builtin-apply.c

   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 <fnmatch.h>
  10 #include "cache.h"
  11 #include "cache-tree.h"
  12 #include "quote.h"
  13 #include "blob.h"
  14 #include "delta.h"
  15 #include "builtin.h"
  16
  17 /*
  18  *  --check turns on checking that the working tree matches the
  19  *    files that are being modified, but doesn't apply the patch
  20  *  --stat does just a diffstat, and doesn't actually apply
  21  *  --numstat does numeric diffstat, and doesn't actually apply
  22  *  --index-info shows the old and new index info for paths if available.
  23  *  --index updates the cache as well.
  24  *  --cached updates only the cache without ever touching the working tree.
  25  */
  26 static const char *prefix;
  27 static int prefix_length = -1;
  28 static int newfd = -1;
  29
  30 static int unidiff_zero;
  31 static int p_value = 1;
  32 static int check_index;
  33 static int write_index;
  34 static int cached;
  35 static int diffstat;
  36 static int numstat;
  37 static int summary;
  38 static int check;
  39 static int apply = 1;
  40 static int apply_in_reverse;
  41 static int apply_with_reject;
  42 static int apply_verbosely;
  43 static int no_add;
  44 static int show_index_info;
  45 static int line_termination = '\n';
  46 static unsigned long p_context = ULONG_MAX;
  47 static const char apply_usage[] =
  48 "git-apply [--stat] [--numstat] [--summary] [--check] [--index] [--cached] [--apply] [--no-add] [--index-info] [--allow-binary-replacement] [--reverse] [--reject] [--verbose] [-z] [-pNUM] [-CNUM] [--whitespace=<nowarn|warn|error|error-all|strip>] <patch>...";
  49
  50 static enum whitespace_eol {
  51         nowarn_whitespace,
  52         warn_on_whitespace,
  53         error_on_whitespace,
  54         strip_whitespace,
  55 } new_whitespace = warn_on_whitespace;
  56 static int whitespace_error;
  57 static int squelch_whitespace_errors = 5;
  58 static int applied_after_stripping;
  59 static const char *patch_input_file;
  60
  61 static void parse_whitespace_option(const char *option)
  62 {
  63         if (!option) {
  64                 new_whitespace = warn_on_whitespace;
  65                 return;
  66         }
  67         if (!strcmp(option, "warn")) {
  68                 new_whitespace = warn_on_whitespace;
  69                 return;
  70         }
  71         if (!strcmp(option, "nowarn")) {
  72                 new_whitespace = nowarn_whitespace;
  73                 return;
  74         }
  75         if (!strcmp(option, "error")) {
  76                 new_whitespace = error_on_whitespace;
  77                 return;
  78         }
  79         if (!strcmp(option, "error-all")) {
  80                 new_whitespace = error_on_whitespace;
  81                 squelch_whitespace_errors = 0;
  82                 return;
  83         }
  84         if (!strcmp(option, "strip")) {
  85                 new_whitespace = strip_whitespace;
  86                 return;
  87         }
  88         die("unrecognized whitespace option '%s'", option);
  89 }
  90
  91 static void set_default_whitespace_mode(const char *whitespace_option)
  92 {
  93         if (!whitespace_option && !apply_default_whitespace) {
  94                 new_whitespace = (apply
  95                                   ? warn_on_whitespace
  96                                   : nowarn_whitespace);
  97         }
  98 }
  99
 100 /*
 101  * For "diff-stat" like behaviour, we keep track of the biggest change
 102  * we've seen, and the longest filename. That allows us to do simple
 103  * scaling.
 104  */
 105 static int max_change, max_len;
 106
 107 /*
 108  * Various "current state", notably line numbers and what
 109  * file (and how) we're patching right now.. The "is_xxxx"
 110  * things are flags, where -1 means "don't know yet".
 111  */
 112 static int linenr = 1;
 113
 114 /*
 115  * This represents one "hunk" from a patch, starting with
 116  * "@@ -oldpos,oldlines +newpos,newlines @@" marker.  The
 117  * patch text is pointed at by patch, and its byte length
 118  * is stored in size.  leading and trailing are the number
 119  * of context lines.
 120  */
 121 struct fragment {
 122         unsigned long leading, trailing;
 123         unsigned long oldpos, oldlines;
 124         unsigned long newpos, newlines;
 125         const char *patch;
 126         int size;
 127         int rejected;
 128         struct fragment *next;
 129 };
 130
 131 /*
 132  * When dealing with a binary patch, we reuse "leading" field
 133  * to store the type of the binary hunk, either deflated "delta"
 134  * or deflated "literal".
 135  */
 136 #define binary_patch_method leading
 137 #define BINARY_DELTA_DEFLATED   1
 138 #define BINARY_LITERAL_DEFLATED 2
 139
 140 struct patch {
 141         char *new_name, *old_name, *def_name;
 142         unsigned int old_mode, new_mode;
 143         int is_rename, is_copy, is_new, is_delete, is_binary;
 144         int rejected;
 145         unsigned long deflate_origlen;
 146         int lines_added, lines_deleted;
 147         int score;
 148         int inaccurate_eof:1;
 149         struct fragment *fragments;
 150         char *result;
 151         unsigned long resultsize;
 152         char old_sha1_prefix[41];
 153         char new_sha1_prefix[41];
 154         struct patch *next;
 155 };
 156
 157 static void say_patch_name(FILE *output, const char *pre, struct patch *patch, const char *post)
 158 {
 159         fputs(pre, output);
 160         if (patch->old_name && patch->new_name &&
 161             strcmp(patch->old_name, patch->new_name)) {
 162                 write_name_quoted(NULL, 0, patch->old_name, 1, output);
 163                 fputs(" => ", output);
 164                 write_name_quoted(NULL, 0, patch->new_name, 1, output);
 165         }
 166         else {
 167                 const char *n = patch->new_name;
 168                 if (!n)
 169                         n = patch->old_name;
 170                 write_name_quoted(NULL, 0, n, 1, output);
 171         }
 172         fputs(post, output);
 173 }
 174
 175 #define CHUNKSIZE (8192)
 176 #define SLOP (16)
 177
 178 static void *read_patch_file(int fd, unsigned long *sizep)
 179 {
 180         unsigned long size = 0, alloc = CHUNKSIZE;
 181         void *buffer = xmalloc(alloc);
 182
 183         for (;;) {
 184                 int nr = alloc - size;
 185                 if (nr < 1024) {
 186                         alloc += CHUNKSIZE;
 187                         buffer = xrealloc(buffer, alloc);
 188                         nr = alloc - size;
 189                 }
 190                 nr = xread(fd, (char *) buffer + size, nr);
 191                 if (!nr)
 192                         break;
 193                 if (nr < 0)
 194                         die("git-apply: read returned %s", strerror(errno));
 195                 size += nr;
 196         }
 197         *sizep = size;
 198
 199         /*
 200          * Make sure that we have some slop in the buffer
 201          * so that we can do speculative "memcmp" etc, and
 202          * see to it that it is NUL-filled.
