1/* 2 * apply.c 3 * 4 * Copyright (C) Linus Torvalds, 2005 5 * 6 * This applies patches on top of some (arbitrary) version of the SCM. 7 * 8 */ 9#include "cache.h" 10#include "lockfile.h" 11#include "cache-tree.h" 12#include "quote.h" 13#include "blob.h" 14#include "delta.h" 15#include "builtin.h" 16#include "string-list.h" 17#include "dir.h" 18#include "diff.h" 19#include "parse-options.h" 20#include "xdiff-interface.h" 21#include "ll-merge.h" 22#include "rerere.h" 23 24/* 25 * --check turns on checking that the working tree matches the 26 * files that are being modified, but doesn't apply the patch 27 * --stat does just a diffstat, and doesn't actually apply 28 * --numstat does numeric diffstat, and doesn't actually apply 29 * --index-info shows the old and new index info for paths if available. 30 * --index updates the cache as well. 31 * --cached updates only the cache without ever touching the working tree. 32 */ 33static const char *prefix; 34static int prefix_length = -1; 35static int newfd = -1; 36 37static int unidiff_zero; 38static int p_value = 1; 39static int p_value_known; 40static int check_index; 41static int update_index; 42static int cached; 43static int diffstat; 44static int numstat; 45static int summary; 46static int check; 47static int apply = 1; 48static int apply_in_reverse; 49static int apply_with_reject; 50static int apply_verbosely; 51static int allow_overlap; 52static int no_add; 53static int threeway; 54static int unsafe_paths; 55static const char *fake_ancestor; 56static int line_termination = '\n'; 57static unsigned int p_context = UINT_MAX; 58static const char * const apply_usage[] = { 59 N_("git apply [<options>] [<patch>...]"), 60 NULL 61}; 62 63static enum ws_error_action { 64 nowarn_ws_error, 65 warn_on_ws_error, 66 die_on_ws_error, 67 correct_ws_error 68} ws_error_action = warn_on_ws_error; 69static int whitespace_error; 70static int squelch_whitespace_errors = 5; 71static int applied_after_fixing_ws; 72 73static enum ws_ignore { 74 ignore_ws_none, 75 ignore_ws_change 76} ws_ignore_action = ignore_ws_none; 77 78 79static const char *patch_input_file; 80static struct strbuf root = STRBUF_INIT; 81static int read_stdin = 1; 82static int options; 83 84static void parse_whitespace_option(const char *option) 85{ 86 if (!option) { 87 ws_error_action = warn_on_ws_error; 88 return; 89 } 90 if (!strcmp(option, "warn")) { 91 ws_error_action = warn_on_ws_error; 92 return; 93 } 94 if (!strcmp(option, "nowarn")) { 95 ws_error_action = nowarn_ws_error; 96 return; 97 } 98 if (!strcmp(option, "error")) { 99 ws_error_action = die_on_ws_error; 100 return; 101 } 102 if (!strcmp(option, "error-all")) { 103 ws_error_action = die_on_ws_error; 104 squelch_whitespace_errors = 0; 105 return; 106 } 107 if (!strcmp(option, "strip") || !strcmp(option, "fix")) { 108 ws_error_action = correct_ws_error; 109 return; 110 } 111 die(_("unrecognized whitespace option '%s'"), option); 112} 113 114static void parse_ignorewhitespace_option(const char *option) 115{ 116 if (!option || !strcmp(option, "no") || 117 !strcmp(option, "false") || !strcmp(option, "never") || 118 !strcmp(option, "none")) { 119 ws_ignore_action = ignore_ws_none; 120 return; 121 } 122 if (!strcmp(option, "change")) { 123 ws_ignore_action = ignore_ws_change; 124 return; 125 } 126 die(_("unrecognized whitespace ignore option '%s'"), option); 127} 128 129static void set_default_whitespace_mode(const char *whitespace_option) 130{ 131 if (!whitespace_option && !apply_default_whitespace) 132 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error); 133} 134 135/* 136 * For "diff-stat" like behaviour, we keep track of the biggest change 137 * we've seen, and the longest filename. That allows us to do simple 138 * scaling. 139 */ 140static int max_change, max_len; 141 142/* 143 * Various "current state", notably line numbers and what 144 * file (and how) we're patching right now.. The "is_xxxx" 145 * things are flags, where -1 means "don't know yet". 146 */ 147static int linenr = 1; 148 149/* 150 * This represents one "hunk" from a patch, starting with 151 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The 152 * patch text is pointed at by patch, and its byte length 153 * is stored in size. leading and trailing are the number 154 * of context lines. 155 */ 156struct fragment { 157 unsigned long leading, trailing; 158 unsigned long oldpos, oldlines; 159 unsigned long newpos, newlines; 160 /* 161 * 'patch' is usually borrowed from buf in apply_patch(), 162 * but some codepaths store an allocated buffer. 163 */ 164 const char *patch; 165 unsigned free_patch:1, 166 rejected:1; 167 int size; 168 int linenr; 169 struct fragment *next; 170}; 171 172/* 173 * When dealing with a binary patch, we reuse "leading" field 174 * to store the type of the binary hunk, either deflated "delta" 175 * or deflated "literal". 176 */ 177#define binary_patch_method leading 178#define BINARY_DELTA_DEFLATED 1 179#define BINARY_LITERAL_DEFLATED 2 180 181/* 182 * This represents a "patch" to a file, both metainfo changes 183 * such as creation/deletion, filemode and content changes represented 184 * as a series of fragments. 185 */ 186struct patch { 187 char *new_name, *old_name, *def_name; 188 unsigned int old_mode, new_mode; 189 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */ 190 int rejected; 191 unsigned ws_rule; 192 int lines_added, lines_deleted; 193 int score; 194 unsigned int is_toplevel_relative:1; 195 unsigned int inaccurate_eof:1; 196 unsigned int is_binary:1; 197 unsigned int is_copy:1; 198 unsigned int is_rename:1; 199 unsigned int recount:1; 200 unsigned int conflicted_threeway:1; 201 unsigned int direct_to_threeway:1; 202 struct fragment *fragments; 203 char *result; 204 size_t resultsize; 205 char old_sha1_prefix[41]; 206 char new_sha1_prefix[41]; 207 struct patch *next; 208 209 /* three-way fallback result */ 210 struct object_id threeway_stage[3]; 211}; 212 213static void free_fragment_list(struct fragment *list) 214{ 215 while (list) { 216 struct fragment *next = list->next; 217 if (list->free_patch) 218 free((char *)list->patch); 219 free(list); 220 list = next; 221 } 222} 223 224static void free_patch(struct patch *patch) 225{ 226 free_fragment_list(patch->fragments); 227 free(patch->def_name); 228 free(patch->old_name); 229 free(patch->new_name); 230 free(patch->result); 231 free(patch); 232} 233 234static void free_patch_list(struct patch *list) 235{ 236 while (list) { 237 struct patch *next = list->next; 238 free_patch(list); 239 list = next; 240 } 241} 242 243/* 244 * A line in a file, len-bytes long (includes the terminating LF, 245 * except for an incomplete line at the end if the file ends with 246 * one), and its contents hashes to 'hash'. 247 */ 248struct line { 249 size_t len; 250 unsigned hash : 24; 251 unsigned flag : 8; 252#define LINE_COMMON 1 253#define LINE_PATCHED 2 254}; 255 256/* 257 * This represents a "file", which is an array of "lines". 258 */ 259struct image { 260 char *buf; 261 size_t len; 262 size_t nr; 263 size_t alloc; 264 struct line *line_allocated; 265 struct line *line; 266}; 267 268/* 269 * Records filenames that have been touched, in order to handle 270 * the case where more than one patches touch the same file. 271 */ 272 273static struct string_list fn_table; 274 275static uint32_t hash_line(const char *cp, size_t len) 276{ 277 size_t i; 278 uint32_t h; 279 for (i = 0, h = 0; i < len; i++) { 280 if (!isspace(cp[i])) { 281 h = h * 3 + (cp[i] & 0xff); 282 } 283 } 284 return h; 285} 286 287/* 288 * Compare lines s1 of length n1 and s2 of length n2, ignoring 289 * whitespace difference. Returns 1 if they match, 0 otherwise 290 */ 291static int fuzzy_matchlines(const char *s1, size_t n1, 292 const char *s2, size_t n2) 293{ 294 const char *last1 = s1 + n1 - 1; 295 const char *last2 = s2 + n2 - 1; 296 int result = 0; 297 298 /* ignore line endings */ 299 while ((*last1 == '\r') || (*last1 == '\n')) 300 last1--; 301 while ((*last2 == '\r') || (*last2 == '\n')) 302 last2--; 303 304 /* skip leading whitespaces, if both begin with whitespace */ 305 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) { 306 while (isspace(*s1) && (s1 <= last1)) 307 s1++; 308 while (isspace(*s2) && (s2 <= last2)) 309 s2++; 310 } 311 /* early return if both lines are empty */ 312 if ((s1 > last1) && (s2 > last2)) 313 return 1; 314 while (!result) { 315 result = *s1++ - *s2++; 316 /* 317 * Skip whitespace inside. We check for whitespace on 318 * both buffers because we don't want "a b" to match 319 * "ab" 320 */ 321 if (isspace(*s1) && isspace(*s2)) { 322 while (isspace(*s1) && s1 <= last1) 323 s1++; 324 while (isspace(*s2) && s2 <= last2) 325 s2++; 326 } 327 /* 328 * If we reached the end on one side only, 329 * lines don't match 330 */ 331 if ( 332 ((s2 > last2) && (s1 <= last1)) || 333 ((s1 > last1) && (s2 <= last2))) 334 return 0; 335 if ((s1 > last1) && (s2 > last2)) 336 break; 337 } 338 339 return !result; 340} 341 342static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag) 343{ 344 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc); 345 img->line_allocated[img->nr].len = len; 346 img->line_allocated[img->nr].hash = hash_line(bol, len); 347 img->line_allocated[img->nr].flag = flag; 348 img->nr++; 349} 350 351/* 352 * "buf" has the file contents to be patched (read from various sources). 353 * attach it to "image" and add line-based index to it. 354 * "image" now owns the "buf". 355 */ 356static void prepare_image(struct image *image, char *buf, size_t len, 357 int prepare_linetable) 358{ 359 const char *cp, *ep; 360 361 memset(image, 0, sizeof(*image)); 362 image->buf = buf; 363 image->len = len; 364 365 if (!prepare_linetable) 366 return; 367 368 ep = image->buf + image->len; 369 cp = image->buf; 370 while (cp < ep) { 371 const char *next; 372 for (next = cp; next < ep && *next != '\n'; next++) 373 ; 374 if (next < ep) 375 next++; 376 add_line_info(image, cp, next - cp, 0); 377 cp = next; 378 } 379 image->line = image->line_allocated; 380} 381 382static void clear_image(struct image *image) 383{ 384 free(image->buf); 385 free(image->line_allocated); 386 memset(image, 0, sizeof(*image)); 387} 388 389/* fmt must contain _one_ %s and no other substitution */ 390static void say_patch_name(FILE *output, const char *fmt, struct patch *patch) 391{ 392 struct strbuf sb = STRBUF_INIT; 393 394 if (patch->old_name && patch->new_name && 395 strcmp(patch->old_name, patch->new_name)) { 396 quote_c_style(patch->old_name, &sb, NULL, 0); 397 strbuf_addstr(&sb, " => "); 398 quote_c_style(patch->new_name, &sb, NULL, 0); 399 } else { 400 const char *n = patch->new_name; 401 if (!n) 402 n = patch->old_name; 403 quote_c_style(n, &sb, NULL, 0); 404 } 405 fprintf(output, fmt, sb.buf); 406 fputc('\n', output); 407 strbuf_release(&sb); 408} 409 410#define SLOP (16) 411 412static void read_patch_file(struct strbuf *sb, int fd) 413{ 414 if (strbuf_read(sb, fd, 0) < 0) 415 die_errno("git apply: failed to read"); 416 417 /* 418 * Make sure that we have some slop in the buffer 419 * so that we can do speculative "memcmp" etc, and 420 * see to it that it is NUL-filled. 421 */ 422 strbuf_grow(sb, SLOP); 423 memset(sb->buf + sb->len, 0, SLOP); 424} 425 426static unsigned long linelen(const char *buffer, unsigned long size) 427{ 428 unsigned long len = 0; 429 while (size--) { 430 len++; 431 if (*buffer++ == '\n') 432 break; 433 } 434 return len; 435} 436 437static int is_dev_null(const char *str) 438{ 439 return skip_prefix(str, "/dev/null", &str) && isspace(*str); 440} 441 442#define TERM_SPACE 1 443#define TERM_TAB 2 444 445static int name_terminate(const char *name, int namelen, int c, int terminate) 446{ 447 if (c == ' ' && !(terminate & TERM_SPACE)) 448 return 0; 449 if (c == '\t' && !(terminate & TERM_TAB)) 450 return 0; 451 452 return 1; 453} 454 455/* remove double slashes to make --index work with such filenames */ 456static char *squash_slash(char *name) 457{ 458 int i = 0, j = 0; 459 460 if (!name) 461 return NULL; 462 463 while (name[i]) { 464 if ((name[j++] = name[i++]) == '/') 465 while (name[i] == '/') 466 i++; 467 } 468 name[j] = '\0'; 469 return name; 470} 471 472static char *find_name_gnu(const char *line, const char *def, int p_value) 473{ 474 struct strbuf name = STRBUF_INIT; 475 char *cp; 476 477 /* 478 * Proposed "new-style" GNU patch/diff format; see 479 * http://marc.info/?l=git&m=112927316408690&w=2 480 */ 481 if (unquote_c_style(&name, line, NULL)) { 482 strbuf_release(&name); 483 return NULL; 484 } 485 486 for (cp = name.buf; p_value; p_value--) { 487 cp = strchr(cp, '/'); 488 if (!cp) { 489 strbuf_release(&name); 490 return NULL; 491 } 492 cp++; 493 } 494 495 strbuf_remove(&name, 0, cp - name.buf); 496 if (root.len) 497 strbuf_insert(&name, 0, root.buf, root.len); 498 return squash_slash(strbuf_detach(&name, NULL)); 499} 500 501static size_t sane_tz_len(const char *line, size_t len) 502{ 503 const char *tz, *p; 504 505 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ') 506 return 0; 507 tz = line + len - strlen(" +0500"); 508 509 if (tz[1] != '+' && tz[1] != '-') 510 return 0; 511 512 for (p = tz + 2; p != line + len; p++) 513 if (!isdigit(*p)) 514 return 0; 515 516 return line + len - tz; 517} 518 519static size_t tz_with_colon_len(const char *line, size_t len) 520{ 521 const char *tz, *p; 522 523 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':') 524 return 0; 525 tz = line + len - strlen(" +08:00"); 526 527 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-')) 528 return 0; 529 p = tz + 2; 530 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 531 !isdigit(*p++) || !isdigit(*p++)) 532 return 0; 533 534 return line + len - tz; 535} 536 537static size_t date_len(const char *line, size_t len) 538{ 539 const char *date, *p; 540 541 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-') 542 return 0; 543 p = date = line + len - strlen("72-02-05"); 544 545 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 546 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 547 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */ 548 return 0; 549 550 if (date - line >= strlen("19") && 551 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */ 552 date -= strlen("19"); 553 554 return line + len - date; 555} 556 557static size_t short_time_len(const char *line, size_t len) 558{ 559 const char *time, *p; 560 561 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':') 562 return 0; 563 p = time = line + len - strlen(" 07:01:32"); 564 565 /* Permit 1-digit hours? */ 566 if (*p++ != ' ' || 567 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 568 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 569 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */ 570 return 0; 571 572 return line + len - time; 573} 574 575static size_t fractional_time_len(const char *line, size_t len) 576{ 577 const char *p; 578 size_t n; 579 580 /* Expected format: 19:41:17.620000023 */ 581 if (!len || !isdigit(line[len - 1])) 582 return 0; 583 p = line + len - 1; 584 585 /* Fractional seconds. */ 586 while (p > line && isdigit(*p)) 587 p--; 588 if (*p != '.') 589 return 0; 590 591 /* Hours, minutes, and whole seconds. */ 592 n = short_time_len(line, p - line); 593 if (!n) 594 return 0; 595 596 return line + len - p + n; 597} 598 599static size_t trailing_spaces_len(const char *line, size_t len) 600{ 601 const char *p; 602 603 /* Expected format: ' ' x (1 or more) */ 604 if (!len || line[len - 1] != ' ') 605 return 0; 606 607 p = line + len; 608 while (p != line) { 609 p--; 610 if (*p != ' ') 611 return line + len - (p + 1); 612 } 613 614 /* All spaces! */ 615 return len; 616} 617 618static size_t diff_timestamp_len(const char *line, size_t len) 619{ 620 const char *end = line + len; 621 size_t n; 622 623 /* 624 * Posix: 2010-07-05 19:41:17 625 * GNU: 2010-07-05 19:41:17.620000023 -0500 626 */ 627 628 if (!isdigit(end[-1])) 629 return 0; 630 631 n = sane_tz_len(line, end - line); 632 if (!n) 633 n = tz_with_colon_len(line, end - line); 634 end -= n; 635 636 n = short_time_len(line, end - line); 637 if (!n) 638 n = fractional_time_len(line, end - line); 639 end -= n; 640 641 n = date_len(line, end - line); 642 if (!n) /* No date. Too bad. */ 643 return 0; 644 end -= n; 645 646 if (end == line) /* No space before date. */ 647 return 0; 648 if (end[-1] == '\t') { /* Success! */ 649 end--; 650 return line + len - end; 651 } 652 if (end[-1] != ' ') /* No space before date. */ 653 return 0; 654 655 /* Whitespace damage. */ 656 end -= trailing_spaces_len(line, end - line); 657 return line + len - end; 658} 659 660static char *find_name_common(const char *line, const char *def, 661 int p_value, const char *end, int terminate) 662{ 663 int len; 664 const char *start = NULL; 665 666 if (p_value == 0) 667 start = line; 668 while (line != end) { 669 char c = *line; 670 671 if (!end && isspace(c)) { 672 if (c == '\n') 673 break; 674 if (name_terminate(start, line-start, c, terminate)) 675 break; 676 } 677 line++; 678 if (c == '/' && !