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 const char *root; 81static int root_len; 82static int read_stdin = 1; 83static int options; 84 85static void parse_whitespace_option(const char *option) 86{ 87 if (!option) { 88 ws_error_action = warn_on_ws_error; 89 return; 90 } 91 if (!strcmp(option, "warn")) { 92 ws_error_action = warn_on_ws_error; 93 return; 94 } 95 if (!strcmp(option, "nowarn")) { 96 ws_error_action = nowarn_ws_error; 97 return; 98 } 99 if (!strcmp(option, "error")) { 100 ws_error_action = die_on_ws_error; 101 return; 102 } 103 if (!strcmp(option, "error-all")) { 104 ws_error_action = die_on_ws_error; 105 squelch_whitespace_errors = 0; 106 return; 107 } 108 if (!strcmp(option, "strip") || !strcmp(option, "fix")) { 109 ws_error_action = correct_ws_error; 110 return; 111 } 112 die(_("unrecognized whitespace option '%s'"), option); 113} 114 115static void parse_ignorewhitespace_option(const char *option) 116{ 117 if (!option || !strcmp(option, "no") || 118 !strcmp(option, "false") || !strcmp(option, "never") || 119 !strcmp(option, "none")) { 120 ws_ignore_action = ignore_ws_none; 121 return; 122 } 123 if (!strcmp(option, "change")) { 124 ws_ignore_action = ignore_ws_change; 125 return; 126 } 127 die(_("unrecognized whitespace ignore option '%s'"), option); 128} 129 130static void set_default_whitespace_mode(const char *whitespace_option) 131{ 132 if (!whitespace_option && !apply_default_whitespace) 133 ws_error_action = (apply ? warn_on_ws_error : nowarn_ws_error); 134} 135 136/* 137 * For "diff-stat" like behaviour, we keep track of the biggest change 138 * we've seen, and the longest filename. That allows us to do simple 139 * scaling. 140 */ 141static int max_change, max_len; 142 143/* 144 * Various "current state", notably line numbers and what 145 * file (and how) we're patching right now.. The "is_xxxx" 146 * things are flags, where -1 means "don't know yet". 147 */ 148static int linenr = 1; 149 150/* 151 * This represents one "hunk" from a patch, starting with 152 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The 153 * patch text is pointed at by patch, and its byte length 154 * is stored in size. leading and trailing are the number 155 * of context lines. 156 */ 157struct fragment { 158 unsigned long leading, trailing; 159 unsigned long oldpos, oldlines; 160 unsigned long newpos, newlines; 161 /* 162 * 'patch' is usually borrowed from buf in apply_patch(), 163 * but some codepaths store an allocated buffer. 164 */ 165 const char *patch; 166 unsigned free_patch:1, 167 rejected:1; 168 int size; 169 int linenr; 170 struct fragment *next; 171}; 172 173/* 174 * When dealing with a binary patch, we reuse "leading" field 175 * to store the type of the binary hunk, either deflated "delta" 176 * or deflated "literal". 177 */ 178#define binary_patch_method leading 179#define BINARY_DELTA_DEFLATED 1 180#define BINARY_LITERAL_DEFLATED 2 181 182/* 183 * This represents a "patch" to a file, both metainfo changes 184 * such as creation/deletion, filemode and content changes represented 185 * as a series of fragments. 186 */ 187struct patch { 188 char *new_name, *old_name, *def_name; 189 unsigned int old_mode, new_mode; 190 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */ 191 int rejected; 192 unsigned ws_rule; 193 int lines_added, lines_deleted; 194 int score; 195 unsigned int is_toplevel_relative:1; 196 unsigned int inaccurate_eof:1; 197 unsigned int is_binary:1; 198 unsigned int is_copy:1; 199 unsigned int is_rename:1; 200 unsigned int recount:1; 201 unsigned int conflicted_threeway:1; 202 unsigned int direct_to_threeway:1; 203 struct fragment *fragments; 204 char *result; 205 size_t resultsize; 206 char old_sha1_prefix[41]; 207 char new_sha1_prefix[41]; 208 struct patch *next; 209 210 /* three-way fallback result */ 211 unsigned char threeway_stage[3][20]; 212}; 213 214static void free_fragment_list(struct fragment *list) 215{ 216 while (list) { 217 struct fragment *next = list->next; 218 if (list->free_patch) 219 free((char *)list->patch); 220 free(list); 221 list = next; 222 } 223} 224 225static void free_patch(struct patch *patch) 226{ 227 free_fragment_list(patch->fragments); 228 free(patch->def_name); 229 free(patch->old_name); 230 free(patch->new_name); 231 free(patch->result); 232 free(patch); 233} 234 235static void free_patch_list(struct patch *list) 236{ 237 while (list) { 238 struct patch *next = list->next; 239 free_patch(list); 240 list = next; 241 } 242} 243 244/* 245 * A line in a file, len-bytes long (includes the terminating LF, 246 * except for an incomplete line at the end if the file ends with 247 * one), and its contents hashes to 'hash'. 248 */ 249struct line { 250 size_t len; 251 unsigned hash : 24; 252 unsigned flag : 8; 253#define LINE_COMMON 1 254#define LINE_PATCHED 2 255}; 256 257/* 258 * This represents a "file", which is an array of "lines". 259 */ 260struct image { 261 char *buf; 262 size_t len; 263 size_t nr; 264 size_t alloc; 265 struct line *line_allocated; 266 struct line *line; 267}; 268 269/* 270 * Records filenames that have been touched, in order to handle 271 * the case where more than one patches touch the same file. 272 */ 273 274static struct string_list fn_table; 275 276static uint32_t hash_line(const char *cp, size_t len) 277{ 278 size_t i; 279 uint32_t h; 280 for (i = 0, h = 0; i < len; i++) { 281 if (!isspace(cp[i])) { 282 h = h * 3 + (cp[i] & 0xff); 283 } 284 } 285 return h; 286} 287 288/* 289 * Compare lines s1 of length n1 and s2 of length n2, ignoring 290 * whitespace difference. Returns 1 if they match, 0 otherwise 291 */ 292static int fuzzy_matchlines(const char *s1, size_t n1, 293 const char *s2, size_t n2) 294{ 295 const char *last1 = s1 + n1 - 1; 296 const char *last2 = s2 + n2 - 1; 297 int result = 0; 298 299 /* ignore line endings */ 300 while ((*last1 == '\r') || (*last1 == '\n')) 301 last1--; 302 while ((*last2 == '\r') || (*last2 == '\n')) 303 last2--; 304 305 /* skip leading whitespaces, if both begin with whitespace */ 306 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) { 307 while (isspace(*s1) && (s1 <= last1)) 308 s1++; 309 while (isspace(*s2) && (s2 <= last2)) 310 s2++; 311 } 312 /* early return if both lines are empty */ 313 if ((s1 > last1) && (s2 > last2)) 314 return 1; 315 while (!result) { 316 result = *s1++ - *s2++; 317 /* 318 * Skip whitespace inside. We check for whitespace on 319 * both buffers because we don't want "a b" to match 320 * "ab" 321 */ 322 if (isspace(*s1) && isspace(*s2)) { 323 while (isspace(*s1) && s1 <= last1) 324 s1++; 325 while (isspace(*s2) && s2 <= last2) 326 s2++; 327 } 328 /* 329 * If we reached the end on one side only, 330 * lines don't match 331 */ 332 if ( 333 ((s2 > last2) && (s1 <= last1)) || 334 ((s1 > last1) && (s2 <= last2))) 335 return 0; 336 if ((s1 > last1) && (s2 > last2)) 337 break; 338 } 339 340 return !result; 341} 342 343static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag) 344{ 345 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc); 346 img->line_allocated[img->nr].len = len; 347 img->line_allocated[img->nr].hash = hash_line(bol, len); 348 img->line_allocated[img->nr].flag = flag; 349 img->nr++; 350} 351 352/* 353 * "buf" has the file contents to be patched (read from various sources). 354 * attach it to "image" and add line-based index to it. 355 * "image" now owns the "buf". 356 */ 357static void prepare_image(struct image *image, char *buf, size_t len, 358 int prepare_linetable) 359{ 360 const char *cp, *ep; 361 362 memset(image, 0, sizeof(*image)); 363 image->buf = buf; 364 image->len = len; 365 366 if (!prepare_linetable) 367 return; 368 369 ep = image->buf + image->len; 370 cp = image->buf; 371 while (cp < ep) { 372 const char *next; 373 for (next = cp; next < ep && *next != '\n'; next++) 374 ; 375 if (next < ep) 376 next++; 377 add_line_info(image, cp, next - cp, 0); 378 cp = next; 379 } 380 image->line = image->line_allocated; 381} 382 383static void clear_image(struct image *image) 384{ 385 free(image->buf); 386 free(image->line_allocated); 387 memset(image, 0, sizeof(*image)); 388} 389 390/* fmt must contain _one_ %s and no other substitution */ 391static void say_patch_name(FILE *output, const char *fmt, struct patch *patch) 392{ 393 struct strbuf sb = STRBUF_INIT; 394 395 if (patch->old_name && patch->new_name && 396 strcmp(patch->old_name, patch->new_name)) { 397 quote_c_style(patch->old_name, &sb, NULL, 0); 398 strbuf_addstr(&sb, " => "); 399 quote_c_style(patch->new_name, &sb, NULL, 0); 400 } else { 401 const char *n = patch->new_name; 402 if (!n) 403 n = patch->old_name; 404 quote_c_style(n, &sb, NULL, 0); 405 } 406 fprintf(output, fmt, sb.buf); 407 fputc('\n', output); 408 strbuf_release(&sb); 409} 410 411#define SLOP (16) 412 413static void read_patch_file(struct strbuf *sb, int fd) 414{ 415 if (strbuf_read(sb, fd, 0) < 0) 416 die_errno("git apply: failed to read"); 417 418 /* 419 * Make sure that we have some slop in the buffer 420 * so that we can do speculative "memcmp" etc, and 421 * see to it that it is NUL-filled. 422 */ 423 strbuf_grow(sb, SLOP); 424 memset(sb->buf + sb->len, 0, SLOP); 425} 426 427static unsigned long linelen(const char *buffer, unsigned long size) 428{ 429 unsigned long len = 0; 430 while (size--) { 431 len++; 432 if (*buffer++ == '\n') 433 break; 434 } 435 return len; 436} 437 438static int is_dev_null(const char *str) 439{ 440 return skip_prefix(str, "/dev/null", &str) && isspace(*str); 441} 442 443#define TERM_SPACE 1 444#define TERM_TAB 2 445 446static int name_terminate(const char *name, int namelen, int c, int terminate) 447{ 448 if (c == ' ' && !(terminate & TERM_SPACE)) 449 return 0; 450 if (c == '\t' && !(terminate & TERM_TAB)) 451 return 0; 452 453 return 1; 454} 455 456/* remove double slashes to make --index work with such filenames */ 457static char *squash_slash(char *name) 458{ 459 int i = 0, j = 0; 460 461 if (!name) 462 return NULL; 463 464 while (name[i]) { 465 if ((name[j++] = name[i++]) == '/') 466 while (name[i] == '/') 467 i++; 468 } 469 name[j] = '\0'; 470 return name; 471} 472 473static char *find_name_gnu(const char *line, const char *def, int p_value) 474{ 475 struct strbuf name = STRBUF_INIT; 476 char *cp; 477 478 /* 479 * Proposed "new-style" GNU patch/diff format; see 480 * http://marc.info/?l=git&m=112927316408690&w=2 481 */ 482 if (unquote_c_style(&name, line, NULL)) { 483 strbuf_release(&name); 484 return NULL; 485 } 486 487 for (cp = name.buf; p_value; p_value--) { 488 cp = strchr(cp, '/'); 489 if (!cp) { 490 strbuf_release(&name); 491 return NULL; 492 } 493 cp++; 494 } 495 496 strbuf_remove(&name, 0, cp - name.buf); 497 if (root) 498 strbuf_insert(&name, 0, root, root_len); 499 return squash_slash(strbuf_detach(&name, NULL)); 500} 501 502static size_t sane_tz_len(const char *line, size_t len) 503{ 504 const char *tz, *p; 505 506 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ') 507 return 0; 508 tz = line + len - strlen(" +0500"); 509 510 if (tz[1] != '+' && tz[1] != '-') 511 return 0; 512 513 for (p = tz + 2; p != line + len; p++) 514 if (!isdigit(*p)) 515 return 0; 516 517 return line + len - tz; 518} 519 520static size_t tz_with_colon_len(const char *line, size_t len) 521{ 522 const char *tz, *p; 523 524 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':') 525 return 0; 526 tz = line + len - strlen(" +08:00"); 527 528 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-')) 529 return 0; 530 p = tz + 2; 531 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 532 !isdigit(*p++) || !isdigit(*p++)) 533 return 0; 534 535 return line + len - tz; 536} 537 538static size_t date_len(const char *line, size_t len) 539{ 540 const char *date, *p; 541 542 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-') 543 return 0; 544 p = date = line + len - strlen("72-02-05"); 545 546 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 547 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 548 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */ 549 return 0; 550 551 if (date - line >= strlen("19") && 552 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */ 553 date -= strlen("19"); 554 555 return line + len - date; 556} 557 558static size_t short_time_len(const char *line, size_t len) 559{ 560 const char *time, *p; 561 562 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':') 563 return 0; 564 p = time = line + len - strlen(" 07:01:32"); 565 566 /* Permit 1-digit hours? */ 567 if (*p++ != ' ' || 568 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 569 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 570 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */ 571 return 0; 572 573 return line + len - time; 574} 575 576static size_t fractional_time_len(const char *line, size_t len) 577{ 578 const char *p; 579 size_t n; 580 581 /* Expected format: 19:41:17.620000023 */ 582 if (!len || !isdigit(line[len - 1])) 583 return 0; 584 p = line + len - 1; 585 586 /* Fractional seconds. */ 587 while (p > line && isdigit(*p)) 588 p--; 589 if (*p != '.') 590 return 0; 591 592 /* Hours, minutes, and whole seconds. */ 593 n = short_time_len(line, p - line); 594 if (!n) 595 return 0; 596 597 return line + len - p + n; 598} 599 600static size_t trailing_spaces_len(const char *line, size_t len) 601{ 602 const char *p; 603 604 /* Expected format: ' ' x (1 or more) */ 605 if (!len || line[len - 1] != ' ') 606 return 0; 607 608 p = line + len; 609 while (p != line) { 610 p--; 611 if (*p != ' ') 612 return line + len - (p + 1); 613 } 614 615 /* All spaces! */ 616 return len; 617} 618 619static size_t diff_timestamp_len(const char *line, size_t len) 620{ 621 const char *end = line + len; 622 size_t n; 623 624 /* 625 * Posix: 2010-07-05 19:41:17 626 * GNU: 2010-07-05 19:41:17.620000023 -0500 627 */ 628 629 if (!isdigit(end[-1])) 630 return 0; 631 632 n = sane_tz_len(line, end - line); 633 if (!n) 634 n = tz_with_colon_len(line, end - line); 635 end -= n; 636 637 n = short_time_len(line, end - line); 638 if (!n) 639 n = fractional_time_len(line, end - line); 640 end -= n; 641 642 n = date_len(line, end - line); 643 if (!n) /* No date. Too bad. */ 644 return 0; 645 end -= n; 646 647 if (end == line) /* No space before date. */ 648 return 0; 649 if (end[-1] == '\t') { /* Success! */ 650 end--; 651 return line + len - end; 652 } 653 if (end[-1] != ' ') /* No space before date. */ 654 return 0; 655 656 /* Whitespace damage. */ 657 end -= trailing_spaces_len(line, end - line); 658 return line + len - end; 659} 660 661static char *find_name_common(const char *line, const char *def, 662 int p_value, const char *end, int terminate) 663{ 664 int len; 665 const char *start = NULL; 666 667 if (p_value == 0) 668 start = line; 669 while (line != end) { 670 char c = *line; 671 672 if (!end && isspace(c)) { 673 if (c == '\n') 674 break; 675 if (name_terminate(start, line-start, c, terminate)) 676 break; 677 } 678 line++; 679 if (c == '/' && !--p_value) 680 start = line; 681 } 682 if (!start) 683 return squash_slash(xstrdup_or_null(def)); 684 len = line - start; 685 if (!len) 686 return squash_slash(xstrdup_or_null(def)); 687 688 /* 689 * Generally we prefer the shorter name, especially 690 * if the other one is just a variation of that with 691 * something else tacked on to the end (ie "file.orig" 692 * or "file~"). 693 */ 694 if (def) { 695 int deflen = strlen(def); 696 if (deflen < len && !strncmp(start, def, deflen)) 697 return squash_slash(xstrdup(def)); 698 } 699 700 if (root) { 701 char *ret = xmalloc(root_len + len + 1); 702 strcpy(ret, root); 703 memcpy(ret + root_len, start, len); 704 ret[root_len + len] = '\0'; 705 return squash_slash(ret); 706 } 707 708 return squash_slash(xmemdupz(start, len)); 709} 710 711static char *find_name(const char *line, char *def, int p_value, int terminate) 712{ 713 if (*line == '"') { 714 char *name = find_name_gnu(line, def, p_value); 715 if (name) 716 return name; 717 } 718 719 return find_name_common(line, def, p_value, NULL, terminate); 720} 721 722static char *find_name_traditional(const char *line, char *def, int p_value) 723{ 724 size_t len; 725 size_t date_len; 726 727 if (*line == '"') { 728 char *name = find_name_gnu(line, def, p_value); 729 if (name) 730 return name; 731 } 732 733 len = strchrnul(line, '\n') - line; 734 date_len = diff_timestamp_len(line, len); 735 if (!date_len) 736 return find_name_common(line, def, p_value, NULL, TERM_TAB); 737 len -= date_len; 738 739 return find_name_common(line, def, p_value, line + len, 0); 740} 741 742static int count_slashes(const char *cp) 743{ 744 int cnt = 0; 745 char ch; 746 747 while ((ch = *cp++)) 748 if (ch == '/') 749 cnt++; 750 return cnt; 751} 752 753/* 754 * Given the string after "--- " or "+++ ", guess the appropriate 755 * p_value for the given patch. 