 203          */
 204         if (alloc < size + SLOP)
 205                 buffer = xrealloc(buffer, size + SLOP);
 206         memset((char *) buffer + size, 0, SLOP);
 207         return buffer;
 208 }
 209
 210 static unsigned long linelen(const char *buffer, unsigned long size)
 211 {
 212         unsigned long len = 0;
 213         while (size--) {
 214                 len++;
 215                 if (*buffer++ == '\n')
 216                         break;
 217         }
 218         return len;
 219 }
 220
 221 static int is_dev_null(const char *str)
 222 {
 223         return !memcmp("/dev/null", str, 9) && isspace(str[9]);
 224 }
 225
 226 #define TERM_SPACE      1
 227 #define TERM_TAB        2
 228
 229 static int name_terminate(const char *name, int namelen, int c, int terminate)
 230 {
 231         if (c == ' ' && !(terminate & TERM_SPACE))
 232                 return 0;
 233         if (c == '\t' && !(terminate & TERM_TAB))
 234                 return 0;
 235
 236         return 1;
 237 }
 238
 239 static char * find_name(const char *line, char *def, int p_value, int terminate)
 240 {
 241         int len;
 242         const char *start = line;
 243         char *name;
 244
 245         if (*line == '"') {
 246                 /* Proposed "new-style" GNU patch/diff format; see
 247                  * http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2
 248                  */
 249                 name = unquote_c_style(line, NULL);
 250                 if (name) {
 251                         char *cp = name;
 252                         while (p_value) {
 253                                 cp = strchr(name, '/');
 254                                 if (!cp)
 255                                         break;
 256                                 cp++;
 257                                 p_value--;
 258                         }
 259                         if (cp) {
 260                                 /* name can later be freed, so we need
 261                                  * to memmove, not just return cp
 262                                  */
 263                                 memmove(name, cp, strlen(cp) + 1);
 264                                 free(def);
 265                                 return name;
 266                         }
 267                         else {
 268                                 free(name);
 269                                 name = NULL;
 270                         }
 271                 }
 272         }
 273
 274         for (;;) {
 275                 char c = *line;
 276
 277                 if (isspace(c)) {
 278                         if (c == '\n')
 279                                 break;
 280                         if (name_terminate(start, line-start, c, terminate))
 281                                 break;
 282                 }
 283                 line++;
 284                 if (c == '/' && !--p_value)
 285                         start = line;
 286         }
 287         if (!start)
 288                 return def;
 289         len = line - start;
 290         if (!len)
 291                 return def;
 292
 293         /*
 294          * Generally we prefer the shorter name, especially
 295          * if the other one is just a variation of that with
 296          * something else tacked on to the end (ie "file.orig"
 297          * or "file~").
 298          */
 299         if (def) {
 300                 int deflen = strlen(def);
 301                 if (deflen < len && !strncmp(start, def, deflen))
 302                         return def;
 303         }
 304
 305         name = xmalloc(len + 1);
 306         memcpy(name, start, len);
 307         name[len] = 0;
 308         free(def);
 309         return name;
 310 }
 311
 312 /*
 313  * Get the name etc info from the --/+++ lines of a traditional patch header
 314  *
 315  * NOTE! This hardcodes "-p1" behaviour in filename detection.
 316  *
 317  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 318  * files, we can happily check the index for a match, but for creating a
 319  * new file we should try to match whatever "patch" does. I have no idea.
 320  */
 321 static void parse_traditional_patch(const char *first, const char *second, struct patch *patch)
 322 {
 323         char *name;
 324
 325         first += 4;     /* skip "--- " */
 326         second += 4;    /* skip "+++ " */
 327         if (is_dev_null(first)) {
 328                 patch->is_new = 1;
 329                 patch->is_delete = 0;
 330                 name = find_name(second, NULL, p_value, TERM_SPACE | TERM_TAB);
 331                 patch->new_name = name;
 332         } else if (is_dev_null(second)) {
 333                 patch->is_new = 0;
 334                 patch->is_delete = 1;
 335                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 336                 patch->old_name = name;
 337         } else {
 338                 name = find_name(first, NULL, p_value, TERM_SPACE | TERM_TAB);
 339                 name = find_name(second, name, p_value, TERM_SPACE | TERM_TAB);
 340                 patch->old_name = patch->new_name = name;
 341         }
 342         if (!name)
 343                 die("unable to find filename in patch at line %d", linenr);
 344 }
 345
 346 static int gitdiff_hdrend(const char *line, struct patch *patch)
 347 {
 348         return -1;
 349 }
 350
 351 /*
 352  * We're anal about diff header consistency, to make
 353  * sure that we don't end up having strange ambiguous
 354  * patches floating around.
 355  *
 356  * As a result, gitdiff_{old|new}name() will check
 357  * their names against any previous information, just
 358  * to make sure..
 359  */
 360 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 361 {
 362         if (!orig_name && !isnull)
 363                 return find_name(line, NULL, 1, TERM_TAB);
 364
 365         if (orig_name) {
 366                 int len;
 367                 const char *name;
 368                 char *another;
 369                 name = orig_name;
 370                 len = strlen(name);
 371                 if (isnull)
 372                         die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 373                 another = find_name(line, NULL, 1, TERM_TAB);
 374                 if (!another || memcmp(another, name, len))
 375                         die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 376                 free(another);
 377                 return orig_name;
 378         }
 379         else {
 380                 /* expect "/dev/null" */
 381                 if (memcmp("/dev/null", line, 9) || line[9] != '\n')
 382                         die("git-apply: bad git-diff - expected /dev/null on line %d", linenr);
 383                 return NULL;
 384         }
 385 }
 386
 387 static int gitdiff_oldname(const char *line, struct patch *patch)
 388 {
 389         patch->old_name = gitdiff_verify_name(line, patch->is_new, patch->old_name, "old");
 390         return 0;
 391 }
 392
 393 static int gitdiff_newname(const char *line, struct patch *patch)
 394 {
 395         patch->new_name = gitdiff_verify_name(line, patch->is_delete, patch->new_name, "new");
 396         return 0;
 397 }
 398
 399 static int gitdiff_oldmode(const char *line, struct patch *patch)
 400 {
 401         patch->old_mode = strtoul(line, NULL, 8);
 402         return 0;
 403 }
 404
 405 static int gitdiff_newmode(const char *line, struct patch *patch)
 406 {
 407         patch->new_mode = strtoul(line, NULL, 8);
 408         return 0;
 409 }
 410
 411 static int gitdiff_delete(const char *line, struct patch *patch)
 412 {
 413         patch->is_delete = 1;
 414         patch->old_name = patch->def_name;
 415         return gitdiff_oldmode(line, patch);
 416 }
 417
 418 static int gitdiff_newfile(const char *line, struct patch *patch)
 419 {
 420         patch->is_new = 1;
 421         patch->new_name = patch->def_name;
 422         return gitdiff_newmode(line, patch);
 423 }
 424
 425 static int gitdiff_copysrc(const char *line, struct patch *patch)
 426 {
 427         patch->is_copy = 1;
 428         patch->old_name = find_name(line, NULL, 0, 0);
 429         return 0;
 430 }
 431
 432 static int gitdiff_copydst(const char *line, struct patch *patch)
 433 {
 434         patch->is_copy = 1;
 435         patch->new_name = find_name(line, NULL, 0, 0);
 436         return 0;
 437 }
 438
 439 static int gitdiff_renamesrc(const char *line, struct patch *patch)
 440 {
 441         patch->is_rename = 1;
 442         patch->old_name = find_name(line, NULL, 0, 0);
 443         return 0;
 444 }
 445
 446 static int gitdiff_renamedst(const char *line, struct patch *patch)
 447 {
 448         patch->is_rename = 1;
 449         patch->new_name = find_name(line, NULL, 0, 0);
 450         return 0;
 451 }
 452
 453 static int gitdiff_similarity(const char *line, struct patch *patch)
 454 {
 455         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 456                 patch->score = 0;
 457         return 0;
 458 }
 459
 460 static int gitdiff_dissimilarity(const char *line, struct patch *patch)
 461 {
 462         if ((patch->score = strtoul(line, NULL, 10)) == ULONG_MAX)
 463                 patch->score = 0;
 464         return 0;
 465 }
 466
 467 static int gitdiff_index(const char *line, struct patch *patch)
 468 {
 469         /* index line is N hexadecimal, "..", N hexadecimal,
 470          * and optional space with octal mode.
