall
2287 * beyond the end of the file and make sure that all of them
2288 * are whitespace characters. (This can only happen if
2289 * we are removing blank lines at the end of the file.)
2290 */
2291 buf = preimage_eof = preimage->buf + preoff;
2292 for ( ; i < preimage->nr; i++)
2293 preoff += preimage->line[i].len;
2294 preimage_end = preimage->buf + preoff;
2295 for ( ; buf < preimage_end; buf++)
2296 if (!isspace(*buf))
2297 return 0;
2298
2299 /*
2300 * Update the preimage and the common postimage context
2301 * lines to use the same whitespace as the target.
2302 * If whitespace is missing in the target (i.e.
2303 * if the preimage extends beyond the end of the file),
2304 * use the whitespace from the preimage.
2305 */
2306 extra_chars = preimage_end - preimage_eof;
2307 strbuf_init(&fixed, imgoff + extra_chars);
2308 strbuf_add(&fixed, img->buf + try, imgoff);
2309 strbuf_add(&fixed, preimage_eof, extra_chars);
2310 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2311 update_pre_post_images(preimage, postimage,
2312 fixed_buf, fixed_len, postlen);
2313 return 1;
2314 }
2315
2316 static int match_fragment(struct apply_state *state,
2317 struct image *img,
2318 struct image *preimage,
2319 struct image *postimage,
2320 unsigned long try,
2321 int try_lno,
2322 unsigned ws_rule,
2323 int match_beginning, int match_end)
2324 {
2325 int i;
2326 char *fixed_buf, *buf, *orig, *target;
2327 struct strbuf fixed;
2328 size_t fixed_len, postlen;
2329 int preimage_limit;
2330
2331 if (preimage->nr + try_lno <= img->nr) {
2332 /*
2333 * The hunk falls within the boundaries of img.
2334 */
2335 preimage_limit = preimage->nr;
2336 if (match_end && (preimage->nr + try_lno != img->nr))
2337 return 0;
2338 } else if (state->ws_error_action == correct_ws_error &&
2339 (ws_rule & WS_BLANK_AT_EOF)) {
2340 /*
2341 * This hunk extends beyond the end of img, and we are
2342 * removing blank lines at the end of the file. This
2343 * many lines from the beginning of the preimage must
2344 * match with img, and the remainder of the preimage
2345 * must be blank.
2346 */
2347 preimage_limit = img->nr - try_lno;
2348 } else {
2349 /*
2350 * The hunk extends beyond the end of the img and
2351 * we are not removing blanks at the end, so we
2352 * should reject the hunk at this position.
2353 */
2354 return 0;
2355 }
2356
2357 if (match_beginning && try_lno)
2358 return 0;
2359
2360 /* Quick hash check */
2361 for (i = 0; i < preimage_limit; i++)
2362 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||
2363 (preimage->line[i].hash != img->line[try_lno + i].hash))
2364 return 0;
2365
2366 if (preimage_limit == preimage->nr) {
2367 /*
2368 * Do we have an exact match? If we were told to match
2369 * at the end, size must be exactly at try+fragsize,
2370 * otherwise try+fragsize must be still within the preimage,
2371 * and either case, the old piece should match the preimage
2372 * exactly.
2373 */
2374 if ((match_end
2375 ? (try + preimage->len == img->len)
2376 : (try + preimage->len <= img->len)) &&
2377 !memcmp(img->buf + try, preimage->buf, preimage->len))
2378 return 1;
2379 } else {
2380 /*
2381 * The preimage extends beyond the end of img, so
2382 * there cannot be an exact match.
2383 *
2384 * There must be one non-blank context line that match
2385 * a line before the end of img.
2386 */
2387 char *buf_end;
2388
2389 buf = preimage->buf;
2390 buf_end = buf;
2391 for (i = 0; i < preimage_limit; i++)
2392 buf_end += preimage->line[i].len;
2393
2394 for ( ; buf < buf_end; buf++)
2395 if (!isspace(*buf))
2396 break;
2397 if (buf == buf_end)
2398 return 0;
2399 }
2400
2401 /*
2402 * No exact match. If we are ignoring whitespace, run a line-by-line
2403 * fuzzy matching. We collect all the line length information because
2404 * we need it to adjust whitespace if we match.
2405 */
2406 if (state->ws_ignore_action == ignore_ws_change)
2407 return line_by_line_fuzzy_match(img, preimage, postimage,
2408 try, try_lno, preimage_limit);
2409
2410 if (state->ws_error_action != correct_ws_error)
2411 return 0;
2412
2413 /*
2414 * The hunk does not apply byte-by-byte, but the hash says
2415 * it might with whitespace fuzz. We weren't asked to
2416 * ignore whitespace, we were asked to correct whitespace
2417 * errors, so let's try matching after whitespace correction.
2418 *
2419 * While checking the preimage against the target, whitespace
2420 * errors in both fixed, we count how large the corresponding
2421 * postimage needs to be. The postimage prepared by
2422 * apply_one_fragment() has whitespace errors fixed on added
2423 * lines already, but the common lines were propagated as-is,
2424 * which may become longer when their whitespace errors are
2425 * fixed.
2426 */
2427
2428 /* First count added lines in postimage */
2429 postlen = 0;
2430 for (i = 0; i < postimage->nr; i++) {
2431 if (!(postimage->line[i].flag & LINE_COMMON))
2432 postlen += postimage->line[i].len;
2433 }
2434
2435 /*
2436 * The preimage may extend beyond the end of the file,
2437 * but in this loop we will only handle the part of the
2438 * preimage that falls within the file.
2439 */
2440 strbuf_init(&fixed, preimage->len + 1);
2441 orig = preimage->buf;
2442 target = img->buf + try;
2443 for (i = 0; i < preimage_limit; i++) {
2444 size_t oldlen = preimage->line[i].len;
2445 size_t tgtlen = img->line[try_lno + i].len;
2446 size_t fixstart = fixed.len;
2447 struct strbuf tgtfix;
2448 int match;
2449
2450 /* Try fixing the line in the preimage */
2451 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2452
2453 /* Try fixing the line in the target */
2454 strbuf_init(&tgtfix, tgtlen);
2455 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);
2456
2457 /*
2458 * If they match, either the preimage was based on
2459 * a version before our tree fixed whitespace breakage,
2460 * or we are lacking a whitespace-fix patch the tree
2461 * the preimage was based on already had (i.e. target
2462 * has whitespace breakage, the preimage doesn't).
