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 allow_overlap; 47static int no_add; 48static const char *fake_ancestor; 49static int line_termination = '\n'; 50static unsigned int p_context = UINT_MAX; 51static const char * const apply_usage[] = { 52 "git apply [options] [<patch>...]", 53 NULL 54}; 55 56static enum ws_error_action { 57 nowarn_ws_error, 58 warn_on_ws_error, 59 die_on_ws_error, 60 correct_ws_error 61} ws_error_action = warn_on_ws_error; 62static int whitespace_error; 63static int squelch_whitespace_errors = 5; 64static int applied_after_fixing_ws; 65 66static enum ws_ignore { 67 ignore_ws_none, 68 ignore_ws_change 69} ws_ignore_action = ignore_ws_none; 70 71 72static const char *patch_input_file; 73static const char *root; 74static int root_len; 75static int read_stdin = 1; 76static int options; 77 78static void parse_whitespace_option(const char *option) 79{ 80 if (!option) { 81 ws_error_action = warn_on_ws_error; 82 return; 83 } 84 if (!strcmp(option, "warn")) { 85 ws_error_action = warn_on_ws_error; 86 return; 87 } 88 if (!strcmp(option, "nowarn")) { 89 ws_error_action = nowarn_ws_error; 90 return; 91 } 92 if (!strcmp(option, "error")) { 93 ws_error_action = die_on_ws_error; 94 return; 95 } 96 if (!strcmp(option, "error-all")) { 97 ws_error_action = die_on_ws_error; 98 squelch_whitespace_errors = 0; 99 return; 100 } 101 if (!strcmp(option, "strip") || !strcmp(option, "fix")) { 102 ws_error_action = correct_ws_error; 103 return; 104 } 105 die("unrecognized whitespace option '%s'", option); 106} 107 108static void parse_ignorewhitespace_option(const char *option) 109{ 110 if (!option || !strcmp(option, "no") || 111 !strcmp(option, "false") || !strcmp(option, "never") || 112 !strcmp(option, "none")) { 113 ws_ignore_action = ignore_ws_none; 114 return; 115 } 116 if (!strcmp(option, "change")) { 117 ws_ignore_action = ignore_ws_change; 118 return; 119 } 120 die("unrecognized whitespace ignore option '%s'", option); 121} 122 123static void set_default_whitespace_mode(const char *whitespace_option) 124{ 125 if (!whitespace_option && !apply_default_whitespace) 126 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error); 127} 128 129/* 130 * For "diff-stat" like behaviour, we keep track of the biggest change 131 * we've seen, and the longest filename. That allows us to do simple 132 * scaling. 133 */ 134static int max_change, max_len; 135 136/* 137 * Various "current state", notably line numbers and what 138 * file (and how) we're patching right now.. The "is_xxxx" 139 * things are flags, where -1 means "don't know yet". 140 */ 141static int linenr = 1; 142 143/* 144 * This represents one "hunk" from a patch, starting with 145 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The 146 * patch text is pointed at by patch, and its byte length 147 * is stored in size. leading and trailing are the number 148 * of context lines. 149 */ 150struct fragment { 151 unsigned long leading, trailing; 152 unsigned long oldpos, oldlines; 153 unsigned long newpos, newlines; 154 const char *patch; 155 int size; 156 int rejected; 157 int linenr; 158 struct fragment *next; 159}; 160 161/* 162 * When dealing with a binary patch, we reuse "leading" field 163 * to store the type of the binary hunk, either deflated "delta" 164 * or deflated "literal". 165 */ 166#define binary_patch_method leading 167#define BINARY_DELTA_DEFLATED 1 168#define BINARY_LITERAL_DEFLATED 2 169 170/* 171 * This represents a "patch" to a file, both metainfo changes 172 * such as creation/deletion, filemode and content changes represented 173 * as a series of fragments. 174 */ 175struct patch { 176 char *new_name, *old_name, *def_name; 177 unsigned int old_mode, new_mode; 178 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */ 179 int rejected; 180 unsigned ws_rule; 181 unsigned long deflate_origlen; 182 int lines_added, lines_deleted; 183 int score; 184 unsigned int is_toplevel_relative:1; 185 unsigned int inaccurate_eof:1; 186 unsigned int is_binary:1; 187 unsigned int is_copy:1; 188 unsigned int is_rename:1; 189 unsigned int recount:1; 190 struct fragment *fragments; 191 char *result; 192 size_t resultsize; 193 char old_sha1_prefix[41]; 194 char new_sha1_prefix[41]; 195 struct patch *next; 196}; 197 198/* 199 * A line in a file, len-bytes long (includes the terminating LF, 200 * except for an incomplete line at the end if the file ends with 201 * one), and its contents hashes to 'hash'. 202 */ 203struct line { 204 size_t len; 205 unsigned hash : 24; 206 unsigned flag : 8; 207#define LINE_COMMON 1 208#define LINE_PATCHED 2 209}; 210 211/* 212 * This represents a "file", which is an array of "lines". 213 */ 214struct image { 215 char *buf; 216 size_t len; 217 size_t nr; 218 size_t alloc; 219 struct line *line_allocated; 220 struct line *line; 221}; 222 223/* 224 * Records filenames that have been touched, in order to handle 225 * the case where more than one patches touch the same file. 226 */ 227 228static struct string_list fn_table; 229 230static uint32_t hash_line(const char *cp, size_t len) 231{ 232 size_t i; 233 uint32_t h; 234 for (i = 0, h = 0; i < len; i++) { 235 if (!isspace(cp[i])) { 236 h = h * 3 + (cp[i] & 0xff); 237 } 238 } 239 return h; 240} 241 242/* 243 * Compare lines s1 of length n1 and s2 of length n2, ignoring 244 * whitespace difference. Returns 1 if they match, 0 otherwise 245 */ 246static int fuzzy_matchlines(const char *s1, size_t n1, 247 const char *s2, size_t n2) 248{ 249 const char *last1 = s1 + n1 - 1; 250 const char *last2 = s2 + n2 - 1; 251 int result = 0; 252 253 if (n1 < 0 || n2 < 0) 254 return 0; 255 256 /* ignore line endings */ 257 while ((*last1 == '\r') || (*last1 == '\n')) 258 last1--; 259 while ((*last2 == '\r') || (*last2 == '\n')) 260 last2--; 261 262 /* skip leading whitespace */ 263 while (isspace(*s1) && (s1 <= last1)) 264 s1++; 265 while (isspace(*s2) && (s2 <= last2)) 266 s2++; 267 /* early return if both lines are empty */ 268 if ((s1 > last1) && (s2 > last2)) 269 return 1; 270 while (!result) { 271 result = *s1++ - *s2++; 272 /* 273 * Skip whitespace inside. We check for whitespace on 274 * both buffers because we don't want "a b" to match 275 * "ab" 276 */ 277 if (isspace(*s1) && isspace(*s2)) { 278 while (isspace(*s1) && s1 <= last1) 279 s1++; 280 while (isspace(*s2) && s2 <= last2) 281 s2++; 282 } 283 /* 284 * If we reached the end on one side only, 285 * lines don't match 286 */ 287 if ( 288 ((s2 > last2) && (s1 <= last1)) || 289 ((s1 > last1) && (s2 <= last2))) 290 return 0; 291 if ((s1 > last1) && (s2 > last2)) 292 break; 293 } 294 295 return !result; 296} 297 298static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag) 299{ 300 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc); 301 img->line_allocated[img->nr].len = len; 302 img->line_allocated[img->nr].hash = hash_line(bol, len); 303 img->line_allocated[img->nr].flag = flag; 304 img->nr++; 305} 306 307static void prepare_image(struct image *image, char *buf, size_t len, 308 int prepare_linetable) 309{ 310 const char *cp, *ep; 311 312 memset(image, 0, sizeof(*image)); 313 image->buf = buf; 314 image->len = len; 315 316 if (!prepare_linetable) 317 return; 318 319 ep = image->buf + image->len; 320 cp = image->buf; 321 while (cp < ep) { 322 const char *next; 323 for (next = cp; next < ep && *next != '\n'; next++) 324 ; 325 if (next < ep) 326 next++; 327 add_line_info(image, cp, next - cp, 0); 328 cp = next; 329 } 330 image->line = image->line_allocated; 331} 332 333static void clear_image(struct image *image) 334{ 335 free(image->buf); 336 image->buf = NULL; 337 image->len = 0; 338} 339 340static void say_patch_name(FILE *output, const char *pre, 341 struct patch *patch, const char *post) 342{ 343 fputs(pre, output); 344 if (patch->old_name && patch->new_name && 345 strcmp(patch->old_name, patch->new_name)) { 346 quote_c_style(patch->old_name, NULL, output, 0); 347 fputs(" => ", output); 348 quote_c_style(patch->new_name, NULL, output, 0); 349 } else { 350 const char *n = patch->new_name; 351 if (!n) 352 n = patch->old_name; 353 quote_c_style(n, NULL, output, 0); 354 } 355 fputs(post, output); 356} 357 358#define CHUNKSIZE (8192) 359#define SLOP (16) 360 361static void read_patch_file(struct strbuf *sb, int fd) 362{ 363 if (strbuf_read(sb, fd, 0) < 0) 364 die_errno("git apply: failed to read"); 365 366 /* 367 * Make sure that we have some slop in the buffer 368 * so that we can do speculative "memcmp" etc, and 369 * see to it that it is NUL-filled. 370 */ 371 strbuf_grow(sb, SLOP); 372 memset(sb->buf + sb->len, 0, SLOP); 373} 374 375static unsigned long linelen(const char *buffer, unsigned long size) 376{ 377 unsigned long len = 0; 378 while (size--) { 379 len++; 380 if (*buffer++ == '\n') 381 break; 382 } 383 return len; 384} 385 386static int is_dev_null(const char *str) 387{ 388 return !memcmp("/dev/null", str, 9) && isspace(str[9]); 389} 390 391#define TERM_SPACE 1 392#define TERM_TAB 2 393 394static int name_terminate(const char *name, int namelen, int c, int terminate) 395{ 396 if (c == ' ' && !(terminate & TERM_SPACE)) 397 return 0; 398 if (c == '\t' && !(terminate & TERM_TAB)) 399 return 0; 400 401 return 1; 402} 403 404/* remove double slashes to make --index work with such filenames */ 405static char *squash_slash(char *name) 406{ 407 int i = 0, j = 0; 408 409 if (!name) 410 return NULL; 411 412 while (name[i]) { 413 if ((name[j++] = name[i++]) == '/') 414 while (name[i] == '/') 415 i++; 416 } 417 name[j] = '\0'; 418 return name; 419} 420 421static char *find_name_gnu(const char *line, char *def, int p_value) 422{ 423 struct strbuf name = STRBUF_INIT; 424 char *cp; 425 426 /* 427 * Proposed "new-style" GNU patch/diff format; see 428 * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2 429 */ 430 if (unquote_c_style(&name, line, NULL)) { 431 strbuf_release(&name); 432 return NULL; 433 } 434 435 for (cp = name.buf; p_value; p_value--) { 436 cp = strchr(cp, '/'); 437 if (!cp) { 438 strbuf_release(&name); 439 return NULL; 440 } 441 cp++; 442 } 443 444 /* name can later be freed, so we need 445 * to memmove, not just return cp 446 */ 447 strbuf_remove(&name, 0, cp - name.buf); 448 free(def); 449 if (root) 450 strbuf_insert(&name, 0, root, root_len); 451 return squash_slash(strbuf_detach(&name, NULL)); 452} 453 454static size_t sane_tz_len(const char *line, size_t len) 455{ 456 const char *tz, *p; 457 458 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ') 459 return 0; 460 tz = line + len - strlen(" +0500"); 461 462 if (tz[1] != '+' && tz[1] != '-') 463 return 0; 464 465 for (p = tz + 2; p != line + len; p++) 466 if (!isdigit(*p)) 467 return 0; 468 469 return line + len - tz; 470} 471 472static size_t tz_with_colon_len(const char *line, size_t len) 473{ 474 const char *tz, *p; 475 476 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':') 477 return 0; 478 tz = line + len - strlen(" +08:00"); 479 480 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-')) 481 return 0; 482 p = tz + 2; 483 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 484 !isdigit(*p++) || !isdigit(*p++)) 485 return 0; 486 487 return line + len - tz; 488} 489 490static size_t date_len(const char *line, size_t len) 491{ 492 const char *date, *p; 493 494 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-') 495 return 0; 496 p = date = line + len - strlen("72-02-05"); 497 498 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 499 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 500 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */ 501 return 0; 502 503 if (date - line >= strlen("19") && 504 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */ 505 date -= strlen("19"); 506 507 return line + len - date; 508} 509 510static size_t short_time_len(const char *line, size_t len) 511{ 512 const char *time, *p; 513 514 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':') 515 return 0; 516 p = time = line + len - strlen(" 07:01:32"); 517 518 /* Permit 1-digit hours? */ 519 if (*p++ != ' ' || 520 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 521 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 522 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */ 523 return 0; 524 525 return line + len - time; 526} 527 528static size_t fractional_time_len(const char *line, size_t len) 529{ 530 const char *p; 531 size_t n; 532 533 /* Expected format: 19:41:17.620000023 */ 534 if (!len || !isdigit(line[len - 1])) 535 return 0; 536 p = line + len - 1; 537 538 /* Fractional seconds. */ 539 while (p > line && isdigit(*p)) 540 p--; 541 if (*p != '.') 542 return 0; 543 544 /* Hours, minutes, and whole seconds. */ 545 n = short_time_len(line, p - line); 546 if (!n) 547 return 0; 548 549 return line + len - p + n; 550} 551 552static size_t trailing_spaces_len(const char *line, size_t len) 553{ 554 const char *p; 555 556 /* Expected format: ' ' x (1 or more) */ 557 if (!len || line[len - 1] != ' ') 558 return 0; 559 560 p = line + len; 561 while (p != line) { 562 p--; 563 if (*p != ' ') 564 return line + len - (p + 1); 565 } 566 567 /* All spaces! */ 568 return len; 569} 570 571static size_t diff_timestamp_len(const char *line, size_t len) 572{ 573 const char *end = line + len; 574 size_t n; 575 576 /* 577 * Posix: 2010-07-05 19:41:17 578 * GNU: 2010-07-05 19:41:17.620000023 -0500 579 */ 580 581 if (!isdigit(end[-1])) 582 return 0; 583 584 n = sane_tz_len(line, end - line); 585 if (!n) 586 n = tz_with_colon_len(line, end - line); 587 end -= n; 588 589 n = short_time_len(line, end - line); 590 if (!n) 591 n = fractional_time_len(line, end - line); 592 end -= n; 593 594 n = date_len(line, end - line); 595 if (!n) /* No date. Too bad. */ 596 return 0; 597 end -= n; 598 599 if (end == line) /* No space before date. */ 600 return 0; 601 if (end[-1] == '\t') { /* Success! */ 602 end--; 603 return line + len - end; 604 } 605 if (end[-1] != ' ') /* No space before date. */ 606 return 0; 607 608 /* Whitespace damage. */ 609 end -= trailing_spaces_len(line, end - line); 610 return line + len - end; 611} 612 613static char *find_name_common(const char *line, char *def, int p_value, 614 const char *end, int terminate) 615{ 616 int len; 617 const char *start = NULL; 618 619 if (p_value == 0) 620 start = line; 621 while (line != end) { 622 char c = *line; 623 624 if (!end && isspace(c)) { 625 if (c == '\n') 626 break; 627 if (name_terminate(start, line-start, c, terminate)) 628 break; 629 } 630 line++; 631 if (c == '/' && !