 471          */
 472         const char *ptr, *eol;
 473         int len;
 474
 475         ptr = strchr(line, '.');
 476         if (!ptr || ptr[1] != '.' || 40 < ptr - line)
 477                 return 0;
 478         len = ptr - line;
 479         memcpy(patch->old_sha1_prefix, line, len);
 480         patch->old_sha1_prefix[len] = 0;
 481
 482         line = ptr + 2;
 483         ptr = strchr(line, ' ');
 484         eol = strchr(line, '\n');
 485
 486         if (!ptr || eol < ptr)
 487                 ptr = eol;
 488         len = ptr - line;
 489
 490         if (40 < len)
 491                 return 0;
 492         memcpy(patch->new_sha1_prefix, line, len);
 493         patch->new_sha1_prefix[len] = 0;
 494         if (*ptr == ' ')
 495                 patch->new_mode = patch->old_mode = strtoul(ptr+1, NULL, 8);
 496         return 0;
 497 }
 498
 499 /*
 500  * This is normal for a diff that doesn't change anything: we'll fall through
 501  * into the next diff. Tell the parser to break out.
 502  */
 503 static int gitdiff_unrecognized(const char *line, struct patch *patch)
 504 {
 505         return -1;
 506 }
 507
 508 static const char *stop_at_slash(const char *line, int llen)
 509 {
 510         int i;
 511
 512         for (i = 0; i < llen; i++) {
 513                 int ch = line[i];
 514                 if (ch == '/')
 515                         return line + i;
 516         }
 517         return NULL;
 518 }
 519
 520 /* This is to extract the same name that appears on "diff --git"
 521  * line.  We do not find and return anything if it is a rename
 522  * patch, and it is OK because we will find the name elsewhere.
 523  * We need to reliably find name only when it is mode-change only,
 524  * creation or deletion of an empty file.  In any of these cases,
 525  * both sides are the same name under a/ and b/ respectively.
 526  */
 527 static char *git_header_name(char *line, int llen)
 528 {
 529         int len;
 530         const char *name;
 531         const char *second = NULL;
 532
 533         line += strlen("diff --git ");
 534         llen -= strlen("diff --git ");
 535
 536         if (*line == '"') {
 537                 const char *cp;
 538                 char *first = unquote_c_style(line, &second);
 539                 if (!first)
 540                         return NULL;
 541
 542                 /* advance to the first slash */
 543                 cp = stop_at_slash(first, strlen(first));
 544                 if (!cp || cp == first) {
 545                         /* we do not accept absolute paths */
 546                 free_first_and_fail:
 547                         free(first);
 548                         return NULL;
 549                 }
 550                 len = strlen(cp+1);
 551                 memmove(first, cp+1, len+1); /* including NUL */
 552
 553                 /* second points at one past closing dq of name.
 554                  * find the second name.
 555                  */
 556                 while ((second < line + llen) && isspace(*second))
 557                         second++;
 558
 559                 if (line + llen <= second)
 560                         goto free_first_and_fail;
 561                 if (*second == '"') {
 562                         char *sp = unquote_c_style(second, NULL);
 563                         if (!sp)
 564                                 goto free_first_and_fail;
 565                         cp = stop_at_slash(sp, strlen(sp));
 566                         if (!cp || cp == sp) {
 567                         free_both_and_fail:
 568                                 free(sp);
 569                                 goto free_first_and_fail;
 570                         }
 571                         /* They must match, otherwise ignore */
 572                         if (strcmp(cp+1, first))
 573                                 goto free_both_and_fail;
 574                         free(sp);
 575                         return first;
 576                 }
 577
 578                 /* unquoted second */
 579                 cp = stop_at_slash(second, line + llen - second);
 580                 if (!cp || cp == second)
 581                         goto free_first_and_fail;
 582                 cp++;
 583                 if (line + llen - cp != len + 1 ||
 584                     memcmp(first, cp, len))
 585                         goto free_first_and_fail;
 586                 return first;
 587         }
 588
 589         /* unquoted first name */
 590         name = stop_at_slash(line, llen);
 591         if (!name || name == line)
 592                 return NULL;
 593
 594         name++;
 595
 596         /* since the first name is unquoted, a dq if exists must be
 597          * the beginning of the second name.
 598          */
 599         for (second = name; second < line + llen; second++) {
 600                 if (*second == '"') {
 601                         const char *cp = second;
 602                         const char *np;
 603                         char *sp = unquote_c_style(second, NULL);
 604
 605                         if (!sp)
 606                                 return NULL;
 607                         np = stop_at_slash(sp, strlen(sp));
 608                         if (!np || np == sp) {
 609                         free_second_and_fail:
 610                                 free(sp);
 611                                 return NULL;
 612                         }
 613                         np++;
 614                         len = strlen(np);
 615                         if (len < cp - name &&
 616                             !strncmp(np, name, len) &&
 617                             isspace(name[len])) {
 618                                 /* Good */
 619                                 memmove(sp, np, len + 1);
 620                                 return sp;
 621                         }
 622                         goto free_second_and_fail;
 623                 }
 624         }
 625
 626         /*
 627          * Accept a name only if it shows up twice, exactly the same
 628          * form.
 629          */
 630         for (len = 0 ; ; len++) {
 631                 switch (name[len]) {
 632                 default:
 633                         continue;
 634                 case '\n':
 635                         return NULL;
 636                 case '\t': case ' ':
 637                         second = name+len;
 638                         for (;;) {
 639                                 char c = *second++;
 640                                 if (c == '\n')
 641                                         return NULL;
 642                                 if (c == '/')
 643                                         break;
 644                         }
 645                         if (second[len] == '\n' && !memcmp(name, second, len)) {
 646                                 char *ret = xmalloc(len + 1);
 647                                 memcpy(ret, name, len);
 648                                 ret[len] = 0;
 649                                 return ret;
 650                         }
 651                 }
 652         }
 653         return NULL;
 654 }
 655
 656 /* Verify that we recognize the lines following a git header */
 657 static int parse_git_header(char *line, int len, unsigned int size, struct patch *patch)
 658 {
 659         unsigned long offset;
 660
 661         /* A git diff has explicit new/delete information, so we don't guess */
 662         patch->is_new = 0;
 663         patch->is_delete = 0;
 664
 665         /*
 666          * Some things may not have the old name in the
 667          * rest of the headers anywhere (pure mode changes,
 668          * or removing or adding empty files), so we get
 669          * the default name from the header.
