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 "cache-tree.h" 11#include "quote.h" 12#include "blob.h" 13#include "delta.h" 14#include "builtin.h" 15#include "string-list.h" 16#include "dir.h" 17#include "parse-options.h" 18 19/* 20 * --check turns on checking that the working tree matches the 21 * files that are being modified, but doesn't apply the patch 22 * --stat does just a diffstat, and doesn't actually apply 23 * --numstat does numeric diffstat, and doesn't actually apply 24 * --index-info shows the old and new index info for paths if available. 25 * --index updates the cache as well. 26 * --cached updates only the cache without ever touching the working tree. 27 */ 28static const char *prefix; 29static int prefix_length = -1; 30static int newfd = -1; 31 32static int unidiff_zero; 33static int p_value = 1; 34static int p_value_known; 35static int check_index; 36static int update_index; 37static int cached; 38static int diffstat; 39static int numstat; 40static int summary; 41static int check; 42static int apply = 1; 43static int apply_in_reverse; 44static int apply_with_reject; 45static int apply_verbosely; 46static int no_add; 47static const char *fake_ancestor; 48static int line_termination = '\n'; 49static unsigned int p_context = UINT_MAX; 50static const char * const apply_usage[] = { 51 "git apply [options] [<patch>...]", 52 NULL 53}; 54 55static enum ws_error_action { 56 nowarn_ws_error, 57 warn_on_ws_error, 58 die_on_ws_error, 59 correct_ws_error, 60} ws_error_action = warn_on_ws_error; 61static int whitespace_error; 62static int squelch_whitespace_errors = 5; 63static int applied_after_fixing_ws; 64 65static enum ws_ignore { 66 ignore_ws_none, 67 ignore_ws_change, 68} ws_ignore_action = ignore_ws_none; 69 70 71static const char *patch_input_file; 72static const char *root; 73static int root_len; 74static int read_stdin = 1; 75static int options; 76 77static void parse_whitespace_option(const char *option) 78{ 79 if (!option) { 80 ws_error_action = warn_on_ws_error; 81 return; 82 } 83 if (!strcmp(option, "warn")) { 84 ws_error_action = warn_on_ws_error; 85 return; 86 } 87 if (!strcmp(option, "nowarn")) { 88 ws_error_action = nowarn_ws_error; 89 return; 90 } 91 if (!strcmp(option, "error")) { 92 ws_error_action = die_on_ws_error; 93 return; 94 } 95 if (!strcmp(option, "error-all")) { 96 ws_error_action = die_on_ws_error; 97 squelch_whitespace_errors = 0; 98 return; 99 } 100 if (!strcmp(option, "strip") || !strcmp(option, "fix")) { 101 ws_error_action = correct_ws_error; 102 return; 103 } 104 die("unrecognized whitespace option '%s'", option); 105} 106 107static void parse_ignorewhitespace_option(const char *option) 108{ 109 if (!option || !strcmp(option, "no") || 110 !strcmp(option, "false") || !strcmp(option, "never") || 111 !strcmp(option, "none")) { 112 ws_ignore_action = ignore_ws_none; 113 return; 114 } 115 if (!strcmp(option, "change")) { 116 ws_ignore_action = ignore_ws_change; 117 return; 118 } 119 die("unrecognized whitespace ignore option '%s'", option); 120} 121 122static void set_default_whitespace_mode(const char *whitespace_option) 123{ 124 if (!whitespace_option && !apply_default_whitespace) 125 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error); 126} 127 128/* 129 * For "diff-stat" like behaviour, we keep track of the biggest change 130 * we've seen, and the longest filename. That allows us to do simple 131 * scaling. 132 */ 133static int max_change, max_len; 134 135/* 136 * Various "current state", notably line numbers and what 137 * file (and how) we're patching right now.. The "is_xxxx" 138 * things are flags, where -1 means "don't know yet". 139 */ 140static int linenr = 1; 141 142/* 143 * This represents one "hunk" from a patch, starting with 144 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The 145 * patch text is pointed at by patch, and its byte length 146 * is stored in size. leading and trailing are the number 147 * of context lines. 148 */ 149struct fragment { 150 unsigned long leading, trailing; 151 unsigned long oldpos, oldlines; 152 unsigned long newpos, newlines; 153 const char *patch; 154 int size; 155 int rejected; 156 int linenr; 157 struct fragment *next; 158}; 159 160/* 161 * When dealing with a binary patch, we reuse "leading" field 162 * to store the type of the binary hunk, either deflated "delta" 163 * or deflated "literal". 164 */ 165#define binary_patch_method leading 166#define BINARY_DELTA_DEFLATED 1 167#define BINARY_LITERAL_DEFLATED 2 168 169/* 170 * This represents a "patch" to a file, both metainfo changes 171 * such as creation/deletion, filemode and content changes represented 172 * as a series of fragments. 173 */ 174struct patch { 175 char *new_name, *old_name, *def_name; 176 unsigned int old_mode, new_mode; 177 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */ 178 int rejected; 179 unsigned ws_rule; 180 unsigned long deflate_origlen; 181 int lines_added, lines_deleted; 182 int score; 183 unsigned int is_toplevel_relative:1; 184 unsigned int inaccurate_eof:1; 185 unsigned int is_binary:1; 186 unsigned int is_copy:1; 187 unsigned int is_rename:1; 188 unsigned int recount:1; 189 struct fragment *fragments; 190 char *result; 191 size_t resultsize; 192 char old_sha1_prefix[41]; 193 char new_sha1_prefix[41]; 194 struct patch *next; 195}; 196 197/* 198 * A line in a file, len-bytes long (includes the terminating LF, 199 * except for an incomplete line at the end if the file ends with 200 * one), and its contents hashes to 'hash'. 201 */ 202struct line { 203 size_t len; 204 unsigned hash : 24; 205 unsigned flag : 8; 206#define LINE_COMMON 1 207}; 208 209/* 210 * This represents a "file", which is an array of "lines". 211 */ 212struct image { 213 char *buf; 214 size_t len; 215 size_t nr; 216 size_t alloc; 217 struct line *line_allocated; 218 struct line *line; 219}; 220 221/* 222 * Records filenames that have been touched, in order to handle 223 * the case where more than one patches touch the same file. 224 */ 225 226static struct string_list fn_table; 227 228static uint32_t hash_line(const char *cp, size_t len) 229{ 230 size_t i; 231 uint32_t h; 232 for (i = 0, h = 0; i < len; i++) { 233 if (!isspace(cp[i])) { 234 h = h * 3 + (cp[i] & 0xff); 235 } 236 } 237 return h; 238} 239 240/* 241 * Compare lines s1 of length n1 and s2 of length n2, ignoring 242 * whitespace difference. Returns 1 if they match, 0 otherwise 243 */ 244static int fuzzy_matchlines(const char *s1, size_t n1, 245 const char *s2, size_t n2) 246{ 247 const char *last1 = s1 + n1 - 1; 248 const char *last2 = s2 + n2 - 1; 249 int result = 0; 250 251 if (n1 < 0 || n2 < 0) 252 return 0; 253 254 /* ignore line endings */ 255 while ((*last1 == '\r') || (*last1 == '\n')) 256 last1--; 257 while ((*last2 == '\r') || (*last2 == '\n')) 258 last2--; 259 260 /* skip leading whitespace */ 261 while (isspace(*s1) && (s1 <= last1)) 262 s1++; 263 while (isspace(*s2) && (s2 <= last2)) 264 s2++; 265 /* early return if both lines are empty */ 266 if ((s1 > last1) && (s2 > last2)) 267 return 1; 268 while (!result) { 269 result = *s1++ - *s2++; 270 /* 271 * Skip whitespace inside. We check for whitespace on 272 * both buffers because we don't want "a b" to match 273 * "ab" 274 */ 275 if (isspace(*s1) && isspace(*s2)) { 276 while (isspace(*s1) && s1 <= last1) 277 s1++; 278 while (isspace(*s2) && s2 <= last2) 279 s2++; 280 } 281 /* 282 * If we reached the end on one side only, 283 * lines don't match 284 */ 285 if ( 286 ((s2 > last2) && (s1 <= last1)) || 287 ((s1 > last1) && (s2 <= last2))) 288 return 0; 289 if ((s1 > last1) && (s2 > last2)) 290 break; 291 } 292 293 return !result; 294} 295 296static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag) 297{ 298 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc); 299 img->line_allocated[img->nr].len = len; 300 img->line_allocated[img->nr].hash = hash_line(bol, len); 301 img->line_allocated[img->nr].flag = flag; 302 img->nr++; 303} 304 305static void prepare_image(struct image *image, char *buf, size_t len, 306 int prepare_linetable) 307{ 308 const char *cp, *ep; 309 310 memset(image, 0, sizeof(*image)); 311 image->buf = buf; 312 image->len = len; 313 314 if (!prepare_linetable) 315 return; 316 317 ep = image->buf + image->len; 318 cp = image->buf; 319 while (cp < ep) { 320 const char *next; 321 for (next = cp; next < ep && *next != '\n'; next++) 322 ; 323 if (next < ep) 324 next++; 325 add_line_info(image, cp, next - cp, 0); 326 cp = next; 327 } 328 image->line = image->line_allocated; 329} 330 331static void clear_image(struct image *image) 332{ 333 free(image->buf); 334 image->buf = NULL; 335 image->len = 0; 336} 337 338static void say_patch_name(FILE *output, const char *pre, 339 struct patch *patch, const char *post) 340{ 341 fputs(pre, output); 342 if (patch->old_name && patch->new_name && 343 strcmp(patch->old_name, patch->new_name)) { 344 quote_c_style(patch->old_name, NULL, output, 0); 345 fputs(" => ", output); 346 quote_c_style(patch->new_name, NULL, output, 0); 347 } else { 348 const char *n = patch->new_name; 349 if (!n) 350 n = patch->old_name; 351 quote_c_style(n, NULL, output, 0); 352 } 353 fputs(post, output); 354} 355 356#define CHUNKSIZE (8192) 357#define SLOP (16) 358 359static void read_patch_file(struct strbuf *sb, int fd) 360{ 361 if (strbuf_read(sb, fd, 0) < 0) 362 die_errno("git apply: failed to read"); 363 364 /* 365 * Make sure that we have some slop in the buffer 366 * so that we can do speculative "memcmp" etc, and 367 * see to it that it is NUL-filled. 368 */ 369 strbuf_grow(sb, SLOP); 370 memset(sb->buf + sb->len, 0, SLOP); 371} 372 373static unsigned long linelen(const char *buffer, unsigned long size) 374{ 375 unsigned long len = 0; 376 while (size--) { 377 len++; 378 if (*buffer++ == '\n') 379 break; 380 } 381 return len; 382} 383 384static int is_dev_null(const char *str) 385{ 386 return !memcmp("/dev/null", str, 9) && isspace(str[9]); 387} 388 389#define TERM_SPACE 1 390#define TERM_TAB 2 391 392static int name_terminate(const char *name, int namelen, int c, int terminate) 393{ 394 if (c == ' ' && !(terminate & TERM_SPACE)) 395 return 0; 396 if (c == '\t' && !(terminate & TERM_TAB)) 397 return 0; 398 399 return 1; 400} 401 402/* remove double slashes to make --index work with such filenames */ 403static char *squash_slash(char *name) 404{ 405 int i = 0, j = 0; 406 407 if (!name) 408 return NULL; 409 410 while (name[i]) { 411 if ((name[j++] = name[i++]) == '/') 412 while (name[i] == '/') 413 i++; 414 } 415 name[j] = '\0'; 416 return name; 417} 418 419static char *find_name(const char *line, char *def, int p_value, int terminate) 420{ 421 int len; 422 const char *start = NULL; 423 424 if (p_value == 0) 425 start = line; 426 427 if (*line == '"') { 428 struct strbuf name = STRBUF_INIT; 429 430 /* 431 * Proposed "new-style" GNU patch/diff format; see 432 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2 433 */ 434 if (!unquote_c_style(&name, line, NULL)) { 435 char *cp; 436 437 for (cp = name.buf; p_value; p_value--) { 438 cp = strchr(cp, '/'); 439 if (!cp) 440 break; 441 cp++; 442 } 443 if (cp) { 444 /* name can later be freed, so we need 445 * to memmove, not just return cp 446 */ 447 strbuf_remove(&name, 0, cp - name.buf); 448 free(def); 449 if (root) 450 strbuf_insert(&name, 0, root, root_len); 451 return squash_slash(strbuf_detach(&name, NULL)); 452 } 453 } 454 strbuf_release(&name); 455 } 456 457 for (;;) { 458 char c = *line; 459 460 if (isspace(c)) { 461 if (c == '\n') 462 break; 463 if (name_terminate(start, line-start, c, terminate)) 464 break; 465 } 466 line++; 467 if (c == '/' && !--p_value) 468 start = line; 469 } 470 if (!start) 471 return squash_slash(def); 472 len = line - start; 473 if (!len) 474 return squash_slash(def); 475 476 /* 477 * Generally we prefer the shorter name, especially 478 * if the other one is just a variation of that with 479 * something else tacked on to the end (ie "file.orig" 480 * or "file~"). 481 */ 482 if (def) { 483 int deflen = strlen(def); 484 if (deflen < len && !strncmp(start, def, deflen)) 485 return squash_slash(def); 486 free(def); 487 } 488 489 if (root) { 490 char *ret = xmalloc(root_len + len + 1); 491 strcpy(ret, root); 492 memcpy(ret + root_len, start, len); 493 ret[root_len + len] = '\0'; 494 return squash_slash(ret); 495 } 496 497 return squash_slash(xmemdupz(start, len)); 498} 499 500static int count_slashes(const char *cp) 501{ 502 int cnt = 0; 503 char ch; 504 505 while ((ch = *cp++)) 506 if (ch == '/') 507 cnt++; 508 return cnt; 509} 510 511/* 512 * Given the string after "--- " or "+++ ", guess the appropriate 513 * p_value for the given patch. 514 */ 515static int guess_p_value(const char *nameline) 516{ 517 char *name, *cp; 518 int val = -1; 519 520 if (is_dev_null(nameline)) 521 return -1; 522 name = find_name(nameline, NULL, 0, TERM_SPACE | TERM_TAB); 523 if (!name) 524 return -1; 525 cp = strchr(name, '/'); 526 if (!cp) 527 val = 0; 528 else if (prefix) { 529 /* 530 * Does it begin with "a/$our-prefix" and such? Then this is 531 * very likely to apply to our directory. 