2463 * In either case, we are fixing the whitespace breakages
2464 * so we might as well take the fix together with their
2465 * real change.
2466 */
2467 match = (tgtfix.len == fixed.len - fixstart &&
2468 !memcmp(tgtfix.buf, fixed.buf + fixstart,
2469 fixed.len - fixstart));
2470
2471 /* Add the length if this is common with the postimage */
2472 if (preimage->line[i].flag & LINE_COMMON)
2473 postlen += tgtfix.len;
2474
2475 strbuf_release(&tgtfix);
2476 if (!match)
2477 goto unmatch_exit;
2478
2479 orig += oldlen;
2480 target += tgtlen;
2481 }
2482
2483
2484 /*
2485 * Now handle the lines in the preimage that falls beyond the
2486 * end of the file (if any). They will only match if they are
2487 * empty or only contain whitespace (if WS_BLANK_AT_EOL is
2488 * false).
2489 */
2490 for ( ; i < preimage->nr; i++) {
2491 size_t fixstart = fixed.len; /* start of the fixed preimage */
2492 size_t oldlen = preimage->line[i].len;
2493 int j;
2494
2495 /* Try fixing the line in the preimage */
2496 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);
2497
2498 for (j = fixstart; j < fixed.len; j++)
2499 if (!isspace(fixed.buf[j]))
2500 goto unmatch_exit;
2501
2502 orig += oldlen;
2503 }
2504
2505 /*
2506 * Yes, the preimage is based on an older version that still
2507 * has whitespace breakages unfixed, and fixing them makes the
2508 * hunk match. Update the context lines in the postimage.
2509 */
2510 fixed_buf = strbuf_detach(&fixed, &fixed_len);
2511 if (postlen < postimage->len)
2512 postlen = 0;
2513 update_pre_post_images(preimage, postimage,
2514 fixed_buf, fixed_len, postlen);
2515 return 1;
2516
2517 unmatch_exit:
2518 strbuf_release(&fixed);
2519 return 0;
2520 }
2521
2522 static int find_pos(struct apply_state *state,
2523 struct image *img,
2524 struct image *preimage,
2525 struct image *postimage,
2526 int line,
2527 unsigned ws_rule,
2528 int match_beginning, int match_end)
2529 {
2530 int i;
2531 unsigned long backwards, forwards, try;
2532 int backwards_lno, forwards_lno, try_lno;
2533
2534 /*
2535 * If match_beginning or match_end is specified, there is no
2536 * point starting from a wrong line that will never match and
2537 * wander around and wait for a match at the specified end.
2538 */
2539 if (match_beginning)
2540 line = 0;
2541 else if (match_end)
2542 line = img->nr - preimage->nr;
2543
2544 /*
2545 * Because the comparison is unsigned, the following test
2546 * will also take care of a negative line number that can
2547 * result when match_end and preimage is larger than the target.
2548 */
2549 if ((size_t) line > img->nr)
2550 line = img->nr;
2551
2552 try = 0;
2553 for (i = 0; i < line; i++)
2554 try += img->line[i].len;
2555
2556 /*
2557 * There's probably some smart way to do this, but I'll leave
2558 * that to the smart and beautiful people. I'm simple and stupid.
2559 */
2560 backwards = try;
2561 backwards_lno = line;
2562 forwards = try;
2563 forwards_lno = line;
2564 try_lno = line;
2565
2566 for (i = 0; ; i++) {
2567 if (match_fragment(state, img, preimage, postimage,
2568 try, try_lno, ws_rule,
2569 match_beginning, match_end))
2570 return try_lno;
2571
2572 again:
2573 if (backwards_lno == 0 && forwards_lno == img->nr)
2574 break;
2575
2576 if (i & 1) {
2577 if (backwards_lno == 0) {
2578 i++;
2579 goto again;
2580 }
2581 backwards_lno--;
2582 backwards -= img->line[backwards_lno].len;
2583 try = backwards;
2584 try_lno = backwards_lno;
2585 } else {
2586 if (forwards_lno == img->nr) {
2587 i++;
2588 goto again;
2589 }
2590 forwards += img->line[forwards_lno].len;
2591 forwards_lno++;
2592 try = forwards;
2593 try_lno = forwards_lno;
2594 }
2595
2596 }
2597 return -1;
2598 }
2599
2600 static void remove_first_line(struct image *img)
2601 {
2602 img->buf += img->line[0].len;
2603 img->len -= img->line[0].len;
2604 img->line++;
2605 img->nr--;
2606 }
2607
2608 static void remove_last_line(struct image *img)
2609 {
2610 img->len -= img->line[--img->nr].len;
2611 }
2612
2613 /*
2614 * The change from "preimage" and "postimage" has been found to
2615 * apply at applied_pos (counts in line numbers) in "img".
2616 * Update "img" to remove "preimage" and replace it with "postimage".
2617 */
2618 static void update_image(struct apply_state *state,
2619 struct image *img,
2620 int applied_pos,
2621 struct image *preimage,
2622 struct image *postimage)
2623 {
2624 /*
2625 * remove the copy of preimage at offset in img
2626 * and replace it with postimage
2627 */
2628 int i, nr;
2629 size_t remove_count, insert_count, applied_at = 0;
2630 char *result;
2631 int preimage_limit;
2632
2633 /*
2634 * If we are removing blank lines at the end of img,
2635 * the preimage may extend beyond the end.
2636 * If that is the case, we must be careful only to
2637 * remove the part of the preimage that falls within
2638 * the boundaries of img. Initialize preimage_limit
2639 * to the number of lines in the preimage that falls
2640 * within the boundaries.