 670          */
 671         patch->def_name = git_header_name(line, len);
 672
 673         line += len;
 674         size -= len;
 675         linenr++;
 676         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 677                 static const struct opentry {
 678                         const char *str;
 679                         int (*fn)(const char *, struct patch *);
 680                 } optable[] = {
 681                         { "@@ -", gitdiff_hdrend },
 682                         { "--- ", gitdiff_oldname },
 683                         { "+++ ", gitdiff_newname },
 684                         { "old mode ", gitdiff_oldmode },
 685                         { "new mode ", gitdiff_newmode },
 686                         { "deleted file mode ", gitdiff_delete },
 687                         { "new file mode ", gitdiff_newfile },
 688                         { "copy from ", gitdiff_copysrc },
 689                         { "copy to ", gitdiff_copydst },
 690                         { "rename old ", gitdiff_renamesrc },
 691                         { "rename new ", gitdiff_renamedst },
 692                         { "rename from ", gitdiff_renamesrc },
 693                         { "rename to ", gitdiff_renamedst },
 694                         { "similarity index ", gitdiff_similarity },
 695                         { "dissimilarity index ", gitdiff_dissimilarity },
 696                         { "index ", gitdiff_index },
 697                         { "", gitdiff_unrecognized },
 698                 };
 699                 int i;
 700
 701                 len = linelen(line, size);
 702                 if (!len || line[len-1] != '\n')
 703                         break;
 704                 for (i = 0; i < ARRAY_SIZE(optable); i++) {
 705                         const struct opentry *p = optable + i;
 706                         int oplen = strlen(p->str);
 707                         if (len < oplen || memcmp(p->str, line, oplen))
 708                                 continue;
 709                         if (p->fn(line + oplen, patch) < 0)
 710                                 return offset;
 711                         break;
 712                 }
 713         }
 714
 715         return offset;
 716 }
 717
 718 static int parse_num(const char *line, unsigned long *p)
 719 {
 720         char *ptr;
 721
 722         if (!isdigit(*line))
 723                 return 0;
 724         *p = strtoul(line, &ptr, 10);
 725         return ptr - line;
 726 }
 727
 728 static int parse_range(const char *line, int len, int offset, const char *expect,
 729                         unsigned long *p1, unsigned long *p2)
 730 {
 731         int digits, ex;
 732
 733         if (offset < 0 || offset >= len)
 734                 return -1;
 735         line += offset;
 736         len -= offset;
 737
 738         digits = parse_num(line, p1);
 739         if (!digits)
 740                 return -1;
 741
 742         offset += digits;
 743         line += digits;
 744         len -= digits;
 745
 746         *p2 = 1;
 747         if (*line == ',') {
 748                 digits = parse_num(line+1, p2);
 749                 if (!digits)
 750                         return -1;
 751
 752                 offset += digits+1;
 753                 line += digits+1;
 754                 len -= digits+1;
 755         }
 756
 757         ex = strlen(expect);
 758         if (ex > len)
 759                 return -1;
 760         if (memcmp(line, expect, ex))
 761                 return -1;
 762
 763         return offset + ex;
 764 }
 765
 766 /*
 767  * Parse a unified diff fragment header of the
 768  * form "@@ -a,b +c,d @@"
 769  */
 770 static int parse_fragment_header(char *line, int len, struct fragment *fragment)
 771 {
 772         int offset;
 773
 774         if (!len || line[len-1] != '\n')
 775                 return -1;
 776
 777         /* Figure out the number of lines in a fragment */
 778         offset = parse_range(line, len, 4, " +", &fragment->oldpos, &fragment->oldlines);
 779         offset = parse_range(line, len, offset, " @@", &fragment->newpos, &fragment->newlines);
 780
 781         return offset;
 782 }
 783
 784 static int find_header(char *line, unsigned long size, int *hdrsize, struct patch *patch)
 785 {
 786         unsigned long offset, len;
 787
 788         patch->is_rename = patch->is_copy = 0;
 789         patch->is_new = patch->is_delete = -1;
 790         patch->old_mode = patch->new_mode = 0;
 791         patch->old_name = patch->new_name = NULL;
 792         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
 793                 unsigned long nextlen;
 794
 795                 len = linelen(line, size);
 796                 if (!len)
 797                         break;
 798
 799                 /* Testing this early allows us to take a few shortcuts.. */
 800                 if (len < 6)
 801                         continue;
 802
 803                 /*
 804                  * Make sure we don't find any unconnected patch fragments.
 805                  * That's a sign that we didn't find a header, and that a
 806                  * patch has become corrupted/broken up.
 807                  */
 808                 if (!memcmp("@@ -", line, 4)) {
 809                         struct fragment dummy;
 810                         if (parse_fragment_header(line, len, &dummy) < 0)
 811                                 continue;
 812                         error("patch fragment without header at line %d: %.*s", linenr, (int)len-1, line);
 813                 }
 814
 815                 if (size < len + 6)
 816                         break;
 817
 818                 /*
 819                  * Git patch? It might not have a real patch, just a rename
 820                  * or mode change, so we handle that specially
 821                  */
 822                 if (!memcmp("diff --git ", line, 11)) {
 823                         int git_hdr_len = parse_git_header(line, len, size, patch);
 824                         if (git_hdr_len <= len)
 825                                 continue;
 826                         if (!patch->old_name && !patch->new_name) {
 827                                 if (!patch->def_name)
 828                                         die("git diff header lacks filename information (line %d)", linenr);
 829                                 patch->old_name = patch->new_name = patch->def_name;
 830                         }
 831                         *hdrsize = git_hdr_len;
 832                         return offset;
 833                 }
 834
 835                 /** --- followed by +++ ? */
 836                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
 837                         continue;
 838
 839                 /*
 840                  * We only accept unified patches, so we want it to
 841                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
 842                  * minimum
 843                  */
 844                 nextlen = linelen(line + len, size - len);
 845                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
 846                         continue;
 847
 848                 /* Ok, we'll consider it a patch */
 849                 parse_traditional_patch(line, line+len, patch);
 850                 *hdrsize = len + nextlen;
 851                 linenr += 2;
 852                 return offset;
 853         }
 854         return -1;
 855 }
 856
 857 static void check_whitespace(const char *line, int len)
 858 {
 859         const char *err = "Adds trailing whitespace";
 860         int seen_space = 0;
 861         int i;
 862
 863         /*
 864          * We know len is at least two, since we have a '+' and we
 865          * checked that the last character was a '\n' before calling
 866          * this function.  That is, an addition of an empty line would
 867          * check the '+' here.  Sneaky...
 868          */
 869         if (isspace(line[len-2]))
 870                 goto error;
 871
 872         /*
 873          * Make sure that there is no space followed by a tab in
 874          * indentation.
 875          */
 876         err = "Space in indent is followed by a tab";
 877         for (i = 1; i < len; i++) {
 878                 if (line[i] == '\t') {
 879                         if (seen_space)
 880                                 goto error;
 881                 }
 882                 else if (line[i] == ' ')
 883                         seen_space = 1;
 884                 else
 885                         break;
 886         }
 887         return;
 888
 889  error:
 890         whitespace_error++;
 891         if (squelch_whitespace_errors &&
 892             squelch_whitespace_errors < whitespace_error)
 893                 ;
 894         else
 895                 fprintf(stderr, "%s.\n%s:%d:%.*s\n",
 896                         err, patch_input_file, linenr, len-2, line+1);
 897 }
 898
 899
 900 /*
 901  * Parse a unified diff. Note that this really needs to parse each
 902  * fragment separately, since the only way to know the difference
 903  * between a "---" that is part of a patch, and a "---" that starts
 904  * the next patch is to look at the line counts..
 905  */
 906 static int parse_fragment(char *line, unsigned long size, struct patch *patch, struct fragment *fragment)
 907 {
 908         int added, deleted;
 909         int len = linelen(line, size), offset;
 910         unsigned long oldlines, newlines;
 911         unsigned long leading, trailing;
 912
 913         offset = parse_fragment_header(line, len, fragment);
 914         if (offset < 0)
 915                 return -1;
 916         oldlines = fragment->oldlines;
 917         newlines = fragment->newlines;
 918         leading = 0;
 919         trailing = 0;
 920
 921         /* Parse the thing.. */
 922         line += len;
 923         size -= len;
 924         linenr++;
 925         added = deleted = 0;
 926         for (offset = len;
 927              0 < size;
 928              offset += len, size -= len, line += len, linenr++) {
 929                 if (!oldlines && !newlines)
 930                         break;
 931                 len = linelen(line, size);
 932                 if (!len || line[len-1] != '\n')
 933                         return -1;
 934                 switch (*line) {
 935                 default:
 936                         return -1;
 937                 case '\n': /* newer GNU diff, an empty context line */
 938                 case ' ':
 939                         oldlines--;
 940                         newlines--;
 941                         if (!deleted && !added)
 942                                 leading++;
 943                         trailing++;
 944                         break;
 945                 case '-':
 946                         deleted++;
 947                         oldlines--;
 948                         trailing = 0;
 949                         break;
 950                 case '+':
 951                         if (new_whitespace != nowarn_whitespace)
 952                                 check_whitespace(line, len);
 953                         added++;
 954                         newlines--;
 955                         trailing = 0;
 956                         break;
 957
 958                 /* We allow "\ No newline at end of file". Depending
 959                  * on locale settings when the patch was produced we
 960                  * don't know what this line looks like. The only
 961                  * thing we do know is that it begins with "\ ".
