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#include "apply.h" 24 25static const char * const apply_usage[] = { 26 N_("git apply [<options>] [<patch>...]"), 27 NULL 28}; 29 30static void set_default_whitespace_mode(struct apply_state *state) 31{ 32 if (!state->whitespace_option && !apply_default_whitespace) 33 state->ws_error_action = (state->apply ? warn_on_ws_error : nowarn_ws_error); 34} 35 36/* 37 * This represents one "hunk" from a patch, starting with 38 * "@@ -oldpos,oldlines +newpos,newlines @@" marker. The 39 * patch text is pointed at by patch, and its byte length 40 * is stored in size. leading and trailing are the number 41 * of context lines. 42 */ 43struct fragment { 44 unsigned long leading, trailing; 45 unsigned long oldpos, oldlines; 46 unsigned long newpos, newlines; 47 /* 48 * 'patch' is usually borrowed from buf in apply_patch(), 49 * but some codepaths store an allocated buffer. 50 */ 51 const char *patch; 52 unsigned free_patch:1, 53 rejected:1; 54 int size; 55 int linenr; 56 struct fragment *next; 57}; 58 59/* 60 * When dealing with a binary patch, we reuse "leading" field 61 * to store the type of the binary hunk, either deflated "delta" 62 * or deflated "literal". 63 */ 64#define binary_patch_method leading 65#define BINARY_DELTA_DEFLATED 1 66#define BINARY_LITERAL_DEFLATED 2 67 68/* 69 * This represents a "patch" to a file, both metainfo changes 70 * such as creation/deletion, filemode and content changes represented 71 * as a series of fragments. 72 */ 73struct patch { 74 char *new_name, *old_name, *def_name; 75 unsigned int old_mode, new_mode; 76 int is_new, is_delete; /* -1 = unknown, 0 = false, 1 = true */ 77 int rejected; 78 unsigned ws_rule; 79 int lines_added, lines_deleted; 80 int score; 81 unsigned int is_toplevel_relative:1; 82 unsigned int inaccurate_eof:1; 83 unsigned int is_binary:1; 84 unsigned int is_copy:1; 85 unsigned int is_rename:1; 86 unsigned int recount:1; 87 unsigned int conflicted_threeway:1; 88 unsigned int direct_to_threeway:1; 89 struct fragment *fragments; 90 char *result; 91 size_t resultsize; 92 char old_sha1_prefix[41]; 93 char new_sha1_prefix[41]; 94 struct patch *next; 95 96 /* three-way fallback result */ 97 struct object_id threeway_stage[3]; 98}; 99 100static void free_fragment_list(struct fragment *list) 101{ 102 while (list) { 103 struct fragment *next = list->next; 104 if (list->free_patch) 105 free((char *)list->patch); 106 free(list); 107 list = next; 108 } 109} 110 111static void free_patch(struct patch *patch) 112{ 113 free_fragment_list(patch->fragments); 114 free(patch->def_name); 115 free(patch->old_name); 116 free(patch->new_name); 117 free(patch->result); 118 free(patch); 119} 120 121static void free_patch_list(struct patch *list) 122{ 123 while (list) { 124 struct patch *next = list->next; 125 free_patch(list); 126 list = next; 127 } 128} 129 130/* 131 * A line in a file, len-bytes long (includes the terminating LF, 132 * except for an incomplete line at the end if the file ends with 133 * one), and its contents hashes to 'hash'. 134 */ 135struct line { 136 size_t len; 137 unsigned hash : 24; 138 unsigned flag : 8; 139#define LINE_COMMON 1 140#define LINE_PATCHED 2 141}; 142 143/* 144 * This represents a "file", which is an array of "lines". 145 */ 146struct image { 147 char *buf; 148 size_t len; 149 size_t nr; 150 size_t alloc; 151 struct line *line_allocated; 152 struct line *line; 153}; 154 155static uint32_t hash_line(const char *cp, size_t len) 156{ 157 size_t i; 158 uint32_t h; 159 for (i = 0, h = 0; i < len; i++) { 160 if (!isspace(cp[i])) { 161 h = h * 3 + (cp[i] & 0xff); 162 } 163 } 164 return h; 165} 166 167/* 168 * Compare lines s1 of length n1 and s2 of length n2, ignoring 169 * whitespace difference. Returns 1 if they match, 0 otherwise 170 */ 171static int fuzzy_matchlines(const char *s1, size_t n1, 172 const char *s2, size_t n2) 173{ 174 const char *last1 = s1 + n1 - 1; 175 const char *last2 = s2 + n2 - 1; 176 int result = 0; 177 178 /* ignore line endings */ 179 while ((*last1 == '\r') || (*last1 == '\n')) 180 last1--; 181 while ((*last2 == '\r') || (*last2 == '\n')) 182 last2--; 183 184 /* skip leading whitespaces, if both begin with whitespace */ 185 if (s1 <= last1 && s2 <= last2 && isspace(*s1) && isspace(*s2)) { 186 while (isspace(*s1) && (s1 <= last1)) 187 s1++; 188 while (isspace(*s2) && (s2 <= last2)) 189 s2++; 190 } 191 /* early return if both lines are empty */ 192 if ((s1 > last1) && (s2 > last2)) 193 return 1; 194 while (!result) { 195 result = *s1++ - *s2++; 196 /* 197 * Skip whitespace inside. We check for whitespace on 198 * both buffers because we don't want "a b" to match 199 * "ab" 200 */ 201 if (isspace(*s1) && isspace(*s2)) { 202 while (isspace(*s1) && s1 <= last1) 203 s1++; 204 while (isspace(*s2) && s2 <= last2) 205 s2++; 206 } 207 /* 208 * If we reached the end on one side only, 209 * lines don't match 210 */ 211 if ( 212 ((s2 > last2) && (s1 <= last1)) || 213 ((s1 > last1) && (s2 <= last2))) 214 return 0; 215 if ((s1 > last1) && (s2 > last2)) 216 break; 217 } 218 219 return !result; 220} 221 222static void add_line_info(struct image *img, const char *bol, size_t len, unsigned flag) 223{ 224 ALLOC_GROW(img->line_allocated, img->nr + 1, img->alloc); 225 img->line_allocated[img->nr].len = len; 226 img->line_allocated[img->nr].hash = hash_line(bol, len); 227 img->line_allocated[img->nr].flag = flag; 228 img->nr++; 229} 230 231/* 232 * "buf" has the file contents to be patched (read from various sources). 233 * attach it to "image" and add line-based index to it. 234 * "image" now owns the "buf". 235 */ 236static void prepare_image(struct image *image, char *buf, size_t len, 237 int prepare_linetable) 238{ 239 const char *cp, *ep; 240 241 memset(image, 0, sizeof(*image)); 242 image->buf = buf; 243 image->len = len; 244 245 if (!prepare_linetable) 246 return; 247 248 ep = image->buf + image->len; 249 cp = image->buf; 250 while (cp < ep) { 251 const char *next; 252 for (next = cp; next < ep && *next != '\n'; next++) 253 ; 254 if (next < ep) 255 next++; 256 add_line_info(image, cp, next - cp, 0); 257 cp = next; 258 } 259 image->line = image->line_allocated; 260} 261 262static void clear_image(struct image *image) 263{ 264 free(image->buf); 265 free(image->line_allocated); 266 memset(image, 0, sizeof(*image)); 267} 268 269/* fmt must contain _one_ %s and no other substitution */ 270static void say_patch_name(FILE *output, const char *fmt, struct patch *patch) 271{ 272 struct strbuf sb = STRBUF_INIT; 273 274 if (patch->old_name && patch->new_name && 275 strcmp(patch->old_name, patch->new_name)) { 276 quote_c_style(patch->old_name, &sb, NULL, 0); 277 strbuf_addstr(&sb, " => "); 278 quote_c_style(patch->new_name, &sb, NULL, 0); 279 } else { 280 const char *n = patch->new_name; 281 if (!n) 282 n = patch->old_name; 283 quote_c_style(n, &sb, NULL, 0); 284 } 285 fprintf(output, fmt, sb.buf); 286 fputc('\n', output); 287 strbuf_release(&sb); 288} 289 290#define SLOP (16) 291 292static int read_patch_file(struct strbuf *sb, int fd) 293{ 294 if (strbuf_read(sb, fd, 0) < 0) 295 return error_errno("git apply: failed to read"); 296 297 /* 298 * Make sure that we have some slop in the buffer 299 * so that we can do speculative "memcmp" etc, and 300 * see to it that it is NUL-filled. 301 */ 302 strbuf_grow(sb, SLOP); 303 memset(sb->buf + sb->len, 0, SLOP); 304 return 0; 305} 306 307static unsigned long linelen(const char *buffer, unsigned long size) 308{ 309 unsigned long len = 0; 310 while (size--) { 311 len++; 312 if (*buffer++ == '\n') 313 break; 314 } 315 return len; 316} 317 318static int is_dev_null(const char *str) 319{ 320 return skip_prefix(str, "/dev/null", &str) && isspace(*str); 321} 322 323#define TERM_SPACE 1 324#define TERM_TAB 2 325 326static int name_terminate(int c, int terminate) 327{ 328 if (c == ' ' && !(terminate & TERM_SPACE)) 329 return 0; 330 if (c == '\t' && !(terminate & TERM_TAB)) 331 return 0; 332 333 return 1; 334} 335 336/* remove double slashes to make --index work with such filenames */ 337static char *squash_slash(char *name) 338{ 339 int i = 0, j = 0; 340 341 if (!name) 342 return NULL; 343 344 while (name[i]) { 345 if ((name[j++] = name[i++]) == '/') 346 while (name[i] == '/') 347 i++; 348 } 349 name[j] = '\0'; 350 return name; 351} 352 353static char *find_name_gnu(struct apply_state *state, 354 const char *line, 355 const char *def, 356 int p_value) 357{ 358 struct strbuf name = STRBUF_INIT; 359 char *cp; 360 361 /* 362 * Proposed "new-style" GNU patch/diff format; see 363 * http://marc.info/?l=git&m=112927316408690&w=2 364 */ 365 if (unquote_c_style(&name, line, NULL)) { 366 strbuf_release(&name); 367 return NULL; 368 } 369 370 for (cp = name.buf; p_value; p_value--) { 371 cp = strchr(cp, '/'); 372 if (!cp) { 373 strbuf_release(&name); 374 return NULL; 375 } 376 cp++; 377 } 378 379 strbuf_remove(&name, 0, cp - name.buf); 380 if (state->root.len) 381 strbuf_insert(&name, 0, state->root.buf, state->root.len); 382 return squash_slash(strbuf_detach(&name, NULL)); 383} 384 385static size_t sane_tz_len(const char *line, size_t len) 386{ 387 const char *tz, *p; 388 389 if (len < strlen(" +0500") || line[len-strlen(" +0500")] != ' ') 390 return 0; 391 tz = line + len - strlen(" +0500"); 392 393 if (tz[1] != '+' && tz[1] != '-') 394 return 0; 395 396 for (p = tz + 2; p != line + len; p++) 397 if (!isdigit(*p)) 398 return 0; 399 400 return line + len - tz; 401} 402 403static size_t tz_with_colon_len(const char *line, size_t len) 404{ 405 const char *tz, *p; 406 407 if (len < strlen(" +08:00") || line[len - strlen(":00")] != ':') 408 return 0; 409 tz = line + len - strlen(" +08:00"); 410 411 if (tz[0] != ' ' || (tz[1] != '+' && tz[1] != '-')) 412 return 0; 413 p = tz + 2; 414 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 415 !isdigit(*p++) || !isdigit(*p++)) 416 return 0; 417 418 return line + len - tz; 419} 420 421static size_t date_len(const char *line, size_t len) 422{ 423 const char *date, *p; 424 425 if (len < strlen("72-02-05") || line[len-strlen("-05")] != '-') 426 return 0; 427 p = date = line + len - strlen("72-02-05"); 428 429 if (!isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 430 !isdigit(*p++) || !isdigit(*p++) || *p++ != '-' || 431 !isdigit(*p++) || !isdigit(*p++)) /* Not a date. */ 432 return 0; 433 434 if (date - line >= strlen("19") && 435 isdigit(date[-1]) && isdigit(date[-2])) /* 4-digit year */ 436 date -= strlen("19"); 437 438 return line + len - date; 439} 440 441static size_t short_time_len(const char *line, size_t len) 442{ 443 const char *time, *p; 444 445 if (len < strlen(" 07:01:32") || line[len-strlen(":32")] != ':') 446 return 0; 447 p = time = line + len - strlen(" 07:01:32"); 448 449 /* Permit 1-digit hours? */ 450 if (*p++ != ' ' || 451 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 452 !isdigit(*p++) || !isdigit(*p++) || *p++ != ':' || 453 !isdigit(*p++) || !isdigit(*p++)) /* Not a time. */ 454 return 0; 455 456 return line + len - time; 457} 458 459static size_t fractional_time_len(const char *line, size_t len) 460{ 461 const char *p; 462 size_t n; 463 464 /* Expected format: 19:41:17.620000023 */ 465 if (!len || !isdigit(line[len - 1])) 466 return 0; 467 p = line + len - 1; 468 469 /* Fractional seconds. */ 470 while (p > line && isdigit(*p)) 471 p--; 472 if (*p != '.') 473 return 0; 474 475 /* Hours, minutes, and whole seconds. */ 476 n = short_time_len(line, p - line); 477 if (!n) 478 return 0; 479 480 return line + len - p + n; 481} 482 483static size_t trailing_spaces_len(const char *line, size_t len) 484{ 485 const char *p; 486 487 /* Expected format: ' ' x (1 or more) */ 488 if (!len || line[len - 1] != ' ') 489 return 0; 490 491 p = line + len; 492 while (p != line) { 493 p--; 494 if (*p != ' ') 495 return line + len - (p + 1); 496 } 497 498 /* All spaces! */ 499 return len; 500} 501 502static size_t diff_timestamp_len(const char *line, size_t len) 503{ 504 const char *end = line + len; 505 size_t n; 506 507 /* 508 * Posix: 2010-07-05 19:41:17 509 * GNU: 2010-07-05 19:41:17.620000023 -0500 510 */ 511 512 if (!isdigit(end[-1])) 513 return 0; 514 515 n = sane_tz_len(line, end - line); 516 if (!n) 517 n = tz_with_colon_len(line, end - line); 518 end -= n; 519 520 n = short_time_len(line, end - line); 521 if (!n) 522 n = fractional_time_len(line, end - line); 523 end -= n; 524 525 n = date_len(line, end - line); 526 if (!n) /* No date. Too bad. */ 527 return 0; 528 end -= n; 529 530 if (end == line) /* No space before date. */ 531 return 0; 532 if (end[-1] == '\t') { /* Success! */ 533 end--; 534 return line + len - end; 535 } 536 if (end[-1] != ' ') /* No space before date. */ 537 return 0; 538 539 /* Whitespace damage. */ 540 end -= trailing_spaces_len(line, end - line); 541 return line + len - end; 542} 543 544static char *find_name_common(struct apply_state *state, 545 const char *line, 546 const char *def, 547 int p_value, 548 const char *end, 549 int terminate) 550{ 551 int len; 552 const char *start = NULL; 553 554 if (p_value == 0) 555 start = line; 556 while (line != end) { 557 char c = *line; 558 559 if (!end && isspace(c)) { 560 if (c == '\n') 561 break; 562 if (name_terminate(c, terminate)) 563 break; 564 } 565 line++; 566 if (c == '/' && !--p_value) 567 start = line; 568 } 569 if (!start) 570 return squash_slash(xstrdup_or_null(def)); 571 len = line - start; 572 if (!len) 573 return squash_slash(xstrdup_or_null(def)); 574 575 /* 576 * Generally we prefer the shorter name, especially 577 * if the other one is just a variation of that with 578 * something else tacked on to the end (ie "file.orig" 579 * or "file~"). 580 */ 581 if (def) { 582 int deflen = strlen(def); 583 if (deflen < len && !strncmp(start, def, deflen)) 584 return squash_slash(xstrdup(def)); 585 } 586 587 if (state->root.len) { 588 char *ret = xstrfmt("%s%.*s", state->root.buf, len, start); 589 return squash_slash(ret); 590 } 591 592 return squash_slash(xmemdupz(start, len)); 593} 594 595static char *find_name(struct apply_state *state, 596 const char *line, 597 char *def, 598 int p_value, 599 int terminate) 600{ 601 if (*line == '"') { 602 char *name = find_name_gnu(state, line, def, p_value); 603 if (name) 604 return name; 605 } 606 607 return find_name_common(state, line, def, p_value, NULL, terminate); 608} 609 610static char *find_name_traditional(struct apply_state *state, 611 const char *line, 612 char *def, 613 int p_value) 614{ 615 size_t len; 616 size_t date_len; 617 618 if (*line == '"') { 619 char *name = find_name_gnu(state, line, def, p_value); 620 if (name) 621 return name; 622 } 623 624 len = strchrnul(line, '\n') - line; 625 date_len = diff_timestamp_len(line, len); 626 if (!date_len) 627 return find_name_common(state, line, def, p_value, NULL, TERM_TAB); 628 len -= date_len; 629 630 return find_name_common(state, line, def, p_value, line + len, 0); 631} 632 633static int count_slashes(const char *cp) 634{ 635 int cnt = 0; 636 char ch; 637 638 while ((ch = *cp++)) 639 if (ch == '/') 640 cnt++; 641 return cnt; 642} 643 644/* 645 * Given the string after "--- " or "+++ ", guess the appropriate 646 * p_value for the given patch. 647 */ 648static int guess_p_value(struct apply_state *state, const char *nameline) 649{ 650 char *name, *cp; 651 int val = -1; 652 653 if (is_dev_null(nameline)) 654 return -1; 655 name = find_name_traditional(state, nameline, NULL, 0); 656 if (!name) 657 return -1; 658 cp = strchr(name, '/'); 659 if (!cp) 660 val = 0; 661 else if (state->prefix) { 662 /* 663 * Does it begin with "a/$our-prefix" and such? Then this is 664 * very likely to apply to our directory. 665 */ 666 if (!strncmp(name, state->prefix, state->prefix_length)) 667 val = count_slashes(state->prefix); 668 else { 669 cp++; 670 if (!strncmp(cp, state->prefix, state->prefix_length)) 671 val = count_slashes(state->prefix) + 1; 672 } 673 } 674 free(name); 675 return val; 676} 677 678/* 679 * Does the ---/+++ line have the POSIX timestamp after the last HT? 680 * GNU diff puts epoch there to signal a creation/deletion event. Is 681 * this such a timestamp? 682 */ 683static int has_epoch_timestamp(const char *nameline) 684{ 685 /* 686 * We are only interested in epoch timestamp; any non-zero 687 * fraction cannot be one, hence "(\.0+)?" in the regexp below. 688 * For the same reason, the date must be either 1969-12-31 or 689 * 1970-01-01, and the seconds part must be "00". 690 */ 691 const char stamp_regexp[] = 692 "^(1969-12-31|1970-01-01)" 693 " " 694 "[0-2][0-9]:[0-5][0-9]:00(\\.0+)?" 695 " " 696 "([-+][0-2][0-9]:?