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_gnu(const char *line, char *def, int p_value) 420{ 421 struct strbuf name = STRBUF_INIT; 422 char *cp; 423 424 /* 425 * Proposed "new-style" GNU patch/diff format; see 426 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2 427 */ 428 if (unquote_c_style(&name, line, NULL)) { 429 strbuf_release(&name); 430 return NULL; 431 } 432 433 for (cp = name.buf; p_value; p_value--) { 434 cp = strchr(cp, '/'); 435 if (!cp) { 436 strbuf_release(&name); 437 return NULL; 438 } 439 cp++; 440 } 441 442 /* name can later be freed, so we need 443 * to memmove, not just return cp 444 */ 445 strbuf_remove(&name, 0, cp - name.buf); 446 free(def); 447 if (root) 448 strbuf_insert(&name, 0, root, root_len); 449 return squash_slash(strbuf_detach(&name, NULL)); 450} 451 452static size_t tz_len(const char *line, size_t len) 453{ 454 const char *tz, *p; 455 456 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ') 457 return 0; 458 tz = line + len - strlen(" +0500"); 459 460 if (tz[1] != '+' && tz[1] != '-') 461 return 0; 462 463 for (p = tz + 2; p != line + len; p++) 464 if (!isdigit(*p)) 465 return 0; 466 467 return line + len - tz; 468} 469 470static size_t date_len(const char *line, size_t len) 471{ 472 const char *date, *p; 473 474 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-') 475 return 0; 476 p = date = line + len - strlen("72-02-05"); 477 478 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 479 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 480 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */ 481 return 0; 482 483 if (date - line >= strlen("19") && 484 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */ 485 date -= strlen("19"); 486 487 return line + len - date; 488} 489 490static size_t short_time_len(const char *line, size_t len) 491{ 492 const char *time, *p; 493 494 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':') 495 return 0; 496 p = time = line + len - strlen(" 07:01:32"); 497 498 /* Permit 1-digit hours? */ 499 if (*p++ != ' ' || 500 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 501 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 502 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */ 503 return 0; 504 505 return line + len - time; 506} 507 508static size_t fractional_time_len(const char *line, size_t len) 509{ 510 const char *p; 511 size_t n; 512 513 /* Expected format: 19:41:17.620000023 */ 514 if (!len || !isdigit(line[len - 1])) 515 return 0; 516 p = line + len - 1; 517 518 /* Fractional seconds. */ 519 while (p > line && isdigit(*p)) 520 p--; 521 if (*p != '.') 522 return 0; 523 524 /* Hours, minutes, and whole seconds. */ 525 n = short_time_len(line, p - line); 526 if (!n) 527 return 0; 528 529 return line + len - p + n; 530} 531 532static size_t trailing_spaces_len(const char *line, size_t len) 533{ 534 const char *p; 535 536 /* Expected format: ' ' x (1 or more) */ 537 if (!len || line[len - 1] != ' ') 538 return 0; 539 540 p = line + len; 541 while (p != line) { 542 p--; 543 if (*p != ' ') 544 return line + len - (p + 1); 545 } 546 547 /* All spaces! */ 548 return len; 549} 550 551static size_t diff_timestamp_len(const char *line, size_t len) 552{ 553 const char *end = line + len; 554 size_t n; 555 556 /* 557 * Posix: 2010-07-05 19:41:17 558 * GNU: 2010-07-05 19:41:17.620000023 -0500 559 */ 560 561 if (!isdigit(end[-1])) 562 return 0; 563 564 n = tz_len(line, end - line); 565 end -= n; 566 567 n = short_time_len(line, end - line); 568 if (!n) 569 n = fractional_time_len(line, end - line); 570 end -= n; 571 572 n = date_len(line, end - line); 573 if (!n) /* No date. Too bad. */ 574 return 0; 575 end -= n; 576 577 if (end == line) /* No space before date. */ 578 return 0; 579 if (end[-1] == '\t') { /* Success! */ 580 end--; 581 return line + len - end; 582 } 583 if (end[-1] != ' ') /* No space before date. */ 584 return 0; 585 586 /* Whitespace damage. */ 587 end -= trailing_spaces_len(line, end - line); 588 return line + len - end; 589} 590 591static char *find_name_common(const char *line, char *def, int p_value, 592 const char *end, int terminate) 593{ 594 int len; 595 const char *start = NULL; 596 597 if (p_value == 0) 598 start = line; 599 while (line != end) { 600 char c = *line; 601 602 if (!end && isspace(c)) { 603 if (c == '\n') 604 break; 605 if (name_terminate(start, line-start, c, terminate)) 606 break; 607 } 608 line++; 609 if (c == '/' && !--p_value) 610 start = line; 611 } 612 if (!start) 613 return squash_slash(def); 614 len = line - start; 615 if (!len) 616 return squash_slash(def); 617 618 /* 619 * Generally we prefer the shorter name, especially 620 * if the other one is just a variation of that with 621 * something else tacked on to the end (ie "file.orig" 622 * or "file~"). 623 */ 624 if (def) { 625 int deflen = strlen(def); 626 if (deflen < len && !strncmp(start, def, deflen)) 627 return squash_slash(def); 628 free(def); 629 } 630 631 if (root) { 632 char *ret = xmalloc(root_len + len + 1); 633 strcpy(ret, root); 634 memcpy(ret + root_len, start, len); 635 ret[root_len + len] = '\0'; 636 return squash_slash(ret); 637 } 638 639 return squash_slash(xmemdupz(start, len)); 640} 641 642static char *find_name(const char *line, char *def, int p_value, int terminate) 643{ 644 if (*line == '"') { 645 char *name = find_name_gnu(line, def, p_value); 646 if (name) 647 return name; 648 } 649 650 return find_name_common(line, def, p_value, NULL, terminate); 651} 652 653static char *find_name_traditional(const char *line, char *def, int p_value) 654{ 655 size_t len = strlen(line); 656 size_t date_len; 657 658 if (*line == '"') { 659 char *name = find_name_gnu(line, def, p_value); 660 if (name) 661 return name; 662 } 663 664 len = strchrnul(line, '\n') - line; 665 date_len = diff_timestamp_len(line, len); 666 if (!date_len) 667 return find_name_common(line, def, p_value, NULL, TERM_TAB); 668 len -= date_len; 669 670 return find_name_common(line, def, p_value, line + len, 0); 671} 672 673static int count_slashes(const char *cp) 674{ 675 int cnt = 0; 676 char ch; 677 678 while ((ch = *cp++)) 679 if (ch == '/') 680 cnt++; 681 return cnt; 682} 683 684/* 685 * Given the string after "--- " or "+++ ", guess the appropriate 686 * p_value for the given patch. 687 */ 688static int guess_p_value(const char *nameline) 689{ 690 char *name, *cp; 691 int val = -1; 692 693 if (is_dev_null(nameline)) 694 return -1; 695 name = find_name_traditional(nameline, NULL, 0); 696 if (!name) 697 return -1; 698 cp = strchr(name, '/'); 699 if (!cp) 700 val = 0; 701 else if (prefix) { 702 /* 703 * Does it begin with "a/$our-prefix" and such? Then this is 704 * very likely to apply to our directory. 705 */ 706 if (!strncmp(name, prefix, prefix_length)) 707 val = count_slashes(prefix); 708 else { 709 cp++; 710 if (!strncmp(cp, prefix, prefix_length)) 711 val = count_slashes(prefix) + 1; 712 } 713 } 714 free(name); 715 return val; 716} 717 718/* 719 * Does the ---/+++ line has the POSIX timestamp after the last HT? 720 * GNU diff puts epoch there to signal a creation/deletion event. Is 721 * this such a timestamp? 722 */ 723static int has_epoch_timestamp(const char *nameline) 724{ 725 /* 726 * We are only interested in epoch timestamp; any non-zero 727 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 728 * For the same reason, the date must be either 1969-12-31 or 729 * 1970-01-01, and the seconds part must be "00". 730 */ 731 const char stamp_regexp[] = 732 "^(1969-12-31|1970-01-01)" 733 " " 734 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 735 " " 736 "([-+][0-2][0-9][0-5][0-9])\n"; 737 const char *timestamp = NULL, *cp; 738 static regex_t *stamp; 739 regmatch_t m[10]; 740 int zoneoffset; 741 int hourminute; 742 int status; 743 744 for (cp = nameline; *cp != '\n'; cp++) { 745 if (*cp == '\t') 746 timestamp = cp + 1; 747 } 748 if (!timestamp) 749 return 0; 750 if (!stamp) { 751 stamp = xmalloc(sizeof(*stamp)); 752 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 753 warning("Cannot prepare timestamp regexp %s", 754 stamp_regexp); 755 return 0; 756 } 757 } 758 759 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 760 if (status) { 761 if (status != REG_NOMATCH) 762 warning("regexec returned %d for input: %s", 763 status, timestamp); 764 return 0; 765 } 766 767 zoneoffset = strtol(timestamp + m[3].rm_so + 1, NULL, 10); 768 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 769 if (timestamp[m[3].rm_so] == '-') 770 zoneoffset = -zoneoffset; 771 772 /* 773 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 774 * (west of GMT) or 1970-01-01 (east of GMT) 775 */ 776 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 777 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 778 return 0; 779 780 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 781 strtol(timestamp + 14, NULL, 10) - 782 zoneoffset); 783 784 return ((zoneoffset < 0 && hourminute == 1440) || 785 (0 <= zoneoffset && !hourminute)); 786} 787 788/* 789 * Get the name etc info from the ---/+++ lines of a traditional patch header 790 * 791 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 792 * files, we can happily check the index for a match, but for creating a 793 * new file we should try to match whatever "patch" does. I have no idea. 794 */ 795static void parse_traditional_patch(const char *first, const char *second, struct patch *patch) 796{ 797 char *name; 798 799 first += 4; /* skip "--- " */ 800 second += 4; /* skip "+++ " */ 801 if (!p_value_known) { 802 int p, q; 803 p = guess_p_value(first); 804 q = guess_p_value(second); 805 if (p < 0) p = q; 806 if (0 <= p && p == q) { 807 p_value = p; 808 p_value_known = 1; 809 } 810 } 811 if (is_dev_null(first)) { 812 patch->is_new = 1; 813 patch->is_delete = 0; 814 name = find_name_traditional(second, NULL, p_value); 815 patch->new_name = name; 816 } else if (is_dev_null(second)) { 817 patch->is_new = 0; 818 patch->is_delete = 1; 819 name = find_name_traditional(first, NULL, p_value); 820 patch->old_name = name; 821 } else { 822 name = find_name_traditional(first, NULL, p_value); 823 name = find_name_traditional(second, name, p_value); 824 if (has_epoch_timestamp(first)) { 825 patch->is_new = 1; 826 patch->is_delete = 0; 827 patch->new_name = name; 828 } else if (has_epoch_timestamp(second)) { 829 patch->is_new = 0; 830 patch->is_delete = 1; 831 patch->old_name = name; 832 } else { 833 patch->old_name = patch->new_name = name; 834 } 835 } 836 if (!name) 837 die("unable to find filename in patch at line %d", linenr); 838} 839 840static int gitdiff_hdrend(const char *line, struct patch *patch) 841{ 842 return -1; 843} 844 845/* 846 * We're anal about diff header consistency, to make 847 * sure that we don't end up having strange ambiguous 848 * patches floating around. 