 962                  * Checking for 12 is just for sanity check -- any
 963                  * l10n of "\ No newline..." is at least that long.
 964                  */
 965                 case '\\':
 966                         if (len < 12 || memcmp(line, "\\ ", 2))
 967                                 return -1;
 968                         break;
 969                 }
 970         }
 971         if (oldlines || newlines)
 972                 return -1;
 973         fragment->leading = leading;
 974         fragment->trailing = trailing;
 975
 976         /* If a fragment ends with an incomplete line, we failed to include
 977          * it in the above loop because we hit oldlines == newlines == 0
 978          * before seeing it.
 979          */
 980         if (12 < size && !memcmp(line, "\\ ", 2))
 981                 offset += linelen(line, size);
 982
 983         patch->lines_added += added;
 984         patch->lines_deleted += deleted;
 985
 986         if (0 < patch->is_new && oldlines)
 987                 return error("new file depends on old contents");
 988         if (0 < patch->is_delete && newlines)
 989                 return error("deleted file still has contents");
 990         return offset;
 991 }
 992
 993 static int parse_single_patch(char *line, unsigned long size, struct patch *patch)
 994 {
 995         unsigned long offset = 0;
 996         unsigned long oldlines = 0, newlines = 0, context = 0;
 997         struct fragment **fragp = &patch->fragments;
 998
 999         while (size > 4 && !memcmp(line, "@@ -", 4)) {
1000                 struct fragment *fragment;
1001                 int len;
1002
1003                 fragment = xcalloc(1, sizeof(*fragment));
1004                 len = parse_fragment(line, size, patch, fragment);
1005                 if (len <= 0)
1006                         die("corrupt patch at line %d", linenr);
1007                 fragment->patch = line;
1008                 fragment->size = len;
1009                 oldlines += fragment->oldlines;
1010                 newlines += fragment->newlines;
1011                 context += fragment->leading + fragment->trailing;
1012
1013                 *fragp = fragment;
1014                 fragp = &fragment->next;
1015
1016                 offset += len;
1017                 line += len;
1018                 size -= len;
1019         }
1020
1021         /*
1022          * If something was removed (i.e. we have old-lines) it cannot
1023          * be creation, and if something was added it cannot be
1024          * deletion.  However, the reverse is not true; --unified=0
1025          * patches that only add are not necessarily creation even
1026          * though they do not have any old lines, and ones that only
1027          * delete are not necessarily deletion.
1028          *
1029          * Unfortunately, a real creation/deletion patch do _not_ have
1030          * any context line by definition, so we cannot safely tell it
1031          * apart with --unified=0 insanity.  At least if the patch has
1032          * more than one hunk it is not creation or deletion.
1033          */
1034         if (patch->is_new < 0 &&
1035             (oldlines || (patch->fragments && patch->fragments->next)))
1036                 patch->is_new = 0;
1037         if (patch->is_delete < 0 &&
1038             (newlines || (patch->fragments && patch->fragments->next)))
1039                 patch->is_delete = 0;
1040         if (!unidiff_zero || context) {
1041                 /* If the user says the patch is not generated with
1042                  * --unified=0, or if we have seen context lines,
1043                  * then not having oldlines means the patch is creation,
1044                  * and not having newlines means the patch is deletion.
1045                  */
1046                 if (patch->is_new < 0 && !oldlines) {
1047                         patch->is_new = 1;
1048                         patch->old_name = NULL;
1049                 }
1050                 if (patch->is_delete < 0 && !newlines) {
1051                         patch->is_delete = 1;
1052                         patch->new_name = NULL;
1053                 }
1054         }
1055
1056         if (0 < patch->is_new && oldlines)
1057                 die("new file %s depends on old contents", patch->new_name);
1058         if (0 < patch->is_delete && newlines)
1059                 die("deleted file %s still has contents", patch->old_name);
1060         if (!patch->is_delete && !newlines && context)
1061                 fprintf(stderr, "** warning: file %s becomes empty but "
1062                         "is not deleted\n", patch->new_name);
1063
1064         return offset;
1065 }
1066
1067 static inline int metadata_changes(struct patch *patch)
1068 {
1069         return  patch->is_rename > 0 ||
1070                 patch->is_copy > 0 ||
1071                 patch->is_new > 0 ||
1072                 patch->is_delete ||
1073                 (patch->old_mode && patch->new_mode &&
1074                  patch->old_mode != patch->new_mode);
1075 }
1076
1077 static char *inflate_it(const void *data, unsigned long size,
1078                         unsigned long inflated_size)
1079 {
1080         z_stream stream;
1081         void *out;
1082         int st;
1083
1084         memset(&stream, 0, sizeof(stream));
1085
1086         stream.next_in = (unsigned char *)data;
1087         stream.avail_in = size;
1088         stream.next_out = out = xmalloc(inflated_size);
1089         stream.avail_out = inflated_size;
1090         inflateInit(&stream);
1091         st = inflate(&stream, Z_FINISH);
1092         if ((st != Z_STREAM_END) || stream.total_out != inflated_size) {
1093                 free(out);
1094                 return NULL;
1095         }
1096         return out;
1097 }
1098
1099 static struct fragment *parse_binary_hunk(char **buf_p,
1100                                           unsigned long *sz_p,
1101                                           int *status_p,
1102                                           int *used_p)
1103 {
1104         /* Expect a line that begins with binary patch method ("literal"
1105          * or "delta"), followed by the length of data before deflating.
1106          * a sequence of 'length-byte' followed by base-85 encoded data
1107          * should follow, terminated by a newline.
1108          *
1109          * Each 5-byte sequence of base-85 encodes up to 4 bytes,
1110          * and we would limit the patch line to 66 characters,
1111          * so one line can fit up to 13 groups that would decode
1112          * to 52 bytes max.  The length byte 'A'-'Z' corresponds
1113          * to 1-26 bytes, and 'a'-'z' corresponds to 27-52 bytes.
1114          */
1115         int llen, used;
1116         unsigned long size = *sz_p;
1117         char *buffer = *buf_p;
1118         int patch_method;
1119         unsigned long origlen;
1120         char *data = NULL;
1121         int hunk_size = 0;
1122         struct fragment *frag;
1123
1124         llen = linelen(buffer, size);
1125         used = llen;
1126
1127         *status_p = 0;
1128
1129         if (!strncmp(buffer, "delta ", 6)) {
1130                 patch_method = BINARY_DELTA_DEFLATED;
1131                 origlen = strtoul(buffer + 6, NULL, 10);
1132         }
1133         else if (!strncmp(buffer, "literal ", 8)) {
1134                 patch_method = BINARY_LITERAL_DEFLATED;
1135                 origlen = strtoul(buffer + 8, NULL, 10);
1136         }
1137         else
1138                 return NULL;
1139
1140         linenr++;
1141         buffer += llen;
1142         while (1) {
1143                 int byte_length, max_byte_length, newsize;
1144                 llen = linelen(buffer, size);
1145                 used += llen;
1146                 linenr++;
1147                 if (llen == 1) {
1148                         /* consume the blank line */
1149                         buffer++;
1150                         size--;
1151                         break;
1152                 }
1153                 /* Minimum line is "A00000\n" which is 7-byte long,
1154                  * and the line length must be multiple of 5 plus 2.