--p_value) 679 start = line; 680 } 681 if (!start) 682 return squash_slash(xstrdup_or_null(def)); 683 len = line - start; 684 if (!len) 685 return squash_slash(xstrdup_or_null(def)); 686 687 /* 688 * Generally we prefer the shorter name, especially 689 * if the other one is just a variation of that with 690 * something else tacked on to the end (ie "file.orig" 691 * or "file~"). 692 */ 693 if (def) { 694 int deflen = strlen(def); 695 if (deflen < len && !strncmp(start, def, deflen)) 696 return squash_slash(xstrdup(def)); 697 } 698 699 if (root.len) { 700 char *ret = xstrfmt("%s%.*s", root.buf, len, start); 701 return squash_slash(ret); 702 } 703 704 return squash_slash(xmemdupz(start, len)); 705} 706 707static char *find_name(const char *line, char *def, int p_value, int terminate) 708{ 709 if (*line == '"') { 710 char *name = find_name_gnu(line, def, p_value); 711 if (name) 712 return name; 713 } 714 715 return find_name_common(line, def, p_value, NULL, terminate); 716} 717 718static char *find_name_traditional(const char *line, char *def, int p_value) 719{ 720 size_t len; 721 size_t date_len; 722 723 if (*line == '"') { 724 char *name = find_name_gnu(line, def, p_value); 725 if (name) 726 return name; 727 } 728 729 len = strchrnul(line, '\n') - line; 730 date_len = diff_timestamp_len(line, len); 731 if (!date_len) 732 return find_name_common(line, def, p_value, NULL, TERM_TAB); 733 len -= date_len; 734 735 return find_name_common(line, def, p_value, line + len, 0); 736} 737 738static int count_slashes(const char *cp) 739{ 740 int cnt = 0; 741 char ch; 742 743 while ((ch = *cp++)) 744 if (ch == '/') 745 cnt++; 746 return cnt; 747} 748 749/* 750 * Given the string after "--- " or "+++ ", guess the appropriate 751 * p_value for the given patch. 752 */ 753static int guess_p_value(const char *nameline) 754{ 755 char *name, *cp; 756 int val = -1; 757 758 if (is_dev_null(nameline)) 759 return -1; 760 name = find_name_traditional(nameline, NULL, 0); 761 if (!name) 762 return -1; 763 cp = strchr(name, '/'); 764 if (!cp) 765 val = 0; 766 else if (prefix) { 767 /* 768 * Does it begin with "a/$our-prefix" and such? Then this is 769 * very likely to apply to our directory. 770 */ 771 if (!strncmp(name, prefix, prefix_length)) 772 val = count_slashes(prefix); 773 else { 774 cp++; 775 if (!strncmp(cp, prefix, prefix_length)) 776 val = count_slashes(prefix) + 1; 777 } 778 } 779 free(name); 780 return val; 781} 782 783/* 784 * Does the ---/+++ line have the POSIX timestamp after the last HT? 785 * GNU diff puts epoch there to signal a creation/deletion event. Is 786 * this such a timestamp? 787 */ 788static int has_epoch_timestamp(const char *nameline) 789{ 790 /* 791 * We are only interested in epoch timestamp; any non-zero 792 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 793 * For the same reason, the date must be either 1969-12-31 or 794 * 1970-01-01, and the seconds part must be "00". 795 */ 796 const char stamp_regexp[] = 797 "^(1969-12-31|1970-01-01)" 798 " " 799 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 800 " " 801 "([-+][0-2][0-9]:?[0-5][0-9])\n"; 802 const char *timestamp = NULL, *cp, *colon; 803 static regex_t *stamp; 804 regmatch_t m[10]; 805 int zoneoffset; 806 int hourminute; 807 int status; 808 809 for (cp = nameline; *cp != '\n'; cp++) { 810 if (*cp == '\t') 811 timestamp = cp + 1; 812 } 813 if (!timestamp) 814 return 0; 815 if (!stamp) { 816 stamp = xmalloc(sizeof(*stamp)); 817 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 818 warning(_("Cannot prepare timestamp regexp %s"), 819 stamp_regexp); 820 return 0; 821 } 822 } 823 824 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 825 if (status) { 826 if (status != REG_NOMATCH) 827 warning(_("regexec returned %d for input: %s"), 828 status, timestamp); 829 return 0; 830 } 831 832 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10); 833 if (*colon == ':') 834 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10); 835 else 836 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 837 if (timestamp[m[3].rm_so] == '-') 838 zoneoffset = -zoneoffset; 839 840 /* 841 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 842 * (west of GMT) or 1970-01-01 (east of GMT) 843 */ 844 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 845 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 846 return 0; 847 848 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 849 strtol(timestamp + 14, NULL, 10) - 850 zoneoffset); 851 852 return ((zoneoffset < 0 && hourminute == 1440) || 853 (0 <= zoneoffset && !hourminute)); 854} 855 856/* 857 * Get the name etc info from the ---/+++ lines of a traditional patch header 858 * 859 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 860 * files, we can happily check the index for a match, but for creating a 861 * new file we should try to match whatever "patch" does. I have no idea. 862 */ 863static void parse_traditional_patch(const char *first, const char *second, struct patch *patch) 864{ 865 char *name; 866 867 first += 4; /* skip "--- " */ 868 second += 4; /* skip "+++ " */ 869 if (!p_value_known) { 870 int p, q; 871 p = guess_p_value(first); 872 q = guess_p_value(second); 873 if (p < 0) p = q; 874 if (0 <= p && p == q) { 875 p_value = p; 876 p_value_known = 1; 877 } 878 } 879 if (is_dev_null(first)) { 880 patch->is_new = 1; 881 patch->is_delete = 0; 882 name = find_name_traditional(second, NULL, p_value); 883 patch->new_name = name; 884 } else if (is_dev_null(second)) { 885 patch->is_new = 0; 886 patch->is_delete = 1; 887 name = find_name_traditional(first, NULL, p_value); 888 patch->old_name = name; 889 } else { 890 char *first_name; 891 first_name = find_name_traditional(first, NULL, p_value); 892 name = find_name_traditional(second, first_name, p_value); 893 free(first_name); 894 if (has_epoch_timestamp(first)) { 895 patch->is_new = 1; 896 patch->is_delete = 0; 897 patch->new_name = name; 898 } else if (has_epoch_timestamp(second)) { 899 patch->is_new = 0; 900 patch->is_delete = 1; 901 patch->old_name = name; 902 } else { 903 patch->old_name = name; 904 patch->new_name = xstrdup_or_null(name); 905 } 906 } 907 if (!name) 908 die(_("unable to find filename in patch at line %d"), linenr); 909} 910 911static int gitdiff_hdrend(const char *line, struct patch *patch) 912{ 913 return -1; 914} 915 916/* 917 * We're anal about diff header consistency, to make 918 * sure that we don't end up having strange ambiguous 919 * patches floating around. 920 * 921 * As a result, gitdiff_{old|new}name() will check 922 * their names against any previous information, just 923 * to make sure.. 924 */ 925#define DIFF_OLD_NAME 0 926#define DIFF_NEW_NAME 1 927 928static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side) 929{ 930 if (!orig_name && !isnull) 931 return find_name(line, NULL, p_value, TERM_TAB); 932 933 if (orig_name) { 934 int len = strlen(orig_name); 935 char *another; 936 if (isnull) 937 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), 938 orig_name, linenr); 939 another = find_name(line, NULL, p_value, TERM_TAB); 940 if (!another || memcmp(another, orig_name, len + 1)) 941 die((side == DIFF_NEW_NAME) ? 942 _("git apply: bad git-diff - inconsistent new filename on line %d") : 943 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr); 944 free(another); 945 return orig_name; 946 } else { 947 /* expect "/dev/null" */ 948 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 949 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr); 950 return NULL; 951 } 952} 953 954static int gitdiff_oldname(const char *line, struct patch *patch) 955{ 956 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, 957 DIFF_OLD_NAME); 958 return 0; 959} 960 961static int gitdiff_newname(const char *line, struct patch *patch) 962{ 963 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, 964 DIFF_NEW_NAME); 965 return 0; 966} 967 968static int gitdiff_oldmode(const char *line, struct patch *patch) 969{ 970 patch->old_mode = strtoul(line, NULL, 8); 971 return 0; 972} 973 974static int gitdiff_newmode(const char *line, struct patch *patch) 975{ 976 patch->new_mode = strtoul(line, NULL, 8); 977 return 0; 978} 979 980static int gitdiff_delete(const char *line, struct patch *patch) 981{ 982 patch->is_delete = 1; 983 free(patch->old_name); 984 patch->old_name = xstrdup_or_null(patch->def_name); 985 return gitdiff_oldmode(line, patch); 986} 987 988static int gitdiff_newfile(const char *line, struct patch *patch) 989{ 990 patch->is_new = 1; 991 free(patch->new_name); 992 patch->new_name = xstrdup_or_null(patch->def_name); 993 return gitdiff_newmode(line, patch); 994} 995 996static int gitdiff_copysrc(const char *line, struct patch *patch) 997{ 998 patch->is_copy = 1; 999 free(patch->old_name);1000 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1001 return 0;1002}10031004static int gitdiff_copydst(const char *line, struct patch *patch)1005{1006 patch->is_copy = 1;1007 free(patch->new_name);1008 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1009 return 0;1010}10111012static int gitdiff_renamesrc(const char *line, struct patch *patch)1013{1014 patch->is_rename = 1;1015 free(patch->old_name);1016 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1017 return 0;1018}10191020static int gitdiff_renamedst(const char *line, struct patch *patch)1021{1022 patch->is_rename = 1;1023 free(patch->new_name);1024 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1025 return 0;1026}10271028static int gitdiff_similarity(const char *line, struct patch *patch)1029{1030 unsigned long val = strtoul(line, NULL, 10);1031 if (val <= 100)1032 patch->score = val;1033 return 0;1034}10351036static int gitdiff_dissimilarity(const char *line, struct patch *patch)1037{1038 unsigned long val = strtoul(line, NULL, 10);1039 if (val <= 100)1040 patch->score = val;1041 return 0;1042}10431044static int gitdiff_index(const char *line, struct patch *patch)1045{1046 /*1047 * index line is N hexadecimal, "..", N hexadecimal,1048 * and optional space with octal mode.1049 */1050 const char *ptr, *eol;1051 int len;10521053 ptr = strchr(line, '.');1054 if (!ptr || ptr[1] != '.' || 40 < ptr - line)1055 return 0;1056 len = ptr - line;1057 memcpy(patch->old_sha1_prefix, line, len);1058 patch->old_sha1_prefix[len] = 0;10591060 line = ptr + 2;1061 ptr = strchr(line, ' ');1062 eol = strchrnul(line, '\n');10631064 if (!ptr || eol < ptr)1065 ptr = eol;1066 len = ptr - line;10671068 if (40 < len)1069 return 0;1070 memcpy(patch->new_sha1_prefix, line, len);1071 patch->new_sha1_prefix[len] = 0;1072 if (*ptr == ' ')1073 patch->old_mode = strtoul(ptr+1, NULL, 8);1074 return 0;1075}10761077/*1078 * This is normal for a diff that doesn't change anything: we'll fall through1079 * into the next diff. Tell the parser to break out.1080 */1081static int gitdiff_unrecognized(const char *line, struct patch *patch)1082{1083 return -1;1084}10851086/*1087 * Skip p_value leading components from "line"; as we do not accept1088 * absolute paths, return NULL in that case.1089 */1090static const char *skip_tree_prefix(const char *line, int llen)1091{1092 int nslash;1093 int i;10941095 if (!p_value)1096 return (llen && line[0] == '/') ? NULL : line;10971098 nslash = p_value;1099 for (i = 0; i < llen; i++) {1100 int ch = line[i];1101 if (ch == '/' && --nslash <= 0)1102 return (i == 0) ? NULL : &line[i + 1];1103 }1104 return NULL;1105}11061107/*1108 * This is to extract the same name that appears on "diff --git"1109 * line. We do not find and return anything if it is a rename1110 * patch, and it is OK because we will find the name elsewhere.1111 * We need to reliably find name only when it is mode-change only,1112 * creation or deletion of an empty file. In any of these cases,1113 * both sides are the same name under a/ and b/ respectively.1114 */1115static char *git_header_name(const char *line, int llen)1116{1117 const char *name;1118 const char *second = NULL;1119 size_t len, line_len;11201121 line += strlen("diff --git ");1122 llen -= strlen("diff --git ");11231124 if (*line == '"') {1125 const char *cp;1126 struct strbuf first = STRBUF_INIT;1127 struct strbuf sp = STRBUF_INIT;11281129 if (unquote_c_style(&first, line, &second))1130 goto free_and_fail1;11311132 /* strip the a/b prefix including trailing slash */1133 cp = skip_tree_prefix(first.buf, first.len);1134 if (!cp)1135 goto free_and_fail1;1136 strbuf_remove(&first, 0, cp - first.buf);11371138 /*1139 * second points at one past closing dq of name.1140 * find the second name.1141 */1142 while ((second < line + llen) && isspace(*second))1143 second++;11441145 if (line + llen <= second)1146 goto free_and_fail1;1147 if (*second == '"') {1148 if (unquote_c_style(&sp, second, NULL))1149 goto free_and_fail1;1150 cp = skip_tree_prefix(sp.buf, sp.len);1151 if (!cp)1152 goto free_and_fail1;1153 /* They must match, otherwise ignore */1154 if (strcmp(cp, first.buf))1155 goto free_and_fail1;1156 strbuf_release(&sp);1157 return strbuf_detach(&first, NULL);1158 }11591160 /* unquoted second */1161 cp = skip_tree_prefix(second, line + llen - second);1162 if (!cp)1163 goto free_and_fail1;1164 if (line + llen - cp != first.len ||1165 memcmp(first.buf, cp, first.len))1166 goto free_and_fail1;1167 return strbuf_detach(&first, NULL);11681169 free_and_fail1:1170 strbuf_release(&first);1171 strbuf_release(&sp);1172 return NULL;1173 }11741175 /* unquoted first name */1176 name = skip_tree_prefix(line, llen);1177 if (!name)1178 return NULL;11791180 /*1181 * since the first name is unquoted, a dq if exists must be1182 * the beginning of the second name.1183 */1184 for (second = name; second < line + llen; second++) {1185 if (*second == '"') {1186 struct strbuf sp = STRBUF_INIT;1187 const char *np;11881189 if (unquote_c_style(&sp, second, NULL))1190 goto free_and_fail2;11911192 np = skip_tree_prefix(sp.buf, sp.len);1193 if (!np)1194 goto free_and_fail2;11951196 len = sp.buf + sp.len - np;1197 if (len < second - name &&1198 !strncmp(np, name, len) &&1199 isspace(name[len])) {1200 /* Good */1201 strbuf_remove(&sp, 0, np - sp.buf);1202 return strbuf_detach(&sp, NULL);1203 }12041205 free_and_fail2:1206 strbuf_release(&sp);1207 return NULL;1208 }1209 }12101211 /*1212 * Accept a name only if it shows up twice, exactly the same1213 * form.1214 */1215 second = strchr(name, '\n');1216 if (!second)1217 return NULL;1218 line_len = second - name;1219 for (len = 0 ; ; len++) {1220 switch (name[len]) {1221 default:1222 continue;1223 case '\n':1224 return NULL;1225 case '\t': case ' ':1226 /*1227 * Is this the separator between the preimage1228 * and the postimage pathname? Again, we are1229 * only interested in the case where there is1230 * no rename, as this is only to set def_name1231 * and a rename patch has the names elsewhere1232 * in an unambiguous form.1233 */1234 if (!name[len + 1])1235 return NULL; /* no postimage name */1236 second = skip_tree_prefix(name + len + 1,1237 line_len - (len + 1));1238 if (!second)1239 return NULL;1240 /*1241 * Does len bytes starting at "name" and "second"1242 * (that are separated by one HT or SP we just1243 * found) exactly match?1244 */1245 if (second[len] == '\n' && !strncmp(name, second, len))1246 return xmemdupz(name, len);1247 }1248 }1249}12501251/* Verify that we recognize the lines following a git header */1252static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)1253{1254 unsigned long offset;12551256 /* A git diff has explicit new/delete information, so we don't guess */1257 patch->is_new = 0;1258 patch->is_delete = 0;12591260 /*1261 * Some things may not have the old name in the1262 * rest of the headers anywhere (pure mode changes,1263 * or removing or adding empty files), so we get1264 * the default name from the header.1265 */1266 patch->def_name = git_header_name(line, len);1267 if (patch->def_name && root.len) {1268 char *s = xstrfmt("%s%s", root.buf, patch->def_name);1269 free(patch->def_name);1270 patch->def_name = s;1271 }12721273 line += len;1274 size -= len;1275 linenr++;1276 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {1277 static const struct opentry {1278 const char *str;1279 int (*fn)(const char *, struct patch *);1280 } optable[] = {1281 { "@@ -", gitdiff_hdrend },1282 { "--- ", gitdiff_oldname },1283 { "+++ ", gitdiff_newname },1284 { "old mode ", gitdiff_oldmode },1285 { "new mode ", gitdiff_newmode },1286 { "deleted file mode ", gitdiff_delete },1287 { "new file mode ", gitdiff_newfile },1288 { "copy from ", gitdiff_copysrc },1289 { "copy to ", gitdiff_copydst },1290 { "rename old ", gitdiff_renamesrc },1291 { "rename new ", gitdiff_renamedst },1292 { "rename from ", gitdiff_renamesrc },1293 { "rename to ", gitdiff_renamedst },1294 { "similarity index ", gitdiff_similarity },1295 { "dissimilarity index ", gitdiff_dissimilarity },1296 { "index ", gitdiff_index },1297 { "", gitdiff_unrecognized },1298 };1299 int i;13001301 len = linelen(line, size);1302 if (!len || line[len-1] != '\n')1303 break;1304 for (i = 0; i < ARRAY_SIZE(optable); i++) {1305 const struct opentry *p = optable + i;1306 int oplen = strlen(p->str);1307 if (len < oplen || memcmp(p->str, line, oplen))1308 continue;1309 if (p->fn(line + oplen, patch) < 0)1310 return offset;1311 break;1312 }1313 }13141315 return offset;1316}13171318static int parse_num(const char *line, unsigned long *p)1319{1320 char *ptr;13211322 if (!isdigit(*line))1323 return 0;1324 *p = strtoul(line, &ptr, 10);1325 return ptr - line;1326}13271328static int parse_range(const char *line, int len, int offset, const char *expect,1329 unsigned long *p1, unsigned long *p2)1330{1331 int digits, ex;13321333 if (offset < 0 || offset >= len)1334 return -1;1335 line += offset;1336 len -= offset;13371338 digits = parse_num(line, p1);1339 if (!digits)1340 return -1;13411342 offset += digits;1343 line += digits;1344 len -= digits;13451346 *p2 = 1;1347 if (*line == ',') {1348 digits = parse_num(line+1, p2);1349 if (!digits)1350 return -1;13511352 offset += digits+1;1353 line += digits+1;1354 len -= digits+1;1355 }13561357 ex = strlen(expect);1358 if (ex > len)1359 return -1;1360 if (memcmp(line, expect, ex))1361 return -1;13621363 return offset + ex;1364}13651366static void recount_diff(const char *line, int size, struct fragment *fragment)1367{1368 int oldlines = 0, newlines = 0, ret = 0;13691370 if (size < 1) {1371 warning("recount: ignore empty hunk");1372 return;1373 }13741375 for (;;) {1376 int len = linelen(line, size);1377 size -= len;1378 line += len;13791380 if (size < 1)1381 break;13821383 switch (*line) {1384 case ' ': case '\n':1385 newlines++;1386 /* fall through */1387 case '-':1388 oldlines++;1389 continue;1390 case '+':1391 newlines++;1392 continue;1393 case '\\':1394 continue;1395 case '@':1396 ret = size < 3 || !starts_with(line, "@@ ");1397 break;1398 case 'd':1399 ret = size < 5 || !starts_with(line, "diff ");1400 break;1401 default:1402 ret = -1;1403 break;1404 }1405 if (ret) {1406 warning(_("recount: unexpected line: %.*s"),1407 (int)linelen(line, size), line);1408 return;1409 }1410 break;1411 }1412 fragment->oldlines = oldlines;1413 fragment->newlines = newlines;1414}14151416/*1417 * Parse a unified diff fragment header of the1418 * form "@@ -a,b +c,d @@"1419 */1420static int parse_fragment_header(const char *line, int len, struct fragment *fragment)1421{1422 int offset;14231424 if (!len || line[len-1] != '\n')1425 return -1;14261427 /* Figure out the number of lines in a fragment */1428 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1429 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);14301431 return offset;1432}14331434static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)1435{1436 unsigned long offset, len;14371438 patch->is_toplevel_relative = 0;1439 patch->is_rename = patch->is_copy = 0;1440 patch->is_new = patch->is_delete = -1;1441 patch->old_mode = patch->new_mode = 0;1442 patch->old_name = patch->new_name = NULL;1443 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {1444 unsigned long nextlen;14451446 len = linelen(line, size);1447 if (!len)1448 break;14491450 /* Testing this early allows us to take a few shortcuts.. */1451 if (len < 6)1452 continue;14531454 /*1455 * Make sure we don't find any unconnected patch fragments.1456 * That's a sign that we didn't find a header, and that a1457 * patch has become corrupted/broken up.1458 */1459 if (!memcmp("@@ -", line, 4)) {1460 struct fragment dummy;1461 if (parse_fragment_header(line, len, &dummy) < 0)1462 continue;1463 die(_("patch fragment without header at line %d: %.*s"),1464 linenr, (int)len-1, line);1465 }14661467 if (size < len + 6)1468 break;14691470 /*1471 * Git patch? It might not have a real patch, just a rename1472 * or mode change, so we handle that specially1473 */1474 if (!memcmp("diff --git ", line, 11)) {1475 int git_hdr_len = parse_git_header(line, len, size, patch);1476 if (git_hdr_len <= len)1477 continue;1478 if (!patch->old_name && !patch->new_name) {1479 if (!patch->def_name)1480 die(Q_("git diff header lacks filename information when removing "1481 "%d leading pathname component (line %d)",1482 "git diff header lacks filename information when removing "1483 "%d leading pathname components (line %d)",1484 p_value),1485 p_value, linenr);1486 patch->old_name = xstrdup(patch->def_name);1487 patch->new_name = xstrdup(patch->def_name);1488 }1489 if (!patch->is_delete && !patch->new_name)1490 die("git diff header lacks filename information "1491 "(line %d)", linenr);1492 patch->is_toplevel_relative = 1;1493 *hdrsize = git_hdr_len;1494 return offset;1495 }14961497 /* --- followed by +++ ? */1498 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1499 continue;15001501 /*1502 * We only accept unified patches, so we want it to1503 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1504 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1505 */1506 nextlen = linelen(line + len, size - len);1507 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1508 continue;15091510 /* Ok, we'll consider it a patch */1511 parse_traditional_patch(line, line+len, patch);1512 *hdrsize = len + nextlen;1513 linenr += 2;1514 return offset;1515 }1516 return -1;1517}15181519static void record_ws_error(unsigned result, const char *line, int len, int linenr)1520{1521 char *err;15221523 if (!result)1524 return;15251526 whitespace_error++;1527 if (squelch_whitespace_errors &&1528 squelch_whitespace_errors < whitespace_error)1529 return;15301531 err = whitespace_error_string(result);1532 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1533 patch_input_file, linenr, err, len, line);1534 free(err);1535}15361537static void check_whitespace(const char *line, int len, unsigned ws_rule)1538{1539 unsigned result = ws_check(line + 1, len - 1, ws_rule);15401541 record_ws_error(result, line + 1, len - 2, linenr);1542}15431544/*1545 * Parse a unified diff. Note that this really needs to parse each1546 * fragment separately, since the only way to know the difference1547 * between a "---" that is part of a patch, and a "---" that starts1548 * the next patch is to look at the line counts..1549 */1550static int parse_fragment(const char *line, unsigned long size,1551 struct patch *patch, struct fragment *fragment)1552{1553 int added, deleted;1554 int len = linelen(line, size), offset;1555 unsigned long oldlines, newlines;1556 unsigned long leading, trailing;15571558 offset = parse_fragment_header(line, len, fragment);1559 if (offset < 0)1560 return -1;1561 if (offset > 0 && patch->recount)1562 recount_diff(line + offset, size - offset, fragment);1563 oldlines = fragment->oldlines;1564 newlines = fragment->newlines;1565 leading = 0;1566 trailing = 0;15671568 /* Parse the thing.. */1569 line += len;1570 size -= len;1571 linenr++;1572 added = deleted = 0;1573 for (offset = len;1574 0 < size;1575 offset += len, size -= len, line += len, linenr++) {1576 if (!oldlines && !newlines)1577 break;1578 len = linelen(line, size);1579 if (!len || line[len-1] != '\n')1580 return -1;1581 switch (*line) {1582 default:1583 return -1;1584 case '\n': /* newer GNU diff, an empty context line */1585 case ' ':1586 oldlines--;1587 newlines--;1588 if (!deleted && !added)1589 leading++;1590 trailing++;1591 if (!apply_in_reverse &&1592 ws_error_action == correct_ws_error)1593 check_whitespace(line, len, patch->ws_rule);1594 break;1595 case '-':1596 if (apply_in_reverse &&1597 ws_error_action != nowarn_ws_error)1598 check_whitespace(line, len, patch->ws_rule);1599 deleted++;1600 oldlines--;1601 trailing = 0;1602 break;1603 case '+':1604 if (!apply_in_reverse &&1605 ws_error_action != nowarn_ws_error)1606 check_whitespace(line, len, patch->ws_rule);1607 added++;1608 newlines--;1609 trailing = 0;1610 break;16111612 /*1613 * We allow "\ No newline at end of file". Depending1614 * on locale settings when the patch was produced we1615 * don't know what this line looks like. The only1616 * thing we do know is that it begins with "\ ".1617 * Checking for 12 is just for sanity check -- any1618 * l10n of "\ No newline..." is at least that long.1619 */1620 case '\\':1621 if (len < 12 || memcmp(line, "\\ ", 2))1622 return -1;1623 break;1624 }1625 }1626 if (oldlines || newlines)1627 return -1;1628 if (!deleted && !added)1629 return -1;16301631 fragment->leading = leading;1632 fragment->trailing = trailing;16331634 /*1635 * If a fragment ends with an incomplete line, we failed to include1636 * it in the above loop because we hit oldlines == newlines == 01637 * before seeing it.1638 */1639 if (12 < size && !memcmp(line, "\\ ", 2))1640 offset += linelen(line, size);16411642 patch->lines_added += added;1643 patch->lines_deleted += deleted;16441645 if (0 < patch->is_new && oldlines)1646 return error(_("new file depends on old contents"));1647 if (0 < patch->is_delete && newlines)1648 return error(_("deleted file still has contents"));1649 return offset;1650}16511652/*1653 * We have seen "diff --git a/... b/..." header (or a traditional patch1654 * header). Read hunks that belong to this patch into fragments and hang1655 * them to the given patch structure.1656 *1657 * The (fragment->patch, fragment->size) pair points into the memory given1658 * by the caller, not a copy, when we return.1659 */1660static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)1661{1662 unsigned long offset = 0;1663 unsigned long oldlines = 0, newlines = 0, context = 0;1664 struct fragment **fragp = &patch->fragments;16651666 while (size > 4 && !memcmp(line, "@@ -", 4)) {1667 struct fragment *fragment;1668 int len;16691670 fragment = xcalloc(1, sizeof(*fragment));1671 fragment->linenr = linenr;1672 len = parse_fragment(line, size, patch, fragment);1673 if (len <= 0)1674 die(_("corrupt patch at line %d"), linenr);1675 fragment->patch = line;1676 fragment->size = len;1677 oldlines += fragment->oldlines;1678 newlines += fragment->newlines;1679 context += fragment->leading + fragment->trailing;16801681 *fragp = fragment;1682 fragp = &fragment->next;16831684 offset += len;1685 line += len;1686 size -= len;1687 }16881689 /*1690 * If something was removed (i.e. we have old-lines) it cannot1691 * be creation, and if something was added it cannot be1692 * deletion. However, the reverse is not true; --unified=01693 * patches that only add are not necessarily creation even1694 * though they do not have any old lines, and ones that only1695 * delete are not necessarily deletion.1696 *1697 * Unfortunately, a real creation/deletion patch do _not_ have1698 * any context line by definition, so we cannot safely tell it1699 * apart with --unified=0 insanity. At least if the patch has1700 * more than one hunk it is not creation or deletion.1701 */1702 if (patch->is_new < 0 &&1703 (oldlines || (patch->fragments && patch->fragments->next)))1704 patch->is_new = 0;1705 if (patch->is_delete < 0 &&1706 (newlines || (patch->fragments && patch->fragments->next)))1707 patch->is_delete = 0;17081709 if (0 < patch->is_new && oldlines)1710 die(_("new file %s depends on old contents"), patch->new_name);1711 if (0 < patch->is_delete && newlines)1712 die(_("deleted file %s still has contents"), patch->old_name);1713 if (!patch->is_delete && !newlines && context)1714 fprintf_ln(stderr,1715 _("** warning: "1716 "file %s becomes empty but is not deleted"),1717 patch->new_name);17181719 return offset;1720}17211722static inline int metadata_changes(struct patch *patch)1723{1724 return patch->is_rename > 0 ||1725 patch->is_copy > 0 ||1726 patch->is_new > 0 ||1727 patch->is_delete ||1728 (patch->old_mode && patch->new_mode &&1729 patch->old_mode != patch->new_mode);1730}17311732static char *inflate_it(const void *data, unsigned long size,1733 unsigned long inflated_size)1734{1735 git_zstream stream;1736 void *out;1737 int st;17381739 memset(&stream, 0, sizeof(stream));17401741 stream.next_in = (unsigned char *)data;1742 stream.avail_in = size;1743 stream.next_out = out = xmalloc(inflated_size);1744 stream.avail_out = inflated_size;1745 git_inflate_init(&stream);1746 st = git_inflate(&stream, Z_FINISH);1747 git_inflate_end(&stream);1748 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1749 free(out);1750 return NULL;1751 }1752 return out;1753}17541755/*1756 * Read a binary hunk and return a new fragment; fragment->patch1757 * points at an allocated memory that the caller must free, so1758 * it is marked as "->free_patch = 1".1759 */1760static struct fragment *parse_binary_hunk(char **buf_p,1761 unsigned long *sz_p,1762 int *status_p,1763 int *used_p)1764{1765 /*1766 * Expect a line that begins with binary patch method ("literal"1767 * or "delta"), followed by the length of data before deflating.1768 * a sequence of 'length-byte' followed by base-85 encoded data1769 * should follow, terminated by a newline.1770 *1771 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1772 * and we would limit the patch line to 66 characters,1773 * so one line can fit up to 13 groups that would decode1774 * to 52 bytes max. The length byte 'A'-'Z' corresponds1775 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1776 */1777 int llen, used;1778 unsigned long size = *sz_p;1779 char *buffer = *buf_p;1780 int patch_method;1781 unsigned long origlen;1782 char *data = NULL;1783 int hunk_size = 0;1784 struct fragment *frag;17851786 llen = linelen(buffer, size);1787 used = llen;17881789 *status_p = 0;17901791 if (starts_with(buffer, "delta ")) {1792 patch_method = BINARY_DELTA_DEFLATED;1793 origlen = strtoul(buffer + 6, NULL, 10);1794 }1795 else if (starts_with(buffer, "literal ")) {1796 patch_method = BINARY_LITERAL_DEFLATED;1797 origlen = strtoul(buffer + 8, NULL, 10);1798 }1799 else1800 return NULL;18011802 linenr++;1803 buffer += llen;1804 while (1) {1805 int byte_length, max_byte_length, newsize;1806 llen = linelen(buffer, size);1807 used += llen;1808 linenr++;1809 if (llen == 1) {1810 /* consume the blank line */1811 buffer++;1812 size--;1813 break;1814 }1815 /*1816 * Minimum line is "A00000\n" which is 7-byte long,1817 * and the line length must be multiple of 5 plus 2.1818 */1819 if ((llen < 7) || (llen-2) % 5)1820 goto corrupt;1821 max_byte_length = (llen - 2) / 5 * 4;1822 byte_length = *buffer;1823 if ('A' <= byte_length && byte_length <= 'Z')1824 byte_length = byte_length - 'A' + 1;1825 else if ('a' <= byte_length && byte_length <= 'z')1826 byte_length = byte_length - 'a' + 27;1827 else1828 goto corrupt;1829 /* if the input length was not multiple of 4, we would1830 * have filler at the end but the filler should never1831 * exceed 3 bytes1832 */1833 if (max_byte_length < byte_length ||1834 byte_length <= max_byte_length - 4)1835 goto corrupt;1836 newsize = hunk_size + byte_length;1837 data = xrealloc(data, newsize);1838 if (decode_85(data + hunk_size, buffer + 1, byte_length))1839 goto corrupt;1840 hunk_size = newsize;1841 buffer += llen;1842 size -= llen;1843 }18441845 frag = xcalloc(1, sizeof(*frag));1846 frag->patch = inflate_it(data, hunk_size, origlen);1847 frag->free_patch = 1;1848 if (!frag->patch)1849 goto corrupt;1850 free(data);1851 frag->size = origlen;1852 *buf_p = buffer;1853 *sz_p = size;1854 *used_p = used;1855 frag->binary_patch_method = patch_method;1856 return frag;18571858 corrupt:1859 free(data);1860 *status_p = -1;1861 error(_("corrupt binary patch at line %d: %.