756 */ 757static int guess_p_value(const char *nameline) 758{ 759 char *name, *cp; 760 int val = -1; 761 762 if (is_dev_null(nameline)) 763 return -1; 764 name = find_name_traditional(nameline, NULL, 0); 765 if (!name) 766 return -1; 767 cp = strchr(name, '/'); 768 if (!cp) 769 val = 0; 770 else if (prefix) { 771 /* 772 * Does it begin with "a/$our-prefix" and such? Then this is 773 * very likely to apply to our directory. 774 */ 775 if (!strncmp(name, prefix, prefix_length)) 776 val = count_slashes(prefix); 777 else { 778 cp++; 779 if (!strncmp(cp, prefix, prefix_length)) 780 val = count_slashes(prefix) + 1; 781 } 782 } 783 free(name); 784 return val; 785} 786 787/* 788 * Does the ---/+++ line has the POSIX timestamp after the last HT? 789 * GNU diff puts epoch there to signal a creation/deletion event. Is 790 * this such a timestamp? 791 */ 792static int has_epoch_timestamp(const char *nameline) 793{ 794 /* 795 * We are only interested in epoch timestamp; any non-zero 796 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 797 * For the same reason, the date must be either 1969-12-31 or 798 * 1970-01-01, and the seconds part must be "00". 799 */ 800 const char stamp_regexp[] = 801 "^(1969-12-31|1970-01-01)" 802 " " 803 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 804 " " 805 "([-+][0-2][0-9]:?[0-5][0-9])\n"; 806 const char *timestamp = NULL, *cp, *colon; 807 static regex_t *stamp; 808 regmatch_t m[10]; 809 int zoneoffset; 810 int hourminute; 811 int status; 812 813 for (cp = nameline; *cp != '\n'; cp++) { 814 if (*cp == '\t') 815 timestamp = cp + 1; 816 } 817 if (!timestamp) 818 return 0; 819 if (!stamp) { 820 stamp = xmalloc(sizeof(*stamp)); 821 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 822 warning(_("Cannot prepare timestamp regexp %s"), 823 stamp_regexp); 824 return 0; 825 } 826 } 827 828 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 829 if (status) { 830 if (status != REG_NOMATCH) 831 warning(_("regexec returned %d for input: %s"), 832 status, timestamp); 833 return 0; 834 } 835 836 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10); 837 if (*colon == ':') 838 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10); 839 else 840 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 841 if (timestamp[m[3].rm_so] == '-') 842 zoneoffset = -zoneoffset; 843 844 /* 845 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 846 * (west of GMT) or 1970-01-01 (east of GMT) 847 */ 848 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 849 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 850 return 0; 851 852 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 853 strtol(timestamp + 14, NULL, 10) - 854 zoneoffset); 855 856 return ((zoneoffset < 0 && hourminute == 1440) || 857 (0 <= zoneoffset && !hourminute)); 858} 859 860/* 861 * Get the name etc info from the ---/+++ lines of a traditional patch header 862 * 863 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 864 * files, we can happily check the index for a match, but for creating a 865 * new file we should try to match whatever "patch" does. I have no idea. 866 */ 867static void parse_traditional_patch(const char *first, const char *second, struct patch *patch) 868{ 869 char *name; 870 871 first += 4; /* skip "--- " */ 872 second += 4; /* skip "+++ " */ 873 if (!p_value_known) { 874 int p, q; 875 p = guess_p_value(first); 876 q = guess_p_value(second); 877 if (p < 0) p = q; 878 if (0 <= p && p == q) { 879 p_value = p; 880 p_value_known = 1; 881 } 882 } 883 if (is_dev_null(first)) { 884 patch->is_new = 1; 885 patch->is_delete = 0; 886 name = find_name_traditional(second, NULL, p_value); 887 patch->new_name = name; 888 } else if (is_dev_null(second)) { 889 patch->is_new = 0; 890 patch->is_delete = 1; 891 name = find_name_traditional(first, NULL, p_value); 892 patch->old_name = name; 893 } else { 894 char *first_name; 895 first_name = find_name_traditional(first, NULL, p_value); 896 name = find_name_traditional(second, first_name, p_value); 897 free(first_name); 898 if (has_epoch_timestamp(first)) { 899 patch->is_new = 1; 900 patch->is_delete = 0; 901 patch->new_name = name; 902 } else if (has_epoch_timestamp(second)) { 903 patch->is_new = 0; 904 patch->is_delete = 1; 905 patch->old_name = name; 906 } else { 907 patch->old_name = name; 908 patch->new_name = xstrdup_or_null(name); 909 } 910 } 911 if (!name) 912 die(_("unable to find filename in patch at line %d"), linenr); 913} 914 915static int gitdiff_hdrend(const char *line, struct patch *patch) 916{ 917 return -1; 918} 919 920/* 921 * We're anal about diff header consistency, to make 922 * sure that we don't end up having strange ambiguous 923 * patches floating around. 924 * 925 * As a result, gitdiff_{old|new}name() will check 926 * their names against any previous information, just 927 * to make sure.. 928 */ 929#define DIFF_OLD_NAME 0 930#define DIFF_NEW_NAME 1 931 932static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, int side) 933{ 934 if (!orig_name && !isnull) 935 return find_name(line, NULL, p_value, TERM_TAB); 936 937 if (orig_name) { 938 int len; 939 const char *name; 940 char *another; 941 name = orig_name; 942 len = strlen(name); 943 if (isnull) 944 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), name, linenr); 945 another = find_name(line, NULL, p_value, TERM_TAB); 946 if (!another || memcmp(another, name, len + 1)) 947 die((side == DIFF_NEW_NAME) ? 948 _("git apply: bad git-diff - inconsistent new filename on line %d") : 949 _("git apply: bad git-diff - inconsistent old filename on line %d"), linenr); 950 free(another); 951 return orig_name; 952 } 953 else { 954 /* expect "/dev/null" */ 955 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 956 die(_("git apply: bad git-diff - expected /dev/null on line %d"), linenr); 957 return NULL; 958 } 959} 960 961static int gitdiff_oldname(const char *line, struct patch *patch) 962{ 963 char *orig = patch->old_name; 964 patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, 965 DIFF_OLD_NAME); 966 if (orig != patch->old_name) 967 free(orig); 968 return 0; 969} 970 971static int gitdiff_newname(const char *line, struct patch *patch) 972{ 973 char *orig = patch->new_name; 974 patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, 975 DIFF_NEW_NAME); 976 if (orig != patch->new_name) 977 free(orig); 978 return 0; 979} 980 981static int gitdiff_oldmode(const char *line, struct patch *patch) 982{ 983 patch->old_mode = strtoul(line, NULL, 8); 984 return 0; 985} 986 987static int gitdiff_newmode(const char *line, struct patch *patch) 988{ 989 patch->new_mode = strtoul(line, NULL, 8); 990 return 0; 991} 992 993static int gitdiff_delete(const char *line, struct patch *patch) 994{ 995 patch->is_delete = 1; 996 free(patch->old_name); 997 patch->old_name = xstrdup_or_null(patch->def_name); 998 return gitdiff_oldmode(line, patch); 999}10001001static int gitdiff_newfile(const char *line, struct patch *patch)1002{1003 patch->is_new = 1;1004 free(patch->new_name);1005 patch->new_name = xstrdup_or_null(patch->def_name);1006 return gitdiff_newmode(line, patch);1007}10081009static int gitdiff_copysrc(const char *line, struct patch *patch)1010{1011 patch->is_copy = 1;1012 free(patch->old_name);1013 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1014 return 0;1015}10161017static int gitdiff_copydst(const char *line, struct patch *patch)1018{1019 patch->is_copy = 1;1020 free(patch->new_name);1021 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1022 return 0;1023}10241025static int gitdiff_renamesrc(const char *line, struct patch *patch)1026{1027 patch->is_rename = 1;1028 free(patch->old_name);1029 patch->old_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1030 return 0;1031}10321033static int gitdiff_renamedst(const char *line, struct patch *patch)1034{1035 patch->is_rename = 1;1036 free(patch->new_name);1037 patch->new_name = find_name(line, NULL, p_value ? p_value - 1 : 0, 0);1038 return 0;1039}10401041static int gitdiff_similarity(const char *line, struct patch *patch)1042{1043 unsigned long val = strtoul(line, NULL, 10);1044 if (val <= 100)1045 patch->score = val;1046 return 0;1047}10481049static int gitdiff_dissimilarity(const char *line, struct patch *patch)1050{1051 unsigned long val = strtoul(line, NULL, 10);1052 if (val <= 100)1053 patch->score = val;1054 return 0;1055}10561057static int gitdiff_index(const char *line, struct patch *patch)1058{1059 /*1060 * index line is N hexadecimal, "..", N hexadecimal,1061 * and optional space with octal mode.1062 */1063 const char *ptr, *eol;1064 int len;10651066 ptr = strchr(line, '.');1067 if (!ptr || ptr[1] != '.' || 40 < ptr - line)1068 return 0;1069 len = ptr - line;1070 memcpy(patch->old_sha1_prefix, line, len);1071 patch->old_sha1_prefix[len] = 0;10721073 line = ptr + 2;1074 ptr = strchr(line, ' ');1075 eol = strchrnul(line, '\n');10761077 if (!ptr || eol < ptr)1078 ptr = eol;1079 len = ptr - line;10801081 if (40 < len)1082 return 0;1083 memcpy(patch->new_sha1_prefix, line, len);1084 patch->new_sha1_prefix[len] = 0;1085 if (*ptr == ' ')1086 patch->old_mode = strtoul(ptr+1, NULL, 8);1087 return 0;1088}10891090/*1091 * This is normal for a diff that doesn't change anything: we'll fall through1092 * into the next diff. Tell the parser to break out.1093 */1094static int gitdiff_unrecognized(const char *line, struct patch *patch)1095{1096 return -1;1097}10981099/*1100 * Skip p_value leading components from "line"; as we do not accept1101 * absolute paths, return NULL in that case.1102 */1103static const char *skip_tree_prefix(const char *line, int llen)1104{1105 int nslash;1106 int i;11071108 if (!p_value)1109 return (llen && line[0] == '/') ? NULL : line;11101111 nslash = p_value;1112 for (i = 0; i < llen; i++) {1113 int ch = line[i];1114 if (ch == '/' && --nslash <= 0)1115 return (i == 0) ? NULL : &line[i + 1];1116 }1117 return NULL;1118}11191120/*1121 * This is to extract the same name that appears on "diff --git"1122 * line. We do not find and return anything if it is a rename1123 * patch, and it is OK because we will find the name elsewhere.1124 * We need to reliably find name only when it is mode-change only,1125 * creation or deletion of an empty file. In any of these cases,1126 * both sides are the same name under a/ and b/ respectively.1127 */1128static char *git_header_name(const char *line, int llen)1129{1130 const char *name;1131 const char *second = NULL;1132 size_t len, line_len;11331134 line += strlen("diff --git ");1135 llen -= strlen("diff --git ");11361137 if (*line == '"') {1138 const char *cp;1139 struct strbuf first = STRBUF_INIT;1140 struct strbuf sp = STRBUF_INIT;11411142 if (unquote_c_style(&first, line, &second))1143 goto free_and_fail1;11441145 /* strip the a/b prefix including trailing slash */1146 cp = skip_tree_prefix(first.buf, first.len);1147 if (!cp)1148 goto free_and_fail1;1149 strbuf_remove(&first, 0, cp - first.buf);11501151 /*1152 * second points at one past closing dq of name.1153 * find the second name.1154 */1155 while ((second < line + llen) && isspace(*second))1156 second++;11571158 if (line + llen <= second)1159 goto free_and_fail1;1160 if (*second == '"') {1161 if (unquote_c_style(&sp, second, NULL))1162 goto free_and_fail1;1163 cp = skip_tree_prefix(sp.buf, sp.len);1164 if (!cp)1165 goto free_and_fail1;1166 /* They must match, otherwise ignore */1167 if (strcmp(cp, first.buf))1168 goto free_and_fail1;1169 strbuf_release(&sp);1170 return strbuf_detach(&first, NULL);1171 }11721173 /* unquoted second */1174 cp = skip_tree_prefix(second, line + llen - second);1175 if (!cp)1176 goto free_and_fail1;1177 if (line + llen - cp != first.len ||1178 memcmp(first.buf, cp, first.len))1179 goto free_and_fail1;1180 return strbuf_detach(&first, NULL);11811182 free_and_fail1:1183 strbuf_release(&first);1184 strbuf_release(&sp);1185 return NULL;1186 }11871188 /* unquoted first name */1189 name = skip_tree_prefix(line, llen);1190 if (!name)1191 return NULL;11921193 /*1194 * since the first name is unquoted, a dq if exists must be1195 * the beginning of the second name.1196 */1197 for (second = name; second < line + llen; second++) {1198 if (*second == '"') {1199 struct strbuf sp = STRBUF_INIT;1200 const char *np;12011202 if (unquote_c_style(&sp, second, NULL))1203 goto free_and_fail2;12041205 np = skip_tree_prefix(sp.buf, sp.len);1206 if (!np)1207 goto free_and_fail2;12081209 len = sp.buf + sp.len - np;1210 if (len < second - name &&1211 !strncmp(np, name, len) &&1212 isspace(name[len])) {1213 /* Good */1214 strbuf_remove(&sp, 0, np - sp.buf);1215 return strbuf_detach(&sp, NULL);1216 }12171218 free_and_fail2:1219 strbuf_release(&sp);1220 return NULL;1221 }1222 }12231224 /*1225 * Accept a name only if it shows up twice, exactly the same1226 * form.1227 */1228 second = strchr(name, '\n');1229 if (!second)1230 return NULL;1231 line_len = second - name;1232 for (len = 0 ; ; len++) {1233 switch (name[len]) {1234 default:1235 continue;1236 case '\n':1237 return NULL;1238 case '\t': case ' ':1239 /*1240 * Is this the separator between the preimage1241 * and the postimage pathname? Again, we are1242 * only interested in the case where there is1243 * no rename, as this is only to set def_name1244 * and a rename patch has the names elsewhere1245 * in an unambiguous form.1246 */1247 if (!name[len + 1])1248 return NULL; /* no postimage name */1249 second = skip_tree_prefix(name + len + 1,1250 line_len - (len + 1));1251 if (!second)1252 return NULL;1253 /*1254 * Does len bytes starting at "name" and "second"1255 * (that are separated by one HT or SP we just1256 * found) exactly match?1257 */1258 if (second[len] == '\n' && !strncmp(name, second, len))1259 return xmemdupz(name, len);1260 }1261 }1262}12631264/* Verify that we recognize the lines following a git header */1265static int parse_git_header(const char *line, int len, unsigned int size, struct patch *patch)1266{1267 unsigned long offset;12681269 /* A git diff has explicit new/delete information, so we don't guess */1270 patch->is_new = 0;1271 patch->is_delete = 0;12721273 /*1274 * Some things may not have the old name in the1275 * rest of the headers anywhere (pure mode changes,1276 * or removing or adding empty files), so we get1277 * the default name from the header.1278 */1279 patch->def_name = git_header_name(line, len);1280 if (patch->def_name && root) {1281 char *s = xstrfmt("%s%s", root, patch->def_name);1282 free(patch->def_name);1283 patch->def_name = s;1284 }12851286 line += len;1287 size -= len;1288 linenr++;1289 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {1290 static const struct opentry {1291 const char *str;1292 int (*fn)(const char *, struct patch *);1293 } optable[] = {1294 { "@@ -", gitdiff_hdrend },1295 { "--- ", gitdiff_oldname },1296 { "+++ ", gitdiff_newname },1297 { "old mode ", gitdiff_oldmode },1298 { "new mode ", gitdiff_newmode },1299 { "deleted file mode ", gitdiff_delete },1300 { "new file mode ", gitdiff_newfile },1301 { "copy from ", gitdiff_copysrc },1302 { "copy to ", gitdiff_copydst },1303 { "rename old ", gitdiff_renamesrc },1304 { "rename new ", gitdiff_renamedst },1305 { "rename from ", gitdiff_renamesrc },1306 { "rename to ", gitdiff_renamedst },1307 { "similarity index ", gitdiff_similarity },1308 { "dissimilarity index ", gitdiff_dissimilarity },1309 { "index ", gitdiff_index },1310 { "", gitdiff_unrecognized },1311 };1312 int i;13131314 len = linelen(line, size);1315 if (!len || line[len-1] != '\n')1316 break;1317 for (i = 0; i < ARRAY_SIZE(optable); i++) {1318 const struct opentry *p = optable + i;1319 int oplen = strlen(p->str);1320 if (len < oplen || memcmp(p->str, line, oplen))1321 continue;1322 if (p->fn(line + oplen, patch) < 0)1323 return offset;1324 break;1325 }1326 }13271328 return offset;1329}13301331static int parse_num(const char *line, unsigned long *p)1332{1333 char *ptr;13341335 if (!isdigit(*line))1336 return 0;1337 *p = strtoul(line, &ptr, 10);1338 return ptr - line;1339}13401341static int parse_range(const char *line, int len, int offset, const char *expect,1342 unsigned long *p1, unsigned long *p2)1343{1344 int digits, ex;13451346 if (offset < 0 || offset >= len)1347 return -1;1348 line += offset;1349 len -= offset;13501351 digits = parse_num(line, p1);1352 if (!digits)1353 return -1;13541355 offset += digits;1356 line += digits;1357 len -= digits;13581359 *p2 = 1;1360 if (*line == ',') {1361 digits = parse_num(line+1, p2);1362 if (!digits)1363 return -1;13641365 offset += digits+1;1366 line += digits+1;1367 len -= digits+1;1368 }13691370 ex = strlen(expect);1371 if (ex > len)1372 return -1;1373 if (memcmp(line, expect, ex))1374 return -1;13751376 return offset + ex;1377}13781379static void recount_diff(const char *line, int size, struct fragment *fragment)1380{1381 int oldlines = 0, newlines = 0, ret = 0;13821383 if (size < 1) {1384 warning("recount: ignore empty hunk");1385 return;1386 }13871388 for (;;) {1389 int len = linelen(line, size);1390 size -= len;1391 line += len;13921393 if (size < 1)1394 break;13951396 switch (*line) {1397 case ' ': case '\n':1398 newlines++;1399 /* fall through */1400 case '-':1401 oldlines++;1402 continue;1403 case '+':1404 newlines++;1405 continue;1406 case '\\':1407 continue;1408 case '@':1409 ret = size < 3 || !starts_with(line, "@@ ");1410 break;1411 case 'd':1412 ret = size < 5 || !starts_with(line, "diff ");1413 break;1414 default:1415 ret = -1;1416 break;1417 }1418 if (ret) {1419 warning(_("recount: unexpected line: %.