[0-5][0-9])\n"; 697 const char *timestamp = NULL, *cp, *colon; 698 static regex_t *stamp; 699 regmatch_t m[10]; 700 int zoneoffset; 701 int hourminute; 702 int status; 703 704 for (cp = nameline; *cp != '\n'; cp++) { 705 if (*cp == '\t') 706 timestamp = cp + 1; 707 } 708 if (!timestamp) 709 return 0; 710 if (!stamp) { 711 stamp = xmalloc(sizeof(*stamp)); 712 if (regcomp(stamp, stamp_regexp, REG_EXTENDED)) { 713 warning(_("Cannot prepare timestamp regexp %s"), 714 stamp_regexp); 715 return 0; 716 } 717 } 718 719 status = regexec(stamp, timestamp, ARRAY_SIZE(m), m, 0); 720 if (status) { 721 if (status != REG_NOMATCH) 722 warning(_("regexec returned %d for input: %s"), 723 status, timestamp); 724 return 0; 725 } 726 727 zoneoffset = strtol(timestamp + m[3].rm_so + 1, (char **) &colon, 10); 728 if (*colon == ':') 729 zoneoffset = zoneoffset * 60 + strtol(colon + 1, NULL, 10); 730 else 731 zoneoffset = (zoneoffset / 100) * 60 + (zoneoffset % 100); 732 if (timestamp[m[3].rm_so] == '-') 733 zoneoffset = -zoneoffset; 734 735 /* 736 * YYYY-MM-DD hh:mm:ss must be from either 1969-12-31 737 * (west of GMT) or 1970-01-01 (east of GMT) 738 */ 739 if ((zoneoffset < 0 && memcmp(timestamp, "1969-12-31", 10)) || 740 (0 <= zoneoffset && memcmp(timestamp, "1970-01-01", 10))) 741 return 0; 742 743 hourminute = (strtol(timestamp + 11, NULL, 10) * 60 + 744 strtol(timestamp + 14, NULL, 10) - 745 zoneoffset); 746 747 return ((zoneoffset < 0 && hourminute == 1440) || 748 (0 <= zoneoffset && !hourminute)); 749} 750 751/* 752 * Get the name etc info from the ---/+++ lines of a traditional patch header 753 * 754 * FIXME! The end-of-filename heuristics are kind of screwy. For existing 755 * files, we can happily check the index for a match, but for creating a 756 * new file we should try to match whatever "patch" does. I have no idea. 757 */ 758static int parse_traditional_patch(struct apply_state *state, 759 const char *first, 760 const char *second, 761 struct patch *patch) 762{ 763 char *name; 764 765 first += 4; /* skip "--- " */ 766 second += 4; /* skip "+++ " */ 767 if (!state->p_value_known) { 768 int p, q; 769 p = guess_p_value(state, first); 770 q = guess_p_value(state, second); 771 if (p < 0) p = q; 772 if (0 <= p && p == q) { 773 state->p_value = p; 774 state->p_value_known = 1; 775 } 776 } 777 if (is_dev_null(first)) { 778 patch->is_new = 1; 779 patch->is_delete = 0; 780 name = find_name_traditional(state, second, NULL, state->p_value); 781 patch->new_name = name; 782 } else if (is_dev_null(second)) { 783 patch->is_new = 0; 784 patch->is_delete = 1; 785 name = find_name_traditional(state, first, NULL, state->p_value); 786 patch->old_name = name; 787 } else { 788 char *first_name; 789 first_name = find_name_traditional(state, first, NULL, state->p_value); 790 name = find_name_traditional(state, second, first_name, state->p_value); 791 free(first_name); 792 if (has_epoch_timestamp(first)) { 793 patch->is_new = 1; 794 patch->is_delete = 0; 795 patch->new_name = name; 796 } else if (has_epoch_timestamp(second)) { 797 patch->is_new = 0; 798 patch->is_delete = 1; 799 patch->old_name = name; 800 } else { 801 patch->old_name = name; 802 patch->new_name = xstrdup_or_null(name); 803 } 804 } 805 if (!name) 806 return error(_("unable to find filename in patch at line %d"), state->linenr); 807 808 return 0; 809} 810 811static int gitdiff_hdrend(struct apply_state *state, 812 const char *line, 813 struct patch *patch) 814{ 815 return 1; 816} 817 818/* 819 * We're anal about diff header consistency, to make 820 * sure that we don't end up having strange ambiguous 821 * patches floating around. 822 * 823 * As a result, gitdiff_{old|new}name() will check 824 * their names against any previous information, just 825 * to make sure.. 826 */ 827#define DIFF_OLD_NAME 0 828#define DIFF_NEW_NAME 1 829 830static void gitdiff_verify_name(struct apply_state *state, 831 const char *line, 832 int isnull, 833 char **name, 834 int side) 835{ 836 if (!*name && !isnull) { 837 *name = find_name(state, line, NULL, state->p_value, TERM_TAB); 838 return; 839 } 840 841 if (*name) { 842 int len = strlen(*name); 843 char *another; 844 if (isnull) 845 die(_("git apply: bad git-diff - expected /dev/null, got %s on line %d"), 846 *name, state->linenr); 847 another = find_name(state, line, NULL, state->p_value, TERM_TAB); 848 if (!another || memcmp(another, *name, len + 1)) 849 die((side == DIFF_NEW_NAME) ? 850 _("git apply: bad git-diff - inconsistent new filename on line %d") : 851 _("git apply: bad git-diff - inconsistent old filename on line %d"), state->linenr); 852 free(another); 853 } else { 854 /* expect "/dev/null" */ 855 if (memcmp("/dev/null", line, 9) || line[9] != '\n') 856 die(_("git apply: bad git-diff - expected /dev/null on line %d"), state->linenr); 857 } 858} 859 860static int gitdiff_oldname(struct apply_state *state, 861 const char *line, 862 struct patch *patch) 863{ 864 gitdiff_verify_name(state, line, 865 patch->is_new, &patch->old_name, 866 DIFF_OLD_NAME); 867 return 0; 868} 869 870static int gitdiff_newname(struct apply_state *state, 871 const char *line, 872 struct patch *patch) 873{ 874 gitdiff_verify_name(state, line, 875 patch->is_delete, &patch->new_name, 876 DIFF_NEW_NAME); 877 return 0; 878} 879 880static int gitdiff_oldmode(struct apply_state *state, 881 const char *line, 882 struct patch *patch) 883{ 884 patch->old_mode = strtoul(line, NULL, 8); 885 return 0; 886} 887 888static int gitdiff_newmode(struct apply_state *state, 889 const char *line, 890 struct patch *patch) 891{ 892 patch->new_mode = strtoul(line, NULL, 8); 893 return 0; 894} 895 896static int gitdiff_delete(struct apply_state *state, 897 const char *line, 898 struct patch *patch) 899{ 900 patch->is_delete = 1; 901 free(patch->old_name); 902 patch->old_name = xstrdup_or_null(patch->def_name); 903 return gitdiff_oldmode(state, line, patch); 904} 905 906static int gitdiff_newfile(struct apply_state *state, 907 const char *line, 908 struct patch *patch) 909{ 910 patch->is_new = 1; 911 free(patch->new_name); 912 patch->new_name = xstrdup_or_null(patch->def_name); 913 return gitdiff_newmode(state, line, patch); 914} 915 916static int gitdiff_copysrc(struct apply_state *state, 917 const char *line, 918 struct patch *patch) 919{ 920 patch->is_copy = 1; 921 free(patch->old_name); 922 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0); 923 return 0; 924} 925 926static int gitdiff_copydst(struct apply_state *state, 927 const char *line, 928 struct patch *patch) 929{ 930 patch->is_copy = 1; 931 free(patch->new_name); 932 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0); 933 return 0; 934} 935 936static int gitdiff_renamesrc(struct apply_state *state, 937 const char *line, 938 struct patch *patch) 939{ 940 patch->is_rename = 1; 941 free(patch->old_name); 942 patch->old_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0); 943 return 0; 944} 945 946static int gitdiff_renamedst(struct apply_state *state, 947 const char *line, 948 struct patch *patch) 949{ 950 patch->is_rename = 1; 951 free(patch->new_name); 952 patch->new_name = find_name(state, line, NULL, state->p_value ? state->p_value - 1 : 0, 0); 953 return 0; 954} 955 956static int gitdiff_similarity(struct apply_state *state, 957 const char *line, 958 struct patch *patch) 959{ 960 unsigned long val = strtoul(line, NULL, 10); 961 if (val <= 100) 962 patch->score = val; 963 return 0; 964} 965 966static int gitdiff_dissimilarity(struct apply_state *state, 967 const char *line, 968 struct patch *patch) 969{ 970 unsigned long val = strtoul(line, NULL, 10); 971 if (val <= 100) 972 patch->score = val; 973 return 0; 974} 975 976static int gitdiff_index(struct apply_state *state, 977 const char *line, 978 struct patch *patch) 979{ 980 /* 981 * index line is N hexadecimal, "..", N hexadecimal, 982 * and optional space with octal mode. 983 */ 984 const char *ptr, *eol; 985 int len; 986 987 ptr = strchr(line, '.'); 988 if (!ptr || ptr[1] != '.' || 40 < ptr - line) 989 return 0; 990 len = ptr - line; 991 memcpy(patch->old_sha1_prefix, line, len); 992 patch->old_sha1_prefix[len] = 0; 993 994 line = ptr + 2; 995 ptr = strchr(line, ' '); 996 eol = strchrnul(line, '\n'); 997 998 if (!ptr || eol < ptr) 999 ptr = eol;1000 len = ptr - line;10011002 if (40 < len)1003 return 0;1004 memcpy(patch->new_sha1_prefix, line, len);1005 patch->new_sha1_prefix[len] = 0;1006 if (*ptr == ' ')1007 patch->old_mode = strtoul(ptr+1, NULL, 8);1008 return 0;1009}10101011/*1012 * This is normal for a diff that doesn't change anything: we'll fall through1013 * into the next diff. Tell the parser to break out.1014 */1015static int gitdiff_unrecognized(struct apply_state *state,1016 const char *line,1017 struct patch *patch)1018{1019 return 1;1020}10211022/*1023 * Skip p_value leading components from "line"; as we do not accept1024 * absolute paths, return NULL in that case.1025 */1026static const char *skip_tree_prefix(struct apply_state *state,1027 const char *line,1028 int llen)1029{1030 int nslash;1031 int i;10321033 if (!state->p_value)1034 return (llen && line[0] == '/') ? NULL : line;10351036 nslash = state->p_value;1037 for (i = 0; i < llen; i++) {1038 int ch = line[i];1039 if (ch == '/' && --nslash <= 0)1040 return (i == 0) ? NULL : &line[i + 1];1041 }1042 return NULL;1043}10441045/*1046 * This is to extract the same name that appears on "diff --git"1047 * line. We do not find and return anything if it is a rename1048 * patch, and it is OK because we will find the name elsewhere.1049 * We need to reliably find name only when it is mode-change only,1050 * creation or deletion of an empty file. In any of these cases,1051 * both sides are the same name under a/ and b/ respectively.1052 */1053static char *git_header_name(struct apply_state *state,1054 const char *line,1055 int llen)1056{1057 const char *name;1058 const char *second = NULL;1059 size_t len, line_len;10601061 line += strlen("diff --git ");1062 llen -= strlen("diff --git ");10631064 if (*line == '"') {1065 const char *cp;1066 struct strbuf first = STRBUF_INIT;1067 struct strbuf sp = STRBUF_INIT;10681069 if (unquote_c_style(&first, line, &second))1070 goto free_and_fail1;10711072 /* strip the a/b prefix including trailing slash */1073 cp = skip_tree_prefix(state, first.buf, first.len);1074 if (!cp)1075 goto free_and_fail1;1076 strbuf_remove(&first, 0, cp - first.buf);10771078 /*1079 * second points at one past closing dq of name.1080 * find the second name.1081 */1082 while ((second < line + llen) && isspace(*second))1083 second++;10841085 if (line + llen <= second)1086 goto free_and_fail1;1087 if (*second == '"') {1088 if (unquote_c_style(&sp, second, NULL))1089 goto free_and_fail1;1090 cp = skip_tree_prefix(state, sp.buf, sp.len);1091 if (!cp)1092 goto free_and_fail1;1093 /* They must match, otherwise ignore */1094 if (strcmp(cp, first.buf))1095 goto free_and_fail1;1096 strbuf_release(&sp);1097 return strbuf_detach(&first, NULL);1098 }10991100 /* unquoted second */1101 cp = skip_tree_prefix(state, second, line + llen - second);1102 if (!cp)1103 goto free_and_fail1;1104 if (line + llen - cp != first.len ||1105 memcmp(first.buf, cp, first.len))1106 goto free_and_fail1;1107 return strbuf_detach(&first, NULL);11081109 free_and_fail1:1110 strbuf_release(&first);1111 strbuf_release(&sp);1112 return NULL;1113 }11141115 /* unquoted first name */1116 name = skip_tree_prefix(state, line, llen);1117 if (!name)1118 return NULL;11191120 /*1121 * since the first name is unquoted, a dq if exists must be1122 * the beginning of the second name.1123 */1124 for (second = name; second < line + llen; second++) {1125 if (*second == '"') {1126 struct strbuf sp = STRBUF_INIT;1127 const char *np;11281129 if (unquote_c_style(&sp, second, NULL))1130 goto free_and_fail2;11311132 np = skip_tree_prefix(state, sp.buf, sp.len);1133 if (!np)1134 goto free_and_fail2;11351136 len = sp.buf + sp.len - np;1137 if (len < second - name &&1138 !strncmp(np, name, len) &&1139 isspace(name[len])) {1140 /* Good */1141 strbuf_remove(&sp, 0, np - sp.buf);1142 return strbuf_detach(&sp, NULL);1143 }11441145 free_and_fail2:1146 strbuf_release(&sp);1147 return NULL;1148 }1149 }11501151 /*1152 * Accept a name only if it shows up twice, exactly the same1153 * form.1154 */1155 second = strchr(name, '\n');1156 if (!second)1157 return NULL;1158 line_len = second - name;1159 for (len = 0 ; ; len++) {1160 switch (name[len]) {1161 default:1162 continue;1163 case '\n':1164 return NULL;1165 case '\t': case ' ':1166 /*1167 * Is this the separator between the preimage1168 * and the postimage pathname? Again, we are1169 * only interested in the case where there is1170 * no rename, as this is only to set def_name1171 * and a rename patch has the names elsewhere1172 * in an unambiguous form.1173 */1174 if (!name[len + 1])1175 return NULL; /* no postimage name */1176 second = skip_tree_prefix(state, name + len + 1,1177 line_len - (len + 1));1178 if (!second)1179 return NULL;1180 /*1181 * Does len bytes starting at "name" and "second"1182 * (that are separated by one HT or SP we just1183 * found) exactly match?1184 */1185 if (second[len] == '\n' && !strncmp(name, second, len))1186 return xmemdupz(name, len);1187 }1188 }1189}11901191/* Verify that we recognize the lines following a git header */1192static int parse_git_header(struct apply_state *state,1193 const char *line,1194 int len,1195 unsigned int size,1196 struct patch *patch)1197{1198 unsigned long offset;11991200 /* A git diff has explicit new/delete information, so we don't guess */1201 patch->is_new = 0;1202 patch->is_delete = 0;12031204 /*1205 * Some things may not have the old name in the1206 * rest of the headers anywhere (pure mode changes,1207 * or removing or adding empty files), so we get1208 * the default name from the header.1209 */1210 patch->def_name = git_header_name(state, line, len);1211 if (patch->def_name && state->root.len) {1212 char *s = xstrfmt("%s%s", state->root.buf, patch->def_name);1213 free(patch->def_name);1214 patch->def_name = s;1215 }12161217 line += len;1218 size -= len;1219 state->linenr++;1220 for (offset = len ; size > 0 ; offset += len, size -= len, line += len, state->linenr++) {1221 static const struct opentry {1222 const char *str;1223 int (*fn)(struct apply_state *, const char *, struct patch *);1224 } optable[] = {1225 { "@@ -", gitdiff_hdrend },1226 { "--- ", gitdiff_oldname },1227 { "+++ ", gitdiff_newname },1228 { "old mode ", gitdiff_oldmode },1229 { "new mode ", gitdiff_newmode },1230 { "deleted file mode ", gitdiff_delete },1231 { "new file mode ", gitdiff_newfile },1232 { "copy from ", gitdiff_copysrc },1233 { "copy to ", gitdiff_copydst },1234 { "rename old ", gitdiff_renamesrc },1235 { "rename new ", gitdiff_renamedst },1236 { "rename from ", gitdiff_renamesrc },1237 { "rename to ", gitdiff_renamedst },1238 { "similarity index ", gitdiff_similarity },1239 { "dissimilarity index ", gitdiff_dissimilarity },1240 { "index ", gitdiff_index },1241 { "", gitdiff_unrecognized },1242 };1243 int i;12441245 len = linelen(line, size);1246 if (!len || line[len-1] != '\n')1247 break;1248 for (i = 0; i < ARRAY_SIZE(optable); i++) {1249 const struct opentry *p = optable + i;1250 int oplen = strlen(p->str);1251 int res;1252 if (len < oplen || memcmp(p->str, line, oplen))1253 continue;1254 res = p->fn(state, line + oplen, patch);1255 if (res < 0)1256 return -1;1257 if (res > 0)1258 return offset;1259 break;1260 }1261 }12621263 return offset;1264}12651266static int parse_num(const char *line, unsigned long *p)1267{1268 char *ptr;12691270 if (!isdigit(*line))1271 return 0;1272 *p = strtoul(line, &ptr, 10);1273 return ptr - line;1274}12751276static int parse_range(const char *line, int len, int offset, const char *expect,1277 unsigned long *p1, unsigned long *p2)1278{1279 int digits, ex;12801281 if (offset < 0 || offset >= len)1282 return -1;1283 line += offset;1284 len -= offset;12851286 digits = parse_num(line, p1);1287 if (!digits)1288 return -1;12891290 offset += digits;1291 line += digits;1292 len -= digits;12931294 *p2 = 1;1295 if (*line == ',') {1296 digits = parse_num(line+1, p2);1297 if (!digits)1298 return -1;12991300 offset += digits+1;1301 line += digits+1;1302 len -= digits+1;1303 }13041305 ex = strlen(expect);1306 if (ex > len)1307 return -1;1308 if (memcmp(line, expect, ex))1309 return -1;13101311 return offset + ex;1312}13131314static void recount_diff(const char *line, int size, struct fragment *fragment)1315{1316 int oldlines = 0, newlines = 0, ret = 0;13171318 if (size < 1) {1319 warning("recount: ignore empty hunk");1320 return;1321 }13221323 for (;;) {1324 int len = linelen(line, size);1325 size -= len;1326 line += len;13271328 if (size < 1)1329 break;13301331 switch (*line) {1332 case ' ': case '\n':1333 newlines++;1334 /* fall through */1335 case '-':1336 oldlines++;1337 continue;1338 case '+':1339 newlines++;1340 continue;1341 case '\\':1342 continue;1343 case '@':1344 ret = size < 3 || !starts_with(line, "@@ ");1345 break;1346 case 'd':1347 ret = size < 5 || !starts_with(line, "diff ");1348 break;1349 default:1350 ret = -1;1351 break;1352 }1353 if (ret) {1354 warning(_("recount: unexpected line: %.