532 */ 533 if (!strncmp(name, prefix, prefix_length)) 534 val = count_slashes(prefix); 535 else { 536 cp++; 537 if (!strncmp(cp, prefix, prefix_length)) 538 val = count_slashes(prefix) + 1; 539 } 540 } 541 free(name); 542 return val; 543} 544 545/* 546 * Does the ---/+++ line has the POSIX timestamp after the last HT? 547 * GNU diff puts epoch there to signal a creation/deletion event. Is 548 * this such a timestamp? 549 */ 550static int has_epoch_timestamp(const char *nameline) 551{ 552 /* 553 * We are only interested in epoch timestamp; any non-zero 554 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 555 * For the same reason, the date must be either 1969-12-31 or 556 * 1970-01-01, and the seconds part must be "00". 557 */ 558 const char stamp_regexp[] = 559 "^(1969-12-31|1970-01-01)" 560 " " 561 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 562 " " 563 "([-+][0-2][0-9][0-5][0-9])\n"; 564 const char *timestamp = NULL, *cp; 565 static regex_t *stamp; 566 regmatch_t m[10]; 567 int zoneoffset; 568 int hourminute; 569 int status; 570 571 for (cp = nameline; *cp != '\n'; cp++) { 572 if (*cp == '\t') 573 timestamp = cp + 1; 574 } 575 if (!timestamp) 576 return 0; 577 if (!stamp) { 578 stamp = xmalloc(sizeof(*stamp)); 579 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 580 warning("Cannot prepare timestamp regexp %s", 581 stamp_regexp); 582 return 0; 583 } 584 } 585 586 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 587 if (status) { 588 if (status != REG_NOMATCH) 589 warning("regexec returned %d for input: %s", 590 status, timestamp); 591 return 0; 592 } 593 594 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10); 595 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 596 if (timestamp[m[3].rm_so] == '-') 597 zoneoffset = -zoneoffset; 598 599 /* 600 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 601 * (west of GMT) or 1970-01-01 (east of GMT) 602 */ 603 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 604 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 605 return 0; 606 607 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 608 strtol(timestamp + 14, NULL, 10) - 609 zoneoffset); 610 611 return ((zoneoffset < 0 && hourminute == 1440) || 612 (0 <= zoneoffset && !hourminute)); 613} 614 615/* 616 * Get the name etc info from the ---/+++ lines of a traditional patch header 617 * 618 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 619 * files, we can happily check the index for a match, but for creating a 620 * new file we should try to match whatever "patch" does. I have no idea. 621 */ 622static void parse_traditional_patch(const char *first, const char *second, struct patch *patch) 623{ 624 char *name; 625 626 first += 4; /* skip "--- " */ 627 second += 4; /* skip "+++ " */ 628 if (!p_value_known) { 629 int p, q; 630 p = guess_p_value(first); 631 q = guess_p_value(second); 632 if (p < 0) p = q; 633 if (0 <= p && p == q) { 634 p_value = p; 635 p_value_known = 1; 636 } 637 } 638 if (is_dev_null(first)) { 639 patch->is_new = 1; 640 patch->is_delete = 0; 641 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB); 642 patch->new_name = name; 643 } else if (is_dev_null(second)) { 644 patch->is_new = 0; 645 patch->is_delete = 1; 646 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB); 647 patch->old_name = name; 648 } else { 649 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB); 650 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB); 651 if (has_epoch_timestamp(first)) { 652 patch->is_new = 1; 653 patch->is_delete = 0; 654 patch->new_name = name; 655 } else if (has_epoch_timestamp(second)) { 656 patch->is_new = 0; 657 patch->is_delete = 1; 658 patch->old_name = name; 659 } else { 660 patch->old_name = patch->new_name = name; 661 } 662 } 663 if (!name) 664 die("unable to find filename in patch at line %d", linenr); 665} 666 667static int gitdiff_hdrend(const char *line, struct patch *patch) 668{ 669 return -1; 670} 671 672/* 673 * We're anal about diff header consistency, to make 674 * sure that we don't end up having strange ambiguous 675 * patches floating around. 676 * 677 * As a result, gitdiff_{old|new}name() will check 678 * their names against any previous information, just 679 * to make sure.. 680 */ 681static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew) 682{ 683 if (!orig_name && !isnull) 684 return find_name(line, NULL, p_value, TERM_TAB); 685 686 if (orig_name) { 687 int len; 688 const char *name; 689 char *another; 690 name = orig_name; 691 len = strlen(name); 692 if (isnull) 693 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr); 694 another = find_name(line, NULL, p_value, TERM_TAB); 695 if (!another || memcmp(another, name, len + 1)) 696 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr); 697 free(another); 698 return orig_name; 699 } 700 else { 701 /* expect "/dev/null" */ 702 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 703 die("git apply: bad git-diff - expected /dev/null on line %d", linenr); 704 return NULL; 705 } 706} 707 708static int gitdiff_oldname(const char *line, struct patch *patch) 709{ 710 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old"); 711 return 0; 712} 713 714static int gitdiff_newname(const char *line, struct patch *patch) 715{ 716 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new"); 717 return 0; 718} 719 720static int gitdiff_oldmode(const char *line, struct patch *patch) 721{ 722 patch->old_mode = strtoul(line, NULL, 8); 723 return 0; 724} 725 726static int gitdiff_newmode(const char *line, struct patch *patch) 727{ 728 patch->new_mode = strtoul(line, NULL, 8); 729 return 0; 730} 731 732static int gitdiff_delete(const char *line, struct patch *patch) 733{ 734 patch->is_delete = 1; 735 patch->old_name = patch->def_name; 736 return gitdiff_oldmode(line, patch); 737} 738 739static int gitdiff_newfile(const char *line, struct patch *patch) 740{ 741 patch->is_new = 1; 742 patch->new_name = patch->def_name; 743 return gitdiff_newmode(line, patch); 744} 745 746static int gitdiff_copysrc(const char *line, struct patch *patch) 747{ 748 patch->is_copy = 1; 749 patch->old_name = find_name(line, NULL, 0, 0); 750 return 0; 751} 752 753static int gitdiff_copydst(const char *line, struct patch *patch) 754{ 755 patch->is_copy = 1; 756 patch->new_name = find_name(line, NULL, 0, 0); 757 return 0; 758} 759 760static int gitdiff_renamesrc(const char *line, struct patch *patch) 761{ 762 patch->is_rename = 1; 763 patch->old_name = find_name(line, NULL, 0, 0); 764 return 0; 765} 766 767static int gitdiff_renamedst(const char *line, struct patch *patch) 768{ 769 patch->is_rename = 1; 770 patch->new_name = find_name(line, NULL, 0, 0); 771 return 0; 772} 773 774static int gitdiff_similarity(const char *line, struct patch *patch) 775{ 776 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 777 patch->score = 0; 778 return 0; 779} 780 781static int gitdiff_dissimilarity(const char *line, struct patch *patch) 782{ 783 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 784 patch->score = 0; 785 return 0; 786} 787 788static int gitdiff_index(const char *line, struct patch *patch) 789{ 790 /* 791 * index line is N hexadecimal, "..", N hexadecimal, 792 * and optional space with octal mode. 793 */ 794 const char *ptr, *eol; 795 int len; 796 797 ptr = strchr(line, '.'); 798 if (!ptr || ptr[1] != '.' || 40 < ptr - line) 799 return 0; 800 len = ptr - line; 801 memcpy(patch->old_sha1_prefix, line, len); 802 patch->old_sha1_prefix[len] = 0; 803 804 line = ptr + 2; 805 ptr = strchr(line, ' '); 806 eol = strchr(line, '\n'); 807 808 if (!ptr || eol < ptr) 809 ptr = eol; 810 len = ptr - line; 811 812 if (40 < len) 813 return 0; 814 memcpy(patch->new_sha1_prefix, line, len); 815 patch->new_sha1_prefix[len] = 0; 816 if (*ptr == ' ') 817 patch->old_mode = strtoul(ptr+1, NULL, 8); 818 return 0; 819} 820 821/* 822 * This is normal for a diff that doesn't change anything: we'll fall through 823 * into the next diff. Tell the parser to break out. 824 */ 825static int gitdiff_unrecognized(const char *line, struct patch *patch) 826{ 827 return -1; 828} 829 830static const char *stop_at_slash(const char *line, int llen) 831{ 832 int nslash = p_value; 833 int i; 834 835 for (i = 0; i < llen; i++) { 836 int ch = line[i]; 837 if (ch == '/' && --nslash <= 0) 838 return &line[i]; 839 } 840 return NULL; 841} 842 843/* 844 * This is to extract the same name that appears on "diff --git" 845 * line. We do not find and return anything if it is a rename 846 * patch, and it is OK because we will find the name elsewhere. 847 * We need to reliably find name only when it is mode-change only, 848 * creation or deletion of an empty file. In any of these cases, 849 * both sides are the same name under a/ and b/ respectively. 850 */ 851static char *git_header_name(char *line, int llen) 852{ 853 const char *name; 854 const char *second = NULL; 855 size_t len; 856 857 line += strlen("diff --git "); 858 llen -= strlen("diff --git "); 859 860 if (*line == '"') { 861 const char *cp; 862 struct strbuf first = STRBUF_INIT; 863 struct strbuf sp = STRBUF_INIT; 864 865 if (unquote_c_style(&first, line, &second)) 866 goto free_and_fail1; 867 868 /* advance to the first slash */ 869 cp = stop_at_slash(first.buf, first.len); 870 /* we do not accept absolute paths */ 871 if (!cp || cp == first.buf) 872 goto free_and_fail1; 873 strbuf_remove(&first, 0, cp + 1 - first.buf); 874 875 /* 876 * second points at one past closing dq of name. 877 * find the second name. 878 */ 879 while ((second < line + llen) && isspace(*second)) 880 second++; 881 882 if (line + llen <= second) 883 goto free_and_fail1; 884 if (*second == '"') { 885 if (unquote_c_style(&sp, second, NULL)) 886 goto free_and_fail1; 887 cp = stop_at_slash(sp.buf, sp.len); 888 if (!cp || cp == sp.buf) 889 goto free_and_fail1; 890 /* They must match, otherwise ignore */ 891 if (strcmp(cp + 1, first.buf)) 892 goto free_and_fail1; 893 strbuf_release(&sp); 894 return strbuf_detach(&first, NULL); 895 } 896 897 /* unquoted second */ 898 cp = stop_at_slash(second, line + llen - second); 899 if (!cp || cp == second) 900 goto free_and_fail1; 901 cp++; 902 if (line + llen - cp != first.len + 1 || 903 memcmp(first.buf, cp, first.len)) 904 goto free_and_fail1; 905 return strbuf_detach(&first, NULL); 906 907 free_and_fail1: 908 strbuf_release(&first); 909 strbuf_release(&sp); 910 return NULL; 911 } 912 913 /* unquoted first name */ 914 name = stop_at_slash(line, llen); 915 if (!name || name == line) 916 return NULL; 917 name++; 918 919 /* 920 * since the first name is unquoted, a dq if exists must be 921 * the beginning of the second name. 922 */ 923 for (second = name; second < line + llen; second++) { 924 if (*second == '"') { 925 struct strbuf sp = STRBUF_INIT; 926 const char *np; 927 928 if (unquote_c_style(&sp, second, NULL)) 929 goto free_and_fail2; 930 931 np = stop_at_slash(sp.buf, sp.len); 932 if (!np || np == sp.buf) 933 goto free_and_fail2; 934 np++; 935 936 len = sp.buf + sp.len - np; 937 if (len < second - name && 938 !strncmp(np, name, len) && 939 isspace(name[len])) { 940 /* Good */ 941 strbuf_remove(&sp, 0, np - sp.buf); 942 return strbuf_detach(&sp, NULL); 943 } 944 945 free_and_fail2: 946 strbuf_release(&sp); 947 return NULL; 948 } 949 } 950 951 /* 952 * Accept a name only if it shows up twice, exactly the same 953 * form. 954 */ 955 for (len = 0 ; ; len++) { 956 switch (name[len]) { 957 default: 958 continue; 959 case '\n': 960 return NULL; 961 case '\t': case ' ': 962 second = name+len; 963 for (;;) { 964 char c = *second++; 965 if (c == '\n') 966 return NULL; 967 if (c == '/') 968 break; 969 } 970 if (second[len] == '\n' && !memcmp(name, second, len)) { 971 return xmemdupz(name, len); 972 } 973 } 974 } 975} 976 977/* Verify that we recognize the lines following a git header */ 978static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch) 979{ 980 unsigned long offset; 981 982 /* A git diff has explicit new/delete information, so we don't guess */ 983 patch->is_new = 0; 984 patch->is_delete = 0; 985 986 /* 987 * Some things may not have the old name in the 988 * rest of the headers anywhere (pure mode changes, 989 * or removing or adding empty files), so we get 990 * the default name from the header. 