*s"),1420 (int)linelen(line, size), line);1421 return;1422 }1423 break;1424 }1425 fragment->oldlines = oldlines;1426 fragment->newlines = newlines;1427}14281429/*1430 * Parse a unified diff fragment header of the1431 * form "@@ -a,b +c,d @@"1432 */1433static int parse_fragment_header(const char *line, int len, struct fragment *fragment)1434{1435 int offset;14361437 if (!len || line[len-1] != '\n')1438 return -1;14391440 /* Figure out the number of lines in a fragment */1441 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1442 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);14431444 return offset;1445}14461447static int find_header(const char *line, unsigned long size, int *hdrsize, struct patch *patch)1448{1449 unsigned long offset, len;14501451 patch->is_toplevel_relative = 0;1452 patch->is_rename = patch->is_copy = 0;1453 patch->is_new = patch->is_delete = -1;1454 patch->old_mode = patch->new_mode = 0;1455 patch->old_name = patch->new_name = NULL;1456 for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {1457 unsigned long nextlen;14581459 len = linelen(line, size);1460 if (!len)1461 break;14621463 /* Testing this early allows us to take a few shortcuts.. */1464 if (len < 6)1465 continue;14661467 /*1468 * Make sure we don't find any unconnected patch fragments.1469 * That's a sign that we didn't find a header, and that a1470 * patch has become corrupted/broken up.1471 */1472 if (!memcmp("@@ -", line, 4)) {1473 struct fragment dummy;1474 if (parse_fragment_header(line, len, &dummy) < 0)1475 continue;1476 die(_("patch fragment without header at line %d: %.*s"),1477 linenr, (int)len-1, line);1478 }14791480 if (size < len + 6)1481 break;14821483 /*1484 * Git patch? It might not have a real patch, just a rename1485 * or mode change, so we handle that specially1486 */1487 if (!memcmp("diff --git ", line, 11)) {1488 int git_hdr_len = parse_git_header(line, len, size, patch);1489 if (git_hdr_len <= len)1490 continue;1491 if (!patch->old_name && !patch->new_name) {1492 if (!patch->def_name)1493 die(Q_("git diff header lacks filename information when removing "1494 "%d leading pathname component (line %d)",1495 "git diff header lacks filename information when removing "1496 "%d leading pathname components (line %d)",1497 p_value),1498 p_value, linenr);1499 patch->old_name = xstrdup(patch->def_name);1500 patch->new_name = xstrdup(patch->def_name);1501 }1502 if (!patch->is_delete && !patch->new_name)1503 die("git diff header lacks filename information "1504 "(line %d)", linenr);1505 patch->is_toplevel_relative = 1;1506 *hdrsize = git_hdr_len;1507 return offset;1508 }15091510 /* --- followed by +++ ? */1511 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1512 continue;15131514 /*1515 * We only accept unified patches, so we want it to1516 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1517 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1518 */1519 nextlen = linelen(line + len, size - len);1520 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1521 continue;15221523 /* Ok, we'll consider it a patch */1524 parse_traditional_patch(line, line+len, patch);1525 *hdrsize = len + nextlen;1526 linenr += 2;1527 return offset;1528 }1529 return -1;1530}15311532static void record_ws_error(unsigned result, const char *line, int len, int linenr)1533{1534 char *err;15351536 if (!result)1537 return;15381539 whitespace_error++;1540 if (squelch_whitespace_errors &&1541 squelch_whitespace_errors < whitespace_error)1542 return;15431544 err = whitespace_error_string(result);1545 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1546 patch_input_file, linenr, err, len, line);1547 free(err);1548}15491550static void check_whitespace(const char *line, int len, unsigned ws_rule)1551{1552 unsigned result = ws_check(line + 1, len - 1, ws_rule);15531554 record_ws_error(result, line + 1, len - 2, linenr);1555}15561557/*1558 * Parse a unified diff. Note that this really needs to parse each1559 * fragment separately, since the only way to know the difference1560 * between a "---" that is part of a patch, and a "---" that starts1561 * the next patch is to look at the line counts..1562 */1563static int parse_fragment(const char *line, unsigned long size,1564 struct patch *patch, struct fragment *fragment)1565{1566 int added, deleted;1567 int len = linelen(line, size), offset;1568 unsigned long oldlines, newlines;1569 unsigned long leading, trailing;15701571 offset = parse_fragment_header(line, len, fragment);1572 if (offset < 0)1573 return -1;1574 if (offset > 0 && patch->recount)1575 recount_diff(line + offset, size - offset, fragment);1576 oldlines = fragment->oldlines;1577 newlines = fragment->newlines;1578 leading = 0;1579 trailing = 0;15801581 /* Parse the thing.. */1582 line += len;1583 size -= len;1584 linenr++;1585 added = deleted = 0;1586 for (offset = len;1587 0 < size;1588 offset += len, size -= len, line += len, linenr++) {1589 if (!oldlines && !newlines)1590 break;1591 len = linelen(line, size);1592 if (!len || line[len-1] != '\n')1593 return -1;1594 switch (*line) {1595 default:1596 return -1;1597 case '\n': /* newer GNU diff, an empty context line */1598 case ' ':1599 oldlines--;1600 newlines--;1601 if (!deleted && !added)1602 leading++;1603 trailing++;1604 if (!apply_in_reverse &&1605 ws_error_action == correct_ws_error)1606 check_whitespace(line, len, patch->ws_rule);1607 break;1608 case '-':1609 if (apply_in_reverse &&1610 ws_error_action != nowarn_ws_error)1611 check_whitespace(line, len, patch->ws_rule);1612 deleted++;1613 oldlines--;1614 trailing = 0;1615 break;1616 case '+':1617 if (!apply_in_reverse &&1618 ws_error_action != nowarn_ws_error)1619 check_whitespace(line, len, patch->ws_rule);1620 added++;1621 newlines--;1622 trailing = 0;1623 break;16241625 /*1626 * We allow "\ No newline at end of file". Depending1627 * on locale settings when the patch was produced we1628 * don't know what this line looks like. The only1629 * thing we do know is that it begins with "\ ".1630 * Checking for 12 is just for sanity check -- any1631 * l10n of "\ No newline..." is at least that long.1632 */1633 case '\\':1634 if (len < 12 || memcmp(line, "\\ ", 2))1635 return -1;1636 break;1637 }1638 }1639 if (oldlines || newlines)1640 return -1;1641 fragment->leading = leading;1642 fragment->trailing = trailing;16431644 /*1645 * If a fragment ends with an incomplete line, we failed to include1646 * it in the above loop because we hit oldlines == newlines == 01647 * before seeing it.1648 */1649 if (12 < size && !memcmp(line, "\\ ", 2))1650 offset += linelen(line, size);16511652 patch->lines_added += added;1653 patch->lines_deleted += deleted;16541655 if (0 < patch->is_new && oldlines)1656 return error(_("new file depends on old contents"));1657 if (0 < patch->is_delete && newlines)1658 return error(_("deleted file still has contents"));1659 return offset;1660}16611662/*1663 * We have seen "diff --git a/... b/..." header (or a traditional patch1664 * header). Read hunks that belong to this patch into fragments and hang1665 * them to the given patch structure.1666 *1667 * The (fragment->patch, fragment->size) pair points into the memory given1668 * by the caller, not a copy, when we return.1669 */1670static int parse_single_patch(const char *line, unsigned long size, struct patch *patch)1671{1672 unsigned long offset = 0;1673 unsigned long oldlines = 0, newlines = 0, context = 0;1674 struct fragment **fragp = &patch->fragments;16751676 while (size > 4 && !memcmp(line, "@@ -", 4)) {1677 struct fragment *fragment;1678 int len;16791680 fragment = xcalloc(1, sizeof(*fragment));1681 fragment->linenr = linenr;1682 len = parse_fragment(line, size, patch, fragment);1683 if (len <= 0)1684 die(_("corrupt patch at line %d"), linenr);1685 fragment->patch = line;1686 fragment->size = len;1687 oldlines += fragment->oldlines;1688 newlines += fragment->newlines;1689 context += fragment->leading + fragment->trailing;16901691 *fragp = fragment;1692 fragp = &fragment->next;16931694 offset += len;1695 line += len;1696 size -= len;1697 }16981699 /*1700 * If something was removed (i.e. we have old-lines) it cannot1701 * be creation, and if something was added it cannot be1702 * deletion. However, the reverse is not true; --unified=01703 * patches that only add are not necessarily creation even1704 * though they do not have any old lines, and ones that only1705 * delete are not necessarily deletion.1706 *1707 * Unfortunately, a real creation/deletion patch do _not_ have1708 * any context line by definition, so we cannot safely tell it1709 * apart with --unified=0 insanity. At least if the patch has1710 * more than one hunk it is not creation or deletion.1711 */1712 if (patch->is_new < 0 &&1713 (oldlines || (patch->fragments && patch->fragments->next)))1714 patch->is_new = 0;1715 if (patch->is_delete < 0 &&1716 (newlines || (patch->fragments && patch->fragments->next)))1717 patch->is_delete = 0;17181719 if (0 < patch->is_new && oldlines)1720 die(_("new file %s depends on old contents"), patch->new_name);1721 if (0 < patch->is_delete && newlines)1722 die(_("deleted file %s still has contents"), patch->old_name);1723 if (!patch->is_delete && !newlines && context)1724 fprintf_ln(stderr,1725 _("** warning: "1726 "file %s becomes empty but is not deleted"),1727 patch->new_name);17281729 return offset;1730}17311732static inline int metadata_changes(struct patch *patch)1733{1734 return patch->is_rename > 0 ||1735 patch->is_copy > 0 ||1736 patch->is_new > 0 ||1737 patch->is_delete ||1738 (patch->old_mode && patch->new_mode &&1739 patch->old_mode != patch->new_mode);1740}17411742static char *inflate_it(const void *data, unsigned long size,1743 unsigned long inflated_size)1744{1745 git_zstream stream;1746 void *out;1747 int st;17481749 memset(&stream, 0, sizeof(stream));17501751 stream.next_in = (unsigned char *)data;1752 stream.avail_in = size;1753 stream.next_out = out = xmalloc(inflated_size);1754 stream.avail_out = inflated_size;1755 git_inflate_init(&stream);1756 st = git_inflate(&stream, Z_FINISH);1757 git_inflate_end(&stream);1758 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1759 free(out);1760 return NULL;1761 }1762 return out;1763}17641765/*1766 * Read a binary hunk and return a new fragment; fragment->patch1767 * points at an allocated memory that the caller must free, so1768 * it is marked as "->free_patch = 1".1769 */1770static struct fragment *parse_binary_hunk(char **buf_p,1771 unsigned long *sz_p,1772 int *status_p,1773 int *used_p)1774{1775 /*1776 * Expect a line that begins with binary patch method ("literal"1777 * or "delta"), followed by the length of data before deflating.1778 * a sequence of 'length-byte' followed by base-85 encoded data1779 * should follow, terminated by a newline.1780 *1781 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1782 * and we would limit the patch line to 66 characters,1783 * so one line can fit up to 13 groups that would decode1784 * to 52 bytes max. The length byte 'A'-'Z' corresponds1785 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1786 */1787 int llen, used;1788 unsigned long size = *sz_p;1789 char *buffer = *buf_p;1790 int patch_method;1791 unsigned long origlen;1792 char *data = NULL;1793 int hunk_size = 0;1794 struct fragment *frag;17951796 llen = linelen(buffer, size);1797 used = llen;17981799 *status_p = 0;18001801 if (starts_with(buffer, "delta ")) {1802 patch_method = BINARY_DELTA_DEFLATED;1803 origlen = strtoul(buffer + 6, NULL, 10);1804 }1805 else if (starts_with(buffer, "literal ")) {1806 patch_method = BINARY_LITERAL_DEFLATED;1807 origlen = strtoul(buffer + 8, NULL, 10);1808 }1809 else1810 return NULL;18111812 linenr++;1813 buffer += llen;1814 while (1) {1815 int byte_length, max_byte_length, newsize;1816 llen = linelen(buffer, size);1817 used += llen;1818 linenr++;1819 if (llen == 1) {1820 /* consume the blank line */1821 buffer++;1822 size--;1823 break;1824 }1825 /*1826 * Minimum line is "A00000\n" which is 7-byte long,1827 * and the line length must be multiple of 5 plus 2.1828 */1829 if ((llen < 7) || (llen-2) % 5)1830 goto corrupt;1831 max_byte_length = (llen - 2) / 5 * 4;1832 byte_length = *buffer;1833 if ('A' <= byte_length && byte_length <= 'Z')1834 byte_length = byte_length - 'A' + 1;1835 else if ('a' <= byte_length && byte_length <= 'z')1836 byte_length = byte_length - 'a' + 27;1837 else1838 goto corrupt;1839 /* if the input length was not multiple of 4, we would1840 * have filler at the end but the filler should never1841 * exceed 3 bytes1842 */1843 if (max_byte_length < byte_length ||1844 byte_length <= max_byte_length - 4)1845 goto corrupt;1846 newsize = hunk_size + byte_length;1847 data = xrealloc(data, newsize);1848 if (decode_85(data + hunk_size, buffer + 1, byte_length))1849 goto corrupt;1850 hunk_size = newsize;1851 buffer += llen;1852 size -= llen;1853 }18541855 frag = xcalloc(1, sizeof(*frag));1856 frag->patch = inflate_it(data, hunk_size, origlen);1857 frag->free_patch = 1;1858 if (!frag->patch)1859 goto corrupt;1860 free(data);1861 frag->size = origlen;1862 *buf_p = buffer;1863 *sz_p = size;1864 *used_p = used;1865 frag->binary_patch_method = patch_method;1866 return frag;18671868 corrupt:1869 free(data);1870 *status_p = -1;1871 error(_("corrupt binary patch at line %d: %.