*s"),1862 linenr-1, llen-1, buffer);1863 return NULL;1864}18651866static int parse_binary(char *buffer, unsigned long size, struct patch *patch)1867{1868 /*1869 * We have read "GIT binary patch\n"; what follows is a line1870 * that says the patch method (currently, either "literal" or1871 * "delta") and the length of data before deflating; a1872 * sequence of 'length-byte' followed by base-85 encoded data1873 * follows.1874 *1875 * When a binary patch is reversible, there is another binary1876 * hunk in the same format, starting with patch method (either1877 * "literal" or "delta") with the length of data, and a sequence1878 * of length-byte + base-85 encoded data, terminated with another1879 * empty line. This data, when applied to the postimage, produces1880 * the preimage.1881 */1882 struct fragment *forward;1883 struct fragment *reverse;1884 int status;1885 int used, used_1;18861887 forward = parse_binary_hunk(&buffer, &size, &status, &used);1888 if (!forward && !status)1889 /* there has to be one hunk (forward hunk) */1890 return error(_("unrecognized binary patch at line %d"), linenr-1);1891 if (status)1892 /* otherwise we already gave an error message */1893 return status;18941895 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);1896 if (reverse)1897 used += used_1;1898 else if (status) {1899 /*1900 * Not having reverse hunk is not an error, but having1901 * a corrupt reverse hunk is.1902 */1903 free((void*) forward->patch);1904 free(forward);1905 return status;1906 }1907 forward->next = reverse;1908 patch->fragments = forward;1909 patch->is_binary = 1;1910 return used;1911}19121913static void prefix_one(char **name)1914{1915 char *old_name = *name;1916 if (!old_name)1917 return;1918 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));1919 free(old_name);1920}19211922static void prefix_patch(struct patch *p)1923{1924 if (!prefix || p->is_toplevel_relative)1925 return;1926 prefix_one(&p->new_name);1927 prefix_one(&p->old_name);1928}19291930/*1931 * include/exclude1932 */19331934static struct string_list limit_by_name;1935static int has_include;1936static void add_name_limit(const char *name, int exclude)1937{1938 struct string_list_item *it;19391940 it = string_list_append(&limit_by_name, name);1941 it->util = exclude ? NULL : (void *) 1;1942}19431944static int use_patch(struct patch *p)1945{1946 const char *pathname = p->new_name ? p->new_name : p->old_name;1947 int i;19481949 /* Paths outside are not touched regardless of "--include" */1950 if (0 < prefix_length) {1951 int pathlen = strlen(pathname);1952 if (pathlen <= prefix_length ||1953 memcmp(prefix, pathname, prefix_length))1954 return 0;1955 }19561957 /* See if it matches any of exclude/include rule */1958 for (i = 0; i < limit_by_name.nr; i++) {1959 struct string_list_item *it = &limit_by_name.items[i];1960 if (!wildmatch(it->string, pathname, 0, NULL))1961 return (it->util != NULL);1962 }19631964 /*1965 * If we had any include, a path that does not match any rule is1966 * not used. Otherwise, we saw bunch of exclude rules (or none)1967 * and such a path is used.1968 */1969 return !has_include;1970}197119721973/*1974 * Read the patch text in "buffer" that extends for "size" bytes; stop1975 * reading after seeing a single patch (i.e. changes to a single file).1976 * Create fragments (i.e. patch hunks) and hang them to the given patch.1977 * Return the number of bytes consumed, so that the caller can call us1978 * again for the next patch.1979 */1980static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)1981{1982 int hdrsize, patchsize;1983 int offset = find_header(buffer, size, &hdrsize, patch);19841985 if (offset < 0)1986 return offset;19871988 prefix_patch(patch);19891990 if (!use_patch(patch))1991 patch->ws_rule = 0;1992 else1993 patch->ws_rule = whitespace_rule(patch->new_name1994 ? patch->new_name1995 : patch->old_name);19961997 patchsize = parse_single_patch(buffer + offset + hdrsize,1998 size - offset - hdrsize, patch);19992000 if (!patchsize) {2001 static const char git_binary[] = "GIT binary patch\n";2002 int hd = hdrsize + offset;2003 unsigned long llen = linelen(buffer + hd, size - hd);20042005 if (llen == sizeof(git_binary) - 1 &&2006 !memcmp(git_binary, buffer + hd, llen)) {2007 int used;2008 linenr++;2009 used = parse_binary(buffer + hd + llen,2010 size - hd - llen, patch);2011 if (used)2012 patchsize = used + llen;2013 else2014 patchsize = 0;2015 }2016 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {2017 static const char *binhdr[] = {2018 "Binary files ",2019 "Files ",2020 NULL,2021 };2022 int i;2023 for (i = 0; binhdr[i]; i++) {2024 int len = strlen(binhdr[i]);2025 if (len < size - hd &&2026 !memcmp(binhdr[i], buffer + hd, len)) {2027 linenr++;2028 patch->is_binary = 1;2029 patchsize = llen;2030 break;2031 }2032 }2033 }20342035 /* Empty patch cannot be applied if it is a text patch2036 * without metadata change. A binary patch appears2037 * empty to us here.2038 */2039 if ((apply || check) &&2040 (!patch->is_binary && !metadata_changes(patch)))2041 die(_("patch with only garbage at line %d"), linenr);2042 }20432044 return offset + hdrsize + patchsize;2045}20462047#define swap(a,b) myswap((a),(b),sizeof(a))20482049#define myswap(a, b, size) do { \2050 unsigned char mytmp[size]; \2051 memcpy(mytmp, &a, size); \2052 memcpy(&a, &b, size); \2053 memcpy(&b, mytmp, size); \2054} while (0)20552056static void reverse_patches(struct patch *p)2057{2058 for (; p; p = p->next) {2059 struct fragment *frag = p->fragments;20602061 swap(p->new_name, p->old_name);2062 swap(p->new_mode, p->old_mode);2063 swap(p->is_new, p->is_delete);2064 swap(p->lines_added, p->lines_deleted);2065 swap(p->old_sha1_prefix, p->new_sha1_prefix);20662067 for (; frag; frag = frag->next) {2068 swap(frag->newpos, frag->oldpos);2069 swap(frag->newlines, frag->oldlines);2070 }2071 }2072}20732074static const char pluses[] =2075"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";2076static const char minuses[]=2077"----------------------------------------------------------------------";20782079static void show_stats(struct patch *patch)2080{2081 struct strbuf qname = STRBUF_INIT;2082 char *cp = patch->new_name ? patch->new_name : patch->old_name;2083 int max, add, del;20842085 quote_c_style(cp, &qname, NULL, 0);20862087 /*2088 * "scale" the filename2089 */2090 max = max_len;2091 if (max > 50)2092 max = 50;20932094 if (qname.len > max) {2095 cp = strchr(qname.buf + qname.len + 3 - max, '/');2096 if (!cp)2097 cp = qname.buf + qname.len + 3 - max;2098 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);2099 }21002101 if (patch->is_binary) {2102 printf(" %-*s | Bin\n", max, qname.buf);2103 strbuf_release(&qname);2104 return;2105 }21062107 printf(" %-*s |", max, qname.buf);2108 strbuf_release(&qname);21092110 /*2111 * scale the add/delete2112 */2113 max = max + max_change > 70 ? 70 - max : max_change;2114 add = patch->lines_added;2115 del = patch->lines_deleted;21162117 if (max_change > 0) {2118 int total = ((add + del) * max + max_change / 2) / max_change;2119 add = (add * max + max_change / 2) / max_change;2120 del = total - add;2121 }2122 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,2123 add, pluses, del, minuses);2124}21252126static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)2127{2128 switch (st->st_mode & S_IFMT) {2129 case S_IFLNK:2130 if (strbuf_readlink(buf, path, st->st_size) < 0)2131 return error(_("unable to read symlink %s"), path);2132 return 0;2133 case S_IFREG:2134 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)2135 return error(_("unable to open or read %s"), path);2136 convert_to_git(path, buf->buf, buf->len, buf, 0);2137 return 0;2138 default:2139 return -1;2140 }2141}21422143/*2144 * Update the preimage, and the common lines in postimage,2145 * from buffer buf of length len. If postlen is 0 the postimage2146 * is updated in place, otherwise it's updated on a new buffer2147 * of length postlen2148 */21492150static void update_pre_post_images(struct image *preimage,2151 struct image *postimage,2152 char *buf,2153 size_t len, size_t postlen)2154{2155 int i, ctx, reduced;2156 char *new, *old, *fixed;2157 struct image fixed_preimage;21582159 /*2160 * Update the preimage with whitespace fixes. Note that we2161 * are not losing preimage->buf -- apply_one_fragment() will2162 * free "oldlines".2163 */2164 prepare_image(&fixed_preimage, buf, len, 1);2165 assert(postlen2166 ? fixed_preimage.nr == preimage->nr2167 : fixed_preimage.nr <= preimage->nr);2168 for (i = 0; i < fixed_preimage.nr; i++)2169 fixed_preimage.line[i].flag = preimage->line[i].flag;2170 free(preimage->line_allocated);2171 *preimage = fixed_preimage;21722173 /*2174 * Adjust the common context lines in postimage. This can be2175 * done in-place when we are shrinking it with whitespace2176 * fixing, but needs a new buffer when ignoring whitespace or2177 * expanding leading tabs to spaces.2178 *2179 * We trust the caller to tell us if the update can be done2180 * in place (postlen==0) or not.2181 */2182 old = postimage->buf;2183 if (postlen)2184 new = postimage->buf = xmalloc(postlen);2185 else2186 new = old;2187 fixed = preimage->buf;21882189 for (i = reduced = ctx = 0; i < postimage->nr; i++) {2190 size_t len = postimage->line[i].len;2191 if (!(postimage->line[i].flag & LINE_COMMON)) {2192 /* an added line -- no counterparts in preimage */2193 memmove(new, old, len);2194 old += len;2195 new += len;2196 continue;2197 }21982199 /* a common context -- skip it in the original postimage */2200 old += len;22012202 /* and find the corresponding one in the fixed preimage */2203 while (ctx < preimage->nr &&2204 !(preimage->line[ctx].flag & LINE_COMMON)) {2205 fixed += preimage->line[ctx].len;2206 ctx++;2207 }22082209 /*2210 * preimage is expected to run out, if the caller2211 * fixed addition of trailing blank lines.2212 */2213 if (preimage->nr <= ctx) {2214 reduced++;2215 continue;2216 }22172218 /* and copy it in, while fixing the line length */2219 len = preimage->line[ctx].len;2220 memcpy(new, fixed, len);2221 new += len;2222 fixed += len;2223 postimage->line[i].len = len;2224 ctx++;2225 }22262227 if (postlen2228 ? postlen < new - postimage->buf2229 : postimage->len < new - postimage->buf)2230 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",2231 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));22322233 /* Fix the length of the whole thing */2234 postimage->len = new - postimage->buf;2235 postimage->nr -= reduced;2236}22372238static int match_fragment(struct image *img,2239 struct image *preimage,2240 struct image *postimage,2241 unsigned long try,2242 int try_lno,2243 unsigned ws_rule,2244 int match_beginning, int match_end)2245{2246 int i;2247 char *fixed_buf, *buf, *orig, *target;2248 struct strbuf fixed;2249 size_t fixed_len, postlen;2250 int preimage_limit;22512252 if (preimage->nr + try_lno <= img->nr) {2253 /*2254 * The hunk falls within the boundaries of img.2255 */2256 preimage_limit = preimage->nr;2257 if (match_end && (preimage->nr + try_lno != img->nr))2258 return 0;2259 } else if (ws_error_action == correct_ws_error &&2260 (ws_rule & WS_BLANK_AT_EOF)) {2261 /*2262 * This hunk extends beyond the end of img, and we are2263 * removing blank lines at the end of the file. This2264 * many lines from the beginning of the preimage must2265 * match with img, and the remainder of the preimage2266 * must be blank.2267 */2268 preimage_limit = img->nr - try_lno;2269 } else {2270 /*2271 * The hunk extends beyond the end of the img and2272 * we are not removing blanks at the end, so we2273 * should reject the hunk at this position.2274 */2275 return 0;2276 }22772278 if (match_beginning && try_lno)2279 return 0;22802281 /* Quick hash check */2282 for (i = 0; i < preimage_limit; i++)2283 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||2284 (preimage->line[i].hash != img->line[try_lno + i].hash))2285 return 0;22862287 if (preimage_limit == preimage->nr) {2288 /*2289 * Do we have an exact match? If we were told to match2290 * at the end, size must be exactly at try+fragsize,2291 * otherwise try+fragsize must be still within the preimage,2292 * and either case, the old piece should match the preimage2293 * exactly.2294 */2295 if ((match_end2296 ? (try + preimage->len == img->len)2297 : (try + preimage->len <= img->len)) &&2298 !memcmp(img->buf + try, preimage->buf, preimage->len))2299 return 1;2300 } else {2301 /*2302 * The preimage extends beyond the end of img, so2303 * there cannot be an exact match.2304 *2305 * There must be one non-blank context line that match2306 * a line before the end of img.2307 */2308 char *buf_end;23092310 buf = preimage->buf;2311 buf_end = buf;2312 for (i = 0; i < preimage_limit; i++)2313 buf_end += preimage->line[i].len;23142315 for ( ; buf < buf_end; buf++)2316 if (!isspace(*buf))2317 break;2318 if (buf == buf_end)2319 return 0;2320 }23212322 /*2323 * No exact match. If we are ignoring whitespace, run a line-by-line2324 * fuzzy matching. We collect all the line length information because2325 * we need it to adjust whitespace if we match.2326 */2327 if (ws_ignore_action == ignore_ws_change) {2328 size_t imgoff = 0;2329 size_t preoff = 0;2330 size_t postlen = postimage->len;2331 size_t extra_chars;2332 char *preimage_eof;2333 char *preimage_end;2334 for (i = 0; i < preimage_limit; i++) {2335 size_t prelen = preimage->line[i].len;2336 size_t imglen = img->line[try_lno+i].len;23372338 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,2339 preimage->buf + preoff, prelen))2340 return 0;2341 if (preimage->line[i].flag & LINE_COMMON)2342 postlen += imglen - prelen;2343 imgoff += imglen;2344 preoff += prelen;2345 }23462347 /*2348 * Ok, the preimage matches with whitespace fuzz.2349 *2350 * imgoff now holds the true length of the target that2351 * matches the preimage before the end of the file.2352 *2353 * Count the number of characters in the preimage that fall2354 * beyond the end of the file and make sure that all of them2355 * are whitespace characters. (This can only happen if2356 * we are removing blank lines at the end of the file.)2357 */2358 buf = preimage_eof = preimage->buf + preoff;2359 for ( ; i < preimage->nr; i++)2360 preoff += preimage->line[i].len;2361 preimage_end = preimage->buf + preoff;2362 for ( ; buf < preimage_end; buf++)2363 if (!isspace(*buf))2364 return 0;23652366 /*2367 * Update the preimage and the common postimage context2368 * lines to use the same whitespace as the target.2369 * If whitespace is missing in the target (i.e.2370 * if the preimage extends beyond the end of the file),2371 * use the whitespace from the preimage.2372 */2373 extra_chars = preimage_end - preimage_eof;2374 strbuf_init(&fixed, imgoff + extra_chars);2375 strbuf_add(&fixed, img->buf + try, imgoff);2376 strbuf_add(&fixed, preimage_eof, extra_chars);2377 fixed_buf = strbuf_detach(&fixed, &fixed_len);2378 update_pre_post_images(preimage, postimage,2379 fixed_buf, fixed_len, postlen);2380 return 1;2381 }23822383 if (ws_error_action != correct_ws_error)2384 return 0;23852386 /*2387 * The hunk does not apply byte-by-byte, but the hash says2388 * it might with whitespace fuzz. We weren't asked to2389 * ignore whitespace, we were asked to correct whitespace2390 * errors, so let's try matching after whitespace correction.2391 *2392 * While checking the preimage against the target, whitespace2393 * errors in both fixed, we count how large the corresponding2394 * postimage needs to be. The postimage prepared by2395 * apply_one_fragment() has whitespace errors fixed on added2396 * lines already, but the common lines were propagated as-is,2397 * which may become longer when their whitespace errors are2398 * fixed.2399 */24002401 /* First count added lines in postimage */2402 postlen = 0;2403 for (i = 0; i < postimage->nr; i++) {2404 if (!