991 */ 992 patch->def_name = git_header_name(line, len); 993 if (patch->def_name && root) { 994 char *s = xmalloc(root_len + strlen(patch->def_name) + 1); 995 strcpy(s, root); 996 strcpy(s + root_len, patch->def_name); 997 free(patch->def_name); 998 patch->def_name = s; 999 }10001001 line += len;1002 size -= len;1003 linenr++;1004 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {1005 static const struct opentry {1006 const char *str;1007 int (*fn)(const char *, struct patch *);1008 } optable[] = {1009 { "@@ -", gitdiff_hdrend },1010 { "--- ", gitdiff_oldname },1011 { "+++ ", gitdiff_newname },1012 { "old mode ", gitdiff_oldmode },1013 { "new mode ", gitdiff_newmode },1014 { "deleted file mode ", gitdiff_delete },1015 { "new file mode ", gitdiff_newfile },1016 { "copy from ", gitdiff_copysrc },1017 { "copy to ", gitdiff_copydst },1018 { "rename old ", gitdiff_renamesrc },1019 { "rename new ", gitdiff_renamedst },1020 { "rename from ", gitdiff_renamesrc },1021 { "rename to ", gitdiff_renamedst },1022 { "similarity index ", gitdiff_similarity },1023 { "dissimilarity index ", gitdiff_dissimilarity },1024 { "index ", gitdiff_index },1025 { "", gitdiff_unrecognized },1026 };1027 int i;10281029 len = linelen(line, size);1030 if (!len || line[len-1] != '\n')1031 break;1032 for (i = 0; i < ARRAY_SIZE(optable); i++) {1033 const struct opentry *p = optable + i;1034 int oplen = strlen(p->str);1035 if (len < oplen || memcmp(p->str, line, oplen))1036 continue;1037 if (p->fn(line + oplen, patch) < 0)1038 return offset;1039 break;1040 }1041 }10421043 return offset;1044}10451046static int parse_num(const char *line, unsigned long *p)1047{1048 char *ptr;10491050 if (!isdigit(*line))1051 return 0;1052 *p = strtoul(line, &ptr, 10);1053 return ptr - line;1054}10551056static int parse_range(const char *line, int len, int offset, const char *expect,1057 unsigned long *p1, unsigned long *p2)1058{1059 int digits, ex;10601061 if (offset < 0 || offset >= len)1062 return -1;1063 line += offset;1064 len -= offset;10651066 digits = parse_num(line, p1);1067 if (!digits)1068 return -1;10691070 offset += digits;1071 line += digits;1072 len -= digits;10731074 *p2 = 1;1075 if (*line == ',') {1076 digits = parse_num(line+1, p2);1077 if (!digits)1078 return -1;10791080 offset += digits+1;1081 line += digits+1;1082 len -= digits+1;1083 }10841085 ex = strlen(expect);1086 if (ex > len)1087 return -1;1088 if (memcmp(line, expect, ex))1089 return -1;10901091 return offset + ex;1092}10931094static void recount_diff(char *line, int size, struct fragment *fragment)1095{1096 int oldlines = 0, newlines = 0, ret = 0;10971098 if (size < 1) {1099 warning("recount: ignore empty hunk");1100 return;1101 }11021103 for (;;) {1104 int len = linelen(line, size);1105 size -= len;1106 line += len;11071108 if (size < 1)1109 break;11101111 switch (*line) {1112 case ' ': case '\n':1113 newlines++;1114 /* fall through */1115 case '-':1116 oldlines++;1117 continue;1118 case '+':1119 newlines++;1120 continue;1121 case '\\':1122 continue;1123 case '@':1124 ret = size < 3 || prefixcmp(line, "@@ ");1125 break;1126 case 'd':1127 ret = size < 5 || prefixcmp(line, "diff ");1128 break;1129 default:1130 ret = -1;1131 break;1132 }1133 if (ret) {1134 warning("recount: unexpected line: %.*s",1135 (int)linelen(line, size), line);1136 return;1137 }1138 break;1139 }1140 fragment->oldlines = oldlines;1141 fragment->newlines = newlines;1142}11431144/*1145 * Parse a unified diff fragment header of the1146 * form "@@ -a,b +c,d @@"1147 */1148static int parse_fragment_header(char *line, int len, struct fragment *fragment)1149{1150 int offset;11511152 if (!len || line[len-1] != '\n')1153 return -1;11541155 /* Figure out the number of lines in a fragment */1156 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1157 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);11581159 return offset;1160}11611162static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)1163{1164 unsigned long offset, len;11651166 patch->is_toplevel_relative = 0;1167 patch->is_rename = patch->is_copy = 0;1168 patch->is_new = patch->is_delete = -1;1169 patch->old_mode = patch->new_mode = 0;1170 patch->old_name = patch->new_name = NULL;1171 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {1172 unsigned long nextlen;11731174 len = linelen(line, size);1175 if (!len)1176 break;11771178 /* Testing this early allows us to take a few shortcuts.. */1179 if (len < 6)1180 continue;11811182 /*1183 * Make sure we don't find any unconnected patch fragments.1184 * That's a sign that we didn't find a header, and that a1185 * patch has become corrupted/broken up.1186 */1187 if (!memcmp("@@ -", line, 4)) {1188 struct fragment dummy;1189 if (parse_fragment_header(line, len, &dummy) < 0)1190 continue;1191 die("patch fragment without header at line %d: %.*s",1192 linenr, (int)len-1, line);1193 }11941195 if (size < len + 6)1196 break;11971198 /*1199 * Git patch? It might not have a real patch, just a rename1200 * or mode change, so we handle that specially1201 */1202 if (!memcmp("diff --git ", line, 11)) {1203 int git_hdr_len = parse_git_header(line, len, size, patch);1204 if (git_hdr_len <= len)1205 continue;1206 if (!patch->old_name && !patch->new_name) {1207 if (!patch->def_name)1208 die("git diff header lacks filename information when removing "1209 "%d leading pathname components (line %d)" , p_value, linenr);1210 patch->old_name = patch->new_name = patch->def_name;1211 }1212 patch->is_toplevel_relative = 1;1213 *hdrsize = git_hdr_len;1214 return offset;1215 }12161217 /* --- followed by +++ ? */1218 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1219 continue;12201221 /*1222 * We only accept unified patches, so we want it to1223 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1224 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1225 */1226 nextlen = linelen(line + len, size - len);1227 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1228 continue;12291230 /* Ok, we'll consider it a patch */1231 parse_traditional_patch(line, line+len, patch);1232 *hdrsize = len + nextlen;1233 linenr += 2;1234 return offset;1235 }1236 return -1;1237}12381239static void record_ws_error(unsigned result, const char *line, int len, int linenr)1240{1241 char *err;12421243 if (!result)1244 return;12451246 whitespace_error++;1247 if (squelch_whitespace_errors &&1248 squelch_whitespace_errors < whitespace_error)1249 return;12501251 err = whitespace_error_string(result);1252 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1253 patch_input_file, linenr, err, len, line);1254 free(err);1255}12561257static void check_whitespace(const char *line, int len, unsigned ws_rule)1258{1259 unsigned result = ws_check(line + 1, len - 1, ws_rule);12601261 record_ws_error(result, line + 1, len - 2, linenr);1262}12631264/*1265 * Parse a unified diff. Note that this really needs to parse each1266 * fragment separately, since the only way to know the difference1267 * between a "---" that is part of a patch, and a "---" that starts1268 * the next patch is to look at the line counts..1269 */1270static int parse_fragment(char *line, unsigned long size,1271 struct patch *patch, struct fragment *fragment)1272{1273 int added, deleted;1274 int len = linelen(line, size), offset;1275 unsigned long oldlines, newlines;1276 unsigned long leading, trailing;12771278 offset = parse_fragment_header(line, len, fragment);1279 if (offset < 0)1280 return -1;1281 if (offset > 0 && patch->recount)1282 recount_diff(line + offset, size - offset, fragment);1283 oldlines = fragment->oldlines;1284 newlines = fragment->newlines;1285 leading = 0;1286 trailing = 0;12871288 /* Parse the thing.. */1289 line += len;1290 size -= len;1291 linenr++;1292 added = deleted = 0;1293 for (offset = len;1294 0 < size;1295 offset += len, size -= len, line += len, linenr++) {1296 if (!oldlines && !newlines)1297 break;1298 len = linelen(line, size);1299 if (!len || line[len-1] != '\n')1300 return -1;1301 switch (*line) {1302 default:1303 return -1;1304 case '\n': /* newer GNU diff, an empty context line */1305 case ' ':1306 oldlines--;1307 newlines--;1308 if (!deleted && !added)1309 leading++;1310 trailing++;1311 break;1312 case '-':1313 if (apply_in_reverse &&1314 ws_error_action != nowarn_ws_error)1315 check_whitespace(line, len, patch->ws_rule);1316 deleted++;1317 oldlines--;1318 trailing = 0;1319 break;1320 case '+':1321 if (!apply_in_reverse &&1322 ws_error_action != nowarn_ws_error)1323 check_whitespace(line, len, patch->ws_rule);1324 added++;1325 newlines--;1326 trailing = 0;1327 break;13281329 /*1330 * We allow "\ No newline at end of file". Depending1331 * on locale settings when the patch was produced we1332 * don't know what this line looks like. The only1333 * thing we do know is that it begins with "\ ".1334 * Checking for 12 is just for sanity check -- any1335 * l10n of "\ No newline..." is at least that long.1336 */1337 case '\\':1338 if (len < 12 || memcmp(line, "\\ ", 2))1339 return -1;1340 break;1341 }1342 }1343 if (oldlines || newlines)1344 return -1;1345 fragment->leading = leading;1346 fragment->trailing = trailing;13471348 /*1349 * If a fragment ends with an incomplete line, we failed to include1350 * it in the above loop because we hit oldlines == newlines == 01351 * before seeing it.1352 */1353 if (12 < size && !memcmp(line, "\\ ", 2))1354 offset += linelen(line, size);13551356 patch->lines_added += added;1357 patch->lines_deleted += deleted;13581359 if (0 < patch->is_new && oldlines)1360 return error("new file depends on old contents");1361 if (0 < patch->is_delete && newlines)1362 return error("deleted file still has contents");1363 return offset;1364}13651366static int parse_single_patch(char *line, unsigned long size, struct patch *patch)1367{1368 unsigned long offset = 0;1369 unsigned long oldlines = 0, newlines = 0, context = 0;1370 struct fragment **fragp = &patch->fragments;13711372 while (size > 4 && !memcmp(line, "@@ -", 4)) {1373 struct fragment *fragment;1374 int len;13751376 fragment = xcalloc(1, sizeof(*fragment));1377 fragment->linenr = linenr;1378 len = parse_fragment(line, size, patch, fragment);1379 if (len <= 0)1380 die("corrupt patch at line %d", linenr);1381 fragment->patch = line;1382 fragment->size = len;1383 oldlines += fragment->oldlines;1384 newlines += fragment->newlines;1385 context += fragment->leading + fragment->trailing;13861387 *fragp = fragment;1388 fragp = &fragment->next;13891390 offset += len;1391 line += len;1392 size -= len;1393 }13941395 /*1396 * If something was removed (i.e. we have old-lines) it cannot1397 * be creation, and if something was added it cannot be1398 * deletion. However, the reverse is not true; --unified=01399 * patches that only add are not necessarily creation even1400 * though they do not have any old lines, and ones that only1401 * delete are not necessarily deletion.1402 *1403 * Unfortunately, a real creation/deletion patch do _not_ have1404 * any context line by definition, so we cannot safely tell it1405 * apart with --unified=0 insanity. At least if the patch has1406 * more than one hunk it is not creation or deletion.1407 */1408 if (patch->is_new < 0 &&1409 (oldlines || (patch->fragments && patch->fragments->next)))1410 patch->is_new = 0;1411 if (patch->is_delete < 0 &&1412 (newlines || (patch->fragments && patch->fragments->next)))1413 patch->is_delete = 0;14141415 if (0 < patch->is_new && oldlines)1416 die("new file %s depends on old contents", patch->new_name);1417 if (0 < patch->is_delete && newlines)1418 die("deleted file %s still has contents", patch->old_name);1419 if (!patch->is_delete && !newlines && context)1420 fprintf(stderr, "** warning: file %s becomes empty but "1421 "is not deleted\n", patch->new_name);14221423 return offset;1424}14251426static inline int metadata_changes(struct patch *patch)1427{1428 return patch->is_rename > 0 ||1429 patch->is_copy > 0 ||1430 patch->is_new > 0 ||1431 patch->is_delete ||1432 (patch->old_mode && patch->new_mode &&1433 patch->old_mode != patch->new_mode);1434}14351436static char *inflate_it(const void *data, unsigned long size,1437 unsigned long inflated_size)1438{1439 z_stream stream;1440 void *out;1441 int st;14421443 memset(&stream, 0, sizeof(stream));14441445 stream.next_in = (unsigned char *)data;1446 stream.avail_in = size;1447 stream.next_out = out = xmalloc(inflated_size);1448 stream.avail_out = inflated_size;1449 git_inflate_init(&stream);1450 st = git_inflate(&stream, Z_FINISH);1451 git_inflate_end(&stream);1452 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1453 free(out);1454 return NULL;1455 }1456 return out;1457}14581459static struct fragment *parse_binary_hunk(char **buf_p,1460 unsigned long *sz_p,1461 int *status_p,1462 int *used_p)1463{1464 /*1465 * Expect a line that begins with binary patch method ("literal"1466 * or "delta"), followed by the length of data before deflating.1467 * a sequence of 'length-byte' followed by base-85 encoded data1468 * should follow, terminated by a newline.1469 *1470 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1471 * and we would limit the patch line to 66 characters,1472 * so one line can fit up to 13 groups that would decode1473 * to 52 bytes max. The length byte 'A'-'Z' corresponds1474 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1475 */1476 int llen, used;1477 unsigned long size = *sz_p;1478 char *buffer = *buf_p;1479 int patch_method;1480 unsigned long origlen;1481 char *data = NULL;1482 int hunk_size = 0;1483 struct fragment *frag;14841485 llen = linelen(buffer, size);1486 used = llen;14871488 *status_p = 0;14891490 if (!prefixcmp(buffer, "delta ")) {1491 patch_method = BINARY_DELTA_DEFLATED;1492 origlen = strtoul(buffer + 6, NULL, 10);1493 }1494 else if (!prefixcmp(buffer, "literal ")) {1495 patch_method = BINARY_LITERAL_DEFLATED;1496 origlen = strtoul(buffer + 8, NULL, 10);1497 }1498 else1499 return NULL;15001501 linenr++;1502 buffer += llen;1503 while (1) {1504 int byte_length, max_byte_length, newsize;1505 llen = linelen(buffer, size);1506 used += llen;1507 linenr++;1508 if (llen == 1) {1509 /* consume the blank line */1510 buffer++;1511 size--;1512 break;1513 }1514 /*1515 * Minimum line is "A00000\n" which is 7-byte long,1516 * and the line length must be multiple of 5 plus 2.1517 */1518 if ((llen < 7) || (llen-2) % 5)1519 goto corrupt;1520 max_byte_length = (llen - 2) / 5 * 4;1521 byte_length = *buffer;1522 if ('A' <= byte_length && byte_length <= 'Z')1523 byte_length = byte_length - 'A' + 1;1524 else if ('a' <= byte_length && byte_length <= 'z')1525 byte_length = byte_length - 'a' + 27;1526 else1527 goto corrupt;1528 /* if the input length was not multiple of 4, we would1529 * have filler at the end but the filler should never1530 * exceed 3 bytes1531 */1532 if (max_byte_length < byte_length ||1533 byte_length <= max_byte_length - 4)1534 goto corrupt;1535 newsize = hunk_size + byte_length;1536 data = xrealloc(data, newsize);1537 if (decode_85(data + hunk_size, buffer + 1, byte_length))1538 goto corrupt;1539 hunk_size = newsize;1540 buffer += llen;1541 size -= llen;1542 }15431544 frag = xcalloc(1, sizeof(*frag));1545 frag->patch = inflate_it(data, hunk_size, origlen);1546 if (!frag->patch)1547 goto corrupt;1548 free(data);1549 frag->size = origlen;1550 *buf_p = buffer;1551 *sz_p = size;1552 *used_p = used;1553 frag->binary_patch_method = patch_method;1554 return frag;15551556 corrupt:1557 free(data);1558 *status_p = -1;1559 error("corrupt binary patch at line %d: %.*s",1560 linenr-1, llen-1, buffer);1561 return NULL;1562}15631564static int parse_binary(char *buffer, unsigned long size, struct patch *patch)1565{1566 /*1567 * We have read "GIT binary patch\n"; what follows is a line1568 * that says the patch method (currently, either "literal" or1569 * "delta") and the length of data before deflating; a1570 * sequence of 'length-byte' followed by base-85 encoded data1571 * follows.1572 *1573 * When a binary patch is reversible, there is another binary1574 * hunk in the same format, starting with patch method (either1575 * "literal" or "delta") with the length of data, and a sequence1576 * of length-byte + base-85 encoded data, terminated with another1577 * empty line. This data, when applied to the postimage, produces1578 * the preimage.1579 */1580 struct fragment *forward;1581 struct fragment *reverse;1582 int status;1583 int used, used_1;15841585 forward = parse_binary_hunk(&buffer, &size, &status, &used);1586 if (!forward && !status)1587 /* there has to be one hunk (forward hunk) */1588 return error("unrecognized binary patch at line %d", linenr-1);1589 if (status)1590 /* otherwise we already gave an error message */1591 return status;15921593 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);1594 if (reverse)1595 used += used_1;1596 else if (status) {1597 /*1598 * Not having reverse hunk is not an error, but having1599 * a corrupt reverse hunk is.1600 */1601 free((void*) forward->patch);1602 free(forward);1603 return status;1604 }1605 forward->next = reverse;1606 patch->fragments = forward;1607 patch->is_binary = 1;1608 return used;1609}16101611static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)1612{1613 int hdrsize, patchsize;1614 int offset = find_header(buffer, size, &hdrsize, patch);16151616 if (offset < 0)1617 return offset;16181619 patch->ws_rule = whitespace_rule(patch->new_name1620 ? patch->new_name1621 : patch->old_name);16221623 patchsize = parse_single_patch(buffer + offset + hdrsize,1624 size - offset - hdrsize, patch);16251626 if (!patchsize) {1627 static const char *binhdr[] = {1628 "Binary files ",1629 "Files ",1630 NULL,1631 };1632 static const char git_binary[] = "GIT binary patch\n";1633 int i;1634 int hd = hdrsize + offset;1635 unsigned long llen = linelen(buffer + hd, size - hd);16361637 if (llen == sizeof(git_binary) - 1 &&1638 !memcmp(git_binary, buffer + hd, llen)) {1639 int used;1640 linenr++;1641 used = parse_binary(buffer + hd + llen,1642 size - hd - llen, patch);1643 if (used)1644 patchsize = used + llen;1645 else1646 patchsize = 0;1647 }1648 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {1649 for (i = 0; binhdr[i]; i++) {1650 int len = strlen(binhdr[i]);1651 if (len < size - hd &&1652 !memcmp(binhdr[i], buffer + hd, len)) {1653 linenr++;1654 patch->is_binary = 1;1655 patchsize = llen;1656 break;1657 }1658 }1659 }16601661 /* Empty patch cannot be applied if it is a text patch1662 * without metadata change. A binary patch appears1663 * empty to us here.1664 */1665 if ((apply || check) &&1666 (!patch->is_binary && !metadata_changes(patch)))1667 die("patch with only garbage at line %d", linenr);1668 }16691670 return offset + hdrsize + patchsize;1671}16721673#define swap(a,b) myswap((a),(b),sizeof(a))16741675#define myswap(a, b, size) do { \1676 unsigned char mytmp[size]; \1677 memcpy(mytmp, &a, size); \1678 memcpy(&a, &b, size); \1679 memcpy(&b, mytmp, size); \1680} while (0)16811682static void reverse_patches(struct patch *p)1683{1684 for (; p; p = p->next) {1685 struct fragment *frag = p->fragments;16861687 swap(p->new_name, p->old_name);1688 swap(p->new_mode, p->old_mode);1689 swap(p->is_new, p->is_delete);1690 swap(p->lines_added, p->lines_deleted);1691 swap(p->old_sha1_prefix, p->new_sha1_prefix);16921693 for (; frag; frag = frag->next) {1694 swap(frag->newpos, frag->oldpos);1695 swap(frag->newlines, frag->oldlines);1696 }1697 }1698}16991700static const char pluses[] =1701"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";1702static const char minuses[]=1703"----------------------------------------------------------------------";17041705static void show_stats(struct patch *patch)1706{1707 struct strbuf qname = STRBUF_INIT;1708 char *cp = patch->new_name ? patch->new_name : patch->old_name;1709 int max, add, del;17101711 quote_c_style(cp, &qname, NULL, 0);17121713 /*1714 * "scale" the filename1715 */1716 max = max_len;1717 if (max > 50)1718 max = 50;17191720 if (qname.len > max) {1721 cp = strchr(qname.buf + qname.len + 3 - max, '/');1722 if (!cp)1723 cp = qname.buf + qname.len + 3 - max;1724 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);1725 }17261727 if (patch->is_binary) {1728 printf(" %-*s | Bin\n", max, qname.buf);1729 strbuf_release(&qname);1730 return;1731 }17321733 printf(" %-*s |", max, qname.buf);1734 strbuf_release(&qname);17351736 /*1737 * scale the add/delete1738 */1739 max = max + max_change > 70 ? 70 - max : max_change;1740 add = patch->lines_added;1741 del = patch->lines_deleted;17421743 if (max_change > 0) {1744 int total = ((add + del) * max + max_change / 2) / max_change;1745 add = (add * max + max_change / 2) / max_change;1746 del = total - add;1747 }1748 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,1749 add, pluses, del, minuses);1750}17511752static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)1753{1754 switch (st->st_mode & S_IFMT) {1755 case S_IFLNK:1756 if (strbuf_readlink(buf, path, st->st_size) < 0)1757 return error("unable to read symlink %s", path);1758 return 0;1759 case S_IFREG:1760 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)1761 return error("unable to open or read %s", path);1762 convert_to_git(path, buf->buf, buf->len, buf, 0);1763 return 0;1764 default:1765 return -1;1766 }1767}17681769/*1770 * Update the preimage, and the common lines in postimage,1771 * from buffer buf of length len. If postlen is 0 the postimage1772 * is updated in place, otherwise it's updated on a new buffer1773 * of length postlen1774 */17751776static void update_pre_post_images(struct image *preimage,1777 struct image *postimage,1778 char *buf,1779 size_t len, size_t postlen)1780{1781 int i, ctx;1782 char *new, *old, *fixed;1783 struct image fixed_preimage;17841785 /*1786 * Update the preimage with whitespace fixes. Note that we1787 * are not losing preimage->buf -- apply_one_fragment() will1788 * free "oldlines".1789 */1790 prepare_image(&fixed_preimage, buf, len, 1);1791 assert(fixed_preimage.nr == preimage->nr);1792 for (i = 0; i < preimage->nr; i++)1793 fixed_preimage.line[i].flag = preimage->line[i].flag;1794 free(preimage->line_allocated);1795 *preimage = fixed_preimage;17961797 /*1798 * Adjust the common context lines in postimage. This can be1799 * done in-place when we are just doing whitespace fixing,1800 * which does not make the string grow, but needs a new buffer1801 * when ignoring whitespace causes the update, since in this case1802 * we could have e.g. tabs converted to multiple spaces.1803 * We trust the caller to tell us if the update can be done1804 * in place (postlen==0) or not.1805 */1806 old = postimage->buf;1807 if (postlen)1808 new = postimage->buf = xmalloc(postlen);1809 else1810 new = old;1811 fixed = preimage->buf;1812 for (i = ctx = 0; i < postimage->nr; i++) {1813 size_t len = postimage->line[i].len;1814 if (!(postimage->line[i].flag & LINE_COMMON)) {1815 /* an added line -- no counterparts in preimage */1816 memmove(new, old, len);1817 old += len;1818 new += len;1819 continue;1820 }18211822 /* a common context -- skip it in the original postimage */1823 old += len;18241825 /* and find the corresponding one in the fixed preimage */1826 while (ctx < preimage->nr &&1827 !(preimage->line[ctx].flag & LINE_COMMON)) {1828 fixed += preimage->line[ctx].len;1829 ctx++;1830 }1831 if (preimage->nr <= ctx)1832 die("oops");18331834 /* and copy it in, while fixing the line length */1835 len = preimage->line[ctx].len;1836 memcpy(new, fixed, len);1837 new += len;1838 fixed += len;1839 postimage->line[i].len = len;1840 ctx++;1841 }18421843 /* Fix the length of the whole thing */1844 postimage->len = new - postimage->buf;1845}18461847static int match_fragment(struct image *img,1848 struct image *preimage,1849 struct image *postimage,1850 unsigned long try,1851 int try_lno,1852 unsigned ws_rule,1853 int match_beginning, int match_end)1854{1855 int i;1856 char *fixed_buf, *buf, *orig, *target;1857 int preimage_limit;18581859 if (preimage->nr + try_lno <= img->nr) {1860 /*1861 * The hunk falls within the boundaries of img.1862 */1863 preimage_limit = preimage->nr;1864 if (match_end && (preimage->nr + try_lno != img->nr))1865 return 0;1866 } else if (ws_error_action == correct_ws_error &&1867 (ws_rule & WS_BLANK_AT_EOF) && match_end) {1868 /*1869 * This hunk that matches at the end extends beyond1870 * the end of img, and we are removing blank lines1871 * at the end of the file. This many lines from the1872 * beginning of the preimage must match with img, and1873 * the remainder of the preimage must be blank.1874 */1875 preimage_limit = img->nr - try_lno;1876 } else {1877 /*1878 * The hunk extends beyond the end of the img and1879 * we are not removing blanks at the end, so we1880 * should reject the hunk at this position.1881 */1882 return 0;1883 }18841885 if (match_beginning && try_lno)1886 return 0;18871888 /* Quick hash check */1889 for (i = 0; i < preimage_limit; i++)1890 if (preimage->line[i].hash != img->line[try_lno + i].hash)1891 return 0;18921893 if (preimage_limit == preimage->nr) {1894 /*1895 * Do we have an exact match? If we were told to match1896 * at the end, size must be exactly at try+fragsize,1897 * otherwise try+fragsize must be still within the preimage,1898 * and either case, the old piece should match the preimage1899 * exactly.1900 */1901 if ((match_end1902 ? (try + preimage->len == img->len)1903 : (try + preimage->len <= img->len)) &&1904 !memcmp(img->buf + try, preimage->buf, preimage->len))1905 return 1;1906 } else {1907 /*1908 * The preimage extends beyond the end of img, so1909 * there cannot be an exact match.1910 *1911 * There must be one non-blank context line that match1912 * a line before the end of img.1913 */1914 char *buf_end;19151916 buf = preimage->buf;1917 buf_end = buf;1918 for (i = 0; i < preimage_limit; i++)1919 buf_end += preimage->line[i].len;19201921 for ( ; buf < buf_end; buf++)1922 if (!isspace(*buf))1923 break;1924 if (buf == buf_end)1925 return 0;1926 }19271928 /*1929 * No exact match. If we are ignoring whitespace, run a line-by-line1930 * fuzzy matching. We collect all the line length information because1931 * we need it to adjust whitespace if we match.1932 */1933 if (ws_ignore_action == ignore_ws_change) {1934 size_t imgoff = 0;1935 size_t preoff = 0;1936 size_t postlen = postimage->len;1937 size_t extra_chars;1938 char *preimage_eof;1939 char *preimage_end;1940 for (i = 0; i < preimage_limit; i++) {1941 size_t prelen = preimage->line[i].len;1942 size_t imglen = img->line[try_lno+i].len;19431944 