2641 */
2642 preimage_limit = preimage->nr;
2643 if (preimage_limit > img->nr - applied_pos)
2644 preimage_limit = img->nr - applied_pos;
2645
2646 for (i = 0; i < applied_pos; i++)
2647 applied_at += img->line[i].len;
2648
2649 remove_count = 0;
2650 for (i = 0; i < preimage_limit; i++)
2651 remove_count += img->line[applied_pos + i].len;
2652 insert_count = postimage->len;
2653
2654 /* Adjust the contents */
2655 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));
2656 memcpy(result, img->buf, applied_at);
2657 memcpy(result + applied_at, postimage->buf, postimage->len);
2658 memcpy(result + applied_at + postimage->len,
2659 img->buf + (applied_at + remove_count),
2660 img->len - (applied_at + remove_count));
2661 free(img->buf);
2662 img->buf = result;
2663 img->len += insert_count - remove_count;
2664 result[img->len] = '\0';
2665
2666 /* Adjust the line table */
2667 nr = img->nr + postimage->nr - preimage_limit;
2668 if (preimage_limit < postimage->nr) {
2669 /*
2670 * NOTE: this knows that we never call remove_first_line()
2671 * on anything other than pre/post image.
2672 */
2673 REALLOC_ARRAY(img->line, nr);
2674 img->line_allocated = img->line;
2675 }
2676 if (preimage_limit != postimage->nr)
2677 memmove(img->line + applied_pos + postimage->nr,
2678 img->line + applied_pos + preimage_limit,
2679 (img->nr - (applied_pos + preimage_limit)) *
2680 sizeof(*img->line));
2681 memcpy(img->line + applied_pos,
2682 postimage->line,
2683 postimage->nr * sizeof(*img->line));
2684 if (!state->allow_overlap)
2685 for (i = 0; i < postimage->nr; i++)
2686 img->line[applied_pos + i].flag |= LINE_PATCHED;
2687 img->nr = nr;
2688 }
2689
2690 /*
2691 * Use the patch-hunk text in "frag" to prepare two images (preimage and
2692 * postimage) for the hunk. Find lines that match "preimage" in "img" and
2693 * replace the part of "img" with "postimage" text.
2694 */
2695 static int apply_one_fragment(struct apply_state *state,
2696 struct image *img, struct fragment *frag,
2697 int inaccurate_eof, unsigned ws_rule,
2698 int nth_fragment)
2699 {
2700 int match_beginning, match_end;
2701 const char *patch = frag->patch;
2702 int size = frag->size;
2703 char *old, *oldlines;
2704 struct strbuf newlines;
2705 int new_blank_lines_at_end = 0;
2706 int found_new_blank_lines_at_end = 0;
2707 int hunk_linenr = frag->linenr;
2708 unsigned long leading, trailing;
2709 int pos, applied_pos;
2710 struct image preimage;
2711 struct image postimage;
2712
2713 memset(&preimage, 0, sizeof(preimage));
2714 memset(&postimage, 0, sizeof(postimage));
2715 oldlines = xmalloc(size);
2716 strbuf_init(&newlines, size);
2717
2718 old = oldlines;
2719 while (size > 0) {
2720 char first;
2721 int len = linelen(patch, size);
2722 int plen;
2723 int added_blank_line = 0;
2724 int is_blank_context = 0;
2725 size_t start;
2726
2727 if (!len)
2728 break;
2729
2730 /*
2731 * "plen" is how much of the line we should use for
2732 * the actual patch data. Normally we just remove the
2733 * first character on the line, but if the line is
2734 * followed by "\ No newline", then we also remove the
2735 * last one (which is the newline, of course).
2736 */
2737 plen = len - 1;
2738 if (len < size && patch[len] == '\\')
2739 plen--;
2740 first = *patch;
2741 if (state->apply_in_reverse) {
2742 if (first == '-')
2743 first = '+';
2744 else if (first == '+')
2745 first = '-';
2746 }
2747
2748 switch (first) {
2749 case '\n':
2750 /* Newer GNU diff, empty context line */
2751 if (plen < 0)
2752 /* ... followed by '\No newline'; nothing */
2753 break;
2754 *old++ = '\n';
2755 strbuf_addch(&newlines, '\n');
2756 add_line_info(&preimage, "\n", 1, LINE_COMMON);
2757 add_line_info(&postimage, "\n", 1, LINE_COMMON);
2758 is_blank_context = 1;
2759 break;
2760 case ' ':
2761 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&
2762 ws_blank_line(patch + 1, plen, ws_rule))
2763 is_blank_context = 1;
2764 case '-':
2765 memcpy(old, patch + 1, plen);
2766 add_line_info(&preimage, old, plen,
2767 (first == ' ' ? LINE_COMMON : 0));
2768 old += plen;
2769 if (first == '-')
2770 break;
2771 /* Fall-through for ' ' */
2772 case '+':
2773 /* --no-add does not add new lines */
2774 if (first == '+' && state->no_add)
2775 break;
2776
2777 start = newlines.len;
2778 if (first != '+' ||
2779 !state->whitespace_error ||
2780 state->ws_error_action != correct_ws_error) {
2781 strbuf_add(&newlines, patch + 1, plen);
2782 }
2783 else {
2784 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);
2785 }
2786 add_line_info(&postimage, newlines.buf + start, newlines.len - start,
2787 (first == '+' ? 0 : LINE_COMMON));
2788 if (first == '+' &&
2789 (ws_rule & WS_BLANK_AT_EOF) &&
2790 ws_blank_line(patch + 1, plen, ws_rule))
2791 added_blank_line = 1;
2792 break;
2793 case '@': case '\\':
2794 /* Ignore it, we already handled it */
2795 break;
2796 default:
2797 if (state->apply_verbosely)
2798 error(_("invalid start of line: '%c'"), first);
2799 applied_pos = -1;
2800 goto out;
2801 }
2802 if (added_blank_line) {
2803 if (!new_blank_lines_at_end)
2804 found_new_blank_lines_at_end = hunk_linenr;
2805 new_blank_lines_at_end++;
2806 }
2807 else if (is_blank_context)
2808 ;
2809 else
2810 new_blank_lines_at_end = 0;
2811 patch += len;
2812 size -= len;
2813 hunk_linenr++;
2814 }
2815 if (inaccurate_eof &&
2816 old > oldlines && old[-1] == '\n' &&
2817 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {
2818 old--;
2819 strbuf_setlen(&newlines, newlines.len - 1);
2820 }
2821
2822 leading = frag->leading;
2823 trailing = frag->trailing;
2824
2825 /*
2826 * A hunk to change lines at the beginning would begin with
2827 * @@ -1,L +N,M @@
2828 * but we need to be careful. -U0 that inserts before the second
2829 * line also has this pattern.