*s"),1355 (int)linelen(line, size), line);1356 return;1357 }1358 break;1359 }1360 fragment->oldlines = oldlines;1361 fragment->newlines = newlines;1362}13631364/*1365 * Parse a unified diff fragment header of the1366 * form "@@ -a,b +c,d @@"1367 */1368static int parse_fragment_header(const char *line, int len, struct fragment *fragment)1369{1370 int offset;13711372 if (!len || line[len-1] != '\n')1373 return -1;13741375 /* Figure out the number of lines in a fragment */1376 offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);1377 offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);13781379 return offset;1380}13811382/*1383 * Find file diff header1384 *1385 * Returns:1386 * -1 if no header was found1387 * -128 in case of error1388 * the size of the header in bytes (called "offset") otherwise1389 */1390static int find_header(struct apply_state *state,1391 const char *line,1392 unsigned long size,1393 int *hdrsize,1394 struct patch *patch)1395{1396 unsigned long offset, len;13971398 patch->is_toplevel_relative = 0;1399 patch->is_rename = patch->is_copy = 0;1400 patch->is_new = patch->is_delete = -1;1401 patch->old_mode = patch->new_mode = 0;1402 patch->old_name = patch->new_name = NULL;1403 for (offset = 0; size > 0; offset += len, size -= len, line += len, state->linenr++) {1404 unsigned long nextlen;14051406 len = linelen(line, size);1407 if (!len)1408 break;14091410 /* Testing this early allows us to take a few shortcuts.. */1411 if (len < 6)1412 continue;14131414 /*1415 * Make sure we don't find any unconnected patch fragments.1416 * That's a sign that we didn't find a header, and that a1417 * patch has become corrupted/broken up.1418 */1419 if (!memcmp("@@ -", line, 4)) {1420 struct fragment dummy;1421 if (parse_fragment_header(line, len, &dummy) < 0)1422 continue;1423 error(_("patch fragment without header at line %d: %.*s"),1424 state->linenr, (int)len-1, line);1425 return -128;1426 }14271428 if (size < len + 6)1429 break;14301431 /*1432 * Git patch? It might not have a real patch, just a rename1433 * or mode change, so we handle that specially1434 */1435 if (!memcmp("diff --git ", line, 11)) {1436 int git_hdr_len = parse_git_header(state, line, len, size, patch);1437 if (git_hdr_len < 0)1438 return -128;1439 if (git_hdr_len <= len)1440 continue;1441 if (!patch->old_name && !patch->new_name) {1442 if (!patch->def_name) {1443 error(Q_("git diff header lacks filename information when removing "1444 "%d leading pathname component (line %d)",1445 "git diff header lacks filename information when removing "1446 "%d leading pathname components (line %d)",1447 state->p_value),1448 state->p_value, state->linenr);1449 return -128;1450 }1451 patch->old_name = xstrdup(patch->def_name);1452 patch->new_name = xstrdup(patch->def_name);1453 }1454 if (!patch->is_delete && !patch->new_name) {1455 error("git diff header lacks filename information "1456 "(line %d)", state->linenr);1457 return -128;1458 }1459 patch->is_toplevel_relative = 1;1460 *hdrsize = git_hdr_len;1461 return offset;1462 }14631464 /* --- followed by +++ ? */1465 if (memcmp("--- ", line, 4) || memcmp("+++ ", line + len, 4))1466 continue;14671468 /*1469 * We only accept unified patches, so we want it to1470 * at least have "@@ -a,b +c,d @@\n", which is 14 chars1471 * minimum ("@@ -0,0 +1 @@\n" is the shortest).1472 */1473 nextlen = linelen(line + len, size - len);1474 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))1475 continue;14761477 /* Ok, we'll consider it a patch */1478 if (parse_traditional_patch(state, line, line+len, patch))1479 return -128;1480 *hdrsize = len + nextlen;1481 state->linenr += 2;1482 return offset;1483 }1484 return -1;1485}14861487static void record_ws_error(struct apply_state *state,1488 unsigned result,1489 const char *line,1490 int len,1491 int linenr)1492{1493 char *err;14941495 if (!result)1496 return;14971498 state->whitespace_error++;1499 if (state->squelch_whitespace_errors &&1500 state->squelch_whitespace_errors < state->whitespace_error)1501 return;15021503 err = whitespace_error_string(result);1504 fprintf(stderr, "%s:%d: %s.\n%.*s\n",1505 state->patch_input_file, linenr, err, len, line);1506 free(err);1507}15081509static void check_whitespace(struct apply_state *state,1510 const char *line,1511 int len,1512 unsigned ws_rule)1513{1514 unsigned result = ws_check(line + 1, len - 1, ws_rule);15151516 record_ws_error(state, result, line + 1, len - 2, state->linenr);1517}15181519/*1520 * Parse a unified diff. Note that this really needs to parse each1521 * fragment separately, since the only way to know the difference1522 * between a "---" that is part of a patch, and a "---" that starts1523 * the next patch is to look at the line counts..1524 */1525static int parse_fragment(struct apply_state *state,1526 const char *line,1527 unsigned long size,1528 struct patch *patch,1529 struct fragment *fragment)1530{1531 int added, deleted;1532 int len = linelen(line, size), offset;1533 unsigned long oldlines, newlines;1534 unsigned long leading, trailing;15351536 offset = parse_fragment_header(line, len, fragment);1537 if (offset < 0)1538 return -1;1539 if (offset > 0 && patch->recount)1540 recount_diff(line + offset, size - offset, fragment);1541 oldlines = fragment->oldlines;1542 newlines = fragment->newlines;1543 leading = 0;1544 trailing = 0;15451546 /* Parse the thing.. */1547 line += len;1548 size -= len;1549 state->linenr++;1550 added = deleted = 0;1551 for (offset = len;1552 0 < size;1553 offset += len, size -= len, line += len, state->linenr++) {1554 if (!oldlines && !newlines)1555 break;1556 len = linelen(line, size);1557 if (!len || line[len-1] != '\n')1558 return -1;1559 switch (*line) {1560 default:1561 return -1;1562 case '\n': /* newer GNU diff, an empty context line */1563 case ' ':1564 oldlines--;1565 newlines--;1566 if (!deleted && !added)1567 leading++;1568 trailing++;1569 if (!state->apply_in_reverse &&1570 state->ws_error_action == correct_ws_error)1571 check_whitespace(state, line, len, patch->ws_rule);1572 break;1573 case '-':1574 if (state->apply_in_reverse &&1575 state->ws_error_action != nowarn_ws_error)1576 check_whitespace(state, line, len, patch->ws_rule);1577 deleted++;1578 oldlines--;1579 trailing = 0;1580 break;1581 case '+':1582 if (!state->apply_in_reverse &&1583 state->ws_error_action != nowarn_ws_error)1584 check_whitespace(state, line, len, patch->ws_rule);1585 added++;1586 newlines--;1587 trailing = 0;1588 break;15891590 /*1591 * We allow "\ No newline at end of file". Depending1592 * on locale settings when the patch was produced we1593 * don't know what this line looks like. The only1594 * thing we do know is that it begins with "\ ".1595 * Checking for 12 is just for sanity check -- any1596 * l10n of "\ No newline..." is at least that long.1597 */1598 case '\\':1599 if (len < 12 || memcmp(line, "\\ ", 2))1600 return -1;1601 break;1602 }1603 }1604 if (oldlines || newlines)1605 return -1;1606 if (!deleted && !added)1607 return -1;16081609 fragment->leading = leading;1610 fragment->trailing = trailing;16111612 /*1613 * If a fragment ends with an incomplete line, we failed to include1614 * it in the above loop because we hit oldlines == newlines == 01615 * before seeing it.1616 */1617 if (12 < size && !memcmp(line, "\\ ", 2))1618 offset += linelen(line, size);16191620 patch->lines_added += added;1621 patch->lines_deleted += deleted;16221623 if (0 < patch->is_new && oldlines)1624 return error(_("new file depends on old contents"));1625 if (0 < patch->is_delete && newlines)1626 return error(_("deleted file still has contents"));1627 return offset;1628}16291630/*1631 * We have seen "diff --git a/... b/..." header (or a traditional patch1632 * header). Read hunks that belong to this patch into fragments and hang1633 * them to the given patch structure.1634 *1635 * The (fragment->patch, fragment->size) pair points into the memory given1636 * by the caller, not a copy, when we return.1637 *1638 * Returns:1639 * -1 in case of error,1640 * the number of bytes in the patch otherwise.1641 */1642static int parse_single_patch(struct apply_state *state,1643 const char *line,1644 unsigned long size,1645 struct patch *patch)1646{1647 unsigned long offset = 0;1648 unsigned long oldlines = 0, newlines = 0, context = 0;1649 struct fragment **fragp = &patch->fragments;16501651 while (size > 4 && !memcmp(line, "@@ -", 4)) {1652 struct fragment *fragment;1653 int len;16541655 fragment = xcalloc(1, sizeof(*fragment));1656 fragment->linenr = state->linenr;1657 len = parse_fragment(state, line, size, patch, fragment);1658 if (len <= 0) {1659 free(fragment);1660 return error(_("corrupt patch at line %d"), state->linenr);1661 }1662 fragment->patch = line;1663 fragment->size = len;1664 oldlines += fragment->oldlines;1665 newlines += fragment->newlines;1666 context += fragment->leading + fragment->trailing;16671668 *fragp = fragment;1669 fragp = &fragment->next;16701671 offset += len;1672 line += len;1673 size -= len;1674 }16751676 /*1677 * If something was removed (i.e. we have old-lines) it cannot1678 * be creation, and if something was added it cannot be1679 * deletion. However, the reverse is not true; --unified=01680 * patches that only add are not necessarily creation even1681 * though they do not have any old lines, and ones that only1682 * delete are not necessarily deletion.1683 *1684 * Unfortunately, a real creation/deletion patch do _not_ have1685 * any context line by definition, so we cannot safely tell it1686 * apart with --unified=0 insanity. At least if the patch has1687 * more than one hunk it is not creation or deletion.1688 */1689 if (patch->is_new < 0 &&1690 (oldlines || (patch->fragments && patch->fragments->next)))1691 patch->is_new = 0;1692 if (patch->is_delete < 0 &&1693 (newlines || (patch->fragments && patch->fragments->next)))1694 patch->is_delete = 0;16951696 if (0 < patch->is_new && oldlines)1697 return error(_("new file %s depends on old contents"), patch->new_name);1698 if (0 < patch->is_delete && newlines)1699 return error(_("deleted file %s still has contents"), patch->old_name);1700 if (!patch->is_delete && !newlines && context)1701 fprintf_ln(stderr,1702 _("** warning: "1703 "file %s becomes empty but is not deleted"),1704 patch->new_name);17051706 return offset;1707}17081709static inline int metadata_changes(struct patch *patch)1710{1711 return patch->is_rename > 0 ||1712 patch->is_copy > 0 ||1713 patch->is_new > 0 ||1714 patch->is_delete ||1715 (patch->old_mode && patch->new_mode &&1716 patch->old_mode != patch->new_mode);1717}17181719static char *inflate_it(const void *data, unsigned long size,1720 unsigned long inflated_size)1721{1722 git_zstream stream;1723 void *out;1724 int st;17251726 memset(&stream, 0, sizeof(stream));17271728 stream.next_in = (unsigned char *)data;1729 stream.avail_in = size;1730 stream.next_out = out = xmalloc(inflated_size);1731 stream.avail_out = inflated_size;1732 git_inflate_init(&stream);1733 st = git_inflate(&stream, Z_FINISH);1734 git_inflate_end(&stream);1735 if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {1736 free(out);1737 return NULL;1738 }1739 return out;1740}17411742/*1743 * Read a binary hunk and return a new fragment; fragment->patch1744 * points at an allocated memory that the caller must free, so1745 * it is marked as "->free_patch = 1".1746 */1747static struct fragment *parse_binary_hunk(struct apply_state *state,1748 char **buf_p,1749 unsigned long *sz_p,1750 int *status_p,1751 int *used_p)1752{1753 /*1754 * Expect a line that begins with binary patch method ("literal"1755 * or "delta"), followed by the length of data before deflating.1756 * a sequence of 'length-byte' followed by base-85 encoded data1757 * should follow, terminated by a newline.1758 *1759 * Each 5-byte sequence of base-85 encodes up to 4 bytes,1760 * and we would limit the patch line to 66 characters,1761 * so one line can fit up to 13 groups that would decode1762 * to 52 bytes max. The length byte 'A'-'Z' corresponds1763 * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.1764 */1765 int llen, used;1766 unsigned long size = *sz_p;1767 char *buffer = *buf_p;1768 int patch_method;1769 unsigned long origlen;1770 char *data = NULL;1771 int hunk_size = 0;1772 struct fragment *frag;17731774 llen = linelen(buffer, size);1775 used = llen;17761777 *status_p = 0;17781779 if (starts_with(buffer, "delta ")) {1780 patch_method = BINARY_DELTA_DEFLATED;1781 origlen = strtoul(buffer + 6, NULL, 10);1782 }1783 else if (starts_with(buffer, "literal ")) {1784 patch_method = BINARY_LITERAL_DEFLATED;1785 origlen = strtoul(buffer + 8, NULL, 10);1786 }1787 else1788 return NULL;17891790 state->linenr++;1791 buffer += llen;1792 while (1) {1793 int byte_length, max_byte_length, newsize;1794 llen = linelen(buffer, size);1795 used += llen;1796 state->linenr++;1797 if (llen == 1) {1798 /* consume the blank line */1799 buffer++;1800 size--;1801 break;1802 }1803 /*1804 * Minimum line is "A00000\n" which is 7-byte long,1805 * and the line length must be multiple of 5 plus 2.1806 */1807 if ((llen < 7) || (llen-2) % 5)1808 goto corrupt;1809 max_byte_length = (llen - 2) / 5 * 4;1810 byte_length = *buffer;1811 if ('A' <= byte_length && byte_length <= 'Z')1812 byte_length = byte_length - 'A' + 1;1813 else if ('a' <= byte_length && byte_length <= 'z')1814 byte_length = byte_length - 'a' + 27;1815 else1816 goto corrupt;1817 /* if the input length was not multiple of 4, we would1818 * have filler at the end but the filler should never1819 * exceed 3 bytes1820 */1821 if (max_byte_length < byte_length ||1822 byte_length <= max_byte_length - 4)1823 goto corrupt;1824 newsize = hunk_size + byte_length;1825 data = xrealloc(data, newsize);1826 if (decode_85(data + hunk_size, buffer + 1, byte_length))1827 goto corrupt;1828 hunk_size = newsize;1829 buffer += llen;1830 size -= llen;1831 }18321833 frag = xcalloc(1, sizeof(*frag));1834 frag->patch = inflate_it(data, hunk_size, origlen);1835 frag->free_patch = 1;1836 if (!frag->patch)1837 goto corrupt;1838 free(data);1839 frag->size = origlen;1840 *buf_p = buffer;1841 *sz_p = size;1842 *used_p = used;1843 frag->binary_patch_method = patch_method;1844 return frag;18451846 corrupt:1847 free(data);1848 *status_p = -1;1849 error(_("corrupt binary patch at line %d: %.