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,1945 preimage->buf + preoff, prelen))1946 return 0;1947 if (preimage->line[i].flag & LINE_COMMON)1948 postlen += imglen - prelen;1949 imgoff += imglen;1950 preoff += prelen;1951 }19521953 /*1954 * Ok, the preimage matches with whitespace fuzz.1955 *1956 * imgoff now holds the true length of the target that1957 * matches the preimage before the end of the file.1958 *1959 * Count the number of characters in the preimage that fall1960 * beyond the end of the file and make sure that all of them1961 * are whitespace characters. (This can only happen if1962 * we are removing blank lines at the end of the file.)1963 */1964 buf = preimage_eof = preimage->buf + preoff;1965 for ( ; i < preimage->nr; i++)1966 preoff += preimage->line[i].len;1967 preimage_end = preimage->buf + preoff;1968 for ( ; buf < preimage_end; buf++)1969 if (!isspace(*buf))1970 return 0;19711972 /*1973 * Update the preimage and the common postimage context1974 * lines to use the same whitespace as the target.1975 * If whitespace is missing in the target (i.e.1976 * if the preimage extends beyond the end of the file),1977 * use the whitespace from the preimage.1978 */1979 extra_chars = preimage_end - preimage_eof;1980 fixed_buf = xmalloc(imgoff + extra_chars);1981 memcpy(fixed_buf, img->buf + try, imgoff);1982 memcpy(fixed_buf + imgoff, preimage_eof, extra_chars);1983 imgoff += extra_chars;1984 update_pre_post_images(preimage, postimage,1985 fixed_buf, imgoff, postlen);1986 return 1;1987 }19881989 if (ws_error_action != correct_ws_error)1990 return 0;19911992 /*1993 * The hunk does not apply byte-by-byte, but the hash says1994 * it might with whitespace fuzz. We haven't been asked to1995 * ignore whitespace, we were asked to correct whitespace1996 * errors, so let's try matching after whitespace correction.1997 *1998 * The preimage may extend beyond the end of the file,1999 * but in this loop we will only handle the part of the2000 * preimage that falls within the file.2001 */2002 fixed_buf = xmalloc(preimage->len + 1);2003 buf = fixed_buf;2004 orig = preimage->buf;2005 target = img->buf + try;2006 for (i = 0; i < preimage_limit; i++) {2007 size_t fixlen; /* length after fixing the preimage */2008 size_t oldlen = preimage->line[i].len;2009 size_t tgtlen = img->line[try_lno + i].len;2010 size_t tgtfixlen; /* length after fixing the target line */2011 char tgtfixbuf[1024], *tgtfix;2012 int match;20132014 /* Try fixing the line in the preimage */2015 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);20162017 /* Try fixing the line in the target */2018 if (sizeof(tgtfixbuf) > tgtlen)2019 tgtfix = tgtfixbuf;2020 else2021 tgtfix = xmalloc(tgtlen);2022 tgtfixlen = ws_fix_copy(tgtfix, target, tgtlen, ws_rule, NULL);20232024 /*2025 * If they match, either the preimage was based on2026 * a version before our tree fixed whitespace breakage,2027 * or we are lacking a whitespace-fix patch the tree2028 * the preimage was based on already had (i.e. target2029 * has whitespace breakage, the preimage doesn't).2030 * In either case, we are fixing the whitespace breakages2031 * so we might as well take the fix together with their2032 * real change.2033 */2034 match = (tgtfixlen == fixlen && !memcmp(tgtfix, buf, fixlen));20352036 if (tgtfix != tgtfixbuf)2037 free(tgtfix);2038 if (!match)2039 goto unmatch_exit;20402041 orig += oldlen;2042 buf += fixlen;2043 target += tgtlen;2044 }204520462047 /*2048 * Now handle the lines in the preimage that falls beyond the2049 * end of the file (if any). They will only match if they are2050 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2051 * false).2052 */2053 for ( ; i < preimage->nr; i++) {2054 size_t fixlen; /* length after fixing the preimage */2055 size_t oldlen = preimage->line[i].len;2056 int j;20572058 /* Try fixing the line in the preimage */2059 fixlen = ws_fix_copy(buf, orig, oldlen, ws_rule, NULL);20602061 for (j = 0; j < fixlen; j++)2062 if (!isspace(buf[j]))2063 goto unmatch_exit;20642065 orig += oldlen;2066 buf += fixlen;2067 }20682069 /*2070 * Yes, the preimage is based on an older version that still2071 * has whitespace breakages unfixed, and fixing them makes the2072 * hunk match. Update the context lines in the postimage.2073 */2074 update_pre_post_images(preimage, postimage,2075 fixed_buf, buf - fixed_buf, 0);2076 return 1;20772078 unmatch_exit:2079 free(fixed_buf);2080 return 0;2081}20822083static int find_pos(struct image *img,2084 struct image *preimage,2085 struct image *postimage,2086 int line,2087 unsigned ws_rule,2088 int match_beginning, int match_end)2089{2090 int i;2091 unsigned long backwards, forwards, try;2092 int backwards_lno, forwards_lno, try_lno;20932094 /*2095 * If match_beginning or match_end is specified, there is no2096 * point starting from a wrong line that will never match and2097 * wander around and wait for a match at the specified end.2098 */2099 if (match_beginning)2100 line = 0;2101 else if (match_end)2102 line = img->nr - preimage->nr;21032104 /*2105 * Because the comparison is unsigned, the following test2106 * will also take care of a negative line number that can2107 * result when match_end and preimage is larger than the target.2108 */2109 if ((size_t) line > img->nr)2110 line = img->nr;21112112 try = 0;2113 for (i = 0; i < line; i++)2114 try += img->line[i].len;21152116 /*2117 * There's probably some smart way to do this, but I'll leave2118 * that to the smart and beautiful people. I'm simple and stupid.2119 */2120 backwards = try;2121 backwards_lno = line;2122 forwards = try;2123 forwards_lno = line;2124 try_lno = line;21252126 for (i = 0; ; i++) {2127 if (match_fragment(img, preimage, postimage,2128 try, try_lno, ws_rule,2129 match_beginning, match_end))2130 return try_lno;21312132 again:2133 if (backwards_lno == 0 && forwards_lno == img->nr)2134 break;21352136 if (i & 1) {2137 if (backwards_lno == 0) {2138 i++;2139 goto again;2140 }2141 backwards_lno--;2142 backwards -= img->line[backwards_lno].len;2143 try = backwards;2144 try_lno = backwards_lno;2145 } else {2146 if (forwards_lno == img->nr) {2147 i++;2148 goto again;2149 }2150 forwards += img->line[forwards_lno].len;2151 forwards_lno++;2152 try = forwards;2153 try_lno = forwards_lno;2154 }21552156 }2157 return -1;2158}21592160static void remove_first_line(struct image *img)2161{2162 img->buf += img->line[0].len;2163 img->len -= img->line[0].len;2164 img->line++;2165 img->nr--;2166}21672168static void remove_last_line(struct image *img)2169{2170 img->len -= img->line[--img->nr].len;2171}21722173static void update_image(struct image *img,2174 int applied_pos,2175 struct image *preimage,2176 struct image *postimage)2177{2178 /*2179 * remove the copy of preimage at offset in img2180 * and replace it with postimage2181 */2182 int i, nr;2183 size_t remove_count, insert_count, applied_at = 0;2184 char *result;2185 int preimage_limit;21862187 /*2188 * If we are removing blank lines at the end of img,2189 * the preimage may extend beyond the end.2190 * If that is the case, we must be careful only to2191 * remove the part of the preimage that falls within2192 * the boundaries of img. Initialize preimage_limit2193 * to the number of lines in the preimage that falls2194 * within the boundaries.2195 */2196 preimage_limit = preimage->nr;2197 if (preimage_limit > img->nr - applied_pos)2198 preimage_limit = img->nr - applied_pos;21992200 for (i = 0; i < applied_pos; i++)2201 applied_at += img->line[i].len;22022203 remove_count = 0;2204 for (i = 0; i < preimage_limit; i++)2205 remove_count += img->line[applied_pos + i].len;2206 insert_count = postimage->len;22072208 /* Adjust the contents */2209 result = xmalloc(img->len + insert_count - remove_count + 1);2210 memcpy(result, img->buf, applied_at);2211 memcpy(result + applied_at, postimage->buf, postimage->len);2212 memcpy(result + applied_at + postimage->len,2213 img->buf + (applied_at + remove_count),2214 img->len - (applied_at + remove_count));2215 free(img->buf);2216 img->buf = result;2217 img->len += insert_count - remove_count;2218 result[img->len] = '\0';22192220 /* Adjust the line table */2221 nr = img->nr + postimage->nr - preimage_limit;2222 if (preimage_limit < postimage->nr) {2223 /*2224 * NOTE: this knows that we never call remove_first_line()2225 * on anything other than pre/post image.2226 */2227 img->line = xrealloc(img->line, nr * sizeof(*img->line));2228 img->line_allocated = img->line;2229 }2230 if (preimage_limit != postimage->nr)2231 memmove(img->line + applied_pos + postimage->nr,2232 img->line + applied_pos + preimage_limit,2233 (img->nr - (applied_pos + preimage_limit)) *2234 sizeof(*img->line));2235 memcpy(img->line + applied_pos,2236 postimage->line,2237 postimage->nr * sizeof(*img->line));2238 img->nr = nr;2239}22402241static int apply_one_fragment(struct image *img, struct fragment *frag,2242 int inaccurate_eof, unsigned ws_rule)2243{2244 int match_beginning, match_end;2245 const char *patch = frag->patch;2246 int size = frag->size;2247 char *old, *new, *oldlines, *newlines;2248 int new_blank_lines_at_end = 0;2249 unsigned long leading, trailing;2250 int pos, applied_pos;2251 struct image preimage;2252 struct image postimage;22532254 memset(&preimage, 0, sizeof(preimage));2255 memset(&postimage, 0, sizeof(postimage));2256 oldlines = xmalloc(size);2257 newlines = xmalloc(size);22582259 old = oldlines;2260 new = newlines;2261 while (size > 0) {2262 char first;2263 int len = linelen(patch, size);2264 int plen, added;2265 int added_blank_line = 0;2266 int is_blank_context = 0;22672268 if (!len)2269 break;22702271 /*2272 * "plen" is how much of the line we should use for2273 * the actual patch data. Normally we just remove the2274 * first character on the line, but if the line is2275 * followed by "\ No newline", then we also remove the2276 * last one (which is the newline, of course).2277 */2278 plen = len - 1;2279 if (len < size && patch[len] == '\\')2280 plen--;2281 first = *patch;2282 if (apply_in_reverse) {2283 if (first == '-')2284 first = '+';2285 else if (first == '+')2286 first = '-';2287 }22882289 switch (first) {2290 case '\n':2291 /* Newer GNU diff, empty context line */2292 if (plen < 0)2293 /* ... followed by '\No newline'; nothing */2294 break;2295 *old++ = '\n';2296 *new++ = '\n';2297 add_line_info(&preimage, "\n", 1, LINE_COMMON);2298 add_line_info(&postimage, "\n", 1, LINE_COMMON);2299 is_blank_context = 1;2300 break;2301 case ' ':2302 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2303 ws_blank_line(patch + 1, plen, ws_rule))2304 is_blank_context = 1;2305 case '-':2306 memcpy(old, patch + 1, plen);2307 add_line_info(&preimage, old, plen,2308 (first == ' ' ? LINE_COMMON : 0));2309 old += plen;2310 if (first == '-')2311 break;2312 /* Fall-through for ' ' */2313 case '+':2314 /* --no-add does not add new lines */2315 if (first == '+' && no_add)2316 break;23172318 if (first != '+' ||2319 !whitespace_error ||2320 ws_error_action != correct_ws_error) {2321 memcpy(new, patch + 1, plen);2322 added = plen;2323 }2324 else {2325 added = ws_fix_copy(new, patch + 1, plen, ws_rule, &applied_after_fixing_ws);2326 }2327 add_line_info(&postimage, new, added,2328 (first == '+' ? 