2830 *
2831 * And a hunk to add to an empty file would begin with
2832 * @@ -0,0 +N,M @@
2833 *
2834 * In other words, a hunk that is (frag->oldpos <= 1) with or
2835 * without leading context must match at the beginning.
2836 */
2837 match_beginning = (!frag->oldpos ||
2838 (frag->oldpos == 1 && !state->unidiff_zero));
2839
2840 /*
2841 * A hunk without trailing lines must match at the end.
2842 * However, we simply cannot tell if a hunk must match end
2843 * from the lack of trailing lines if the patch was generated
2844 * with unidiff without any context.
2845 */
2846 match_end = !state->unidiff_zero && !trailing;
2847
2848 pos = frag->newpos ? (frag->newpos - 1) : 0;
2849 preimage.buf = oldlines;
2850 preimage.len = old - oldlines;
2851 postimage.buf = newlines.buf;
2852 postimage.len = newlines.len;
2853 preimage.line = preimage.line_allocated;
2854 postimage.line = postimage.line_allocated;
2855
2856 for (;;) {
2857
2858 applied_pos = find_pos(state, img, &preimage, &postimage, pos,
2859 ws_rule, match_beginning, match_end);
2860
2861 if (applied_pos >= 0)
2862 break;
2863
2864 /* Am I at my context limits? */
2865 if ((leading <= state->p_context) && (trailing <= state->p_context))
2866 break;
2867 if (match_beginning || match_end) {
2868 match_beginning = match_end = 0;
2869 continue;
2870 }
2871
2872 /*
2873 * Reduce the number of context lines; reduce both
2874 * leading and trailing if they are equal otherwise
2875 * just reduce the larger context.
2876 */
2877 if (leading >= trailing) {
2878 remove_first_line(&preimage);
2879 remove_first_line(&postimage);
2880 pos--;
2881 leading--;
2882 }
2883 if (trailing > leading) {
2884 remove_last_line(&preimage);
2885 remove_last_line(&postimage);
2886 trailing--;
2887 }
2888 }
2889
2890 if (applied_pos >= 0) {
2891 if (new_blank_lines_at_end &&
2892 preimage.nr + applied_pos >= img->nr &&
2893 (ws_rule & WS_BLANK_AT_EOF) &&
2894 state->ws_error_action != nowarn_ws_error) {
2895 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,
2896 found_new_blank_lines_at_end);
2897 if (state->ws_error_action == correct_ws_error) {
2898 while (new_blank_lines_at_end--)
2899 remove_last_line(&postimage);
2900 }
2901 /*
2902 * We would want to prevent write_out_results()
2903 * from taking place in apply_patch() that follows
2904 * the callchain led us here, which is:
2905 * apply_patch->check_patch_list->check_patch->
2906 * apply_data->apply_fragments->apply_one_fragment
2907 */
2908 if (state->ws_error_action == die_on_ws_error)
2909 state->apply = 0;
2910 }
2911
2912 if (state->apply_verbosely && applied_pos != pos) {
2913 int offset = applied_pos - pos;
2914 if (state->apply_in_reverse)
2915 offset = 0 - offset;
2916 fprintf_ln(stderr,
2917 Q_("Hunk #%d succeeded at %d (offset %d line).",
2918 "Hunk #%d succeeded at %d (offset %d lines).",
2919 offset),
2920 nth_fragment, applied_pos + 1, offset);
2921 }
2922
2923 /*
2924 * Warn if it was necessary to reduce the number
2925 * of context lines.
2926 */
2927 if ((leading != frag->leading) ||
2928 (trailing != frag->trailing))
2929 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"
2930 " to apply fragment at %d"),
2931 leading, trailing, applied_pos+1);
2932 update_image(state, img, applied_pos, &preimage, &postimage);
2933 } else {
2934 if (state->apply_verbosely)
2935 error(_("while searching for:\n%.*s"),
2936 (int)(old - oldlines), oldlines);
2937 }
2938
2939 out:
2940 free(oldlines);
2941 strbuf_release(&newlines);
2942 free(preimage.line_allocated);
2943 free(postimage.line_allocated);
2944
2945 return (applied_pos < 0);
2946 }
2947
2948 static int apply_binary_fragment(struct apply_state *state,
2949 struct image *img,
2950 struct patch *patch)
2951 {
2952 struct fragment *fragment = patch->fragments;
2953 unsigned long len;
2954 void *dst;
2955
2956 if (!fragment)
2957 return error(_("missing binary patch data for '%s'"),
2958 patch->new_name ?
2959 patch->new_name :
2960 patch->old_name);
2961
2962 /* Binary patch is irreversible without the optional second hunk */
2963 if (state->apply_in_reverse) {
2964 if (!fragment->next)
2965 return error("cannot reverse-apply a binary patch "
2966 "without the reverse hunk to '%s'",
2967 patch->new_name
2968 ? patch->new_name : patch->old_name);
2969 fragment = fragment->next;
2970 }
2971 switch (fragment->binary_patch_method) {
2972 case BINARY_DELTA_DEFLATED:
2973 dst = patch_delta(img->buf, img->len, fragment->patch,
2974 fragment->size, &len);
2975 if (!dst)
2976 return -1;
2977 clear_image(img);
2978 img->buf = dst;
2979 img->len = len;
2980 return 0;
2981 case BINARY_LITERAL_DEFLATED:
2982 clear_image(img);
2983 img->len = fragment->size;
2984 img->buf = xmemdupz(fragment->patch, img->len);
2985 return 0;
2986 }
2987 return -1;
2988 }
2989
2990 /*
2991 * Replace "img" with the result of applying the binary patch.
2992 * The binary patch data itself in patch->fragment is still kept
2993 * but the preimage prepared by the caller in "img" is freed here
2994 * or in the helper function apply_binary_fragment() this calls.
2995 */
2996 static int apply_binary(struct apply_state *state,
2997 struct image *img,
2998 struct patch *patch)
2999 {
3000 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3001 unsigned char sha1[20];
3002
3003 /*
3004 * For safety, we require patch index line to contain
3005 * full 40-byte textual SHA1 for old and new, at least for now.