849 * 850 * As a result, gitdiff_{old|new}name() will check 851 * their names against any previous information, just 852 * to make sure.. 853 */ 854static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew) 855{ 856 if (!orig_name && !isnull) 857 return find_name(line, NULL, p_value, TERM_TAB); 858 859 if (orig_name) { 860 int len; 861 const char *name; 862 char *another; 863 name = orig_name; 864 len = strlen(name); 865 if (isnull) 866 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr); 867 another = find_name(line, NULL, p_value, TERM_TAB); 868 if (!another || memcmp(another, name, len + 1)) 869 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr); 870 free(another); 871 return orig_name; 872 } 873 else { 874 /* expect "/dev/null" */ 875 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 876 die("git apply: bad git-diff - expected /dev/null on line %d", linenr); 877 return NULL; 878 } 879} 880 881static int gitdiff_oldname(const char *line, struct patch *patch) 882{ 883 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old"); 884 return 0; 885} 886 887static int gitdiff_newname(const char *line, struct patch *patch) 888{ 889 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new"); 890 return 0; 891} 892 893static int gitdiff_oldmode(const char *line, struct patch *patch) 894{ 895 patch->old_mode = strtoul(line, NULL, 8); 896 return 0; 897} 898 899static int gitdiff_newmode(const char *line, struct patch *patch) 900{ 901 patch->new_mode = strtoul(line, NULL, 8); 902 return 0; 903} 904 905static int gitdiff_delete(const char *line, struct patch *patch) 906{ 907 patch->is_delete = 1; 908 patch->old_name = patch->def_name; 909 return gitdiff_oldmode(line, patch); 910} 911 912static int gitdiff_newfile(const char *line, struct patch *patch) 913{ 914 patch->is_new = 1; 915 patch->new_name = patch->def_name; 916 return gitdiff_newmode(line, patch); 917} 918 919static int gitdiff_copysrc(const char *line, struct patch *patch) 920{ 921 patch->is_copy = 1; 922 patch->old_name = find_name(line, NULL, 0, 0); 923 return 0; 924} 925 926static int gitdiff_copydst(const char *line, struct patch *patch) 927{ 928 patch->is_copy = 1; 929 patch->new_name = find_name(line, NULL, 0, 0); 930 return 0; 931} 932 933static int gitdiff_renamesrc(const char *line, struct patch *patch) 934{ 935 patch->is_rename = 1; 936 patch->old_name = find_name(line, NULL, 0, 0); 937 return 0; 938} 939 940static int gitdiff_renamedst(const char *line, struct patch *patch) 941{ 942 patch->is_rename = 1; 943 patch->new_name = find_name(line, NULL, 0, 0); 944 return 0; 945} 946 947static int gitdiff_similarity(const char *line, struct patch *patch) 948{ 949 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 950 patch->score = 0; 951 return 0; 952} 953 954static int gitdiff_dissimilarity(const char *line, struct patch *patch) 955{ 956 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 957 patch->score = 0; 958 return 0; 959} 960 961static int gitdiff_index(const char *line, struct patch *patch) 962{ 963 /* 964 * index line is N hexadecimal, "..", N hexadecimal, 965 * and optional space with octal mode. 966 */ 967 const char *ptr, *eol; 968 int len; 969 970 ptr = strchr(line, '.'); 971 if (!ptr || ptr[1] != '.' || 40 < ptr - line) 972 return 0; 973 len = ptr - line; 974 memcpy(patch->old_sha1_prefix, line, len); 975 patch->old_sha1_prefix[len] = 0; 976 977 line = ptr + 2; 978 ptr = strchr(line, ' '); 979 eol = strchr(line, '\n'); 980 981 if (!ptr || eol < ptr) 982 ptr = eol; 983 len = ptr - line; 984 985 if (40 < len) 986 return 0; 987 memcpy(patch->new_sha1_prefix, line, len); 988 patch->new_sha1_prefix[len] = 0; 989 if (*ptr == ' ') 990 patch->old_mode = strtoul(ptr+1, NULL, 8); 991 return 0; 992} 993 994/* 995 * This is normal for a diff that doesn't change anything: we'll fall through 996 * into the next diff. Tell the parser to break out. 997 */ 998static int gitdiff_unrecognized(const char *line, struct patch *patch) 999{1000 return -1;1001}10021003static const char *stop_at_slash(const char *line, int llen)1004{1005 int nslash = p_value;1006 int i;10071008 for (i = 0; i < llen; i++) {1009 int ch = line[i];1010 if (ch == '/' && --nslash <= 0)1011 return &line[i];1012 }1013 return NULL;1014}10151016/*1017 * This is to extract the same name that appears on "diff --git"1018 * line. We do not find and return anything if it is a rename1019 * patch, and it is OK because we will find the name elsewhere.1020 * We need to reliably find name only when it is mode-change only,1021 * creation or deletion of an empty file. In any of these cases,1022 * both sides are the same name under a/ and b/ respectively.1023 */1024static char *git_header_name(char *line, int llen)1025{1026 const char *name;1027 const char *second = NULL;1028 size_t len;10291030 line += strlen("diff --git ");1031 llen -= strlen("diff --git ");10321033 if (*line == '"') {1034 const char *cp;1035 struct strbuf first = STRBUF_INIT;1036 struct strbuf sp = STRBUF_INIT;10371038 if (unquote_c_style(&first, line, &second))1039 goto free_and_fail1;10401041 /* advance to the first slash */1042 cp = stop_at_slash(first.buf, first.len);1043 /* we do not accept absolute paths */1044 if (!cp || cp == first.buf)1045 goto free_and_fail1;1046 strbuf_remove(&first, 0, cp + 1 - first.buf);10471048 /*1049 * second points at one past closing dq of name.1050 * find the second name.1051 */1052 while ((second < line + llen) && isspace(*second))1053 second++;10541055 if (line + llen <= second)1056 goto free_and_fail1;1057 if (*second == '"') {1058 if (unquote_c_style(&sp, second, NULL))1059 goto free_and_fail1;1060 cp = stop_at_slash(sp.buf, sp.len);1061 if (!cp || cp == sp.buf)1062 goto free_and_fail1;1063 /* They must match, otherwise ignore */1064 if (strcmp(cp + 1, first.buf))1065 goto free_and_fail1;1066 strbuf_release(&sp);1067 return strbuf_detach(&first, NULL);1068 }10691070 /* unquoted second */1071 cp = stop_at_slash(second, line + llen - second);1072 if (!cp || cp == second)1073 goto free_and_fail1;1074 cp++;1075 if (line + llen - cp != first.len + 1 ||1076 memcmp(first.buf, cp, first.len))1077 goto free_and_fail1;1078 return strbuf_detach(&first, NULL);10791080 free_and_fail1:1081 strbuf_release(&first);1082 strbuf_release(&sp);1083 return NULL;1084 }10851086 /* unquoted first name */1087 name = stop_at_slash(line, llen);1088 if (!name || name == line)1089 return NULL;1090 name++;10911092 /*1093 * since the first name is unquoted, a dq if exists must be1094 * the beginning of the second name.1095 */1096 for (second = name; second < line + llen; second++) {1097 if (*second == '"') {1098 struct strbuf sp = STRBUF_INIT;1099 const char *np;11001101 if (unquote_c_style(&sp, second, NULL))1102 goto free_and_fail2;11031104 np = stop_at_slash(sp.buf, sp.len);1105 if (!np || np == sp.buf)1106 goto free_and_fail2;1107 np++;11081109 len = sp.buf + sp.len - np;1110 if (len < second - name &&1111 !strncmp(np, name, len) &&1112 isspace(name[len])) {1113 /* Good */1114 strbuf_remove(&sp, 0, np - sp.buf);1115 return strbuf_detach(&sp, NULL);1116 }11171118 free_and_fail2:1119 strbuf_release(&sp);1120 return NULL;1121 }1122 }11231124 /*1125 * Accept a name only if it shows up twice, exactly the same1126 * form.1127 */1128 for (len = 0 ; ; len++) {1129 switch (name[len]) {1130 default:1131 continue;1132 case '\n':1133 return NULL;1134 case '\t': case ' ':1135 second = name+len;1136 for (;;) {1137 char c = *second++;1138 if (c == '\n')1139 return NULL;1140 if (c == '/')1141 break;1142 }1143 if (second[len] == '\n' && !memcmp(name, second, len)) {1144 return xmemdupz(name, len);1145 }1146 }1147 }1148}11491150/* Verify that we recognize the lines following a git header */1151static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)1152{1153 unsigned long offset;11541155 /* A git diff has explicit new/delete information, so we don't guess */1156 patch->is_new = 0;1157 patch->is_delete = 0;11581159 /*1160 * Some things may not have the old name in the1161 * rest of the headers anywhere (pure mode changes,1162 * or removing or adding empty files), so we get1163 * the default name from the header.1164 */1165 patch->def_name = git_header_name(line, len);1166 if (patch->def_name && root) {1167 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);1168 strcpy(s, root);1169 strcpy(s + root_len, patch->def_name);1170 free(patch->def_name);1171 patch->def_name = s;1172 }11731174 line += len;1175 size -= len;1176 linenr++;1177 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {1178 static const struct opentry {1179 const char *str;1180 int (*fn)(const char *, struct patch *);1181 } optable[] = {1182 { "@@ -", gitdiff_hdrend },1183 { "--- ", gitdiff_oldname },1184 { "+++ ", gitdiff_newname },1185 { "old mode ", gitdiff_oldmode },1186 { "new mode ", gitdiff_newmode },1187 { "deleted file mode ", gitdiff_delete },1188 { "new file mode ", gitdiff_newfile },1189 { "copy from ", gitdiff_copysrc },1190 { "copy to ", gitdiff_copydst },1191 { "rename old ", gitdiff_renamesrc },1192 { "rename new ", gitdiff_renamedst },1193 { "rename from ", gitdiff_renamesrc },1194 { "rename to ", gitdiff_renamedst },1195 { "similarity index ", gitdiff_similarity },1196 { "dissimilarity index ", gitdiff_dissimilarity },1197 { "index ", gitdiff_index },1198 { "", gitdiff_unrecognized },1199 };1200 int i;12011202 len = linelen(line, size);1203 if (!len || line[len-1] != '\n')1204 break;1205 for (i = 0; i < ARRAY_SIZE(optable); i++) {1206 const struct opentry *p = optable + i;1207 int oplen = strlen(p->str);1208 if (len < oplen || memcmp(p->str, line, oplen))1209 continue;1210 if (p->fn(line + oplen, patch) < 0)1211 return offset;1212 break;1213 }1214 }12151216 return offset;1217}12181219static int parse_num(const char *line, unsigned long *p)1220{1221 char *ptr;12221223 if (!isdigit(*line))1224 return 0;1225 *p = strtoul(line, &ptr, 10);1226 return ptr - line;1227}12281229static int parse_range(const char *line, int len, int offset, const char *expect,1230 unsigned long *p1, unsigned long *p2)1231{1232 int digits, ex;12331234 if (offset < 0 || offset >= len)1235 return -1;1236 line += offset;1237 len -= offset;12381239 digits = parse_num(line, p1);1240 if (!digits)1241 return -1;12421243 offset += digits;1244 line += digits;1245 len -= digits;12461247 *p2 = 1;1248 if (*line == ',') {1249 digits = parse_num(line+1, p2);1250 if (!digits)1251 return -1;12521253 offset += digits+1;1254 line += digits+1;1255 len -= digits+1;1256 }12571258 ex = strlen(expect);1259 if (ex > len)1260 return -1;1261 if (memcmp(line, expect, ex))1262 return -1;12631264 return offset + ex;1265}12661267static void recount_diff(char *line, int size, struct fragment *fragment)1268{1269 int oldlines = 0, newlines = 0, ret = 0;12701271 if (size < 1) {1272 warning("recount: ignore empty hunk");1273 return;1274 }12751276 for (;;) {1277 int len = linelen(line, size);1278 size -= len;1279 line += len;12801281 if (size < 1)1282 break;12831284 switch (*line) {1285 case ' ': case '\n':1286 newlines++;1287 /* fall through */1288 case '-':1289 oldlines++;1290 continue;1291 case '+':1292 newlines++;1293 continue;1294 case '\\':1295 continue;1296 case '@':1297 ret = size < 3 || prefixcmp(line, "@@ ");1298 break;1299 case 'd':1300 ret = size < 5 || prefixcmp(line, "diff ");1301 break;1302 default:1303 ret = -1;1304 break;1305 }1306 if (ret) {1307 warning("recount: unexpected line: %.*s",1308 (int)linelen(line, size), line);1309 return;1310 }1311 break;1312 }1313 fragment->oldlines = oldlines;1314 fragment->newlines = newlines;1315}13161317/*1318 * Parse a unified diff fragment header of the1319 * form "@@ -a,b +c,d @@"1320 */1321static int parse_fragment_header(char *line, int len, struct fragment *fragment)1322{1323 int offset;13241325 if (!len || line[len-1] != '\n')1326 return -1;13271328 /* Figure out the number of lines in a fragment */1329 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1330 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);13311332 return offset;1333}13341335static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)1336{1337 unsigned long offset, len;13381339 patch->is_toplevel_relative = 0;1340 patch->is_rename = patch->is_copy = 0;1341 patch->is_new = patch->is_delete = -1;1342 patch->old_mode = patch->new_mode = 0;1343 patch->old_name = patch->new_name = NULL;1344 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {1345 unsigned long nextlen;13461347 len = linelen(line, size);1348 if (!len)1349 break;13501351 /* Testing this early allows us to take a few shortcuts.. */1352 if (len < 6)1353 continue;13541355 /*1356 * Make sure we don't find any unconnected patch fragments.1357 * That's a sign that we didn't find a header, and that a1358 * patch has become corrupted/broken up.1359 */1360 if (!memcmp("@@ -", line, 4)) {1361 struct fragment dummy;1362 if (parse_fragment_header(line, len, &dummy) < 0)1363 continue;1364 die("patch fragment without header at line %d: %.