--p_value) 632 start = line; 633 } 634 if (!start) 635 return squash_slash(def); 636 len = line - start; 637 if (!len) 638 return squash_slash(def); 639 640 /* 641 * Generally we prefer the shorter name, especially 642 * if the other one is just a variation of that with 643 * something else tacked on to the end (ie "file.orig" 644 * or "file~"). 645 */ 646 if (def) { 647 int deflen = strlen(def); 648 if (deflen < len && !strncmp(start, def, deflen)) 649 return squash_slash(def); 650 free(def); 651 } 652 653 if (root) { 654 char *ret = xmalloc(root_len + len + 1); 655 strcpy(ret, root); 656 memcpy(ret + root_len, start, len); 657 ret[root_len + len] = '\0'; 658 return squash_slash(ret); 659 } 660 661 return squash_slash(xmemdupz(start, len)); 662} 663 664static char *find_name(const char *line, char *def, int p_value, int terminate) 665{ 666 if (*line == '"') { 667 char *name = find_name_gnu(line, def, p_value); 668 if (name) 669 return name; 670 } 671 672 return find_name_common(line, def, p_value, NULL, terminate); 673} 674 675static char *find_name_traditional(const char *line, char *def, int p_value) 676{ 677 size_t len = strlen(line); 678 size_t date_len; 679 680 if (*line == '"') { 681 char *name = find_name_gnu(line, def, p_value); 682 if (name) 683 return name; 684 } 685 686 len = strchrnul(line, '\n') - line; 687 date_len = diff_timestamp_len(line, len); 688 if (!date_len) 689 return find_name_common(line, def, p_value, NULL, TERM_TAB); 690 len -= date_len; 691 692 return find_name_common(line, def, p_value, line + len, 0); 693} 694 695static int count_slashes(const char *cp) 696{ 697 int cnt = 0; 698 char ch; 699 700 while ((ch = *cp++)) 701 if (ch == '/') 702 cnt++; 703 return cnt; 704} 705 706/* 707 * Given the string after "--- " or "+++ ", guess the appropriate 708 * p_value for the given patch. 709 */ 710static int guess_p_value(const char *nameline) 711{ 712 char *name, *cp; 713 int val = -1; 714 715 if (is_dev_null(nameline)) 716 return -1; 717 name = find_name_traditional(nameline, NULL, 0); 718 if (!name) 719 return -1; 720 cp = strchr(name, '/'); 721 if (!cp) 722 val = 0; 723 else if (prefix) { 724 /* 725 * Does it begin with "a/$our-prefix" and such? Then this is 726 * very likely to apply to our directory. 727 */ 728 if (!strncmp(name, prefix, prefix_length)) 729 val = count_slashes(prefix); 730 else { 731 cp++; 732 if (!strncmp(cp, prefix, prefix_length)) 733 val = count_slashes(prefix) + 1; 734 } 735 } 736 free(name); 737 return val; 738} 739 740/* 741 * Does the ---/+++ line has the POSIX timestamp after the last HT? 742 * GNU diff puts epoch there to signal a creation/deletion event. Is 743 * this such a timestamp? 744 */ 745static int has_epoch_timestamp(const char *nameline) 746{ 747 /* 748 * We are only interested in epoch timestamp; any non-zero 749 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 750 * For the same reason, the date must be either 1969-12-31 or 751 * 1970-01-01, and the seconds part must be "00". 752 */ 753 const char stamp_regexp[] = 754 "^(1969-12-31|1970-01-01)" 755 " " 756 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 757 " " 758 "([-+][0-2][0-9]:?[0-5][0-9])\n"; 759 const char *timestamp = NULL, *cp, *colon; 760 static regex_t *stamp; 761 regmatch_t m[10]; 762 int zoneoffset; 763 int hourminute; 764 int status; 765 766 for (cp = nameline; *cp != '\n'; cp++) { 767 if (*cp == '\t') 768 timestamp = cp + 1; 769 } 770 if (!timestamp) 771 return 0; 772 if (!stamp) { 773 stamp = xmalloc(sizeof(*stamp)); 774 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 775 warning("Cannot prepare timestamp regexp %s", 776 stamp_regexp); 777 return 0; 778 } 779 } 780 781 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 782 if (status) { 783 if (status != REG_NOMATCH) 784 warning("regexec returned %d for input: %s", 785 status, timestamp); 786 return 0; 787 } 788 789 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10); 790 if (*colon == ':') 791 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10); 792 else 793 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 794 if (timestamp[m[3].rm_so] == '-') 795 zoneoffset = -zoneoffset; 796 797 /* 798 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 799 * (west of GMT) or 1970-01-01 (east of GMT) 800 */ 801 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 802 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 803 return 0; 804 805 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 806 strtol(timestamp + 14, NULL, 10) - 807 zoneoffset); 808 809 return ((zoneoffset < 0 && hourminute == 1440) || 810 (0 <= zoneoffset && !hourminute)); 811} 812 813/* 814 * Get the name etc info from the ---/+++ lines of a traditional patch header 815 * 816 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 817 * files, we can happily check the index for a match, but for creating a 818 * new file we should try to match whatever "patch" does. I have no idea. 819 */ 820static void parse_traditional_patch(const char *first, const char *second, struct patch *patch) 821{ 822 char *name; 823 824 first += 4; /* skip "--- " */ 825 second += 4; /* skip "+++ " */ 826 if (!p_value_known) { 827 int p, q; 828 p = guess_p_value(first); 829 q = guess_p_value(second); 830 if (p < 0) p = q; 831 if (0 <= p && p == q) { 832 p_value = p; 833 p_value_known = 1; 834 } 835 } 836 if (is_dev_null(first)) { 837 patch->is_new = 1; 838 patch->is_delete = 0; 839 name = find_name_traditional(second, NULL, p_value); 840 patch->new_name = name; 841 } else if (is_dev_null(second)) { 842 patch->is_new = 0; 843 patch->is_delete = 1; 844 name = find_name_traditional(first, NULL, p_value); 845 patch->old_name = name; 846 } else { 847 name = find_name_traditional(first, NULL, p_value); 848 name = find_name_traditional(second, name, p_value); 849 if (has_epoch_timestamp(first)) { 850 patch->is_new = 1; 851 patch->is_delete = 0; 852 patch->new_name = name; 853 } else if (has_epoch_timestamp(second)) { 854 patch->is_new = 0; 855 patch->is_delete = 1; 856 patch->old_name = name; 857 } else { 858 patch->old_name = patch->new_name = name; 859 } 860 } 861 if (!name) 862 die("unable to find filename in patch at line %d", linenr); 863} 864 865static int gitdiff_hdrend(const char *line, struct patch *patch) 866{ 867 return -1; 868} 869 870/* 871 * We're anal about diff header consistency, to make 872 * sure that we don't end up having strange ambiguous 873 * patches floating around. 874 * 875 * As a result, gitdiff_{old|new}name() will check 876 * their names against any previous information, just 877 * to make sure.. 878 */ 879static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew) 880{ 881 if (!orig_name && !isnull) 882 return find_name(line, NULL, p_value, TERM_TAB); 883 884 if (orig_name) { 885 int len; 886 const char *name; 887 char *another; 888 name = orig_name; 889 len = strlen(name); 890 if (isnull) 891 die("git apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr); 892 another = find_name(line, NULL, p_value, TERM_TAB); 893 if (!another || memcmp(another, name, len + 1)) 894 die("git apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr); 895 free(another); 896 return orig_name; 897 } 898 else { 899 /* expect "/dev/null" */ 900 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 901 die("git apply: bad git-diff - expected /dev/null on line %d", linenr); 902 return NULL; 903 } 904} 905 906static int gitdiff_oldname(const char *line, struct patch *patch) 907{ 908 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old"); 909 return 0; 910} 911 912static int gitdiff_newname(const char *line, struct patch *patch) 913{ 914 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new"); 915 return 0; 916} 917 918static int gitdiff_oldmode(const char *line, struct patch *patch) 919{ 920 patch->old_mode = strtoul(line, NULL, 8); 921 return 0; 922} 923 924static int gitdiff_newmode(const char *line, struct patch *patch) 925{ 926 patch->new_mode = strtoul(line, NULL, 8); 927 return 0; 928} 929 930static int gitdiff_delete(const char *line, struct patch *patch) 931{ 932 patch->is_delete = 1; 933 patch->old_name = patch->def_name; 934 return gitdiff_oldmode(line, patch); 935} 936 937static int gitdiff_newfile(const char *line, struct patch *patch) 938{ 939 patch->is_new = 1; 940 patch->new_name = patch->def_name; 941 return gitdiff_newmode(line, patch); 942} 943 944static int gitdiff_copysrc(const char *line, struct patch *patch) 945{ 946 patch->is_copy = 1; 947 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0); 948 return 0; 949} 950 951static int gitdiff_copydst(const char *line, struct patch *patch) 952{ 953 patch->is_copy = 1; 954 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0); 955 return 0; 956} 957 958static int gitdiff_renamesrc(const char *line, struct patch *patch) 959{ 960 patch->is_rename = 1; 961 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0); 962 return 0; 963} 964 965static int gitdiff_renamedst(const char *line, struct patch *patch) 966{ 967 patch->is_rename = 1; 968 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0); 969 return 0; 970} 971 972static int gitdiff_similarity(const char *line, struct patch *patch) 973{ 974 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 975 patch->score = 0; 976 return 0; 977} 978 979static int gitdiff_dissimilarity(const char *line, struct patch *patch) 980{ 981 if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX) 982 patch->score = 0; 983 return 0; 984} 985 986static int gitdiff_index(const char *line, struct patch *patch) 987{ 988 /* 989 * index line is N hexadecimal, "..", N hexadecimal, 990 * and optional space with octal mode. 991 */ 992 const char *ptr, *eol; 993 int len; 994 995 ptr = strchr(line, '.'); 996 if (!ptr || ptr[1] != '.' || 40 < ptr - line) 997 return 0; 998 len = ptr - line; 999 memcpy(patch->old_sha1_prefix, line, len);1000 patch->old_sha1_prefix[len] = 0;10011002 line = ptr + 2;1003 ptr = strchr(line, ' ');1004 eol = strchr(line, '\n');10051006 if (!ptr || eol < ptr)1007 ptr = eol;1008 len = ptr - line;10091010 if (40 < len)1011 return 0;1012 memcpy(patch->new_sha1_prefix, line, len);1013 patch->new_sha1_prefix[len] = 0;1014 if (*ptr == ' ')1015 patch->old_mode = strtoul(ptr+1, NULL, 8);1016 return 0;1017}10181019/*1020 * This is normal for a diff that doesn't change anything: we'll fall through1021 * into the next diff. Tell the parser to break out.1022 */1023static int gitdiff_unrecognized(const char *line, struct patch *patch)1024{1025 return -1;1026}10271028static const char *stop_at_slash(const char *line, int llen)1029{1030 int nslash = p_value;1031 int i;10321033 for (i = 0; i < llen; i++) {1034 int ch = line[i];1035 if (ch == '/' && --nslash <= 0)1036 return &line[i];1037 }1038 return NULL;1039}10401041/*1042 * This is to extract the same name that appears on "diff --git"1043 * line. We do not find and return anything if it is a rename1044 * patch, and it is OK because we will find the name elsewhere.1045 * We need to reliably find name only when it is mode-change only,1046 * creation or deletion of an empty file. In any of these cases,1047 * both sides are the same name under a/ and b/ respectively.1048 */1049static char *git_header_name(char *line, int llen)1050{1051 const char *name;1052 const char *second = NULL;1053 size_t len, line_len;10541055 line += strlen("diff --git ");1056 llen -= strlen("diff --git ");10571058 if (*line == '"') {1059 const char *cp;1060 struct strbuf first = STRBUF_INIT;1061 struct strbuf sp = STRBUF_INIT;10621063 if (unquote_c_style(&first, line, &second))1064 goto free_and_fail1;10651066 /* advance to the first slash */1067 cp = stop_at_slash(first.buf, first.len);1068 /* we do not accept absolute paths */1069 if (!cp || cp == first.buf)1070 goto free_and_fail1;1071 strbuf_remove(&first, 0, cp + 1 - first.buf);10721073 /*1074 * second points at one past closing dq of name.1075 * find the second name.1076 */1077 while ((second < line + llen) && isspace(*second))1078 second++;10791080 if (line + llen <= second)1081 goto free_and_fail1;1082 if (*second == '"') {1083 if (unquote_c_style(&sp, second, NULL))1084 goto free_and_fail1;1085 cp = stop_at_slash(sp.buf, sp.len);1086 if (!cp || cp == sp.buf)1087 goto free_and_fail1;1088 /* They must match, otherwise ignore */1089 if (strcmp(cp + 1, first.buf))1090 goto free_and_fail1;1091 strbuf_release(&sp);1092 return strbuf_detach(&first, NULL);1093 }10941095 /* unquoted second */1096 cp = stop_at_slash(second, line + llen - second);1097 if (!cp || cp == second)1098 goto free_and_fail1;1099 cp++;1100 if (line + llen - cp != first.len + 1 ||1101 memcmp(first.buf, cp, first.len))1102 