3006 */
3007 if (strlen(patch->old_sha1_prefix) != 40 ||
3008 strlen(patch->new_sha1_prefix) != 40 ||
3009 get_sha1_hex(patch->old_sha1_prefix, sha1) ||
3010 get_sha1_hex(patch->new_sha1_prefix, sha1))
3011 return error("cannot apply binary patch to '%s' "
3012 "without full index line", name);
3013
3014 if (patch->old_name) {
3015 /*
3016 * See if the old one matches what the patch
3017 * applies to.
3018 */
3019 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3020 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
3021 return error("the patch applies to '%s' (%s), "
3022 "which does not match the "
3023 "current contents.",
3024 name, sha1_to_hex(sha1));
3025 }
3026 else {
3027 /* Otherwise, the old one must be empty. */
3028 if (img->len)
3029 return error("the patch applies to an empty "
3030 "'%s' but it is not empty", name);
3031 }
3032
3033 get_sha1_hex(patch->new_sha1_prefix, sha1);
3034 if (is_null_sha1(sha1)) {
3035 clear_image(img);
3036 return 0; /* deletion patch */
3037 }
3038
3039 if (has_sha1_file(sha1)) {
3040 /* We already have the postimage */
3041 enum object_type type;
3042 unsigned long size;
3043 char *result;
3044
3045 result = read_sha1_file(sha1, &type, &size);
3046 if (!result)
3047 return error("the necessary postimage %s for "
3048 "'%s' cannot be read",
3049 patch->new_sha1_prefix, name);
3050 clear_image(img);
3051 img->buf = result;
3052 img->len = size;
3053 } else {
3054 /*
3055 * We have verified buf matches the preimage;
3056 * apply the patch data to it, which is stored
3057 * in the patch->fragments->{patch,size}.
3058 */
3059 if (apply_binary_fragment(state, img, patch))
3060 return error(_("binary patch does not apply to '%s'"),
3061 name);
3062
3063 /* verify that the result matches */
3064 hash_sha1_file(img->buf, img->len, blob_type, sha1);
3065 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
3066 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),
3067 name, patch->new_sha1_prefix, sha1_to_hex(sha1));
3068 }
3069
3070 return 0;
3071 }
3072
3073 static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)
3074 {
3075 struct fragment *frag = patch->fragments;
3076 const char *name = patch->old_name ? patch->old_name : patch->new_name;
3077 unsigned ws_rule = patch->ws_rule;
3078 unsigned inaccurate_eof = patch->inaccurate_eof;
3079 int nth = 0;
3080
3081 if (patch->is_binary)
3082 return apply_binary(state, img, patch);
3083
3084 while (frag) {
3085 nth++;
3086 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {
3087 error(_("patch failed: %s:%ld"), name, frag->oldpos);
3088 if (!state->apply_with_reject)
3089 return -1;
3090 frag->rejected = 1;
3091 }
3092 frag = frag->next;
3093 }
3094 return 0;
3095 }
3096
3097 static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)
3098 {
3099 if (S_ISGITLINK(mode)) {
3100 strbuf_grow(buf, 100);
3101 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));
3102 } else {
3103 enum object_type type;
3104 unsigned long sz;
3105 char *result;
3106
3107 result = read_sha1_file(sha1, &type, &sz);
3108 if (!result)
3109 return -1;
3110 /* XXX read_sha1_file NUL-terminates */
3111 strbuf_attach(buf, result, sz, sz + 1);
3112 }
3113 return 0;
3114 }
3115
3116 static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)
3117 {
3118 if (!ce)
3119 return 0;
3120 return read_blob_object(buf, ce->sha1, ce->ce_mode);
3121 }
3122
3123 static struct patch *in_fn_table(struct apply_state *state, const char *name)
3124 {
3125 struct string_list_item *item;
3126
3127 if (name == NULL)
3128 return NULL;
3129
3130 item = string_list_lookup(&state->fn_table, name);
3131 if (item != NULL)
3132 return (struct patch *)item->util;
3133
3134 return NULL;
3135 }
3136
3137 /*
3138 * item->util in the filename table records the status of the path.
3139 * Usually it points at a patch (whose result records the contents
3140 * of it after applying it), but it could be PATH_WAS_DELETED for a
3141 * path that a previously applied patch has already removed, or
3142 * PATH_TO_BE_DELETED for a path that a later patch would remove.
3143 *
3144 * The latter is needed to deal with a case where two paths A and B
3145 * are swapped by first renaming A to B and then renaming B to A;
3146 * moving A to B should not be prevented due to presence of B as we
3147 * will remove it in a later patch.