1155                  */
1156                 if ((llen < 7) || (llen-2) % 5)
1157                         goto corrupt;
1158                 max_byte_length = (llen - 2) / 5 * 4;
1159                 byte_length = *buffer;
1160                 if ('A' <= byte_length && byte_length <= 'Z')
1161                         byte_length = byte_length - 'A' + 1;
1162                 else if ('a' <= byte_length && byte_length <= 'z')
1163                         byte_length = byte_length - 'a' + 27;
1164                 else
1165                         goto corrupt;
1166                 /* if the input length was not multiple of 4, we would
1167                  * have filler at the end but the filler should never
1168                  * exceed 3 bytes
1169                  */
1170                 if (max_byte_length < byte_length ||
1171                     byte_length <= max_byte_length - 4)
1172                         goto corrupt;
1173                 newsize = hunk_size + byte_length;
1174                 data = xrealloc(data, newsize);
1175                 if (decode_85(data + hunk_size, buffer + 1, byte_length))
1176                         goto corrupt;
1177                 hunk_size = newsize;
1178                 buffer += llen;
1179                 size -= llen;
1180         }
1181
1182         frag = xcalloc(1, sizeof(*frag));
1183         frag->patch = inflate_it(data, hunk_size, origlen);
1184         if (!frag->patch)
1185                 goto corrupt;
1186         free(data);
1187         frag->size = origlen;
1188         *buf_p = buffer;
1189         *sz_p = size;
1190         *used_p = used;
1191         frag->binary_patch_method = patch_method;
1192         return frag;
1193
1194  corrupt:
1195         free(data);
1196         *status_p = -1;
1197         error("corrupt binary patch at line %d: %.*s",
1198               linenr-1, llen-1, buffer);
1199         return NULL;
1200 }
1201
1202 static int parse_binary(char *buffer, unsigned long size, struct patch *patch)
1203 {
1204         /* We have read "GIT binary patch\n"; what follows is a line
1205          * that says the patch method (currently, either "literal" or
1206          * "delta") and the length of data before deflating; a
1207          * sequence of 'length-byte' followed by base-85 encoded data
1208          * follows.
1209          *
1210          * When a binary patch is reversible, there is another binary
1211          * hunk in the same format, starting with patch method (either
1212          * "literal" or "delta") with the length of data, and a sequence
1213          * of length-byte + base-85 encoded data, terminated with another
1214          * empty line.  This data, when applied to the postimage, produces
1215          * the preimage.
1216          */
1217         struct fragment *forward;
1218         struct fragment *reverse;
1219         int status;
1220         int used, used_1;
1221
1222         forward = parse_binary_hunk(&buffer, &size, &status, &used);
1223         if (!forward && !status)
1224                 /* there has to be one hunk (forward hunk) */
1225                 return error("unrecognized binary patch at line %d", linenr-1);
1226         if (status)
1227                 /* otherwise we already gave an error message */
1228                 return status;
1229
1230         reverse = parse_binary_hunk(&buffer, &size, &status, &used_1);
1231         if (reverse)
1232                 used += used_1;
1233         else if (status) {
1234                 /* not having reverse hunk is not an error, but having
1235                  * a corrupt reverse hunk is.
1236                  */
1237                 free((void*) forward->patch);
1238                 free(forward);
1239                 return status;
1240         }
1241         forward->next = reverse;
1242         patch->fragments = forward;
1243         patch->is_binary = 1;
1244         return used;
1245 }
1246
1247 static int parse_chunk(char *buffer, unsigned long size, struct patch *patch)
1248 {
1249         int hdrsize, patchsize;
1250         int offset = find_header(buffer, size, &hdrsize, patch);
1251
1252         if (offset < 0)
1253                 return offset;
1254
1255         patchsize = parse_single_patch(buffer + offset + hdrsize, size - offset - hdrsize, patch);
1256
1257         if (!patchsize) {
1258                 static const char *binhdr[] = {
1259                         "Binary files ",
1260                         "Files ",
1261                         NULL,
1262                 };
1263                 static const char git_binary[] = "GIT binary patch\n";
1264                 int i;
1265                 int hd = hdrsize + offset;
1266                 unsigned long llen = linelen(buffer + hd, size - hd);
1267
1268                 if (llen == sizeof(git_binary) - 1 &&
1269                     !memcmp(git_binary, buffer + hd, llen)) {
1270                         int used;
1271                         linenr++;
1272                         used = parse_binary(buffer + hd + llen,
1273                                             size - hd - llen, patch);
1274                         if (used)
1275                                 patchsize = used + llen;
1276                         else
1277                                 patchsize = 0;
1278                 }
1279                 else if (!memcmp(" differ\n", buffer + hd + llen - 8, 8)) {
1280                         for (i = 0; binhdr[i]; i++) {
1281                                 int len = strlen(binhdr[i]);
1282                                 if (len < size - hd &&
1283                                     !memcmp(binhdr[i], buffer + hd, len)) {
1284                                         linenr++;
1285                                         patch->is_binary = 1;
1286                                         patchsize = llen;
1287                                         break;
1288                                 }
1289                         }
1290                 }
1291
1292                 /* Empty patch cannot be applied if it is a text patch
1293                  * without metadata change.  A binary patch appears
1294                  * empty to us here.
1295                  */
1296                 if ((apply || check) &&
1297                     (!patch->is_binary && !metadata_changes(patch)))
1298                         die("patch with only garbage at line %d", linenr);
1299         }
1300
1301         return offset + hdrsize + patchsize;
1302 }
1303
1304 #define swap(a,b) myswap((a),(b),sizeof(a))
1305
1306 #define myswap(a, b, size) do {         \
1307         unsigned char mytmp[size];      \
1308         memcpy(mytmp, &a, size);                \
1309         memcpy(&a, &b, size);           \
1310         memcpy(&b, mytmp, size);                \
1311 } while (0)
1312
1313 static void reverse_patches(struct patch *p)
1314 {
1315         for (; p; p = p->next) {
1316                 struct fragment *frag = p->fragments;
1317
1318                 swap(p->new_name, p->old_name);
1319                 swap(p->new_mode, p->old_mode);
1320                 swap(p->is_new, p->is_delete);
1321                 swap(p->lines_added, p->lines_deleted);
1322                 swap(p->old_sha1_prefix, p->new_sha1_prefix);
1323
1324                 for (; frag; frag = frag->next) {
1325                         swap(frag->newpos, frag->oldpos);
1326                         swap(frag->newlines, frag->oldlines);
1327                 }
1328         }
1329 }
1330
1331 static const char pluses[] = "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++";
1332 static const char minuses[]= "----------------------------------------------------------------------";
1333
1334 static void show_stats(struct patch *patch)
1335 {
1336         const char *prefix = "";
1337         char *name = patch->new_name;
1338         char *qname = NULL;
1339         int len, max, add, del, total;
1340
1341         if (!name)
1342                 name = patch->old_name;
1343
1344         if (0 < (len = quote_c_style(name, NULL, NULL, 0))) {
1345                 qname = xmalloc(len + 1);
1346                 quote_c_style(name, qname, NULL, 0);
1347                 name = qname;
1348         }
1349
1350         /*
1351          * "scale" the filename
1352          */
1353         len = strlen(name);
1354         max = max_len;
1355         if (max > 50)
1356                 max = 50;
1357         if (len > max) {
1358                 char *slash;
1359                 prefix = "...";
1360                 max -= 3;
1361                 name += len - max;
1362                 slash = strchr(name, '/');
1363                 if (slash)
1364                         name = slash;
1365         }
1366         len = max;
1367
1368         /*
1369          * scale the add/delete
1370          */
1371         max = max_change;
1372         if (max + len > 70)
1373                 max = 70 - len;
1374
1375         add = patch->lines_added;
1376         del = patch->lines_deleted;
1377         total = add + del;
1378
1379         if (max_change > 0) {
1380                 total = (total * max + max_change / 2) / max_change;
1381                 add = (add * max + max_change / 2) / max_change;
1382                 del = total - add;
1383         }
1384         if (patch->is_binary)
1385                 printf(" %s%-*s |  Bin\n", prefix, len, name);
1386         else
1387                 printf(" %s%-*s |%5d %.*s%.*s\n", prefix,
1388                        len, name, patch->lines_added + patch->lines_deleted,
1389                        add, pluses, del, minuses);
1390         free(qname);
1391 }
1392
1393 static int read_old_data(struct stat *st, const char *path, void *buf, unsigned long size)
1394 {
1395         int fd;
1396         unsigned long got;
1397
1398         switch (st->st_mode & S_IFMT) {
1399         case S_IFLNK:
1400                 return readlink(path, buf, size);
1401         case S_IFREG:
1402                 fd = open(path, O_RDONLY);
1403                 if (fd < 0)
1404                         return error("unable to open %s", path);
1405                 got = 0;
1406                 for (;;) {
1407                         int ret = xread(fd, (char *) buf + got, size - got);
1408                         if (ret <= 0)
1409                                 break;
1410                         got += ret;
1411                 }
1412                 close(fd);
1413                 return got;
1414
1415         default:
1416                 return -1;
1417         }
1418 }
1419
1420 static int find_offset(const char *buf, unsigned long size, const char *fragment, unsigned long fragsize, int line, int *lines)
1421 {
1422         int i;
1423         unsigned long start, backwards, forwards;
1424
1425         if (fragsize > size)
1426                 return -1;
1427
1428         start = 0;
1429         if (line > 1) {
1430                 unsigned long offset = 0;
1431                 i = line-1;
1432                 while (offset + fragsize <= size) {
1433                         if (buf[offset++] == '\n') {
1434                                 start = offset;
1435                                 if (!--i)
1436                                         break;
1437                         }
1438                 }
1439         }
1440
1441         /* Exact line number? */
1442         if (!memcmp(buf + start, fragment, fragsize))
1443                 return start;
1444
1445         /*
1446          * There's probably some smart way to do this, but I'll leave
1447          * that to the smart and beautiful people. I'm simple and stupid.