goto free_and_fail1;1103 return strbuf_detach(&first, NULL);11041105 free_and_fail1:1106 strbuf_release(&first);1107 strbuf_release(&sp);1108 return NULL;1109 }11101111 /* unquoted first name */1112 name = stop_at_slash(line, llen);1113 if (!name || name == line)1114 return NULL;1115 name++;11161117 /*1118 * since the first name is unquoted, a dq if exists must be1119 * the beginning of the second name.1120 */1121 for (second = name; second < line + llen; second++) {1122 if (*second == '"') {1123 struct strbuf sp = STRBUF_INIT;1124 const char *np;11251126 if (unquote_c_style(&sp, second, NULL))1127 goto free_and_fail2;11281129 np = stop_at_slash(sp.buf, sp.len);1130 if (!np || np == sp.buf)1131 goto free_and_fail2;1132 np++;11331134 len = sp.buf + sp.len - np;1135 if (len < second - name &&1136 !strncmp(np, name, len) &&1137 isspace(name[len])) {1138 /* Good */1139 strbuf_remove(&sp, 0, np - sp.buf);1140 return strbuf_detach(&sp, NULL);1141 }11421143 free_and_fail2:1144 strbuf_release(&sp);1145 return NULL;1146 }1147 }11481149 /*1150 * Accept a name only if it shows up twice, exactly the same1151 * form.1152 */1153 second = strchr(name, '\n');1154 if (!second)1155 return NULL;1156 line_len = second - name;1157 for (len = 0 ; ; len++) {1158 switch (name[len]) {1159 default:1160 continue;1161 case '\n':1162 return NULL;1163 case '\t': case ' ':1164 second = stop_at_slash(name + len, line_len - len);1165 if (!second)1166 return NULL;1167 second++;1168 if (second[len] == '\n' && !strncmp(name, second, len)) {1169 return xmemdupz(name, len);1170 }1171 }1172 }1173}11741175/* Verify that we recognize the lines following a git header */1176static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)1177{1178 unsigned long offset;11791180 /* A git diff has explicit new/delete information, so we don't guess */1181 patch->is_new = 0;1182 patch->is_delete = 0;11831184 /*1185 * Some things may not have the old name in the1186 * rest of the headers anywhere (pure mode changes,1187 * or removing or adding empty files), so we get1188 * the default name from the header.1189 */1190 patch->def_name = git_header_name(line, len);1191 if (patch->def_name && root) {1192 char *s = xmalloc(root_len + strlen(patch->def_name) + 1);1193 strcpy(s, root);1194 strcpy(s + root_len, patch->def_name);1195 free(patch->def_name);1196 patch->def_name = s;1197 }11981199 line += len;1200 size -= len;1201 linenr++;1202 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {1203 static const struct opentry {1204 const char *str;1205 int (*fn)(const char *, struct patch *);1206 } optable[] = {1207 { "@@ -", gitdiff_hdrend },1208 { "--- ", gitdiff_oldname },1209 { "+++ ", gitdiff_newname },1210 { "old mode ", gitdiff_oldmode },1211 { "new mode ", gitdiff_newmode },1212 { "deleted file mode ", gitdiff_delete },1213 { "new file mode ", gitdiff_newfile },1214 { "copy from ", gitdiff_copysrc },1215 { "copy to ", gitdiff_copydst },1216 { "rename old ", gitdiff_renamesrc },1217 { "rename new ", gitdiff_renamedst },1218 { "rename from ", gitdiff_renamesrc },1219 { "rename to ", gitdiff_renamedst },1220 { "similarity index ", gitdiff_similarity },1221 { "dissimilarity index ", gitdiff_dissimilarity },1222 { "index ", gitdiff_index },1223 { "", gitdiff_unrecognized },1224 };1225 int i;12261227 len = linelen(line, size);1228 if (!len || line[len-1] != '\n')1229 break;1230 for (i = 0; i < ARRAY_SIZE(optable); i++) {1231 const struct opentry *p = optable + i;1232 int oplen = strlen(p->str);1233 if (len < oplen || memcmp(p->str, line, oplen))1234 continue;1235 if (p->fn(line + oplen, patch) < 0)1236 return offset;1237 break;1238 }1239 }12401241 return offset;1242}12431244static int parse_num(const char *line, unsigned long *p)1245{1246 char *ptr;12471248 if (!isdigit(*line))1249 return 0;1250 *p = strtoul(line, &ptr, 10);1251 return ptr - line;1252}12531254static int parse_range(const char *line, int len, int offset, const char *expect,1255 unsigned long *p1, unsigned long *p2)1256{1257 int digits, ex;12581259 if (offset < 0 || offset >= len)1260 return -1;1261 line += offset;1262 len -= offset;12631264 digits = parse_num(line, p1);1265 if (!digits)1266 return -1;12671268 offset += digits;1269 line += digits;1270 len -= digits;12711272 *p2 = 1;1273 if (*line == ',') {1274 digits = parse_num(line+1, p2);1275 if (!digits)1276 return -1;12771278 offset += digits+1;1279 line += digits+1;1280 len -= digits+1;1281 }12821283 ex = strlen(expect);1284 if (ex > len)1285 return -1;1286 if (memcmp(line, expect, ex))1287 return -1;12881289 return offset + ex;1290}12911292static void recount_diff(char *line, int size, struct fragment *fragment)1293{1294 int oldlines = 0, newlines = 0, ret = 0;12951296 if (size < 1) {1297 warning("recount: ignore empty hunk");1298 return;1299 }13001301 for (;;) {1302 int len = linelen(line, size);1303 size -= len;1304 line += len;13051306 if (size < 1)1307 break;13081309 switch (*line) {1310 case ' ': case '\n':1311 newlines++;1312 /* fall through */1313 case '-':1314 oldlines++;1315 continue;1316 case '+':1317 newlines++;1318 continue;1319 case '\\':1320 continue;1321 case '@':1322 ret = size < 3 || prefixcmp(line, "@@ ");1323 break;1324 case 'd':1325 ret = size < 5 || prefixcmp(line, "diff ");1326 break;1327 default:1328 ret = -1;1329 break;1330 }1331 if (ret) {1332 warning("recount: unexpected line: %.*s",1333 (int)linelen(line, size), line);1334 return;1335 }1336 break;1337 }1338 fragment->oldlines = oldlines;1339 fragment->newlines = newlines;1340}13411342/*1343 * Parse a unified diff fragment header of the1344 * form "@@ -a,b +c,d @@"1345 */1346static int parse_fragment_header(char *line, int len, struct fragment *fragment)1347{1348 int offset;13491350 if (!len || line[len-1] != '\n')1351 return -1;13521353 /* Figure out the number of lines in a fragment */1354 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1355 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);13561357 return offset;1358}13591360static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)1361{1362 unsigned long offset, len;13631364 patch->is_toplevel_relative = 0;1365 patch->is_rename = patch->is_copy = 0;1366 patch->is_new = patch->is_delete = -1;1367 patch->old_mode = patch->new_mode = 0;1368 patch->old_name = patch->new_name = NULL;1369 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {1370 unsigned long nextlen;13711372 len = linelen(line, size);1373 if (!len)1374 break;13751376 /* Testing this early allows us to take a few shortcuts.. */1377 if (len < 6)1378 continue;13791380 /*1381 * Make sure we don't find any unconnected patch fragments.1382 * That's a sign that we didn't find a header, and that a1383 * patch has become corrupted/broken up.1384 */1385 if (!memcmp("@@ -", line, 4)) {1386 struct fragment dummy;1387 if (parse_fragment_header(line, len, &dummy) < 0)1388 continue;1389 die("patch fragment without header at line %d: %.*s",1390 linenr, (int)len-1, line);1391 }13921393 if (size < len + 6)1394 break;13951396 /*1397 * Git patch? It might not have a real patch, just a rename1398 * or mode change, so we handle that specially1399 */1400 if (!memcmp("diff --git ", line, 11)) {1401 int git_hdr_len = parse_git_header(line, len, size, patch);1402 if (git_hdr_len <= len)1403 continue;1404 if (!patch->old_name && !patch->new_name) {1405 if (!patch->def_name)1406 die("git diff header lacks filename information when removing "1407 "%d leading pathname components (line %d)" , p_value, linenr);1408 patch->old_name = patch->new_name = patch->def_name;1409 }1410 if (!patch->is_delete && !patch->new_name)1411 die("git diff header lacks filename information "1412 "(line %d)", linenr);1413 patch->is_toplevel_relative = 1;1414 *hdrsize = git_hdr_len;1415 return offset;1416 }14171418 /* --- followed by +++ ? */1419 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1420 continue;14211422 /*1423 * We only accept unified patches, so we want it to1424 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1425 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1426 */1427 nextlen = linelen(line + len, size - len);1428 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1429 continue;14301431 /* Ok, we'll consider it a patch */1432 parse_traditional_patch(line, line+len, patch);1433 *hdrsize = len + nextlen;1434 linenr += 2;1435 return offset;1436 }1437 return -1;1438}14391440static void record_ws_error(unsigned result, const char *line, int len, int linenr)1441{1442 char *err;14431444 if (!result)1445 return;14461447 whitespace_error++;1448 if (squelch_whitespace_errors &&1449 squelch_whitespace_errors < whitespace_error)1450 return;14511452 err = whitespace_error_string(result);1453 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1454 patch_input_file, linenr, err, len, line);1455 free(err);1456}14571458static void check_whitespace(const char *line, int len, unsigned ws_rule)1459{1460 unsigned result = ws_check(line + 1, len - 1, ws_rule);14611462 record_ws_error(result, line + 1, len - 2, linenr);1463}14641465/*1466 * Parse a unified diff. Note that this really needs to parse each1467 * fragment separately, since the only way to know the difference1468 * between a "---" that is part of a patch, and a "---" that starts1469 * the next patch is to look at the line counts..1470 */1471static int parse_fragment(char *line, unsigned long size,1472 struct patch *patch, struct fragment *fragment)1473{1474 int added, deleted;1475 int len = linelen(line, size), offset;1476 unsigned long oldlines, newlines;1477 unsigned long leading, trailing;14781479 offset = parse_fragment_header(line, len, fragment);1480 if (offset < 0)1481 return -1;1482 if (offset > 0 && patch->recount)1483 recount_diff(line + offset, size - offset, fragment);1484 oldlines = fragment->oldlines;1485 newlines = fragment->newlines;1486 leading = 0;1487 trailing = 0;14881489 /* Parse the thing.. */1490 line += len;1491 size -= len;1492 linenr++;1493 added = deleted = 0;1494 for (offset = len;1495 0 < size;1496 offset += len, size -= len, line += len, linenr++) {1497 if (!oldlines && !newlines)1498 break;1499 len = linelen(line, size);1500 if (!len || line[len-1] != '\n')1501 return -1;1502 switch (*line) {1503 default:1504 return -1;1505 case '\n': /* newer GNU diff, an empty context line */1506 case ' ':1507 oldlines--;1508 newlines--;1509 if (!deleted && !added)1510 leading++;1511 trailing++;1512 break;1513 case '-':1514 if (apply_in_reverse &&1515 ws_error_action != nowarn_ws_error)1516 check_whitespace(line, len, patch->ws_rule);1517 deleted++;1518 oldlines--;1519 trailing = 0;1520 break;1521 case '+':1522 if (!apply_in_reverse &&1523 ws_error_action != nowarn_ws_error)1524 check_whitespace(line, len, patch->ws_rule);1525 added++;1526 newlines--;1527 trailing = 0;1528 break;15291530 /*1531 * We allow "\ No newline at end of file". Depending1532 * on locale settings when the patch was produced we1533 * don't know what this line looks like. The only1534 * thing we do know is that it begins with "\ ".1535 * Checking for 12 is just for sanity check -- any1536 * l10n of "\ No newline..." is at least that long.1537 */1538 case '\\':1539 if (len < 12 || memcmp(line, "\\ ", 2))1540 return -1;1541 break;1542 }1543 }1544 if (oldlines || newlines)1545 return -1;1546 fragment->leading = leading;1547 fragment->trailing = trailing;15481549 /*1550 * If a fragment ends with an incomplete line, we failed to include1551 * it in the above loop because we hit oldlines == newlines == 01552 * before seeing it.1553 */1554 if (12 < size && !memcmp(line, "\\ ", 2))1555 offset += linelen(line, size);15561557 patch->lines_added += added;1558 patch->lines_deleted += deleted;15591560 if (0 < patch->is_new && oldlines)1561 return error("new file depends on old contents");1562 if (0 < patch->is_delete && newlines)1563 return error("deleted file still has contents");1564 return offset;1565}15661567static int parse_single_patch(char *line, unsigned long size, struct patch *patch)1568{1569 unsigned long offset = 0;1570 unsigned long oldlines = 0, newlines = 0, context = 0;1571 struct fragment **fragp = &patch->fragments;15721573 while (size > 4 && !memcmp(line, "@@ -", 4)) {1574 struct fragment *fragment;1575 int len;15761577 fragment = xcalloc(1, sizeof(*fragment));1578 fragment->linenr = linenr;1579 len = parse_fragment(line, size, patch, fragment);1580 if (len <= 0)1581 die("corrupt patch at line %d", linenr);1582 fragment->patch = line;1583 fragment->size = len;1584 oldlines += fragment->oldlines;1585 newlines += fragment->newlines;1586 context += fragment->leading + fragment->trailing;15871588 *fragp = fragment;1589 fragp = &fragment->next;15901591 offset += len;1592 line += len;1593 size -= len;1594 }15951596 /*1597 * If something was removed (i.e. we have old-lines) it cannot1598 * be creation, and if something was added it cannot be1599 * deletion. However, the reverse is not true; --unified=01600 * patches that only add are not necessarily creation even1601 * though they do not have any old lines, and ones that only1602 * delete are not necessarily deletion.1603 *1604 * Unfortunately, a real creation/deletion patch do _not_ have1605 * any context line by definition, so we cannot safely tell it1606 * apart with --unified=0 insanity. At least if the patch has1607 * more than one hunk it is not creation or deletion.1608 */1609 if (patch->is_new < 0 &&1610 (oldlines || (patch->fragments && patch->fragments->next)))1611 patch->is_new = 0;1612 if (patch->is_delete < 0 &&1613 (newlines || (patch->fragments && patch->fragments->next)))1614 patch->is_delete = 0;16151616 if (0 < patch->is_new && oldlines)1617 die("new file %s depends on old contents", patch->new_name);1618 if (0 < patch->is_delete && newlines)1619 die("deleted file %s still has contents", patch->old_name);1620 if (!patch->is_delete && !newlines && context)1621 fprintf(stderr, "** warning: file %s becomes empty but "1622 "is not deleted\n", patch->new_name);16231624 return offset;1625}16261627static inline int metadata_changes(struct patch *patch)1628{1629 return patch->is_rename > 0 ||1630 patch->is_copy > 0 ||1631 patch->is_new > 0 ||1632 patch->is_delete ||1633 (patch->old_mode && patch->new_mode &&1634 patch->old_mode != patch->new_mode);1635}16361637static char *inflate_it(const void *data, unsigned long size,1638 unsigned long inflated_size)1639{1640 git_zstream stream;1641 void *out;1642 int st;16431644 memset(&stream, 0, sizeof(stream));16451646 stream.next_in = (unsigned char *)data;1647 stream.avail_in = size;1648 stream.next_out = out = xmalloc(inflated_size);1649 stream.avail_out = inflated_size;1650 git_inflate_init(&stream);1651 st = git_inflate(&stream, Z_FINISH);1652 git_inflate_end(&stream);1653 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1654 free(out);1655 return NULL;1656 }1657 return out;1658}16591660static struct fragment *parse_binary_hunk(char **buf_p,1661 unsigned long *sz_p,1662 int *status_p,1663 int *used_p)1664{1665 /*1666 * Expect a line that begins with binary patch method ("literal"1667 * or "delta"), followed by the length of data before deflating.1668 * a sequence of 'length-byte' followed by base-85 encoded data1669 * should follow, terminated by a newline.1670 *1671 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1672 * and we would limit the patch line to 66 characters,1673 * so one line can fit up to 13 groups that would decode1674 * to 52 bytes max. The length byte 'A'-'Z' corresponds1675 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1676 */1677 int llen, used;1678 unsigned long size = *sz_p;1679 char *buffer = *buf_p;1680 int patch_method;1681 unsigned long origlen;1682 char *data = NULL;1683 int hunk_size = 0;1684 struct fragment *frag;16851686 llen = linelen(buffer, size);1687 used = llen;16881689 *status_p = 0;16901691 if (!prefixcmp(buffer, "delta ")) {1692 patch_method = BINARY_DELTA_DEFLATED;1693 origlen = strtoul(buffer + 6, NULL, 10);1694 }1695 else if (!prefixcmp(buffer, "literal ")) {1696 patch_method = BINARY_LITERAL_DEFLATED;1697 origlen = strtoul(buffer + 8, NULL, 10);1698 }1699 else1700 return NULL;17011702 linenr++;1703 buffer += llen;1704 while (1) {1705 int byte_length, max_byte_length, newsize;1706 llen = linelen(buffer, size);1707 used += llen;1708 linenr++;1709 if (llen == 1) {1710 /* consume the blank line */1711 buffer++;1712 size--;1713 break;1714 }1715 /*1716 * Minimum line is "A00000\n" which is 7-byte long,1717 * and the line length must be multiple of 5 plus 2.1718 */1719 if ((llen < 7) || (llen-2) % 5)1720 goto corrupt;1721 max_byte_length = (llen - 2) / 5 * 4;1722 byte_length = *buffer;1723 if ('A' <= byte_length && byte_length <= 'Z')1724 byte_length = byte_length - 'A' + 1;1725 else if ('a' <= byte_length && byte_length <= 'z')1726 byte_length = byte_length - 'a' + 27;1727 else1728 goto corrupt;1729 /* if the input length was not multiple of 4, we would1730 * have filler at the end but the filler should never1731 * exceed 3 bytes1732 */1733 if (max_byte_length < byte_length ||1734 byte_length <= max_byte_length - 4)1735 goto corrupt;1736 newsize = hunk_size + byte_length;1737 data = xrealloc(data, newsize);1738 if (decode_85(data + hunk_size, buffer + 1, byte_length))1739 goto corrupt;1740 hunk_size = newsize;1741 buffer += llen;1742 size -= llen;1743 }17441745 frag = xcalloc(1, sizeof(*frag));1746 frag->patch = inflate_it(data, hunk_size, origlen);1747 if (!frag->patch)1748 goto corrupt;1749 free(data);1750 frag->size = origlen;1751 *buf_p = buffer;1752 *sz_p = size;1753 *used_p = used;1754 frag->binary_patch_method = patch_method;1755 return frag;17561757 corrupt:1758 free(data);1759 *status_p = -1;1760 error("corrupt binary patch at line %d: %.*s",1761 linenr-1, llen-1, buffer);1762 return NULL;1763}17641765static int parse_binary(char *buffer, unsigned long size, struct patch *patch)1766{1767 /*1768 * We have read "GIT binary patch\n"; what follows is a line1769 * that says the patch method (currently, either "literal" or1770 * "delta") and the length of data before deflating; a1771 * sequence of 'length-byte' followed by base-85 encoded data1772 * follows.1773 *1774 * When a binary patch is reversible, there is another binary1775 * hunk in the same format, starting with patch method (either1776 * "literal" or "delta") with the length of data, and a sequence1777 * of length-byte + base-85 encoded data, terminated with another1778 * empty line. This data, when applied to the postimage, produces1779 * the preimage.1780 */1781 struct fragment *forward;1782 struct fragment *reverse;1783 int status;1784 int used, used_1;17851786 forward = parse_binary_hunk(&buffer, &size, &status, &used);1787 if (!forward && !status)1788 /* there has to be one hunk (forward hunk) */1789 return error("unrecognized binary patch at line %d", linenr-1);1790 if (status)1791 /* otherwise we already gave an error message */1792 return status;17931794 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);1795 if (reverse)1796 used += used_1;1797 else if (status) {1798 /*1799 * Not having reverse hunk is not an error, but having1800 * a corrupt reverse hunk is.1801 */1802 free((void*) forward->patch);1803 free(forward);1804 return status;1805 }1806 forward->next = reverse;1807 patch->fragments = forward;1808 patch->is_binary = 1;1809 return used;1810}18111812static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)1813{1814 int hdrsize, patchsize;1815 int offset = find_header(buffer, size, &hdrsize, patch);18161817 if (offset < 0)1818 return offset;18191820 patch->ws_rule = whitespace_rule(patch->new_name1821 ? patch->new_name1822 : patch->old_name);18231824 patchsize = parse_single_patch(buffer + offset + hdrsize,1825 size - offset - hdrsize, patch);18261827 if (!patchsize) {1828 static const char *binhdr[] = {1829 "Binary files ",1830 "Files ",1831 NULL,1832 };1833 static const char git_binary[] = "GIT binary patch\n";1834 int i;1835 int hd = hdrsize + offset;1836 unsigned long llen = linelen(buffer + hd, size - hd);18371838 if (llen == sizeof(git_binary) - 1 &&1839 !memcmp(git_binary, buffer + hd, llen)) {1840 int used;1841 linenr++;1842 used = parse_binary(buffer + hd + llen,1843 size - hd - llen, patch);1844 if (used)1845 patchsize = used + llen;1846 else1847 patchsize = 0;1848 }1849 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {1850 for (i = 0; binhdr[i]; i++) {1851 int len = strlen(binhdr[i]);1852 if (len < size - hd &&1853 !memcmp(binhdr[i], buffer + hd, len)) {1854 linenr++;1855 patch->is_binary = 1;1856 patchsize = llen;1857 break;1858 }1859 }1860 }18611862 /* Empty patch cannot be applied if it is a text patch1863 * without metadata change. A binary patch appears1864 * empty to us here.1865 */1866 if ((apply || check) &&1867 (!patch->is_binary && !metadata_changes(patch)))1868 die("patch with only garbage at line %d", linenr);1869 }18701871 return offset + hdrsize + patchsize;1872}18731874#define swap(a,b) myswap((a),(b),sizeof(a))18751876#define myswap(a, b, size) do { \1877 unsigned char mytmp[size]; \1878 memcpy(mytmp, &a, size); \1879 memcpy(&a, &b, size); \1880 memcpy(&b, mytmp, size); \1881} while (0)18821883static void reverse_patches(struct patch *p)1884{1885 for (; p; p = p->next) {1886 struct fragment *frag = p->fragments;18871888 swap(p->new_name, p->old_name);1889 swap(p->new_mode, p->old_mode);1890 swap(p->is_new, p->is_delete);1891 swap(p->lines_added, p->lines_deleted);1892 swap(p->old_sha1_prefix, p->new_sha1_prefix);18931894 for (; frag; frag = frag->next) {1895 swap(frag->newpos, frag->oldpos);1896 swap(frag->newlines, frag->oldlines);1897 }1898 }1899}19001901static const char pluses[] =1902"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";1903static const char minuses[]=1904"----------------------------------------------------------------------";19051906static void show_stats(struct patch *patch)1907{1908 struct strbuf qname = STRBUF_INIT;1909 char *cp = patch->new_name ? patch->new_name : patch->old_name;1910 int max, add, del;19111912 quote_c_style(cp, &qname, NULL, 0);19131914 /*1915 * "scale" the filename1916 */1917 max = max_len;1918 if (max > 50)1919 max = 50;19201921 if (qname.len > max) {1922 cp = strchr(qname.buf + qname.len + 3 - max, '/');1923 if (!cp)1924 cp = qname.buf + qname.len + 3 - max;1925 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);1926 }19271928 if (patch->is_binary) {1929 printf(" %-*s | Bin\n", max, qname.buf);1930 strbuf_release(&qname);1931 return;1932 }19331934 printf(" %-*s |", max, qname.buf);1935 strbuf_release(&qname);19361937 /*1938 * scale the add/delete1939 */1940 max = max + max_change > 70 ? 70 - max : max_change;1941 add = patch->lines_added;1942 del = patch->lines_deleted;19431944 if (max_change > 0) {1945 int total = ((add + del) * max + max_change / 2) / max_change;1946 add = (add * max + max_change / 2) / max_change;1947 del = total - add;1948 }1949 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,1950 add, pluses, del, minuses);1951}19521953static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)1954{1955 switch (st->st_mode & S_IFMT) {1956 case S_IFLNK:1957 if (strbuf_readlink(buf, path, st->st_size) < 0)1958 return error("unable to read symlink %s", path);1959 return 0;1960 case S_IFREG:1961 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)1962 return error("unable to open or read %s", path);1963 convert_to_git(path, buf->buf, buf->len, buf, 0);1964 return 0;1965 default:1966 return -1;1967 }1968}19691970/*1971 * Update the preimage, and the common lines in postimage,1972 * from buffer buf of length len. If postlen is 0 the postimage1973 * is updated in place, otherwise it's updated on a new buffer1974 * of length postlen1975 */19761977static void update_pre_post_images(struct image *preimage,1978 struct image *postimage,1979 char *buf,1980 size_t len, size_t postlen)1981{1982 int i, ctx;1983 char *new, *old, *fixed;1984 struct image fixed_preimage;19851986 /*1987 * Update the preimage with whitespace fixes. Note that we1988 * are not losing preimage->buf -- apply_one_fragment() will1989 * free "oldlines".1990 */1991 prepare_image(&fixed_preimage, buf, len, 1);1992 assert(fixed_preimage.nr == preimage->nr);1993 for (i = 0; i < preimage->nr; i++)1994 fixed_preimage.line[i].flag = preimage->line[i].flag;1995 free(preimage->line_allocated);1996 *preimage = fixed_preimage;19971998 /*1999 * Adjust the common context lines in postimage. This can be2000 * done in-place when we are just doing whitespace fixing,2001 * which does not make the string grow, but needs a new buffer2002 * when ignoring whitespace causes the update, since in this case2003 * we could have e.g. tabs converted to multiple spaces.2004 * We trust the caller to tell us if the update can be done2005 * in place (postlen==0) or not.2006 */2007 old = postimage->buf;2008 if (postlen)2009 new = postimage->buf = xmalloc(postlen);2010 else2011 new = old;2012 fixed = preimage->buf;2013 for (i = ctx = 0; i < postimage->nr; i++) {2014 size_t len = postimage->line[i].len;2015 if (!(postimage->line[i].flag & LINE_COMMON)) {2016 /* an added line -- no counterparts in preimage */2017 memmove(new, old, len);2018 old += len;2019 new += len;2020 continue;2021 }20222023 /* a common context -- skip it in the original postimage */2024 old += len;20252026 /* and find the corresponding one in the fixed preimage */2027 while (ctx < preimage->nr &&2028 !