3148 */
3149 #define PATH_TO_BE_DELETED ((struct patch *) -2)
3150 #define PATH_WAS_DELETED ((struct patch *) -1)
3151
3152 static int to_be_deleted(struct patch *patch)
3153 {
3154 return patch == PATH_TO_BE_DELETED;
3155 }
3156
3157 static int was_deleted(struct patch *patch)
3158 {
3159 return patch == PATH_WAS_DELETED;
3160 }
3161
3162 static void add_to_fn_table(struct apply_state *state, struct patch *patch)
3163 {
3164 struct string_list_item *item;
3165
3166 /*
3167 * Always add new_name unless patch is a deletion
3168 * This should cover the cases for normal diffs,
3169 * file creations and copies
3170 */
3171 if (patch->new_name != NULL) {
3172 item = string_list_insert(&state->fn_table, patch->new_name);
3173 item->util = patch;
3174 }
3175
3176 /*
3177 * store a failure on rename/deletion cases because
3178 * later chunks shouldn't patch old names
3179 */
3180 if ((patch->new_name == NULL) || (patch->is_rename)) {
3181 item = string_list_insert(&state->fn_table, patch->old_name);
3182 item->util = PATH_WAS_DELETED;
3183 }
3184 }
3185
3186 static void prepare_fn_table(struct apply_state *state, struct patch *patch)
3187 {
3188 /*
3189 * store information about incoming file deletion
3190 */
3191 while (patch) {
3192 if ((patch->new_name == NULL) || (patch->is_rename)) {
3193 struct string_list_item *item;
3194 item = string_list_insert(&state->fn_table, patch->old_name);
3195 item->util = PATH_TO_BE_DELETED;
3196 }
3197 patch = patch->next;
3198 }
3199 }
3200
3201 static int checkout_target(struct index_state *istate,
3202 struct cache_entry *ce, struct stat *st)
3203 {
3204 struct checkout costate;
3205
3206 memset(&costate, 0, sizeof(costate));
3207 costate.base_dir = "";
3208 costate.refresh_cache = 1;
3209 costate.istate = istate;
3210 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))
3211 return error(_("cannot checkout %s"), ce->name);
3212 return 0;
3213 }
3214
3215 static struct patch *previous_patch(struct apply_state *state,
3216 struct patch *patch,
3217 int *gone)
3218 {
3219 struct patch *previous;
3220
3221 *gone = 0;
3222 if (patch->is_copy || patch->is_rename)
3223 return NULL; /* "git" patches do not depend on the order */
3224
3225 previous = in_fn_table(state, patch->old_name);
3226 if (!previous)
3227 return NULL;
3228
3229 if (to_be_deleted(previous))
3230 return NULL; /* the deletion hasn't happened yet */
3231
3232 if (was_deleted(previous))
3233 *gone = 1;
3234
3235 return previous;
3236 }
3237
3238 static int verify_index_match(const struct cache_entry *ce, struct stat *st)
3239 {
3240 if (S_ISGITLINK(ce->ce_mode)) {
3241 if (!S_ISDIR(st->st_mode))
3242 return -1;
3243 return 0;
3244 }
3245 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);
3246 }
3247
3248 #define SUBMODULE_PATCH_WITHOUT_INDEX 1
3249
3250 static int load_patch_target(struct apply_state *state,
3251 struct strbuf *buf,
3252 const struct cache_entry *ce,
3253 struct stat *st,
3254 const char *name,
3255 unsigned expected_mode)
3256 {
3257 if (state->cached || state->check_index) {
3258 if (read_file_or_gitlink(ce, buf))
3259 return error(_("failed to read %s"), name);
3260 } else if (name) {
3261 if (S_ISGITLINK(expected_mode)) {
3262 if (ce)
3263 return read_file_or_gitlink(ce, buf);
3264 else
3265 return SUBMODULE_PATCH_WITHOUT_INDEX;
3266 } else if (has_symlink_leading_path(name, strlen(name))) {
3267 return error(_("reading from '%s' beyond a symbolic link"), name);
3268 } else {
3269 if (read_old_data(st, name, buf))
3270 return error(_("failed to read %s"), name);
3271 }
3272 }
3273 return 0;
3274 }
3275
3276 /*
3277 * We are about to apply "patch"; populate the "image" with the
3278 * current version we have, from the working tree or from the index,
3279 * depending on the situation e.g. --cached/--index. If we are
3280 * applying a non-git patch that incrementally updates the tree,
3281 * we read from the result of a previous diff.
3282 */
3283 static int load_preimage(struct apply_state *state,
3284 struct image *image,
3285 struct patch *patch, struct stat *st,
3286 const struct cache_entry *ce)
3287 {
3288 struct strbuf buf = STRBUF_INIT;
3289 size_t len;
3290 char *img;
3291 struct patch *previous;
3292 int status;
3293
3294 previous = previous_patch(state, patch, &status);
3295 if (status)
3296 return error(_("path %s has been renamed/deleted"),
3297 patch->old_name);
3298 if (previous) {
3299 /* We have a patched copy in memory; use that. */
3300 strbuf_add(&buf, previous->result, previous->resultsize);
3301 } else {
3302 status = load_patch_target(state, &buf, ce, st,
3303 patch->old_name, patch->old_mode);
3304 if (status < 0)
3305 return status;
3306 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {
3307 /*
3308 * There is no way to apply subproject
3309 * patch without looking at the index.
3310 * NEEDSWORK: shouldn't this be flagged
3311 * as an error???
3312 */
3313 free_fragment_list(patch->fragments);
3314 patch->fragments = NULL;
3315 } else if (status) {
3316 return error(_("failed to read %s"), patch->old_name);
3317 }
3318 }
3319
3320 img = strbuf_detach(&buf, &len);
3321 prepare_image(image, img, len, !patch->is_binary);
3322 return 0;
3323 }
3324
3325 static int three_way_merge(struct image *image,
3326 char *path,
3327 const unsigned char *base,
3328 const unsigned char *ours,
3329 const unsigned char *theirs)
3330 {
3331 mmfile_t base_file, our_file, their_file;
3332 mmbuffer_t result = { NULL };
3333 int status;
3334
3335 read_mmblob(&base_file, base);
3336 read_mmblob(&our_file, ours);
3337 read_mmblob(&their_file, theirs);
3338 status = ll_merge(&result, path,
3339 &base_file, "base",
3340 &our_file, "ours",
3341 &their_file, "theirs", NULL);
3342 free(base_file.ptr);
3343 free(our_file.ptr);
3344 free(their_file.ptr);
3345 if (status < 0 || !result.ptr) {
3346 free(result.ptr);
3347 return -1;
3348 }
3349 clear_image(image);
3350 image->buf = result.ptr;
3351 image->len = result.size;
3352
3353 return status;
3354 }
3355
3356 /*
3357 * When directly falling back to add/add three-way merge, we read from
3358 * the current contents of the new_name. In no cases other than that
3359 * this function will be called.