*s"),1850 state->linenr-1, llen-1, buffer);1851 return NULL;1852}18531854/*1855 * Returns:1856 * -1 in case of error,1857 * the length of the parsed binary patch otherwise1858 */1859static int parse_binary(struct apply_state *state,1860 char *buffer,1861 unsigned long size,1862 struct patch *patch)1863{1864 /*1865 * We have read "GIT binary patch\n"; what follows is a line1866 * that says the patch method (currently, either "literal" or1867 * "delta") and the length of data before deflating; a1868 * sequence of 'length-byte' followed by base-85 encoded data1869 * follows.1870 *1871 * When a binary patch is reversible, there is another binary1872 * hunk in the same format, starting with patch method (either1873 * "literal" or "delta") with the length of data, and a sequence1874 * of length-byte + base-85 encoded data, terminated with another1875 * empty line. This data, when applied to the postimage, produces1876 * the preimage.1877 */1878 struct fragment *forward;1879 struct fragment *reverse;1880 int status;1881 int used, used_1;18821883 forward = parse_binary_hunk(state, &buffer, &size, &status, &used);1884 if (!forward && !status)1885 /* there has to be one hunk (forward hunk) */1886 return error(_("unrecognized binary patch at line %d"), state->linenr-1);1887 if (status)1888 /* otherwise we already gave an error message */1889 return status;18901891 reverse = parse_binary_hunk(state, &buffer, &size, &status, &used_1);1892 if (reverse)1893 used += used_1;1894 else if (status) {1895 /*1896 * Not having reverse hunk is not an error, but having1897 * a corrupt reverse hunk is.1898 */1899 free((void*) forward->patch);1900 free(forward);1901 return status;1902 }1903 forward->next = reverse;1904 patch->fragments = forward;1905 patch->is_binary = 1;1906 return used;1907}19081909static void prefix_one(struct apply_state *state, char **name)1910{1911 char *old_name = *name;1912 if (!old_name)1913 return;1914 *name = xstrdup(prefix_filename(state->prefix, state->prefix_length, *name));1915 free(old_name);1916}19171918static void prefix_patch(struct apply_state *state, struct patch *p)1919{1920 if (!state->prefix || p->is_toplevel_relative)1921 return;1922 prefix_one(state, &p->new_name);1923 prefix_one(state, &p->old_name);1924}19251926/*1927 * include/exclude1928 */19291930static void add_name_limit(struct apply_state *state,1931 const char *name,1932 int exclude)1933{1934 struct string_list_item *it;19351936 it = string_list_append(&state->limit_by_name, name);1937 it->util = exclude ? NULL : (void *) 1;1938}19391940static int use_patch(struct apply_state *state, struct patch *p)1941{1942 const char *pathname = p->new_name ? p->new_name : p->old_name;1943 int i;19441945 /* Paths outside are not touched regardless of "--include" */1946 if (0 < state->prefix_length) {1947 int pathlen = strlen(pathname);1948 if (pathlen <= state->prefix_length ||1949 memcmp(state->prefix, pathname, state->prefix_length))1950 return 0;1951 }19521953 /* See if it matches any of exclude/include rule */1954 for (i = 0; i < state->limit_by_name.nr; i++) {1955 struct string_list_item *it = &state->limit_by_name.items[i];1956 if (!wildmatch(it->string, pathname, 0, NULL))1957 return (it->util != NULL);1958 }19591960 /*1961 * If we had any include, a path that does not match any rule is1962 * not used. Otherwise, we saw bunch of exclude rules (or none)1963 * and such a path is used.1964 */1965 return !state->has_include;1966}19671968/*1969 * Read the patch text in "buffer" that extends for "size" bytes; stop1970 * reading after seeing a single patch (i.e. changes to a single file).1971 * Create fragments (i.e. patch hunks) and hang them to the given patch.1972 *1973 * Returns:1974 * -1 if no header was found or parse_binary() failed,1975 * -128 on another error,1976 * the number of bytes consumed otherwise,1977 * so that the caller can call us again for the next patch.1978 */1979static int parse_chunk(struct apply_state *state, char *buffer, unsigned long size, struct patch *patch)1980{1981 int hdrsize, patchsize;1982 int offset = find_header(state, buffer, size, &hdrsize, patch);19831984 if (offset < 0)1985 return offset;19861987 prefix_patch(state, patch);19881989 if (!use_patch(state, patch))1990 patch->ws_rule = 0;1991 else1992 patch->ws_rule = whitespace_rule(patch->new_name1993 ? patch->new_name1994 : patch->old_name);19951996 patchsize = parse_single_patch(state,1997 buffer + offset + hdrsize,1998 size - offset - hdrsize,1999 patch);20002001 if (patchsize < 0)2002 return -128;20032004 if (!patchsize) {2005 static const char git_binary[] = "GIT binary patch\n";2006 int hd = hdrsize + offset;2007 unsigned long llen = linelen(buffer + hd, size - hd);20082009 if (llen == sizeof(git_binary) - 1 &&2010 !memcmp(git_binary, buffer + hd, llen)) {2011 int used;2012 state->linenr++;2013 used = parse_binary(state, buffer + hd + llen,2014 size - hd - llen, patch);2015 if (used < 0)2016 return -1;2017 if (used)2018 patchsize = used + llen;2019 else2020 patchsize = 0;2021 }2022 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {2023 static const char *binhdr[] = {2024 "Binary files ",2025 "Files ",2026 NULL,2027 };2028 int i;2029 for (i = 0; binhdr[i]; i++) {2030 int len = strlen(binhdr[i]);2031 if (len < size - hd &&2032 !memcmp(binhdr[i], buffer + hd, len)) {2033 state->linenr++;2034 patch->is_binary = 1;2035 patchsize = llen;2036 break;2037 }2038 }2039 }20402041 /* Empty patch cannot be applied if it is a text patch2042 * without metadata change. A binary patch appears2043 * empty to us here.2044 */2045 if ((state->apply || state->check) &&2046 (!patch->is_binary && !metadata_changes(patch))) {2047 error(_("patch with only garbage at line %d"), state->linenr);2048 return -128;2049 }2050 }20512052 return offset + hdrsize + patchsize;2053}20542055#define swap(a,b) myswap((a),(b),sizeof(a))20562057#define myswap(a, b, size) do { \2058 unsigned char mytmp[size]; \2059 memcpy(mytmp, &a, size); \2060 memcpy(&a, &b, size); \2061 memcpy(&b, mytmp, size); \2062} while (0)20632064static void reverse_patches(struct patch *p)2065{2066 for (; p; p = p->next) {2067 struct fragment *frag = p->fragments;20682069 swap(p->new_name, p->old_name);2070 swap(p->new_mode, p->old_mode);2071 swap(p->is_new, p->is_delete);2072 swap(p->lines_added, p->lines_deleted);2073 swap(p->old_sha1_prefix, p->new_sha1_prefix);20742075 for (; frag; frag = frag->next) {2076 swap(frag->newpos, frag->oldpos);2077 swap(frag->newlines, frag->oldlines);2078 }2079 }2080}20812082static const char pluses[] =2083"++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";2084static const char minuses[]=2085"----------------------------------------------------------------------";20862087static void show_stats(struct apply_state *state, struct patch *patch)2088{2089 struct strbuf qname = STRBUF_INIT;2090 char *cp = patch->new_name ? patch->new_name : patch->old_name;2091 int max, add, del;20922093 quote_c_style(cp, &qname, NULL, 0);20942095 /*2096 * "scale" the filename2097 */2098 max = state->max_len;2099 if (max > 50)2100 max = 50;21012102 if (qname.len > max) {2103 cp = strchr(qname.buf + qname.len + 3 - max, '/');2104 if (!cp)2105 cp = qname.buf + qname.len + 3 - max;2106 strbuf_splice(&qname, 0, cp - qname.buf, "...", 3);2107 }21082109 if (patch->is_binary) {2110 printf(" %-*s | Bin\n", max, qname.buf);2111 strbuf_release(&qname);2112 return;2113 }21142115 printf(" %-*s |", max, qname.buf);2116 strbuf_release(&qname);21172118 /*2119 * scale the add/delete2120 */2121 max = max + state->max_change > 70 ? 70 - max : state->max_change;2122 add = patch->lines_added;2123 del = patch->lines_deleted;21242125 if (state->max_change > 0) {2126 int total = ((add + del) * max + state->max_change / 2) / state->max_change;2127 add = (add * max + state->max_change / 2) / state->max_change;2128 del = total - add;2129 }2130 printf("%5d %.*s%.*s\n", patch->lines_added + patch->lines_deleted,2131 add, pluses, del, minuses);2132}21332134static int read_old_data(struct stat *st, const char *path, struct strbuf *buf)2135{2136 switch (st->st_mode & S_IFMT) {2137 case S_IFLNK:2138 if (strbuf_readlink(buf, path, st->st_size) < 0)2139 return error(_("unable to read symlink %s"), path);2140 return 0;2141 case S_IFREG:2142 if (strbuf_read_file(buf, path, st->st_size) != st->st_size)2143 return error(_("unable to open or read %s"), path);2144 convert_to_git(path, buf->buf, buf->len, buf, 0);2145 return 0;2146 default:2147 return -1;2148 }2149}21502151/*2152 * Update the preimage, and the common lines in postimage,2153 * from buffer buf of length len. If postlen is 0 the postimage2154 * is updated in place, otherwise it's updated on a new buffer2155 * of length postlen2156 */21572158static void update_pre_post_images(struct image *preimage,2159 struct image *postimage,2160 char *buf,2161 size_t len, size_t postlen)2162{2163 int i, ctx, reduced;2164 char *new, *old, *fixed;2165 struct image fixed_preimage;21662167 /*2168 * Update the preimage with whitespace fixes. Note that we2169 * are not losing preimage->buf -- apply_one_fragment() will2170 * free "oldlines".2171 */2172 prepare_image(&fixed_preimage, buf, len, 1);2173 assert(postlen2174 ? fixed_preimage.nr == preimage->nr2175 : fixed_preimage.nr <= preimage->nr);2176 for (i = 0; i < fixed_preimage.nr; i++)2177 fixed_preimage.line[i].flag = preimage->line[i].flag;2178 free(preimage->line_allocated);2179 *preimage = fixed_preimage;21802181 /*2182 * Adjust the common context lines in postimage. This can be2183 * done in-place when we are shrinking it with whitespace2184 * fixing, but needs a new buffer when ignoring whitespace or2185 * expanding leading tabs to spaces.2186 *2187 * We trust the caller to tell us if the update can be done2188 * in place (postlen==0) or not.2189 */2190 old = postimage->buf;2191 if (postlen)2192 new = postimage->buf = xmalloc(postlen);2193 else2194 new = old;2195 fixed = preimage->buf;21962197 for (i = reduced = ctx = 0; i < postimage->nr; i++) {2198 size_t l_len = postimage->line[i].len;2199 if (!(postimage->line[i].flag & LINE_COMMON)) {2200 /* an added line -- no counterparts in preimage */2201 memmove(new, old, l_len);2202 old += l_len;2203 new += l_len;2204 continue;2205 }22062207 /* a common context -- skip it in the original postimage */2208 old += l_len;22092210 /* and find the corresponding one in the fixed preimage */2211 while (ctx < preimage->nr &&2212 !(preimage->line[ctx].flag & LINE_COMMON)) {2213 fixed += preimage->line[ctx].len;2214 ctx++;2215 }22162217 /*2218 * preimage is expected to run out, if the caller2219 * fixed addition of trailing blank lines.2220 */2221 if (preimage->nr <= ctx) {2222 reduced++;2223 continue;2224 }22252226 /* and copy it in, while fixing the line length */2227 l_len = preimage->line[ctx].len;2228 memcpy(new, fixed, l_len);2229 new += l_len;2230 fixed += l_len;2231 postimage->line[i].len = l_len;2232 ctx++;2233 }22342235 if (postlen2236 ? postlen < new - postimage->buf2237 : postimage->len < new - postimage->buf)2238 die("BUG: caller miscounted postlen: asked %d, orig = %d, used = %d",2239 (int)postlen, (int) postimage->len, (int)(new - postimage->buf));22402241 /* Fix the length of the whole thing */2242 postimage->len = new - postimage->buf;2243 postimage->nr -= reduced;2244}22452246static int line_by_line_fuzzy_match(struct image *img,2247 struct image *preimage,2248 struct image *postimage,2249 unsigned long try,2250 int try_lno,2251 int preimage_limit)2252{2253 int i;2254 size_t imgoff = 0;2255 size_t preoff = 0;2256 size_t postlen = postimage->len;2257 size_t extra_chars;2258 char *buf;2259 char *preimage_eof;2260 char *preimage_end;2261 struct strbuf fixed;2262 char *fixed_buf;2263 size_t fixed_len;22642265 for (i = 0; i < preimage_limit; i++) {2266 size_t prelen = preimage->line[i].len;2267 size_t imglen = img->line[try_lno+i].len;22682269 if (!fuzzy_matchlines(img->buf + try + imgoff, imglen,2270 preimage->buf + preoff, prelen))2271 return 0;2272 if (preimage->line[i].flag & LINE_COMMON)2273 postlen += imglen - prelen;2274 imgoff += imglen;2275 preoff += prelen;2276 }22772278 /*2279 * Ok, the preimage matches with whitespace fuzz.2280 *2281 * imgoff now holds the true length of the target that2282 * matches the preimage before the end of the file.2283 *2284 * Count the number of characters in the preimage that fall2285 * beyond the end of the file and make sure that all of them2286 * are whitespace characters. (This can only happen if2287 * we are removing blank lines at the end of the file.)2288 */2289 buf = preimage_eof = preimage->buf + preoff;2290 for ( ; i < preimage->nr; i++)2291 preoff += preimage->line[i].len;2292 preimage_end = preimage->buf + preoff;2293 for ( ; buf < preimage_end; buf++)2294 if (!isspace(*buf))2295 return 0;22962297 /*2298 * Update the preimage and the common postimage context2299 * lines to use the same whitespace as the target.2300 * If whitespace is missing in the target (i.e.2301 * if the preimage extends beyond the end of the file),2302 * use the whitespace from the preimage.2303 */2304 extra_chars = preimage_end - preimage_eof;2305 strbuf_init(&fixed, imgoff + extra_chars);2306 strbuf_add(&fixed, img->buf + try, imgoff);2307 strbuf_add(&fixed, preimage_eof, extra_chars);2308 fixed_buf = strbuf_detach(&fixed, &fixed_len);2309 update_pre_post_images(preimage, postimage,2310 fixed_buf, fixed_len, postlen);2311 return 1;2312}23132314static int match_fragment(struct apply_state *state,2315 struct image *img,2316 struct image *preimage,2317 struct image *postimage,2318 unsigned long try,2319 int try_lno,2320 unsigned ws_rule,2321 int match_beginning, int match_end)2322{2323 int i;2324 char *fixed_buf, *buf, *orig, *target;2325 struct strbuf fixed;2326 size_t fixed_len, postlen;2327 int preimage_limit;23282329 if (preimage->nr + try_lno <= img->nr) {2330 /*2331 * The hunk falls within the boundaries of img.2332 */2333 preimage_limit = preimage->nr;2334 if (match_end && (preimage->nr + try_lno != img->nr))2335 return 0;2336 } else if (state->ws_error_action == correct_ws_error &&2337 (ws_rule & WS_BLANK_AT_EOF)) {2338 /*2339 * This hunk extends beyond the end of img, and we are2340 * removing blank lines at the end of the file. This2341 * many lines from the beginning of the preimage must2342 * match with img, and the remainder of the preimage2343 * must be blank.2344 */2345 preimage_limit = img->nr - try_lno;2346 } else {2347 /*2348 * The hunk extends beyond the end of the img and2349 * we are not removing blanks at the end, so we2350 * should reject the hunk at this position.2351 */2352 return 0;2353 }23542355 if (match_beginning && try_lno)2356 return 0;23572358 /* Quick hash check */2359 for (i = 0; i < preimage_limit; i++)2360 if ((img->line[try_lno + i].flag & LINE_PATCHED) ||2361 (preimage->line[i].hash != img->line[try_lno + i].hash))2362 return 0;23632364 if (preimage_limit == preimage->nr) {2365 /*2366 * Do we have an exact match? If we were told to match2367 * at the end, size must be exactly at try+fragsize,2368 * otherwise try+fragsize must be still within the preimage,2369 * and either case, the old piece should match the preimage2370 * exactly.2371 */2372 if ((match_end2373 ? (try + preimage->len == img->len)2374 : (try + preimage->len <= img->len)) &&2375 !memcmp(img->buf + try, preimage->buf, preimage->len))2376 return 1;2377 } else {2378 /*2379 * The preimage extends beyond the end of img, so2380 * there cannot be an exact match.2381 *2382 * There must be one non-blank context line that match2383 * a line before the end of img.2384 */2385 char *buf_end;23862387 buf = preimage->buf;2388 buf_end = buf;2389 for (i = 0; i < preimage_limit; i++)2390 buf_end += preimage->line[i].len;23912392 for ( ; buf < buf_end; buf++)2393 if (!isspace(*buf))2394 break;2395 if (buf == buf_end)2396 return 0;2397 }23982399 /*2400 * No exact match. If we are ignoring whitespace, run a line-by-line2401 * fuzzy matching. We collect all the line length information because2402 * we need it to adjust whitespace if we match.2403 */2404 if (state->ws_ignore_action == ignore_ws_change)2405 return line_by_line_fuzzy_match(img, preimage, postimage,2406 try, try_lno, preimage_limit);24072408 if (state->ws_error_action != correct_ws_error)2409 return 0;24102411 /*2412 * The hunk does not apply byte-by-byte, but the hash says2413 * it might with whitespace fuzz. We weren't asked to2414 * ignore whitespace, we were asked to correct whitespace2415 * errors, so let's try matching after whitespace correction.2416 *2417 * While checking the preimage against the target, whitespace2418 * errors in both fixed, we count how large the corresponding2419 * postimage needs to be. The postimage prepared by2420 * apply_one_fragment() has whitespace errors fixed on added2421 * lines already, but the common lines were propagated as-is,2422 * which may become longer when their whitespace errors are2423 * fixed.2424 */24252426 /* First count added lines in postimage */2427 postlen = 0;2428 for (i = 0; i < postimage->nr; i++) {2429 if (!