(preimage->line[ctx].flag & LINE_COMMON)) {2029 fixed += preimage->line[ctx].len;2030 ctx++;2031 }2032 if (preimage->nr <= ctx)2033 die("oops");20342035 /* and copy it in, while fixing the line length */2036 len = preimage->line[ctx].len;2037 memcpy(new, fixed, len);2038 new += len;2039 fixed += len;2040 postimage->line[i].len = len;2041 ctx++;2042 }20432044 /* Fix the length of the whole thing */2045 postimage->len = new - postimage->buf;2046}20472048static int match_fragment(struct image *img,2049 struct image *preimage,2050 struct image *postimage,2051 unsigned long try,2052 int try_lno,2053 unsigned ws_rule,2054 int match_beginning, int match_end)2055{2056 int i;2057 char *fixed_buf, *buf, *orig, *target;2058 struct strbuf fixed;2059 size_t fixed_len;2060 int preimage_limit;20612062 if (preimage->nr + try_lno <= img->nr) {2063 /*2064 * The hunk falls within the boundaries of img.2065 */2066 preimage_limit = preimage->nr;2067 if (match_end && (preimage->nr + try_lno != img->nr))2068 return 0;2069 } else if (ws_error_action == correct_ws_error &&2070 (ws_rule & WS_BLANK_AT_EOF)) {2071 /*2072 * This hunk extends beyond the end of img, and we are2073 * removing blank lines at the end of the file. This2074 * many lines from the beginning of the preimage must2075 * match with img, and the remainder of the preimage2076 * must be blank.2077 */2078 preimage_limit = img->nr - try_lno;2079 } else {2080 /*2081 * The hunk extends beyond the end of the img and2082 * we are not removing blanks at the end, so we2083 * should reject the hunk at this position.2084 */2085 return 0;2086 }20872088 if (match_beginning && try_lno)2089 return 0;20902091 /* Quick hash check */2092 for (i = 0; i < preimage_limit; i++)2093 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||2094 (preimage->line[i].hash != img->line[try_lno + i].hash))2095 return 0;20962097 if (preimage_limit == preimage->nr) {2098 /*2099 * Do we have an exact match? If we were told to match2100 * at the end, size must be exactly at try+fragsize,2101 * otherwise try+fragsize must be still within the preimage,2102 * and either case, the old piece should match the preimage2103 * exactly.2104 */2105 if ((match_end2106 ? (try + preimage->len == img->len)2107 : (try + preimage->len <= img->len)) &&2108 !memcmp(img->buf + try, preimage->buf, preimage->len))2109 return 1;2110 } else {2111 /*2112 * The preimage extends beyond the end of img, so2113 * there cannot be an exact match.2114 *2115 * There must be one non-blank context line that match2116 * a line before the end of img.2117 */2118 char *buf_end;21192120 buf = preimage->buf;2121 buf_end = buf;2122 for (i = 0; i < preimage_limit; i++)2123 buf_end += preimage->line[i].len;21242125 for ( ; buf < buf_end; buf++)2126 if (!isspace(*buf))2127 break;2128 if (buf == buf_end)2129 return 0;2130 }21312132 /*2133 * No exact match. If we are ignoring whitespace, run a line-by-line2134 * fuzzy matching. We collect all the line length information because2135 * we need it to adjust whitespace if we match.2136 */2137 if (ws_ignore_action == ignore_ws_change) {2138 size_t imgoff = 0;2139 size_t preoff = 0;2140 size_t postlen = postimage->len;2141 size_t extra_chars;2142 char *preimage_eof;2143 char *preimage_end;2144 for (i = 0; i < preimage_limit; i++) {2145 size_t prelen = preimage->line[i].len;2146 size_t imglen = img->line[try_lno+i].len;21472148 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,2149 preimage->buf + preoff, prelen))2150 return 0;2151 if (preimage->line[i].flag & LINE_COMMON)2152 postlen += imglen - prelen;2153 imgoff += imglen;2154 preoff += prelen;2155 }21562157 /*2158 * Ok, the preimage matches with whitespace fuzz.2159 *2160 * imgoff now holds the true length of the target that2161 * matches the preimage before the end of the file.2162 *2163 * Count the number of characters in the preimage that fall2164 * beyond the end of the file and make sure that all of them2165 * are whitespace characters. (This can only happen if2166 * we are removing blank lines at the end of the file.)2167 */2168 buf = preimage_eof = preimage->buf + preoff;2169 for ( ; i < preimage->nr; i++)2170 preoff += preimage->line[i].len;2171 preimage_end = preimage->buf + preoff;2172 for ( ; buf < preimage_end; buf++)2173 if (!isspace(*buf))2174 return 0;21752176 /*2177 * Update the preimage and the common postimage context2178 * lines to use the same whitespace as the target.2179 * If whitespace is missing in the target (i.e.2180 * if the preimage extends beyond the end of the file),2181 * use the whitespace from the preimage.2182 */2183 extra_chars = preimage_end - preimage_eof;2184 strbuf_init(&fixed, imgoff + extra_chars);2185 strbuf_add(&fixed, img->buf + try, imgoff);2186 strbuf_add(&fixed, preimage_eof, extra_chars);2187 fixed_buf = strbuf_detach(&fixed, &fixed_len);2188 update_pre_post_images(preimage, postimage,2189 fixed_buf, fixed_len, postlen);2190 return 1;2191 }21922193 if (ws_error_action != correct_ws_error)2194 return 0;21952196 /*2197 * The hunk does not apply byte-by-byte, but the hash says2198 * it might with whitespace fuzz. We haven't been asked to2199 * ignore whitespace, we were asked to correct whitespace2200 * errors, so let's try matching after whitespace correction.2201 *2202 * The preimage may extend beyond the end of the file,2203 * but in this loop we will only handle the part of the2204 * preimage that falls within the file.2205 */2206 strbuf_init(&fixed, preimage->len + 1);2207 orig = preimage->buf;2208 target = img->buf + try;2209 for (i = 0; i < preimage_limit; i++) {2210 size_t oldlen = preimage->line[i].len;2211 size_t tgtlen = img->line[try_lno + i].len;2212 size_t fixstart = fixed.len;2213 struct strbuf tgtfix;2214 int match;22152216 /* Try fixing the line in the preimage */2217 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);22182219 /* Try fixing the line in the target */2220 strbuf_init(&tgtfix, tgtlen);2221 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);22222223 /*2224 * If they match, either the preimage was based on2225 * a version before our tree fixed whitespace breakage,2226 * or we are lacking a whitespace-fix patch the tree2227 * the preimage was based on already had (i.e. target2228 * has whitespace breakage, the preimage doesn't).2229 * In either case, we are fixing the whitespace breakages2230 * so we might as well take the fix together with their2231 * real change.2232 */2233 match = (tgtfix.len == fixed.len - fixstart &&2234 !memcmp(tgtfix.buf, fixed.buf + fixstart,2235 fixed.len - fixstart));22362237 strbuf_release(&tgtfix);2238 if (!match)2239 goto unmatch_exit;22402241 orig += oldlen;2242 target += tgtlen;2243 }224422452246 /*2247 * Now handle the lines in the preimage that falls beyond the2248 * end of the file (if any). They will only match if they are2249 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2250 * false).2251 */2252 for ( ; i < preimage->nr; i++) {2253 size_t fixstart = fixed.len; /* start of the fixed preimage */2254 size_t oldlen = preimage->line[i].len;2255 int j;22562257 /* Try fixing the line in the preimage */2258 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);22592260 for (j = fixstart; j < fixed.len; j++)2261 if (!isspace(fixed.buf[j]))2262 goto unmatch_exit;22632264 orig += oldlen;2265 }22662267 /*2268 * Yes, the preimage is based on an older version that still2269 * has whitespace breakages unfixed, and fixing them makes the2270 * hunk match. Update the context lines in the postimage.2271 */2272 fixed_buf = strbuf_detach(&fixed, &fixed_len);2273 update_pre_post_images(preimage, postimage,2274 fixed_buf, fixed_len, 0);2275 return 1;22762277 unmatch_exit:2278 strbuf_release(&fixed);2279 return 0;2280}22812282static int find_pos(struct image *img,2283 struct image *preimage,2284 struct image *postimage,2285 int line,2286 unsigned ws_rule,2287 int match_beginning, int match_end)2288{2289 int i;2290 unsigned long backwards, forwards, try;2291 int backwards_lno, forwards_lno, try_lno;22922293 /*2294 * If match_beginning or match_end is specified, there is no2295 * point starting from a wrong line that will never match and2296 * wander around and wait for a match at the specified end.2297 */2298 if (match_beginning)2299 line = 0;2300 else if (match_end)2301 line = img->nr - preimage->nr;23022303 /*2304 * Because the comparison is unsigned, the following test2305 * will also take care of a negative line number that can2306 * result when match_end and preimage is larger than the target.2307 */2308 if ((size_t) line > img->nr)2309 line = img->nr;23102311 try = 0;2312 for (i = 0; i < line; i++)2313 try += img->line[i].len;23142315 /*2316 * There's probably some smart way to do this, but I'll leave2317 * that to the smart and beautiful people. I'm simple and stupid.2318 */2319 backwards = try;2320 backwards_lno = line;2321 forwards = try;2322 forwards_lno = line;2323 try_lno = line;23242325 for (i = 0; ; i++) {2326 if (match_fragment(img, preimage, postimage,2327 try, try_lno, ws_rule,2328 match_beginning, match_end))2329 return try_lno;23302331 again:2332 if (backwards_lno == 0 && forwards_lno == img->nr)2333 break;23342335 if (i & 1) {2336 if (backwards_lno == 0) {2337 i++;2338 goto again;2339 }2340 backwards_lno--;2341 backwards -= img->line[backwards_lno].len;2342 try = backwards;2343 try_lno = backwards_lno;2344 } else {2345 if (forwards_lno == img->nr) {2346 i++;2347 goto again;2348 }2349 forwards += img->line[forwards_lno].len;2350 forwards_lno++;2351 try = forwards;2352 try_lno = forwards_lno;2353 }23542355 }2356 return -1;2357}23582359static void remove_first_line(struct image *img)2360{2361 img->buf += img->line[0].len;2362 img->len -= img->line[0].len;2363 img->line++;2364 img->nr--;2365}23662367static void remove_last_line(struct image *img)2368{2369 img->len -= img->line[--img->nr].len;2370}23712372static void update_image(struct image *img,2373 int applied_pos,2374 struct image *preimage,2375 struct image *postimage)2376{2377 /*2378 * remove the copy of preimage at offset in img2379 * and replace it with postimage2380 */2381 int i, nr;2382 size_t remove_count, insert_count, applied_at = 0;2383 char *result;2384 int preimage_limit;23852386 /*2387 * If we are removing blank lines at the end of img,2388 * the preimage may extend beyond the end.2389 * If that is the case, we must be careful only to2390 * remove the part of the preimage that falls within2391 * the boundaries of img. Initialize preimage_limit2392 * to the number of lines in the preimage that falls2393 * within the boundaries.2394 */2395 preimage_limit = preimage->nr;2396 if (preimage_limit > img->nr - applied_pos)2397 preimage_limit = img->nr - applied_pos;23982399 for (i = 0; i < applied_pos; i++)2400 applied_at += img->line[i].len;24012402 remove_count = 0;2403 for (i = 0; i < preimage_limit; i++)2404 remove_count += img->line[applied_pos + i].len;2405 insert_count = postimage->len;24062407 /* Adjust the contents */2408 result = xmalloc(img->len + insert_count - remove_count + 1);2409 memcpy(result, img->buf, applied_at);2410 memcpy(result + applied_at, postimage->buf, postimage->len);2411 memcpy(result + applied_at + postimage->len,2412 img->buf + (applied_at + remove_count),2413 img->len - (applied_at + remove_count));2414 free(img->buf);2415 img->buf = result;2416 img->len += insert_count - remove_count;2417 result[img->len] = '\0';24182419 /* Adjust the line table */2420 nr = img->nr + postimage->nr - preimage_limit;2421 if (preimage_limit < postimage->nr) {2422 /*2423 * NOTE: this knows that we never call remove_first_line()2424 * on anything other than pre/post image.2425 */2426 img->line = xrealloc(img->line, nr * sizeof(*img->line));2427 img->line_allocated = img->line;2428 }2429 if (preimage_limit != postimage->nr)2430 memmove(img->line + applied_pos + postimage->nr,2431 img->line + applied_pos + preimage_limit,2432 (img->nr - (applied_pos + preimage_limit)) *2433 sizeof(*img->line));2434 memcpy(img->line + applied_pos,2435 postimage->line,2436 postimage->nr * sizeof(*img->line));2437 if (!allow_overlap)2438 for (i = 0; i < postimage->nr; i++)2439 img->line[applied_pos + i].flag |= LINE_PATCHED;2440 img->nr = nr;2441}24422443static int apply_one_fragment(struct image *img, struct fragment *frag,2444 int inaccurate_eof, unsigned ws_rule,2445 int nth_fragment)2446{2447 int match_beginning, match_end;2448 const char *patch = frag->patch;2449 int size = frag->size;2450 char *old, *oldlines;2451 struct strbuf newlines;2452 int new_blank_lines_at_end = 0;2453 int found_new_blank_lines_at_end = 0;2454 int hunk_linenr = frag->linenr;2455 unsigned long leading, trailing;2456 int pos, applied_pos;2457 struct image preimage;2458 struct image postimage;24592460 memset(&preimage, 0, sizeof(preimage));2461 memset(&postimage, 0, sizeof(postimage));2462 oldlines = xmalloc(size);2463 strbuf_init(&newlines, size);24642465 old = oldlines;2466 while (size > 0) {2467 char first;2468 int len = linelen(patch, size);2469 int plen;2470 int added_blank_line = 0;2471 int is_blank_context = 0;2472 size_t start;24732474 if (!len)2475 break;24762477 /*2478 * "plen" is how much of the line we should use for2479 * the actual patch data. Normally we just remove the2480 * first character on the line, but if the line is2481 * followed by "\ No newline", then we also remove the2482 * last one (which is the newline, of course).2483 */2484 plen = len - 1;2485 if (len < size && patch[len] == '\\')2486 plen--;2487 first = *patch;2488 if (apply_in_reverse) {2489 if (first == '-')2490 first = '+';2491 else if (first == '+')2492 first = '-';2493 }24942495 switch (first) {2496 case '\n':2497 /* Newer GNU diff, empty context line */2498 if (plen < 0)2499 /* ... followed by '\No newline'; nothing */2500 break;2501 *old++ = '\n';2502 strbuf_addch(&newlines, '\n');2503 add_line_info(&preimage, "\n", 1, LINE_COMMON);2504 add_line_info(&postimage, "\n", 1, LINE_COMMON);2505 is_blank_context = 1;2506 break;2507 case ' ':2508 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2509 ws_blank_line(patch + 1, plen, ws_rule))2510 is_blank_context = 1;2511 case '-':2512 memcpy(old, patch + 1, plen);2513 add_line_info(&preimage, old, plen,2514 (first == ' ' ? LINE_COMMON : 0));2515 old += plen;2516 if (first == '-')2517 break;2518 /* Fall-through for ' ' */2519 case '+':2520 /* --no-add does not add new lines */2521 if (first == '+' && no_add)2522 break;25232524 start = newlines.len;2525 if (first != '+' ||2526 !whitespace_error ||2527 ws_error_action != correct_ws_error) {2528 strbuf_add(&newlines, patch + 1, plen);2529 }2530 else {2531 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);2532 }2533 add_line_info(&postimage, newlines.buf + start, newlines.len - start,2534 (first == '+' ? 