3360 */
3361 static int load_current(struct apply_state *state,
3362 struct image *image,
3363 struct patch *patch)
3364 {
3365 struct strbuf buf = STRBUF_INIT;
3366 int status, pos;
3367 size_t len;
3368 char *img;
3369 struct stat st;
3370 struct cache_entry *ce;
3371 char *name = patch->new_name;
3372 unsigned mode = patch->new_mode;
3373
3374 if (!patch->is_new)
3375 die("BUG: patch to %s is not a creation", patch->old_name);
3376
3377 pos = cache_name_pos(name, strlen(name));
3378 if (pos < 0)
3379 return error(_("%s: does not exist in index"), name);
3380 ce = active_cache[pos];
3381 if (lstat(name, &st)) {
3382 if (errno != ENOENT)
3383 return error(_("%s: %s"), name, strerror(errno));
3384 if (checkout_target(&the_index, ce, &st))
3385 return -1;
3386 }
3387 if (verify_index_match(ce, &st))
3388 return error(_("%s: does not match index"), name);
3389
3390 status = load_patch_target(state, &buf, ce, &st, name, mode);
3391 if (status < 0)
3392 return status;
3393 else if (status)
3394 return -1;
3395 img = strbuf_detach(&buf, &len);
3396 prepare_image(image, img, len, !patch->is_binary);
3397 return 0;
3398 }
3399
3400 static int try_threeway(struct apply_state *state,
3401 struct image *image,
3402 struct patch *patch,
3403 struct stat *st,
3404 const struct cache_entry *ce)
3405 {
3406 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];
3407 struct strbuf buf = STRBUF_INIT;
3408 size_t len;
3409 int status;
3410 char *img;
3411 struct image tmp_image;
3412
3413 /* No point falling back to 3-way merge in these cases */
3414 if (patch->is_delete ||
3415 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))
3416 return -1;
3417
3418 /* Preimage the patch was prepared for */
3419 if (patch->is_new)
3420 write_sha1_file("", 0, blob_type, pre_sha1);
3421 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||
3422 read_blob_object(&buf, pre_sha1, patch->old_mode))
3423 return error("repository lacks the necessary blob to fall back on 3-way merge.");
3424
3425 fprintf(stderr, "Falling back to three-way merge...\n");
3426
3427 img = strbuf_detach(&buf, &len);
3428 prepare_image(&tmp_image, img, len, 1);
3429 /* Apply the patch to get the post image */
3430 if (apply_fragments(state, &tmp_image, patch) < 0) {
3431 clear_image(&tmp_image);
3432 return -1;
3433 }
3434 /* post_sha1[] is theirs */
3435 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);
3436 clear_image(&tmp_image);
3437
3438 /* our_sha1[] is ours */
3439 if (patch->is_new) {
3440 if (load_current(state, &tmp_image, patch))
3441 return error("cannot read the current contents of '%s'",
3442 patch->new_name);
3443 } else {
3444 if (load_preimage(state, &tmp_image, patch, st, ce))
3445 return error("cannot read the current contents of '%s'",
3446 patch->old_name);
3447 }
3448 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);
3449 clear_image(&tmp_image);
3450
3451 /* in-core three-way merge between post and our using pre as base */
3452 status = three_way_merge(image, patch->new_name,
3453 pre_sha1, our_sha1, post_sha1);
3454 if (status < 0) {
3455 fprintf(stderr, "Failed to fall back on three-way merge...\n");
3456 return status;
3457 }
3458
3459 if (status) {
3460 patch->conflicted_threeway = 1;
3461 if (patch->is_new)
3462 oidclr(&patch->threeway_stage[0]);
3463 else
3464 hashcpy(patch->threeway_stage[0].hash, pre_sha1);
3465 hashcpy(patch->threeway_stage[1].hash, our_sha1);
3466 hashcpy(patch->threeway_stage[2].hash, post_sha1);
3467 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);
3468 } else {
3469 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);
3470 }
3471 return 0;
3472 }
3473
3474 static int apply_data(struct apply_state *state, struct patch *patch,
3475 struct stat *st, const struct cache_entry *ce)
3476 {
3477 struct image image;
3478
3479 if (load_preimage(state, &image, patch, st, ce) < 0)
3480 return -1;
3481
3482 if (patch->direct_to_threeway ||
3483 apply_fragments(state, &image, patch) < 0) {
3484 /* Note: with --reject, apply_fragments() returns 0 */
3485 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)
3486 return -1;
3487 }
3488 patch->result = image.buf;
3489 patch->resultsize = image.len;
3490 add_to_fn_table(state, patch);
3491 free(image.line_allocated);
3492
3493 if (0 < patch->is_delete && patch->resultsize)
3494 return error(_("removal patch leaves file contents"));
3495
3496 return 0;
3497 }
3498
3499 /*
3500 * If "patch" that we are looking at modifies or deletes what we have,
3501 * we would want it not to lose any local modification we have, either
3502 * in the working tree or in the index.
3503 *
3504 * This also decides if a non-git patch is a creation patch or a
3505 * modification to an existing empty file. We do not check the state
3506 * of the current tree for a creation patch in this function; the caller
3507 * check_patch() separately makes sure (and errors out otherwise) that
3508 * the path the patch creates does not exist in the current tree.
3509 */
3510 static int check_preimage(struct apply_state *state,
3511 struct patch *patch,
3512 struct cache_entry **ce,
3513 struct stat *st)
3514 {
3515 const char *old_name = patch->old_name;
3516 struct patch *previous = NULL;
3517 int stat_ret = 0, status;
3518 unsigned st_mode = 0;
3519
3520 if (!old_name)
3521 return 0;
3522
3523 assert(patch->is_new <= 0);
3524 previous = previous_patch(state, patch, &status);
3525
3526 if (status)
3527 return error(_("path %s has been renamed/deleted"), old_name);
3528 if (previous) {
3529 st_mode = previous->new_mode;
3530 } else if (!state->cached) {
3531 stat_ret = lstat(old_name, st);
3532 if (stat_ret && errno != ENOENT)
3533 return error(_("%s: %s"), old_name, strerror(errno));
3534 }
3535
3536 if (state->check_index && !previous) {
3537 int pos = cache_name_pos(old_name, strlen(old_name));
3538 if (pos < 0) {
3539 if (patch->is_new < 0)
3540 goto is_new;
3541 return error(_("%s: does not exist in index"), old_name);
3542 }
3543 *ce = active_cache[pos];
3544 if (stat_ret < 0) {
3545 if (checkout_target(&the_index, *ce, st))
3546 return -1;
3547 }
3548 if (!state->cached && verify_index_match(*ce, st))
3549 return error(_("%s: does not match index"), old_name);
3550 if (state->cached)
3551 st_mode = (*ce)->ce_mode;
3552 } else if (stat_ret < 0) {
3553 if (patch->is_new < 0)
3554 goto is_new;
3555 return error(_("%s: %s"), old_name, strerror(errno));
3556 }
3557
3558 if (!state->cached && !previous)
3559 st_mode = ce_mode_from_stat(*ce, st->st_mode);
3560
3561 if (patch->is_new < 0)
3562 patch->is_new = 0;
3563 if (!patch->old_mode)
3564 patch->old_mode = st_mode;
3565 if ((st_mode ^ patch->old_mode) & S_IFMT)
3566 return error(_("%s: wrong type"), old_name);
3567 if (st_mode != patch->old_mode)
3568 warning(_("%s has type %o, expected %o"),
3569 old_name, st_mode, patch->old_mode);
3570 if (!patch->new_mode && !patch->is_delete)
3571 patch->new_mode = st_mode;
3572 return 0;
3573
3574 is_new:
3575 patch->is_new = 1;
3576 patch->is_delete = 0;
3577 free(patch->old_name);
3578 patch->old_name = NULL;
3579 return 0;
3580 }
3581
3582
3583 #define EXISTS_IN_INDEX 1
3584 #define EXISTS_IN_WORKTREE 2
3585
3586 static int check_to_create(struct apply_state *state,
3587 const char *new_name,
3588 int ok_if_exists)
3589 {
3590 struct stat nst;
3591
3592 if (state->check_index &&
3593 cache_name_pos(new_name, strlen(new_name)) >= 0 &&
3594 !ok_if_exists)
3595 return EXISTS_IN_INDEX;
3596 if (state->cached)
3597 return 0;
3598
3599 if (!lstat(new_name, &nst)) {
3600 if (S_ISDIR(nst.st_mode) || ok_if_exists)
3601 return 0;
3602 /*
3603 * A leading component of new_name might be a symlink
3604 * that is going to be removed with this patch, but
3605 * still pointing at somewhere that has the path.