(postimage->line[i].flag & LINE_COMMON))2405 postlen += postimage->line[i].len;2406 }24072408 /*2409 * The preimage may extend beyond the end of the file,2410 * but in this loop we will only handle the part of the2411 * preimage that falls within the file.2412 */2413 strbuf_init(&fixed, preimage->len + 1);2414 orig = preimage->buf;2415 target = img->buf + try;2416 for (i = 0; i < preimage_limit; i++) {2417 size_t oldlen = preimage->line[i].len;2418 size_t tgtlen = img->line[try_lno + i].len;2419 size_t fixstart = fixed.len;2420 struct strbuf tgtfix;2421 int match;24222423 /* Try fixing the line in the preimage */2424 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24252426 /* Try fixing the line in the target */2427 strbuf_init(&tgtfix, tgtlen);2428 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);24292430 /*2431 * If they match, either the preimage was based on2432 * a version before our tree fixed whitespace breakage,2433 * or we are lacking a whitespace-fix patch the tree2434 * the preimage was based on already had (i.e. target2435 * has whitespace breakage, the preimage doesn't).2436 * In either case, we are fixing the whitespace breakages2437 * so we might as well take the fix together with their2438 * real change.2439 */2440 match = (tgtfix.len == fixed.len - fixstart &&2441 !memcmp(tgtfix.buf, fixed.buf + fixstart,2442 fixed.len - fixstart));24432444 /* Add the length if this is common with the postimage */2445 if (preimage->line[i].flag & LINE_COMMON)2446 postlen += tgtfix.len;24472448 strbuf_release(&tgtfix);2449 if (!match)2450 goto unmatch_exit;24512452 orig += oldlen;2453 target += tgtlen;2454 }245524562457 /*2458 * Now handle the lines in the preimage that falls beyond the2459 * end of the file (if any). They will only match if they are2460 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2461 * false).2462 */2463 for ( ; i < preimage->nr; i++) {2464 size_t fixstart = fixed.len; /* start of the fixed preimage */2465 size_t oldlen = preimage->line[i].len;2466 int j;24672468 /* Try fixing the line in the preimage */2469 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24702471 for (j = fixstart; j < fixed.len; j++)2472 if (!isspace(fixed.buf[j]))2473 goto unmatch_exit;24742475 orig += oldlen;2476 }24772478 /*2479 * Yes, the preimage is based on an older version that still2480 * has whitespace breakages unfixed, and fixing them makes the2481 * hunk match. Update the context lines in the postimage.2482 */2483 fixed_buf = strbuf_detach(&fixed, &fixed_len);2484 if (postlen < postimage->len)2485 postlen = 0;2486 update_pre_post_images(preimage, postimage,2487 fixed_buf, fixed_len, postlen);2488 return 1;24892490 unmatch_exit:2491 strbuf_release(&fixed);2492 return 0;2493}24942495static int find_pos(struct image *img,2496 struct image *preimage,2497 struct image *postimage,2498 int line,2499 unsigned ws_rule,2500 int match_beginning, int match_end)2501{2502 int i;2503 unsigned long backwards, forwards, try;2504 int backwards_lno, forwards_lno, try_lno;25052506 /*2507 * If match_beginning or match_end is specified, there is no2508 * point starting from a wrong line that will never match and2509 * wander around and wait for a match at the specified end.2510 */2511 if (match_beginning)2512 line = 0;2513 else if (match_end)2514 line = img->nr - preimage->nr;25152516 /*2517 * Because the comparison is unsigned, the following test2518 * will also take care of a negative line number that can2519 * result when match_end and preimage is larger than the target.2520 */2521 if ((size_t) line > img->nr)2522 line = img->nr;25232524 try = 0;2525 for (i = 0; i < line; i++)2526 try += img->line[i].len;25272528 /*2529 * There's probably some smart way to do this, but I'll leave2530 * that to the smart and beautiful people. I'm simple and stupid.2531 */2532 backwards = try;2533 backwards_lno = line;2534 forwards = try;2535 forwards_lno = line;2536 try_lno = line;25372538 for (i = 0; ; i++) {2539 if (match_fragment(img, preimage, postimage,2540 try, try_lno, ws_rule,2541 match_beginning, match_end))2542 return try_lno;25432544 again:2545 if (backwards_lno == 0 && forwards_lno == img->nr)2546 break;25472548 if (i & 1) {2549 if (backwards_lno == 0) {2550 i++;2551 goto again;2552 }2553 backwards_lno--;2554 backwards -= img->line[backwards_lno].len;2555 try = backwards;2556 try_lno = backwards_lno;2557 } else {2558 if (forwards_lno == img->nr) {2559 i++;2560 goto again;2561 }2562 forwards += img->line[forwards_lno].len;2563 forwards_lno++;2564 try = forwards;2565 try_lno = forwards_lno;2566 }25672568 }2569 return -1;2570}25712572static void remove_first_line(struct image *img)2573{2574 img->buf += img->line[0].len;2575 img->len -= img->line[0].len;2576 img->line++;2577 img->nr--;2578}25792580static void remove_last_line(struct image *img)2581{2582 img->len -= img->line[--img->nr].len;2583}25842585/*2586 * The change from "preimage" and "postimage" has been found to2587 * apply at applied_pos (counts in line numbers) in "img".2588 * Update "img" to remove "preimage" and replace it with "postimage".2589 */2590static void update_image(struct image *img,2591 int applied_pos,2592 struct image *preimage,2593 struct image *postimage)2594{2595 /*2596 * remove the copy of preimage at offset in img2597 * and replace it with postimage2598 */2599 int i, nr;2600 size_t remove_count, insert_count, applied_at = 0;2601 char *result;2602 int preimage_limit;26032604 /*2605 * If we are removing blank lines at the end of img,2606 * the preimage may extend beyond the end.2607 * If that is the case, we must be careful only to2608 * remove the part of the preimage that falls within2609 * the boundaries of img. Initialize preimage_limit2610 * to the number of lines in the preimage that falls2611 * within the boundaries.2612 */2613 preimage_limit = preimage->nr;2614 if (preimage_limit > img->nr - applied_pos)2615 preimage_limit = img->nr - applied_pos;26162617 for (i = 0; i < applied_pos; i++)2618 applied_at += img->line[i].len;26192620 remove_count = 0;2621 for (i = 0; i < preimage_limit; i++)2622 remove_count += img->line[applied_pos + i].len;2623 insert_count = postimage->len;26242625 /* Adjust the contents */2626 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));2627 memcpy(result, img->buf, applied_at);2628 memcpy(result + applied_at, postimage->buf, postimage->len);2629 memcpy(result + applied_at + postimage->len,2630 img->buf + (applied_at + remove_count),2631 img->len - (applied_at + remove_count));2632 free(img->buf);2633 img->buf = result;2634 img->len += insert_count - remove_count;2635 result[img->len] = '\0';26362637 /* Adjust the line table */2638 nr = img->nr + postimage->nr - preimage_limit;2639 if (preimage_limit < postimage->nr) {2640 /*2641 * NOTE: this knows that we never call remove_first_line()2642 * on anything other than pre/post image.2643 */2644 REALLOC_ARRAY(img->line, nr);2645 img->line_allocated = img->line;2646 }2647 if (preimage_limit != postimage->nr)2648 memmove(img->line + applied_pos + postimage->nr,2649 img->line + applied_pos + preimage_limit,2650 (img->nr - (applied_pos + preimage_limit)) *2651 sizeof(*img->line));2652 memcpy(img->line + applied_pos,2653 postimage->line,2654 postimage->nr * sizeof(*img->line));2655 if (!allow_overlap)2656 for (i = 0; i < postimage->nr; i++)2657 img->line[applied_pos + i].flag |= LINE_PATCHED;2658 img->nr = nr;2659}26602661/*2662 * Use the patch-hunk text in "frag" to prepare two images (preimage and2663 * postimage) for the hunk. Find lines that match "preimage" in "img" and2664 * replace the part of "img" with "postimage" text.2665 */2666static int apply_one_fragment(struct image *img, struct fragment *frag,2667 int inaccurate_eof, unsigned ws_rule,2668 int nth_fragment)2669{2670 int match_beginning, match_end;2671 const char *patch = frag->patch;2672 int size = frag->size;2673 char *old, *oldlines;2674 struct strbuf newlines;2675 int new_blank_lines_at_end = 0;2676 int found_new_blank_lines_at_end = 0;2677 int hunk_linenr = frag->linenr;2678 unsigned long leading, trailing;2679 int pos, applied_pos;2680 struct image preimage;2681 struct image postimage;26822683 memset(&preimage, 0, sizeof(preimage));2684 memset(&postimage, 0, sizeof(postimage));2685 oldlines = xmalloc(size);2686 strbuf_init(&newlines, size);26872688 old = oldlines;2689 while (size > 0) {2690 char first;2691 int len = linelen(patch, size);2692 int plen;2693 int added_blank_line = 0;2694 int is_blank_context = 0;2695 size_t start;26962697 if (!len)2698 break;26992700 /*2701 * "plen" is how much of the line we should use for2702 * the actual patch data. Normally we just remove the2703 * first character on the line, but if the line is2704 * followed by "\ No newline", then we also remove the2705 * last one (which is the newline, of course).2706 */2707 plen = len - 1;2708 if (len < size && patch[len] == '\\')2709 plen--;2710 first = *patch;2711 if (apply_in_reverse) {2712 if (first == '-')2713 first = '+';2714 else if (first == '+')2715 first = '-';2716 }27172718 switch (first) {2719 case '\n':2720 /* Newer GNU diff, empty context line */2721 if (plen < 0)2722 /* ... followed by '\No newline'; nothing */2723 break;2724 *old++ = '\n';2725 strbuf_addch(&newlines, '\n');2726 add_line_info(&preimage, "\n", 1, LINE_COMMON);2727 add_line_info(&postimage, "\n", 1, LINE_COMMON);2728 is_blank_context = 1;2729 break;2730 case ' ':2731 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2732 ws_blank_line(patch + 1, plen, ws_rule))2733 is_blank_context = 1;2734 case '-':2735 memcpy(old, patch + 1, plen);2736 add_line_info(&preimage, old, plen,2737 (first == ' ' ? LINE_COMMON : 0));2738 old += plen;2739 if (first == '-')2740 break;2741 /* Fall-through for ' ' */2742 case '+':2743 /* --no-add does not add new lines */2744 if (first == '+' && no_add)2745 break;27462747 start = newlines.len;2748 if (first != '+' ||2749 !whitespace_error ||2750 ws_error_action != correct_ws_error) {2751 strbuf_add(&newlines, patch + 1, plen);2752 }2753 else {2754 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);2755 }2756 add_line_info(&postimage, newlines.buf + start, newlines.len - start,2757 (first == '+' ? 0 : LINE_COMMON));2758 if (first == '+' &&2759 (ws_rule & WS_BLANK_AT_EOF) &&2760 ws_blank_line(patch + 1, plen, ws_rule))2761 added_blank_line = 1;2762 break;2763 case '@': case '\\':2764 /* Ignore it, we already handled it */2765 break;2766 default:2767 if (apply_verbosely)2768 error(_("invalid start of line: '%c'"), first);2769 applied_pos = -1;2770 goto out;2771 }2772 if (added_blank_line) {2773 if (!new_blank_lines_at_end)2774 found_new_blank_lines_at_end = hunk_linenr;2775 new_blank_lines_at_end++;2776 }2777 else if (is_blank_context)2778 ;2779 else2780 new_blank_lines_at_end = 0;2781 patch += len;2782 size -= len;2783 hunk_linenr++;2784 }2785 if (inaccurate_eof &&2786 old > oldlines && old[-1] == '\n' &&2787 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {2788 old--;2789 strbuf_setlen(&newlines, newlines.len - 1);2790 }27912792 leading = frag->leading;2793 trailing = frag->trailing;27942795 /*2796 * A hunk to change lines at the beginning would begin with2797 * @@ -1,L +N,M @@2798 * but we need to be careful. -U0 that inserts before the second2799 * line also has this pattern.2800 *2801 * And a hunk to add to an empty file would begin with2802 * @@ -0,0 +N,M @@2803 *2804 * In other words, a hunk that is (frag->oldpos <= 1) with or2805 * without leading context must match at the beginning.2806 */2807 match_beginning = (!frag->oldpos ||2808 (frag->oldpos == 1 && !unidiff_zero));28092810 /*2811 * A hunk without trailing lines must match at the end.2812 * However, we simply cannot tell if a hunk must match end2813 * from the lack of trailing lines if the patch was generated2814 * with unidiff without any context.2815 */2816 match_end = !unidiff_zero && !trailing;28172818 pos = frag->newpos ? (frag->newpos - 1) : 0;2819 preimage.buf = oldlines;2820 preimage.len = old - oldlines;2821 postimage.buf = newlines.buf;2822 postimage.len = newlines.len;2823 preimage.line = preimage.line_allocated;2824 postimage.line = postimage.line_allocated;28252826 for (;;) {28272828 applied_pos = find_pos(img, &preimage, &postimage, pos,2829 ws_rule, match_beginning, match_end);28302831 if (applied_pos >= 0)2832 break;28332834 /* Am I at my context limits? */2835 if ((leading <= p_context) && (trailing <= p_context))2836 break;2837 if (match_beginning || match_end) {2838 match_beginning = match_end = 0;2839 continue;2840 }28412842 /*2843 * Reduce the number of context lines; reduce both2844 * leading and trailing if they are equal otherwise2845 * just reduce the larger context.2846 */2847 if (leading >= trailing) {2848 remove_first_line(&preimage);2849 remove_first_line(&postimage);2850 pos--;2851 leading--;2852 }2853 if (trailing > leading) {2854 remove_last_line(&preimage);2855 remove_last_line(&postimage);2856 trailing--;2857 }2858 }28592860 if (applied_pos >= 0) {2861 if (new_blank_lines_at_end &&2862 preimage.nr + applied_pos >= img->nr &&2863 (ws_rule & WS_BLANK_AT_EOF) &&2864 ws_error_action != nowarn_ws_error) {2865 record_ws_error(WS_BLANK_AT_EOF, "+", 1,2866 found_new_blank_lines_at_end);2867 if (ws_error_action == correct_ws_error) {2868 while (new_blank_lines_at_end--)2869 remove_last_line(&postimage);2870 }2871 /*2872 * We would want to prevent write_out_results()2873 * from taking place in apply_patch() that follows2874 * the callchain led us here, which is:2875 * apply_patch->check_patch_list->check_patch->2876 * apply_data->apply_fragments->apply_one_fragment2877 */2878 if (ws_error_action == die_on_ws_error)2879 apply = 0;2880 }28812882 if (apply_verbosely && applied_pos != pos) {2883 int offset = applied_pos - pos;2884 if (apply_in_reverse)2885 offset = 0 - offset;2886 fprintf_ln(stderr,2887 Q_("Hunk #%d succeeded at %d (offset %d line).",2888 "Hunk #%d succeeded at %d (offset %d lines).",2889 offset),2890 nth_fragment, applied_pos + 1, offset);2891 }28922893 /*2894 * Warn if it was necessary to reduce the number2895 * of context lines.2896 */2897 if ((leading != frag->leading) ||2898 (trailing != frag->trailing))2899 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"2900 " to apply fragment at %d"),2901 leading, trailing, applied_pos+1);2902 update_image(img, applied_pos, &preimage, &postimage);2903 } else {2904 if (apply_verbosely)2905 error(_("while searching for:\n%.