*s",1365 linenr, (int)len-1, line);1366 }13671368 if (size < len + 6)1369 break;13701371 /*1372 * Git patch? It might not have a real patch, just a rename1373 * or mode change, so we handle that specially1374 */1375 if (!memcmp("diff --git ", line, 11)) {1376 int git_hdr_len = parse_git_header(line, len, size, patch);1377 if (git_hdr_len <= len)1378 continue;1379 if (!patch->old_name && !patch->new_name) {1380 if (!patch->def_name)1381 die("git diff header lacks filename information when removing "1382 "%d leading pathname components (line %d)" , p_value, linenr);1383 patch->old_name = patch->new_name = patch->def_name;1384 }1385 patch->is_toplevel_relative = 1;1386 *hdrsize = git_hdr_len;1387 return offset;1388 }13891390 /* --- followed by +++ ? */1391 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1392 continue;13931394 /*1395 * We only accept unified patches, so we want it to1396 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1397 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1398 */1399 nextlen = linelen(line + len, size - len);1400 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1401 continue;14021403 /* Ok, we'll consider it a patch */1404 parse_traditional_patch(line, line+len, patch);1405 *hdrsize = len + nextlen;1406 linenr += 2;1407 return offset;1408 }1409 return -1;1410}14111412static void record_ws_error(unsigned result, const char *line, int len, int linenr)1413{1414 char *err;14151416 if (!result)1417 return;14181419 whitespace_error++;1420 if (squelch_whitespace_errors &&1421 squelch_whitespace_errors < whitespace_error)1422 return;14231424 err = whitespace_error_string(result);1425 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1426 patch_input_file, linenr, err, len, line);1427 free(err);1428}14291430static void check_whitespace(const char *line, int len, unsigned ws_rule)1431{1432 unsigned result = ws_check(line + 1, len - 1, ws_rule);14331434 record_ws_error(result, line + 1, len - 2, linenr);1435}14361437/*1438 * Parse a unified diff. Note that this really needs to parse each1439 * fragment separately, since the only way to know the difference1440 * between a "---" that is part of a patch, and a "---" that starts1441 * the next patch is to look at the line counts..1442 */1443static int parse_fragment(char *line, unsigned long size,1444 struct patch *patch, struct fragment *fragment)1445{1446 int added, deleted;1447 int len = linelen(line, size), offset;1448 unsigned long oldlines, newlines;1449 unsigned long leading, trailing;14501451 offset = parse_fragment_header(line, len, fragment);1452 if (offset < 0)1453 return -1;1454 if (offset > 0 && patch->recount)1455 recount_diff(line + offset, size - offset, fragment);1456 oldlines = fragment->oldlines;1457 newlines = fragment->newlines;1458 leading = 0;1459 trailing = 0;14601461 /* Parse the thing.. */1462 line += len;1463 size -= len;1464 linenr++;1465 added = deleted = 0;1466 for (offset = len;1467 0 < size;1468 offset += len, size -= len, line += len, linenr++) {1469 if (!oldlines && !newlines)1470 break;1471 len = linelen(line, size);1472 if (!len || line[len-1] != '\n')1473 return -1;1474 switch (*line) {1475 default:1476 return -1;1477 case '\n': /* newer GNU diff, an empty context line */1478 case ' ':1479 oldlines--;1480 newlines--;1481 if (!deleted && !added)1482 leading++;1483 trailing++;1484 break;1485 case '-':1486 if (apply_in_reverse &&1487 ws_error_action != nowarn_ws_error)1488 check_whitespace(line, len, patch->ws_rule);1489 deleted++;1490 oldlines--;1491 trailing = 0;1492 break;1493 case '+':1494 if (!apply_in_reverse &&1495 ws_error_action != nowarn_ws_error)1496 check_whitespace(line, len, patch->ws_rule);1497 added++;1498 newlines--;1499 trailing = 0;1500 break;15011502 /*1503 * We allow "\ No newline at end of file". Depending1504 * on locale settings when the patch was produced we1505 * don't know what this line looks like. The only1506 * thing we do know is that it begins with "\ ".1507 * Checking for 12 is just for sanity check -- any1508 * l10n of "\ No newline..." is at least that long.1509 */1510 case '\\':1511 if (len < 12 || memcmp(line, "\\ ", 2))1512 return -1;1513 break;1514 }1515 }1516 if (oldlines || newlines)1517 return -1;1518 fragment->leading = leading;1519 fragment->trailing = trailing;15201521 /*1522 * If a fragment ends with an incomplete line, we failed to include1523 * it in the above loop because we hit oldlines == newlines == 01524 * before seeing it.1525 */1526 if (12 < size && !memcmp(line, "\\ ", 2))1527 offset += linelen(line, size);15281529 patch->lines_added += added;1530 patch->lines_deleted += deleted;15311532 if (0 < patch->is_new && oldlines)1533 return error("new file depends on old contents");1534 if (0 < patch->is_delete && newlines)1535 return error("deleted file still has contents");1536 return offset;1537}15381539static int parse_single_patch(char *line, unsigned long size, struct patch *patch)1540{1541 unsigned long offset = 0;1542 unsigned long oldlines = 0, newlines = 0, context = 0;1543 struct fragment **fragp = &patch->fragments;15441545 while (size > 4 && !memcmp(line, "@@ -", 4)) {1546 struct fragment *fragment;1547 int len;15481549 fragment = xcalloc(1, sizeof(*fragment));1550 fragment->linenr = linenr;1551 len = parse_fragment(line, size, patch, fragment);1552 if (len <= 0)1553 die("corrupt patch at line %d", linenr);1554 fragment->patch = line;1555 fragment->size = len;1556 oldlines += fragment->oldlines;1557 newlines += fragment->newlines;1558 context += fragment->leading + fragment->trailing;15591560 *fragp = fragment;1561 fragp = &fragment->next;15621563 offset += len;1564 line += len;1565 size -= len;1566 }15671568 /*1569 * If something was removed (i.e. we have old-lines) it cannot1570 * be creation, and if something was added it cannot be1571 * deletion. However, the reverse is not true; --unified=01572 * patches that only add are not necessarily creation even1573 * though they do not have any old lines, and ones that only1574 * delete are not necessarily deletion.1575 *1576 * Unfortunately, a real creation/deletion patch do _not_ have1577 * any context line by definition, so we cannot safely tell it1578 * apart with --unified=0 insanity. At least if the patch has1579 * more than one hunk it is not creation or deletion.1580 */1581 if (patch->is_new < 0 &&1582 (oldlines || (patch->fragments && patch->fragments->next)))1583 patch->is_new = 0;1584 if (patch->is_delete < 0 &&1585 (newlines || (patch->fragments && patch->fragments->next)))1586 patch->is_delete = 0;15871588 if (0 < patch->is_new && oldlines)1589 die("new file %s depends on old contents", patch->new_name);1590 if (0 < patch->is_delete && newlines)1591 die("deleted file %s still has contents", patch->old_name);1592 if (!patch->is_delete && !newlines && context)1593 fprintf(stderr, "** warning: file %s becomes empty but "1594 "is not deleted\n", patch->new_name);15951596 return offset;1597}15981599static inline int metadata_changes(struct patch *patch)1600{1601 return patch->is_rename > 0 ||1602 patch->is_copy > 0 ||1603 patch->is_new > 0 ||1604 patch->is_delete ||1605 (patch->old_mode && patch->new_mode &&1606 patch->old_mode != patch->new_mode);1607}16081609static char *inflate_it(const void *data, unsigned long size,1610 unsigned long inflated_size)1611{1612 z_stream stream;1613 void *out;1614 int st;16151616 memset(&stream, 0, sizeof(stream));16171618 stream.next_in = (unsigned char *)data;1619 stream.avail_in = size;1620 stream.next_out = out = xmalloc(inflated_size);1621 stream.avail_out = inflated_size;1622 git_inflate_init(&stream);1623 st = git_inflate(&stream, Z_FINISH);1624 git_inflate_end(&stream);1625 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1626 free(out);1627 return NULL;1628 }1629 return out;1630}16311632static struct fragment *parse_binary_hunk(char **buf_p,1633 unsigned long *sz_p,1634 int *status_p,1635 int *used_p)1636{1637 /*1638 * Expect a line that begins with binary patch method ("literal"1639 * or "delta"), followed by the length of data before deflating.1640 * a sequence of 'length-byte' followed by base-85 encoded data1641 * should follow, terminated by a newline.1642 *1643 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1644 * and we would limit the patch line to 66 characters,1645 * so one line can fit up to 13 groups that would decode1646 * to 52 bytes max. The length byte 'A'-'Z' corresponds1647 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1648 */1649 int llen, used;1650 unsigned long size = *sz_p;1651 char *buffer = *buf_p;1652 int patch_method;1653 unsigned long origlen;1654 char *data = NULL;1655 int hunk_size = 0;1656 struct fragment *frag;16571658 llen = linelen(buffer, size);1659 used = llen;16601661 *status_p = 0;16621663 if (!prefixcmp(buffer, "delta ")) {1664 patch_method = BINARY_DELTA_DEFLATED;1665 origlen = strtoul(buffer + 6, NULL, 10);1666 }1667 else if (!prefixcmp(buffer, "literal ")) {1668 patch_method = BINARY_LITERAL_DEFLATED;1669 origlen = strtoul(buffer + 8, NULL, 10);1670 }1671 else1672 return NULL;16731674 linenr++;1675 buffer += llen;1676 while (1) {1677 int byte_length, max_byte_length, newsize;1678 llen = linelen(buffer, size);1679 used += llen;1680 linenr++;1681 if (llen == 1) {1682 /* consume the blank line */1683 buffer++;1684 size--;1685 break;1686 }1687 /*1688 * Minimum line is "A00000\n" which is 7-byte long,1689 * and the line length must be multiple of 5 plus 2.1690 */1691 if ((llen < 7) || (llen-2) % 5)1692 goto corrupt;1693 max_byte_length = (llen - 2) / 5 * 4;1694 byte_length = *buffer;1695 if ('A' <= byte_length && byte_length <= 'Z')1696 byte_length = byte_length - 'A' + 1;1697 else if ('a' <= byte_length && byte_length <= 'z')1698 byte_length = byte_length - 'a' + 27;1699 else1700 goto corrupt;1701 /* if the input length was not multiple of 4, we would1702 * have filler at the end but the filler should never1703 * exceed 3 bytes1704 */1705 if (max_byte_length < byte_length ||1706 byte_length <= max_byte_length - 4)1707 goto corrupt;1708 newsize = hunk_size + byte_length;1709 data = xrealloc(data, newsize);1710 if (decode_85(data + hunk_size, buffer + 1, byte_length))1711 goto corrupt;1712 hunk_size = newsize;1713 buffer += llen;1714 size -= llen;1715 }17161717 frag = xcalloc(1, sizeof(*frag));1718 frag->patch = inflate_it(data, hunk_size, origlen);1719 if (!frag->patch)1720 goto corrupt;1721 free(data);1722 frag->size = origlen;1723 *buf_p = buffer;1724 *sz_p = size;1725 *used_p = used;1726 frag->binary_patch_method = patch_method;1727 return frag;17281729 corrupt:1730 free(data);1731 *status_p = -1;1732 error("corrupt binary patch at line %d: %.*s",1733 linenr-1, llen-1, buffer);1734 return NULL;1735}17361737static int parse_binary(char *buffer, unsigned long size, struct patch *patch)1738{1739 /*1740 * We have read "GIT binary patch\n"; what follows is a line1741 * that says the patch method (currently, either "literal" or1742 * "delta") and the length of data before deflating; a1743 * sequence of 'length-byte' followed by base-85 encoded data1744 * follows.1745 *1746 * When a binary patch is reversible, there is another binary1747 * hunk in the same format, starting with patch method (either1748 * "literal" or "delta") with the length of data, and a sequence1749 * of length-byte + base-85 encoded data, terminated with another1750 * empty line. This data, when applied to the postimage, produces1751 * the preimage.1752 */1753 struct fragment *forward;1754 struct fragment *reverse;1755 int status;1756 int used, used_1;17571758 forward = parse_binary_hunk(&buffer, &size, &status, &used);1759 if (!forward && !status)1760 /* there has to be one hunk (forward hunk) */1761 return error("unrecognized binary patch at line %d", linenr-1);1762 if (status)1763 /* otherwise we already gave an error message */1764 return status;17651766 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);1767 if (reverse)1768 used += used_1;1769 else if (status) {1770 /*1771 * Not having reverse hunk is not an error, but having1772 * a corrupt reverse hunk is.1773 */1774 free((void*) forward->patch);1775 free(forward);1776 return status;1777 }1778 forward->next = reverse;1779 patch->fragments = forward;1780 patch->is_binary = 1;1781 return used;1782}17831784static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)1785{1786 int hdrsize, patchsize;1787 int offset = find_header(buffer, size, &hdrsize, patch);17881789 if (offset < 0)1790 return offset;17911792 patch->ws_rule = whitespace_rule(patch->new_name1793 ? patch->new_name1794 : patch->old_name);17951796 patchsize = parse_single_patch(buffer + offset + hdrsize,1797 size - offset - hdrsize, patch);17981799 if (!patchsize) {1800 static const char *binhdr[] = {1801 "Binary files ",1802 "Files ",1803 NULL,1804 };1805 static const char git_binary[] = "GIT binary patch\n";1806 int i;1807 int hd = hdrsize + offset;1808 unsigned long llen = linelen(buffer + hd, size - hd);18091810 if (llen == sizeof(git_binary) - 1 &&1811 !memcmp(git_binary, buffer + hd, llen)) {1812 int used;1813 linenr++;1814 used = parse_binary(buffer + hd + llen,1815 size - hd - llen, patch);1816 if (used)1817 patchsize = used + llen;1818 else1819 patchsize = 0;1820 }1821 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {1822 for (i = 0; binhdr[i]; i++) {1823 int len = strlen(binhdr[i]);1824 if (len < size - hd &&1825 !memcmp(binhdr[i], buffer + hd, len)) {1826 linenr++;1827 patch->is_binary = 1;1828 patchsize = llen;1829 break;1830 }1831 }1832 }18331834 /* Empty patch cannot be applied if it is a text patch1835 * without metadata change. A binary patch appears1836 * empty to us here.1837 */1838 if ((apply || check) &&1839 (!patch->is_binary && !metadata_changes(patch)))1840 die("patch with only garbage at line %d", linenr);1841 }18421843 return offset + hdrsize + patchsize;1844}18451846#define swap(a,b) myswap((a),(b),sizeof(a))18471848#define myswap(a, b, size) do { \1849 unsigned char mytmp[size]; \1850 memcpy(mytmp, &a, size); \1851 memcpy(&a, &b, size); \1852 memcpy(&b, mytmp, size); \1853} while (0)18541855static void reverse_patches(struct patch *p)1856{1857 for (; p; p = p->next) {1858 struct fragment *frag = p->fragments;18591860 swap(p->new_name, p->old_name);1861 swap(p->new_mode, p->old_mode);1862 swap(p->is_new, p->is_delete);1863 swap(p->lines_added, p->lines_deleted);1864 swap(p->old_sha1_prefix, p->new_sha1_prefix);18651866 for (; frag; frag = frag->next) {1867 swap(frag->newpos, frag->oldpos);1868 swap(frag->newlines, frag->oldlines);1869 }1870 }1871}18721873static const char pluses[] =1874"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";1875static const char minuses[]=1876"----------------------------------------------------------------------";18771878static void show_stats(struct patch *patch)1879{1880 struct strbuf qname = STRBUF_INIT;1881 char *cp = patch->new_name ? patch->new_name : patch->old_name;1882 int max, add, del;18831884 quote_c_style(cp, &qname, NULL, 0);18851886 /*1887 * "scale" the filename1888 */1889 max = max_len;1890 if (max > 50)1891 max = 50;18921893 if (qname.len > max) {1894 cp = strchr(qname.buf + qname.len + 3 - max, '/');1895 if (!cp)1896 cp = qname.buf + qname.len + 3 - max;1897 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);1898 }18991900 if (patch->is_binary) {1901 printf(" %-*s | Bin\n", max, qname.buf);1902 strbuf_release(&qname);1903 return;1904 }19051906 printf(" %-*s |", max, qname.buf);1907 strbuf_release(&qname);19081909 /*1910 * scale the add/delete1911 */1912 max = max + max_change > 70 ? 70 - max : max_change;1913 add = patch->lines_added;1914 del = patch->lines_deleted;19151916 if (max_change > 0) {1917 int total = ((add + del) * max + max_change / 2) / max_change;1918 add = (add * max + max_change / 2) / max_change;1919 del = total - add;1920 }1921 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,1922 add, pluses, del, minuses);1923}19241925static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)1926{1927 switch (st->st_mode & S_IFMT) {1928 case S_IFLNK:1929 if (strbuf_readlink(buf, path, st->st_size) < 0)1930 return error("unable to read symlink %s", path);1931 return 0;1932 case S_IFREG:1933 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)1934 return error("unable to open or read %s", path);1935 convert_to_git(path, buf->buf, buf->len, buf, 0);1936 return 0;1937 default:1938 return -1;1939 }1940}19411942/*1943 * Update the preimage, and the common lines in postimage,1944 * from buffer buf of length len. If postlen is 0 the postimage1945 * is updated in place, otherwise it's updated on a new buffer1946 * of length postlen1947 */19481949static void update_pre_post_images(struct image *preimage,1950 struct image *postimage,1951 char *buf,1952 size_t len, size_t postlen)1953{1954 int i, ctx;1955 char *new, *old, *fixed;1956 struct image fixed_preimage;19571958 /*1959 * Update the preimage with whitespace fixes. Note that we1960 * are not losing preimage->buf -- apply_one_fragment() will1961 * free "oldlines".1962 */1963 prepare_image(&fixed_preimage, buf, len, 1);1964 assert(fixed_preimage.nr == preimage->nr);1965 for (i = 0; i < preimage->nr; i++)1966 fixed_preimage.line[i].flag = preimage->line[i].flag;1967 free(preimage->line_allocated);1968 *preimage = fixed_preimage;19691970 /*1971 * Adjust the common context lines in postimage. This can be1972 * done in-place when we are just doing whitespace fixing,1973 * which does not make the string grow, but needs a new buffer1974 * when ignoring whitespace causes the update, since in this case1975 * we could have e.g. tabs converted to multiple spaces.1976 * We trust the caller to tell us if the update can be done1977 * in place (postlen==0) or not.1978 */1979 old = postimage->buf;1980 if (postlen)1981 new = postimage->buf = xmalloc(postlen);1982 else1983 new = old;1984 fixed = preimage->buf;1985 for (i = ctx = 0; i < postimage->nr; i++) {1986 size_t len = postimage->line[i].len;1987 if (!(postimage->line[i].flag & LINE_COMMON)) {1988 /* an added line -- no counterparts in preimage */1989 memmove(new, old, len);1990 old += len;1991 new += len;1992 continue;1993 }19941995 /* a common context -- skip it in the original postimage */1996 old += len;19971998 /* and find the corresponding one in the fixed preimage */1999 while (ctx < preimage->nr &&2000 !(preimage->line[ctx].flag & LINE_COMMON)) {2001 fixed += preimage->line[ctx].len;2002 ctx++;2003 }2004 if (preimage->nr <= ctx)2005 die("oops");20062007 /* and copy it in, while fixing the line length */2008 len = preimage->line[ctx].len;2009 memcpy(new, fixed, len);2010 new += len;2011 fixed += len;2012 postimage->line[i].len = len;2013 ctx++;2014 }20152016 /* Fix the length of the whole thing */2017 postimage->len = new - postimage->buf;2018}20192020static int match_fragment(struct image *img,2021 struct image *preimage,2022 struct image *postimage,2023 unsigned long try,2024 int try_lno,2025 unsigned ws_rule,2026 int match_beginning, int match_end)2027{2028 int i;2029 char *fixed_buf, *buf, *orig, *target;2030 struct strbuf fixed;2031 size_t fixed_len;2032 int preimage_limit;20332034 if (preimage->nr + try_lno <= img->nr) {2035 /*2036 * The hunk falls within the boundaries of img.2037 */2038 preimage_limit = preimage->nr;2039 if (match_end && (preimage->nr + try_lno != img->nr))2040 return 0;2041 } else if (ws_error_action == correct_ws_error &&2042 (ws_rule & WS_BLANK_AT_EOF)) {2043 /*2044 * This hunk extends beyond the end of img, and we are2045 * removing blank lines at the end of the file. This2046 * many lines from the beginning of the preimage must2047 * match with img, and the remainder of the preimage2048 * must be blank.2049 */2050 preimage_limit = img->nr - try_lno;2051 } else {2052 /*2053 * The hunk extends beyond the end of the img and2054 * we are not removing blanks at the end, so we2055 * should reject the hunk at this position.2056 */2057 return 0;2058 }20592060 if (match_beginning && try_lno)2061 return 0;20622063 /* Quick hash check */2064 for (i = 0; i < preimage_limit; i++)2065 if (preimage->line[i].hash != img->line[try_lno + i].hash)2066 return 0;20672068 if (preimage_limit == preimage->nr) {2069 /*2070 * Do we have an exact match? If we were told to match2071 * at the end, size must be exactly at try+fragsize,2072 * otherwise try+fragsize must be still within the preimage,2073 * and either case, the old piece should match the preimage2074 * exactly.2075 */2076 if ((match_end2077 ? (try + preimage->len == img->len)2078 : (try + preimage->len <= img->len)) &&2079 !memcmp(img->buf + try, preimage->buf, preimage->len))2080 return 1;2081 } else {2082 /*2083 * The preimage extends beyond the end of img, so2084 * there cannot be an exact match.2085 *2086 * There must be one non-blank context line that match2087 * a line before the end of img.2088 */2089 char *buf_end;20902091 buf = preimage->buf;2092 buf_end = buf;2093 for (i = 0; i < preimage_limit; i++)2094 buf_end += preimage->line[i].len;20952096 for ( ; buf < buf_end; buf++)2097 if (!isspace(*buf))2098 break;2099 if (buf == buf_end)2100 return 0;2101 }21022103 /*2104 * No exact match. If we are ignoring whitespace, run a line-by-line2105 * fuzzy matching. We collect all the line length information because2106 * we need it to adjust whitespace if we match.2107 */2108 if (ws_ignore_action == ignore_ws_change) {2109 size_t imgoff = 0;2110 size_t preoff = 0;2111 size_t postlen = postimage->len;2112 size_t extra_chars;2113 char *preimage_eof;2114 char *preimage_end;2115 for (i = 0; i < preimage_limit; i++) {2116 size_t prelen = preimage->line[i].len;2117 size_t imglen = img->line[try_lno+i].len;21182119 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,2120 preimage->buf + preoff, prelen))2121 return 0;2122 if (preimage->line[i].flag & LINE_COMMON)2123 postlen += imglen - prelen;2124 imgoff += imglen;2125 preoff += prelen;2126 }21272128 /*2129 * Ok, the preimage matches with whitespace fuzz.2130 *2131 * imgoff now holds the true length of the target that2132 * matches the preimage before the end of the file.2133 *2134 * Count the number of characters in the preimage that fall2135 * beyond the end of the file and make sure that all of them2136 * are whitespace characters. (This can only happen if2137 * we are removing blank lines at the end of the file.)2138 */2139 buf = preimage_eof = preimage->buf + preoff;2140 for ( ; i < preimage->nr; i++)2141 preoff += preimage->line[i].len;2142 preimage_end = preimage->buf + preoff;2143 for ( ; buf < preimage_end; buf++)2144 if (!isspace(*buf))2145 return 0;21462147 /*2148 * Update the preimage and the common postimage context2149 * lines to use the same whitespace as the target.2150 * If whitespace is missing in the target (i.e.2151 * if the preimage extends beyond the end of the file),2152 * use the whitespace from the preimage.2153 */2154 extra_chars = preimage_end - preimage_eof;2155 strbuf_init(&fixed, imgoff + extra_chars);2156 strbuf_add(&fixed, img->buf + try, imgoff);2157 strbuf_add(&fixed, preimage_eof, extra_chars);2158 fixed_buf = strbuf_detach(&fixed, &fixed_len);2159 update_pre_post_images(preimage, postimage,2160 fixed_buf, fixed_len, postlen);2161 return 1;2162 }21632164 if (ws_error_action != correct_ws_error)2165 return 0;21662167 /*2168 * The hunk does not apply byte-by-byte, but the hash says2169 * it might with whitespace fuzz. We haven't been asked to2170 * ignore whitespace, we were asked to correct whitespace2171 * errors, so let's try matching after whitespace correction.2172 *2173 * The preimage may extend beyond the end of the file,2174 * but in this loop we will only handle the part of the2175 * preimage that falls within the file.2176 */2177 strbuf_init(&fixed, preimage->len + 1);2178 orig = preimage->buf;2179 target = img->buf + try;2180 for (i = 0; i < preimage_limit; i++) {2181 size_t oldlen = preimage->line[i].len;2182 size_t tgtlen = img->line[try_lno + i].len;2183 size_t fixstart = fixed.len;2184 struct strbuf tgtfix;2185 int match;21862187 /* Try fixing the line in the preimage */2188 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);21892190 /* Try fixing the line in the target */2191 strbuf_init(&tgtfix, tgtlen);2192 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);21932194 /*2195 * If they match, either the preimage was based on2196 * a version before our tree fixed whitespace breakage,2197 * or we are lacking a whitespace-fix patch the tree2198 * the preimage was based on already had (i.e. target2199 * has whitespace breakage, the preimage doesn't).2200 * In either case, we are fixing the whitespace breakages2201 * so we might as well take the fix together with their2202 * real change.2203 */2204 match = (tgtfix.len == fixed.len - fixstart &&2205 !memcmp(tgtfix.buf, fixed.buf + fixstart,2206 fixed.len - fixstart));22072208 strbuf_release(&tgtfix);2209 if (!match)2210 goto unmatch_exit;22112212 orig += oldlen;2213 target += tgtlen;2214 }221522162217 /*2218 * Now handle the lines in the preimage that falls beyond the2219 * end of the file (if any). They will only match if they are2220 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2221 * false).2222 */2223 for ( ; i < preimage->nr; i++) {2224 size_t fixstart = fixed.len; /* start of the fixed preimage */2225 size_t oldlen = preimage->line[i].len;2226 int j;22272228 /* Try fixing the line in the preimage */2229 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);22302231 for (j = fixstart; j < fixed.len; j++)2232 if (!isspace(fixed.buf[j]))2233 goto unmatch_exit;22342235 orig += oldlen;2236 }22372238 /*2239 * Yes, the preimage is based on an older version that still2240 * has whitespace breakages unfixed, and fixing them makes the2241 * hunk match. Update the context lines in the postimage.2242 */2243 fixed_buf = strbuf_detach(&fixed, &fixed_len);2244 update_pre_post_images(preimage, postimage,2245 fixed_buf, fixed_len, 0);2246 return 1;22472248 unmatch_exit:2249 strbuf_release(&fixed);2250 return 0;2251}22522253static int find_pos(struct image *img,2254 struct image *preimage,2255 struct image *postimage,2256 int line,2257 unsigned ws_rule,2258 int match_beginning, int match_end)2259{2260 int i;2261 unsigned long backwards, forwards, try;2262 int backwards_lno, forwards_lno, try_lno;22632264 /*2265 * If match_beginning or match_end is specified, there is no2266 * point starting from a wrong line that will never match and2267 * wander around and wait for a match at the specified end.2268 */2269 if (match_beginning)2270 line = 0;2271 else if (match_end)2272 line = img->nr - preimage->nr;22732274 /*2275 * Because the comparison is unsigned, the following test2276 * will also take care of a negative line number that can2277 * result when match_end and preimage is larger than the target.2278 */2279 if ((size_t) line > img->nr)2280 line = img->nr;22812282 try = 0;2283 for (i = 0; i < line; i++)2284 try += img->line[i].len;22852286 /*2287 * There's probably some smart way to do this, but I'll leave2288 * that to the smart and beautiful people. I'm simple and stupid.2289 */2290 backwards = try;2291 backwards_lno = line;2292 forwards = try;2293 forwards_lno = line;2294 try_lno = line;22952296 for (i = 0; ; i++) {2297 if (match_fragment(img, preimage, postimage,2298 try, try_lno, ws_rule,2299 match_beginning, match_end))2300 return try_lno;23012302 again:2303 if (backwards_lno == 0 && forwards_lno == img->nr)2304 break;23052306 if (i & 1) {2307 if (backwards_lno == 0) {2308 i++;2309 goto again;2310 }2311 backwards_lno--;2312 backwards -= img->line[backwards_lno].len;2313 try = backwards;2314 try_lno = backwards_lno;2315 } else {2316 if (forwards_lno == img->nr) {2317 i++;2318 goto again;2319 }2320 forwards += img->line[forwards_lno].len;2321 forwards_lno++;2322 try = forwards;2323 try_lno = forwards_lno;2324 }23252326 }2327 return -1;2328}23292330static void remove_first_line(struct image *img)2331{2332 img->buf += img->line[0].len;2333 img->len -= img->line[0].len;2334 img->line++;2335 img->nr--;2336}23372338static void remove_last_line(struct image *img)2339{2340 img->len -= img->line[--img->nr].len;2341}23422343static void update_image(struct image *img,2344 int applied_pos,2345 struct image *preimage,2346 struct image *postimage)2347{2348 /*2349 * remove the copy of preimage at offset in img2350 * and replace it with postimage2351 */2352 int i, nr;2353 size_t remove_count, insert_count, applied_at = 0;2354 char *result;2355 int preimage_limit;23562357 /*2358 * If we are removing blank lines at the end of img,2359 * the preimage may extend beyond the end.2360 * If that is the case, we must be careful only to2361 * remove the part of the preimage that falls within2362 * the boundaries of img. Initialize preimage_limit2363 * to the number of lines in the preimage that falls2364 * within the boundaries.2365 */2366 preimage_limit = preimage->nr;2367 if (preimage_limit > img->nr - applied_pos)2368 preimage_limit = img->nr - applied_pos;23692370 for (i = 0; i < applied_pos; i++)2371 applied_at += img->line[i].len;23722373 remove_count = 0;2374 for (i = 0; i < preimage_limit; i++)2375 remove_count += img->line[applied_pos + i].len;2376 insert_count = postimage->len;23772378 /* Adjust the contents */2379 result = xmalloc(img->len + insert_count - remove_count + 1);2380 memcpy(result, img->buf, applied_at);2381 memcpy(result + applied_at, postimage->buf, postimage->len);2382 memcpy(result + applied_at + postimage->len,2383 img->buf + (applied_at + remove_count),2384 img->len - (applied_at + remove_count));2385 free(img->buf);2386 img->buf = result;2387 img->len += insert_count - remove_count;2388 result[img->len] = '\0';23892390 /* Adjust the line table */2391 nr = img->nr + postimage->nr - preimage_limit;2392 if (preimage_limit < postimage->nr) {2393 /*2394 * NOTE: this knows that we never call remove_first_line()2395 * on anything other than pre/post image.2396 */2397 img->line = xrealloc(img->line, nr * sizeof(*img->line));2398 img->line_allocated = img->line;2399 }2400 if (preimage_limit != postimage->nr)2401 memmove(img->line + applied_pos + postimage->nr,2402 img->line + applied_pos + preimage_limit,2403 (img->nr - (applied_pos + preimage_limit)) *2404 sizeof(*img->line));2405 memcpy(img->line + applied_pos,2406 postimage->line,2407 postimage->nr * sizeof(*img->line));2408 img->nr = nr;2409}24102411static int apply_one_fragment(struct image *img, struct fragment *frag,2412 int inaccurate_eof, unsigned ws_rule)2413{2414 int match_beginning, match_end;2415 const char *patch = frag->patch;2416 int size = frag->size;2417 char *old, *oldlines;2418 struct strbuf newlines;2419 int new_blank_lines_at_end = 0;2420 unsigned long leading, trailing;2421 int pos, applied_pos;2422 struct image preimage;2423 struct image postimage;24242425 memset(&preimage, 0, sizeof(preimage));2426 memset(&postimage, 0, sizeof(postimage));2427 oldlines = xmalloc(size);2428 strbuf_init(&newlines, size);24292430 old = oldlines;2431 while (size > 0) {2432 char first;2433 int len = linelen(patch, size);2434 int plen;2435 int added_blank_line = 0;2436 int is_blank_context = 0;2437 size_t start;24382439 if (!len)2440 break;24412442 /*2443 * "plen" is how much of the line we should use for2444 * the actual patch data. Normally we just remove the2445 * first character on the line, but if the line is2446 * followed by "\ No newline", then we also remove the2447 * last one (which is the newline, of course).2448 */2449 plen = len - 1;2450 if (len < size && patch[len] == '\\')2451 plen--;2452 first = *patch;2453 if (apply_in_reverse) {2454 if (first == '-')2455 first = '+';2456 else if (first == '+')2457 first = '-';2458 }24592460 switch (first) {2461 case '\n':2462 /* Newer GNU diff, empty context line */2463 if (plen < 0)2464 /* ... followed by '\No newline'; nothing */2465 break;2466 *old++ = '\n';2467 strbuf_addch(&newlines, '\n');2468 add_line_info(&preimage, "\n", 1, LINE_COMMON);2469 add_line_info(&postimage, "\n", 1, LINE_COMMON);2470 is_blank_context = 1;2471 break;2472 case ' ':2473 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2474 ws_blank_line(patch + 1, plen, ws_rule))2475 is_blank_context = 1;2476 case '-':2477 memcpy(old, patch + 1, plen);2478 add_line_info(&preimage, old, plen,2479 (first == ' ' ? LINE_COMMON : 0));2480 old += plen;2481 if (first == '-')2482 break;2483 /* Fall-through for ' ' */2484 case '+':2485 /* --no-add does not add new lines */2486 if (first == '+' && no_add)2487 break;24882489 start = newlines.len;2490 if (first != '+' ||2491 !whitespace_error ||2492 ws_error_action != correct_ws_error) {2493 strbuf_add(&newlines, patch + 1, plen);2494 }2495 else {2496 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);2497 }2498 add_line_info(&postimage, newlines.buf + start, newlines.len - start,2499 (first == '+' ? 0 : LINE_COMMON));2500 if (first == '+' &&2501 (ws_rule & WS_BLANK_AT_EOF) &&2502 ws_blank_line(patch + 1, plen, ws_rule))2503 added_blank_line = 1;2504 break;2505 case '@': case '\\':2506 /* Ignore it, we already handled it */2507 break;2508 default:2509 if (apply_verbosely)2510 error("invalid start of line: '%c'", first);2511 return -1;2512 }2513 if (added_blank_line)2514 new_blank_lines_at_end++;2515 else if (is_blank_context)2516 ;2517 else2518 new_blank_lines_at_end = 0;2519 patch += len;2520 size -= len;2521 }2522 if (inaccurate_eof &&2523 old > oldlines && old[-1] == '\n' &&2524 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {2525 old--;2526 strbuf_setlen(&newlines, newlines.len - 1);2527 }25282529 leading = frag->leading;2530 trailing = frag->trailing;25312532 /*2533 * A hunk to change lines at the beginning would begin with2534 * @@ -1,L +N,M @@2535 * but we need to be careful. -U0 that inserts before the second2536 * line also has this pattern.2537 *2538 * And a hunk to add to an empty file would begin with2539 * @@ -0,0 +N,M @@2540 *2541 * In other words, a hunk that is (frag->oldpos <= 1) with or2542 * without leading context must match at the beginning.2543 */2544 match_beginning = (!frag->oldpos ||2545 (frag->oldpos == 1 && !unidiff_zero));25462547 /*2548 * A hunk without trailing lines must match at the end.2549 * However, we simply cannot tell if a hunk must match end2550 * from the lack of trailing lines if the patch was generated2551 * with unidiff without any context.2552 */2553 match_end = !unidiff_zero && !trailing;25542555 pos = frag->newpos ? (frag->newpos - 1) : 0;2556 preimage.buf = oldlines;2557 preimage.len = old - oldlines;2558 postimage.buf = newlines.buf;2559 postimage.len = newlines.len;2560 preimage.line = preimage.line_allocated;2561 postimage.line = postimage.line_allocated;25622563 for (;;) {25642565 applied_pos = find_pos(img, &preimage, &postimage, pos,2566 ws_rule, match_beginning, match_end);25672568 if (applied_pos >= 0)2569 break;25702571 /* Am I at my context limits? */2572 if ((leading <= p_context) && (trailing <= p_context))2573 break;2574 if (match_beginning || match_end) {2575 match_beginning = match_end = 0;2576 continue;2577 }25782579 /*2580 * Reduce the number of context lines; reduce both2581 * leading and trailing if they are equal otherwise2582 * just reduce the larger context.2583 */2584 if (leading >= trailing) {2585 remove_first_line(&preimage);2586 remove_first_line(&postimage);2587 pos--;2588 leading--;2589 }2590 if (trailing > leading) {2591 remove_last_line(&preimage);2592 remove_last_line(&postimage);2593 trailing--;2594 }2595 }25962597 if (applied_pos >= 0) {2598 if (new_blank_lines_at_end &&2599 preimage.nr + applied_pos >= img->nr &&2600 (ws_rule & WS_BLANK_AT_EOF) &&2601 ws_error_action != nowarn_ws_error) {2602 record_ws_error(WS_BLANK_AT_EOF, "+", 1, frag->linenr);2603 if (ws_error_action == correct_ws_error) {2604 while (new_blank_lines_at_end--)2605 remove_last_line(&postimage);2606 }2607 /*2608 * We would want to prevent write_out_results()2609 * from taking place in apply_patch() that follows2610 * the callchain led us here, which is:2611 * apply_patch->check_patch_list->check_patch->2612 * apply_data->apply_fragments->apply_one_fragment2613 */2614 if (ws_error_action == die_on_ws_error)2615 apply = 0;2616 }26172618 /*2619 * Warn if it was necessary to reduce the number2620 * of context lines.2621 */2622 if ((leading != frag->leading) ||2623 (trailing != frag->trailing))2624 fprintf(stderr, "Context reduced to (%ld/%ld)"2625 " to apply fragment at %d\n",2626 leading, trailing, applied_pos+1);2627 update_image(img, applied_pos, &preimage, &postimage);2628 } else {2629 if (apply_verbosely)2630 error("while searching for:\n%.