*s"),1872 linenr-1, llen-1, buffer);1873 return NULL;1874}18751876static int parse_binary(char *buffer, unsigned long size, struct patch *patch)1877{1878 /*1879 * We have read "GIT binary patch\n"; what follows is a line1880 * that says the patch method (currently, either "literal" or1881 * "delta") and the length of data before deflating; a1882 * sequence of 'length-byte' followed by base-85 encoded data1883 * follows.1884 *1885 * When a binary patch is reversible, there is another binary1886 * hunk in the same format, starting with patch method (either1887 * "literal" or "delta") with the length of data, and a sequence1888 * of length-byte + base-85 encoded data, terminated with another1889 * empty line. This data, when applied to the postimage, produces1890 * the preimage.1891 */1892 struct fragment *forward;1893 struct fragment *reverse;1894 int status;1895 int used, used_1;18961897 forward = parse_binary_hunk(&buffer, &size, &status, &used);1898 if (!forward && !status)1899 /* there has to be one hunk (forward hunk) */1900 return error(_("unrecognized binary patch at line %d"), linenr-1);1901 if (status)1902 /* otherwise we already gave an error message */1903 return status;19041905 reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);1906 if (reverse)1907 used += used_1;1908 else if (status) {1909 /*1910 * Not having reverse hunk is not an error, but having1911 * a corrupt reverse hunk is.1912 */1913 free((void*) forward->patch);1914 free(forward);1915 return status;1916 }1917 forward->next = reverse;1918 patch->fragments = forward;1919 patch->is_binary = 1;1920 return used;1921}19221923static void prefix_one(char **name)1924{1925 char *old_name = *name;1926 if (!old_name)1927 return;1928 *name = xstrdup(prefix_filename(prefix, prefix_length, *name));1929 free(old_name);1930}19311932static void prefix_patch(struct patch *p)1933{1934 if (!prefix || p->is_toplevel_relative)1935 return;1936 prefix_one(&p->new_name);1937 prefix_one(&p->old_name);1938}19391940/*1941 * include/exclude1942 */19431944static struct string_list limit_by_name;1945static int has_include;1946static void add_name_limit(const char *name, int exclude)1947{1948 struct string_list_item *it;19491950 it = string_list_append(&limit_by_name, name);1951 it->util = exclude ? NULL : (void *) 1;1952}19531954static int use_patch(struct patch *p)1955{1956 const char *pathname = p->new_name ? p->new_name : p->old_name;1957 int i;19581959 /* Paths outside are not touched regardless of "--include" */1960 if (0 < prefix_length) {1961 int pathlen = strlen(pathname);1962 if (pathlen <= prefix_length ||1963 memcmp(prefix, pathname, prefix_length))1964 return 0;1965 }19661967 /* See if it matches any of exclude/include rule */1968 for (i = 0; i < limit_by_name.nr; i++) {1969 struct string_list_item *it = &limit_by_name.items[i];1970 if (!wildmatch(it->string, pathname, 0, NULL))1971 return (it->util != NULL);1972 }19731974 /*1975 * If we had any include, a path that does not match any rule is1976 * not used. Otherwise, we saw bunch of exclude rules (or none)1977 * and such a path is used.1978 */1979 return !has_include;1980}198119821983/*1984 * Read the patch text in "buffer" that extends for "size" bytes; stop1985 * reading after seeing a single patch (i.e. changes to a single file).1986 * Create fragments (i.e. patch hunks) and hang them to the given patch.1987 * Return the number of bytes consumed, so that the caller can call us1988 * again for the next patch.1989 */1990static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)1991{1992 int hdrsize, patchsize;1993 int offset = find_header(buffer, size, &hdrsize, patch);19941995 if (offset < 0)1996 return offset;19971998 prefix_patch(patch);19992000 if (!use_patch(patch))2001 patch->ws_rule = 0;2002 else2003 patch->ws_rule = whitespace_rule(patch->new_name2004 ? patch->new_name2005 : patch->old_name);20062007 patchsize = parse_single_patch(buffer + offset + hdrsize,2008 size - offset - hdrsize, patch);20092010 if (!patchsize) {2011 static const char git_binary[] = "GIT binary patch\n";2012 int hd = hdrsize + offset;2013 unsigned long llen = linelen(buffer + hd, size - hd);20142015 if (llen == sizeof(git_binary) - 1 &&2016 !memcmp(git_binary, buffer + hd, llen)) {2017 int used;2018 linenr++;2019 used = parse_binary(buffer + hd + llen,2020 size - hd - llen, patch);2021 if (used)2022 patchsize = used + llen;2023 else2024 patchsize = 0;2025 }2026 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {2027 static const char *binhdr[] = {2028 "Binary files ",2029 "Files ",2030 NULL,2031 };2032 int i;2033 for (i = 0; binhdr[i]; i++) {2034 int len = strlen(binhdr[i]);2035 if (len < size - hd &&2036 !memcmp(binhdr[i], buffer + hd, len)) {2037 linenr++;2038 patch->is_binary = 1;2039 patchsize = llen;2040 break;2041 }2042 }2043 }20442045 /* Empty patch cannot be applied if it is a text patch2046 * without metadata change. A binary patch appears2047 * empty to us here.2048 */2049 if ((apply || check) &&2050 (!patch->is_binary && !metadata_changes(patch)))2051 die(_("patch with only garbage at line %d"), linenr);2052 }20532054 return offset + hdrsize + patchsize;2055}20562057#define swap(a,b) myswap((a),(b),sizeof(a))20582059#define myswap(a, b, size) do { \2060 unsigned char mytmp[size]; \2061 memcpy(mytmp, &a, size); \2062 memcpy(&a, &b, size); \2063 memcpy(&b, mytmp, size); \2064} while (0)20652066static void reverse_patches(struct patch *p)2067{2068 for (; p; p = p->next) {2069 struct fragment *frag = p->fragments;20702071 swap(p->new_name, p->old_name);2072 swap(p->new_mode, p->old_mode);2073 swap(p->is_new, p->is_delete);2074 swap(p->lines_added, p->lines_deleted);2075 swap(p->old_sha1_prefix, p->new_sha1_prefix);20762077 for (; frag; frag = frag->next) {2078 swap(frag->newpos, frag->oldpos);2079 swap(frag->newlines, frag->oldlines);2080 }2081 }2082}20832084static const char pluses[] =2085"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";2086static const char minuses[]=2087"----------------------------------------------------------------------";20882089static void show_stats(struct patch *patch)2090{2091 struct strbuf qname = STRBUF_INIT;2092 char *cp = patch->new_name ? patch->new_name : patch->old_name;2093 int max, add, del;20942095 quote_c_style(cp, &qname, NULL, 0);20962097 /*2098 * "scale" the filename2099 */2100 max = max_len;2101 if (max > 50)2102 max = 50;21032104 if (qname.len > max) {2105 cp = strchr(qname.buf + qname.len + 3 - max, '/');2106 if (!cp)2107 cp = qname.buf + qname.len + 3 - max;2108 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);2109 }21102111 if (patch->is_binary) {2112 printf(" %-*s | Bin\n", max, qname.buf);2113 strbuf_release(&qname);2114 return;2115 }21162117 printf(" %-*s |", max, qname.buf);2118 strbuf_release(&qname);21192120 /*2121 * scale the add/delete2122 */2123 max = max + max_change > 70 ? 70 - max : max_change;2124 add = patch->lines_added;2125 del = patch->lines_deleted;21262127 if (max_change > 0) {2128 int total = ((add + del) * max + max_change / 2) / max_change;2129 add = (add * max + max_change / 2) / max_change;2130 del = total - add;2131 }2132 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,2133 add, pluses, del, minuses);2134}21352136static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)2137{2138 switch (st->st_mode & S_IFMT) {2139 case S_IFLNK:2140 if (strbuf_readlink(buf, path, st->st_size) < 0)2141 return error(_("unable to read symlink %s"), path);2142 return 0;2143 case S_IFREG:2144 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)2145 return error(_("unable to open or read %s"), path);2146 convert_to_git(path, buf->buf, buf->len, buf, 0);2147 return 0;2148 default:2149 return -1;2150 }2151}21522153/*2154 * Update the preimage, and the common lines in postimage,2155 * from buffer buf of length len. If postlen is 0 the postimage2156 * is updated in place, otherwise it's updated on a new buffer2157 * of length postlen2158 */21592160static void update_pre_post_images(struct image *preimage,2161 struct image *postimage,2162 char *buf,2163 size_t len, size_t postlen)2164{2165 int i, ctx, reduced;2166 char *new, *old, *fixed;2167 struct image fixed_preimage;21682169 /*2170 * Update the preimage with whitespace fixes. Note that we2171 * are not losing preimage->buf -- apply_one_fragment() will2172 * free "oldlines".2173 */2174 prepare_image(&fixed_preimage, buf, len, 1);2175 assert(postlen2176 ? fixed_preimage.nr == preimage->nr2177 : fixed_preimage.nr <= preimage->nr);2178 for (i = 0; i < fixed_preimage.nr; i++)2179 fixed_preimage.line[i].flag = preimage->line[i].flag;2180 free(preimage->line_allocated);2181 *preimage = fixed_preimage;21822183 /*2184 * Adjust the common context lines in postimage. This can be2185 * done in-place when we are shrinking it with whitespace2186 * fixing, but needs a new buffer when ignoring whitespace or2187 * expanding leading tabs to spaces.2188 *2189 * We trust the caller to tell us if the update can be done2190 * in place (postlen==0) or not.2191 */2192 old = postimage->buf;2193 if (postlen)2194 new = postimage->buf = xmalloc(postlen);2195 else2196 new = old;2197 fixed = preimage->buf;21982199 for (i = reduced = ctx = 0; i < postimage->nr; i++) {2200 size_t len = postimage->line[i].len;2201 if (!(postimage->line[i].flag & LINE_COMMON)) {2202 /* an added line -- no counterparts in preimage */2203 memmove(new, old, len);2204 old += len;2205 new += len;2206 continue;2207 }22082209 /* a common context -- skip it in the original postimage */2210 old += len;22112212 /* and find the corresponding one in the fixed preimage */2213 while (ctx < preimage->nr &&2214 !(preimage->line[ctx].flag & LINE_COMMON)) {2215 fixed += preimage->line[ctx].len;2216 ctx++;2217 }22182219 /*2220 * preimage is expected to run out, if the caller2221 * fixed addition of trailing blank lines.2222 */2223 if (preimage->nr <= ctx) {2224 reduced++;2225 continue;2226 }22272228 /* and copy it in, while fixing the line length */2229 len = preimage->line[ctx].len;2230 memcpy(new, fixed, len);2231 new += len;2232 fixed += len;2233 postimage->line[i].len = len;2234 ctx++;2235 }22362237 if (postlen2238 ? postlen < new - postimage->buf2239 : postimage->len < new - postimage->buf)2240 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",2241 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));22422243 /* Fix the length of the whole thing */2244 postimage->len = new - postimage->buf;2245 postimage->nr -= reduced;2246}22472248static int match_fragment(struct image *img,2249 struct image *preimage,2250 struct image *postimage,2251 unsigned long try,2252 int try_lno,2253 unsigned ws_rule,2254 int match_beginning, int match_end)2255{2256 int i;2257 char *fixed_buf, *buf, *orig, *target;2258 struct strbuf fixed;2259 size_t fixed_len, postlen;2260 int preimage_limit;22612262 if (preimage->nr + try_lno <= img->nr) {2263 /*2264 * The hunk falls within the boundaries of img.2265 */2266 preimage_limit = preimage->nr;2267 if (match_end && (preimage->nr + try_lno != img->nr))2268 return 0;2269 } else if (ws_error_action == correct_ws_error &&2270 (ws_rule & WS_BLANK_AT_EOF)) {2271 /*2272 * This hunk extends beyond the end of img, and we are2273 * removing blank lines at the end of the file. This2274 * many lines from the beginning of the preimage must2275 * match with img, and the remainder of the preimage2276 * must be blank.2277 */2278 preimage_limit = img->nr - try_lno;2279 } else {2280 /*2281 * The hunk extends beyond the end of the img and2282 * we are not removing blanks at the end, so we2283 * should reject the hunk at this position.2284 */2285 return 0;2286 }22872288 if (match_beginning && try_lno)2289 return 0;22902291 /* Quick hash check */2292 for (i = 0; i < preimage_limit; i++)2293 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||2294 (preimage->line[i].hash != img->line[try_lno + i].hash))2295 return 0;22962297 if (preimage_limit == preimage->nr) {2298 /*2299 * Do we have an exact match? If we were told to match2300 * at the end, size must be exactly at try+fragsize,2301 * otherwise try+fragsize must be still within the preimage,2302 * and either case, the old piece should match the preimage2303 * exactly.2304 */2305 if ((match_end2306 ? (try + preimage->len == img->len)2307 : (try + preimage->len <= img->len)) &&2308 !memcmp(img->buf + try, preimage->buf, preimage->len))2309 return 1;2310 } else {2311 /*2312 * The preimage extends beyond the end of img, so2313 * there cannot be an exact match.2314 *2315 * There must be one non-blank context line that match2316 * a line before the end of img.2317 */2318 char *buf_end;23192320 buf = preimage->buf;2321 buf_end = buf;2322 for (i = 0; i < preimage_limit; i++)2323 buf_end += preimage->line[i].len;23242325 for ( ; buf < buf_end; buf++)2326 if (!isspace(*buf))2327 break;2328 if (buf == buf_end)2329 return 0;2330 }23312332 /*2333 * No exact match. If we are ignoring whitespace, run a line-by-line2334 * fuzzy matching. We collect all the line length information because2335 * we need it to adjust whitespace if we match.2336 */2337 if (ws_ignore_action == ignore_ws_change) {2338 size_t imgoff = 0;2339 size_t preoff = 0;2340 size_t postlen = postimage->len;2341 size_t extra_chars;2342 char *preimage_eof;2343 char *preimage_end;2344 for (i = 0; i < preimage_limit; i++) {2345 size_t prelen = preimage->line[i].len;2346 size_t imglen = img->line[try_lno+i].len;23472348 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,2349 preimage->buf + preoff, prelen))2350 return 0;2351 if (preimage->line[i].flag & LINE_COMMON)2352 postlen += imglen - prelen;2353 imgoff += imglen;2354 preoff += prelen;2355 }23562357 /*2358 * Ok, the preimage matches with whitespace fuzz.2359 *2360 * imgoff now holds the true length of the target that2361 * matches the preimage before the end of the file.2362 *2363 * Count the number of characters in the preimage that fall2364 * beyond the end of the file and make sure that all of them2365 * are whitespace characters. (This can only happen if2366 * we are removing blank lines at the end of the file.)2367 */2368 buf = preimage_eof = preimage->buf + preoff;2369 for ( ; i < preimage->nr; i++)2370 preoff += preimage->line[i].len;2371 preimage_end = preimage->buf + preoff;2372 for ( ; buf < preimage_end; buf++)2373 if (!isspace(*buf))2374 return 0;23752376 /*2377 * Update the preimage and the common postimage context2378 * lines to use the same whitespace as the target.2379 * If whitespace is missing in the target (i.e.2380 * if the preimage extends beyond the end of the file),2381 * use the whitespace from the preimage.2382 */2383 extra_chars = preimage_end - preimage_eof;2384 strbuf_init(&fixed, imgoff + extra_chars);2385 strbuf_add(&fixed, img->buf + try, imgoff);2386 strbuf_add(&fixed, preimage_eof, extra_chars);2387 fixed_buf = strbuf_detach(&fixed, &fixed_len);2388 update_pre_post_images(preimage, postimage,2389 fixed_buf, fixed_len, postlen);2390 return 1;2391 }23922393 if (ws_error_action != correct_ws_error)2394 return 0;23952396 /*2397 * The hunk does not apply byte-by-byte, but the hash says2398 * it might with whitespace fuzz. We weren't asked to2399 * ignore whitespace, we were asked to correct whitespace2400 * errors, so let's try matching after whitespace correction.2401 *2402 * While checking the preimage against the target, whitespace2403 * errors in both fixed, we count how large the corresponding2404 * postimage needs to be. The postimage prepared by2405 * apply_one_fragment() has whitespace errors fixed on added2406 * lines already, but the common lines were propagated as-is,2407 * which may become longer when their whitespace errors are2408 * fixed.2409 */24102411 /* First count added lines in postimage */2412 postlen = 0;2413 for (i = 0; i < postimage->nr; i++) {2414 if (!