0 : LINE_COMMON));2329 new += added;2330 if (first == '+' &&2331 (ws_rule & WS_BLANK_AT_EOF) &&2332 ws_blank_line(patch + 1, plen, ws_rule))2333 added_blank_line = 1;2334 break;2335 case '@': case '\\':2336 /* Ignore it, we already handled it */2337 break;2338 default:2339 if (apply_verbosely)2340 error("invalid start of line: '%c'", first);2341 return -1;2342 }2343 if (added_blank_line)2344 new_blank_lines_at_end++;2345 else if (is_blank_context)2346 ;2347 else2348 new_blank_lines_at_end = 0;2349 patch += len;2350 size -= len;2351 }2352 if (inaccurate_eof &&2353 old > oldlines && old[-1] == '\n' &&2354 new > newlines && new[-1] == '\n') {2355 old--;2356 new--;2357 }23582359 leading = frag->leading;2360 trailing = frag->trailing;23612362 /*2363 * A hunk to change lines at the beginning would begin with2364 * @@ -1,L +N,M @@2365 * but we need to be careful. -U0 that inserts before the second2366 * line also has this pattern.2367 *2368 * And a hunk to add to an empty file would begin with2369 * @@ -0,0 +N,M @@2370 *2371 * In other words, a hunk that is (frag->oldpos <= 1) with or2372 * without leading context must match at the beginning.2373 */2374 match_beginning = (!frag->oldpos ||2375 (frag->oldpos == 1 && !unidiff_zero));23762377 /*2378 * A hunk without trailing lines must match at the end.2379 * However, we simply cannot tell if a hunk must match end2380 * from the lack of trailing lines if the patch was generated2381 * with unidiff without any context.2382 */2383 match_end = !unidiff_zero && !trailing;23842385 pos = frag->newpos ? (frag->newpos - 1) : 0;2386 preimage.buf = oldlines;2387 preimage.len = old - oldlines;2388 postimage.buf = newlines;2389 postimage.len = new - newlines;2390 preimage.line = preimage.line_allocated;2391 postimage.line = postimage.line_allocated;23922393 for (;;) {23942395 applied_pos = find_pos(img, &preimage, &postimage, pos,2396 ws_rule, match_beginning, match_end);23972398 if (applied_pos >= 0)2399 break;24002401 /* Am I at my context limits? */2402 if ((leading <= p_context) && (trailing <= p_context))2403 break;2404 if (match_beginning || match_end) {2405 match_beginning = match_end = 0;2406 continue;2407 }24082409 /*2410 * Reduce the number of context lines; reduce both2411 * leading and trailing if they are equal otherwise2412 * just reduce the larger context.2413 */2414 if (leading >= trailing) {2415 remove_first_line(&preimage);2416 remove_first_line(&postimage);2417 pos--;2418 leading--;2419 }2420 if (trailing > leading) {2421 remove_last_line(&preimage);2422 remove_last_line(&postimage);2423 trailing--;2424 }2425 }24262427 if (applied_pos >= 0) {2428 if (new_blank_lines_at_end &&2429 preimage.nr + applied_pos >= img->nr &&2430 (ws_rule & WS_BLANK_AT_EOF) &&2431 ws_error_action != nowarn_ws_error) {2432 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);2433 if (ws_error_action == correct_ws_error) {2434 while (new_blank_lines_at_end--)2435 remove_last_line(&postimage);2436 }2437 /*2438 * We would want to prevent write_out_results()2439 * from taking place in apply_patch() that follows2440 * the callchain led us here, which is:2441 * apply_patch->check_patch_list->check_patch->2442 * apply_data->apply_fragments->apply_one_fragment2443 */2444 if (ws_error_action == die_on_ws_error)2445 apply = 0;2446 }24472448 /*2449 * Warn if it was necessary to reduce the number2450 * of context lines.2451 */2452 if ((leading != frag->leading) ||2453 (trailing != frag->trailing))2454 fprintf(stderr, "Context reduced to (%ld/%ld)"2455 " to apply fragment at %d\n",2456 leading, trailing, applied_pos+1);2457 update_image(img, applied_pos, &preimage, &postimage);2458 } else {2459 if (apply_verbosely)2460 error("while searching for:\n%.*s",2461 (int)(old - oldlines), oldlines);2462 }24632464 free(oldlines);2465 free(newlines);2466 free(preimage.line_allocated);2467 free(postimage.line_allocated);24682469 return (applied_pos < 0);2470}24712472static int apply_binary_fragment(struct image *img, struct patch *patch)2473{2474 struct fragment *fragment = patch->fragments;2475 unsigned long len;2476 void *dst;24772478 /* Binary patch is irreversible without the optional second hunk */2479 if (apply_in_reverse) {2480 if (!fragment->next)2481 return error("cannot reverse-apply a binary patch "2482 "without the reverse hunk to '%s'",2483 patch->new_name2484 ? patch->new_name : patch->old_name);2485 fragment = fragment->next;2486 }2487 switch (fragment->binary_patch_method) {2488 case BINARY_DELTA_DEFLATED:2489 dst = patch_delta(img->buf, img->len, fragment->patch,2490 fragment->size, &len);2491 if (!dst)2492 return -1;2493 clear_image(img);2494 img->buf = dst;2495 img->len = len;2496 return 0;2497 case BINARY_LITERAL_DEFLATED:2498 clear_image(img);2499 img->len = fragment->size;2500 img->buf = xmalloc(img->len+1);2501 memcpy(img->buf, fragment->patch, img->len);2502 img->buf[img->len] = '\0';2503 return 0;2504 }2505 return -1;2506}25072508static int apply_binary(struct image *img, struct patch *patch)2509{2510 const char *name = patch->old_name ? patch->old_name : patch->new_name;2511 unsigned char sha1[20];25122513 /*2514 * For safety, we require patch index line to contain2515 * full 40-byte textual SHA1 for old and new, at least for now.2516 */2517 if (strlen(patch->old_sha1_prefix) != 40 ||2518 strlen(patch->new_sha1_prefix) != 40 ||2519 get_sha1_hex(patch->old_sha1_prefix, sha1) ||2520 get_sha1_hex(patch->new_sha1_prefix, sha1))2521 return error("cannot apply binary patch to '%s' "2522 "without full index line", name);25232524 if (patch->old_name) {2525 /*2526 * See if the old one matches what the patch2527 * applies to.2528 */2529 hash_sha1_file(img->buf, img->len, blob_type, sha1);2530 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))2531 return error("the patch applies to '%s' (%s), "2532 "which does not match the "2533 "current contents.",2534 name, sha1_to_hex(sha1));2535 }2536 else {2537 /* Otherwise, the old one must be empty. */2538 if (img->len)2539 return error("the patch applies to an empty "2540 "'%s' but it is not empty", name);2541 }25422543 get_sha1_hex(patch->new_sha1_prefix, sha1);2544 if (is_null_sha1(sha1)) {2545 clear_image(img);2546 return 0; /* deletion patch */2547 }25482549 if (has_sha1_file(sha1)) {2550 /* We already have the postimage */2551 enum object_type type;2552 unsigned long size;2553 char *result;25542555 result = read_sha1_file(sha1, &type, &size);2556 if (!result)2557 return error("the necessary postimage %s for "2558 "'%s' cannot be read",2559 patch->new_sha1_prefix, name);2560 clear_image(img);2561 img->buf = result;2562 img->len = size;2563 } else {2564 /*2565 * We have verified buf matches the preimage;2566 * apply the patch data to it, which is stored2567 * in the patch->fragments->{patch,size}.2568 */2569 if (apply_binary_fragment(img, patch))2570 return error("binary patch does not apply to '%s'",2571 name);25722573 /* verify that the result matches */2574 hash_sha1_file(img->buf, img->len, blob_type, sha1);2575 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))2576 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",2577 name, patch->new_sha1_prefix, sha1_to_hex(sha1));2578 }25792580 return 0;2581}25822583static int apply_fragments(struct image *img, struct patch *patch)2584{2585 struct fragment *frag = patch->fragments;2586 const char *name = patch->old_name ? patch->old_name : patch->new_name;2587 unsigned ws_rule = patch->ws_rule;2588 unsigned inaccurate_eof = patch->inaccurate_eof;25892590 if (patch->is_binary)2591 return apply_binary(img, patch);25922593 while (frag) {2594 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {2595 error("patch failed: %s:%ld", name, frag->oldpos);2596 if (!apply_with_reject)2597 return -1;2598 frag->rejected = 1;2599 }2600 frag = frag->next;2601 }2602 return 0;2603}26042605static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)2606{2607 if (!ce)2608 return 0;26092610 if (S_ISGITLINK(ce->ce_mode)) {2611 strbuf_grow(buf, 100);2612 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));2613 } else {2614 enum object_type type;2615 unsigned long sz;2616 char *result;26172618 result = read_sha1_file(ce->sha1, &type, &sz);2619 if (!result)2620 return -1;2621 /* XXX read_sha1_file NUL-terminates */2622 strbuf_attach(buf, result, sz, sz + 1);2623 }2624 return 0;2625}26262627static struct patch *in_fn_table(const char *name)2628{2629 struct string_list_item *item;26302631 if (name == NULL)2632 return NULL;26332634 item = string_list_lookup(name, &fn_table);2635 if (item != NULL)2636 return (struct patch *)item->util;26372638 return NULL;2639}26402641/*2642 * item->util in the filename table records the status of the path.2643 * Usually it points at a patch (whose result records the contents2644 * of it after applying it), but it could be PATH_WAS_DELETED for a2645 * path that a previously applied patch has already removed.2646 */2647 #define PATH_TO_BE_DELETED ((struct patch *) -2)2648#define PATH_WAS_DELETED ((struct patch *) -1)26492650static int to_be_deleted(struct patch *patch)2651{2652 return patch == PATH_TO_BE_DELETED;2653}26542655static int was_deleted(struct patch *patch)2656{2657 return patch == PATH_WAS_DELETED;2658}26592660static void add_to_fn_table(struct patch *patch)2661{2662 struct string_list_item *item;26632664 /*2665 * Always add new_name unless patch is a deletion2666 * This should cover the cases for normal diffs,2667 * file creations and copies2668 */2669 if (patch->new_name != NULL) {2670 item = string_list_insert(patch->new_name, &fn_table);2671 item->util = patch;2672 }26732674 /*2675 * store a failure on rename/deletion cases because2676 * later chunks shouldn't patch old names2677 */2678 if ((patch->new_name == NULL) || (patch->is_rename)) {2679 item = string_list_insert(patch->old_name, &fn_table);2680 item->util = PATH_WAS_DELETED;2681 }2682}26832684static void prepare_fn_table(struct patch *patch)2685{2686 /*2687 * store information about incoming file deletion2688 */2689 while (patch) {2690 if ((patch->new_name == NULL) || (patch->is_rename)) {2691 struct string_list_item *item;2692 item = string_list_insert(patch->old_name, &fn_table);2693 item->util = PATH_TO_BE_DELETED;2694 }2695 patch = patch->next;2696 }2697}26982699static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)2700{2701 struct strbuf buf = STRBUF_INIT;2702 struct image image;2703 size_t len;2704 char *img;2705 struct patch *tpatch;27062707 if (!(patch->is_copy || patch->is_rename) &&2708 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {2709 if (was_deleted(tpatch)) {2710 return error("patch %s has been renamed/deleted",2711 patch->old_name);2712 }2713 /* We have a patched copy in memory use that */2714 strbuf_add(&buf, tpatch->result, tpatch->resultsize);2715 } else if (cached) {2716 if (read_file_or_gitlink(ce, &buf))2717 return error("read of %s failed", patch->old_name);2718 } else if (patch->old_name) {2719 if (S_ISGITLINK(patch->old_mode)) {2720 if (ce) {2721 read_file_or_gitlink(ce, &buf);2722 } else {2723 /*2724 * There is no way to apply subproject2725 * patch without looking at the index.2726 */2727 patch->fragments = NULL;2728 }2729 } else {2730 if (read_old_data(st, patch->old_name, &buf))2731 return error("read of %s failed", patch->old_name);2732 }2733 }27342735 img = strbuf_detach(&buf, &len);2736 prepare_image(&image, img, len, !patch->is_binary);27372738 if (apply_fragments(&image, patch) < 0)2739 return -1; /* note with --reject this succeeds. */2740 patch->result = image.buf;2741 patch->resultsize = image.len;2742 add_to_fn_table(patch);2743 free(image.line_allocated);27442745 if (0 < patch->is_delete && patch->resultsize)2746 return error("removal patch leaves file contents");27472748 return 0;2749}27502751static int check_to_create_blob(const char *new_name, int ok_if_exists)2752{2753 struct stat nst;2754 if (!lstat(new_name, &nst)) {2755 if (S_ISDIR(nst.st_mode) || ok_if_exists)2756 return 0;2757 /*2758 * A leading component of new_name might be a symlink2759 * that is going to be removed with this patch, but2760 * still pointing at somewhere that has the path.2761 * In such a case, path "new_name" does not exist as2762 * far as git is concerned.2763 */2764 if (has_symlink_leading_path(new_name, strlen(new_name)))2765 return 0;27662767 return error("%s: already exists in working directory", new_name);2768 }2769 else if ((errno != ENOENT) && (errno != ENOTDIR))2770 return error("%s: %s", new_name, strerror(errno));2771 return 0;2772}27732774static int verify_index_match(struct cache_entry *ce, struct stat *st)2775{2776 if (S_ISGITLINK(ce->ce_mode)) {2777 if (!S_ISDIR(st->st_mode))2778 return -1;2779 return 0;2780 }2781 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);2782}27832784static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)2785{2786 const char *old_name = patch->old_name;2787 struct patch *tpatch = NULL;2788 int stat_ret = 0;2789 unsigned st_mode = 0;27902791 /*2792 * Make sure that we do not have local modifications from the2793 * index when we are looking at the index. Also make sure2794 * we have the preimage file to be patched in the work tree,2795 * unless --cached, which tells git to apply only in the index.2796 */2797 if (!old_name)2798 return 0;27992800 assert(patch->is_new <= 0);28012802 if (!