*s"),2906 (int)(old - oldlines), oldlines);2907 }29082909out:2910 free(oldlines);2911 strbuf_release(&newlines);2912 free(preimage.line_allocated);2913 free(postimage.line_allocated);29142915 return (applied_pos < 0);2916}29172918static int apply_binary_fragment(struct image *img, struct patch *patch)2919{2920 struct fragment *fragment = patch->fragments;2921 unsigned long len;2922 void *dst;29232924 if (!fragment)2925 return error(_("missing binary patch data for '%s'"),2926 patch->new_name ?2927 patch->new_name :2928 patch->old_name);29292930 /* Binary patch is irreversible without the optional second hunk */2931 if (apply_in_reverse) {2932 if (!fragment->next)2933 return error("cannot reverse-apply a binary patch "2934 "without the reverse hunk to '%s'",2935 patch->new_name2936 ? patch->new_name : patch->old_name);2937 fragment = fragment->next;2938 }2939 switch (fragment->binary_patch_method) {2940 case BINARY_DELTA_DEFLATED:2941 dst = patch_delta(img->buf, img->len, fragment->patch,2942 fragment->size, &len);2943 if (!dst)2944 return -1;2945 clear_image(img);2946 img->buf = dst;2947 img->len = len;2948 return 0;2949 case BINARY_LITERAL_DEFLATED:2950 clear_image(img);2951 img->len = fragment->size;2952 img->buf = xmemdupz(fragment->patch, img->len);2953 return 0;2954 }2955 return -1;2956}29572958/*2959 * Replace "img" with the result of applying the binary patch.2960 * The binary patch data itself in patch->fragment is still kept2961 * but the preimage prepared by the caller in "img" is freed here2962 * or in the helper function apply_binary_fragment() this calls.2963 */2964static int apply_binary(struct image *img, struct patch *patch)2965{2966 const char *name = patch->old_name ? patch->old_name : patch->new_name;2967 unsigned char sha1[20];29682969 /*2970 * For safety, we require patch index line to contain2971 * full 40-byte textual SHA1 for old and new, at least for now.2972 */2973 if (strlen(patch->old_sha1_prefix) != 40 ||2974 strlen(patch->new_sha1_prefix) != 40 ||2975 get_sha1_hex(patch->old_sha1_prefix, sha1) ||2976 get_sha1_hex(patch->new_sha1_prefix, sha1))2977 return error("cannot apply binary patch to '%s' "2978 "without full index line", name);29792980 if (patch->old_name) {2981 /*2982 * See if the old one matches what the patch2983 * applies to.2984 */2985 hash_sha1_file(img->buf, img->len, blob_type, sha1);2986 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))2987 return error("the patch applies to '%s' (%s), "2988 "which does not match the "2989 "current contents.",2990 name, sha1_to_hex(sha1));2991 }2992 else {2993 /* Otherwise, the old one must be empty. */2994 if (img->len)2995 return error("the patch applies to an empty "2996 "'%s' but it is not empty", name);2997 }29982999 get_sha1_hex(patch->new_sha1_prefix, sha1);3000 if (is_null_sha1(sha1)) {3001 clear_image(img);3002 return 0; /* deletion patch */3003 }30043005 if (has_sha1_file(sha1)) {3006 /* We already have the postimage */3007 enum object_type type;3008 unsigned long size;3009 char *result;30103011 result = read_sha1_file(sha1, &type, &size);3012 if (!result)3013 return error("the necessary postimage %s for "3014 "'%s' cannot be read",3015 patch->new_sha1_prefix, name);3016 clear_image(img);3017 img->buf = result;3018 img->len = size;3019 } else {3020 /*3021 * We have verified buf matches the preimage;3022 * apply the patch data to it, which is stored3023 * in the patch->fragments->{patch,size}.3024 */3025 if (apply_binary_fragment(img, patch))3026 return error(_("binary patch does not apply to '%s'"),3027 name);30283029 /* verify that the result matches */3030 hash_sha1_file(img->buf, img->len, blob_type, sha1);3031 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))3032 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),3033 name, patch->new_sha1_prefix, sha1_to_hex(sha1));3034 }30353036 return 0;3037}30383039static int apply_fragments(struct image *img, struct patch *patch)3040{3041 struct fragment *frag = patch->fragments;3042 const char *name = patch->old_name ? patch->old_name : patch->new_name;3043 unsigned ws_rule = patch->ws_rule;3044 unsigned inaccurate_eof = patch->inaccurate_eof;3045 int nth = 0;30463047 if (patch->is_binary)3048 return apply_binary(img, patch);30493050 while (frag) {3051 nth++;3052 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {3053 error(_("patch failed: %s:%ld"), name, frag->oldpos);3054 if (!apply_with_reject)3055 return -1;3056 frag->rejected = 1;3057 }3058 frag = frag->next;3059 }3060 return 0;3061}30623063static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)3064{3065 if (S_ISGITLINK(mode)) {3066 strbuf_grow(buf, 100);3067 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));3068 } else {3069 enum object_type type;3070 unsigned long sz;3071 char *result;30723073 result = read_sha1_file(sha1, &type, &sz);3074 if (!result)3075 return -1;3076 /* XXX read_sha1_file NUL-terminates */3077 strbuf_attach(buf, result, sz, sz + 1);3078 }3079 return 0;3080}30813082static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)3083{3084 if (!ce)3085 return 0;3086 return read_blob_object(buf, ce->sha1, ce->ce_mode);3087}30883089static struct patch *in_fn_table(const char *name)3090{3091 struct string_list_item *item;30923093 if (name == NULL)3094 return NULL;30953096 item = string_list_lookup(&fn_table, name);3097 if (item != NULL)3098 return (struct patch *)item->util;30993100 return NULL;3101}31023103/*3104 * item->util in the filename table records the status of the path.3105 * Usually it points at a patch (whose result records the contents3106 * of it after applying it), but it could be PATH_WAS_DELETED for a3107 * path that a previously applied patch has already removed, or3108 * PATH_TO_BE_DELETED for a path that a later patch would remove.3109 *3110 * The latter is needed to deal with a case where two paths A and B3111 * are swapped by first renaming A to B and then renaming B to A;3112 * moving A to B should not be prevented due to presence of B as we3113 * will remove it in a later patch.3114 */3115#define PATH_TO_BE_DELETED ((struct patch *) -2)3116#define PATH_WAS_DELETED ((struct patch *) -1)31173118static int to_be_deleted(struct patch *patch)3119{3120 return patch == PATH_TO_BE_DELETED;3121}31223123static int was_deleted(struct patch *patch)3124{3125 return patch == PATH_WAS_DELETED;3126}31273128static void add_to_fn_table(struct patch *patch)3129{3130 struct string_list_item *item;31313132 /*3133 * Always add new_name unless patch is a deletion3134 * This should cover the cases for normal diffs,3135 * file creations and copies3136 */3137 if (patch->new_name != NULL) {3138 item = string_list_insert(&fn_table, patch->new_name);3139 item->util = patch;3140 }31413142 /*3143 * store a failure on rename/deletion cases because3144 * later chunks shouldn't patch old names3145 */3146 if ((patch->new_name == NULL) || (patch->is_rename)) {3147 item = string_list_insert(&fn_table, patch->old_name);3148 item->util = PATH_WAS_DELETED;3149 }3150}31513152static void prepare_fn_table(struct patch *patch)3153{3154 /*3155 * store information about incoming file deletion3156 */3157 while (patch) {3158 if ((patch->new_name == NULL) || (patch->is_rename)) {3159 struct string_list_item *item;3160 item = string_list_insert(&fn_table, patch->old_name);3161 item->util = PATH_TO_BE_DELETED;3162 }3163 patch = patch->next;3164 }3165}31663167static int checkout_target(struct index_state *istate,3168 struct cache_entry *ce, struct stat *st)3169{3170 struct checkout costate;31713172 memset(&costate, 0, sizeof(costate));3173 costate.base_dir = "";3174 costate.refresh_cache = 1;3175 costate.istate = istate;3176 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))3177 return error(_("cannot checkout %s"), ce->name);3178 return 0;3179}31803181static struct patch *previous_patch(struct patch *patch, int *gone)3182{3183 struct patch *previous;31843185 *gone = 0;3186 if (patch->is_copy || patch->is_rename)3187 return NULL; /* "git" patches do not depend on the order */31883189 previous = in_fn_table(patch->old_name);3190 if (!previous)3191 return NULL;31923193 if (to_be_deleted(previous))3194 return NULL; /* the deletion hasn't happened yet */31953196 if (was_deleted(previous))3197 *gone = 1;31983199 return previous;3200}32013202static int verify_index_match(const struct cache_entry *ce, struct stat *st)3203{3204 if (S_ISGITLINK(ce->ce_mode)) {3205 if (!S_ISDIR(st->st_mode))3206 return -1;3207 return 0;3208 }3209 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);3210}32113212#define SUBMODULE_PATCH_WITHOUT_INDEX 132133214static int load_patch_target(struct strbuf *buf,3215 const struct cache_entry *ce,3216 struct stat *st,3217 const char *name,3218 unsigned expected_mode)3219{3220 if (cached || check_index) {3221 if (read_file_or_gitlink(ce, buf))3222 return error(_("read of %s failed"), name);3223 } else if (name) {3224 if (S_ISGITLINK(expected_mode)) {3225 if (ce)3226 return read_file_or_gitlink(ce, buf);3227 else3228 return SUBMODULE_PATCH_WITHOUT_INDEX;3229 } else if (has_symlink_leading_path(name, strlen(name))) {3230 return error(_("reading from '%s' beyond a symbolic link"), name);3231 } else {3232 if (read_old_data(st, name, buf))3233 return error(_("read of %s failed"), name);3234 }3235 }3236 return 0;3237}32383239/*3240 * We are about to apply "patch"; populate the "image" with the3241 * current version we have, from the working tree or from the index,3242 * depending on the situation e.g. --cached/--index. If we are3243 * applying a non-git patch that incrementally updates the tree,3244 * we read from the result of a previous diff.3245 */3246static int load_preimage(struct image *image,3247 struct patch *patch, struct stat *st,3248 const struct cache_entry *ce)3249{3250 struct strbuf buf = STRBUF_INIT;3251 size_t len;3252 char *img;3253 struct patch *previous;3254 int status;32553256 previous = previous_patch(patch, &status);3257 if (status)3258 return error(_("path %s has been renamed/deleted"),3259 patch->old_name);3260 if (previous) {3261 /* We have a patched copy in memory; use that. */3262 strbuf_add(&buf, previous->result, previous->resultsize);3263 } else {3264 status = load_patch_target(&buf, ce, st,3265 patch->old_name, patch->old_mode);3266 if (status < 0)3267 return status;3268 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {3269 /*3270 * There is no way to apply subproject3271 * patch without looking at the index.3272 * NEEDSWORK: shouldn't this be flagged3273 * as an error???3274 */3275 free_fragment_list(patch->fragments);3276 patch->fragments = NULL;3277 } else if (status) {3278 return error(_("read of %s failed"), patch->old_name);3279 }3280 }32813282 img = strbuf_detach(&buf, &len);3283 prepare_image(image, img, len, !patch->is_binary);3284 return 0;3285}32863287static int three_way_merge(struct image *image,3288 char *path,3289 const unsigned char *base,3290 const unsigned char *ours,3291 const unsigned char *theirs)3292{3293 mmfile_t base_file, our_file, their_file;3294 mmbuffer_t result = { NULL };3295 int status;32963297 read_mmblob(&base_file, base);3298 read_mmblob(&our_file, ours);3299 read_mmblob(&their_file, theirs);3300 status = ll_merge(&result, path,3301 &base_file, "base",3302 &our_file, "ours",3303 &their_file, "theirs", NULL);3304 free(base_file.ptr);3305 free(our_file.ptr);3306 free(their_file.ptr);3307 if (status < 0 || !result.ptr) {3308 free(result.ptr);3309 return -1;3310 }3311 clear_image(image);3312 image->buf = result.ptr;3313 image->len = result.size;33143315 return status;3316}33173318/*3319 * When directly falling back to add/add three-way merge, we read from3320 * the current contents of the new_name. In no cases other than that3321 * this function will be called.3322 */3323static int load_current(struct image *image, struct patch *patch)3324{3325 struct strbuf buf = STRBUF_INIT;3326 int status, pos;3327 size_t len;3328 char *img;3329 struct stat st;3330 struct cache_entry *ce;3331 char *name = patch->new_name;3332 unsigned mode = patch->new_mode;33333334 if (!patch->is_new)3335 die("BUG: patch to %s is not a creation", patch->old_name);33363337 pos = cache_name_pos(name, strlen(name));3338 if (pos < 0)3339 return error(_("%s: does not exist in index"), name);3340 ce = active_cache[pos];3341 if (lstat(name, &st)) {3342 if (errno != ENOENT)3343 return error(_("%s: %s"), name, strerror(errno));3344 if (checkout_target(&the_index, ce, &st))3345 return -1;3346 }3347 if (verify_index_match(ce, &st))3348 return error(_("%s: does not match index"), name);33493350 status = load_patch_target(&buf, ce, &st, name, mode);3351 if (status < 0)3352 return status;3353 else if (status)3354 return -1;3355 img = strbuf_detach(&buf, &len);3356 prepare_image(image, img, len, !patch->is_binary);3357 return 0;3358}33593360static int try_threeway(struct image *image, struct patch *patch,3361 struct stat *st, const struct cache_entry *ce)3362{3363 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];3364 struct strbuf buf = STRBUF_INIT;3365 size_t len;3366 int status;3367 char *img;3368 struct image tmp_image;33693370 /* No point falling back to 3-way merge in these cases */3371 if (patch->is_delete ||3372 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))3373 return -1;33743375 /* Preimage the patch was prepared for */3376 if (patch->is_new)3377 write_sha1_file("", 0, blob_type, pre_sha1);3378 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||3379 read_blob_object(&buf, pre_sha1, patch->old_mode))3380 return error("repository lacks the necessary blob to fall back on 3-way merge.");33813382 fprintf(stderr, "Falling back to three-way merge...\n");33833384 img = strbuf_detach(&buf, &len);3385 prepare_image(&tmp_image, img, len, 1);3386 /* Apply the patch to get the post image */3387 if (apply_fragments(&tmp_image, patch) < 0) {3388 clear_image(&tmp_image);3389 return -1;3390 }3391 /* post_sha1[] is theirs */3392 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);3393 clear_image(&tmp_image);33943395 /* our_sha1[] is ours */3396 if (patch->is_new) {3397 if (load_current(&tmp_image, patch))3398 return error("cannot read the current contents of '%s'",3399 patch->new_name);3400 } else {3401 if (load_preimage(&tmp_image, patch, st, ce))3402 return error("cannot read the current contents of '%s'",3403 patch->old_name);3404 }3405 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);3406 clear_image(&tmp_image);34073408 /* in-core three-way merge between post and our using pre as base */3409 status = three_way_merge(image, patch->new_name,3410 pre_sha1, our_sha1, post_sha1);3411 if (status < 0) {3412 fprintf(stderr, "Failed to fall back on three-way merge...\n");3413 return status;3414 }34153416 if (status) {3417 patch->conflicted_threeway = 1;3418 if (patch->is_new)3419 oidclr(&patch->threeway_stage[0]);3420 else3421 hashcpy(patch->threeway_stage[0].hash, pre_sha1);3422 hashcpy(patch->threeway_stage[1].hash, our_sha1);3423 hashcpy(patch->threeway_stage[2].hash, post_sha1);3424 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);3425 } else {3426 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);3427 }3428 return 0;3429}34303431static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)3432{3433 struct image image;34343435 if (load_preimage(&image, patch, st, ce) < 0)3436 return -1;34373438 if (patch->direct_to_threeway ||3439 apply_fragments(&image, patch) < 0) {3440 /* Note: with --reject, apply_fragments() returns 0 */3441 if (!threeway || try_threeway(&image, patch, st, ce) < 0)3442 return -1;3443 }3444 patch->result = image.buf;3445 patch->resultsize = image.len;3446 add_to_fn_table(patch);3447 free(image.line_allocated);34483449 if (0 < patch->is_delete && patch->resultsize)3450 return error(_("removal patch leaves file contents"));34513452 return 0;3453}34543455/*3456 * If "patch" that we are looking at modifies or deletes what we have,3457 * we would want it not to lose any local modification we have, either3458 * in the working tree or in the index.3459 *3460 * This also decides if a non-git patch is a creation patch or a3461 * modification to an existing empty file. We do not check the state3462 * of the current tree for a creation patch in this function; the caller3463 * check_patch() separately makes sure (and errors out otherwise) that3464 * the path the patch creates does not exist in the current tree.3465 */3466static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)3467{3468 const char *old_name = patch->old_name;3469 struct patch *previous = NULL;3470 int stat_ret = 0, status;3471 unsigned st_mode = 0;34723473 if (!old_name)3474 return 0;34753476 assert(patch->is_new <= 0);3477 previous = previous_patch(patch, &status);34783479 if (status)3480 return error(_("path %s has been renamed/deleted"), old_name);3481 if (previous) {3482 st_mode = previous->new_mode;3483 } else if (!cached) {3484 stat_ret = lstat(old_name, st);3485 if (stat_ret && errno != ENOENT)3486 return error(_("%s: %s"), old_name, strerror(errno));3487 }34883489 if (check_index && !previous) {3490 int pos = cache_name_pos(old_name, strlen(old_name));3491 if (pos < 0) {3492 if (patch->is_new < 0)3493 goto is_new;3494 return error(_("%s: does not exist in index"), old_name);3495 }3496 *ce = active_cache[pos];3497 if (stat_ret < 0) {3498 if (checkout_target(&the_index, *ce, st))3499 return -1;3500 }3501 if (!cached && verify_index_match(*ce, st))3502 return error(_("%s: does not match index"), old_name);3503 if (cached)3504 st_mode = (*ce)->ce_mode;3505 } else if (stat_ret < 0) {3506 if (patch->is_new < 0)3507 goto is_new;3508 return error(_("%s: %s"), old_name, strerror(errno));3509 }35103511 if (!cached && !previous)3512 st_mode = ce_mode_from_stat(*ce, st->st_mode);35133514 if (patch->is_new < 0)3515 patch->is_new = 0;3516 if (!patch->old_mode)3517 patch->old_mode = st_mode;3518 if ((st_mode ^ patch->old_mode) & S_IFMT)3519 return error(_("%s: wrong type"), old_name);3520 if (st_mode != patch->old_mode)3521 warning(_("%s has type %o, expected %o"),3522 old_name, st_mode, patch->old_mode);3523 if (!patch->new_mode && !patch->is_delete)3524 patch->new_mode = st_mode;3525 return 0;35263527 is_new:3528 patch->is_new = 1;3529 patch->is_delete = 0;3530 free(patch->old_name);3531 patch->old_name = NULL;3532 return 0;3533}353435353536#define EXISTS_IN_INDEX 13537#define EXISTS_IN_WORKTREE 235383539static int check_to_create(const char *new_name, int ok_if_exists)3540{3541 struct stat nst;35423543 if (check_index &&3544 cache_name_pos(new_name, strlen(new_name)) >= 0 &&3545 !ok_if_exists)3546 return EXISTS_IN_INDEX;3547 if (cached)3548 return 0;35493550 if (!lstat(new_name, &nst)) {3551 if (S_ISDIR(nst.st_mode) || ok_if_exists)3552 return 0;3553 /*3554 * A leading component of new_name might be a symlink3555 * that is going to be removed with this patch, but3556 * still pointing at somewhere that has the path.3557 * In such a case, path "new_name" does not exist as3558 * far as git is concerned.3559 */3560 if (has_symlink_leading_path(new_name, strlen(new_name)))3561 return 0;35623563 return EXISTS_IN_WORKTREE;3564 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {3565 return error("%s: %s", new_name, strerror(errno));3566 }3567 return 0;3568}35693570/*3571 * We need to keep track of how symlinks in the preimage are3572 * manipulated by the patches. A patch to add a/b/c where a/b3573 * is a symlink should not be allowed to affect the directory3574 * the symlink points at, but if the same patch removes a/b,3575 * it is perfectly fine, as the patch removes a/b to make room3576 * to create a directory a/b so that a/b/c can be created.3577 */3578static struct string_list symlink_changes;3579#define SYMLINK_GOES_AWAY 013580#define SYMLINK_IN_RESULT 0235813582static uintptr_t register_symlink_changes(const char *path, uintptr_t what)3583{3584 struct string_list_item *ent;35853586 ent = string_list_lookup(&symlink_changes, path);3587 if (!ent) {3588 ent = string_list_insert(&symlink_changes, path);3589 ent->util = (void *)0;3590 }3591 ent->util = (void *)(what | ((uintptr_t)ent->util));3592 return (uintptr_t)ent->util;3593}35943595static uintptr_t check_symlink_changes(const char *path)3596{3597 struct string_list_item *ent;35983599 ent = string_list_lookup(&symlink_changes, path);3600 if (!ent)3601 return 0;3602 return (uintptr_t)ent->util;3603}36043605static void prepare_symlink_changes(struct patch *patch)3606{3607 for ( ; patch; patch = patch->next) {3608 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&3609 (patch->is_rename || patch->is_delete))3610 /* the symlink at patch->old_name is removed */3611 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);36123613 if (patch->new_name && S_ISLNK(patch->new_mode))3614 /* the symlink at patch->new_name is created or remains */3615 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);3616 }3617}36183619static int path_is_beyond_symlink_1(struct strbuf *name)3620{3621 do {3622 unsigned int change;36233624 while (--name->len && name->buf[name->len] != '/')3625 ; /* scan backwards */3626 if (!name->len)3627 break;3628 name->buf[name->len] = '\0';3629 change = check_symlink_changes(name->buf);3630 if (change & SYMLINK_IN_RESULT)3631 return 1;3632 if (change & SYMLINK_GOES_AWAY)3633 /*3634 * This cannot be "return 0", because we may3635 * see a new one created at a higher level.3636 */3637 continue;36383639 /* otherwise, check the preimage */3640 if (check_index) {3641 struct cache_entry *ce;36423643 ce = cache_file_exists(name->buf, name->len, ignore_case);3644 if (ce && S_ISLNK(ce->ce_mode))3645 return 1;3646 } else {3647 struct stat st;3648 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))3649 return 1;3650 }3651 } while (1);3652 return 0;3653}36543655static int path_is_beyond_symlink(const char *name_)3656{3657 int ret;3658 struct strbuf name = STRBUF_INIT;36593660 assert(*name_ != '\0');3661 strbuf_addstr(&name, name_);3662 ret = path_is_beyond_symlink_1(&name);3663 strbuf_release(&name);36643665 return ret;3666}36673668static void die_on_unsafe_path(struct patch *patch)3669{3670 const char *old_name = NULL;3671 const char *new_name = NULL;3672 if (patch->is_delete)3673 old_name = patch->old_name;3674 else if (!patch->is_new && !patch->is_copy)3675 old_name = patch->old_name;3676 if (!patch->is_delete)3677 new_name = patch->new_name;36783679 if (old_name && !verify_path(old_name))3680 die(_("invalid path '%s'"), old_name);3681 if (new_name && !verify_path(new_name))3682 die(_("invalid path '%s'"), new_name);3683}36843685/*3686 * Check and apply the patch in-core; leave the result in patch->result3687 * for the caller to write it out to the final destination.3688 */3689static int check_patch(struct patch *patch)3690{3691 struct stat st;3692 const char *old_name = patch->old_name;3693 const char *new_name = patch->new_name;3694 const char *name = old_name ? old_name : new_name;3695 struct cache_entry *ce = NULL;3696 struct patch *tpatch;3697 int ok_if_exists;3698 int status;36993700 patch->rejected = 1; /* we will drop this after we succeed */37013702 status = check_preimage(patch, &ce, &st);3703 if (status)3704 return status;3705 old_name = patch->old_name;37063707 /*3708 * A type-change diff is always split into a patch to delete3709 * old, immediately followed by a patch to create new (see3710 * diff.c::run_diff()); in such a case it is Ok that the entry3711 * to be deleted by the previous patch is still in the working3712 * tree and in the index.3713 *3714 * A patch to swap-rename between A and B would first rename A3715 * to B and then rename B to A. While applying the first one,3716 * the presence of B should not stop A from getting renamed to3717 * B; ask to_be_deleted() about the later rename. Removal of3718 * B and rename from A to B is handled the same way by asking3719 * was_deleted().3720 */3721 if ((tpatch = in_fn_table(new_name)) &&3722 (was_deleted(tpatch) || to_be_deleted(tpatch)))3723 ok_if_exists = 1;3724 else3725 ok_if_exists = 0;37263727 if (new_name &&3728 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {3729 int err = check_to_create(new_name, ok_if_exists);37303731 if (err && threeway) {3732 patch->direct_to_threeway = 1;3733 } else switch (err) {3734 case 0:3735 break; /* happy */3736 case EXISTS_IN_INDEX:3737 return error(_("%s: already exists in index"), new_name);3738 break;3739 case EXISTS_IN_WORKTREE:3740 return error(_("%s: already exists in working directory"),3741 new_name);3742 default:3743 return err;3744 }37453746 if (!patch->new_mode) {3747 if (0 < patch->is_new)3748 patch->new_mode = S_IFREG | 0644;3749 else3750 patch->new_mode = patch->old_mode;3751 }3752 }37533754 if (new_name && old_name) {3755 int same = !strcmp(old_name, new_name);3756 if (!patch->new_mode)3757 patch->new_mode = patch->old_mode;3758 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {3759 if (same)3760 return error(_("new mode (%o) of %s does not "3761 "match old mode (%o)"),3762 patch->new_mode, new_name,3763 patch->old_mode);3764 else3765 return error(_("new mode (%o) of %s does not "3766 "match old mode (%o) of %s"),3767 patch->new_mode, new_name,3768 patch->old_mode, old_name);3769 }3770 }37713772 if (!unsafe_paths)3773 die_on_unsafe_path(patch);37743775 /*3776 * An attempt to read from or delete a path that is beyond a3777 * symbolic link will be prevented by load_patch_target() that3778 * is called at the beginning of apply_data() so we do not3779 * have to worry about a patch marked with "is_delete" bit3780 * here. We however need to make sure that the patch result3781 * is not deposited to a path that is beyond a symbolic link3782 * here.3783 */3784 if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))3785 return error(_("affected file '%s' is beyond a symbolic link"),3786 patch->new_name);37873788 if (apply_data(patch, &st, ce) < 0)3789 return error(_("%s: patch does not apply"), name);3790 patch->rejected = 0;3791 return 0;3792}37933794static int check_patch_list(struct patch *patch)3795{3796 int err = 0;37973798 prepare_symlink_changes(patch);3799 prepare_fn_table(patch);3800 while (patch) {3801 if (apply_verbosely)3802 say_patch_name(stderr,3803 _("Checking patch %s..."), patch);3804 err |= check_patch(patch);3805 patch = patch->next;3806 }3807 return err;3808}38093810/* This function tries to read the sha1 from the current index */3811static int get_current_sha1(const char *path, unsigned char *sha1)3812{3813 int pos;38143815 if (read_cache() < 0)3816 return -1;3817 pos = cache_name_pos(path, strlen(path));3818 if (pos < 0)3819 return -1;3820 hashcpy(sha1, active_cache[pos]->sha1);3821 return 0;3822}38233824static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])3825{3826 /*3827 * A usable gitlink patch has only one fragment (hunk) that looks like:3828 * @@ -1 +1 @@3829 * -Subproject commit <old sha1>3830 * +Subproject commit <new sha1>3831 * or3832 * @@ -1 +0,0 @@3833 * -Subproject commit <old sha1>3834 * for a removal patch.3835 */3836 struct fragment *hunk = p->fragments;3837 static const char heading[] = "-Subproject commit ";3838 char *preimage;38393840 if (/* does the patch have only one hunk? */3841 hunk && !hunk->next &&3842 /* is its preimage one line? */3843 hunk->oldpos == 1 && hunk->oldlines == 1 &&3844 /* does preimage begin with the heading? */3845 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&3846 starts_with(++preimage, heading) &&3847 /* does it record full SHA-1? */3848 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&3849 preimage[sizeof(heading) + 40 - 1] == '\n' &&3850 /* does the abbreviated name on the index line agree with it? */3851 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))3852 return 0; /* it all looks fine */38533854 /* we may have full object name on the index line */3855 return get_sha1_hex(p->old_sha1_prefix, sha1);3856}38573858/* Build an index that contains the just the files needed for a 3way merge */3859static void build_fake_ancestor(struct patch *list, const char *filename)3860{3861 struct patch *patch;3862 struct index_state result = { NULL };3863 static struct lock_file lock;38643865 /* Once we start supporting the reverse patch, it may be3866 * worth showing the new sha1 prefix, but until then...3867 */3868 for (patch = list; patch; patch = patch->next) {3869 unsigned char sha1[20];3870 struct cache_entry *ce;3871 const char *name;38723873 name = patch->old_name ? patch->old_name : patch->new_name;3874 if (0 < patch->is_new)3875 continue;38763877 if (S_ISGITLINK(patch->old_mode)) {3878 if (!preimage_sha1_in_gitlink_patch(patch, sha1))3879 ; /* ok, the textual part looks sane */3880 else3881 die("sha1 information is lacking or useless for submodule %s",3882 name);3883 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {3884 ; /* ok */3885 } else if (!patch->lines_added && !patch->lines_deleted) {3886 /* mode-only change: update the current */3887 if (get_current_sha1(patch->old_name, sha1))3888 die("mode change for %s, which is not "3889 "in current HEAD", name);3890 } else3891 die("sha1 information is lacking or useless "3892 "(%s).", name);38933894 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);3895 if (!ce)3896 die(_("make_cache_entry failed for path '%s'"), name);3897 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3898 die ("Could not add %s to temporary index", name);3899 }39003901 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);3902 if (write_locked_index(&result, &lock, COMMIT_LOCK))3903 die ("Could not write temporary index to %s", filename);39043905 discard_index(&result);3906}39073908static void stat_patch_list(struct patch *patch)3909{3910 int files, adds, dels;39113912 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3913 files++;3914 adds += patch->lines_added;3915 dels += patch->lines_deleted;3916 show_stats(patch);3917 }39183919 print_stat_summary(stdout, files, adds, dels);3920}39213922static void numstat_patch_list(struct patch *patch)3923{3924 for ( ; patch; patch = patch->next) {3925 const char *name;3926 name = patch->new_name ? patch->new_name : patch->old_name;3927 if (patch->is_binary)3928 printf("-\t-\t");3929 else3930 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3931 write_name_quoted(name, stdout, line_termination);3932 }3933}39343935static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3936{3937 if (mode)3938 printf(" %s mode %06o %s\n", newdelete, mode, name);3939 else3940 printf(" %s %s\n", newdelete, name);3941}39423943static void show_mode_change(struct patch *p, int show_name)3944{3945 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3946 if (show_name)3947 printf(" mode change %06o => %06o %s\n",3948 p->old_mode, p->new_mode, p->new_name);3949 else3950 printf(" mode change %06o => %06o\n",3951 p->old_mode, p->new_mode);3952 }3953}39543955static void show_rename_copy(struct patch *p)3956{3957 const char *renamecopy = p->is_rename ? "rename" : "copy";3958 const char *old, *new;39593960 /* Find common prefix */3961 old = p->old_name;3962 new = p->new_name;3963 while (1) {3964 const char *slash_old, *slash_new;3965 slash_old = strchr(old, '/');3966 slash_new = strchr(new, '/');3967 if (!slash_old ||3968 !slash_new ||3969 slash_old - old != slash_new - new ||3970 memcmp(old, new, slash_new - new))3971 break;3972 old = slash_old + 1;3973 new = slash_new + 1;3974 }3975 /* p->old_name thru old is the common prefix, and old and new3976 * through the end of names are renames3977 */3978 if (old != p->old_name)3979 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,3980 (int)(old - p->old_name), p->old_name,3981 old, new, p->score);3982 else3983 printf(" %s %s => %s (%d%%)\n", renamecopy,3984 p->old_name, p->new_name, p->score);3985 show_mode_change(p, 0);3986}39873988static void summary_patch_list(struct patch *patch)3989{3990 struct patch *p;39913992 for (p = patch; p; p = p->next) {3993 if (p->is_new)3994 show_file_mode_name("create", p->new_mode, p->new_name);3995 else if (p->is_delete)3996 show_file_mode_name("delete", p->old_mode, p->old_name);3997 else {3998 if (p->is_rename || p->is_copy)3999 show_rename_copy(p);4000 else {4001 if (p->score) {4002 printf(" rewrite %s (%d%%)\n",4003 p->new_name, p->score);4004 show_mode_change(p, 0);4005 }4006 else4007 show_mode_change(p, 1);4008 }4009 }4010 }4011}40124013static void patch_stats(struct patch *patch)4014{4015 int lines = patch->lines_added + patch->lines_deleted;40164017 if (lines > max_change)4018 max_change = lines;4019 if (patch->old_name) {4020 int len = quote_c_style(patch->old_name, NULL, NULL, 0);4021 if (!len)4022 len = strlen(patch->old_name);4023 if (len > max_len)4024 max_len = len;4025 }4026 if (patch->new_name) {4027 int len = quote_c_style(patch->new_name, NULL, NULL, 0);4028 if (!len)4029 len = strlen(patch->new_name);4030 if (len > max_len)4031 max_len = len;4032 }4033}40344035static void remove_file(struct patch *patch, int rmdir_empty)4036{4037 if (update_index) {4038 if (remove_file_from_cache(patch->old_name) < 0)4039 die(_("unable to remove %s from index"), patch->old_name);4040 }4041 if (!cached) {4042 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {4043 remove_path(patch->old_name);4044 }4045 }4046}40474048static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)4049{4050 struct stat st;4051 struct cache_entry *ce;4052 int namelen = strlen(path);4053 unsigned ce_size = cache_entry_size(namelen);40544055 if (!update_index)4056 return;40574058 ce = xcalloc(1, ce_size);4059 memcpy(ce->name, path, namelen);4060 ce->ce_mode = create_ce_mode(mode);4061 ce->ce_flags = create_ce_flags(0);4062 ce->ce_namelen = namelen;4063 if (S_ISGITLINK(mode)) {4064 const char *s;40654066 if (!skip_prefix(buf, "Subproject commit ", &s) ||4067 get_sha1_hex(s, ce->sha1))4068 die(_("corrupt patch for submodule %s"), path);4069 } else {4070 if (!cached) {4071 if (lstat(path, &st) < 0)4072 die_errno(_("unable to stat newly created file '%s'"),4073 path);4074 fill_stat_cache_info(ce, &st);4075 }4076 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)4077 die(_("unable to create backing store for newly created file %s"), path);4078 }4079 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4080 die(_("unable to add cache entry for %s"), path);4081}40824083static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)4084{4085 