1448          */
1449         backwards = start;
1450         forwards = start;
1451         for (i = 0; ; i++) {
1452                 unsigned long try;
1453                 int n;
1454
1455                 /* "backward" */
1456                 if (i & 1) {
1457                         if (!backwards) {
1458                                 if (forwards + fragsize > size)
1459                                         break;
1460                                 continue;
1461                         }
1462                         do {
1463                                 --backwards;
1464                         } while (backwards && buf[backwards-1] != '\n');
1465                         try = backwards;
1466                 } else {
1467                         while (forwards + fragsize <= size) {
1468                                 if (buf[forwards++] == '\n')
1469                                         break;
1470                         }
1471                         try = forwards;
1472                 }
1473
1474                 if (try + fragsize > size)
1475                         continue;
1476                 if (memcmp(buf + try, fragment, fragsize))
1477                         continue;
1478                 n = (i >> 1)+1;
1479                 if (i & 1)
1480                         n = -n;
1481                 *lines = n;
1482                 return try;
1483         }
1484
1485         /*
1486          * We should start searching forward and backward.
1487          */
1488         return -1;
1489 }
1490
1491 static void remove_first_line(const char **rbuf, int *rsize)
1492 {
1493         const char *buf = *rbuf;
1494         int size = *rsize;
1495         unsigned long offset;
1496         offset = 0;
1497         while (offset <= size) {
1498                 if (buf[offset++] == '\n')
1499                         break;
1500         }
1501         *rsize = size - offset;
1502         *rbuf = buf + offset;
1503 }
1504
1505 static void remove_last_line(const char **rbuf, int *rsize)
1506 {
1507         const char *buf = *rbuf;
1508         int size = *rsize;
1509         unsigned long offset;
1510         offset = size - 1;
1511         while (offset > 0) {
1512                 if (buf[--offset] == '\n')
1513                         break;
1514         }
1515         *rsize = offset + 1;
1516 }
1517
1518 struct buffer_desc {
1519         char *buffer;
1520         unsigned long size;
1521         unsigned long alloc;
1522 };
1523
1524 static int apply_line(char *output, const char *patch, int plen)
1525 {
1526         /* plen is number of bytes to be copied from patch,
1527          * starting at patch+1 (patch[0] is '+').  Typically
1528          * patch[plen] is '\n', unless this is the incomplete
1529          * last line.
1530          */
1531         int i;
1532         int add_nl_to_tail = 0;
1533         int fixed = 0;
1534         int last_tab_in_indent = -1;
1535         int last_space_in_indent = -1;
1536         int need_fix_leading_space = 0;
1537         char *buf;
1538
1539         if ((new_whitespace != strip_whitespace) || !whitespace_error) {
1540                 memcpy(output, patch + 1, plen);
1541                 return plen;
1542         }
1543
1544         if (1 < plen && isspace(patch[plen-1])) {
1545                 if (patch[plen] == '\n')
1546                         add_nl_to_tail = 1;
1547                 plen--;
1548                 while (0 < plen && isspace(patch[plen]))
1549                         plen--;
1550                 fixed = 1;
1551         }
1552
1553         for (i = 1; i < plen; i++) {
1554                 char ch = patch[i];
1555                 if (ch == '\t') {
1556                         last_tab_in_indent = i;
1557                         if (0 <= last_space_in_indent)
1558                                 need_fix_leading_space = 1;
1559                 }
1560                 else if (ch == ' ')
1561                         last_space_in_indent = i;
1562                 else
1563                         break;
1564         }
1565
1566         buf = output;
1567         if (need_fix_leading_space) {
1568                 /* between patch[1..last_tab_in_indent] strip the
1569                  * funny spaces, updating them to tab as needed.
1570                  */
1571                 for (i = 1; i < last_tab_in_indent; i++, plen--) {
1572                         char ch = patch[i];
1573                         if (ch != ' ')
1574                                 *output++ = ch;
1575                         else if ((i % 8) == 0)
1576                                 *output++ = '\t';
1577                 }
1578                 fixed = 1;
1579                 i = last_tab_in_indent;
1580         }
1581         else
1582                 i = 1;
1583
1584         memcpy(output, patch + i, plen);
1585         if (add_nl_to_tail)
1586                 output[plen++] = '\n';
1587         if (fixed)
1588                 applied_after_stripping++;
1589         return output + plen - buf;
1590 }
1591
1592 static int apply_one_fragment(struct buffer_desc *desc, struct fragment *frag, int inaccurate_eof)
1593 {
1594         int match_beginning, match_end;
1595         char *buf = desc->buffer;
1596         const char *patch = frag->patch;
1597         int offset, size = frag->size;
1598         char *old = xmalloc(size);
1599         char *new = xmalloc(size);
1600         const char *oldlines, *newlines;
1601         int oldsize = 0, newsize = 0;
1602         unsigned long leading, trailing;
1603         int pos, lines;
1604
1605         while (size > 0) {
1606                 char first;
1607                 int len = linelen(patch, size);
1608                 int plen;
1609
1610                 if (!len)
1611                         break;
1612
1613                 /*
1614                  * "plen" is how much of the line we should use for
1615                  * the actual patch data. Normally we just remove the
1616                  * first character on the line, but if the line is
1617                  * followed by "\ No newline", then we also remove the
1618                  * last one (which is the newline, of course).