0 : LINE_COMMON));2535 if (first == '+' &&2536 (ws_rule & WS_BLANK_AT_EOF) &&2537 ws_blank_line(patch + 1, plen, ws_rule))2538 added_blank_line = 1;2539 break;2540 case '@': case '\\':2541 /* Ignore it, we already handled it */2542 break;2543 default:2544 if (apply_verbosely)2545 error("invalid start of line: '%c'", first);2546 return -1;2547 }2548 if (added_blank_line) {2549 if (!new_blank_lines_at_end)2550 found_new_blank_lines_at_end = hunk_linenr;2551 new_blank_lines_at_end++;2552 }2553 else if (is_blank_context)2554 ;2555 else2556 new_blank_lines_at_end = 0;2557 patch += len;2558 size -= len;2559 hunk_linenr++;2560 }2561 if (inaccurate_eof &&2562 old > oldlines && old[-1] == '\n' &&2563 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {2564 old--;2565 strbuf_setlen(&newlines, newlines.len - 1);2566 }25672568 leading = frag->leading;2569 trailing = frag->trailing;25702571 /*2572 * A hunk to change lines at the beginning would begin with2573 * @@ -1,L +N,M @@2574 * but we need to be careful. -U0 that inserts before the second2575 * line also has this pattern.2576 *2577 * And a hunk to add to an empty file would begin with2578 * @@ -0,0 +N,M @@2579 *2580 * In other words, a hunk that is (frag->oldpos <= 1) with or2581 * without leading context must match at the beginning.2582 */2583 match_beginning = (!frag->oldpos ||2584 (frag->oldpos == 1 && !unidiff_zero));25852586 /*2587 * A hunk without trailing lines must match at the end.2588 * However, we simply cannot tell if a hunk must match end2589 * from the lack of trailing lines if the patch was generated2590 * with unidiff without any context.2591 */2592 match_end = !unidiff_zero && !trailing;25932594 pos = frag->newpos ? (frag->newpos - 1) : 0;2595 preimage.buf = oldlines;2596 preimage.len = old - oldlines;2597 postimage.buf = newlines.buf;2598 postimage.len = newlines.len;2599 preimage.line = preimage.line_allocated;2600 postimage.line = postimage.line_allocated;26012602 for (;;) {26032604 applied_pos = find_pos(img, &preimage, &postimage, pos,2605 ws_rule, match_beginning, match_end);26062607 if (applied_pos >= 0)2608 break;26092610 /* Am I at my context limits? */2611 if ((leading <= p_context) && (trailing <= p_context))2612 break;2613 if (match_beginning || match_end) {2614 match_beginning = match_end = 0;2615 continue;2616 }26172618 /*2619 * Reduce the number of context lines; reduce both2620 * leading and trailing if they are equal otherwise2621 * just reduce the larger context.2622 */2623 if (leading >= trailing) {2624 remove_first_line(&preimage);2625 remove_first_line(&postimage);2626 pos--;2627 leading--;2628 }2629 if (trailing > leading) {2630 remove_last_line(&preimage);2631 remove_last_line(&postimage);2632 trailing--;2633 }2634 }26352636 if (applied_pos >= 0) {2637 if (new_blank_lines_at_end &&2638 preimage.nr + applied_pos >= img->nr &&2639 (ws_rule & WS_BLANK_AT_EOF) &&2640 ws_error_action != nowarn_ws_error) {2641 record_ws_error(WS_BLANK_AT_EOF, "+", 1,2642 found_new_blank_lines_at_end);2643 if (ws_error_action == correct_ws_error) {2644 while (new_blank_lines_at_end--)2645 remove_last_line(&postimage);2646 }2647 /*2648 * We would want to prevent write_out_results()2649 * from taking place in apply_patch() that follows2650 * the callchain led us here, which is:2651 * apply_patch->check_patch_list->check_patch->2652 * apply_data->apply_fragments->apply_one_fragment2653 */2654 if (ws_error_action == die_on_ws_error)2655 apply = 0;2656 }26572658 if (apply_verbosely && applied_pos != pos) {2659 int offset = applied_pos - pos;2660 if (apply_in_reverse)2661 offset = 0 - offset;2662 fprintf(stderr,2663 "Hunk #%d succeeded at %d (offset %d lines).\n",2664 nth_fragment, applied_pos + 1, offset);2665 }26662667 /*2668 * Warn if it was necessary to reduce the number2669 * of context lines.2670 */2671 if ((leading != frag->leading) ||2672 (trailing != frag->trailing))2673 fprintf(stderr, "Context reduced to (%ld/%ld)"2674 " to apply fragment at %d\n",2675 leading, trailing, applied_pos+1);2676 update_image(img, applied_pos, &preimage, &postimage);2677 } else {2678 if (apply_verbosely)2679 error("while searching for:\n%.*s",2680 (int)(old - oldlines), oldlines);2681 }26822683 free(oldlines);2684 strbuf_release(&newlines);2685 free(preimage.line_allocated);2686 free(postimage.line_allocated);26872688 return (applied_pos < 0);2689}26902691static int apply_binary_fragment(struct image *img, struct patch *patch)2692{2693 struct fragment *fragment = patch->fragments;2694 unsigned long len;2695 void *dst;26962697 if (!fragment)2698 return error("missing binary patch data for '%s'",2699 patch->new_name ?2700 patch->new_name :2701 patch->old_name);27022703 /* Binary patch is irreversible without the optional second hunk */2704 if (apply_in_reverse) {2705 if (!fragment->next)2706 return error("cannot reverse-apply a binary patch "2707 "without the reverse hunk to '%s'",2708 patch->new_name2709 ? patch->new_name : patch->old_name);2710 fragment = fragment->next;2711 }2712 switch (fragment->binary_patch_method) {2713 case BINARY_DELTA_DEFLATED:2714 dst = patch_delta(img->buf, img->len, fragment->patch,2715 fragment->size, &len);2716 if (!dst)2717 return -1;2718 clear_image(img);2719 img->buf = dst;2720 img->len = len;2721 return 0;2722 case BINARY_LITERAL_DEFLATED:2723 clear_image(img);2724 img->len = fragment->size;2725 img->buf = xmalloc(img->len+1);2726 memcpy(img->buf, fragment->patch, img->len);2727 img->buf[img->len] = '\0';2728 return 0;2729 }2730 return -1;2731}27322733static int apply_binary(struct image *img, struct patch *patch)2734{2735 const char *name = patch->old_name ? patch->old_name : patch->new_name;2736 unsigned char sha1[20];27372738 /*2739 * For safety, we require patch index line to contain2740 * full 40-byte textual SHA1 for old and new, at least for now.2741 */2742 if (strlen(patch->old_sha1_prefix) != 40 ||2743 strlen(patch->new_sha1_prefix) != 40 ||2744 get_sha1_hex(patch->old_sha1_prefix, sha1) ||2745 get_sha1_hex(patch->new_sha1_prefix, sha1))2746 return error("cannot apply binary patch to '%s' "2747 "without full index line", name);27482749 if (patch->old_name) {2750 /*2751 * See if the old one matches what the patch2752 * applies to.2753 */2754 hash_sha1_file(img->buf, img->len, blob_type, sha1);2755 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))2756 return error("the patch applies to '%s' (%s), "2757 "which does not match the "2758 "current contents.",2759 name, sha1_to_hex(sha1));2760 }2761 else {2762 /* Otherwise, the old one must be empty. */2763 if (img->len)2764 return error("the patch applies to an empty "2765 "'%s' but it is not empty", name);2766 }27672768 get_sha1_hex(patch->new_sha1_prefix, sha1);2769 if (is_null_sha1(sha1)) {2770 clear_image(img);2771 return 0; /* deletion patch */2772 }27732774 if (has_sha1_file(sha1)) {2775 /* We already have the postimage */2776 enum object_type type;2777 unsigned long size;2778 char *result;27792780 result = read_sha1_file(sha1, &type, &size);2781 if (!result)2782 return error("the necessary postimage %s for "2783 "'%s' cannot be read",2784 patch->new_sha1_prefix, name);2785 clear_image(img);2786 img->buf = result;2787 img->len = size;2788 } else {2789 /*2790 * We have verified buf matches the preimage;2791 * apply the patch data to it, which is stored2792 * in the patch->fragments->{patch,size}.2793 */2794 if (apply_binary_fragment(img, patch))2795 return error("binary patch does not apply to '%s'",2796 name);27972798 /* verify that the result matches */2799 hash_sha1_file(img->buf, img->len, blob_type, sha1);2800 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))2801 return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)",2802 name, patch->new_sha1_prefix, sha1_to_hex(sha1));2803 }28042805 return 0;2806}28072808static int apply_fragments(struct image *img, struct patch *patch)2809{2810 struct fragment *frag = patch->fragments;2811 const char *name = patch->old_name ? patch->old_name : patch->new_name;2812 unsigned ws_rule = patch->ws_rule;2813 unsigned inaccurate_eof = patch->inaccurate_eof;2814 int nth = 0;28152816 if (patch->is_binary)2817 return apply_binary(img, patch);28182819 while (frag) {2820 nth++;2821 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {2822 error("patch failed: %s:%ld", name, frag->oldpos);2823 if (!apply_with_reject)2824 return -1;2825 frag->rejected = 1;2826 }2827 frag = frag->next;2828 }2829 return 0;2830}28312832static int read_file_or_gitlink(struct cache_entry *ce, struct strbuf *buf)2833{2834 if (!ce)2835 return 0;28362837 if (S_ISGITLINK(ce->ce_mode)) {2838 strbuf_grow(buf, 100);2839 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(ce->sha1));2840 } else {2841 enum object_type type;2842 unsigned long sz;2843 char *result;28442845 result = read_sha1_file(ce->sha1, &type, &sz);2846 if (!result)2847 return -1;2848 /* XXX read_sha1_file NUL-terminates */2849 strbuf_attach(buf, result, sz, sz + 1);2850 }2851 return 0;2852}28532854static struct patch *in_fn_table(const char *name)2855{2856 struct string_list_item *item;28572858 if (name == NULL)2859 return NULL;28602861 item = string_list_lookup(&fn_table, name);2862 if (item != NULL)2863 return (struct patch *)item->util;28642865 return NULL;2866}28672868/*2869 * item->util in the filename table records the status of the path.2870 * Usually it points at a patch (whose result records the contents2871 * of it after applying it), but it could be PATH_WAS_DELETED for a2872 * path that a previously applied patch has already removed.2873 */2874 #define PATH_TO_BE_DELETED ((struct patch *) -2)2875#define PATH_WAS_DELETED ((struct patch *) -1)28762877static int to_be_deleted(struct patch *patch)2878{2879 return patch == PATH_TO_BE_DELETED;2880}28812882static int was_deleted(struct patch *patch)2883{2884 return patch == PATH_WAS_DELETED;2885}28862887static void add_to_fn_table(struct patch *patch)2888{2889 struct string_list_item *item;28902891 /*2892 * Always add new_name unless patch is a deletion2893 * This should cover the cases for normal diffs,2894 * file creations and copies2895 */2896 if (patch->new_name != NULL) {2897 item = string_list_insert(&fn_table, patch->new_name);2898 item->util = patch;2899 }29002901 /*2902 * store a failure on rename/deletion cases because2903 * later chunks shouldn't patch old names2904 */2905 if ((patch->new_name == NULL) || (patch->is_rename)) {2906 item = string_list_insert(&fn_table, patch->old_name);2907 item->util = PATH_WAS_DELETED;2908 }2909}29102911static void prepare_fn_table(struct patch *patch)2912{2913 /*2914 * store information about incoming file deletion2915 */2916 while (patch) {2917 if ((patch->new_name == NULL) || (patch->is_rename)) {2918 struct string_list_item *item;2919 item = string_list_insert(&fn_table, patch->old_name);2920 item->util = PATH_TO_BE_DELETED;2921 }2922 patch = patch->next;2923 }2924}29252926static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)2927{2928 struct strbuf buf = STRBUF_INIT;2929 struct image image;2930 size_t len;2931 char *img;2932 struct patch *tpatch;29332934 if (!(patch->is_copy || patch->is_rename) &&2935 (tpatch = in_fn_table(patch->old_name)) != NULL && !to_be_deleted(tpatch)) {2936 if (was_deleted(tpatch)) {2937 return error("patch %s has been renamed/deleted",2938 patch->old_name);2939 }2940 /* We have a patched copy in memory use that */2941 strbuf_add(&buf, tpatch->result, tpatch->resultsize);2942 } else if (cached) {2943 if (read_file_or_gitlink(ce, &buf))2944 return error("read of %s failed", patch->old_name);2945 } else if (patch->old_name) {2946 if (S_ISGITLINK(patch->old_mode)) {2947 if (ce) {2948 read_file_or_gitlink(ce, &buf);2949 } else {2950 /*2951 * There is no way to apply subproject2952 * patch without looking at the index.2953 */2954 patch->fragments = NULL;2955 }2956 } else {2957 if (read_old_data(st, patch->old_name, &buf))2958 return error("read of %s failed", patch->old_name);2959 }2960 }29612962 img = strbuf_detach(&buf, &len);2963 prepare_image(&image, img, len, !patch->is_binary);29642965 if (apply_fragments(&image, patch) < 0)2966 return -1; /* note with --reject this succeeds. */2967 patch->result = image.buf;2968 patch->resultsize = image.len;2969 add_to_fn_table(patch);2970 free(image.line_allocated);29712972 if (0 < patch->is_delete && patch->resultsize)2973 return error("removal patch leaves file contents");29742975 return 0;2976}29772978static int check_to_create_blob(const char *new_name, int ok_if_exists)2979{2980 struct stat nst;2981 if (!lstat(new_name, &nst)) {2982 if (S_ISDIR(nst.st_mode) || ok_if_exists)2983 return 0;2984 /*2985 * A leading component of new_name might be a symlink2986 * that is going to be removed with this patch, but2987 * still pointing at somewhere that has the path.2988 * In such a case, path "new_name" does not exist as2989 * far as git is concerned.2990 */2991 if (has_symlink_leading_path(new_name, strlen(new_name)))2992 return 0;29932994 return error("%s: already exists in working directory", new_name);2995 }2996 else if ((errno != ENOENT) && (errno != ENOTDIR))2997 return error("%s: %s", new_name, strerror(errno));2998 return 0;2999}30003001static int verify_index_match(struct cache_entry *ce, struct stat *st)3002{3003 if (S_ISGITLINK(ce->ce_mode)) {3004 if (!S_ISDIR(st->st_mode))3005 return -1;3006 return 0;3007 }3008 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);3009}30103011static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)3012{3013 const char *old_name = patch->old_name;3014 struct patch *tpatch = NULL;3015 int stat_ret = 0;3016 unsigned st_mode = 0;30173018 /*3019 * Make sure that we do not have local modifications from the3020 * index when we are looking at the index. Also make sure3021 * we have the preimage file to be patched in the work tree,3022 * unless --cached, which tells git to apply only in the index.3023 */3024 if (!old_name)3025 return 0;30263027 assert(patch->is_new <= 0);30283029 if (!