(postimage->line[i].flag & LINE_COMMON))2415 postlen += postimage->line[i].len;2416 }24172418 /*2419 * The preimage may extend beyond the end of the file,2420 * but in this loop we will only handle the part of the2421 * preimage that falls within the file.2422 */2423 strbuf_init(&fixed, preimage->len + 1);2424 orig = preimage->buf;2425 target = img->buf + try;2426 for (i = 0; i < preimage_limit; i++) {2427 size_t oldlen = preimage->line[i].len;2428 size_t tgtlen = img->line[try_lno + i].len;2429 size_t fixstart = fixed.len;2430 struct strbuf tgtfix;2431 int match;24322433 /* Try fixing the line in the preimage */2434 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24352436 /* Try fixing the line in the target */2437 strbuf_init(&tgtfix, tgtlen);2438 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);24392440 /*2441 * If they match, either the preimage was based on2442 * a version before our tree fixed whitespace breakage,2443 * or we are lacking a whitespace-fix patch the tree2444 * the preimage was based on already had (i.e. target2445 * has whitespace breakage, the preimage doesn't).2446 * In either case, we are fixing the whitespace breakages2447 * so we might as well take the fix together with their2448 * real change.2449 */2450 match = (tgtfix.len == fixed.len - fixstart &&2451 !memcmp(tgtfix.buf, fixed.buf + fixstart,2452 fixed.len - fixstart));24532454 /* Add the length if this is common with the postimage */2455 if (preimage->line[i].flag & LINE_COMMON)2456 postlen += tgtfix.len;24572458 strbuf_release(&tgtfix);2459 if (!match)2460 goto unmatch_exit;24612462 orig += oldlen;2463 target += tgtlen;2464 }246524662467 /*2468 * Now handle the lines in the preimage that falls beyond the2469 * end of the file (if any). They will only match if they are2470 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2471 * false).2472 */2473 for ( ; i < preimage->nr; i++) {2474 size_t fixstart = fixed.len; /* start of the fixed preimage */2475 size_t oldlen = preimage->line[i].len;2476 int j;24772478 /* Try fixing the line in the preimage */2479 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24802481 for (j = fixstart; j < fixed.len; j++)2482 if (!isspace(fixed.buf[j]))2483 goto unmatch_exit;24842485 orig += oldlen;2486 }24872488 /*2489 * Yes, the preimage is based on an older version that still2490 * has whitespace breakages unfixed, and fixing them makes the2491 * hunk match. Update the context lines in the postimage.2492 */2493 fixed_buf = strbuf_detach(&fixed, &fixed_len);2494 if (postlen < postimage->len)2495 postlen = 0;2496 update_pre_post_images(preimage, postimage,2497 fixed_buf, fixed_len, postlen);2498 return 1;24992500 unmatch_exit:2501 strbuf_release(&fixed);2502 return 0;2503}25042505static int find_pos(struct image *img,2506 struct image *preimage,2507 struct image *postimage,2508 int line,2509 unsigned ws_rule,2510 int match_beginning, int match_end)2511{2512 int i;2513 unsigned long backwards, forwards, try;2514 int backwards_lno, forwards_lno, try_lno;25152516 /*2517 * If match_beginning or match_end is specified, there is no2518 * point starting from a wrong line that will never match and2519 * wander around and wait for a match at the specified end.2520 */2521 if (match_beginning)2522 line = 0;2523 else if (match_end)2524 line = img->nr - preimage->nr;25252526 /*2527 * Because the comparison is unsigned, the following test2528 * will also take care of a negative line number that can2529 * result when match_end and preimage is larger than the target.2530 */2531 if ((size_t) line > img->nr)2532 line = img->nr;25332534 try = 0;2535 for (i = 0; i < line; i++)2536 try += img->line[i].len;25372538 /*2539 * There's probably some smart way to do this, but I'll leave2540 * that to the smart and beautiful people. I'm simple and stupid.2541 */2542 backwards = try;2543 backwards_lno = line;2544 forwards = try;2545 forwards_lno = line;2546 try_lno = line;25472548 for (i = 0; ; i++) {2549 if (match_fragment(img, preimage, postimage,2550 try, try_lno, ws_rule,2551 match_beginning, match_end))2552 return try_lno;25532554 again:2555 if (backwards_lno == 0 && forwards_lno == img->nr)2556 break;25572558 if (i & 1) {2559 if (backwards_lno == 0) {2560 i++;2561 goto again;2562 }2563 backwards_lno--;2564 backwards -= img->line[backwards_lno].len;2565 try = backwards;2566 try_lno = backwards_lno;2567 } else {2568 if (forwards_lno == img->nr) {2569 i++;2570 goto again;2571 }2572 forwards += img->line[forwards_lno].len;2573 forwards_lno++;2574 try = forwards;2575 try_lno = forwards_lno;2576 }25772578 }2579 return -1;2580}25812582static void remove_first_line(struct image *img)2583{2584 img->buf += img->line[0].len;2585 img->len -= img->line[0].len;2586 img->line++;2587 img->nr--;2588}25892590static void remove_last_line(struct image *img)2591{2592 img->len -= img->line[--img->nr].len;2593}25942595/*2596 * The change from "preimage" and "postimage" has been found to2597 * apply at applied_pos (counts in line numbers) in "img".2598 * Update "img" to remove "preimage" and replace it with "postimage".2599 */2600static void update_image(struct image *img,2601 int applied_pos,2602 struct image *preimage,2603 struct image *postimage)2604{2605 /*2606 * remove the copy of preimage at offset in img2607 * and replace it with postimage2608 */2609 int i, nr;2610 size_t remove_count, insert_count, applied_at = 0;2611 char *result;2612 int preimage_limit;26132614 /*2615 * If we are removing blank lines at the end of img,2616 * the preimage may extend beyond the end.2617 * If that is the case, we must be careful only to2618 * remove the part of the preimage that falls within2619 * the boundaries of img. Initialize preimage_limit2620 * to the number of lines in the preimage that falls2621 * within the boundaries.2622 */2623 preimage_limit = preimage->nr;2624 if (preimage_limit > img->nr - applied_pos)2625 preimage_limit = img->nr - applied_pos;26262627 for (i = 0; i < applied_pos; i++)2628 applied_at += img->line[i].len;26292630 remove_count = 0;2631 for (i = 0; i < preimage_limit; i++)2632 remove_count += img->line[applied_pos + i].len;2633 insert_count = postimage->len;26342635 /* Adjust the contents */2636 result = xmalloc(img->len + insert_count - remove_count + 1);2637 memcpy(result, img->buf, applied_at);2638 memcpy(result + applied_at, postimage->buf, postimage->len);2639 memcpy(result + applied_at + postimage->len,2640 img->buf + (applied_at + remove_count),2641 img->len - (applied_at + remove_count));2642 free(img->buf);2643 img->buf = result;2644 img->len += insert_count - remove_count;2645 result[img->len] = '\0';26462647 /* Adjust the line table */2648 nr = img->nr + postimage->nr - preimage_limit;2649 if (preimage_limit < postimage->nr) {2650 /*2651 * NOTE: this knows that we never call remove_first_line()2652 * on anything other than pre/post image.2653 */2654 REALLOC_ARRAY(img->line, nr);2655 img->line_allocated = img->line;2656 }2657 if (preimage_limit != postimage->nr)2658 memmove(img->line + applied_pos + postimage->nr,2659 img->line + applied_pos + preimage_limit,2660 (img->nr - (applied_pos + preimage_limit)) *2661 sizeof(*img->line));2662 memcpy(img->line + applied_pos,2663 postimage->line,2664 postimage->nr * sizeof(*img->line));2665 if (!allow_overlap)2666 for (i = 0; i < postimage->nr; i++)2667 img->line[applied_pos + i].flag |= LINE_PATCHED;2668 img->nr = nr;2669}26702671/*2672 * Use the patch-hunk text in "frag" to prepare two images (preimage and2673 * postimage) for the hunk. Find lines that match "preimage" in "img" and2674 * replace the part of "img" with "postimage" text.2675 */2676static int apply_one_fragment(struct image *img, struct fragment *frag,2677 int inaccurate_eof, unsigned ws_rule,2678 int nth_fragment)2679{2680 int match_beginning, match_end;2681 const char *patch = frag->patch;2682 int size = frag->size;2683 char *old, *oldlines;2684 struct strbuf newlines;2685 int new_blank_lines_at_end = 0;2686 int found_new_blank_lines_at_end = 0;2687 int hunk_linenr = frag->linenr;2688 unsigned long leading, trailing;2689 int pos, applied_pos;2690 struct image preimage;2691 struct image postimage;26922693 memset(&preimage, 0, sizeof(preimage));2694 memset(&postimage, 0, sizeof(postimage));2695 oldlines = xmalloc(size);2696 strbuf_init(&newlines, size);26972698 old = oldlines;2699 while (size > 0) {2700 char first;2701 int len = linelen(patch, size);2702 int plen;2703 int added_blank_line = 0;2704 int is_blank_context = 0;2705 size_t start;27062707 if (!len)2708 break;27092710 /*2711 * "plen" is how much of the line we should use for2712 * the actual patch data. Normally we just remove the2713 * first character on the line, but if the line is2714 * followed by "\ No newline", then we also remove the2715 * last one (which is the newline, of course).2716 */2717 plen = len - 1;2718 if (len < size && patch[len] == '\\')2719 plen--;2720 first = *patch;2721 if (apply_in_reverse) {2722 if (first == '-')2723 first = '+';2724 else if (first == '+')2725 first = '-';2726 }27272728 switch (first) {2729 case '\n':2730 /* Newer GNU diff, empty context line */2731 if (plen < 0)2732 /* ... followed by '\No newline'; nothing */2733 break;2734 *old++ = '\n';2735 strbuf_addch(&newlines, '\n');2736 add_line_info(&preimage, "\n", 1, LINE_COMMON);2737 add_line_info(&postimage, "\n", 1, LINE_COMMON);2738 is_blank_context = 1;2739 break;2740 case ' ':2741 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2742 ws_blank_line(patch + 1, plen, ws_rule))2743 is_blank_context = 1;2744 case '-':2745 memcpy(old, patch + 1, plen);2746 add_line_info(&preimage, old, plen,2747 (first == ' ' ? LINE_COMMON : 0));2748 old += plen;2749 if (first == '-')2750 break;2751 /* Fall-through for ' ' */2752 case '+':2753 /* --no-add does not add new lines */2754 if (first == '+' && no_add)2755 break;27562757 start = newlines.len;2758 if (first != '+' ||2759 !whitespace_error ||2760 ws_error_action != correct_ws_error) {2761 strbuf_add(&newlines, patch + 1, plen);2762 }2763 else {2764 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &applied_after_fixing_ws);2765 }2766 add_line_info(&postimage, newlines.buf + start, newlines.len - start,2767 (first == '+' ? 0 : LINE_COMMON));2768 if (first == '+' &&2769 (ws_rule & WS_BLANK_AT_EOF) &&2770 ws_blank_line(patch + 1, plen, ws_rule))2771 added_blank_line = 1;2772 break;2773 case '@': case '\\':2774 /* Ignore it, we already handled it */2775 break;2776 default:2777 if (apply_verbosely)2778 error(_("invalid start of line: '%c'"), first);2779 return -1;2780 }2781 if (added_blank_line) {2782 if (!new_blank_lines_at_end)2783 found_new_blank_lines_at_end = hunk_linenr;2784 new_blank_lines_at_end++;2785 }2786 else if (is_blank_context)2787 ;2788 else2789 new_blank_lines_at_end = 0;2790 patch += len;2791 size -= len;2792 hunk_linenr++;2793 }2794 if (inaccurate_eof &&2795 old > oldlines && old[-1] == '\n' &&2796 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {2797 old--;2798 strbuf_setlen(&newlines, newlines.len - 1);2799 }28002801 leading = frag->leading;2802 trailing = frag->trailing;28032804 /*2805 * A hunk to change lines at the beginning would begin with2806 * @@ -1,L +N,M @@2807 * but we need to be careful. -U0 that inserts before the second2808 * line also has this pattern.2809 *2810 * And a hunk to add to an empty file would begin with2811 * @@ -0,0 +N,M @@2812 *2813 * In other words, a hunk that is (frag->oldpos <= 1) with or2814 * without leading context must match at the beginning.2815 */2816 match_beginning = (!frag->oldpos ||2817 (frag->oldpos == 1 && !unidiff_zero));28182819 /*2820 * A hunk without trailing lines must match at the end.2821 * However, we simply cannot tell if a hunk must match end2822 * from the lack of trailing lines if the patch was generated2823 * with unidiff without any context.2824 */2825 match_end = !unidiff_zero && !trailing;28262827 pos = frag->newpos ? (frag->newpos - 1) : 0;2828 preimage.buf = oldlines;2829 preimage.len = old - oldlines;2830 postimage.buf = newlines.buf;2831 postimage.len = newlines.len;2832 preimage.line = preimage.line_allocated;2833 postimage.line = postimage.line_allocated;28342835 for (;;) {28362837 applied_pos = find_pos(img, &preimage, &postimage, pos,2838 ws_rule, match_beginning, match_end);28392840 if (applied_pos >= 0)2841 break;28422843 /* Am I at my context limits? */2844 if ((leading <= p_context) && (trailing <= p_context))2845 break;2846 if (match_beginning || match_end) {2847 match_beginning = match_end = 0;2848 continue;2849 }28502851 /*2852 * Reduce the number of context lines; reduce both2853 * leading and trailing if they are equal otherwise2854 * just reduce the larger context.2855 */2856 if (leading >= trailing) {2857 remove_first_line(&preimage);2858 remove_first_line(&postimage);2859 pos--;2860 leading--;2861 }2862 if (trailing > leading) {2863 remove_last_line(&preimage);2864 remove_last_line(&postimage);2865 trailing--;2866 }2867 }28682869 if (applied_pos >= 0) {2870 if (new_blank_lines_at_end &&2871 preimage.nr + applied_pos >= img->nr &&2872 (ws_rule & WS_BLANK_AT_EOF) &&2873 ws_error_action != nowarn_ws_error) {2874 record_ws_error(WS_BLANK_AT_EOF, "+", 1,2875 found_new_blank_lines_at_end);2876 if (ws_error_action == correct_ws_error) {2877 while (new_blank_lines_at_end--)2878 remove_last_line(&postimage);2879 }2880 /*2881 * We would want to prevent write_out_results()2882 * from taking place in apply_patch() that follows2883 * the callchain led us here, which is:2884 * apply_patch->check_patch_list->check_patch->2885 * apply_data->apply_fragments->apply_one_fragment2886 */2887 if (ws_error_action == die_on_ws_error)2888 apply = 0;2889 }28902891 if (apply_verbosely && applied_pos != pos) {2892 int offset = applied_pos - pos;2893 if (apply_in_reverse)2894 offset = 0 - offset;2895 fprintf_ln(stderr,2896 Q_("Hunk #%d succeeded at %d (offset %d line).",2897 "Hunk #%d succeeded at %d (offset %d lines).",2898 offset),2899 nth_fragment, applied_pos + 1, offset);2900 }29012902 /*2903 * Warn if it was necessary to reduce the number2904 * of context lines.2905 */2906 if ((leading != frag->leading) ||2907 (trailing != frag->trailing))2908 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"2909 " to apply fragment at %d"),2910 leading, trailing, applied_pos+1);2911 update_image(img, applied_pos, &preimage, &postimage);2912 } else {2913 if (apply_verbosely)2914 error(_("while searching for:\n%.*s"),2915 (int)(old - oldlines), oldlines);2916 }29172918 free(oldlines);2919 strbuf_release(&newlines);2920 free(preimage.line_allocated);2921 free(postimage.line_allocated);29222923 return (applied_pos < 0);2924}29252926static int apply_binary_fragment(struct image *img, struct patch *patch)2927{2928 struct fragment *fragment = patch->fragments;2929 unsigned long len;2930 void *dst;29312932 if (!fragment)2933 return error(_("missing binary patch data for '%s'"),2934 patch->new_name ?2935 patch->new_name :2936 patch->old_name);29372938 /* Binary patch is irreversible without the optional second hunk */2939 if (apply_in_reverse) {2940 if (!fragment->next)2941 return error("cannot reverse-apply a binary patch "2942 "without the reverse hunk to '%s'",2943 patch->new_name2944 ? patch->new_name : patch->old_name);2945 fragment = fragment->next;2946 }2947 switch (fragment->binary_patch_method) {2948 case BINARY_DELTA_DEFLATED:2949 dst = patch_delta(img->buf, img->len, fragment->patch,2950 fragment->size, &len);2951 if (!dst)2952 return -1;2953 clear_image(img);2954 img->buf = dst;2955 img->len = len;2956 return 0;2957 case BINARY_LITERAL_DEFLATED:2958 clear_image(img);2959 img->len = fragment->size;2960 img->buf = xmemdupz(fragment->patch, img->len);2961 return 0;2962 }2963 return -1;2964}29652966/*2967 * Replace "img" with the result of applying the binary patch.2968 * The binary patch data itself in patch->fragment is still kept2969 * but the preimage prepared by the caller in "img" is freed here2970 * or in the helper function apply_binary_fragment() this calls.2971 */2972static int apply_binary(struct image *img, struct patch *patch)2973{2974 const char *name = patch->old_name ? patch->old_name : patch->new_name;2975 unsigned char sha1[20];29762977 /*2978 * For safety, we require patch index line to contain2979 * full 40-byte textual SHA1 for old and new, at least for now.2980 */2981 if (strlen(patch->old_sha1_prefix) != 40 ||2982 strlen(patch->new_sha1_prefix) != 40 ||2983 get_sha1_hex(patch->old_sha1_prefix, sha1) ||2984 get_sha1_hex(patch->new_sha1_prefix, sha1))2985 return error("cannot apply binary patch to '%s' "2986 "without full index line", name);29872988 if (patch->old_name) {2989 /*2990 * See if the old one matches what the patch2991 * applies to.2992 */2993 hash_sha1_file(img->buf, img->len, blob_type, sha1);2994 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))2995 return error("the patch applies to '%s' (%s), "2996 "which does not match the "2997 "current contents.",2998 name, sha1_to_hex(sha1));2999 }3000 else {3001 /* Otherwise, the old one must be empty. */3002 if (img->len)3003 return error("the