*s",2631 (int)(old - oldlines), oldlines);2632 }26332634 free(oldlines);2635 strbuf_release(&newlines);2636 free(preimage.line_allocated);2637 free(postimage.line_allocated);26382639 return (applied_pos < 0);2640}26412642static int apply_binary_fragment(struct image *img, struct patch *patch)2643{2644 struct fragment *fragment = patch->fragments;2645 unsigned long len;2646 void *dst;26472648 /* Binary patch is irreversible without the optional second hunk */2649 if (apply_in_reverse) {2650 if (!fragment->next)2651 return error("cannot reverse-apply a binary patch "2652 "without the reverse hunk to '%s'",2653 patch->new_name2654 ? patch->new_name : patch->old_name);2655 fragment = fragment->next;2656 }2657 switch (fragment->binary_patch_method) {2658 case BINARY_DELTA_DEFLATED:2659 dst = patch_delta(img->buf, img->len, fragment->patch,2660 fragment->size, &len);2661 if (!dst)2662 return -1;2663 clear_image(img);2664 img->buf = dst;2665 img->len = len;2666 return 0;2667 case BINARY_LITERAL_DEFLATED:2668 clear_image(img);2669 img->len = fragment->size;2670 img->buf = xmalloc(img->len+1);2671 memcpy(img->buf, fragment->patch, img->len);2672 img->buf[img->len] = '\0';2673 return 0;2674 }2675 return -1;2676}26772678static int apply_binary(struct image *img, struct patch *patch)2679{2680 const char *name = patch->old_name ? patch->old_name : patch->new_name;2681 unsigned char sha1[20];26822683 /*2684 * For safety, we require patch index line to contain2685 * full 40-byte textual SHA1 for old and new, at least for now.2686 */2687 if (strlen(patch->old_sha1_prefix) != 40 ||2688 strlen(patch->new_sha1_prefix) != 40 ||2689 get_sha1_hex(patch->old_sha1_prefix, sha1) ||2690 get_sha1_hex(patch->new_sha1_prefix, sha1))2691 return error("cannot apply binary patch to '%s' "2692 "without full index line", name);26932694 if (patch->old_name) {2695 /*2696 * See if the old one matches what the patch2697 * applies to.2698 */2699 hash_sha1_file(img->buf, img->len, blob_type, sha1);2700 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))2701 return error("the patch applies to '%s' (%s), "2702 "which does not match the "2703 "current contents.",2704 name, sha1_to_hex(sha1));2705 }2706 else {2707 /* Otherwise, the old one must be empty. */2708 if (img->len)2709 return error("the patch applies to an empty "2710 "'%s' but it is not empty", name);2711 }27122713 get_sha1_hex(patch->new_sha1_prefix, sha1);2714 if (is_null_sha1(sha1)) {2715 clear_image(img);2716 return 0; /* deletion patch */2717 }27182719 if (has_sha1_file(sha1)) {2720 /* We already have the postimage */2721 enum object_type type;2722 unsigned long size;2723 char *result;27242725 result = read_sha1_file(sha1, &type, &size);2726 if (!result)2727 return error("the necessary postimage %s for "2728 "'%s' cannot be read",2729 patch->new_sha1_prefix, name);2730 clear_image(img);2731 img->buf = result;2732 img->len = size;2733 } else {2734 /*2735 * We have verified buf matches the preimage;2736 * apply the patch data to it, which is stored2737 * in the patch->fragments->{patch,size}.2738 */2739 if (apply_binary_fragment(img, patch))2740 return error("binary patch does not apply to '%s'",2741 name);27422743 /* verify that the result matches */2744 hash_sha1_file(img->buf, img->len, blob_type, sha1);2745 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))2746 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",2747 name, patch->new_sha1_prefix, sha1_to_hex(sha1));2748 }27492750 return 0;2751}27522753static int apply_fragments(struct image *img, struct patch *patch)2754{2755 struct fragment *frag = patch->fragments;2756 const char *name = patch->old_name ? patch->old_name : patch->new_name;2757 unsigned ws_rule = patch->ws_rule;2758 unsigned inaccurate_eof = patch->inaccurate_eof;27592760 if (patch->is_binary)2761 return apply_binary(img, patch);27622763 while (frag) {2764 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule)) {2765 error("patch failed: %s:%ld", name, frag->oldpos);2766 if (!apply_with_reject)2767 return -1;2768 frag->rejected = 1;2769 }2770 frag = frag->next;2771 }2772 return 0;2773}27742775static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)2776{2777 if (!ce)2778 return 0;27792780 if (S_ISGITLINK(ce->ce_mode)) {2781 strbuf_grow(buf, 100);2782 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));2783 } else {2784 enum object_type type;2785 unsigned long sz;2786 char *result;27872788 result = read_sha1_file(ce->sha1, &type, &sz);2789 if (!result)2790 return -1;2791 /* XXX read_sha1_file NUL-terminates */2792 strbuf_attach(buf, result, sz, sz + 1);2793 }2794 return 0;2795}27962797static struct patch *in_fn_table(const char *name)2798{2799 struct string_list_item *item;28002801 if (name == NULL)2802 return NULL;28032804 item = string_list_lookup(&fn_table, name);2805 if (item != NULL)2806 return (struct patch *)item->util;28072808 return NULL;2809}28102811/*2812 * item->util in the filename table records the status of the path.2813 * Usually it points at a patch (whose result records the contents2814 * of it after applying it), but it could be PATH_WAS_DELETED for a2815 * path that a previously applied patch has already removed.2816 */2817 #define PATH_TO_BE_DELETED ((struct patch *) -2)2818#define PATH_WAS_DELETED ((struct patch *) -1)28192820static int to_be_deleted(struct patch *patch)2821{2822 return patch == PATH_TO_BE_DELETED;2823}28242825static int was_deleted(struct patch *patch)2826{2827 return patch == PATH_WAS_DELETED;2828}28292830static void add_to_fn_table(struct patch *patch)2831{2832 struct string_list_item *item;28332834 /*2835 * Always add new_name unless patch is a deletion2836 * This should cover the cases for normal diffs,2837 * file creations and copies2838 */2839 if (patch->new_name != NULL) {2840 item = string_list_insert(&fn_table, patch->new_name);2841 item->util = patch;2842 }28432844 /*2845 * store a failure on rename/deletion cases because2846 * later chunks shouldn't patch old names2847 */2848 if ((patch->new_name == NULL) || (patch->is_rename)) {2849 item = string_list_insert(&fn_table, patch->old_name);2850 item->util = PATH_WAS_DELETED;2851 }2852}28532854static void prepare_fn_table(struct patch *patch)2855{2856 /*2857 * store information about incoming file deletion2858 */2859 while (patch) {2860 if ((patch->new_name == NULL) || (patch->is_rename)) {2861 struct string_list_item *item;2862 item = string_list_insert(&fn_table, patch->old_name);2863 item->util = PATH_TO_BE_DELETED;2864 }2865 patch = patch->next;2866 }2867}28682869static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)2870{2871 struct strbuf buf = STRBUF_INIT;2872 struct image image;2873 size_t len;2874 char *img;2875 struct patch *tpatch;28762877 if (!(patch->is_copy || patch->is_rename) &&2878 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {2879 if (was_deleted(tpatch)) {2880 return error("patch %s has been renamed/deleted",2881 patch->old_name);2882 }2883 /* We have a patched copy in memory use that */2884 strbuf_add(&buf, tpatch->result, tpatch->resultsize);2885 } else if (cached) {2886 if (read_file_or_gitlink(ce, &buf))2887 return error("read of %s failed", patch->old_name);2888 } else if (patch->old_name) {2889 if (S_ISGITLINK(patch->old_mode)) {2890 if (ce) {2891 read_file_or_gitlink(ce, &buf);2892 } else {2893 /*2894 * There is no way to apply subproject2895 * patch without looking at the index.2896 */2897 patch->fragments = NULL;2898 }2899 } else {2900 if (read_old_data(st, patch->old_name, &buf))2901 return error("read of %s failed", patch->old_name);2902 }2903 }29042905 img = strbuf_detach(&buf, &len);2906 prepare_image(&image, img, len, !patch->is_binary);29072908 if (apply_fragments(&image, patch) < 0)2909 return -1; /* note with --reject this succeeds. */2910 patch->result = image.buf;2911 patch->resultsize = image.len;2912 add_to_fn_table(patch);2913 free(image.line_allocated);29142915 if (0 < patch->is_delete && patch->resultsize)2916 return error("removal patch leaves file contents");29172918 return 0;2919}29202921static int check_to_create_blob(const char *new_name, int ok_if_exists)2922{2923 struct stat nst;2924 if (!lstat(new_name, &nst)) {2925 if (S_ISDIR(nst.st_mode) || ok_if_exists)2926 return 0;2927 /*2928 * A leading component of new_name might be a symlink2929 * that is going to be removed with this patch, but2930 * still pointing at somewhere that has the path.2931 * In such a case, path "new_name" does not exist as2932 * far as git is concerned.2933 */2934 if (has_symlink_leading_path(new_name, strlen(new_name)))2935 return 0;29362937 return error("%s: already exists in working directory", new_name);2938 }2939 else if ((errno != ENOENT) && (errno != ENOTDIR))2940 return error("%s: %s", new_name, strerror(errno));2941 return 0;2942}29432944static int verify_index_match(struct cache_entry *ce, struct stat *st)2945{2946 if (S_ISGITLINK(ce->ce_mode)) {2947 if (!S_ISDIR(st->st_mode))2948 return -1;2949 return 0;2950 }2951 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);2952}29532954static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)2955{2956 const char *old_name = patch->old_name;2957 struct patch *tpatch = NULL;2958 int stat_ret = 0;2959 unsigned st_mode = 0;29602961 /*2962 * Make sure that we do not have local modifications from the2963 * index when we are looking at the index. Also make sure2964 * we have the preimage file to be patched in the work tree,2965 * unless --cached, which tells git to apply only in the index.2966 */2967 if (!old_name)2968 return 0;29692970 assert(patch->is_new <= 0);29712972 if (!(patch->is_copy || patch->is_rename) &&2973 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {2974 if (was_deleted(tpatch))2975 return error("%s: has been deleted/renamed", old_name);2976 st_mode = tpatch->new_mode;2977 } else if (!cached) {2978 stat_ret = lstat(old_name, st);2979 if (stat_ret && errno != ENOENT)2980 return error("%s: %s", old_name, strerror(errno));2981 }29822983 if (to_be_deleted(tpatch))2984 tpatch = NULL;29852986 if (check_index && !tpatch) {2987 int pos = cache_name_pos(old_name, strlen(old_name));2988 if (pos < 0) {2989 if (patch->is_new < 0)2990 goto is_new;2991 return error("%s: does not exist in index", old_name);2992 }2993 *ce = active_cache[pos];2994 if (stat_ret < 0) {2995 struct checkout costate;2996 /* checkout */2997 memset(&costate, 0, sizeof(costate));2998 costate.base_dir = "";2999 costate.refresh_cache = 1;3000 if (checkout_entry(*ce, &costate, NULL) ||3001 lstat(old_name, st))3002 return -1;3003 }3004 if (!cached && verify_index_match(*ce, st))3005 return error("%s: does not match index", old_name);3006 if (cached)3007 st_mode = (*ce)->ce_mode;3008 } else if (stat_ret < 0) {3009 if (patch->is_new < 0)3010 goto is_new;3011 return error("%s: %s", old_name, strerror(errno));3012 }30133014 if (!cached && !tpatch)3015 st_mode = ce_mode_from_stat(*ce, st->st_mode);30163017 if (patch->is_new < 0)3018 patch->is_new = 0;3019 if (!patch->old_mode)3020 patch->old_mode = st_mode;3021 if ((st_mode ^ patch->old_mode) & S_IFMT)3022 return error("%s: wrong type", old_name);3023 if (st_mode != patch->old_mode)3024 warning("%s has type %o, expected %o",3025 old_name, st_mode, patch->old_mode);3026 if (!patch->new_mode && !patch->is_delete)3027 patch->new_mode = st_mode;3028 return 0;30293030 is_new:3031 patch->is_new = 1;3032 patch->is_delete = 0;3033 patch->old_name = NULL;3034 return 0;3035}30363037static int check_patch(struct patch *patch)3038{3039 struct stat st;3040 const char *old_name = patch->old_name;3041 const char *new_name = patch->new_name;3042 const char *name = old_name ? old_name : new_name;3043 struct cache_entry *ce = NULL;3044 struct patch *tpatch;3045 int ok_if_exists;3046 int status;30473048 patch->rejected = 1; /* we will drop this after we succeed */30493050 status = check_preimage(patch, &ce, &st);3051 if (status)3052 return status;3053 old_name = patch->old_name;30543055 if ((tpatch = in_fn_table(new_name)) &&3056 (was_deleted(tpatch) || to_be_deleted(tpatch)))3057 /*3058 * A type-change diff is always split into a patch to3059 * delete old, immediately followed by a patch to3060 * create new (see diff.c::run_diff()); in such a case3061 * it is Ok that the entry to be deleted by the3062 * previous patch is still in the working tree and in3063 * the index.3064 */3065 ok_if_exists = 1;3066 else3067 ok_if_exists = 0;30683069 if (new_name &&3070 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {3071 if (check_index &&3072 cache_name_pos(new_name, strlen(new_name)) >= 0 &&3073 !ok_if_exists)3074 return error("%s: already exists in index", new_name);3075 if (!cached) {3076 int err = check_to_create_blob(new_name, ok_if_exists);3077 if (err)3078 return err;3079 }3080 if (!patch->new_mode) {3081 if (0 < patch->is_new)3082 patch->new_mode = S_IFREG | 0644;3083 else3084 patch->new_mode = patch->old_mode;3085 }3086 }30873088 if (new_name && old_name) {3089 int same = !strcmp(old_name, new_name);3090 if (!patch->new_mode)3091 patch->new_mode = patch->old_mode;3092 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)3093 return error("new mode (%o) of %s does not match old mode (%o)%s%s",3094 patch->new_mode, new_name, patch->old_mode,3095 same ? "" : " of ", same ? "" : old_name);3096 }30973098 if (apply_data(patch, &st, ce) < 0)3099 return error("%s: patch does not apply", name);3100 patch->rejected = 0;3101 return 0;3102}31033104static int check_patch_list(struct patch *patch)3105{3106 int err = 0;31073108 prepare_fn_table(patch);3109 while (patch) {3110 if (apply_verbosely)3111 say_patch_name(stderr,3112 "Checking patch ", patch, "...