int fd;4086 struct strbuf nbuf = STRBUF_INIT;40874088 if (S_ISGITLINK(mode)) {4089 struct stat st;4090 if (!lstat(path, &st) && S_ISDIR(st.st_mode))4091 return 0;4092 return mkdir(path, 0777);4093 }40944095 if (has_symlinks && S_ISLNK(mode))4096 /* Although buf:size is counted string, it also is NUL4097 * terminated.4098 */4099 return symlink(buf, path);41004101 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);4102 if (fd < 0)4103 return -1;41044105 if (convert_to_working_tree(path, buf, size, &nbuf)) {4106 size = nbuf.len;4107 buf = nbuf.buf;4108 }4109 write_or_die(fd, buf, size);4110 strbuf_release(&nbuf);41114112 if (close(fd) < 0)4113 die_errno(_("closing file '%s'"), path);4114 return 0;4115}41164117/*4118 * We optimistically assume that the directories exist,4119 * which is true 99% of the time anyway. If they don't,4120 * we create them and try again.4121 */4122static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)4123{4124 if (cached)4125 return;4126 if (!try_create_file(path, mode, buf, size))4127 return;41284129 if (errno == ENOENT) {4130 if (safe_create_leading_directories(path))4131 return;4132 if (!try_create_file(path, mode, buf, size))4133 return;4134 }41354136 if (errno == EEXIST || errno == EACCES) {4137 /* We may be trying to create a file where a directory4138 * used to be.4139 */4140 struct stat st;4141 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))4142 errno = EEXIST;4143 }41444145 if (errno == EEXIST) {4146 unsigned int nr = getpid();41474148 for (;;) {4149 char newpath[PATH_MAX];4150 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);4151 if (!try_create_file(newpath, mode, buf, size)) {4152 if (!rename(newpath, path))4153 return;4154 unlink_or_warn(newpath);4155 break;4156 }4157 if (errno != EEXIST)4158 break;4159 ++nr;4160 }4161 }4162 die_errno(_("unable to write file '%s' mode %o"), path, mode);4163}41644165static void add_conflicted_stages_file(struct patch *patch)4166{4167 int stage, namelen;4168 unsigned ce_size, mode;4169 struct cache_entry *ce;41704171 if (!update_index)4172 return;4173 namelen = strlen(patch->new_name);4174 ce_size = cache_entry_size(namelen);4175 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);41764177 remove_file_from_cache(patch->new_name);4178 for (stage = 1; stage < 4; stage++) {4179 if (is_null_oid(&patch->threeway_stage[stage - 1]))4180 continue;4181 ce = xcalloc(1, ce_size);4182 memcpy(ce->name, patch->new_name, namelen);4183 ce->ce_mode = create_ce_mode(mode);4184 ce->ce_flags = create_ce_flags(stage);4185 ce->ce_namelen = namelen;4186 hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);4187 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4188 die(_("unable to add cache entry for %s"), patch->new_name);4189 }4190}41914192static void create_file(struct patch *patch)4193{4194 char *path = patch->new_name;4195 unsigned mode = patch->new_mode;4196 unsigned long size = patch->resultsize;4197 char *buf = patch->result;41984199 if (!mode)4200 mode = S_IFREG | 0644;4201 create_one_file(path, mode, buf, size);42024203 if (patch->conflicted_threeway)4204 add_conflicted_stages_file(patch);4205 else4206 add_index_file(path, mode, buf, size);4207}42084209/* phase zero is to remove, phase one is to create */4210static void write_out_one_result(struct patch *patch, int phase)4211{4212 if (patch->is_delete > 0) {4213 if (phase == 0)4214 remove_file(patch, 1);4215 return;4216 }4217 if (patch->is_new > 0 || patch->is_copy) {4218 if (phase == 1)4219 create_file(patch);4220 return;4221 }4222 /*4223 * Rename or modification boils down to the same4224 * thing: remove the old, write the new4225 */4226 if (phase == 0)4227 remove_file(patch, patch->is_rename);4228 if (phase == 1)4229 create_file(patch);4230}42314232static int write_out_one_reject(struct patch *patch)4233{4234 FILE *rej;4235 char namebuf[PATH_MAX];4236 struct fragment *frag;4237 int cnt = 0;4238 struct strbuf sb = STRBUF_INIT;42394240 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {4241 if (!frag->rejected)4242 continue;4243 cnt++;4244 }42454246 if (!cnt) {4247 if (apply_verbosely)4248 say_patch_name(stderr,4249 _("Applied patch %s cleanly."), patch);4250 return 0;4251 }42524253 /* This should not happen, because a removal patch that leaves4254 * contents are marked "rejected" at the patch level.4255 */4256 if (!patch->new_name)4257 die(_("internal error"));42584259 /* Say this even without --verbose */4260 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",4261 "Applying patch %%s with %d rejects...",4262 cnt),4263 cnt);4264 say_patch_name(stderr, sb.buf, patch);4265 strbuf_release(&sb);42664267 cnt = strlen(patch->new_name);4268 if (ARRAY_SIZE(namebuf) <= cnt + 5) {4269 cnt = ARRAY_SIZE(namebuf) - 5;4270 warning(_("truncating .rej filename to %.*s.rej"),4271 cnt - 1, patch->new_name);4272 }4273 memcpy(namebuf, patch->new_name, cnt);4274 memcpy(namebuf + cnt, ".rej", 5);42754276 rej = fopen(namebuf, "w");4277 if (!rej)4278 return error(_("cannot open %s: %s"), namebuf, strerror(errno));42794280 /* Normal git tools never deal with .rej, so do not pretend4281 * this is a git patch by saying --git or giving extended4282 * headers. While at it, maybe please "kompare" that wants4283 * the trailing TAB and some garbage at the end of line ;-).4284 */4285 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",4286 patch->new_name, patch->new_name);4287 for (cnt = 1, frag = patch->fragments;4288 frag;4289 cnt++, frag = frag->next) {4290 if (!frag->rejected) {4291 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);4292 continue;4293 }4294 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);4295 fprintf(rej, "%.*s", frag->size, frag->patch);4296 if (frag->patch[frag->size-1] != '\n')4297 fputc('\n', rej);4298 }4299 fclose(rej);4300 return -1;4301}43024303static int write_out_results(struct patch *list)4304{4305 int phase;4306 int errs = 0;4307 struct patch *l;4308 struct string_list cpath = STRING_LIST_INIT_DUP;43094310 for (phase = 0; phase < 2; phase++) {4311 l = list;4312 while (l) {4313 if (l->rejected)4314 errs = 1;4315 else {4316 write_out_one_result(l, phase);4317 if (phase == 1) {4318 if (write_out_one_reject(l))4319 errs = 1;4320 if (l->conflicted_threeway) {4321 string_list_append(&cpath, l->new_name);4322 errs = 1;4323 }4324 }4325 }4326 l = l->next;4327 }4328 }43294330 if (cpath.nr) {4331 struct string_list_item *item;43324333 string_list_sort(&cpath);4334 for_each_string_list_item(item, &cpath)4335 fprintf(stderr, "U %s\n", item->string);4336 string_list_clear(&cpath, 0);43374338 rerere(0);4339 }43404341 return errs;4342}43434344static struct lock_file lock_file;43454346#define INACCURATE_EOF (1<<0)4347#define RECOUNT (1<<1)43484349static int apply_patch(int fd, const char *filename, int options)4350{4351 size_t offset;4352 struct strbuf buf = STRBUF_INIT; /* owns the patch text */4353 struct patch *list = NULL, **listp = &list;4354 int skipped_patch = 0;43554356 patch_input_file = filename;4357 read_patch_file(&buf, fd);4358 offset = 0;4359 while (offset < buf.len) {4360 struct patch *patch;4361 int nr;43624363 patch = xcalloc(1, sizeof(*patch));4364 patch->inaccurate_eof = !!(options & INACCURATE_EOF);4365 patch->recount = !!(options & RECOUNT);4366 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);4367 if (nr < 0)4368 break;4369 if (apply_in_reverse)4370 reverse_patches(patch);4371 if (use_patch(patch)) {4372 patch_stats(patch);4373 *listp = patch;4374 listp = &patch->next;4375 }4376 else {4377 free_patch(patch);4378 skipped_patch++;4379 }4380 offset += nr;4381 }43824383 if (!list && !skipped_patch)4384 die(_("unrecognized input"));43854386 if (whitespace_error && (ws_error_action == die_on_ws_error))4387 apply = 0;43884389 update_index = check_index && apply;4390 if (update_index && newfd < 0)4391 newfd = hold_locked_index(&lock_file, 1);43924393 if (check_index) {4394 if (read_cache() < 0)4395 die(_("unable to read index file"));4396 }43974398 if ((check || apply) &&4399 check_patch_list(list) < 0 &&4400 !apply_with_reject)4401 exit(1);44024403 if (apply && write_out_results(list)) {4404 if (apply_with_reject)4405 exit(1);4406 /* with --3way, we still need to write the index out */4407 return 1;4408 }44094410 if (fake_ancestor)4411 build_fake_ancestor(list, fake_ancestor);44124413 if (diffstat)4414 stat_patch_list(list);44154416 if (numstat)4417 numstat_patch_list(list);44184419 if (summary)4420 summary_patch_list(list);44214422 free_patch_list(list);4423 strbuf_release(&buf);4424 string_list_clear(&fn_table, 0);4425 return 0;4426}44274428static void git_apply_config(void)4429{4430 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);4431 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);4432 git_config(git_default_config, NULL);4433}44344435static int option_parse_exclude(const struct option *opt,4436 const char *arg, int unset)4437{4438 add_name_limit(arg, 1);4439 return 0;4440}44414442static int option_parse_include(const struct option *opt,4443 const char *arg, int unset)4444{4445 add_name_limit(arg, 0);4446 has_include = 1;4447 return 0;4448}44494450static int option_parse_p(const struct option *opt,4451 const char *arg, int unset)4452{4453 p_value = atoi(arg);4454 p_value_known = 1;4455 return 0;4456}44574458static int option_parse_space_change(const struct option *opt,4459 const char *arg, int unset)4460{4461 if (unset)4462 ws_ignore_action = ignore_ws_none;4463 else4464 ws_ignore_action = ignore_ws_change;4465 return 0;4466}44674468static int option_parse_whitespace(const struct option *opt,4469 const char *arg, int unset)4470{4471 const char **whitespace_option = opt->value;44724473 *whitespace_option = arg;4474 parse_whitespace_option(arg);4475 return 0;4476}44774478static int option_parse_directory(const struct option *opt,4479 const char *arg, int unset)4480{4481 strbuf_reset(&root);4482 strbuf_addstr(&root, arg);4483 strbuf_complete(&root, '/');4484 return 0;4485}44864487int cmd_apply(int argc, const char **argv, const char *prefix_)4488{4489 int i;4490 int errs = 0;4491 int is_not_gitdir = !startup_info->have_repository;4492 int force_apply = 0;44934494 const char *whitespace_option = NULL;44954496 struct option builtin_apply_options[] = {4497 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),4498 N_("don't apply changes matching the given path"),4499 0, option_parse_exclude },4500 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),4501 N_("apply changes matching the given path"),4502 0, option_parse_include },4503 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),4504 N_("remove <num> leading slashes from traditional diff paths"),4505 0, option_parse_p },4506 OPT_BOOL(0, "no-add", &no_add,4507 N_("ignore additions made by the patch")),4508 OPT_BOOL(0, "stat", &diffstat,4509 N_("instead of applying the patch, output diffstat for the input")),4510 OPT_NOOP_NOARG(0, "allow-binary-replacement"),4511 OPT_NOOP_NOARG(0, "binary"),4512 OPT_BOOL(0, "numstat", &numstat,4513 N_("show number of added and deleted lines in decimal notation")),4514 OPT_BOOL(0, "summary", &summary,4515 N_("instead of applying the patch, output a summary for the input")),4516 OPT_BOOL(0, "check", &check,4517 N_("instead of applying the patch, see if the patch is applicable")),4518 OPT_BOOL(0, "index", &check_index,4519 N_("make sure the patch is applicable to the current index")),4520 OPT_BOOL(0, "cached", &cached,4521 N_("apply a patch without touching the working tree")),4522 OPT_BOOL(0, "unsafe-paths", &unsafe_paths,4523 N_("accept a patch that touches outside the working area")),4524 OPT_BOOL(0, "apply", &force_apply,4525 N_("also apply the patch (use with --stat/--summary/--check)")),4526 OPT_BOOL('3', "3way", &threeway,4527 N_( "attempt three-way merge if a patch does not apply")),4528 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,4529 N_("build a temporary index based on embedded index information")),4530 /* Think twice before adding "--nul" synonym to this */4531 OPT_SET_INT('z', NULL, &line_termination,4532 N_("paths are separated with NUL character"), '\0'),4533 OPT_INTEGER('C', NULL, &p_context,4534 N_("ensure at least <n> lines of context match")),4535 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),4536 N_("detect new or modified lines that have whitespace errors"),4537 0, option_parse_whitespace },4538 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,4539 N_("ignore changes in whitespace when finding context"),4540 PARSE_OPT_NOARG, option_parse_space_change },4541 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,4542 N_("ignore changes in whitespace when finding context"),4543 PARSE_OPT_NOARG, option_parse_space_change },4544 OPT_BOOL('R', "reverse", &apply_in_reverse,4545 N_("apply the patch in reverse")),4546 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,4547 N_("don't expect at least one line of context")),4548 OPT_BOOL(0, "reject", &apply_with_reject,4549 N_("leave the rejected hunks in corresponding *.rej files")),4550 OPT_BOOL(0, "allow-overlap", &allow_overlap,4551 N_("allow overlapping hunks")),4552 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),4553 OPT_BIT(0, "inaccurate-eof", &options,4554 N_("tolerate incorrectly detected missing new-line at the end of file"),4555 INACCURATE_EOF),4556 OPT_BIT(0, "recount", &options,4557 N_("do not trust the line counts in the hunk headers"),4558 RECOUNT),4559 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),4560 N_("prepend <root> to all filenames"),4561 0, option_parse_directory },4562 OPT_END()4563 };45644565 prefix = prefix_;4566 prefix_length = prefix ? strlen(prefix) : 0;4567 git_apply_config();4568 if (apply_default_whitespace)4569 parse_whitespace_option(apply_default_whitespace);4570 if (apply_default_ignorewhitespace)4571 parse_ignorewhitespace_option(apply_default_ignorewhitespace);45724573 argc = parse_options(argc, argv, prefix, builtin_apply_options,4574 apply_usage, 0);45754576 if (apply_with_reject && threeway)4577 die("--reject and --3way cannot be used together.");4578 if (cached && threeway)4579 die("--cached and --3way cannot be used together.");4580 if (threeway) {4581 if (is_not_gitdir)4582 die(_("--3way outside a repository"));4583 check_index = 1;4584 }4585 if (apply_with_reject)4586 apply = apply_verbosely = 1;4587 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))4588 apply = 0;4589 if (check_index && is_not_gitdir)4590 die(_("--index outside a repository"));4591 if (cached) {4592 if (is_not_gitdir)4593 die(_("--cached outside a repository"));4594 check_index = 1;4595 }4596 if (check_index)4597 unsafe_paths = 0;45984599 for (i = 0; i < argc; i++) {4600 const char *arg = argv[i];4601 int fd;46024603 if (!strcmp(arg, "-")) {4604 errs |= apply_patch(0, "<stdin>", options);4605 read_stdin = 0;4606 continue;4607 } else if (0 < prefix_length)4608 arg = prefix_filename(prefix, prefix_length, arg);46094610 fd = open(arg, O_RDONLY);4611 if (fd < 0)4612 die_errno(_("can't open patch '%s'"), arg);4613 read_stdin = 0;4614 set_default_whitespace_mode(whitespace_option);4615 errs |= apply_patch(fd, arg, options);4616 close(fd);4617 }4618 set_default_whitespace_mode(whitespace_option);4619 if (read_stdin)4620 errs |= apply_patch(0, "<stdin>", options);4621 if (whitespace_error) {4622 if (squelch_whitespace_errors &&4623 squelch_whitespace_errors < whitespace_error) {4624 int squelched =4625 whitespace_error - squelch_whitespace_errors;4626 warning(Q_("squelched %d whitespace error",4627 "squelched %d whitespace errors",4628 squelched),4629 squelched);4630 }4631 if (ws_error_action == die_on_ws_error)4632 die(Q_("%d line adds whitespace errors.",4633 "%d lines add whitespace errors.",4634 whitespace_error),4635 whitespace_error);4636 if (applied_after_fixing_ws && apply)4637 warning("%d line%s applied after"4638 " fixing whitespace errors.",4639 applied_after_fixing_ws,4640 applied_after_fixing_ws == 1 ? "" : "s");4641 else if (whitespace_error)4642 warning(Q_("%d line adds whitespace errors.",4643 "%d lines add whitespace errors.",4644 whitespace_error),4645 whitespace_error);4646 }46474648 if (update_index) {4649 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))4650 die(_("Unable to write new index file"));4651 }46524653 return !!errs;4654}