(postimage->line[i].flag & LINE_COMMON))2430 postlen += postimage->line[i].len;2431 }24322433 /*2434 * The preimage may extend beyond the end of the file,2435 * but in this loop we will only handle the part of the2436 * preimage that falls within the file.2437 */2438 strbuf_init(&fixed, preimage->len + 1);2439 orig = preimage->buf;2440 target = img->buf + try;2441 for (i = 0; i < preimage_limit; i++) {2442 size_t oldlen = preimage->line[i].len;2443 size_t tgtlen = img->line[try_lno + i].len;2444 size_t fixstart = fixed.len;2445 struct strbuf tgtfix;2446 int match;24472448 /* Try fixing the line in the preimage */2449 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24502451 /* Try fixing the line in the target */2452 strbuf_init(&tgtfix, tgtlen);2453 ws_fix_copy(&tgtfix, target, tgtlen, ws_rule, NULL);24542455 /*2456 * If they match, either the preimage was based on2457 * a version before our tree fixed whitespace breakage,2458 * or we are lacking a whitespace-fix patch the tree2459 * the preimage was based on already had (i.e. target2460 * has whitespace breakage, the preimage doesn't).2461 * In either case, we are fixing the whitespace breakages2462 * so we might as well take the fix together with their2463 * real change.2464 */2465 match = (tgtfix.len == fixed.len - fixstart &&2466 !memcmp(tgtfix.buf, fixed.buf + fixstart,2467 fixed.len - fixstart));24682469 /* Add the length if this is common with the postimage */2470 if (preimage->line[i].flag & LINE_COMMON)2471 postlen += tgtfix.len;24722473 strbuf_release(&tgtfix);2474 if (!match)2475 goto unmatch_exit;24762477 orig += oldlen;2478 target += tgtlen;2479 }248024812482 /*2483 * Now handle the lines in the preimage that falls beyond the2484 * end of the file (if any). They will only match if they are2485 * empty or only contain whitespace (if WS_BLANK_AT_EOL is2486 * false).2487 */2488 for ( ; i < preimage->nr; i++) {2489 size_t fixstart = fixed.len; /* start of the fixed preimage */2490 size_t oldlen = preimage->line[i].len;2491 int j;24922493 /* Try fixing the line in the preimage */2494 ws_fix_copy(&fixed, orig, oldlen, ws_rule, NULL);24952496 for (j = fixstart; j < fixed.len; j++)2497 if (!isspace(fixed.buf[j]))2498 goto unmatch_exit;24992500 orig += oldlen;2501 }25022503 /*2504 * Yes, the preimage is based on an older version that still2505 * has whitespace breakages unfixed, and fixing them makes the2506 * hunk match. Update the context lines in the postimage.2507 */2508 fixed_buf = strbuf_detach(&fixed, &fixed_len);2509 if (postlen < postimage->len)2510 postlen = 0;2511 update_pre_post_images(preimage, postimage,2512 fixed_buf, fixed_len, postlen);2513 return 1;25142515 unmatch_exit:2516 strbuf_release(&fixed);2517 return 0;2518}25192520static int find_pos(struct apply_state *state,2521 struct image *img,2522 struct image *preimage,2523 struct image *postimage,2524 int line,2525 unsigned ws_rule,2526 int match_beginning, int match_end)2527{2528 int i;2529 unsigned long backwards, forwards, try;2530 int backwards_lno, forwards_lno, try_lno;25312532 /*2533 * If match_beginning or match_end is specified, there is no2534 * point starting from a wrong line that will never match and2535 * wander around and wait for a match at the specified end.2536 */2537 if (match_beginning)2538 line = 0;2539 else if (match_end)2540 line = img->nr - preimage->nr;25412542 /*2543 * Because the comparison is unsigned, the following test2544 * will also take care of a negative line number that can2545 * result when match_end and preimage is larger than the target.2546 */2547 if ((size_t) line > img->nr)2548 line = img->nr;25492550 try = 0;2551 for (i = 0; i < line; i++)2552 try += img->line[i].len;25532554 /*2555 * There's probably some smart way to do this, but I'll leave2556 * that to the smart and beautiful people. I'm simple and stupid.2557 */2558 backwards = try;2559 backwards_lno = line;2560 forwards = try;2561 forwards_lno = line;2562 try_lno = line;25632564 for (i = 0; ; i++) {2565 if (match_fragment(state, img, preimage, postimage,2566 try, try_lno, ws_rule,2567 match_beginning, match_end))2568 return try_lno;25692570 again:2571 if (backwards_lno == 0 && forwards_lno == img->nr)2572 break;25732574 if (i & 1) {2575 if (backwards_lno == 0) {2576 i++;2577 goto again;2578 }2579 backwards_lno--;2580 backwards -= img->line[backwards_lno].len;2581 try = backwards;2582 try_lno = backwards_lno;2583 } else {2584 if (forwards_lno == img->nr) {2585 i++;2586 goto again;2587 }2588 forwards += img->line[forwards_lno].len;2589 forwards_lno++;2590 try = forwards;2591 try_lno = forwards_lno;2592 }25932594 }2595 return -1;2596}25972598static void remove_first_line(struct image *img)2599{2600 img->buf += img->line[0].len;2601 img->len -= img->line[0].len;2602 img->line++;2603 img->nr--;2604}26052606static void remove_last_line(struct image *img)2607{2608 img->len -= img->line[--img->nr].len;2609}26102611/*2612 * The change from "preimage" and "postimage" has been found to2613 * apply at applied_pos (counts in line numbers) in "img".2614 * Update "img" to remove "preimage" and replace it with "postimage".2615 */2616static void update_image(struct apply_state *state,2617 struct image *img,2618 int applied_pos,2619 struct image *preimage,2620 struct image *postimage)2621{2622 /*2623 * remove the copy of preimage at offset in img2624 * and replace it with postimage2625 */2626 int i, nr;2627 size_t remove_count, insert_count, applied_at = 0;2628 char *result;2629 int preimage_limit;26302631 /*2632 * If we are removing blank lines at the end of img,2633 * the preimage may extend beyond the end.2634 * If that is the case, we must be careful only to2635 * remove the part of the preimage that falls within2636 * the boundaries of img. Initialize preimage_limit2637 * to the number of lines in the preimage that falls2638 * within the boundaries.2639 */2640 preimage_limit = preimage->nr;2641 if (preimage_limit > img->nr - applied_pos)2642 preimage_limit = img->nr - applied_pos;26432644 for (i = 0; i < applied_pos; i++)2645 applied_at += img->line[i].len;26462647 remove_count = 0;2648 for (i = 0; i < preimage_limit; i++)2649 remove_count += img->line[applied_pos + i].len;2650 insert_count = postimage->len;26512652 /* Adjust the contents */2653 result = xmalloc(st_add3(st_sub(img->len, remove_count), insert_count, 1));2654 memcpy(result, img->buf, applied_at);2655 memcpy(result + applied_at, postimage->buf, postimage->len);2656 memcpy(result + applied_at + postimage->len,2657 img->buf + (applied_at + remove_count),2658 img->len - (applied_at + remove_count));2659 free(img->buf);2660 img->buf = result;2661 img->len += insert_count - remove_count;2662 result[img->len] = '\0';26632664 /* Adjust the line table */2665 nr = img->nr + postimage->nr - preimage_limit;2666 if (preimage_limit < postimage->nr) {2667 /*2668 * NOTE: this knows that we never call remove_first_line()2669 * on anything other than pre/post image.2670 */2671 REALLOC_ARRAY(img->line, nr);2672 img->line_allocated = img->line;2673 }2674 if (preimage_limit != postimage->nr)2675 memmove(img->line + applied_pos + postimage->nr,2676 img->line + applied_pos + preimage_limit,2677 (img->nr - (applied_pos + preimage_limit)) *2678 sizeof(*img->line));2679 memcpy(img->line + applied_pos,2680 postimage->line,2681 postimage->nr * sizeof(*img->line));2682 if (!state->allow_overlap)2683 for (i = 0; i < postimage->nr; i++)2684 img->line[applied_pos + i].flag |= LINE_PATCHED;2685 img->nr = nr;2686}26872688/*2689 * Use the patch-hunk text in "frag" to prepare two images (preimage and2690 * postimage) for the hunk. Find lines that match "preimage" in "img" and2691 * replace the part of "img" with "postimage" text.2692 */2693static int apply_one_fragment(struct apply_state *state,2694 struct image *img, struct fragment *frag,2695 int inaccurate_eof, unsigned ws_rule,2696 int nth_fragment)2697{2698 int match_beginning, match_end;2699 const char *patch = frag->patch;2700 int size = frag->size;2701 char *old, *oldlines;2702 struct strbuf newlines;2703 int new_blank_lines_at_end = 0;2704 int found_new_blank_lines_at_end = 0;2705 int hunk_linenr = frag->linenr;2706 unsigned long leading, trailing;2707 int pos, applied_pos;2708 struct image preimage;2709 struct image postimage;27102711 memset(&preimage, 0, sizeof(preimage));2712 memset(&postimage, 0, sizeof(postimage));2713 oldlines = xmalloc(size);2714 strbuf_init(&newlines, size);27152716 old = oldlines;2717 while (size > 0) {2718 char first;2719 int len = linelen(patch, size);2720 int plen;2721 int added_blank_line = 0;2722 int is_blank_context = 0;2723 size_t start;27242725 if (!len)2726 break;27272728 /*2729 * "plen" is how much of the line we should use for2730 * the actual patch data. Normally we just remove the2731 * first character on the line, but if the line is2732 * followed by "\ No newline", then we also remove the2733 * last one (which is the newline, of course).2734 */2735 plen = len - 1;2736 if (len < size && patch[len] == '\\')2737 plen--;2738 first = *patch;2739 if (state->apply_in_reverse) {2740 if (first == '-')2741 first = '+';2742 else if (first == '+')2743 first = '-';2744 }27452746 switch (first) {2747 case '\n':2748 /* Newer GNU diff, empty context line */2749 if (plen < 0)2750 /* ... followed by '\No newline'; nothing */2751 break;2752 *old++ = '\n';2753 strbuf_addch(&newlines, '\n');2754 add_line_info(&preimage, "\n", 1, LINE_COMMON);2755 add_line_info(&postimage, "\n", 1, LINE_COMMON);2756 is_blank_context = 1;2757 break;2758 case ' ':2759 if (plen && (ws_rule & WS_BLANK_AT_EOF) &&2760 ws_blank_line(patch + 1, plen, ws_rule))2761 is_blank_context = 1;2762 case '-':2763 memcpy(old, patch + 1, plen);2764 add_line_info(&preimage, old, plen,2765 (first == ' ' ? LINE_COMMON : 0));2766 old += plen;2767 if (first == '-')2768 break;2769 /* Fall-through for ' ' */2770 case '+':2771 /* --no-add does not add new lines */2772 if (first == '+' && state->no_add)2773 break;27742775 start = newlines.len;2776 if (first != '+' ||2777 !state->whitespace_error ||2778 state->ws_error_action != correct_ws_error) {2779 strbuf_add(&newlines, patch + 1, plen);2780 }2781 else {2782 ws_fix_copy(&newlines, patch + 1, plen, ws_rule, &state->applied_after_fixing_ws);2783 }2784 add_line_info(&postimage, newlines.buf + start, newlines.len - start,2785 (first == '+' ? 0 : LINE_COMMON));2786 if (first == '+' &&2787 (ws_rule & WS_BLANK_AT_EOF) &&2788 ws_blank_line(patch + 1, plen, ws_rule))2789 added_blank_line = 1;2790 break;2791 case '@': case '\\':2792 /* Ignore it, we already handled it */2793 break;2794 default:2795 if (state->apply_verbosely)2796 error(_("invalid start of line: '%c'"), first);2797 applied_pos = -1;2798 goto out;2799 }2800 if (added_blank_line) {2801 if (!new_blank_lines_at_end)2802 found_new_blank_lines_at_end = hunk_linenr;2803 new_blank_lines_at_end++;2804 }2805 else if (is_blank_context)2806 ;2807 else2808 new_blank_lines_at_end = 0;2809 patch += len;2810 size -= len;2811 hunk_linenr++;2812 }2813 if (inaccurate_eof &&2814 old > oldlines && old[-1] == '\n' &&2815 newlines.len > 0 && newlines.buf[newlines.len - 1] == '\n') {2816 old--;2817 strbuf_setlen(&newlines, newlines.len - 1);2818 }28192820 leading = frag->leading;2821 trailing = frag->trailing;28222823 /*2824 * A hunk to change lines at the beginning would begin with2825 * @@ -1,L +N,M @@2826 * but we need to be careful. -U0 that inserts before the second2827 * line also has this pattern.2828 *2829 * And a hunk to add to an empty file would begin with2830 * @@ -0,0 +N,M @@2831 *2832 * In other words, a hunk that is (frag->oldpos <= 1) with or2833 * without leading context must match at the beginning.2834 */2835 match_beginning = (!frag->oldpos ||2836 (frag->oldpos == 1 && !state->unidiff_zero));28372838 /*2839 * A hunk without trailing lines must match at the end.2840 * However, we simply cannot tell if a hunk must match end2841 * from the lack of trailing lines if the patch was generated2842 * with unidiff without any context.2843 */2844 match_end = !state->unidiff_zero && !trailing;28452846 pos = frag->newpos ? (frag->newpos - 1) : 0;2847 preimage.buf = oldlines;2848 preimage.len = old - oldlines;2849 postimage.buf = newlines.buf;2850 postimage.len = newlines.len;2851 preimage.line = preimage.line_allocated;2852 postimage.line = postimage.line_allocated;28532854 for (;;) {28552856 applied_pos = find_pos(state, img, &preimage, &postimage, pos,2857 ws_rule, match_beginning, match_end);28582859 if (applied_pos >= 0)2860 break;28612862 /* Am I at my context limits? */2863 if ((leading <= state->p_context) && (trailing <= state->p_context))2864 break;2865 if (match_beginning || match_end) {2866 match_beginning = match_end = 0;2867 continue;2868 }28692870 /*2871 * Reduce the number of context lines; reduce both2872 * leading and trailing if they are equal otherwise2873 * just reduce the larger context.2874 */2875 if (leading >= trailing) {2876 remove_first_line(&preimage);2877 remove_first_line(&postimage);2878 pos--;2879 leading--;2880 }2881 if (trailing > leading) {2882 remove_last_line(&preimage);2883 remove_last_line(&postimage);2884 trailing--;2885 }2886 }28872888 if (applied_pos >= 0) {2889 if (new_blank_lines_at_end &&2890 preimage.nr + applied_pos >= img->nr &&2891 (ws_rule & WS_BLANK_AT_EOF) &&2892 state->ws_error_action != nowarn_ws_error) {2893 record_ws_error(state, WS_BLANK_AT_EOF, "+", 1,2894 found_new_blank_lines_at_end);2895 if (state->ws_error_action == correct_ws_error) {2896 while (new_blank_lines_at_end--)2897 remove_last_line(&postimage);2898 }2899 /*2900 * We would want to prevent write_out_results()2901 * from taking place in apply_patch() that follows2902 * the callchain led us here, which is:2903 * apply_patch->check_patch_list->check_patch->2904 * apply_data->apply_fragments->apply_one_fragment2905 */2906 if (state->ws_error_action == die_on_ws_error)2907 state->apply = 0;2908 }29092910 if (state->apply_verbosely && applied_pos != pos) {2911 int offset = applied_pos - pos;2912 if (state->apply_in_reverse)2913 offset = 0 - offset;2914 fprintf_ln(stderr,2915 Q_("Hunk #%d succeeded at %d (offset %d line).",2916 "Hunk #%d succeeded at %d (offset %d lines).",2917 offset),2918 nth_fragment, applied_pos + 1, offset);2919 }29202921 /*2922 * Warn if it was necessary to reduce the number2923 * of context lines.2924 */2925 if ((leading != frag->leading) ||2926 (trailing != frag->trailing))2927 fprintf_ln(stderr, _("Context reduced to (%ld/%ld)"2928 " to apply fragment at %d"),2929 leading, trailing, applied_pos+1);2930 update_image(state, img, applied_pos, &preimage, &postimage);2931 } else {2932 if (state->apply_verbosely)2933 error(_("while searching for:\n%.*s"),2934 (int)(old - oldlines), oldlines);2935 }29362937out:2938 free(oldlines);2939 strbuf_release(&newlines);2940 free(preimage.line_allocated);2941 free(postimage.line_allocated);29422943 return (applied_pos < 0);2944}29452946static int apply_binary_fragment(struct apply_state *state,2947 struct image *img,2948 struct patch *patch)2949{2950 struct fragment *fragment = patch->fragments;2951 unsigned long len;2952 void *dst;29532954 if (!fragment)2955 return error(_("missing binary patch data for '%s'"),2956 patch->new_name ?2957 patch->new_name :2958 patch->old_name);29592960 /* Binary patch is irreversible without the optional second hunk */2961 if (state->apply_in_reverse) {2962 if (!fragment->next)2963 return error("cannot reverse-apply a binary patch "2964 "without the reverse hunk to '%s'",2965 patch->new_name2966 ? patch->new_name : patch->old_name);2967 fragment = fragment->next;2968 }2969 switch (fragment->binary_patch_method) {2970 case BINARY_DELTA_DEFLATED:2971 dst = patch_delta(img->buf, img->len, fragment->patch,2972 fragment->size, &len);2973 if (!dst)2974 return -1;2975 clear_image(img);2976 img->buf = dst;2977 img->len = len;2978 return 0;2979 case BINARY_LITERAL_DEFLATED:2980 clear_image(img);2981 img->len = fragment->size;2982 img->buf = xmemdupz(fragment->patch, img->len);2983 return 0;2984 }2985 return -1;2986}29872988/*2989 * Replace "img" with the result of applying the binary patch.2990 * The binary patch data itself in patch->fragment is still kept2991 * but the preimage prepared by the caller in "img" is freed here2992 * or in the helper function apply_binary_fragment() this calls.2993 */2994static int apply_binary(struct apply_state *state,2995 struct image *img,2996 struct patch *patch)2997{2998 const char *name = patch->old_name ? patch->old_name : patch->new_name;2999 unsigned char sha1[20];30003001 /*3002 * For safety, we require patch index line to contain3003 * full 40-byte textual SHA1 for old and new, at least for now.3004 */3005 if (strlen(patch->old_sha1_prefix) != 40 ||3006 strlen(patch->new_sha1_prefix) != 40 ||3007 get_sha1_hex(patch->old_sha1_prefix, sha1) ||3008 get_sha1_hex(patch->new_sha1_prefix, sha1))3009 return error("cannot apply binary patch to '%s' "3010 "without full index line", name);30113012 if (patch->old_name) {3013 /*3014 * See if the old one matches what the patch3015 * applies to.3016 */3017 hash_sha1_file(img->buf, img->len, blob_type, sha1);3018 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))3019 return error("the patch applies to '%s' (%s), "3020 "which does not match the "3021 "current contents.",3022 name, sha1_to_hex(sha1));3023 }3024 else {3025 /* Otherwise, the old one must be empty. */3026 if (img->len)3027 return error("the patch applies to an empty "3028 "'%s' but it is not empty", name);3029 }30303031 get_sha1_hex(patch->new_sha1_prefix, sha1);3032 if (is_null_sha1(sha1)) {3033 clear_image(img);3034 return 0; /* deletion patch */3035 }30363037 if (has_sha1_file(sha1)) {3038 /* We already have the postimage */3039 enum object_type type;3040 unsigned long size;3041 char *result;30423043 result = read_sha1_file(sha1, &type, &size);3044 if (!result)3045 return error("the necessary postimage %s for "3046 "'%s' cannot be read",3047 patch->new_sha1_prefix, name);3048 clear_image(img);3049 img->buf = result;3050 img->len = size;3051 } else {3052 /*3053 * We have verified buf matches the preimage;3054 * apply the patch data to it, which is stored3055 * in the patch->fragments->{patch,size}.3056 */3057 if (apply_binary_fragment(state, img, patch))3058 return error(_("binary patch does not apply to '%s'"),3059 name);30603061 /* verify that the result matches */3062 hash_sha1_file(img->buf, img->len, blob_type, sha1);3063 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))3064 return error(_("binary patch to '%s' creates incorrect result (expecting %s, got %s)"),3065 name, patch->new_sha1_prefix, sha1_to_hex(sha1));3066 }30673068 return 0;3069}30703071static int apply_fragments(struct apply_state *state, struct image *img, struct patch *patch)3072{3073 struct fragment *frag = patch->fragments;3074 const char *name = patch->old_name ? patch->old_name : patch->new_name;3075 unsigned ws_rule = patch->ws_rule;3076 unsigned inaccurate_eof = patch->inaccurate_eof;3077 int nth = 0;30783079 if (patch->is_binary)3080 return apply_binary(state, img, patch);30813082 while (frag) {3083 nth++;3084 if (apply_one_fragment(state, img, frag, inaccurate_eof, ws_rule, nth)) {3085 error(_("patch failed: %s:%ld"), name, frag->oldpos);3086 if (!state->apply_with_reject)3087 return -1;3088 frag->rejected = 1;3089 }3090 frag = frag->next;3091 }3092 return 0;3093}30943095static int read_blob_object(struct strbuf *buf, const unsigned char *sha1, unsigned mode)3096{3097 if (S_ISGITLINK(mode)) {3098 strbuf_grow(buf, 100);3099 strbuf_addf(buf, "Subproject commit %s\n", sha1_to_hex(sha1));3100 } else {3101 enum object_type type;3102 unsigned long sz;3103 char *result;31043105 result = read_sha1_file(sha1, &type, &sz);3106 if (!result)3107 return -1;3108 /* XXX read_sha1_file NUL-terminates */3109 strbuf_attach(buf, result, sz, sz + 1);3110 }3111 return 0;3112}31133114static int read_file_or_gitlink(const struct cache_entry *ce, struct strbuf *buf)3115{3116 if (!ce)3117 return 0;3118 return read_blob_object(buf, ce->sha1, ce->ce_mode);3119}31203121static struct patch *in_fn_table(struct apply_state *state, const char *name)3122{3123 struct string_list_item *item;31243125 if (name == NULL)3126 return NULL;31273128 item = string_list_lookup(&state->fn_table, name);3129 if (item != NULL)3130 return (struct patch *)item->util;31313132 return NULL;3133}31343135/*3136 * item->util in the filename table records the status of the path.3137 * Usually it points at a patch (whose result records the contents3138 * of it after applying it), but it could be PATH_WAS_DELETED for a3139 * path that a previously applied patch has already removed, or3140 * PATH_TO_BE_DELETED for a path that a later patch would remove.3141 *3142 * The latter is needed to deal with a case where two paths A and B3143 * are swapped by first renaming A to B and then renaming B to A;3144 * moving A to B should not be prevented due to presence of B as we3145 * will remove it in a later patch.3146 */3147#define PATH_TO_BE_DELETED ((struct patch *) -2)3148#define PATH_WAS_DELETED ((struct patch *) -1)31493150static int to_be_deleted(struct patch *patch)3151{3152 return patch == PATH_TO_BE_DELETED;3153}31543155static int was_deleted(struct patch *patch)3156{3157 return patch == PATH_WAS_DELETED;3158}31593160static void add_to_fn_table(struct apply_state *state, struct patch *patch)3161{3162 struct string_list_item *item;31633164 /*3165 * Always add new_name unless patch is a deletion3166 * This should cover the cases for normal diffs,3167 * file creations and copies3168 */3169 if (patch->new_name != NULL) {3170 item = string_list_insert(&state->fn_table, patch->new_name);3171 item->util = patch;3172 }31733174 /*3175 * store a failure on rename/deletion cases because3176 * later chunks shouldn't patch old names3177 */3178 if ((patch->new_name == NULL) || (patch->is_rename)) {3179 item = string_list_insert(&state->fn_table, patch->old_name);3180 item->util = PATH_WAS_DELETED;3181 }3182}31833184static void prepare_fn_table(struct apply_state *state, struct patch *patch)3185{3186 /*3187 * store information about incoming file deletion3188 */3189 while (patch) {3190 if ((patch->new_name == NULL) || (patch->is_rename)) {3191 struct string_list_item *item;3192 item = string_list_insert(&state->fn_table, patch->old_name);3193 item->util = PATH_TO_BE_DELETED;3194 }3195 patch = patch->next;3196 }3197}31983199static int checkout_target(struct index_state *istate,3200 struct cache_entry *ce, struct stat *st)3201{3202 struct checkout costate;32033204 memset(&costate, 0, sizeof(costate));3205 costate.base_dir = "";3206 costate.refresh_cache = 1;3207 costate.istate = istate;3208 if (checkout_entry(ce, &costate, NULL) || lstat(ce->name, st))3209 return error(_("cannot checkout %s"), ce->name);3210 return 0;3211}32123213static struct patch *previous_patch(struct apply_state *state,3214 struct patch *patch,3215 int *gone)3216{3217 struct patch *previous;32183219 *gone = 0;3220 if (patch->is_copy || patch->is_rename)3221 return NULL; /* "git" patches do not depend on the order */32223223 previous = in_fn_table(state, patch->old_name);3224 if (!previous)3225 return NULL;32263227 if (to_be_deleted(previous))3228 return NULL; /* the deletion hasn't happened yet */32293230 if (was_deleted(previous))3231 *gone = 1;32323233 return previous;3234}32353236static int verify_index_match(const struct cache_entry *ce, struct stat *st)3237{3238 if (S_ISGITLINK(ce->ce_mode)) {3239 if (!S_ISDIR(st->st_mode))3240 return -1;3241 return 0;3242 }3243 return ce_match_stat(ce, st, CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE);3244}32453246#define SUBMODULE_PATCH_WITHOUT_INDEX 132473248static int load_patch_target(struct apply_state *state,3249 struct strbuf *buf,3250 const struct cache_entry *ce,3251 struct stat *st,3252 const char *name,3253 unsigned expected_mode)3254{3255 if (state->cached || state->check_index) {3256 if (read_file_or_gitlink(ce, buf))3257 return error(_("failed to read %s"), name);3258 } else if (name) {3259 if (S_ISGITLINK(expected_mode)) {3260 if (ce)3261 return read_file_or_gitlink(ce, buf);3262 else3263 return SUBMODULE_PATCH_WITHOUT_INDEX;3264 } else if (has_symlink_leading_path(name, strlen(name))) {3265 return error(_("reading from '%s' beyond a symbolic link"), name);3266 } else {3267 if (read_old_data(st, name, buf))3268 return error(_("failed to read %s"), name);3269 }3270 }3271 return 0;3272}32733274/*3275 * We are about to apply "patch"; populate the "image" with the3276 * current version we have, from the working tree or from the index,3277 * depending on the situation e.g. --cached/--index. If we are3278 * applying a non-git patch that incrementally updates the tree,3279 * we read from the result of a previous diff.3280 */3281static int load_preimage(struct apply_state *state,3282 struct image *image,3283 struct patch *patch, struct stat *st,3284 const struct cache_entry *ce)3285{3286 struct strbuf buf = STRBUF_INIT;3287 size_t len;3288 char *img;3289 struct patch *previous;3290 int status;32913292 previous = previous_patch(state, patch, &status);3293 if (status)3294 return error(_("path %s has been renamed/deleted"),3295 patch->old_name);3296 if (previous) {3297 /* We have a patched copy in memory; use that. */3298 strbuf_add(&buf, previous->result, previous->resultsize);3299 } else {3300 status = load_patch_target(state, &buf, ce, st,3301 patch->old_name, patch->old_mode);3302 if (status < 0)3303 return status;3304 else if (status == SUBMODULE_PATCH_WITHOUT_INDEX) {3305 /*3306 * There is no way to apply subproject3307 * patch without looking at the index.3308 * NEEDSWORK: shouldn't this be flagged3309 * as an error???3310 */3311 free_fragment_list(patch->fragments);3312 patch->fragments = NULL;3313 } else if (status) {3314 return error(_("failed to read %s"), patch->old_name);3315 }3316 }33173318 img = strbuf_detach(&buf, &len);3319 prepare_image(image, img, len, !patch->is_binary);3320 return 0;3321}33223323static int three_way_merge(struct image *image,3324 char *path,3325 const unsigned char *base,3326 const unsigned char *ours,3327 const unsigned char *theirs)3328{3329 mmfile_t base_file, our_file, their_file;3330 mmbuffer_t result = { NULL };3331 int status;33323333 read_mmblob(&base_file, base);3334 read_mmblob(&our_file, ours);3335 read_mmblob(&their_file, theirs);3336 status = ll_merge(&result, path,3337 &base_file, "base",3338 &our_file, "ours",3339 &their_file, "theirs", NULL);3340 free(base_file.ptr);3341 free(our_file.ptr);3342 free(their_file.ptr);3343 if (status < 0 || !result.ptr) {3344 free(result.ptr);3345 return -1;3346 }3347 clear_image(image);3348 image->buf = result.ptr;3349 image->len = result.size;33503351 return status;3352}33533354/*3355 * When directly falling back to add/add three-way merge, we read from3356 * the current contents of the new_name. In no cases other than that3357 * this function will be called.3358 */3359static int load_current(struct apply_state *state,3360 struct image *image,3361 struct patch *patch)3362{3363 struct strbuf buf = STRBUF_INIT;3364 int status, pos;3365 size_t len;3366 char *img;3367 struct stat st;3368 struct cache_entry *ce;3369 char *name = patch->new_name;3370 unsigned mode = patch->new_mode;33713372 if (!patch->is_new)3373 die("BUG: patch to %s is not a creation", patch->old_name);33743375 pos = cache_name_pos(name, strlen(name));3376 if (pos < 0)3377 return error(_("%s: does not exist in index"), name);3378 ce = active_cache[pos];3379 if (lstat(name, &st)) {3380 if (errno != ENOENT)3381 return error(_("%s: %s"), name, strerror(errno));3382 if (checkout_target(&the_index, ce, &st))3383 return -1;3384 }3385 if (verify_index_match(ce, &st))3386 return error(_("%s: does not match index"), name);33873388 status = load_patch_target(state, &buf, ce, &st, name, mode);3389 if (status < 0)3390 return status;3391 else if (status)3392 return -1;3393 img = strbuf_detach(&buf, &len);3394 prepare_image(image, img, len, !patch->is_binary);3395 return 0;3396}33973398static int try_threeway(struct apply_state *state,3399 struct image *image,3400 struct patch *patch,3401 struct stat *st,3402 const struct cache_entry *ce)3403{3404 unsigned char pre_sha1[20], post_sha1[20], our_sha1[20];3405 struct strbuf buf = STRBUF_INIT;3406 size_t len;3407 int status;3408 char *img;3409 struct image tmp_image;34103411 /* No point falling back to 3-way merge in these cases */3412 if (patch->is_delete ||3413 S_ISGITLINK(patch->old_mode) || S_ISGITLINK(patch->new_mode))3414 return -1;34153416 /* Preimage the patch was prepared for */3417 if (patch->is_new)3418 write_sha1_file("", 0, blob_type, pre_sha1);3419 else if (get_sha1(patch->old_sha1_prefix, pre_sha1) ||3420 read_blob_object(&buf, pre_sha1, patch->old_mode))3421 return error("repository lacks the necessary blob to fall back on 3-way merge.");34223423 fprintf(stderr, "Falling back to three-way merge...\n");34243425 img = strbuf_detach(&buf, &len);3426 prepare_image(&tmp_image, img, len, 1);3427 /* Apply the patch to get the post image */3428 if (apply_fragments(state, &tmp_image, patch) < 0) {3429 clear_image(&tmp_image);3430 return -1;3431 }3432 /* post_sha1[] is theirs */3433 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, post_sha1);3434 clear_image(&tmp_image);34353436 /* our_sha1[] is ours */3437 if (patch->is_new) {3438 if (load_current(state, &tmp_image, patch))3439 return error("cannot read the current contents of '%s'",3440 patch->new_name);3441 } else {3442 if (load_preimage(state, &tmp_image, patch, st, ce))3443 return error("cannot read the current contents of '%s'",3444 patch->old_name);3445 }3446 write_sha1_file(tmp_image.buf, tmp_image.len, blob_type, our_sha1);3447 clear_image(&tmp_image);34483449 /* in-core three-way merge between post and our using pre as base */3450 status = three_way_merge(image, patch->new_name,3451 pre_sha1, our_sha1, post_sha1);3452 if (status < 0) {3453 fprintf(stderr, "Failed to fall back on three-way merge...\n");3454 return status;3455 }34563457 if (status) {3458 patch->conflicted_threeway = 1;3459 if (patch->is_new)3460 oidclr(&patch->threeway_stage[0]);3461 else3462 hashcpy(patch->threeway_stage[0].hash, pre_sha1);3463 hashcpy(patch->threeway_stage[1].hash, our_sha1);3464 hashcpy(patch->threeway_stage[2].hash, post_sha1);3465 fprintf(stderr, "Applied patch to '%s' with conflicts.\n", patch->new_name);3466 } else {3467 fprintf(stderr, "Applied patch to '%s' cleanly.\n", patch->new_name);3468 }3469 return 0;3470}34713472static int apply_data(struct apply_state *state, struct patch *patch,3473 struct stat *st, const struct cache_entry *ce)3474{3475 struct image image;34763477 if (load_preimage(state, &image, patch, st, ce) < 0)3478 return -1;34793480 if (patch->direct_to_threeway ||3481 apply_fragments(state, &image, patch) < 0) {3482 /* Note: with --reject, apply_fragments() returns 0 */3483 if (!state->threeway || try_threeway(state, &image, patch, st, ce) < 0)3484 return -1;3485 }3486 patch->result = image.buf;3487 patch->resultsize = image.len;3488 add_to_fn_table(state, patch);3489 free(image.line_allocated);34903491 if (0 < patch->is_delete && patch->resultsize)3492 return error(_("removal patch leaves file contents"));34933494 return 0;3495}34963497/*3498 * If "patch" that we are looking at modifies or deletes what we have,3499 * we would want it not to lose any local modification we have, either3500 * in the working tree or in the index.3501 *3502 * This also decides if a non-git patch is a creation patch or a3503 * modification to an existing empty file. We do not check the state3504 * of the current tree for a creation patch in this function; the caller3505 * check_patch() separately makes sure (and errors out otherwise) that3506 * the path the patch creates does not exist in the current tree.3507 */3508static int check_preimage(struct apply_state *state,3509 struct patch *patch,3510 struct cache_entry **ce,3511 struct stat *st)3512{3513 const char *old_name = patch->old_name;3514 struct patch *previous = NULL;3515 int stat_ret = 0, status;3516 unsigned st_mode = 0;35173518 if (!old_name)3519 return 0;35203521 assert(patch->is_new <= 0);3522 previous = previous_patch(state, patch, &status);35233524 if (status)3525 return error(_("path %s has been renamed/deleted"), old_name);3526 if (previous) {3527 st_mode = previous->new_mode;3528 } else if (!state->cached) {3529 stat_ret = lstat(old_name, st);3530 if (stat_ret && errno != ENOENT)3531 return error(_("%s: %s"), old_name, strerror(errno));3532 }35333534 if (state->check_index && !previous) {3535 int pos = cache_name_pos(old_name, strlen(old_name));3536 if (pos < 0) {3537 if (patch->is_new < 0)3538 goto is_new;3539 return error(_("%s: does not exist in index"), old_name);3540 }3541 *ce = active_cache[pos];3542 if (stat_ret < 0) {3543 if (checkout_target(&the_index, *ce, st))3544 return -1;3545 }3546 if (!state->cached && verify_index_match(*ce, st))3547 return error(_("%s: does not match index"), old_name);3548 if (state->cached)3549 st_mode = (*ce)->ce_mode;3550 } else if (stat_ret < 0) {3551 if (patch->is_new < 0)3552 goto is_new;3553 return error(_("%s: %s"), old_name, strerror(errno));3554 }35553556 if (!state->cached && !previous)3557 st_mode = ce_mode_from_stat(*ce, st->st_mode);35583559 if (patch->is_new < 0)3560 patch->is_new = 0;3561 if (!patch->old_mode)3562 patch->old_mode = st_mode;3563 if ((st_mode ^ patch->old_mode) & S_IFMT)3564 return error(_("%s: wrong type"), old_name);3565 if (st_mode != patch->old_mode)3566 warning(_("%s has type %o, expected %o"),3567 old_name, st_mode, patch->old_mode);3568 if (!patch->new_mode && !patch->is_delete)3569 patch->new_mode = st_mode;3570 return 0;35713572 is_new:3573 patch->is_new = 1;3574 patch->is_delete = 0;3575 free(patch->old_name);3576 patch->old_name = NULL;3577 return 0;3578}357935803581#define EXISTS_IN_INDEX 13582#define EXISTS_IN_WORKTREE 235833584static int check_to_create(struct apply_state *state,3585 const char *new_name,3586 int ok_if_exists)3587{3588 struct stat nst;35893590 if (state->check_index &&3591 cache_name_pos(new_name, strlen(new_name)) >= 0 &&3592 !ok_if_exists)3593 return EXISTS_IN_INDEX;3594 if (state->cached)3595 return 0;35963597 if (!lstat(new_name, &nst)) {3598 if (S_ISDIR(nst.st_mode) || ok_if_exists)3599 return 0;3600 /*3601 * A leading component of new_name might be a symlink3602 * that is going to be removed with this patch, but3603 * still pointing at somewhere that has the path.3604 * In such