3606 * In such a case, path "new_name" does not exist as
3607 * far as git is concerned.
3608 */
3609 if (has_symlink_leading_path(new_name, strlen(new_name)))
3610 return 0;
3611
3612 return EXISTS_IN_WORKTREE;
3613 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {
3614 return error("%s: %s", new_name, strerror(errno));
3615 }
3616 return 0;
3617 }
3618
3619 static uintptr_t register_symlink_changes(struct apply_state *state,
3620 const char *path,
3621 uintptr_t what)
3622 {
3623 struct string_list_item *ent;
3624
3625 ent = string_list_lookup(&state->symlink_changes, path);
3626 if (!ent) {
3627 ent = string_list_insert(&state->symlink_changes, path);
3628 ent->util = (void *)0;
3629 }
3630 ent->util = (void *)(what | ((uintptr_t)ent->util));
3631 return (uintptr_t)ent->util;
3632 }
3633
3634 static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)
3635 {
3636 struct string_list_item *ent;
3637
3638 ent = string_list_lookup(&state->symlink_changes, path);
3639 if (!ent)
3640 return 0;
3641 return (uintptr_t)ent->util;
3642 }
3643
3644 static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)
3645 {
3646 for ( ; patch; patch = patch->next) {
3647 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&
3648 (patch->is_rename || patch->is_delete))
3649 /* the symlink at patch->old_name is removed */
3650 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);
3651
3652 if (patch->new_name && S_ISLNK(patch->new_mode))
3653 /* the symlink at patch->new_name is created or remains */
3654 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);
3655 }
3656 }
3657
3658 static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)
3659 {
3660 do {
3661 unsigned int change;
3662
3663 while (--name->len && name->buf[name->len] != '/')
3664 ; /* scan backwards */
3665 if (!name->len)
3666 break;
3667 name->buf[name->len] = '\0';
3668 change = check_symlink_changes(state, name->buf);
3669 if (change & APPLY_SYMLINK_IN_RESULT)
3670 return 1;
3671 if (change & APPLY_SYMLINK_GOES_AWAY)
3672 /*
3673 * This cannot be "return 0", because we may
3674 * see a new one created at a higher level.
3675 */
3676 continue;
3677
3678 /* otherwise, check the preimage */
3679 if (state->check_index) {
3680 struct cache_entry *ce;
3681
3682 ce = cache_file_exists(name->buf, name->len, ignore_case);
3683 if (ce && S_ISLNK(ce->ce_mode))
3684 return 1;
3685 } else {
3686 struct stat st;
3687 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))
3688 return 1;
3689 }
3690 } while (1);
3691 return 0;
3692 }
3693
3694 static int path_is_beyond_symlink(struct apply_state *state, const char *name_)
3695 {
3696 int ret;
3697 struct strbuf name = STRBUF_INIT;
3698
3699 assert(*name_ != '\0');
3700 strbuf_addstr(&name, name_);
3701 ret = path_is_beyond_symlink_1(state, &name);
3702 strbuf_release(&name);
3703
3704 return ret;
3705 }
3706
3707 static int check_unsafe_path(struct patch *patch)
3708 {
3709 const char *old_name = NULL;
3710 const char *new_name = NULL;
3711 if (patch->is_delete)
3712 old_name = patch->old_name;
3713 else if (!patch->is_new && !patch->is_copy)
3714 old_name = patch->old_name;
3715 if (!patch->is_delete)
3716 new_name = patch->new_name;
3717
3718 if (old_name && !verify_path(old_name))
3719 return error(_("invalid path '%s'"), old_name);
3720 if (new_name && !verify_path(new_name))
3721 return error(_("invalid path '%s'"), new_name);
3722 return 0;
3723 }
3724
3725 /*
3726 * Check and apply the patch in-core; leave the result in patch->result
3727 * for the caller to write it out to the final destination.
3728 */
3729 static int check_patch(struct apply_state *state, struct patch *patch)
3730 {
3731 struct stat st;
3732 const char *old_name = patch->old_name;
3733 const char *new_name = patch->new_name;
3734 const char *name = old_name ? old_name : new_name;
3735 struct cache_entry *ce = NULL;
3736 struct patch *tpatch;
3737 int ok_if_exists;
3738 int status;
3739
3740 patch->rejected = 1; /* we will drop this after we succeed */
3741
3742 status = check_preimage(state, patch, &ce, &st);
3743 if (status)
3744 return status;
3745 old_name = patch->old_name;
3746
3747 /*
3748 * A type-change diff is always split into a patch to delete
3749 * old, immediately followed by a patch to create new (see
3750 * diff.c::run_diff()); in such a case it is Ok that the entry
3751 * to be deleted by the previous patch is still in the working
3752 * tree and in the index.
3753 *
3754 * A pat