patch applies to an empty "3004 "'%s' but it is not empty", name);3005 }30063007 get_sha1_hex(patch->new_sha1_prefix, sha1);3008 if (is_null_sha1(sha1)) {3009 clear_image(img);3010 return 0; /* deletion patch */3011 }30123013 if (has_sha1_file(sha1)) {3014 /* We already have the postimage */3015 enum object_type type;3016 unsigned long size;3017 char *result;30183019 result = read_sha1_file(sha1, &type, &size);3020 if (!result)3021 return error("the necessary postimage %s for "3022 "'%s' cannot be read",3023 patch->new_sha1_prefix, name);3024 clear_image(img);3025 img->buf = result;3026 img->len = size;3027 } else {3028 /*3029 * We have verified buf matches the preimage;3030 * apply the patch data to it, which is stored3031 * in the patch->fragments->{patch,size}.3032 */3033 if (apply_binary_fragment(img, patch))3034 return error(_("binary patch does not apply to '%s'"),3035 name);30363037 /* verify that the result matches */3038 hash_sha1_file(img->buf, img->len, blob_type, sha1);3039 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))3040 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),3041 name, patch->new_sha1_prefix, sha1_to_hex(sha1));3042 }30433044 return 0;3045}30463047static int apply_fragments(struct image *img, struct patch *patch)3048{3049 struct fragment *frag = patch->fragments;3050 const char *name = patch->old_name ? patch->old_name : patch->new_name;3051 unsigned ws_rule = patch->ws_rule;3052 unsigned inaccurate_eof = patch->inaccurate_eof;3053 int nth = 0;30543055 if (patch->is_binary)3056 return apply_binary(img, patch);30573058 while (frag) {3059 nth++;3060 if (apply_one_fragment(img, frag, inaccurate_eof, ws_rule, nth)) {3061 error(_("patch failed: %s:%ld"), name, frag->oldpos);3062 if (!apply_with_reject)3063 return -1;3064 frag->rejected = 1;3065 }3066 frag = frag->next;3067 }3068 return 0;3069}30703071static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)3072{3073 if (S_ISGITLINK(mode)) {3074 strbuf_grow(buf, 100);3075 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));3076 } else {3077 enum object_type type;3078 unsigned long sz;3079 char *result;30803081 result = read_sha1_file(sha1, &type, &sz);3082 if (!result)3083 return -1;3084 /* XXX read_sha1_file NUL-terminates */3085 strbuf_attach(buf, result, sz, sz + 1);3086 }3087 return 0;3088}30893090static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)3091{3092 if (!ce)3093 return 0;3094 return read_blob_object(buf, ce->sha1, ce->ce_mode);3095}30963097static struct patch *in_fn_table(const char *name)3098{3099 struct string_list_item *item;31003101 if (name == NULL)3102 return NULL;31033104 item = string_list_lookup(&fn_table, name);3105 if (item != NULL)3106 return (struct patch *)item->util;31073108 return NULL;3109}31103111/*3112 * item->util in the filename table records the status of the path.3113 * Usually it points at a patch (whose result records the contents3114 * of it after applying it), but it could be PATH_WAS_DELETED for a3115 * path that a previously applied patch has already removed, or3116 * PATH_TO_BE_DELETED for a path that a later patch would remove.3117 *3118 * The latter is needed to deal with a case where two paths A and B3119 * are swapped by first renaming A to B and then renaming B to A;3120 * moving A to B should not be prevented due to presence of B as we3121 * will remove it in a later patch.3122 */3123#define PATH_TO_BE_DELETED ((struct patch *) -2)3124#define PATH_WAS_DELETED ((struct patch *) -1)31253126static int to_be_deleted(struct patch *patch)3127{3128 return patch == PATH_TO_BE_DELETED;3129}31303131static int was_deleted(struct patch *patch)3132{3133 return patch == PATH_WAS_DELETED;3134}31353136static void add_to_fn_table(struct patch *patch)3137{3138 struct string_list_item *item;31393140 /*3141 * Always add new_name unless patch is a deletion3142 * This should cover the cases for normal diffs,3143 * file creations and copies3144 */3145 if (patch->new_name != NULL) {3146 item = string_list_insert(&fn_table, patch->new_name);3147 item->util = patch;3148 }31493150 /*3151 * store a failure on rename/deletion cases because3152 * later chunks shouldn't patch old names3153 */3154 if ((patch->new_name == NULL) || (patch->is_rename)) {3155 item = string_list_insert(&fn_table, patch->old_name);3156 item->util = PATH_WAS_DELETED;3157 }3158}31593160static void prepare_fn_table(struct patch *patch)3161{3162 /*3163 * store information about incoming file deletion3164 */3165 while (patch) {3166 if ((patch->new_name == NULL) || (patch->is_rename)) {3167 struct string_list_item *item;3168 item = string_list_insert(&fn_table, patch->old_name);3169 item->util = PATH_TO_BE_DELETED;3170 }3171 patch = patch->next;3172 }3173}31743175static int checkout_target(struct index_state *istate,3176 struct cache_entry *ce, struct stat *st)3177{3178 struct checkout costate;31793180 memset(&costate, 0, sizeof(costate));3181 costate.base_dir = "";3182 costate.refresh_cache = 1;3183 costate.istate = istate;3184 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))3185 return error(_("cannot checkout %s"), ce->name);3186 return 0;3187}31883189static struct patch *previous_patch(struct patch *patch, int *gone)3190{3191 struct patch *previous;31923193 *gone = 0;3194 if (patch->is_copy || patch->is_rename)3195 return NULL; /* "git" patches do not depend on the order */31963197 previous = in_fn_table(patch->old_name);3198 if (!previous)3199 return NULL;32003201 if (to_be_deleted(previous))3202 return NULL; /* the deletion hasn't happened yet */32033204 if (was_deleted(previous))3205 *gone = 1;32063207 return previous;3208}32093210static int verify_index_match(const struct cache_entry *ce, struct stat *st)3211{3212 if (S_ISGITLINK(ce->ce_mode)) {3213 if (!S_ISDIR(st->st_mode))3214 return -1;3215 return 0;3216 }3217 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);3218}32193220#define SUBMODULE_PATCH_WITHOUT_INDEX 132213222static int load_patch_target(struct strbuf *buf,3223 const struct cache_entry *ce,3224 struct stat *st,3225 const char *name,3226 unsigned expected_mode)3227{3228 if (cached || check_index) {3229 if (read_file_or_gitlink(ce, buf))3230 return error(_("read of %s failed"), name);3231 } else if (name) {3232 if (S_ISGITLINK(expected_mode)) {3233 if (ce)3234 return read_file_or_gitlink(ce, buf);3235 else3236 return SUBMODULE_PATCH_WITHOUT_INDEX;3237 } else if (has_symlink_leading_path(name, strlen(name))) {3238 return error(_("reading from '%s' beyond a symbolic link"), name);3239 } else {3240 if (read_old_data(st, name, buf))3241 return error(_("read of %s failed"), name);3242 }3243 }3244 return 0;3245}32463247/*3248 * We are about to apply "patch"; populate the "image" with the3249 * current version we have, from the working tree or from the index,3250 * depending on the situation e.g. --cached/--index. If we are3251 * applying a non-git patch that incrementally updates the tree,3252 * we read from the result of a previous diff.3253 */3254static int load_preimage(struct image *image,3255 struct patch *patch, struct stat *st,3256 const struct cache_entry *ce)3257{3258 struct strbuf buf = STRBUF_INIT;3259 size_t len;3260 char *img;3261 struct patch *previous;3262 int status;32633264 previous = previous_patch(patch, &status);3265 if (status)3266 return error(_("path %s has been renamed/deleted"),3267 patch->old_name);3268 if (previous) {3269 /* We have a patched copy in memory; use that. */3270 strbuf_add(&buf, previous->result, previous->resultsize);3271 } else {3272 status = load_patch_target(&buf, ce, st,3273 patch->old_name, patch->old_mode);3274 if (status < 0)3275 return status;3276 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {3277 /*3278 * There is no way to apply subproject3279 * patch without looking at the index.3280 * NEEDSWORK: shouldn't this be flagged3281 * as an error???3282 */3283 free_fragment_list(patch->fragments);3284 patch->fragments = NULL;3285 } else if (status) {3286 return error(_("read of %s failed"), patch->old_name);3287 }3288 }32893290 img = strbuf_detach(&buf, &len);3291 prepare_image(image, img, len, !patch->is_binary);3292 return 0;3293}32943295static int three_way_merge(struct image *image,3296 char *path,3297 const unsigned char *base,3298 const unsigned char *ours,3299 const unsigned char *theirs)3300{3301 mmfile_t base_file, our_file, their_file;3302 mmbuffer_t result = { NULL };3303 int status;33043305 read_mmblob(&base_file, base);3306 read_mmblob(&our_file, ours);3307 read_mmblob(&their_file, theirs);3308 status = ll_merge(&result, path,3309 &base_file, "base",3310 &our_file, "ours",3311 &their_file, "theirs", NULL);3312 free(base_file.ptr);3313 free(our_file.ptr);3314 free(their_file.ptr);3315 if (status < 0 || !result.ptr) {3316 free(result.ptr);3317 return -1;3318 }3319 clear_image(image);3320 image->buf = result.ptr;3321 image->len = result.size;33223323 return status;3324}33253326/*3327 * When directly falling back to add/add three-way merge, we read from3328 * the current contents of the new_name. In no cases other than that3329 * this function will be called.3330 */3331static int load_current(struct image *image, struct patch *patch)3332{3333 struct strbuf buf = STRBUF_INIT;3334 int status, pos;3335 size_t len;3336 char *img;3337 struct stat st;3338 struct cache_entry *ce;3339 char *name = patch->new_name;3340 unsigned mode = patch->new_mode;33413342 if (!patch->is_new)3343 die("BUG: patch to %s is not a creation", patch->old_name);33443345 pos = cache_name_pos(name, strlen(name));3346 if (pos < 0)3347 return error(_("%s: does not exist in index"), name);3348 ce = active_cache[pos];3349 if (lstat(name, &st)) {3350 if (errno != ENOENT)3351 return error(_("%s: %s"), name, strerror(errno));3352 if (checkout_target(&the_index, ce, &st))3353 return -1;3354 }3355 if (verify_index_match(ce, &st))3356 return error(_("%s: does not match index"), name);33573358 status = load_patch_target(&buf, ce, &st, name, mode);3359 if (status < 0)3360 return status;3361 else if (status)3362 return -1;3363 img = strbuf_detach(&buf, &len);3364 prepare_image(image, img, len, !patch->is_binary);3365 return 0;3366}33673368static int try_threeway(struct image *image, struct patch *patch,3369 struct stat *st, const struct cache_entry *ce)3370{3371 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];3372 struct strbuf buf = STRBUF_INIT;3373 size_t len;3374 int status;3375 char *img;3376 struct image tmp_image;33773378 /* No point falling back to 3-way merge in these cases */3379 if (patch->is_delete ||3380 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))3381 return -1;33823383 /* Preimage the patch was prepared for */3384 if (patch->is_new)3385 write_sha1_file("", 0, blob_type, pre_sha1);3386 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||3387 read_blob_object(&buf, pre_sha1, patch->old_mode))3388 return error("repository lacks the necessary blob to fall back on 3-way merge.");33893390 fprintf(stderr, "Falling back to three-way merge...\n");33913392 img = strbuf_detach(&buf, &len);3393 prepare_image(&tmp_image, img, len, 1);3394 /* Apply the patch to get the post image */3395 if (apply_fragments(&tmp_image, patch) < 0) {3396 clear_image(&tmp_image);3397 return -1;3398 }3399 /* post_sha1[] is theirs */3400 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);3401 clear_image(&tmp_image);34023403 /* our_sha1[] is ours */3404 if (patch->is_new) {3405 if (load_current(&tmp_image, patch))3406 return error("cannot read the current contents of '%s'",3407 patch->new_name);3408 } else {3409 if (load_preimage(&tmp_image, patch, st, ce))3410 return error("cannot read the current contents of '%s'",3411 patch->old_name);3412 }3413 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);3414 clear_image(&tmp_image);34153416 /* in-core three-way merge between post and our using pre as base */3417 status = three_way_merge(image, patch->new_name,3418 pre_sha1, our_sha1, post_sha1);3419 if (status < 0) {3420 fprintf(stderr, "Failed to fall back on three-way merge...\n");3421 return status;3422 }34233424 if (status) {3425 patch->conflicted_threeway = 1;3426 if (patch->is_new)3427 hashclr(patch->threeway_stage[0]);3428 else3429 hashcpy(patch->threeway_stage[0], pre_sha1);3430 hashcpy(patch->threeway_stage[1], our_sha1);3431 hashcpy(patch->threeway_stage[2], post_sha1);3432 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);3433 } else {3434 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);3435 }3436 return 0;3437}34383439static int apply_data(struct patch *patch, struct stat *st, const struct cache_entry *ce)3440{3441 struct image image;34423443 if (load_preimage(&image, patch, st, ce) < 0)3444 return -1;34453446 if (patch->direct_to_threeway ||3447 apply_fragments(&image, patch) < 0) {3448 /* Note: with --reject, apply_fragments() returns 0 */3449 if (!threeway || try_threeway(&image, patch, st, ce) < 0)3450 return -1;3451 }3452 patch->result = image.buf;3453 patch->resultsize = image.len;3454 add_to_fn_table(patch);3455 free(image.line_allocated);34563457 if (0 < patch->is_delete && patch->resultsize)3458 return error(_("removal patch leaves file contents"));34593460 return 0;3461}34623463/*3464 * If "patch" that we are looking at modifies or deletes what we have,3465 * we would want it not to lose any local modification we have, either3466 * in the working tree or in the index.3467 *3468 * This also decides if a non-git patch is a creation patch or a3469 * modification to an existing empty file. We do not check the state3470 * of the current tree for a creation patch in this function; the caller3471 * check_patch() separately makes sure (and errors out otherwise) that3472 * the path the patch creates does not exist in the current tree.3473 */3474static int check_preimage(struct patch *patch, struct cache_entry **ce, struct stat *st)3475{3476 const char *old_name = patch->old_name;3477 struct patch *previous = NULL;3478 int stat_ret = 0, status;3479 unsigned st_mode = 0;34803481 if (!old_name)3482 return 0;34833484 assert(patch->is_new <= 0);3485 previous = previous_patch(patch, &status);34863487 if (status)3488 return error(_("path %s has been renamed/deleted"), old_name);3489 if (previous) {3490 st_mode = previous->new_mode;3491 } else if (!cached) {3492 stat_ret = lstat(old_name, st);3493 if (stat_ret && errno != ENOENT)3494 return error(_("%s: %s"), old_name, strerror(errno));3495 }34963497 if (check_index && !previous) {3498 int pos = cache_name_pos(old_name, strlen(old_name));3499 if (pos < 0) {3500 if (patch->is_new < 0)3501 goto is_new;3502 return error(_("%s: does not exist in index"), old_name);3503 }3504 *ce = active_cache[pos];3505 if (stat_ret < 0) {3506 if (checkout_target(&the_index, *ce, st))3507 return -1;3508 }3509 if (!cached && verify_index_match(*ce, st))3510 return error(_("%s: does not match index"), old_name);3511 if (cached)3512 st_mode = (*ce)->ce_mode;3513 } else if (stat_ret < 0) {3514 if (patch->is_new < 0)3515 goto is_new;3516 return error(_("%s: %s"), old_name, strerror(errno));3517 }35183519 if (!cached && !previous)3520 st_mode = ce_mode_from_stat(*ce, st->st_mode);35213522 if (patch->is_new < 0)3523 patch->is_new = 0;3524 if (!patch->old_mode)3525 patch->old_mode = st_mode;3526 if ((st_mode ^ patch->old_mode) & S_IFMT)3527 return error(_("%s: wrong type"), old_name);3528 if (st_mode != patch->old_mode)3529 warning(_("%s has type %o, expected %o"),3530 old_name, st_mode, patch->old_mode);3531 if (!patch->new_mode && !patch->is_delete)3532 patch->new_mode = st_mode;3533 return 0;35343535 is_new:3536 patch->is_new = 1;3537 patch->is_delete = 0;3538 free(patch->old_name);3539 patch->old_name = NULL;3540 return 0;3541}354235433544#define EXISTS_IN_INDEX 13545#define EXISTS_IN_WORKTREE 235463547static int check_to_create(const char *new_name, int ok_if_exists)3548{3549 struct stat nst;35503551 if (check_index &&3552 cache_name_pos(new_name, strlen(new_name)) >= 0 &&3553 !ok_if_exists)3554 return EXISTS_IN_INDEX;3555 if (cached)3556 return 0;35573558 if (!lstat(new_name, &nst)) {3559 if (S_ISDIR(nst.st_mode) || ok_if_exists)3560 