(patch->is_copy || patch->is_rename) &&3030 (tpatch = in_fn_table(old_name)) != NULL && !to_be_deleted(tpatch)) {3031 if (was_deleted(tpatch))3032 return error("%s: has been deleted/renamed", old_name);3033 st_mode = tpatch->new_mode;3034 } else if (!cached) {3035 stat_ret = lstat(old_name, st);3036 if (stat_ret && errno != ENOENT)3037 return error("%s: %s", old_name, strerror(errno));3038 }30393040 if (to_be_deleted(tpatch))3041 tpatch = NULL;30423043 if (check_index && !tpatch) {3044 int pos = cache_name_pos(old_name, strlen(old_name));3045 if (pos < 0) {3046 if (patch->is_new < 0)3047 goto is_new;3048 return error("%s: does not exist in index", old_name);3049 }3050 *ce = active_cache[pos];3051 if (stat_ret < 0) {3052 struct checkout costate;3053 /* checkout */3054 memset(&costate, 0, sizeof(costate));3055 costate.base_dir = "";3056 costate.refresh_cache = 1;3057 if (checkout_entry(*ce, &costate, NULL) ||3058 lstat(old_name, st))3059 return -1;3060 }3061 if (!cached && verify_index_match(*ce, st))3062 return error("%s: does not match index", old_name);3063 if (cached)3064 st_mode = (*ce)->ce_mode;3065 } else if (stat_ret < 0) {3066 if (patch->is_new < 0)3067 goto is_new;3068 return error("%s: %s", old_name, strerror(errno));3069 }30703071 if (!cached && !tpatch)3072 st_mode = ce_mode_from_stat(*ce, st->st_mode);30733074 if (patch->is_new < 0)3075 patch->is_new = 0;3076 if (!patch->old_mode)3077 patch->old_mode = st_mode;3078 if ((st_mode ^ patch->old_mode) & S_IFMT)3079 return error("%s: wrong type", old_name);3080 if (st_mode != patch->old_mode)3081 warning("%s has type %o, expected %o",3082 old_name, st_mode, patch->old_mode);3083 if (!patch->new_mode && !patch->is_delete)3084 patch->new_mode = st_mode;3085 return 0;30863087 is_new:3088 patch->is_new = 1;3089 patch->is_delete = 0;3090 patch->old_name = NULL;3091 return 0;3092}30933094static int check_patch(struct patch *patch)3095{3096 struct stat st;3097 const char *old_name = patch->old_name;3098 const char *new_name = patch->new_name;3099 const char *name = old_name ? old_name : new_name;3100 struct cache_entry *ce = NULL;3101 struct patch *tpatch;3102 int ok_if_exists;3103 int status;31043105 patch->rejected = 1; /* we will drop this after we succeed */31063107 status = check_preimage(patch, &ce, &st);3108 if (status)3109 return status;3110 old_name = patch->old_name;31113112 if ((tpatch = in_fn_table(new_name)) &&3113 (was_deleted(tpatch) || to_be_deleted(tpatch)))3114 /*3115 * A type-change diff is always split into a patch to3116 * delete old, immediately followed by a patch to3117 * create new (see diff.c::run_diff()); in such a case3118 * it is Ok that the entry to be deleted by the3119 * previous patch is still in the working tree and in3120 * the index.3121 */3122 ok_if_exists = 1;3123 else3124 ok_if_exists = 0;31253126 if (new_name &&3127 ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {3128 if (check_index &&3129 cache_name_pos(new_name, strlen(new_name)) >= 0 &&3130 !ok_if_exists)3131 return error("%s: already exists in index", new_name);3132 if (!cached) {3133 int err = check_to_create_blob(new_name, ok_if_exists);3134 if (err)3135 return err;3136 }3137 if (!patch->new_mode) {3138 if (0 < patch->is_new)3139 patch->new_mode = S_IFREG | 0644;3140 else3141 patch->new_mode = patch->old_mode;3142 }3143 }31443145 if (new_name && old_name) {3146 int same = !strcmp(old_name, new_name);3147 if (!patch->new_mode)3148 patch->new_mode = patch->old_mode;3149 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)3150 return error("new mode (%o) of %s does not match old mode (%o)%s%s",3151 patch->new_mode, new_name, patch->old_mode,3152 same ? "" : " of ", same ? "" : old_name);3153 }31543155 if (apply_data(patch, &st, ce) < 0)3156 return error("%s: patch does not apply", name);3157 patch->rejected = 0;3158 return 0;3159}31603161static int check_patch_list(struct patch *patch)3162{3163 int err = 0;31643165 prepare_fn_table(patch);3166 while (patch) {3167 if (apply_verbosely)3168 say_patch_name(stderr,3169 "Checking patch ", patch, "...\n");3170 err |= check_patch(patch);3171 patch = patch->next;3172 }3173 return err;3174}31753176/* This function tries to read the sha1 from the current index */3177static int get_current_sha1(const char *path, unsigned char *sha1)3178{3179 int pos;31803181 if (read_cache() < 0)3182 return -1;3183 pos = cache_name_pos(path, strlen(path));3184 if (pos < 0)3185 return -1;3186 hashcpy(sha1, active_cache[pos]->sha1);3187 return 0;3188}31893190/* Build an index that contains the just the files needed for a 3way merge */3191static void build_fake_ancestor(struct patch *list, const char *filename)3192{3193 struct patch *patch;3194 struct index_state result = { NULL };3195 int fd;31963197 /* Once we start supporting the reverse patch, it may be3198 * worth showing the new sha1 prefix, but until then...3199 */3200 for (patch = list; patch; patch = patch->next) {3201 const unsigned char *sha1_ptr;3202 unsigned char sha1[20];3203 struct cache_entry *ce;3204 const char *name;32053206 name = patch->old_name ? patch->old_name : patch->new_name;3207 if (0 < patch->is_new)3208 continue;3209 else if (get_sha1(patch->old_sha1_prefix, sha1))3210 /* git diff has no index line for mode/type changes */3211 if (!patch->lines_added && !patch->lines_deleted) {3212 if (get_current_sha1(patch->old_name, sha1))3213 die("mode change for %s, which is not "3214 "in current HEAD", name);3215 sha1_ptr = sha1;3216 } else3217 die("sha1 information is lacking or useless "3218 "(%s).", name);3219 else3220 sha1_ptr = sha1;32213222 ce = make_cache_entry(patch->old_mode, sha1_ptr, name, 0, 0);3223 if (!ce)3224 die("make_cache_entry failed for path '%s'", name);3225 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3226 die ("Could not add %s to temporary index", name);3227 }32283229 fd = open(filename, O_WRONLY | O_CREAT, 0666);3230 if (fd < 0 || write_index(&result, fd) || close(fd))3231 die ("Could not write temporary index to %s", filename);32323233 discard_index(&result);3234}32353236static void stat_patch_list(struct patch *patch)3237{3238 int files, adds, dels;32393240 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3241 files++;3242 adds += patch->lines_added;3243 dels += patch->lines_deleted;3244 show_stats(patch);3245 }32463247 printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);3248}32493250static void numstat_patch_list(struct patch *patch)3251{3252 for ( ; patch; patch = patch->next) {3253 const char *name;3254 name = patch->new_name ? patch->new_name : patch->old_name;3255 if (patch->is_binary)3256 printf("-\t-\t");3257 else3258 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3259 write_name_quoted(name, stdout, line_termination);3260 }3261}32623263static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3264{3265 if (mode)3266 printf(" %s mode %06o %s\n", newdelete, mode, name);3267 else3268 printf(" %s %s\n", newdelete, name);3269}32703271static void show_mode_change(struct patch *p, int show_name)3272{3273 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3274 if (show_name)3275 printf(" mode change %06o => %06o %s\n",3276 p->old_mode, p->new_mode, p->new_name);3277 else3278 printf(" mode change %06o => %06o\n",3279 p->old_mode, p->new_mode);3280 }3281}32823283static void show_rename_copy(struct patch *p)3284{3285 const char *renamecopy = p->is_rename ? "rename" : "copy";3286 const char *old, *new;32873288 /* Find common prefix */3289 old = p->old_name;3290 new = p->new_name;3291 while (1) {3292 const char *slash_old, *slash_new;3293 slash_old = strchr(old, '/');3294 slash_new = strchr(new, '/');3295 if (!slash_old ||3296 !slash_new ||3297 slash_old - old != slash_new - new ||3298 memcmp(old, new, slash_new - new))3299 break;3300 old = slash_old + 1;3301 new = slash_new + 1;3302 }3303 /* p->old_name thru old is the common prefix, and old and new3304 * through the end of names are renames3305 */3306 if (old != p->old_name)3307 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,3308 (int)(old - p->old_name), p->old_name,3309 old, new, p->score);3310 else3311 printf(" %s %s => %s (%d%%)\n", renamecopy,3312 p->old_name, p->new_name, p->score);3313 show_mode_change(p, 0);3314}33153316static void summary_patch_list(struct patch *patch)3317{3318 struct patch *p;33193320 for (p = patch; p; p = p->next) {3321 if (p->is_new)3322 show_file_mode_name("create", p->new_mode, p->new_name);3323 else if (p->is_delete)3324 show_file_mode_name("delete", p->old_mode, p->old_name);3325 else {3326 if (p->is_rename || p->is_copy)3327 show_rename_copy(p);3328 else {3329 if (p->score) {3330 printf(" rewrite %s (%d%%)\n",3331 p->new_name, p->score);3332 show_mode_change(p, 0);3333 }3334 else3335 show_mode_change(p, 1);3336 }3337 }3338 }3339}33403341static void patch_stats(struct patch *patch)3342{3343 int lines = patch->lines_added + patch->lines_deleted;33443345 if (lines > max_change)3346 max_change = lines;3347 if (patch->old_name) {3348 int len = quote_c_style(patch->old_name, NULL, NULL, 0);3349 if (!len)3350 len = strlen(patch->old_name);3351 if (len > max_len)3352 max_len = len;3353 }3354 if (patch->new_name) {3355 int len = quote_c_style(patch->new_name, NULL, NULL, 0);3356 if (!len)3357 len = strlen(patch->new_name);3358 if (len > max_len)3359 max_len = len;3360 }3361}33623363static void remove_file(struct patch *patch, int rmdir_empty)3364{3365 if (update_index) {3366 if (remove_file_from_cache(patch->old_name) < 0)3367 die("unable to remove %s from index", patch->old_name);3368 }3369 if (!cached) {3370 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {3371 remove_path(patch->old_name);3372 }3373 }3374}33753376static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)3377{3378 struct stat st;3379 struct cache_entry *ce;3380 int namelen = strlen(path);3381 unsigned ce_size = cache_entry_size(namelen);33823383 if (!update_index)3384 return;33853386 ce = xcalloc(1, ce_size);3387 memcpy(ce->name, path, namelen);3388 ce->ce_mode = create_ce_mode(mode);3389 ce->ce_flags = namelen;3390 if (S_ISGITLINK(mode)) {3391 const char *s = buf;33923393 if (get_sha1_hex(s + strlen("Subproject commit "), ce->sha1))3394 die("corrupt patch for subproject %s", path);3395 } else {3396 if (!cached) {3397 if (lstat(path, &st) < 0)3398 die_errno("unable to stat newly created file '%s'",3399 path);3400 fill_stat_cache_info(ce, &st);3401 }3402 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)3403 die("unable to create backing store for newly created file %s", path);3404 }3405 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)3406 die("unable to add cache entry for %s", path);3407}34083409static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)3410{3411 int fd;3412 struct strbuf nbuf = STRBUF_INIT;34133414 if (S_ISGITLINK(mode)) {3415 struct stat st;3416 if (!lstat(path, &st) && S_ISDIR(st.st_mode))3417 return 0;3418 return mkdir(path, 0777);3419 }34203421 if (has_symlinks && S_ISLNK(mode))3422 /* Although buf:size is counted string, it also is NUL3423 * terminated.3424 */3425 return symlink(buf, path);34263427 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);3428 if (fd < 0)3429 return -1;34303431 if (convert_to_working_tree(path, buf, size, &nbuf)) {3432 size = nbuf.len;3433 buf = nbuf.buf;3434 }3435 write_or_die(fd, buf, size);3436 strbuf_release(&nbuf);34373438 if (close(fd) < 0)3439 die_errno("closing file '%s'", path);3440 return 0;3441}34423443/*3444 * We optimistically assume that the directories exist,3445 * which is true 99% of the time anyway. If they don't,3446 * we create them and try again.3447 */3448static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)3449{3450 if (cached)3451 return;3452 if (!try_create_file(path, mode, buf, size))3453 return;34543455 if (errno == ENOENT) {3456 if (safe_create_leading_directories(path))3457 return;3458 if (!try_create_file(path, mode, buf, size))3459 return;3460 }34613462 if (errno == EEXIST || errno == EACCES) {3463 /* We may be trying to create a file where a directory3464 * used to be.3465 */3466 struct stat st;3467 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))3468 errno = EEXIST;3469 }34703471 if (errno == EEXIST) {3472 unsigned int nr = getpid();34733474 for (;;) {3475 char newpath[PATH_MAX];3476 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);3477 if (!try_create_file(newpath, mode, buf, size)) {3478 if (!rename(newpath, path))3479 return;3480 unlink_or_warn(newpath);3481 break;3482 }3483 if (errno != EEXIST)3484 break;3485 ++nr;3486 }3487 }3488 die_errno("unable to write file '%s' mode %o", path, mode);3489}34903491static void create_file(struct patch *patch)3492{3493 char *path = patch->new_name;3494 unsigned mode = patch->new_mode;3495 unsigned long size = patch->resultsize;3496 char *buf = patch->result;34973498 if (!mode)3499 mode = S_IFREG | 0644;3500 create_one_file(path, mode, buf, size);3501 add_index_file(path, mode, buf, size);3502}35033504/* phase zero is to remove, phase one is to create */3505static void write_out_one_result(struct patch *patch, int phase)3506{3507 if (patch->is_delete > 0) {3508 if (phase == 0)3509 remove_file(patch, 1);3510 return;3511 }3512 if (patch->is_new > 0 || patch->is_copy) {3513 if (phase == 1)3514 create_file(patch);3515 return;3516 }3517 /*3518 * Rename or modification boils down to the same3519 * thing: remove the old, write the new3520 */3521 if (phase == 0)3522 remove_file(patch, patch->is_rename);3523 if (phase == 1)3524 create_file(patch);3525}35263527static int write_out_one_reject(struct patch *patch)3528{3529 FILE *rej;3530 char namebuf[PATH_MAX];3531 struct fragment *frag;3532 int cnt = 0;35333534 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {3535 if (!frag->rejected)3536 continue;3537 cnt++;3538 }35393540 if (!cnt) {3541 if (apply_verbosely)3542 say_patch_name(stderr,3543 "Applied patch ", patch, " cleanly.