a case, path "new_name" does not exist as3605 * far as git is concerned.3606 */3607 if (has_symlink_leading_path(new_name, strlen(new_name)))3608 return 0;36093610 return EXISTS_IN_WORKTREE;3611 } else if ((errno != ENOENT) && (errno != ENOTDIR)) {3612 return error("%s: %s", new_name, strerror(errno));3613 }3614 return 0;3615}36163617static uintptr_t register_symlink_changes(struct apply_state *state,3618 const char *path,3619 uintptr_t what)3620{3621 struct string_list_item *ent;36223623 ent = string_list_lookup(&state->symlink_changes, path);3624 if (!ent) {3625 ent = string_list_insert(&state->symlink_changes, path);3626 ent->util = (void *)0;3627 }3628 ent->util = (void *)(what | ((uintptr_t)ent->util));3629 return (uintptr_t)ent->util;3630}36313632static uintptr_t check_symlink_changes(struct apply_state *state, const char *path)3633{3634 struct string_list_item *ent;36353636 ent = string_list_lookup(&state->symlink_changes, path);3637 if (!ent)3638 return 0;3639 return (uintptr_t)ent->util;3640}36413642static void prepare_symlink_changes(struct apply_state *state, struct patch *patch)3643{3644 for ( ; patch; patch = patch->next) {3645 if ((patch->old_name && S_ISLNK(patch->old_mode)) &&3646 (patch->is_rename || patch->is_delete))3647 /* the symlink at patch->old_name is removed */3648 register_symlink_changes(state, patch->old_name, APPLY_SYMLINK_GOES_AWAY);36493650 if (patch->new_name && S_ISLNK(patch->new_mode))3651 /* the symlink at patch->new_name is created or remains */3652 register_symlink_changes(state, patch->new_name, APPLY_SYMLINK_IN_RESULT);3653 }3654}36553656static int path_is_beyond_symlink_1(struct apply_state *state, struct strbuf *name)3657{3658 do {3659 unsigned int change;36603661 while (--name->len && name->buf[name->len] != '/')3662 ; /* scan backwards */3663 if (!name->len)3664 break;3665 name->buf[name->len] = '\0';3666 change = check_symlink_changes(state, name->buf);3667 if (change & APPLY_SYMLINK_IN_RESULT)3668 return 1;3669 if (change & APPLY_SYMLINK_GOES_AWAY)3670 /*3671 * This cannot be "return 0", because we may3672 * see a new one created at a higher level.3673 */3674 continue;36753676 /* otherwise, check the preimage */3677 if (state->check_index) {3678 struct cache_entry *ce;36793680 ce = cache_file_exists(name->buf, name->len, ignore_case);3681 if (ce && S_ISLNK(ce->ce_mode))3682 return 1;3683 } else {3684 struct stat st;3685 if (!lstat(name->buf, &st) && S_ISLNK(st.st_mode))3686 return 1;3687 }3688 } while (1);3689 return 0;3690}36913692static int path_is_beyond_symlink(struct apply_state *state, const char *name_)3693{3694 int ret;3695 struct strbuf name = STRBUF_INIT;36963697 assert(*name_ != '\0');3698 strbuf_addstr(&name, name_);3699 ret = path_is_beyond_symlink_1(state, &name);3700 strbuf_release(&name);37013702 return ret;3703}37043705static void die_on_unsafe_path(struct patch *patch)3706{3707 const char *old_name = NULL;3708 const char *new_name = NULL;3709 if (patch->is_delete)3710 old_name = patch->old_name;3711 else if (!patch->is_new && !patch->is_copy)3712 old_name = patch->old_name;3713 if (!patch->is_delete)3714 new_name = patch->new_name;37153716 if (old_name && !verify_path(old_name))3717 die(_("invalid path '%s'"), old_name);3718 if (new_name && !verify_path(new_name))3719 die(_("invalid path '%s'"), new_name);3720}37213722/*3723 * Check and apply the patch in-core; leave the result in patch->result3724 * for the caller to write it out to the final destination.3725 */3726static int check_patch(struct apply_state *state, struct patch *patch)3727{3728 struct stat st;3729 const char *old_name = patch->old_name;3730 const char *new_name = patch->new_name;3731 const char *name = old_name ? old_name : new_name;3732 struct cache_entry *ce = NULL;3733 struct patch *tpatch;3734 int ok_if_exists;3735 int status;37363737 patch->rejected = 1; /* we will drop this after we succeed */37383739 status = check_preimage(state, patch, &ce, &st);3740 if (status)3741 return status;3742 old_name = patch->old_name;37433744 /*3745 * A type-change diff is always split into a patch to delete3746 * old, immediately followed by a patch to create new (see3747 * diff.c::run_diff()); in such a case it is Ok that the entry3748 * to be deleted by the previous patch is still in the working3749 * tree and in the index.3750 *3751 * A patch to swap-rename between A and B would first rename A3752 * to B and then rename B to A. While applying the first one,3753 * the presence of B should not stop A from getting renamed to3754 * B; ask to_be_deleted() about the later rename. Removal of3755 * B and rename from A to B is handled the same way by asking3756 * was_deleted().3757 */3758 if ((tpatch = in_fn_table(state, new_name)) &&3759 (was_deleted(tpatch) || to_be_deleted(tpatch)))3760 ok_if_exists = 1;3761 else3762 ok_if_exists = 0;37633764 if (new_name &&3765 ((0 < patch->is_new) || patch->is_rename || patch->is_copy)) {3766 int err = check_to_create(state, new_name, ok_if_exists);37673768 if (err && state->threeway) {3769 patch->direct_to_threeway = 1;3770 } else switch (err) {3771 case 0:3772 break; /* happy */3773 case EXISTS_IN_INDEX:3774 return error(_("%s: already exists in index"), new_name);3775 break;3776 case EXISTS_IN_WORKTREE:3777 return error(_("%s: already exists in working directory"),3778 new_name);3779 default:3780 return err;3781 }37823783 if (!patch->new_mode) {3784 if (0 < patch->is_new)3785 patch->new_mode = S_IFREG | 0644;3786 else3787 patch->new_mode = patch->old_mode;3788 }3789 }37903791 if (new_name && old_name) {3792 int same = !strcmp(old_name, new_name);3793 if (!patch->new_mode)3794 patch->new_mode = patch->old_mode;3795 if ((patch->old_mode ^ patch->new_mode) & S_IFMT) {3796 if (same)3797 return error(_("new mode (%o) of %s does not "3798 "match old mode (%o)"),3799 patch->new_mode, new_name,3800 patch->old_mode);3801 else3802 return error(_("new mode (%o) of %s does not "3803 "match old mode (%o) of %s"),3804 patch->new_mode, new_name,3805 patch->old_mode, old_name);3806 }3807 }38083809 if (!state->unsafe_paths)3810 die_on_unsafe_path(patch);38113812 /*3813 * An attempt to read from or delete a path that is beyond a3814 * symbolic link will be prevented by load_patch_target() that3815 * is called at the beginning of apply_data() so we do not3816 * have to worry about a patch marked with "is_delete" bit3817 * here. We however need to make sure that the patch result3818 * is not deposited to a path that is beyond a symbolic link3819 * here.3820 */3821 if (!patch->is_delete && path_is_beyond_symlink(state, patch->new_name))3822 return error(_("affected file '%s' is beyond a symbolic link"),3823 patch->new_name);38243825 if (apply_data(state, patch, &st, ce) < 0)3826 return error(_("%s: patch does not apply"), name);3827 patch->rejected = 0;3828 return 0;3829}38303831static int check_patch_list(struct apply_state *state, struct patch *patch)3832{3833 int err = 0;38343835 prepare_symlink_changes(state, patch);3836 prepare_fn_table(state, patch);3837 while (patch) {3838 if (state->apply_verbosely)3839 say_patch_name(stderr,3840 _("Checking patch %s..."), patch);3841 err |= check_patch(state, patch);3842 patch = patch->next;3843 }3844 return err;3845}38463847/* This function tries to read the sha1 from the current index */3848static int get_current_sha1(const char *path, unsigned char *sha1)3849{3850 int pos;38513852 if (read_cache() < 0)3853 return -1;3854 pos = cache_name_pos(path, strlen(path));3855 if (pos < 0)3856 return -1;3857 hashcpy(sha1, active_cache[pos]->sha1);3858 return 0;3859}38603861static int preimage_sha1_in_gitlink_patch(struct patch *p, unsigned char sha1[20])3862{3863 /*3864 * A usable gitlink patch has only one fragment (hunk) that looks like:3865 * @@ -1 +1 @@3866 * -Subproject commit <old sha1>3867 * +Subproject commit <new sha1>3868 * or3869 * @@ -1 +0,0 @@3870 * -Subproject commit <old sha1>3871 * for a removal patch.3872 */3873 struct fragment *hunk = p->fragments;3874 static const char heading[] = "-Subproject commit ";3875 char *preimage;38763877 if (/* does the patch have only one hunk? */3878 hunk && !hunk->next &&3879 /* is its preimage one line? */3880 hunk->oldpos == 1 && hunk->oldlines == 1 &&3881 /* does preimage begin with the heading? */3882 (preimage = memchr(hunk->patch, '\n', hunk->size)) != NULL &&3883 starts_with(++preimage, heading) &&3884 /* does it record full SHA-1? */3885 !get_sha1_hex(preimage + sizeof(heading) - 1, sha1) &&3886 preimage[sizeof(heading) + 40 - 1] == '\n' &&3887 /* does the abbreviated name on the index line agree with it? */3888 starts_with(preimage + sizeof(heading) - 1, p->old_sha1_prefix))3889 return 0; /* it all looks fine */38903891 /* we may have full object name on the index line */3892 return get_sha1_hex(p->old_sha1_prefix, sha1);3893}38943895/* Build an index that contains the just the files needed for a 3way merge */3896static void build_fake_ancestor(struct patch *list, const char *filename)3897{3898 struct patch *patch;3899 struct index_state result = { NULL };3900 static struct lock_file lock;39013902 /* Once we start supporting the reverse patch, it may be3903 * worth showing the new sha1 prefix, but until then...3904 */3905 for (patch = list; patch; patch = patch->next) {3906 unsigned char sha1[20];3907 struct cache_entry *ce;3908 const char *name;39093910 name = patch->old_name ? patch->old_name : patch->new_name;3911 if (0 < patch->is_new)3912 continue;39133914 if (S_ISGITLINK(patch->old_mode)) {3915 if (!preimage_sha1_in_gitlink_patch(patch, sha1))3916 ; /* ok, the textual part looks sane */3917 else3918 die("sha1 information is lacking or useless for submodule %s",3919 name);3920 } else if (!get_sha1_blob(patch->old_sha1_prefix, sha1)) {3921 ; /* ok */3922 } else if (!patch->lines_added && !patch->lines_deleted) {3923 /* mode-only change: update the current */3924 if (get_current_sha1(patch->old_name, sha1))3925 die("mode change for %s, which is not "3926 "in current HEAD", name);3927 } else3928 die("sha1 information is lacking or useless "3929 "(%s).", name);39303931 ce = make_cache_entry(patch->old_mode, sha1, name, 0, 0);3932 if (!ce)3933 die(_("make_cache_entry failed for path '%s'"), name);3934 if (add_index_entry(&result, ce, ADD_CACHE_OK_TO_ADD))3935 die ("Could not add %s to temporary index", name);3936 }39373938 hold_lock_file_for_update(&lock, filename, LOCK_DIE_ON_ERROR);3939 if (write_locked_index(&result, &lock, COMMIT_LOCK))3940 die ("Could not write temporary index to %s", filename);39413942 discard_index(&result);3943}39443945static void stat_patch_list(struct apply_state *state, struct patch *patch)3946{3947 int files, adds, dels;39483949 for (files = adds = dels = 0 ; patch ; patch = patch->next) {3950 files++;3951 adds += patch->lines_added;3952 dels += patch->lines_deleted;3953 show_stats(state, patch);3954 }39553956 print_stat_summary(stdout, files, adds, dels);3957}39583959static void numstat_patch_list(struct apply_state *state,3960 struct patch *patch)3961{3962 for ( ; patch; patch = patch->next) {3963 const char *name;3964 name = patch->new_name ? patch->new_name : patch->old_name;3965 if (patch->is_binary)3966 printf("-\t-\t");3967 else3968 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);3969 write_name_quoted(name, stdout, state->line_termination);3970 }3971}39723973static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)3974{3975 if (mode)3976 printf(" %s mode %06o %s\n", newdelete, mode, name);3977 else3978 printf(" %s %s\n", newdelete, name);3979}39803981static void show_mode_change(struct patch *p, int show_name)3982{3983 if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {3984 if (show_name)3985 printf(" mode change %06o => %06o %s\n",3986 p->old_mode, p->new_mode, p->new_name);3987 else3988 printf(" mode change %06o => %06o\n",3989 p->old_mode, p->new_mode);3990 }3991}39923993static void show_rename_copy(struct patch *p)3994{3995 const char *renamecopy = p->is_rename ? "rename" : "copy";3996 const char *old, *new;39973998 /* Find common prefix */3999 old = p->old_name;4000 new = p->new_name;4001 while (1) {4002 const char *slash_old, *slash_new;4003 slash_old = strchr(old, '/');4004 slash_new = strchr(new, '/');4005 if (!slash_old ||4006 !slash_new ||4007 slash_old - old != slash_new - new ||4008 memcmp(old, new, slash_new - new))4009 break;4010 old = slash_old + 1;4011 new = slash_new + 1;4012 }4013 /* p->old_name thru old is the common prefix, and old and new4014 * through the end of names are renames4015 */4016 if (old != p->old_name)4017 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,4018 (int)(old - p->old_name), p->old_name,4019 old, new, p->score);4020 else4021 printf(" %s %s => %s (%d%%)\n", renamecopy,4022 p->old_name, p->new_name, p->score);4023 show_mode_change(p, 0);4024}40254026static void summary_patch_list(struct patch *patch)4027{4028 struct patch *p;40294030 for (p = patch; p; p = p->next) {4031 if (p->is_new)4032 show_file_mode_name("create", p->new_mode, p->new_name);4033 else if (p->is_delete)4034 show_file_mode_name("delete", p->old_mode, p->old_name);4035 else {4036 if (p->is_rename || p->is_copy)4037 show_rename_copy(p);4038 else {4039 if (p->score) {4040 printf(" rewrite %s (%d%%)\n",4041 p->new_name, p->score);4042 show_mode_change(p, 0);4043 }4044 else4045 show_mode_change(p, 1);4046 }4047 }4048 }4049}40504051static void patch_stats(struct apply_state *state, struct patch *patch)4052{4053 int lines = patch->lines_added + patch->lines_deleted;40544055 if (lines > state->max_change)4056 state->max_change = lines;4057 if (patch->old_name) {4058 int len = quote_c_style(patch->old_name, NULL, NULL, 0);4059 if (!len)4060 len = strlen(patch->old_name);4061 if (len > state->max_len)4062 state->max_len = len;4063 }4064 if (patch->new_name) {4065 int len = quote_c_style(patch->new_name, NULL, NULL, 0);4066 if (!len)4067 len = strlen(patch->new_name);4068 if (len > state->max_len)4069 state->max_len = len;4070 }4071}40724073static void remove_file(struct apply_state *state, struct patch *patch, int rmdir_empty)4074{4075 if (state->update_index) {4076 if (remove_file_from_cache(patch->old_name) < 0)4077 die(_("unable to remove %s from index"), patch->old_name);4078 }4079 if (!state->cached) {4080 if (!remove_or_warn(patch->old_mode, patch->old_name) && rmdir_empty) {4081 remove_path(patch->old_name);4082 }4083 }4084}40854086static void add_index_file(struct apply_state *state,4087 const char *path,4088 unsigned mode,4089 void *buf,4090 unsigned long size)4091{4092 struct stat st;4093 struct cache_entry *ce;4094 int namelen = strlen(path);4095 unsigned ce_size = cache_entry_size(namelen);40964097 if (!state->update_index)4098 return;40994100 ce = xcalloc(1, ce_size);4101 memcpy(ce->name, path, namelen);4102 ce->ce_mode = create_ce_mode(mode);4103 ce->ce_flags = create_ce_flags(0);4104 ce->ce_namelen = namelen;4105 if (S_ISGITLINK(mode)) {4106 const char *s;41074108 if (!skip_prefix(buf, "Subproject commit ", &s) ||4109 get_sha1_hex(s, ce->sha1))4110 die(_("corrupt patch for submodule %s"), path);4111 } else {4112 if (!state->cached) {4113 if (lstat(path, &st) < 0)4114 die_errno(_("unable to stat newly created file '%s'"),4115 path);4116 fill_stat_cache_info(ce, &st);4117 }4118 if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)4119 die(_("unable to create backing store for newly created file %s"), path);4120 }4121 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4122 die(_("unable to add cache entry for %s"), path);4123}41244125static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)4126{4127 int fd;4128 struct strbuf nbuf = STRBUF_INIT;41294130 if (S_ISGITLINK(mode)) {4131 struct stat st;4132 if (!lstat(path, &st) && S_ISDIR(st.st_mode))4133 return 0;4134 return mkdir(path, 0777);4135 }41364137 if (has_symlinks && S_ISLNK(mode))4138 /* Although buf:size is counted string, it also is NUL4139 * terminated.4140 */4141 return symlink(buf, path);41424143 fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);4144 if (fd < 0)4145 return -1;41464147 if (convert_to_working_tree(path, buf, size, &nbuf)) {4148 size = nbuf.len;4149 buf = nbuf.buf;4150 }4151 write_or_die(fd, buf, size);4152 strbuf_release(&nbuf);41534154 if (close(fd) < 0)4155 die_errno(_("closing file '%s'"), path);4156 return 0;4157}41584159/*4160 * We optimistically assume that the directories exist,4161 * which is true 99% of the time anyway. If they don't,4162 * we create them and try again.4163 */4164static void create_one_file(struct apply_state *state,4165 char *path,4166 unsigned mode,4167 const char *buf,4168 unsigned long size)4169{4170 