return 0;3561 /*3562 * A leading component of new_name might be a symlink3563 * that is going to be removed with this patch, but3564 * still pointing at somewhere that has the path.3565 * In such a case, path "new_name" does not exist as3566 * far as git is concerned.3567 */3568 if (has_symlink_leading_path(new_name, strlen(new_name)))3569 return 0;35703571 return EXISTS_IN_WORKTREE;3572 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {3573 return error("%s: %s", new_name, strerror(errno));3574 }3575 return 0;3576}35773578/*3579 * We need to keep track of how symlinks in the preimage are3580 * manipulated by the patches. A patch to add a/b/c where a/b3581 * is a symlink should not be allowed to affect the directory3582 * the symlink points at, but if the same patch removes a/b,3583 * it is perfectly fine, as the patch removes a/b to make room3584 * to create a directory a/b so that a/b/c can be created.3585 */3586static struct string_list symlink_changes;3587#define SYMLINK_GOES_AWAY 013588#define SYMLINK_IN_RESULT 0235893590static uintptr_t register_symlink_changes(const char *path, uintptr_t what)3591{3592 struct string_list_item *ent;35933594 ent = string_list_lookup(&symlink_changes, path);3595 if (!ent) {3596 ent = string_list_insert(&symlink_changes, path);3597 ent->util = (void *)0;3598 }3599 ent->util = (void *)(what | ((uintptr_t)ent->util));3600 return (uintptr_t)ent->util;3601}36023603static uintptr_t check_symlink_changes(const char *path)3604{3605 struct string_list_item *ent;36063607 ent = string_list_lookup(&symlink_changes, path);3608 if (!ent)3609 return 0;3610 return (uintptr_t)ent->util;3611}36123613static void prepare_symlink_changes(struct patch *patch)3614{3615 for ( ; patch; patch = patch->next) {3616 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&3617 (patch->is_rename || patch->is_delete))3618 /* the symlink at patch->old_name is removed */3619 register_symlink_changes(patch->old_name, SYMLINK_GOES_AWAY);36203621 if (patch->new_name && S_ISLNK(patch->new_mode))3622 /* the symlink at patch->new_name is created or remains */3623 register_symlink_changes(patch->new_name, SYMLINK_IN_RESULT);3624 }3625}36263627static int path_is_beyond_symlink_1(struct strbuf *name)3628{3629 do {3630 unsigned int change;36313632 while (--name->len && name->buf[name->len] != '/')3633 ; /* scan backwards */3634 if (!name->len)3635 break;3636 name->buf[name->len] = '\0';3637 change = check_symlink_changes(name->buf);3638 if (change & SYMLINK_IN_RESULT)3639 return 1;3640 if (change & SYMLINK_GOES_AWAY)3641 /*3642 * This cannot be "return 0", because we may3643 * see a new one created at a higher level.3644 */3645 continue;36463647 /* otherwise, check the preimage */3648 if (check_index) {3649 struct cache_entry *ce;36503651 ce = cache_file_exists(name->buf, name->len, ignore_case);3652 if (ce && S_ISLNK(ce->ce_mode))3653 return 1;3654 } else {3655 struct stat st;3656 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))3657 return 1;3658 }3659 } while (1);3660 return 0;3661}36623663static int path_is_beyond_symlink(const char *name_)3664{3665 int ret;3666 struct strbuf name = STRBUF_INIT;36673668 assert(*name_ != '\0');3669 strbuf_addstr(&name, name_);3670 ret = path_is_beyond_symlink_1(&name);3671 strbuf_release(&name);36723673 return ret;3674}36753676static void die_on_unsafe_path(struct patch *patch)3677{3678 const char *old_name = NULL;3679 const char *new_name = NULL;3680 if (patch->is_delete)3681 old_name = patch->old_name;3682 else if (!patch->is_new && !patch->is_copy)3683 old_name = patch->old_name;3684 if (!patch->is_delete)3685 new_name = patch->new_name;36863687 if (old_name && !verify_path(old_name))3688 die(_("invalid path '%s'"), old_name);3689 if (new_name && !verify_path(new_name))3690 die(_("invalid path '%s'"), new_name);3691}36923693/*3694 * Check and apply the patch in-core; leave the result in patch->result3695 * for the caller to write it out to the final destination.3696 */3697static int check_patch(struct patch *patch)3698{3699 struct stat st;3700 const char *old_name = patch->old_name;3701 const char *new_name = patch->new_name;3702 const char *name = old_name ? old_name : new_name;3703 struct cache_entry *ce = NULL;3704 struct patch *tpatch;3705 int ok_if_exists;3706 int status;37073708 patch->rejected = 1; /* we will drop this after we succeed */37093710 status = check_preimage(patch, &ce, &st);3711 if (status)3712 return status;3713 old_name = patch->old_name;37143715 /*3716 * A type-change diff is always split into a patch to delete3717 * old, immediately followed by a patch to create new (see3718 * diff.c::run_diff()); in such a case it is Ok that the entry3719 * to be deleted by the previous patch is still in the working3720 * tree and in the index.3721 *3722 * A patch to swap-rename between A and B would first rename A3723 * to B and then rename B to A. While applying the first one,3724 * the presence of B should not stop A from getting renamed to3725 * B; ask to_be_deleted() about the later rename. Removal of3726 * B and rename from A to B is handled the same way by asking3727 * was_deleted().3728 */3729 if ((tpatch = in_fn_table(new_name)) &&3730 (was_deleted(tpatch) || to_be_deleted(tpatch)))3731 ok_if_exists = 1;3732 else3733 ok_if_exists = 0;37343735 if (new_name &&3736 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {3737 int err = check_to_create(new_name, ok_if_exists);37383739 if (err && threeway) {3740 patch->direct_to_threeway = 1;3741 } else switch (err) {3742 case 0:3743 break; /* happy */3744 case EXISTS_IN_INDEX:3745 return error(_("%s: already exists in index"), new_name);3746 break;3747 case EXISTS_IN_WORKTREE:3748 return error(_("%s: already exists in working directory"),3749 new_name);3750 default:3751 return err;3752 }37533754 if (!patch->new_mode) {3755 if (0 < patch->is_new)3756 patch->new_mode = S_IFREG | 0644;3757 else3758 patch->new_mode = patch->old_mode;3759 }3760 }37613762 if (new_name && old_name) {3763 int same = !strcmp(old_name, new_name);3764 if (!patch->new_mode)3765 patch->new_mode = patch->old_mode;3766 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {3767 if (same)3768 return error(_("new mode (%o) of %s does not "3769 "match old mode (%o)"),3770 patch->new_mode, new_name,3771 patch->old_mode);3772 else3773 return error(_("new mode (%o) of %s does not "3774 "match old mode (%o) of %s"),3775 patch->new_mode, new_name,3776 patch->old_mode, old_name);3777 }3778 }37793780 if (!unsafe_paths)3781 die_on_unsafe_path(patch);37823783 /*3784 * An attempt to read from or delete a path that is beyond a3785 * symbolic link will be prevented by load_patch_target() that3786 * is called at the beginning of apply_data() so we do not3787 * have to worry about a patch marked with "is_delete" bit3788 * here. We however need to make sure that the patch result3789 * is not deposited to a path that is beyond a symbolic link3790 * here.3791 */3792 if (!patch->is_delete && path_is_beyond_symlink(patch->new_name))3793 return error(_("affected file '%s' is beyond a symbolic link"),3794 patch->new_name);37953796 if (apply_data(patch, &st, ce) < 0)3797 return error(_("%s: patch does not apply"), name);3798 patch->rejected = 0;3799 return 0;3800}38013802static int check_patch_list(struct patch *patch)3803{3804 int err = 0;38053806 prepare_symlink_changes(patch);3807 prepare_fn_table(patch);3808 while (patch) {3809 if (apply_verbosely)3810 say_patch_name(stderr,3811 _("Checking patch %s..."), patch);3812 err |= check_patch(patch);3813 patch = patch->next;3814 }3815 return err;3816}38173818/* This function tries to read the sha1 from the current index */3819static int get_current_sha1(const char *path, unsigned char *sha1)3820{3821 int pos;38223823 if (read_cache() < 0)3824 return -1;3825 pos = cache_name_pos(path, strlen(path));3826 if (pos < 0)3827 return -1;3828 hashcpy(sha1, active_cache[pos]->sha1);3829 return 0;3830}38313832static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])3833{3834 /*3835 * A usable gitlink patch has only one fragment (hunk) that looks like:3836 * @@ -1 +1 @@3837 * -Subproject commit <old sha1>3838 * +Subproject commit <new sha1>3839 * or3840 * @@ -1 +0,0 @@3841 * -Subproject commit <old sha1>3842 * for a removal patch.3843 */3844 struct fragment *hunk = p->fragments;3845 static const char heading[] = "-Subproject commit ";3846 char *preimage;38473848 if (/* does the patch have only one hunk? */3849 hunk && !hunk->next &&3850 /* is its preimage one line? */3851 hunk->oldpos == 1 && hunk->oldlines == 1 &&3852 /* does preimage begin with the heading? */3853 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&3854 starts_with(++preimage, heading) &&3855 /* does it record full SHA-1? */3856 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&3857 preimage[sizeof(heading) + 40 - 1] == '\n' &&3858 /* does the abbreviated name on the index line agree with it? */3859 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))3860 return 0; /* it all looks fine */38613862 /* we may have full object name on the index line */3863 return get_sha1_hex(p->old_sha1_prefix, sha1);3864}38653866/* Build an index that contains the just the files needed for a 3way merge */3867static void build_fake_ancestor(struct patch *list, const char *filename)3868{3869 struct patch *patch;3870 struct index_state result = { NULL };3871 static struct lock_file lock;38723873 /* Once we start supporting the reverse patch, it may be3874 * worth showing the new sha1 prefix, but until then...3875 */3876 for (patch = list; patch; patch = patch->next) {3877 unsigned char sha1[20];3878 struct cache_entry *ce;3879 const char *name;38803881 name = patch->old_name ? patch->old_name : patch->new_name;3882 if (0 < patch->is_new)3883 continue;38843885 if (S_ISGITLINK(patch->old_mode)) {3886 if (!preimage_sha1_in_gitlink_patch(patch, sha1))3887 ; /* ok, the textual part looks sane */3888 else3889 die("sha1 information is lacking or useless for submodule %s",3890 name);3891 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {3892 ; /* ok */3893 } else if (!patch->lines_added && !patch->lines_deleted) {3894 /* mode-only change: update the current */3895 if (get_current_sha1(patch->old_name, sha1))3896 die("mode change for %s, which is not "3897 "in current HEAD", name);3898 } else3899 die("sha1 information is lacking or useless "3900 "(%s).", name);39013902 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);3903 if (!ce)3904 die(_("make_cache_entry failed for path '%s'"), name);3905 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3906 die ("Could not add %s to temporary index", name);3907 }39083909 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);3910 if (write_locked_index(&result, &lock, COMMIT_LOCK))3911 die ("Could not write temporary index to %s", filename);39123913 discard_index(&result);3914}39153916static void stat_patch_list(struct patch *patch)3917{3918 int files, adds, dels;39193920 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3921 files++;3922 adds += patch->lines_added;3923 dels += patch->lines_deleted;3924 show_stats(patch);3925 }39263927 print_stat_summary(stdout, files, adds, dels);3928}39293930static void numstat_patch_list(struct patch *patch)3931{3932 for ( ; patch; patch = patch->next) {3933 const char *name;3934 name = patch->new_name ? patch->new_name : patch->old_name;3935 if (patch->is_binary)3936 printf("-\t-\t");3937 else3938 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3939 write_name_quoted(name, stdout, line_termination);3940 }3941}39423943static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3944{3945 if (mode)3946 printf(" %s mode %06o %s\n", newdelete, mode, name);3947 else3948 printf(" %s %s\n", newdelete, name);3949}39503951static void show_mode_change(struct patch *p, int show_name)3952{3953 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3954 if (show_name)3955 printf(" mode change %06o => %06o %s\n",3956 p->old_mode, p->new_mode, p->new_name);3957 else3958 printf(" mode change %06o => %06o\n",3959 p->old_mode, p->new_mode);3960 }3961}39623963static void show_rename_copy(struct patch *p)3964{3965 const char *renamecopy = p->is_rename ? "rename" : "copy";3966 const char *old, *new;39673968 /* Find common prefix */3969 old = p->old_name;3970 new = p->new_name;3971 while (1) {3972 const char *slash_old, *slash_new;3973 slash_old = strchr(old, '/');3974 slash_new = strchr(new, '/');3975 if (!slash_old ||3976 !slash_new ||3977 slash_old - old != slash_new - new ||3978 memcmp(old, new, slash_new - new))3979 break;3980 old = slash_old + 1;3981 new = slash_new + 1;3982 }3983 /* p->old_name thru old is the common prefix, and old and new3984 * through the end of names are renames3985 */3986 if (old != p->old_name)3987 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,3988 (int)(old - p->old_name), p->old_name,3989 old, new, p->score);3990 else3991 printf(" %s %s => %s (%d%%)\n", renamecopy,3992 p->old_name, p->new_name, p->score);3993 show_mode_change(p, 0);3994}39953996static void summary_patch_list(struct patch *patch)3997{3998 struct patch *p;39994000 for (p = patch; p; p = p->next) {4001 if (p->is_new)4002 show_file_mode_name("create", p->new_mode, p->new_name);4003 else if (p->is_delete)4004 show_file_mode_name("delete", p->old_mode, p->old_name);4005 else {4006 if (p->is_rename || p->is_copy)4007 show_rename_copy(p);4008 else {4009 if (p->score) {4010 printf(" rewrite %s (%d%%)\n",4011 p->new_name, p->score);4012 show_mode_change(p, 0);4013 }4014 else4015 show_mode_change(p, 1);4016 }4017 }4018 }4019}40204021static void patch_stats(struct patch *patch)4022{4023 int lines = patch->lines_added + patch->lines_deleted;40244025 if (lines > max_change)4026 max_change = lines;4027 if (patch->old_name) {4028 int len = quote_c_style(patch->old_name, NULL, NULL, 0);4029 if (!len)4030 len = strlen(patch->old_name);4031 if (len > max_len)4032 max_len = len;4033 }4034 if (patch->new_name) {4035 int len = quote_c_style(patch->new_name, NULL, NULL, 0);4036 if (!len)4037 len = strlen(patch->new_name);4038 if (len > max_len)4039 max_len = len;4040 }4041}40424043static void remove_file(struct patch *patch, int rmdir_empty)4044{4045 if (update_index) {4046 if (remove_file_from_cache(patch->old_name) < 0)4047 die(_("unable to remove %s from index"), patch->old_name);4048 }4049 if (!cached) {4050 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {4051 remove_path(patch->old_name);4052 }4053 }4054}40554056static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)4057{4058 struct stat st;4059 struct cache_entry *ce;4060 int namelen = strlen(path);4061 unsigned ce_size = cache_entry_size(namelen);40624063 if (!update_index)4064 return;40654066 ce = xcalloc(1, ce_size);4067 memcpy(ce->name, path, namelen);4068 ce->ce_mode = create_ce_mode(mode);4069 ce->ce_flags = create_ce_flags(0);4070 ce->ce_namelen = namelen;4071 if (S_ISGITLINK(mode)) {4072 const char *s;40734074 if (!skip_prefix(buf, "Subproject commit ", &s) ||4075 get_sha1_hex(s, ce->sha1))4076 die(_("corrupt patch for submodule %s"), path);4077 } else {4078 if (!cached) {4079 if (lstat(path, &st) < 0)4080 die_errno(_("unable to stat newly created file '%s'"),4081 path);4082 fill_stat_cache_info(ce, &st);4083 }4084 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)4085 die(_("unable to create backing store for newly created file %s"), path);4086 }4087 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4088 die(_("unable to add cache entry for %s"), path);4089}40904091static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)4092{4093 int fd;4094 struct strbuf nbuf = STRBUF_INIT;40954096 if (S_ISGITLINK(mode)) {4097 struct stat st;4098 if (!lstat(path, &st) && S_ISDIR(st.st_mode))4099 return 0;4100 return mkdir(path, 0777);4101 }41024103 if (has_symlinks && S_ISLNK(mode))4104 /* Although buf:size is counted string, it also is NUL4105 * terminated.4106 */4107 return symlink(buf, path);41084109 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);4110 if (fd < 0)4111 return -1;41124113 if (convert_to_working_tree(path, buf, size, &nbuf)) {4114 