\n");3544 return 0;3545 }35463547 /* This should not happen, because a removal patch that leaves3548 * contents are marked "rejected" at the patch level.3549 */3550 if (!patch->new_name)3551 die("internal error");35523553 /* Say this even without --verbose */3554 say_patch_name(stderr, "Applying patch ", patch, " with");3555 fprintf(stderr, " %d rejects...\n", cnt);35563557 cnt = strlen(patch->new_name);3558 if (ARRAY_SIZE(namebuf) <= cnt + 5) {3559 cnt = ARRAY_SIZE(namebuf) - 5;3560 warning("truncating .rej filename to %.*s.rej",3561 cnt - 1, patch->new_name);3562 }3563 memcpy(namebuf, patch->new_name, cnt);3564 memcpy(namebuf + cnt, ".rej", 5);35653566 rej = fopen(namebuf, "w");3567 if (!rej)3568 return error("cannot open %s: %s", namebuf, strerror(errno));35693570 /* Normal git tools never deal with .rej, so do not pretend3571 * this is a git patch by saying --git nor give extended3572 * headers. While at it, maybe please "kompare" that wants3573 * the trailing TAB and some garbage at the end of line ;-).3574 */3575 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",3576 patch->new_name, patch->new_name);3577 for (cnt = 1, frag = patch->fragments;3578 frag;3579 cnt++, frag = frag->next) {3580 if (!frag->rejected) {3581 fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);3582 continue;3583 }3584 fprintf(stderr, "Rejected hunk #%d.\n", cnt);3585 fprintf(rej, "%.*s", frag->size, frag->patch);3586 if (frag->patch[frag->size-1] != '\n')3587 fputc('\n', rej);3588 }3589 fclose(rej);3590 return -1;3591}35923593static int write_out_results(struct patch *list, int skipped_patch)3594{3595 int phase;3596 int errs = 0;3597 struct patch *l;35983599 if (!list && !skipped_patch)3600 return error("No changes");36013602 for (phase = 0; phase < 2; phase++) {3603 l = list;3604 while (l) {3605 if (l->rejected)3606 errs = 1;3607 else {3608 write_out_one_result(l, phase);3609 if (phase == 1 && write_out_one_reject(l))3610 errs = 1;3611 }3612 l = l->next;3613 }3614 }3615 return errs;3616}36173618static struct lock_file lock_file;36193620static struct string_list limit_by_name;3621static int has_include;3622static void add_name_limit(const char *name, int exclude)3623{3624 struct string_list_item *it;36253626 it = string_list_append(&limit_by_name, name);3627 it->util = exclude ? NULL : (void *) 1;3628}36293630static int use_patch(struct patch *p)3631{3632 const char *pathname = p->new_name ? p->new_name : p->old_name;3633 int i;36343635 /* Paths outside are not touched regardless of "--include" */3636 if (0 < prefix_length) {3637 int pathlen = strlen(pathname);3638 if (pathlen <= prefix_length ||3639 memcmp(prefix, pathname, prefix_length))3640 return 0;3641 }36423643 /* See if it matches any of exclude/include rule */3644 for (i = 0; i < limit_by_name.nr; i++) {3645 struct string_list_item *it = &limit_by_name.items[i];3646 if (!fnmatch(it->string, pathname, 0))3647 return (it->util != NULL);3648 }36493650 /*3651 * If we had any include, a path that does not match any rule is3652 * not used. Otherwise, we saw bunch of exclude rules (or none)3653 * and such a path is used.3654 */3655 return !has_include;3656}365736583659static void prefix_one(char **name)3660{3661 char *old_name = *name;3662 if (!old_name)3663 return;3664 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));3665 free(old_name);3666}36673668static void prefix_patches(struct patch *p)3669{3670 if (!prefix || p->is_toplevel_relative)3671 return;3672 for ( ; p; p = p->next) {3673 if (p->new_name == p->old_name) {3674 char *prefixed = p->new_name;3675 prefix_one(&prefixed);3676 p->new_name = p->old_name = prefixed;3677 }3678 else {3679 prefix_one(&p->new_name);3680 prefix_one(&p->old_name);3681 }3682 }3683}36843685#define INACCURATE_EOF (1<<0)3686#define RECOUNT (1<<1)36873688static int apply_patch(int fd, const char *filename, int options)3689{3690 size_t offset;3691 struct strbuf buf = STRBUF_INIT;3692 struct patch *list = NULL, **listp = &list;3693 int skipped_patch = 0;36943695 /* FIXME - memory leak when using multiple patch files as inputs */3696 memset(&fn_table, 0, sizeof(struct string_list));3697 patch_input_file = filename;3698 read_patch_file(&buf, fd);3699 offset = 0;3700 while (offset < buf.len) {3701 struct patch *patch;3702 int nr;37033704 patch = xcalloc(1, sizeof(*patch));3705 patch->inaccurate_eof = !!(options & INACCURATE_EOF);3706 patch->recount = !!(options & RECOUNT);3707 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);3708 if (nr < 0)3709 break;3710 if (apply_in_reverse)3711 reverse_patches(patch);3712 if (prefix)3713 prefix_patches(patch);3714 if (use_patch(patch)) {3715 patch_stats(patch);3716 *listp = patch;3717 listp = &patch->next;3718 }3719 else {3720 /* perhaps free it a bit better? */3721 free(patch);3722 skipped_patch++;3723 }3724 offset += nr;3725 }37263727 if (whitespace_error && (ws_error_action == die_on_ws_error))3728 apply = 0;37293730 update_index = check_index && apply;3731 if (update_index && newfd < 0)3732 newfd = hold_locked_index(&lock_file, 1);37333734 if (check_index) {3735 if (read_cache() < 0)3736 die("unable to read index file");3737 }37383739 if ((check || apply) &&3740 check_patch_list(list) < 0 &&3741 !apply_with_reject)3742 exit(1);37433744 if (apply && write_out_results(list, skipped_patch))3745 exit(1);37463747 if (fake_ancestor)3748 build_fake_ancestor(list, fake_ancestor);37493750 if (diffstat)3751 stat_patch_list(list);37523753 if (numstat)3754 numstat_patch_list(list);37553756 if (summary)3757 summary_patch_list(list);37583759 strbuf_release(&buf);3760 return 0;3761}37623763static int git_apply_config(const char *var, const char *value, void *cb)3764{3765 if (!strcmp(var, "apply.whitespace"))3766 return git_config_string(&apply_default_whitespace, var, value);3767 else if (!strcmp(var, "apply.ignorewhitespace"))3768 return git_config_string(&apply_default_ignorewhitespace, var, value);3769 return git_default_config(var, value, cb);3770}37713772static int option_parse_exclude(const struct option *opt,3773 const char *arg, int unset)3774{3775 add_name_limit(arg, 1);3776 return 0;3777}37783779static int option_parse_include(const struct option *opt,3780 const char *arg, int unset)3781{3782 add_name_limit(arg, 0);3783 has_include = 1;3784 return 0;3785}37863787static int option_parse_p(const struct option *opt,3788 const char *arg, int unset)3789{3790 p_value = atoi(arg);3791 p_value_known = 1;3792 return 0;3793}37943795static int option_parse_z(const struct option *opt,3796 const char *arg, int unset)3797{3798 if (unset)3799 line_termination = '\n';3800 else3801 line_termination = 0;3802 return 0;3803}38043805static int option_parse_space_change(const struct option *opt,3806 const char *arg, int unset)3807{3808 if (unset)3809 ws_ignore_action = ignore_ws_none;3810 else3811 ws_ignore_action = ignore_ws_change;3812 return 0;3813}38143815static int option_parse_whitespace(const struct option *opt,3816 const char *arg, int unset)3817{3818 const char **whitespace_option = opt->value;38193820 *whitespace_option = arg;3821 parse_whitespace_option(arg);3822 return 0;3823}38243825static int option_parse_directory(const struct option *opt,3826 const char *arg, int unset)3827{3828 root_len = strlen(arg);3829 if (root_len && arg[root_len - 1] != '/') {3830 char *new_root;3831 root = new_root = xmalloc(root_len + 2);3832 strcpy(new_root, arg);3833 strcpy(new_root + root_len++, "/");3834 } else3835 root = arg;3836 return 0;3837}38383839int cmd_apply(int argc, const char **argv, const char *prefix_)3840{3841 int i;3842 int errs = 0;3843 int is_not_gitdir = !startup_info->have_repository;3844 int binary;3845 int force_apply = 0;38463847 const char *whitespace_option = NULL;38483849 struct option builtin_apply_options[] = {3850 { OPTION_CALLBACK, 0, "exclude", NULL, "path",3851 "don't apply changes matching the given path",3852 0, option_parse_exclude },3853 { OPTION_CALLBACK, 0, "include", NULL, "path",3854 "apply changes matching the given path",3855 0, option_parse_include },3856 { OPTION_CALLBACK, 'p', NULL, NULL, "num",3857 "remove <num> leading slashes from traditional diff paths",3858 0, option_parse_p },3859 OPT_BOOLEAN(0, "no-add", &no_add,3860 "ignore additions made by the patch"),3861 OPT_BOOLEAN(0, "stat", &diffstat,3862 "instead of applying the patch, output diffstat for the input"),3863 { OPTION_BOOLEAN, 0, "allow-binary-replacement", &binary,3864 NULL, "old option, now no-op",3865 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3866 { OPTION_BOOLEAN, 0, "binary", &binary,3867 NULL, "old option, now no-op",3868 PARSE_OPT_HIDDEN | PARSE_OPT_NOARG },3869 OPT_BOOLEAN(0, "numstat", &numstat,3870 "shows number of added and deleted lines in decimal notation"),3871 OPT_BOOLEAN(0, "summary", &summary,3872 "instead of applying the patch, output a summary for the input"),3873 OPT_BOOLEAN(0, "check", &check,3874 "instead of applying the patch, see if the patch is applicable"),3875 OPT_BOOLEAN(0, "index", &check_index,3876 "make sure the patch is applicable to the current index"),3877 OPT_BOOLEAN(0, "cached", &cached,3878 "apply a patch without touching the working tree"),3879 OPT_BOOLEAN(0, "apply", &force_apply,3880 "also apply the patch (use with --stat/--summary/--check)"),3881 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,3882 "build a temporary index based on embedded index information"),3883 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,3884 "paths are separated with NUL character",3885 PARSE_OPT_NOARG, option_parse_z },3886 OPT_INTEGER('C', NULL, &p_context,3887 "ensure at least <n> lines of context match"),3888 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, "action",3889 "detect new or modified lines that have whitespace errors",3890 0, option_parse_whitespace },3891 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,3892 "ignore changes in whitespace when finding context",3893 PARSE_OPT_NOARG, option_parse_space_change },3894 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,3895 "ignore changes in whitespace when finding context",3896 PARSE_OPT_NOARG, option_parse_space_change },3897 OPT_BOOLEAN('R', "reverse", &apply_in_reverse,3898 "apply the patch in reverse"),3899 OPT_BOOLEAN(0, "unidiff-zero", &unidiff_zero,3900 "don't expect at least one line of context"),3901 OPT_BOOLEAN(0, "reject", &apply_with_reject,3902 "leave the rejected hunks in corresponding *.rej files"),3903 OPT_BOOLEAN(0, "allow-overlap", &allow_overlap,3904 "allow overlapping hunks"),3905 OPT__VERBOSE(&apply_verbosely, "be verbose"),3906 OPT_BIT(0, "inaccurate-eof", &options,3907 "tolerate incorrectly detected missing new-line at the end of file",3908 INACCURATE_EOF),3909 OPT_BIT(0, "recount", &options,3910 "do not trust the line counts in the hunk headers",3911 RECOUNT),3912 { OPTION_CALLBACK, 0, "directory", NULL, "root",3913 "prepend <root> to all filenames",3914 0, option_parse_directory },3915 OPT_END()3916 };39173918 prefix = prefix_;3919 prefix_length = prefix ? strlen(prefix) : 0;3920 git_config(git_apply_config, NULL);3921 if (apply_default_whitespace)3922 parse_whitespace_option(apply_default_whitespace);3923 if (apply_default_ignorewhitespace)3924 parse_ignorewhitespace_option(apply_default_ignorewhitespace);39253926 argc = parse_options(argc, argv, prefix, builtin_apply_options,3927 apply_usage, 0);39283929 if (apply_with_reject)3930 apply = apply_verbosely = 1;3931 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))3932 apply = 0;3933 if (check_index && is_not_gitdir)3934 die("--index outside a repository");3935 if (cached) {3936 if (is_not_gitdir)3937 die("--cached outside a repository");3938 check_index = 1;3939 }3940 for (i = 0; i < argc; i++) {3941 const char *arg = argv[i];3942 int fd;39433944 if (!strcmp(arg, "-")) {3945 errs |= apply_patch(0, "<stdin>", options);3946 read_stdin = 0;3947 continue;3948 } else if (0 < prefix_length)3949 arg = prefix_filename(prefix, prefix_length, arg);39503951 fd = open(arg, O_RDONLY);3952 if (fd < 0)3953 die_errno("can't open patch '%s'", arg);3954 read_stdin = 0;3955 set_default_whitespace_mode(whitespace_option);3956 errs |= apply_patch(fd, arg, options);3957 close(fd);3958 }3959 set_default_whitespace_mode(whitespace_option);3960 if (read_stdin)3961 errs |= apply_patch(0, "<stdin>", options);3962 if (whitespace_error) {3963 if (squelch_whitespace_errors &&3964 squelch_whitespace_errors < whitespace_error) {3965 int squelched =3966 whitespace_error - squelch_whitespace_errors;3967 warning("squelched %d "3968 "whitespace error%s",3969 squelched,3970 squelched == 1 ? "" : "s");3971 }3972 if (ws_error_action == die_on_ws_error)3973 die("%d line%s add%s whitespace errors.",3974 whitespace_error,3975 whitespace_error == 1 ? "" : "s",3976 whitespace_error == 1 ? "s" : "");3977 if (applied_after_fixing_ws && apply)3978 warning("%d line%s applied after"3979 " fixing whitespace errors.",3980 applied_after_fixing_ws,3981 applied_after_fixing_ws == 1 ? "" : "s");3982 else if (whitespace_error)3983 warning("%d line%s add%s whitespace errors.",3984 whitespace_error,3985 whitespace_error == 1 ? "" : "s",3986 whitespace_error == 1 ? "s" : "");3987 }39883989 if (update_index) {3990 if (write_cache(newfd, active_cache, active_nr) ||3991 commit_locked_index(&lock_file))3992 die("Unable to write new index file");3993 }39943995 return !!errs;3996}