\n");3113 err |= check_patch(patch);3114 patch = patch->next;3115 }3116 return err;3117}31183119/* This function tries to read the sha1 from the current index */3120static int get_current_sha1(const char *path, unsigned char *sha1)3121{3122 int pos;31233124 if (read_cache() < 0)3125 return -1;3126 pos = cache_name_pos(path, strlen(path));3127 if (pos < 0)3128 return -1;3129 hashcpy(sha1, active_cache[pos]->sha1);3130 return 0;3131}31323133/* Build an index that contains the just the files needed for a 3way merge */3134static void build_fake_ancestor(struct patch *list, const char *filename)3135{3136 struct patch *patch;3137 struct index_state result = { NULL };3138 int fd;31393140 /* Once we start supporting the reverse patch, it may be3141 * worth showing the new sha1 prefix, but until then...3142 */3143 for (patch = list; patch; patch = patch->next) {3144 const unsigned char *sha1_ptr;3145 unsigned char sha1[20];3146 struct cache_entry *ce;3147 const char *name;31483149 name = patch->old_name ? patch->old_name : patch->new_name;3150 if (0 < patch->is_new)3151 continue;3152 else if (get_sha1(patch->old_sha1_prefix, sha1))3153 /* git diff has no index line for mode/type changes */3154 if (!patch->lines_added && !patch->lines_deleted) {3155 if (get_current_sha1(patch->old_name, sha1))3156 die("mode change for %s, which is not "3157 "in current HEAD", name);3158 sha1_ptr = sha1;3159 } else3160 die("sha1 information is lacking or useless "3161 "(%s).", name);3162 else3163 sha1_ptr = sha1;31643165 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);3166 if (!ce)3167 die("make_cache_entry failed for path '%s'", name);3168 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3169 die ("Could not add %s to temporary index", name);3170 }31713172 fd = open(filename, O_WRONLY | O_CREAT, 0666);3173 if (fd < 0 || write_index(&result, fd) || close(fd))3174 die ("Could not write temporary index to %s", filename);31753176 discard_index(&result);3177}31783179static void stat_patch_list(struct patch *patch)3180{3181 int files, adds, dels;31823183 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3184 files++;3185 adds += patch->lines_added;3186 dels += patch->lines_deleted;3187 show_stats(patch);3188 }31893190 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);3191}31923193static void numstat_patch_list(struct patch *patch)3194{3195 for ( ; patch; patch = patch->next) {3196 const char *name;3197 name = patch->new_name ? patch->new_name : patch->old_name;3198 if (patch->is_binary)3199 printf("-\t-\t");3200 else3201 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3202 write_name_quoted(name, stdout, line_termination);3203 }3204}32053206static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3207{3208 if (mode)3209 printf(" %s mode %06o %s\n", newdelete, mode, name);3210 else3211 printf(" %s %s\n", newdelete, name);3212}32133214static void show_mode_change(struct patch *p, int show_name)3215{3216 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3217 if (show_name)3218 printf(" mode change %06o => %06o %s\n",3219 p->old_mode, p->new_mode, p->new_name);3220 else3221 printf(" mode change %06o => %06o\n",3222 p->old_mode, p->new_mode);3223 }3224}32253226static void show_rename_copy(struct patch *p)3227{3228 const char *renamecopy = p->is_rename ? "rename" : "copy";3229 const char *old, *new;32303231 /* Find common prefix */3232 old = p->old_name;3233 new = p->new_name;3234 while (1) {3235 const char *slash_old, *slash_new;3236 slash_old = strchr(old, '/');3237 slash_new = strchr(new, '/');3238 if (!slash_old ||3239 !slash_new ||3240 slash_old - old != slash_new - new ||3241 memcmp(old, new, slash_new - new))3242 break;3243 old = slash_old + 1;3244 new = slash_new + 1;3245 }3246 /* p->old_name thru old is the common prefix, and old and new3247 * through the end of names are renames3248 */3249 if (old != p->old_name)3250 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,3251 (int)(old - p->old_name), p->old_name,3252 old, new, p->score);3253 else3254 printf(" %s %s => %s (%d%%)\n", renamecopy,3255 p->old_name, p->new_name, p->score);3256 show_mode_change(p, 0);3257}32583259static void summary_patch_list(struct patch *patch)3260{3261 struct patch *p;32623263 for (p = patch; p; p = p->next) {3264 if (p->is_new)3265 show_file_mode_name("create", p->new_mode, p->new_name);3266 else if (p->is_delete)3267 show_file_mode_name("delete", p->old_mode, p->old_name);3268 else {3269 if (p->is_rename || p->is_copy)3270 show_rename_copy(p);3271 else {3272 if (p->score) {3273 printf(" rewrite %s (%d%%)\n",3274 p->new_name, p->score);3275 show_mode_change(p, 0);3276 }3277 else3278 show_mode_change(p, 1);3279 }3280 }3281 }3282}32833284static void patch_stats(struct patch *patch)3285{3286 int lines = patch->lines_added + patch->lines_deleted;32873288 if (lines > max_change)3289 max_change = lines;3290 if (patch->old_name) {3291 int len = quote_c_style(patch->old_name, NULL, NULL, 0);3292 if (!len)3293 len = strlen(patch->old_name);3294 if (len > max_len)3295 max_len = len;3296 }3297 if (patch->new_name) {3298 int len = quote_c_style(patch->new_name, NULL, NULL, 0);3299 if (!len)3300 len = strlen(patch->new_name);3301 if (len > max_len)3302 max_len = len;3303 }3304}33053306static void remove_file(struct patch *patch, int rmdir_empty)3307{3308 if (update_index) {3309 if (remove_file_from_cache(patch->old_name) < 0)3310 die("unable to remove %s from index", patch->old_name);3311 }3312 if (!cached) {3313 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {3314 remove_path(patch->old_name);3315 }3316 }3317}33183319static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)3320{3321 struct stat st;3322 struct cache_entry *ce;3323 int namelen = strlen(path);3324 unsigned ce_size = cache_entry_size(namelen);33253326 if (!update_index)3327 return;33283329 ce = xcalloc(1, ce_size);3330 memcpy(ce->name, path, namelen);3331 ce->ce_mode = create_ce_mode(mode);3332 ce->ce_flags = namelen;3333 if (S_ISGITLINK(mode)) {3334 const char *s = buf;33353336 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))3337 die("corrupt patch for subproject %s", path);3338 } else {3339 if (!cached) {3340 if (lstat(path, &st) < 0)3341 die_errno("unable to stat newly created file '%s'",3342 path);3343 fill_stat_cache_info(ce, &st);3344 }3345 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)3346 die("unable to create backing store for newly created file %s", path);3347 }3348 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)3349 die("unable to add cache entry for %s", path);3350}33513352static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)3353{3354 int fd;3355 struct strbuf nbuf = STRBUF_INIT;33563357 if (S_ISGITLINK(mode)) {3358 struct stat st;3359 if (!lstat(path, &st) && S_ISDIR(st.st_mode))3360 return 0;3361 return mkdir(path, 0777);3362 }33633364 if (has_symlinks && S_ISLNK(mode))3365 /* Although buf:size is counted string, it also is NUL3366 * terminated.3367 */3368 return symlink(buf, path);33693370 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);3371 if (fd < 0)3372 return -1;33733374 if (convert_to_working_tree(path, buf, size, &nbuf)) {3375 size = nbuf.len;3376 buf = nbuf.buf;3377 }3378 write_or_die(fd, buf, size);3379 strbuf_release(&nbuf);33803381 if (close(fd) < 0)3382 die_errno("closing file '%s'", path);3383 return 0;3384}33853386/*3387 * We optimistically assume that the directories exist,3388 * which is true 99% of the time anyway. If they don't,3389 * we create them and try again.3390 */3391static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)3392{3393 if (cached)3394 return;3395 if (!try_create_file(path, mode, buf, size))3396 return;33973398 if (errno == ENOENT) {3399 if (safe_create_leading_directories(path))3400 return;3401 if (!try_create_file(path, mode, buf, size))3402 return;3403 }34043405 if (errno == EEXIST || errno == EACCES) {3406 /* We may be trying to create a file where a directory3407 * used to be.3408 */3409 struct stat st;3410 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))3411 errno = EEXIST;3412 }34133414 if (errno == EEXIST) {3415 unsigned int nr = getpid();34163417 for (;;) {3418 char newpath[PATH_MAX];3419 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);3420 if (!try_create_file(newpath, mode, buf, size)) {3421 if (!rename(newpath, path))3422 return;3423 unlink_or_warn(newpath);3424 break;3425 }3426 if (errno != EEXIST)3427 break;3428 ++nr;3429 }3430 }3431 die_errno("unable to write file '%s' mode %o", path, mode);3432}34333434static void create_file(struct patch *patch)3435{3436 char *path = patch->new_name;3437 unsigned mode = patch->new_mode;3438 unsigned long size = patch->resultsize;3439 char *buf = patch->result;34403441 if (!mode)3442 mode = S_IFREG | 0644;3443 create_one_file(path, mode, buf, size);3444 add_index_file(path, mode, buf, size);3445}34463447/* phase zero is to remove, phase one is to create */3448static void write_out_one_result(struct patch *patch, int phase)3449{3450 if (patch->is_delete > 0) {3451 if (phase == 0)3452 remove_file(patch, 1);3453 return;3454 }3455 if (patch->is_new > 0 || patch->is_copy) {3456 if (phase == 1)3457 create_file(patch);3458 return;3459 }3460 /*3461 * Rename or modification boils down to the same3462 * thing: remove the old, write the new3463 */3464 if (phase == 0)3465 remove_file(patch, patch->is_rename);3466 if (phase == 1)3467 create_file(patch);3468}34693470static int write_out_one_reject(struct patch *patch)3471{3472 FILE *rej;3473 char namebuf[PATH_MAX];3474 struct fragment *frag;3475 int cnt = 0;34763477 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {3478 if (!frag->rejected)3479 continue;3480 cnt++;3481 }34823483 if (!cnt) {3484 if (apply_verbosely)3485 say_patch_name(stderr,3486 "Applied patch ", patch, " cleanly.\n");3487 return 0;3488 }34893490 /* This should not happen, because a removal patch that leaves3491 * contents are marked "rejected" at the patch level.3492 */3493 if (!patch->new_name)3494 die("internal error");34953496 /* Say this even without --verbose */3497 say_patch_name(stderr, "Applying patch ", patch, " with");3498 fprintf(stderr, " %d rejects...\n", cnt);34993500 cnt = strlen(patch->new_name);3501 if (ARRAY_SIZE(namebuf) <= cnt + 5) {3502 cnt = ARRAY_SIZE(namebuf) - 5;3503 warning("truncating .rej filename to %.*s.rej",3504 cnt - 1, patch->new_name);3505 }3506 memcpy(namebuf, patch->new_name, cnt);3507 memcpy(namebuf + cnt, ".rej", 5);35083509 rej = fopen(namebuf, "w");3510 if (!rej)3511 return error("cannot open %s: %s", namebuf, strerror(errno));35123513 /* Normal git tools never deal with .rej, so do not pretend3514 * this is a git patch by saying --git nor give extended3515 * headers. While at it, maybe please "kompare" that wants3516 * the trailing TAB and some garbage at the end of line ;-).3517 */3518 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",3519 patch->new_name, patch->new_name);3520 for (cnt = 1, frag = patch->fragments;3521 frag;3522 cnt++, frag = frag->next) {3523 if (!frag->rejected) {3524 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);3525 continue;3526 }3527 fprintf(stderr, "Rejected hunk #%d.