size = nbuf.len;4115 buf = nbuf.buf;4116 }4117 write_or_die(fd, buf, size);4118 strbuf_release(&nbuf);41194120 if (close(fd) < 0)4121 die_errno(_("closing file '%s'"), path);4122 return 0;4123}41244125/*4126 * We optimistically assume that the directories exist,4127 * which is true 99% of the time anyway. If they don't,4128 * we create them and try again.4129 */4130static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)4131{4132 if (cached)4133 return;4134 if (!try_create_file(path, mode, buf, size))4135 return;41364137 if (errno == ENOENT) {4138 if (safe_create_leading_directories(path))4139 return;4140 if (!try_create_file(path, mode, buf, size))4141 return;4142 }41434144 if (errno == EEXIST || errno == EACCES) {4145 /* We may be trying to create a file where a directory4146 * used to be.4147 */4148 struct stat st;4149 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))4150 errno = EEXIST;4151 }41524153 if (errno == EEXIST) {4154 unsigned int nr = getpid();41554156 for (;;) {4157 char newpath[PATH_MAX];4158 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);4159 if (!try_create_file(newpath, mode, buf, size)) {4160 if (!rename(newpath, path))4161 return;4162 unlink_or_warn(newpath);4163 break;4164 }4165 if (errno != EEXIST)4166 break;4167 ++nr;4168 }4169 }4170 die_errno(_("unable to write file '%s' mode %o"), path, mode);4171}41724173static void add_conflicted_stages_file(struct patch *patch)4174{4175 int stage, namelen;4176 unsigned ce_size, mode;4177 struct cache_entry *ce;41784179 if (!update_index)4180 return;4181 namelen = strlen(patch->new_name);4182 ce_size = cache_entry_size(namelen);4183 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);41844185 remove_file_from_cache(patch->new_name);4186 for (stage = 1; stage < 4; stage++) {4187 if (is_null_sha1(patch->threeway_stage[stage - 1]))4188 continue;4189 ce = xcalloc(1, ce_size);4190 memcpy(ce->name, patch->new_name, namelen);4191 ce->ce_mode = create_ce_mode(mode);4192 ce->ce_flags = create_ce_flags(stage);4193 ce->ce_namelen = namelen;4194 hashcpy(ce->sha1, patch->threeway_stage[stage - 1]);4195 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4196 die(_("unable to add cache entry for %s"), patch->new_name);4197 }4198}41994200static void create_file(struct patch *patch)4201{4202 char *path = patch->new_name;4203 unsigned mode = patch->new_mode;4204 unsigned long size = patch->resultsize;4205 char *buf = patch->result;42064207 if (!mode)4208 mode = S_IFREG | 0644;4209 create_one_file(path, mode, buf, size);42104211 if (patch->conflicted_threeway)4212 add_conflicted_stages_file(patch);4213 else4214 add_index_file(path, mode, buf, size);4215}42164217/* phase zero is to remove, phase one is to create */4218static void write_out_one_result(struct patch *patch, int phase)4219{4220 if (patch->is_delete > 0) {4221 if (phase == 0)4222 remove_file(patch, 1);4223 return;4224 }4225 if (patch->is_new > 0 || patch->is_copy) {4226 if (phase == 1)4227 create_file(patch);4228 return;4229 }4230 /*4231 * Rename or modification boils down to the same4232 * thing: remove the old, write the new4233 */4234 if (phase == 0)4235 remove_file(patch, patch->is_rename);4236 if (phase == 1)4237 create_file(patch);4238}42394240static int write_out_one_reject(struct patch *patch)4241{4242 FILE *rej;4243 char namebuf[PATH_MAX];4244 struct fragment *frag;4245 int cnt = 0;4246 struct strbuf sb = STRBUF_INIT;42474248 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {4249 if (!frag->rejected)4250 continue;4251 cnt++;4252 }42534254 if (!cnt) {4255 if (apply_verbosely)4256 say_patch_name(stderr,4257 _("Applied patch %s cleanly."), patch);4258 return 0;4259 }42604261 /* This should not happen, because a removal patch that leaves4262 * contents are marked "rejected" at the patch level.4263 */4264 if (!patch->new_name)4265 die(_("internal error"));42664267 /* Say this even without --verbose */4268 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",4269 "Applying patch %%s with %d rejects...",4270 cnt),4271 cnt);4272 say_patch_name(stderr, sb.buf, patch);4273 strbuf_release(&sb);42744275 cnt = strlen(patch->new_name);4276 if (ARRAY_SIZE(namebuf) <= cnt + 5) {4277 cnt = ARRAY_SIZE(namebuf) - 5;4278 warning(_("truncating .rej filename to %.*s.rej"),4279 cnt - 1, patch->new_name);4280 }4281 memcpy(namebuf, patch->new_name, cnt);4282 memcpy(namebuf + cnt, ".rej", 5);42834284 rej = fopen(namebuf, "w");4285 if (!rej)4286 return error(_("cannot open %s: %s"), namebuf, strerror(errno));42874288 /* Normal git tools never deal with .rej, so do not pretend4289 * this is a git patch by saying --git or giving extended4290 * headers. While at it, maybe please "kompare" that wants4291 * the trailing TAB and some garbage at the end of line ;-).4292 */4293 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",4294 patch->new_name, patch->new_name);4295 for (cnt = 1, frag = patch->fragments;4296 frag;4297 cnt++, frag = frag->next) {4298 if (!frag->rejected) {4299 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);4300 continue;4301 }4302 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);4303 fprintf(rej, "%.*s", frag->size, frag->patch);4304 if (frag->patch[frag->size-1] != '\n')4305 fputc('\n', rej);4306 }4307 fclose(rej);4308 return -1;4309}43104311static int write_out_results(struct patch *list)4312{4313 int phase;4314 int errs = 0;4315 struct patch *l;4316 struct string_list cpath = STRING_LIST_INIT_DUP;43174318 for (phase = 0; phase < 2; phase++) {4319 l = list;4320 while (l) {4321 if (l->rejected)4322 errs = 1;4323 else {4324 write_out_one_result(l, phase);4325 if (phase == 1) {4326 if (write_out_one_reject(l))4327 errs = 1;4328 if (l->conflicted_threeway) {4329 string_list_append(&cpath, l->new_name);4330 errs = 1;4331 }4332 }4333 }4334 l = l->next;4335 }4336 }43374338 if (cpath.nr) {4339 struct string_list_item *item;43404341 string_list_sort(&cpath);4342 for_each_string_list_item(item, &cpath)4343 fprintf(stderr, "U %s\n", item->string);4344 string_list_clear(&cpath, 0);43454346 rerere(0);4347 }43484349 return errs;4350}43514352static struct lock_file lock_file;43534354#define INACCURATE_EOF (1<<0)4355#define RECOUNT (1<<1)43564357static int apply_patch(int fd, const char *filename, int options)4358{4359 size_t offset;4360 struct strbuf buf = STRBUF_INIT; /* owns the patch text */4361 struct patch *list = NULL, **listp = &list;4362 int skipped_patch = 0;43634364 patch_input_file = filename;4365 read_patch_file(&buf, fd);4366 offset = 0;4367 while (offset < buf.len) {4368 struct patch *patch;4369 int nr;43704371 patch = xcalloc(1, sizeof(*patch));4372 patch->inaccurate_eof = !!(options & INACCURATE_EOF);4373 patch->recount = !!(options & RECOUNT);4374 nr = parse_chunk(buf.buf + offset, buf.len - offset, patch);4375 if (nr < 0)4376 break;4377 if (apply_in_reverse)4378 reverse_patches(patch);4379 if (use_patch(patch)) {4380 patch_stats(patch);4381 *listp = patch;4382 listp = &patch->next;4383 }4384 else {4385 free_patch(patch);4386 skipped_patch++;4387 }4388 offset += nr;4389 }43904391 if (!list && !skipped_patch)4392 die(_("unrecognized input"));43934394 if (whitespace_error && (ws_error_action == die_on_ws_error))4395 apply = 0;43964397 update_index = check_index && apply;4398 if (update_index && newfd < 0)4399 newfd = hold_locked_index(&lock_file, 1);44004401 if (check_index) {4402 if (read_cache() < 0)4403 die(_("unable to read index file"));4404 }44054406 if ((check || apply) &&4407 check_patch_list(list) < 0 &&4408 !apply_with_reject)4409 exit(1);44104411 if (apply && write_out_results(list)) {4412 if (apply_with_reject)4413 exit(1);4414 /* with --3way, we still need to write the index out */4415 return 1;4416 }44174418 if (fake_ancestor)4419 build_fake_ancestor(list, fake_ancestor);44204421 if (diffstat)4422 stat_patch_list(list);44234424 if (numstat)4425 numstat_patch_list(list);44264427 if (summary)4428 summary_patch_list(list);44294430 free_patch_list(list);4431 strbuf_release(&buf);4432 string_list_clear(&fn_table, 0);4433 return 0;4434}44354436static void git_apply_config(void)4437{4438 git_config_get_string_const("apply.whitespace", &apply_default_whitespace);4439 git_config_get_string_const("apply.ignorewhitespace", &apply_default_ignorewhitespace);4440 git_config(git_default_config, NULL);4441}44424443static int option_parse_exclude(const struct option *opt,4444 const char *arg, int unset)4445{4446 add_name_limit(arg, 1);4447 return 0;4448}44494450static int option_parse_include(const struct option *opt,4451 const char *arg, int unset)4452{4453 add_name_limit(arg, 0);4454 has_include = 1;4455 return 0;4456}44574458static int option_parse_p(const struct option *opt,4459 const char *arg, int unset)4460{4461 p_value = atoi(arg);4462 p_value_known = 1;4463 return 0;4464}44654466static int option_parse_z(const struct option *opt,4467 const char *arg, int unset)4468{4469 if (unset)4470 line_termination = '\n';4471 else4472 line_termination = 0;4473 return 0;4474}44754476static int option_parse_space_change(const struct option *opt,4477 const char *arg, int unset)4478{4479 if (unset)4480 ws_ignore_action = ignore_ws_none;4481 else4482 ws_ignore_action = ignore_ws_change;4483 return 0;4484}44854486static int option_parse_whitespace(const struct option *opt,4487 const char *arg, int unset)4488{4489 const char **whitespace_option = opt->value;44904491 *whitespace_option = arg;4492 parse_whitespace_option(arg);4493 return 0;4494}44954496static int option_parse_directory(const struct option *opt,4497 const char *arg, int unset)4498{4499 root_len = strlen(arg);4500 if (root_len && arg[root_len - 1] != '/') {4501 char *new_root;4502 root = new_root = xmalloc(root_len + 2);4503 strcpy(new_root, arg);4504 strcpy(new_root + root_len++, "/");4505 } else4506 root = arg;4507 return 0;4508}45094510int cmd_apply(int argc, const char **argv, const char *prefix_)4511{4512 int i;4513 int errs = 0;4514 int is_not_gitdir = !startup_info->have_repository;4515 int force_apply = 0;45164517 const char *whitespace_option = NULL;45184519 struct option builtin_apply_options[] = {4520 { OPTION_CALLBACK, 0, "exclude", NULL, N_("path"),4521 N_("don't apply changes matching the given path"),4522 0, option_parse_exclude },4523 { OPTION_CALLBACK, 0, "include", NULL, N_("path"),4524 N_("apply changes matching the given path"),4525 0, option_parse_include },4526 { OPTION_CALLBACK, 'p', NULL, NULL, N_("num"),4527 N_("remove <num> leading slashes from traditional diff paths"),4528 0, option_parse_p },4529 OPT_BOOL(0, "no-add", &no_add,4530 N_("ignore additions made by the patch")),4531 OPT_BOOL(0, "stat", &diffstat,4532 N_("instead of applying the patch, output diffstat for the input")),4533 OPT_NOOP_NOARG(0, "allow-binary-replacement"),4534 OPT_NOOP_NOARG(0, "binary"),4535 OPT_BOOL(0, "numstat", &numstat,4536 N_("show number of added and deleted lines in decimal notation")),4537 OPT_BOOL(0, "summary", &summary,4538 N_("instead of applying the patch, output a summary for the input")),4539 OPT_BOOL(0, "check", &check,4540 N_("instead of applying the patch, see if the patch is applicable")),4541 OPT_BOOL(0, "index", &check_index,4542 N_("make sure the patch is applicable to the current index")),4543 OPT_BOOL(0, "cached", &cached,4544 N_("apply a patch without touching the working tree")),4545 OPT_BOOL(0, "unsafe-paths", &unsafe_paths,4546 N_("accept a patch that touches outside the working area")),4547 OPT_BOOL(0, "apply", &force_apply,4548 N_("also apply the patch (use with --stat/--summary/--check)")),4549 OPT_BOOL('3', "3way", &threeway,4550 N_( "attempt three-way merge if a patch does not apply")),4551 OPT_FILENAME(0, "build-fake-ancestor", &fake_ancestor,4552 N_("build a temporary index based on embedded index information")),4553 { OPTION_CALLBACK, 'z', NULL, NULL, NULL,4554 N_("paths are separated with NUL character"),4555 PARSE_OPT_NOARG, option_parse_z },4556 OPT_INTEGER('C', NULL, &p_context,4557 N_("ensure at least <n> lines of context match")),4558 { OPTION_CALLBACK, 0, "whitespace", &whitespace_option, N_("action"),4559 N_("detect new or modified lines that have whitespace errors"),4560 0, option_parse_whitespace },4561 { OPTION_CALLBACK, 0, "ignore-space-change", NULL, NULL,4562 N_("ignore changes in whitespace when finding context"),4563 PARSE_OPT_NOARG, option_parse_space_change },4564 { OPTION_CALLBACK, 0, "ignore-whitespace", NULL, NULL,4565 N_("ignore changes in whitespace when finding context"),4566 PARSE_OPT_NOARG, option_parse_space_change },4567 OPT_BOOL('R', "reverse", &apply_in_reverse,4568 N_("apply the patch in reverse")),4569 OPT_BOOL(0, "unidiff-zero", &unidiff_zero,4570 N_("don't expect at least one line of context")),4571 OPT_BOOL(0, "reject", &apply_with_reject,4572 N_("leave the rejected hunks in corresponding *.rej files")),4573 OPT_BOOL(0, "allow-overlap", &allow_overlap,4574 N_("allow overlapping hunks")),4575 OPT__VERBOSE(&apply_verbosely, N_("be verbose")),4576 OPT_BIT(0, "inaccurate-eof", &options,4577 N_("tolerate incorrectly detected missing new-line at the end of file"),4578 INACCURATE_EOF),4579 OPT_BIT(0, "recount", &options,4580 N_("do not trust the line counts in the hunk headers"),4581 RECOUNT),4582 { OPTION_CALLBACK, 0, "directory", NULL, N_("root"),4583 N_("prepend <root> to all filenames"),4584 0, option_parse_directory },4585 OPT_END()4586 };45874588 prefix = prefix_;4589 prefix_length = prefix ? strlen(prefix) : 0;4590 git_apply_config();4591 if (apply_default_whitespace)4592 parse_whitespace_option(apply_default_whitespace);4593 if (apply_default_ignorewhitespace)4594 parse_ignorewhitespace_option(apply_default_ignorewhitespace);45954596 argc = parse_options(argc, argv, prefix, builtin_apply_options,4597 apply_usage, 0);45984599 if (apply_with_reject && threeway)4600 die("--reject and --3way cannot be used together.");4601 if (cached && threeway)4602 die("--cached and --3way cannot be used together.");4603 if (threeway) {4604 if (is_not_gitdir)4605 die(_("--3way outside a repository"));4606 check_index = 1;4607 }4608 if (apply_with_reject)4609 apply = apply_verbosely = 1;4610 if (!force_apply && (diffstat || numstat || summary || check || fake_ancestor))4611 apply = 0;4612 if (check_index && is_not_gitdir)4613 die(_("--index outside a repository"));4614 if (cached) {4615 if (is_not_gitdir)4616 die(_("--cached outside a repository"));4617 check_index = 1;4618 }4619 if (check_index)4620 unsafe_paths = 0;46214622 for (i = 0; i < argc; i++) {4623 const char *arg = argv[i];4624 int fd;46254626 if (!strcmp(arg, "-")) {4627 errs |= apply_patch(0, "<stdin>", options);4628 read_stdin = 0;4629 continue;4630 } else if (0 < prefix_length)4631 arg = prefix_filename(prefix, prefix_length, arg);46324633 fd = open(arg, O_RDONLY);4634 if (fd < 0)4635 die_errno(_("can't open patch '%s'"), arg);4636 read_stdin = 0;4637 set_default_whitespace_mode(whitespace_option);4638 errs |= apply_patch(fd, arg, options);4639 close(fd);4640 }4641 set_default_whitespace_mode(whitespace_option);4642 if (read_stdin)4643 errs |= apply_patch(0, "<stdin>", options);4644 if (whitespace_error) {4645 if (squelch_whitespace_errors &&4646 squelch_whitespace_errors < whitespace_error) {4647 int squelched =4648 whitespace_error - squelch_whitespace_errors;4649 warning(Q_("squelched %d whitespace error",4650 "squelched %d whitespace errors",4651 squelched),4652 squelched);4653 }4654 if (ws_error_action == die_on_ws_error)4655 die(Q_("%d line adds whitespace errors.",4656 "%d lines add whitespace errors.",4657 whitespace_error),4658 whitespace_error);4659 if (applied_after_fixing_ws && apply)4660 warning("%d line%s applied after"4661 " fixing whitespace errors.",4662 applied_after_fixing_ws,4663 applied_after_fixing_ws == 1 ? "" : "s");4664 else if (whitespace_error)4665 warning(Q_("%d line adds whitespace errors.",4666 "%d lines add whitespace errors.",4667 whitespace_error),4668 whitespace_error);4669 }46704671 if (update_index) {4672 if (write_locked_index(&the_index, &lock_file, COMMIT_LOCK))4673 die(_("Unable to write new index file"));4674 }46754676 return !!errs;4677}