if (state->cached)4171 return;4172 if (!try_create_file(path, mode, buf, size))4173 return;41744175 if (errno == ENOENT) {4176 if (safe_create_leading_directories(path))4177 return;4178 if (!try_create_file(path, mode, buf, size))4179 return;4180 }41814182 if (errno == EEXIST || errno == EACCES) {4183 /* We may be trying to create a file where a directory4184 * used to be.4185 */4186 struct stat st;4187 if (!lstat(path, &st) && (!S_ISDIR(st.st_mode) || !rmdir(path)))4188 errno = EEXIST;4189 }41904191 if (errno == EEXIST) {4192 unsigned int nr = getpid();41934194 for (;;) {4195 char newpath[PATH_MAX];4196 mksnpath(newpath, sizeof(newpath), "%s~%u", path, nr);4197 if (!try_create_file(newpath, mode, buf, size)) {4198 if (!rename(newpath, path))4199 return;4200 unlink_or_warn(newpath);4201 break;4202 }4203 if (errno != EEXIST)4204 break;4205 ++nr;4206 }4207 }4208 die_errno(_("unable to write file '%s' mode %o"), path, mode);4209}42104211static void add_conflicted_stages_file(struct apply_state *state,4212 struct patch *patch)4213{4214 int stage, namelen;4215 unsigned ce_size, mode;4216 struct cache_entry *ce;42174218 if (!state->update_index)4219 return;4220 namelen = strlen(patch->new_name);4221 ce_size = cache_entry_size(namelen);4222 mode = patch->new_mode ? patch->new_mode : (S_IFREG | 0644);42234224 remove_file_from_cache(patch->new_name);4225 for (stage = 1; stage < 4; stage++) {4226 if (is_null_oid(&patch->threeway_stage[stage - 1]))4227 continue;4228 ce = xcalloc(1, ce_size);4229 memcpy(ce->name, patch->new_name, namelen);4230 ce->ce_mode = create_ce_mode(mode);4231 ce->ce_flags = create_ce_flags(stage);4232 ce->ce_namelen = namelen;4233 hashcpy(ce->sha1, patch->threeway_stage[stage - 1].hash);4234 if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)4235 die(_("unable to add cache entry for %s"), patch->new_name);4236 }4237}42384239static void create_file(struct apply_state *state, struct patch *patch)4240{4241 char *path = patch->new_name;4242 unsigned mode = patch->new_mode;4243 unsigned long size = patch->resultsize;4244 char *buf = patch->result;42454246 if (!mode)4247 mode = S_IFREG | 0644;4248 create_one_file(state, path, mode, buf, size);42494250 if (patch->conflicted_threeway)4251 add_conflicted_stages_file(state, patch);4252 else4253 add_index_file(state, path, mode, buf, size);4254}42554256/* phase zero is to remove, phase one is to create */4257static void write_out_one_result(struct apply_state *state,4258 struct patch *patch,4259 int phase)4260{4261 if (patch->is_delete > 0) {4262 if (phase == 0)4263 remove_file(state, patch, 1);4264 return;4265 }4266 if (patch->is_new > 0 || patch->is_copy) {4267 if (phase == 1)4268 create_file(state, patch);4269 return;4270 }4271 /*4272 * Rename or modification boils down to the same4273 * thing: remove the old, write the new4274 */4275 if (phase == 0)4276 remove_file(state, patch, patch->is_rename);4277 if (phase == 1)4278 create_file(state, patch);4279}42804281static int write_out_one_reject(struct apply_state *state, struct patch *patch)4282{4283 FILE *rej;4284 char namebuf[PATH_MAX];4285 struct fragment *frag;4286 int cnt = 0;4287 struct strbuf sb = STRBUF_INIT;42884289 for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {4290 if (!frag->rejected)4291 continue;4292 cnt++;4293 }42944295 if (!cnt) {4296 if (state->apply_verbosely)4297 say_patch_name(stderr,4298 _("Applied patch %s cleanly."), patch);4299 return 0;4300 }43014302 /* This should not happen, because a removal patch that leaves4303 * contents are marked "rejected" at the patch level.4304 */4305 if (!patch->new_name)4306 die(_("internal error"));43074308 /* Say this even without --verbose */4309 strbuf_addf(&sb, Q_("Applying patch %%s with %d reject...",4310 "Applying patch %%s with %d rejects...",4311 cnt),4312 cnt);4313 say_patch_name(stderr, sb.buf, patch);4314 strbuf_release(&sb);43154316 cnt = strlen(patch->new_name);4317 if (ARRAY_SIZE(namebuf) <= cnt + 5) {4318 cnt = ARRAY_SIZE(namebuf) - 5;4319 warning(_("truncating .rej filename to %.*s.rej"),4320 cnt - 1, patch->new_name);4321 }4322 memcpy(namebuf, patch->new_name, cnt);4323 memcpy(namebuf + cnt, ".rej", 5);43244325 rej = fopen(namebuf, "w");4326 if (!rej)4327 return error(_("cannot open %s: %s"), namebuf, strerror(errno));43284329 /* Normal git tools never deal with .rej, so do not pretend4330 * this is a git patch by saying --git or giving extended4331 * headers. While at it, maybe please "kompare" that wants4332 * the trailing TAB and some garbage at the end of line ;-).4333 */4334 fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",4335 patch->new_name, patch->new_name);4336 for (cnt = 1, frag = patch->fragments;4337 frag;4338 cnt++, frag = frag->next) {4339 if (!frag->rejected) {4340 fprintf_ln(stderr, _("Hunk #%d applied cleanly."), cnt);4341 continue;4342 }4343 fprintf_ln(stderr, _("Rejected hunk #%d."), cnt);4344 fprintf(rej, "%.*s", frag->size, frag->patch);4345 if (frag->patch[frag->size-1] != '\n')4346 fputc('\n', rej);4347 }4348 fclose(rej);4349 return -1;4350}43514352static int write_out_results(struct apply_state *state, struct patch *list)4353{4354 int phase;4355 int errs = 0;4356 struct patch *l;4357 struct string_list cpath = STRING_LIST_INIT_DUP;43584359 for (phase = 0; phase < 2; phase++) {4360 l = list;4361 while (l) {4362 if (l->rejected)4363 errs = 1;4364 else {4365 write_out_one_result(state, l, phase);4366 if (phase == 1) {4367 if (write_out_one_reject(state, l))4368 errs = 1;4369 if (l->conflicted_threeway) {4370 string_list_append(&cpath, l->new_name);4371 errs = 1;4372 }4373 }4374 }4375 l = l->next;4376 }4377 }43784379 if (cpath.nr) {4380 struct string_list_item *item;43814382 string_list_sort(&cpath);4383 for_each_string_list_item(item, &cpath)4384 fprintf(stderr, "U %s\n", item->string);4385 string_list_clear(&cpath, 0);43864387 rerere(0);4388 }43894390 return errs;4391}43924393static struct lock_file lock_file;43944395#define INACCURATE_EOF (1<<0)4396#define RECOUNT (1<<1)43974398/*4399 * Try to apply a patch.4400 *4401 * Returns:4402 * -128 if a bad error happened (like patch unreadable)4403 * -1 if patch did not apply and user cannot deal with it4404 * 0 if the patch applied4405 * 1 if the patch did not apply but user might fix it4406 */4407static int apply_patch(struct apply_state *state,4408 int fd,4409 const char *filename,4410 int options)4411{4412 size_t offset;4413 struct strbuf buf = STRBUF_INIT; /* owns the patch text */4414 struct patch *list = NULL, **listp = &list;4415 int skipped_patch = 0;4416 int res = 0;44174418 state->patch_input_file = filename;4419 if (read_patch_file(&buf, fd) < 0)4420 return -128;4421 offset = 0;4422 while (offset < buf.len) {4423 struct patch *patch;4424 int nr;44254426 patch = xcalloc(1, sizeof(*patch));4427 patch->inaccurate_eof = !!(options & INACCURATE_EOF);4428 patch->recount = !!(options & RECOUNT);4429 nr = parse_chunk(state, buf.buf + offset, buf.len - offset, patch);4430 if (nr < 0) {4431 free_patch(patch);4432 if (nr == -128) {4433 res = -128;4434 goto end;4435 }4436 break;4437 }4438 if (state->apply_in_reverse)4439 reverse_patches(patch);4440 if (use_patch(state, patch)) {4441 patch_stats(state, patch);4442 *listp = patch;4443 listp = &patch->next;4444 }4445 else {4446 if (state->apply_verbosely)4447 say_patch_name(stderr, _("Skipped patch '%s'."), patch);4448 free_patch(patch);4449 skipped_patch++;4450 }4451 offset += nr;4452 }44534454 if (!list && !skipped_patch) {4455 error(_("unrecognized input"));4456 res = -128;4457 goto end;4458 }44594460 if (state->whitespace_error && (state->ws_error_action == die_on_ws_error))4461 state->apply = 0;44624463 state->update_index = state->check_index && state->apply;4464 if (state->update_index && state->newfd < 0)4465 state->newfd = hold_locked_index(state->lock_file, 1);44664467 if (state->check_index && read_cache() < 0) {4468 error(_("unable to read index file"));4469 res = -128;4470 goto end;4471 }44724473 if ((state->check || state->apply) &&4474 check_patch_list(state, list) < 0 &&4475 !state->apply_with_reject) {4476 res = -1;4477 goto end;4478 }44794480 if (state->apply && write_out_results(state, list)) {4481 /* with --3way, we still need to write the index out */4482 res = state->apply_with_reject ? -1 : 1;4483 goto end;4484 }44854486 if (state->fake_ancestor)4487 build_fake_ancestor(list, state->fake_ancestor);44884489 if (state->diffstat)4490 stat_patch_list(state, list);44914492 if (state->numstat)4493 numstat_patch_list(state, list);44944495 if (state->summary)4496 summary_patch_list(list);44974498end:4499 free_patch_list(list);4500 strbuf_release(&buf);4501 string_list_clear(&state->fn_table, 0);4502 return res;4503}45044505static int option_parse_exclude(const struct option *opt,4506 const char *arg, int unset)4507{4508 struct apply_state *state = opt->value;4509 add_name_limit(state, arg, 1);4510 return 0;4511}45124513static int option_parse_include(const struct option *opt,4514 const char *arg, int unset)4515{4516 struct apply_state *state = opt->value;4517 add_name_limit(state, arg, 0);4518 state->has_include = 1;4519 return 0;4520}45214522static int option_parse_p(const struct option *opt,4523 const char *arg,4524 int unset)4525{4526 struct apply_state *state = opt->value;4527 state->p_value = atoi(arg);4528 state->p_value_known = 1;4529 return 0;4530}45314532static int option_parse_space_change(const struct option *opt,4533 const char *arg, int unset)4534{4535 struct apply_state *state = opt->value;4536 if (unset)4537 state->ws_ignore_action = ignore_ws_none;4538 else4539 state->ws_ignore_action = ignore_ws_change;4540 return 0;4541}45424543static int option_parse_whitespace(const struct option *opt,4544 const char *arg, int unset)4545{4546 struct apply_state *state = opt->value;4547 state->whitespace_option = arg;4548 if (parse_whitespace_option(state, arg))4549 exit(1);4550 return 0;4551}45524553static int option_parse_directory(const struct option *opt,4554 const char *arg, int unset)4555{4556 struct apply_state *state = opt->value;4557 strbuf_reset(&state->root);4558 strbuf_addstr(&state->root, arg);4559 strbuf_complete(&state->root, '/');4560 return 0;4561}45624563static int apply_all_patches(struct apply_state *state,4564 int argc,4565 const char **argv,4566 int options)4567{4568 int i;4569 int res;4570 int errs = 0;4571 int read_stdin = 1;45724573 for (i = 0; i < argc; i++) {4574 const char *arg = argv[i];4575 int fd;45764577 if (!strcmp(arg, "-")) {4578 res = apply_patch(state, 0, "<stdin>", options);4579 if (res < 0)4580 goto end;4581 errs |= res;4582 read_stdin = 0;4583 continue;4584 } else if (0 < state->prefix_length)4585 arg = prefix_filename(state->prefix,4586 state->prefix_length,4587 arg);45884589 fd = open(arg, O_RDONLY);4590 if (fd < 0) {4591 error(_("can't open patch '%s': %s"), arg, strerror(errno));4592 res = -128;4593 goto end;4594 }4595 read_stdin = 0;4596 set_default_whitespace_mode(state);4597 res = apply_patch(state, fd, arg, options);4598 close(fd);4599 if (res < 0)4600 goto end;4601 errs |= res;4602 }4603 set_default_whitespace_mode(state);4604 if (read_stdin) {4605 res = apply_patch(state, 0, "<stdin>", options);4606 if (res < 0)4607 goto end;4608 errs |= res;4609 }46104611 if (state->whitespace_error) {4612 if (state->squelch_whitespace_errors &&4613 state->squelch_whitespace_errors < state->whitespace_error) {4614 int squelched =4615 state->whitespace_error - state->squelch_whitespace_errors;4616 warning(Q_("squelched %d whitespace error",4617 "squelched %d whitespace errors",4618 squelched),4619 squelched);4620 }4621 if (state->ws_error_action == die_on_ws_error) {4622 error(Q_("%d line adds whitespace errors.",4623 "%d lines add whitespace errors.",4624 state->whitespace_error),4625 state->whitespace_error);4626 res = -128;4627 goto end;4628 }4629 if (state->applied_after_fixing_ws && state->apply)4630 warning("%d line%s applied after"4631 " fixing whitespace errors.",4632 state->applied_after_fixing_ws,4633 state->applied_after_fixing_ws == 1 ? "" : "s");4634 else if (state->whitespace_error)4635 warning(Q_("%d line adds whitespace errors.",4636 "%d lines add whitespace errors.",4637 state->whitespace_error),4638 state->whitespace_error);4639 }46404641 if (state->update_index) {4642 res = write_locked_index(&the_index, state->lock_file, COMMIT_LOCK);4643 if (res) {4644 error(_("Unable to write new index file"));4645 res = -128;4646 goto end;4647 }4648 state->newfd = -1;4649 }46504651 return !!errs;46524653end:4654 if (state->newfd >= 0) {4655 rollback_lock_file(state->lock_file);4656 state->newfd = -1;4657 }46584659 return (res == -1 ? 1 : 128);4660}46614662int cmd_apply(int argc, const char **argv, const char *prefix)4663{4664 int force_apply = 0;4665 int options = 0;4666 int ret;4667 struct apply_state state;46684669 struct option builtin_apply_options[] = {4670 { OPTION_CALLBACK, 0, "exclude", &state, N_("path"),4671 N_("don't apply changes matching the given path"),4672 0, option_parse_exclude },4673 { OPTION_CALLBACK, 0, "include", &state, N_("path"),4674 N_("apply changes matching the given path"),4675 0, option_parse_include },4676 { OPTION_CALLBACK, 'p', NULL, &state, N_("num"),4677 N_("remove <num> leading slashes from traditional diff paths"),4678 0, option_parse_p },4679 OPT_BOOL(0, "no-add", &state.no_add,4680 N_("ignore additions made by the patch")),4681 OPT_BOOL(0, "stat", &state.diffstat,4682 N_("instead of applying the patch, output diffstat for the input")),4683 OPT_NOOP_NOARG(0, "allow-binary-replacement"),4684 OPT_NOOP_NOARG(0, "binary"),4685 OPT_BOOL(0, "numstat", &state.numstat,4686 N_("show number of added and deleted lines in decimal notation")),4687 OPT_BOOL(0, "summary", &state.summary,4688 N_("instead of applying the patch, output a summary for the input")),4689 OPT_BOOL(0, "check", &state.check,4690 N_("instead of applying the patch, see if the patch is applicable")),4691 OPT_BOOL(0, "index", &state.check_index,4692 N_("make sure the patch is applicable to the current index")),4693 OPT_BOOL(0, "cached", &state.cached,4694 N_("apply a patch without touching the working tree")),4695 OPT_BOOL(0, "unsafe-paths", &state.unsafe_paths,4696 N_("accept a patch that touches outside the working area")),4697 OPT_BOOL(0, "apply", &force_apply,4698 N_("also apply the patch (use with --stat/--summary/--check)")),4699 OPT_BOOL('3', "3way", &state.threeway,4700 N_( "attempt three-way merge if a patch does not apply")),4701 OPT_FILENAME(0, "build-fake-ancestor", &state.fake_ancestor,4702 N_("build a temporary index based on embedded index information")),4703 /* Think twice before adding "--nul" synonym to this */4704 OPT_SET_INT('z', NULL, &state.line_termination,4705 N_("paths are separated with NUL character"), '\0'),4706 OPT_INTEGER('C', NULL, &state.p_context,4707 N_("ensure at least <n> lines of context match")),4708 { OPTION_CALLBACK, 0, "whitespace", &state, N_("action"),4709 N_("detect new or modified lines that have whitespace errors"),4710 0, option_parse_whitespace },4711 { OPTION_CALLBACK, 0, "ignore-space-change", &state, NULL,4712 N_("ignore changes in whitespace when finding context"),4713 PARSE_OPT_NOARG, option_parse_space_change },4714 { OPTION_CALLBACK, 0, "ignore-whitespace", &state, NULL,4715 N_("ignore changes in whitespace when finding context"),4716 PARSE_OPT_NOARG, option_parse_space_change },4717 OPT_BOOL('R', "reverse", &state.apply_in_reverse,4718 N_("apply the patch in reverse")),4719 OPT_BOOL(0, "unidiff-zero", &state.unidiff_zero,4720 N_("don't expect at least one line of context")),4721 OPT_BOOL(0, "reject", &state.apply_with_reject,4722 N_("leave the rejected hunks in corresponding *.rej files")),4723 OPT_BOOL(0, "allow-overlap", &state.allow_overlap,4724 N_("allow overlapping hunks")),4725 OPT__VERBOSE(&state.apply_verbosely, N_("be verbose")),4726 OPT_BIT(0, "inaccurate-eof", &options,4727 N_("tolerate incorrectly detected missing new-line at the end of file"),4728 INACCURATE_EOF),4729 OPT_BIT(0, "recount", &options,4730 N_("do not trust the line counts in the hunk headers"),4731 RECOUNT),4732 { OPTION_CALLBACK, 0, "directory", &state, N_("root"),4733 N_("prepend <root> to all filenames"),4734 0, option_parse_directory },4735 OPT_END()4736 };47374738 if (init_apply_state(&state, prefix, &lock_file))4739 exit(128);47404741 argc = parse_options(argc, argv, state.prefix, builtin_apply_options,4742 apply_usage, 0);47434744 if (check_apply_state(&state, force_apply))4745 exit(128);47464747 ret = apply_all_patches(&state, argc, argv, options);47484749 clear_apply_state(&state);47504751 return ret;4752}