(patch->is_copy || patch->is_rename) &&2803 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {2804 if (was_deleted(tpatch))2805 return error("%s: has been deleted/renamed", old_name);2806 st_mode = tpatch->new_mode;2807 } else if (!cached) {2808 stat_ret = lstat(old_name, st);2809 if (stat_ret && errno != ENOENT)2810 return error("%s: %s", old_name, strerror(errno));2811 }28122813 if (to_be_deleted(tpatch))2814 tpatch = NULL;28152816 if (check_index && !tpatch) {2817 int pos = cache_name_pos(old_name, strlen(old_name));2818 if (pos < 0) {2819 if (patch->is_new < 0)2820 goto is_new;2821 return error("%s: does not exist in index", old_name);2822 }2823 *ce = active_cache[pos];2824 if (stat_ret < 0) {2825 struct checkout costate;2826 /* checkout */2827 costate.base_dir = "";2828 costate.base_dir_len = 0;2829 costate.force = 0;2830 costate.quiet = 0;2831 costate.not_new = 0;2832 costate.refresh_cache = 1;2833 if (checkout_entry(*ce, &costate, NULL) ||2834 lstat(old_name, st))2835 return -1;2836 }2837 if (!cached && verify_index_match(*ce, st))2838 return error("%s: does not match index", old_name);2839 if (cached)2840 st_mode = (*ce)->ce_mode;2841 } else if (stat_ret < 0) {2842 if (patch->is_new < 0)2843 goto is_new;2844 return error("%s: %s", old_name, strerror(errno));2845 }28462847 if (!cached && !tpatch)2848 st_mode = ce_mode_from_stat(*ce, st->st_mode);28492850 if (patch->is_new < 0)2851 patch->is_new = 0;2852 if (!patch->old_mode)2853 patch->old_mode = st_mode;2854 if ((st_mode ^ patch->old_mode) & S_IFMT)2855 return error("%s: wrong type", old_name);2856 if (st_mode != patch->old_mode)2857 warning("%s has type %o, expected %o",2858 old_name, st_mode, patch->old_mode);2859 if (!patch->new_mode && !patch->is_delete)2860 patch->new_mode = st_mode;2861 return 0;28622863 is_new:2864 patch->is_new = 1;2865 patch->is_delete = 0;2866 patch->old_name = NULL;2867 return 0;2868}28692870static int check_patch(struct patch *patch)2871{2872 struct stat st;2873 const char *old_name = patch->old_name;2874 const char *new_name = patch->new_name;2875 const char *name = old_name ? old_name : new_name;2876 struct cache_entry *ce = NULL;2877 struct patch *tpatch;2878 int ok_if_exists;2879 int status;28802881 patch->rejected = 1; /* we will drop this after we succeed */28822883 status = check_preimage(patch, &ce, &st);2884 if (status)2885 return status;2886 old_name = patch->old_name;28872888 if ((tpatch = in_fn_table(new_name)) &&2889 (was_deleted(tpatch) || to_be_deleted(tpatch)))2890 /*2891 * A type-change diff is always split into a patch to2892 * delete old, immediately followed by a patch to2893 * create new (see diff.c::run_diff()); in such a case2894 * it is Ok that the entry to be deleted by the2895 * previous patch is still in the working tree and in2896 * the index.2897 */2898 ok_if_exists = 1;2899 else2900 ok_if_exists = 0;29012902 if (new_name &&2903 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {2904 if (check_index &&2905 cache_name_pos(new_name, strlen(new_name)) >= 0 &&2906 !ok_if_exists)2907 return error("%s: already exists in index", new_name);2908 if (!cached) {2909 int err = check_to_create_blob(new_name, ok_if_exists);2910 if (err)2911 return err;2912 }2913 if (!patch->new_mode) {2914 if (0 < patch->is_new)2915 patch->new_mode = S_IFREG | 0644;2916 else2917 patch->new_mode = patch->old_mode;2918 }2919 }29202921 if (new_name && old_name) {2922 int same = !strcmp(old_name, new_name);2923 if (!patch->new_mode)2924 patch->new_mode = patch->old_mode;2925 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)2926 return error("new mode (%o) of %s does not match old mode (%o)%s%s",2927 patch->new_mode, new_name, patch->old_mode,2928 same ? "" : " of ", same ? "" : old_name);2929 }29302931 if (apply_data(patch, &st, ce) < 0)2932 return error("%s: patch does not apply", name);2933 patch->rejected = 0;2934 return 0;2935}29362937static int check_patch_list(struct patch *patch)2938{2939 int err = 0;29402941 prepare_fn_table(patch);2942 while (patch) {2943 if (apply_verbosely)2944 say_patch_name(stderr,2945 "Checking patch ", patch, "...\n");2946 err |= check_patch(patch);2947 patch = patch->next;2948 }2949 return err;2950}29512952/* This function tries to read the sha1 from the current index */2953static int get_current_sha1(const char *path, unsigned char *sha1)2954{2955 int pos;29562957 if (read_cache() < 0)2958 return -1;2959 pos = cache_name_pos(path, strlen(path));2960 if (pos < 0)2961 return -1;2962 hashcpy(sha1, active_cache[pos]->sha1);2963 return 0;2964}29652966/* Build an index that contains the just the files needed for a 3way merge */2967static void build_fake_ancestor(struct patch *list, const char *filename)2968{2969 struct patch *patch;2970 struct index_state result = { NULL };2971 int fd;29722973 /* Once we start supporting the reverse patch, it may be2974 * worth showing the new sha1 prefix, but until then...2975 */2976 for (patch = list; patch; patch = patch->next) {2977 const unsigned char *sha1_ptr;2978 unsigned char sha1[20];2979 struct cache_entry *ce;2980 const char *name;29812982 name = patch->old_name ? patch->old_name : patch->new_name;2983 if (0 < patch->is_new)2984 continue;2985 else if (get_sha1(patch->old_sha1_prefix, sha1))2986 /* git diff has no index line for mode/type changes */2987 if (!patch->lines_added && !patch->lines_deleted) {2988 if (get_current_sha1(patch->new_name, sha1) ||2989 get_current_sha1(patch->old_name, sha1))2990 die("mode change for %s, which is not "2991 "in current HEAD", name);2992 sha1_ptr = sha1;2993 } else2994 die("sha1 information is lacking or useless "2995 "(%s).", name);2996 else2997 sha1_ptr = sha1;29982999 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);3000 if (!ce)3001 die("make_cache_entry failed for path '%s'", name);3002 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3003 die ("Could not add %s to temporary index", name);3004 }30053006 fd = open(filename, O_WRONLY | O_CREAT, 0666);3007 if (fd < 0 || write_index(&result, fd) || close(fd))3008 die ("Could not write temporary index to %s", filename);30093010 discard_index(&result);3011}30123013static void stat_patch_list(struct patch *patch)3014{3015 int files, adds, dels;30163017 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3018 files++;3019 adds += patch->lines_added;3020 dels += patch->lines_deleted;3021 show_stats(patch);3022 }30233024 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);3025}30263027static void numstat_patch_list(struct patch *patch)3028{3029 for ( ; patch; patch = patch->next) {3030 const char *name;3031 name = patch->new_name ? patch->new_name : patch->old_name;3032 if (patch->is_binary)3033 printf("-\t-\t");3034 else3035 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3036 write_name_quoted(name, stdout, line_termination);3037 }3038}30393040static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3041{3042 if (mode)3043 printf(" %s mode %06o %s\n", newdelete, mode, name);3044 else3045 printf(" %s %s\n", newdelete, name);3046}30473048static void show_mode_change(struct patch *p, int show_name)3049{3050 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3051 if (show_name)3052 printf(" mode change %06o => %06o %s\n",3053 p->old_mode, p->new_mode, p->new_name);3054 else3055 printf(" mode change %06o => %06o\n",3056 p->old_mode, p->new_mode);3057 }3058}30593060static void show_rename_copy(struct patch *p)3061{3062 const char *renamecopy = p->is_rename ? "rename" : "copy";3063 const char *old, *new;30643065 /* Find common prefix */3066 old = p->old_name;3067 new = p->new_name;3068 while (1) {3069 const char *slash_old, *slash_new;3070 slash_old = strchr(old, '/');3071 slash_new = strchr(new, '/');3072 if (!slash_old ||3073 !slash_new ||3074 slash_old - old != slash_new - new ||3075 memcmp(old, new, slash_new - new))3076 break;3077 old = slash_old + 1;3078 new = slash_new + 1;3079 }3080 /* p->old_name thru old is the common prefix, and old and new3081 * through the end of names are renames3082 */3083 if (old != p->old_name)3084 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,3085 (int)(old - p->old_name), p->old_name,3086 old, new, p->score);3087 else3088 printf(" %s %s => %s (%d%%)\n", renamecopy,3089 p->old_name, p->new_name, p->score);3090 show_mode_change(p, 0);3091}30923093static void summary_patch_list(struct patch *patch)3094{3095 struct patch *p;30963097 for (p = patch; p; p = p->next) {3098 if (p->is_new)3099 show_file_mode_name("create", p->new_mode, p->new_name);3100 else if (p->is_delete)3101 show_file_mode_name("delete", p->old_mode, p->old_name);3102 else {3103 if (p->is_rename || p->is_copy)3104 show_rename_copy(p);3105 else {3106 if (p->score) {3107 printf(" rewrite %s (%d%%)\n",3108 p->new_name, p->score);3109 show_mode_change(p, 0);3110 }3111 else3112 show_mode_change(p, 1);3113 }3114 }3115 }3116}31173118static void patch_stats(struct patch *patch)3119{3120 int lines = patch->lines_added + patch->lines_deleted;31213122 if (lines > max_change)3123 max_change = lines;3124 if (patch->old_name) {3125 int len = quote_c_style(patch->old_name, NULL, NULL, 0);3126 if (!len)3127 len = strlen(patch->old_name);3128 if (len > max_len)3129 max_len = len;3130 }3131 if (patch->new_name) {3132 int len = quote_c_style(patch->new_name, NULL, NULL, 0);3133 if (!len)3134 len = strlen(patch->new_name);3135 if (len > max_len)3136 max_len = len;3137 }3138}31393140static void remove_file(struct patch *patch, int rmdir_empty)3141{3142 if (update_index) {3143 if (remove_file_from_cache(patch->old_name) < 0)3144 die("unable to remove %s from index", patch->old_name);3145 }3146 if (!cached) {3147 if (S_ISGITLINK(patch->old_mode)) {3148 if (rmdir(patch->old_name))3149 warning("unable to remove submodule %s",3150 patch->old_name);3151 } else if (!unlink_or_warn(patch->old_name) && rmdir_empty) {3152 remove_path(patch->old_name);3153 }3154 }3155}31563157static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)3158{3159 struct stat st;3160 struct cache_entry *ce;3161 int namelen = strlen(path);3162 unsigned ce_size = cache_entry_size(namelen);31633164 if (!update_index)3165 return;31663167 ce = xcalloc(1, ce_size);3168 memcpy(ce->name, path, namelen);3169 ce->ce_mode = create_ce_mode(mode);3170 ce->ce_flags = namelen;3171 if (S_ISGITLINK(mode)) {3172 const char *s = buf;31733174 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))3175 die("corrupt patch for subproject %s", path);3176 } else {3177 if (!cached) {3178 if (lstat(path, &st) < 0)3179 die_errno("unable to stat newly created file '%s'",3180 path);3181 fill_stat_cache_info(ce, &st);3182 }3183 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)3184 die("unable to create backing store for newly created file %s", path);3185 }3186 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)3187 die("unable to add cache entry for %s", path);3188}31893190static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)3191{3192 int fd;3193 struct strbuf nbuf = STRBUF_INIT;31943195 if (S_ISGITLINK(mode)) {3196 struct stat st;3197 if (!lstat(path, &st) && S_ISDIR(st.st_mode))3198 return 0;3199 return mkdir(path, 0777);3200 }32013202 if (has_symlinks && S_ISLNK(mode))3203 /* Although buf:size is counted string, it also is NUL3204 * terminated.3205 */3206 return symlink(buf, path);32073208 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);3209 if (fd < 0)3210 return -1;32113212 if (convert_to_working_tree(path, buf, size, &nbuf)) {3213 size = nbuf.len;3214 buf = nbuf.buf;3215 }3216 write_or_die(fd, buf, size);3217 strbuf_release(&nbuf);32183219 if (close(fd) < 0)3220 die_errno("closing file '%s'", path);3221 return 0;3222}32233224/*3225 * We optimistically assume that the directories exist,3226 * which is true 99% of the time anyway. If they don't,3227 * we create them and try again.3228 */3229static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)3230{3231 if (cached)3232 return;3233 if (!try_create_file(path, mode, buf, size))3234 return;32353236 if (errno == ENOENT) {3237 if (safe_create_leading_directories(path))3238 return;3239 if (!try_create_file(path, mode, buf, size))3240 return;3241 }32423243 if (errno == EEXIST || errno == EACCES) {3244 /* We may be trying to create a file where a directory3245 * used to be.3246 */3247 struct stat st;3248 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))3249 errno = EEXIST;3250 }32513252 if (errno == EEXIST) {3253 unsigned int nr = getpid();32543255 for (;;) {3256 char newpath[PATH_MAX];3257 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);3258 if (!try_create_file(newpath, mode, buf, size)) {3259 if (!rename(newpath, path))3260 return;3261 unlink_or_warn(newpath);3262 break;3263 }3264 if (errno != EEXIST)3265 break;3266 ++nr;3267 }3268 }3269 die_errno("unable to write file '%s' mode %o", path, mode);3270}32713272static void create_file(struct patch *patch)3273{3274 char *path = patch->new_name;3275 unsigned mode = patch->new_mode;3276 unsigned long size = patch->resultsize;3277 char *buf = patch->result;32783279 if (!mode)3280 mode = S_IFREG | 0644;3281 create_one_file(path, mode, buf, size);3282 add_index_file(path, mode, buf, size);3283}32843285/* phase zero is to remove, phase one is to create */3286static void write_out_one_result(struct patch *patch, int phase)3287{3288 if (patch->is_delete > 0) {3289 if (phase == 0)3290 remove_file(patch, 1);3291 return;3292 }3293 if (patch->is_new > 0 || patch->is_copy) {3294 if (phase == 1)3295 create_file(patch);3296 return;3297 }3298 /*3299 * Rename or modification boils down to the same3300 * thing: remove the old, write the new3301 */3302 if (phase == 0)3303 remove_file(patch, patch->is_rename);3304 if (phase == 1)3305 create_file(patch);3306}33073308static int write_out_one_reject(struct patch *patch)3309{3310 FILE *rej;3311 char namebuf[PATH_MAX];3312 struct fragment *frag;3313 int cnt = 0;33143315 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {3316 if (!frag->rejected)3317 continue;3318 cnt++;3319 }33203321 if (!cnt) {3322 if (apply_verbosely)3323 say_patch_name(stderr,3324 "Applied patch ", patch, " cleanly.\n");3325 return 0;3326 }33273328 /* This should not happen, because a removal patch that leaves3329 * contents are marked "rejected" at the patch level.3330 */3331 if (!patch->new_name)3332 die("internal error");33333334 /* Say this even without --verbose */3335 say_patch_name(stderr, "Applying patch ", patch, " with");3336 fprintf(stderr, " %d rejects...