\n", cnt);3528 fprintf(rej, "%.*s", frag->size, frag->patch);3529 if (frag->patch[frag->size-1] != '\n')3530 fputc('\n', rej);3531 }3532 fclose(rej);3533 return -1;3534}35353536static int write_out_results(struct patch *list, int skipped_patch)3537{3538 int phase;3539 int errs = 0;3540 struct patch *l;35413542 if (!list && !skipped_patch)3543 return error("No changes");35443545 for (phase = 0; phase < 2; phase++) {3546 l = list;3547 while (l) {3548 if (l->rejected)3549 errs = 1;3550 else {3551 write_out_one_result(l, phase);3552 if (phase == 1 && write_out_one_reject(l))3553 errs = 1;3554 }3555 l = l->next;3556 }3557 }3558 return errs;3559}35603561static struct lock_file lock_file;35623563static struct string_list limit_by_name;3564static int has_include;3565static void add_name_limit(const char *name, int exclude)3566{3567 struct string_list_item *it;35683569 it = string_list_append(&limit_by_name, name);3570 it->util = exclude ? NULL : (void *) 1;3571}35723573static int use_patch(struct patch *p)3574{3575 const char *pathname = p->new_name ? p->new_name : p->old_name;3576 int i;35773578 /* Paths outside are not touched regardless of "--include" */3579 if (0 < prefix_length) {3580 int pathlen = strlen(pathname);3581 if (pathlen <= prefix_length ||3582 memcmp(prefix, pathname, prefix_length))3583 return 0;3584 }35853586 /* See if it matches any of exclude/include rule */3587 for (i = 0; i < limit_by_name.nr; i++) {3588 struct string_list_item *it = &limit_by_name.items[i];3589 if (!fnmatch(it->string, pathname, 0))3590 return (it->util != NULL);3591 }35923593 /*3594 * If we had any include, a path that does not match any rule is3595 * not used. Otherwise, we saw bunch of exclude rules (or none)3596 * and such a path is used.3597 */3598 return !has_include;3599}360036013602static void prefix_one(char **name)3603{3604 char *old_name = *name;3605 if (!old_name)3606 return;3607 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));3608 free(old_name);3609}36103611static void prefix_patches(struct patch *p)3612{3613 if (!prefix || p->is_toplevel_relative)3614 return;3615 for ( ; p; p = p->next) {3616 if (p->new_name == p->old_name) {3617 char *prefixed = p->new_name;3618 prefix_one(&prefixed);3619 p->new_name = p->old_name = prefixed;3620 }3621 else {3622 prefix_one(&p->new_name);3623 prefix_one(&p->old_name);3624 }3625 }3626}36273628#define INACCURATE_EOF (1<<0)3629#define RECOUNT (1<<1)36303631static int apply_patch(int fd, const char *filename, int options)3632{3633 size_t offset;3634 struct strbuf buf = STRBUF_INIT;3635 struct patch *list = NULL, **listp = &list;3636 int skipped_patch = 0;36373638 /* FIXME - memory leak when using multiple patch files as inputs */3639 memset(&fn_table, 0, sizeof(struct string_list));3640 patch_input_file = filename;3641 read_patch_file(&buf, fd);3642 offset = 0;3643 while (offset < buf.len) {3644 struct patch *patch;3645 int nr;36463647 patch = xcalloc(1, sizeof(*patch));3648 patch->inaccurate_eof = !!(options & INACCURATE_EOF);3649 patch->recount = !!(options & RECOUNT);3650 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);3651 if (nr < 0)3652 break;3653 if (apply_in_reverse)3654 reverse_patches(patch);3655 if (prefix)3656 prefix_patches(patch);3657 if (use_patch(patch)) {3658 patch_stats(patch);3659 *listp = patch;3660 listp = &patch->next;3661 }3662 else {3663 /* perhaps free it a bit better? */3664 free(patch);3665 skipped_patch++;3666 }3667 offset += nr;3668 }36693670 if (whitespace_error && (ws_error_action == die_on_ws_error))3671 apply = 0;36723673 update_index = check_index && apply;3674 if (update_index && newfd < 0)3675 newfd = hold_locked_index(&lock_file, 1);36763677 if (check_index) {3678 if (read_cache() < 0)3679 die("unable to read index file");3680 }36813682 if ((check || apply) &&3683 check_patch_list(list) < 0 &&3684 !apply_with_reject)3685 exit(1);36863687 if (apply && write_out_results(list, skipped_patch))3688 exit(1);36893690 if (fake_ancestor)3691 build_fake_ancestor(list, fake_ancestor);36923693 if (diffstat)3694 stat_patch_list(list);36953696 if (numstat)3697 numstat_patch_list(list);36983699 if (summary)3700 summary_patch_list(list);37013702 strbuf_release(&buf);3703 return 0;3704}37053706static int git_apply_config(const char *var, const char *value, void *cb)3707{3708 if (!strcmp(var, "apply.whitespace"))3709 return git_config_string(&apply_default_whitespace, var, value);3710 else if (!strcmp(var, "apply.ignorewhitespace"))3711 return git_config_string(&apply_default_ignorewhitespace, var, value);3712 return git_default_config(var, value, cb);3713}37143715static int option_parse_exclude(const struct option *opt,3716 const char *arg, int unset)3717{3718 add_name_limit(arg, 1);3719 return 0;3720}37213722static int option_parse_include(const struct option *opt,3723 const char *arg, int unset)3724{3725 add_name_limit(arg, 0);3726 has_include = 1;3727 return 0;3728}37293730static int option_parse_p(const struct option *opt,3731 const char *arg, int unset)3732{3733 p_value = atoi(arg);3734 p_value_known = 1;3735 return 0;3736}37373738static int option_parse_z(const struct option *opt,3739 const char *arg, int unset)3740{3741 if (unset)3742 line_termination = '\n';3743 else3744 line_termination = 0;3745 return 0;3746}37473748static int option_parse_space_change(const struct option *opt,3749 const char *arg, int unset)3750{3751 if (unset)3752 ws_ignore_action = ignore_ws_none;3753 else3754 ws_ignore_action = ignore_ws_change;3755 return 0;3756}37573758static int option_parse_whitespace(const struct option *opt,3759 const char *arg, int unset)3760{3761 const char **whitespace_option = opt->value;37623763 *whitespace_option = arg;3764 parse_whitespace_option(arg);3765 return 0;3766}37673768static int option_parse_directory(const struct option *opt,3769 const char *arg, int unset)3770{3771 root_len = strlen(arg);3772 if (root_len && arg[root_len - 1] != '/') {3773 char *new_root;3774 root = new_root = xmalloc(root_len + 2);3775 strcpy(new_root, arg);3776 strcpy(new_root + root_len++, "/");3777 } else3778 root = arg;3779 return 0;3780}37813782int cmd_apply(int argc, const char **argv, const char *prefix_)3783{3784 int i;3785 int errs = 0;3786 int is_not_gitdir = !startup_info->have_repository;3787 int binary;3788 int force_apply = 0;37893790 const char *whitespace_option = NULL;37913792 struct option builtin_apply_options[] = {3793 { OPTION_CALLBACK, 0, "exclude", NULL, "path",3794 "don't apply changes matching the given path",3795 0, option_parse_exclude },3796 { OPTION_CALLBACK, 0, "include", NULL, "path",3797 "apply changes matching the given path",3798 0, option_parse_include },3799 { OPTION_CALLBACK, 'p', NULL, NULL, "num",3800 "remove <num> leading slashes from traditional diff paths",3801 0, option_parse_p },3802 OPT_BOOLEAN(0, "no-add", &no_add,3803 "ignore additions made by the patch"),3804 OPT_BOOLEAN(0, "stat", &diffstat,3805 "instead of applying the patch, output diffstat for the input"),3806 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,3807 NULL, "old option, now no-op",3808 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3809 { OPTION_BOOLEAN, 0, "binary", &binary,3810 NULL, "old option, now no-op",3811 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3812 OPT_BOOLEAN(0, "numstat", &numstat,3813 "shows number of added and deleted lines in decimal notation"),3814 OPT_BOOLEAN(0, "summary", &summary,3815 "instead of applying the patch, output a summary for the input"),3816 OPT_BOOLEAN(0, "check", &check,3817 "instead of applying the patch, see if the patch is applicable"),3818 OPT_BOOLEAN(0, "index", &check_index,3819 "make sure the patch is applicable to the current index"),3820 OPT_BOOLEAN(0, "cached", &cached,3821 "apply a patch without touching the working tree"),3822 OPT_BOOLEAN(0, "apply", &force_apply,3823 "also apply the patch (use with --stat/--summary/--check)"),3824 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,3825 "build a temporary index based on embedded index information"),3826 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,3827 "paths are separated with NUL character",3828 PARSE_OPT_NOARG, option_parse_z },3829 OPT_INTEGER('C', NULL, &p_context,3830 "ensure at least <n> lines of context match"),3831 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",3832 "detect new or modified lines that have whitespace errors",3833 0, option_parse_whitespace },3834 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,3835 "ignore changes in whitespace when finding context",3836 PARSE_OPT_NOARG, option_parse_space_change },3837 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,3838 "ignore changes in whitespace when finding context",3839 PARSE_OPT_NOARG, option_parse_space_change },3840 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,3841 "apply the patch in reverse"),3842 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,3843 "don't expect at least one line of context"),3844 OPT_BOOLEAN(0, "reject", &apply_with_reject,3845 "leave the rejected hunks in corresponding *.rej files"),3846 OPT__VERBOSE(&apply_verbosely),3847 OPT_BIT(0, "inaccurate-eof", &options,3848 "tolerate incorrectly detected missing new-line at the end of file",3849 INACCURATE_EOF),3850 OPT_BIT(0, "recount", &options,3851 "do not trust the line counts in the hunk headers",3852 RECOUNT),3853 { OPTION_CALLBACK, 0, "directory", NULL, "root",3854 "prepend <root> to all filenames",3855 0, option_parse_directory },3856 OPT_END()3857 };38583859 prefix = prefix_;3860 prefix_length = prefix ? strlen(prefix) : 0;3861 git_config(git_apply_config, NULL);3862 if (apply_default_whitespace)3863 parse_whitespace_option(apply_default_whitespace);3864 if (apply_default_ignorewhitespace)3865 parse_ignorewhitespace_option(apply_default_ignorewhitespace);38663867 argc = parse_options(argc, argv, prefix, builtin_apply_options,3868 apply_usage, 0);38693870 if (apply_with_reject)3871 apply = apply_verbosely = 1;3872 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))3873 apply = 0;3874 if (check_index && is_not_gitdir)3875 die("--index outside a repository");3876 if (cached) {3877 if (is_not_gitdir)3878 die("--cached outside a repository");3879 check_index = 1;3880 }3881 for (i = 0; i < argc; i++) {3882 const char *arg = argv[i];3883 int fd;38843885 if (!strcmp(arg, "-")) {3886 errs |= apply_patch(0, "<stdin>", options);3887 read_stdin = 0;3888 continue;3889 } else if (0 < prefix_length)3890 arg = prefix_filename(prefix, prefix_length, arg);38913892 fd = open(arg, O_RDONLY);3893 if (fd < 0)3894 die_errno("can't open patch '%s'", arg);3895 read_stdin = 0;3896 set_default_whitespace_mode(whitespace_option);3897 errs |= apply_patch(fd, arg, options);3898 close(fd);3899 }3900 set_default_whitespace_mode(whitespace_option);3901 if (read_stdin)3902 errs |= apply_patch(0, "<stdin>", options);3903 if (whitespace_error) {3904 if (squelch_whitespace_errors &&3905 squelch_whitespace_errors < whitespace_error) {3906 int squelched =3907 whitespace_error - squelch_whitespace_errors;3908 warning("squelched %d "3909 "whitespace error%s",3910 squelched,3911 squelched == 1 ? "" : "s");3912 }3913 if (ws_error_action == die_on_ws_error)3914 die("%d line%s add%s whitespace errors.",3915 whitespace_error,3916 whitespace_error == 1 ? "" : "s",3917 whitespace_error == 1 ? "s" : "");3918 if (applied_after_fixing_ws && apply)3919 warning("%d line%s applied after"3920 " fixing whitespace errors.",3921 applied_after_fixing_ws,3922 applied_after_fixing_ws == 1 ? "" : "s");3923 else if (whitespace_error)3924 warning("%d line%s add%s whitespace errors.",3925 whitespace_error,3926 whitespace_error == 1 ? "" : "s",3927 whitespace_error == 1 ? "s" : "");3928 }39293930 if (update_index) {3931 if (write_cache(newfd, active_cache, active_nr) ||3932 commit_locked_index(&lock_file))3933 die("Unable to write new index file");3934 }39353936 return !!errs;3937}