\n", cnt);33373338 cnt = strlen(patch->new_name);3339 if (ARRAY_SIZE(namebuf) <= cnt + 5) {3340 cnt = ARRAY_SIZE(namebuf) - 5;3341 warning("truncating .rej filename to %.*s.rej",3342 cnt - 1, patch->new_name);3343 }3344 memcpy(namebuf, patch->new_name, cnt);3345 memcpy(namebuf + cnt, ".rej", 5);33463347 rej = fopen(namebuf, "w");3348 if (!rej)3349 return error("cannot open %s: %s", namebuf, strerror(errno));33503351 /* Normal git tools never deal with .rej, so do not pretend3352 * this is a git patch by saying --git nor give extended3353 * headers. While at it, maybe please "kompare" that wants3354 * the trailing TAB and some garbage at the end of line ;-).3355 */3356 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",3357 patch->new_name, patch->new_name);3358 for (cnt = 1, frag = patch->fragments;3359 frag;3360 cnt++, frag = frag->next) {3361 if (!frag->rejected) {3362 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);3363 continue;3364 }3365 fprintf(stderr, "Rejected hunk #%d.\n", cnt);3366 fprintf(rej, "%.*s", frag->size, frag->patch);3367 if (frag->patch[frag->size-1] != '\n')3368 fputc('\n', rej);3369 }3370 fclose(rej);3371 return -1;3372}33733374static int write_out_results(struct patch *list, int skipped_patch)3375{3376 int phase;3377 int errs = 0;3378 struct patch *l;33793380 if (!list && !skipped_patch)3381 return error("No changes");33823383 for (phase = 0; phase < 2; phase++) {3384 l = list;3385 while (l) {3386 if (l->rejected)3387 errs = 1;3388 else {3389 write_out_one_result(l, phase);3390 if (phase == 1 && write_out_one_reject(l))3391 errs = 1;3392 }3393 l = l->next;3394 }3395 }3396 return errs;3397}33983399static struct lock_file lock_file;34003401static struct string_list limit_by_name;3402static int has_include;3403static void add_name_limit(const char *name, int exclude)3404{3405 struct string_list_item *it;34063407 it = string_list_append(name, &limit_by_name);3408 it->util = exclude ? NULL : (void *) 1;3409}34103411static int use_patch(struct patch *p)3412{3413 const char *pathname = p->new_name ? p->new_name : p->old_name;3414 int i;34153416 /* Paths outside are not touched regardless of "--include" */3417 if (0 < prefix_length) {3418 int pathlen = strlen(pathname);3419 if (pathlen <= prefix_length ||3420 memcmp(prefix, pathname, prefix_length))3421 return 0;3422 }34233424 /* See if it matches any of exclude/include rule */3425 for (i = 0; i < limit_by_name.nr; i++) {3426 struct string_list_item *it = &limit_by_name.items[i];3427 if (!fnmatch(it->string, pathname, 0))3428 return (it->util != NULL);3429 }34303431 /*3432 * If we had any include, a path that does not match any rule is3433 * not used. Otherwise, we saw bunch of exclude rules (or none)3434 * and such a path is used.3435 */3436 return !has_include;3437}343834393440static void prefix_one(char **name)3441{3442 char *old_name = *name;3443 if (!old_name)3444 return;3445 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));3446 free(old_name);3447}34483449static void prefix_patches(struct patch *p)3450{3451 if (!prefix || p->is_toplevel_relative)3452 return;3453 for ( ; p; p = p->next) {3454 if (p->new_name == p->old_name) {3455 char *prefixed = p->new_name;3456 prefix_one(&prefixed);3457 p->new_name = p->old_name = prefixed;3458 }3459 else {3460 prefix_one(&p->new_name);3461 prefix_one(&p->old_name);3462 }3463 }3464}34653466#define INACCURATE_EOF (1<<0)3467#define RECOUNT (1<<1)34683469static int apply_patch(int fd, const char *filename, int options)3470{3471 size_t offset;3472 struct strbuf buf = STRBUF_INIT;3473 struct patch *list = NULL, **listp = &list;3474 int skipped_patch = 0;34753476 /* FIXME - memory leak when using multiple patch files as inputs */3477 memset(&fn_table, 0, sizeof(struct string_list));3478 patch_input_file = filename;3479 read_patch_file(&buf, fd);3480 offset = 0;3481 while (offset < buf.len) {3482 struct patch *patch;3483 int nr;34843485 patch = xcalloc(1, sizeof(*patch));3486 patch->inaccurate_eof = !!(options & INACCURATE_EOF);3487 patch->recount = !!(options & RECOUNT);3488 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);3489 if (nr < 0)3490 break;3491 if (apply_in_reverse)3492 reverse_patches(patch);3493 if (prefix)3494 prefix_patches(patch);3495 if (use_patch(patch)) {3496 patch_stats(patch);3497 *listp = patch;3498 listp = &patch->next;3499 }3500 else {3501 /* perhaps free it a bit better? */3502 free(patch);3503 skipped_patch++;3504 }3505 offset += nr;3506 }35073508 if (whitespace_error && (ws_error_action == die_on_ws_error))3509 apply = 0;35103511 update_index = check_index && apply;3512 if (update_index && newfd < 0)3513 newfd = hold_locked_index(&lock_file, 1);35143515 if (check_index) {3516 if (read_cache() < 0)3517 die("unable to read index file");3518 }35193520 if ((check || apply) &&3521 check_patch_list(list) < 0 &&3522 !apply_with_reject)3523 exit(1);35243525 if (apply && write_out_results(list, skipped_patch))3526 exit(1);35273528 if (fake_ancestor)3529 build_fake_ancestor(list, fake_ancestor);35303531 if (diffstat)3532 stat_patch_list(list);35333534 if (numstat)3535 numstat_patch_list(list);35363537 if (summary)3538 summary_patch_list(list);35393540 strbuf_release(&buf);3541 return 0;3542}35433544static int git_apply_config(const char *var, const char *value, void *cb)3545{3546 if (!strcmp(var, "apply.whitespace"))3547 return git_config_string(&apply_default_whitespace, var, value);3548 else if (!strcmp(var, "apply.ignorewhitespace"))3549 return git_config_string(&apply_default_ignorewhitespace, var, value);3550 return git_default_config(var, value, cb);3551}35523553static int option_parse_exclude(const struct option *opt,3554 const char *arg, int unset)3555{3556 add_name_limit(arg, 1);3557 return 0;3558}35593560static int option_parse_include(const struct option *opt,3561 const char *arg, int unset)3562{3563 add_name_limit(arg, 0);3564 has_include = 1;3565 return 0;3566}35673568static int option_parse_p(const struct option *opt,3569 const char *arg, int unset)3570{3571 p_value = atoi(arg);3572 p_value_known = 1;3573 return 0;3574}35753576static int option_parse_z(const struct option *opt,3577 const char *arg, int unset)3578{3579 if (unset)3580 line_termination = '\n';3581 else3582 line_termination = 0;3583 return 0;3584}35853586static int option_parse_space_change(const struct option *opt,3587 const char *arg, int unset)3588{3589 if (unset)3590 ws_ignore_action = ignore_ws_none;3591 else3592 ws_ignore_action = ignore_ws_change;3593 return 0;3594}35953596static int option_parse_whitespace(const struct option *opt,3597 const char *arg, int unset)3598{3599 const char **whitespace_option = opt->value;36003601 *whitespace_option = arg;3602 parse_whitespace_option(arg);3603 return 0;3604}36053606static int option_parse_directory(const struct option *opt,3607 const char *arg, int unset)3608{3609 root_len = strlen(arg);3610 if (root_len && arg[root_len - 1] != '/') {3611 char *new_root;3612 root = new_root = xmalloc(root_len + 2);3613 strcpy(new_root, arg);3614 strcpy(new_root + root_len++, "/");3615 } else3616 root = arg;3617 return 0;3618}36193620int cmd_apply(int argc, const char **argv, const char *unused_prefix)3621{3622 int i;3623 int errs = 0;3624 int is_not_gitdir;3625 int binary;3626 int force_apply = 0;36273628 const char *whitespace_option = NULL;36293630 struct option builtin_apply_options[] = {3631 { OPTION_CALLBACK, 0, "exclude", NULL, "path",3632 "don't apply changes matching the given path",3633 0, option_parse_exclude },3634 { OPTION_CALLBACK, 0, "include", NULL, "path",3635 "apply changes matching the given path",3636 0, option_parse_include },3637 { OPTION_CALLBACK, 'p', NULL, NULL, "num",3638 "remove <num> leading slashes from traditional diff paths",3639 0, option_parse_p },3640 OPT_BOOLEAN(0, "no-add", &no_add,3641 "ignore additions made by the patch"),3642 OPT_BOOLEAN(0, "stat", &diffstat,3643 "instead of applying the patch, output diffstat for the input"),3644 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,3645 NULL, "old option, now no-op",3646 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3647 { OPTION_BOOLEAN, 0, "binary", &binary,3648 NULL, "old option, now no-op",3649 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3650 OPT_BOOLEAN(0, "numstat", &numstat,3651 "shows number of added and deleted lines in decimal notation"),3652 OPT_BOOLEAN(0, "summary", &summary,3653 "instead of applying the patch, output a summary for the input"),3654 OPT_BOOLEAN(0, "check", &check,3655 "instead of applying the patch, see if the patch is applicable"),3656 OPT_BOOLEAN(0, "index", &check_index,3657 "make sure the patch is applicable to the current index"),3658 OPT_BOOLEAN(0, "cached", &cached,3659 "apply a patch without touching the working tree"),3660 OPT_BOOLEAN(0, "apply", &force_apply,3661 "also apply the patch (use with --stat/--summary/--check)"),3662 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,3663 "build a temporary index based on embedded index information"),3664 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,3665 "paths are separated with NUL character",3666 PARSE_OPT_NOARG, option_parse_z },3667 OPT_INTEGER('C', NULL, &p_context,3668 "ensure at least <n> lines of context match"),3669 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",3670 "detect new or modified lines that have whitespace errors",3671 0, option_parse_whitespace },3672 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,3673 "ignore changes in whitespace when finding context",3674 PARSE_OPT_NOARG, option_parse_space_change },3675 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,3676 "ignore changes in whitespace when finding context",3677 PARSE_OPT_NOARG, option_parse_space_change },3678 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,3679 "apply the patch in reverse"),3680 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,3681 "don't expect at least one line of context"),3682 OPT_BOOLEAN(0, "reject", &apply_with_reject,3683 "leave the rejected hunks in corresponding *.rej files"),3684 OPT__VERBOSE(&apply_verbosely),3685 OPT_BIT(0, "inaccurate-eof", &options,3686 "tolerate incorrectly detected missing new-line at the end of file",3687 INACCURATE_EOF),3688 OPT_BIT(0, "recount", &options,3689 "do not trust the line counts in the hunk headers",3690 RECOUNT),3691 { OPTION_CALLBACK, 0, "directory", NULL, "root",3692 "prepend <root> to all filenames",3693 0, option_parse_directory },3694 OPT_END()3695 };36963697 prefix = setup_git_directory_gently(&is_not_gitdir);3698 prefix_length = prefix ? strlen(prefix) : 0;3699 git_config(git_apply_config, NULL);3700 if (apply_default_whitespace)3701 parse_whitespace_option(apply_default_whitespace);3702 if (apply_default_ignorewhitespace)3703 parse_ignorewhitespace_option(apply_default_ignorewhitespace);37043705 argc = parse_options(argc, argv, prefix, builtin_apply_options,3706 apply_usage, 0);37073708 if (apply_with_reject)3709 apply = apply_verbosely = 1;3710 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))3711 apply = 0;3712 if (check_index && is_not_gitdir)3713 die("--index outside a repository");3714 if (cached) {3715 if (is_not_gitdir)3716 die("--cached outside a repository");3717 check_index = 1;3718 }3719 for (i = 0; i < argc; i++) {3720 const char *arg = argv[i];3721 int fd;37223723 if (!strcmp(arg, "-")) {3724 errs |= apply_patch(0, "<stdin>", options);3725 read_stdin = 0;3726 continue;3727 } else if (0 < prefix_length)3728 arg = prefix_filename(prefix, prefix_length, arg);37293730 fd = open(arg, O_RDONLY);3731 if (fd < 0)3732 die_errno("can't open patch '%s'", arg);3733 read_stdin = 0;3734 set_default_whitespace_mode(whitespace_option);3735 errs |= apply_patch(fd, arg, options);3736 close(fd);3737 }3738 set_default_whitespace_mode(whitespace_option);3739 if (read_stdin)3740 errs |= apply_patch(0, "<stdin>", options);3741 if (whitespace_error) {3742 if (squelch_whitespace_errors &&3743 squelch_whitespace_errors < whitespace_error) {3744 int squelched =3745 whitespace_error - squelch_whitespace_errors;3746 warning("squelched %d "3747 "whitespace error%s",3748 squelched,3749 squelched == 1 ? "" : "s");3750 }3751 if (ws_error_action == die_on_ws_error)3752 die("%d line%s add%s whitespace errors.",3753 whitespace_error,3754 whitespace_error == 1 ? "" : "s",3755 whitespace_error == 1 ? "s" : "");3756 if (applied_after_fixing_ws && apply)3757 warning("%d line%s applied after"3758 " fixing whitespace errors.",3759 applied_after_fixing_ws,3760 applied_after_fixing_ws == 1 ? "" : "s");3761 else if (whitespace_error)3762 warning("%d line%s add%s whitespace errors.",3763 whitespace_error,3764 whitespace_error == 1 ? "" : "s",3765 whitespace_error == 1 ? "s" : "");3766 }37673768 if (update_index) {3769 if (write_cache(newfd, active_cache, active_nr) ||3770 commit_locked_index(&lock_file))3771 die("Unable to write new index file");3772 }37733774 return !!errs;3775}