1619                  */
1620                 plen = len-1;
1621                 if (len < size && patch[len] == '\\')
1622                         plen--;
1623                 first = *patch;
1624                 if (apply_in_reverse) {
1625                         if (first == '-')
1626                                 first = '+';
1627                         else if (first == '+')
1628                                 first = '-';
1629                 }
1630                 switch (first) {
1631                 case '\n':
1632                         /* Newer GNU diff, empty context line */
1633                         if (plen < 0)
1634                                 /* ... followed by '\No newline'; nothing */
1635                                 break;
1636                         old[oldsize++] = '\n';
1637                         new[newsize++] = '\n';
1638                         break;
1639                 case ' ':
1640                 case '-':
1641                         memcpy(old + oldsize, patch + 1, plen);
1642                         oldsize += plen;
1643                         if (first == '-')
1644                                 break;
1645                 /* Fall-through for ' ' */
1646                 case '+':
1647                         if (first != '+' || !no_add)
1648                                 newsize += apply_line(new + newsize, patch,
1649                                                       plen);
1650                         break;
1651                 case '@': case '\\':
1652                         /* Ignore it, we already handled it */
1653                         break;
1654                 default:
1655                         return -1;
1656                 }
1657                 patch += len;
1658                 size -= len;
1659         }
1660
1661         if (inaccurate_eof && oldsize > 0 && old[oldsize - 1] == '\n' &&
1662                         newsize > 0 && new[newsize - 1] == '\n') {
1663                 oldsize--;
1664                 newsize--;
1665         }
1666
1667         oldlines = old;
1668         newlines = new;
1669         leading = frag->leading;
1670         trailing = frag->trailing;
1671
1672         /*
1673          * If we don't have any leading/trailing data in the patch,
1674          * we want it to match at the beginning/end of the file.
1675          *
1676          * But that would break if the patch is generated with
1677          * --unified=0; sane people wouldn't do that to cause us
1678          * trouble, but we try to please not so sane ones as well.
1679          */
1680         if (unidiff_zero) {
1681                 match_beginning = (!leading && !frag->oldpos);
1682                 match_end = 0;
1683         }
1684         else {
1685                 match_beginning = !leading && (frag->oldpos == 1);
1686                 match_end = !trailing;
1687         }
1688
1689         lines = 0;
1690         pos = frag->newpos;
1691         for (;;) {
1692                 offset = find_offset(buf, desc->size,
1693                                      oldlines, oldsize, pos, &lines);
1694                 if (match_end && offset + oldsize != desc->size)
1695                         offset = -1;
1696                 if (match_beginning && offset)
1697                         offset = -1;
1698                 if (offset >= 0) {
1699                         int diff = newsize - oldsize;
1700                         unsigned long size = desc->size + diff;
1701                         unsigned long alloc = desc->alloc;
1702
1703                         /* Warn if it was necessary to reduce the number
1704                          * of context lines.
1705                          */
1706                         if ((leading != frag->leading) ||
1707                             (trailing != frag->trailing))
1708                                 fprintf(stderr, "Context reduced to (%ld/%ld)"
1709                                         " to apply fragment at %d\n",
1710                                         leading, trailing, pos + lines);
1711
1712                         if (size > alloc) {
1713                                 alloc = size + 8192;
1714                                 desc->alloc = alloc;
1715                                 buf = xrealloc(buf, alloc);
1716                                 desc->buffer = buf;
1717                         }
1718                         desc->size = size;
1719                         memmove(buf + offset + newsize,
1720                                 buf + offset + oldsize,
1721                                 size - offset - newsize);
1722                         memcpy(buf + offset, newlines, newsize);
1723                         offset = 0;
1724
1725                         break;
1726                 }
1727
1728                 /* Am I at my context limits? */
1729                 if ((leading <= p_context) && (trailing <= p_context))
1730                         break;
1731                 if (match_beginning || match_end) {
1732                         match_beginning = match_end = 0;
1733                         continue;
1734                 }
1735                 /* Reduce the number of context lines
1736                  * Reduce both leading and trailing if they are equal
1737                  * otherwise just reduce the larger context.
1738                  */
1739                 if (leading >= trailing) {
1740                         remove_first_line(&oldlines, &oldsize);
1741                         remove_first_line(&newlines, &newsize);
1742                         pos--;
1743                         leading--;
1744                 }
1745                 if (trailing > leading) {
1746                         remove_last_line(&oldlines, &oldsize);
1747                         remove_last_line(&newlines, &newsize);
1748                         trailing--;
1749                 }
1750         }
1751
1752         free(old);
1753         free(new);
1754         return offset;
1755 }
1756
1757 static int apply_binary_fragment(struct buffer_desc *desc, struct patch *patch)
1758 {
1759         unsigned long dst_size;
1760         struct fragment *fragment = patch->fragments;
1761         void *data;
1762         void *result;
1763
1764         /* Binary patch is irreversible without the optional second hunk */
1765         if (apply_in_reverse) {
1766                 if (!fragment->next)
1767                         return error("cannot reverse-apply a binary patch "
1768                                      "without the reverse hunk to '%s'",
1769                                      patch->new_name
1770                                      ? patch->new_name : patch->old_name);
1771                 fragment = fragment->next;
1772         }
1773         data = (void*) fragment->patch;
1774         switch (fragment->binary_patch_method) {
1775         case BINARY_DELTA_DEFLATED:
1776                 result = patch_delta(desc->buffer, desc->size,
1777                                      data,
1778                                      fragment->size,
1779                                      &dst_size);
1780                 free(desc->buffer);
1781                 desc->buffer = result;
1782                 break;
1783         case BINARY_LITERAL_DEFLATED:
1784                 free(desc->buffer);
1785                 desc->buffer = data;
1786                 dst_size = fragment->size;
1787                 break;
1788         }
1789         if (!desc->buffer)
1790                 return -1;
1791         desc->size = desc->alloc = dst_size;
1792         return 0;
1793 }
1794
1795 static int apply_binary(struct buffer_desc *desc, struct patch *patch)
1796 {
1797         const char *name = patch->old_name ? patch->old_name : patch->new_name;
1798         unsigned char sha1[20];
1799
1800         /* For safety, we require patch index line to contain
1801          * full 40-byte textual SHA1 for old and new, at least for now.
1802          */
1803         if (strlen(patch->old_sha1_prefix) != 40 ||
1804             strlen(patch->new_sha1_prefix) != 40 ||
1805             get_sha1_hex(patch->old_sha1_prefix, sha1) ||
1806             get_sha1_hex(patch->new_sha1_prefix, sha1))
1807                 return error("cannot apply binary patch to '%s' "
1808                              "without full index line", name);
1809
1810         if (patch->old_name) {
1811                 /* See if the old one matches what the patch
1812                  * applies to.
1813                  */
1814                 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1815                 if (strcmp(sha1_to_hex(sha1), patch->old_sha1_prefix))
1816                         return error("the patch applies to '%s' (%s), "
1817                                      "which does not match the "
1818                                      "current contents.",
1819                                      name, sha1_to_hex(sha1));
1820         }
1821         else {
1822                 /* Otherwise, the old one must be empty. */
1823                 if (desc->size)
1824                         return error("the patch applies to an empty "
1825                                      "'%s' but it is not empty", name);
1826         }
1827
1828         get_sha1_hex(patch->new_sha1_prefix, sha1);
1829         if (is_null_sha1(sha1)) {
1830                 free(desc->buffer);
1831                 desc->alloc = desc->size = 0;
1832                 desc->buffer = NULL;
1833                 return 0; /* deletion patch */
1834         }
1835
1836         if (has_sha1_file(sha1)) {
1837                 /* We already have the postimage */
1838                 char type[10];
1839                 unsigned long size;
1840
1841                 free(desc->buffer);
1842                 desc->buffer = read_sha1_file(sha1, type, &size);
1843                 if (!desc->buffer)
1844                         return error("the necessary postimage %s for "
1845                                      "'%s' cannot be read",
1846                                      patch->new_sha1_prefix, name);
1847                 desc->alloc = desc->size = size;
1848         }
1849         else {
1850                 /* We have verified desc matches the preimage;
1851                  * apply the patch data to it, which is stored
1852                  * in the patch->fragments->{patch,size}.
1853                  */
1854                 if (apply_binary_fragment(desc, patch))
1855                         return error("binary patch does not apply to '%s'",
1856                                      name);
1857
1858                 /* verify that the result matches */
1859                 hash_sha1_file(desc->buffer, desc->size, blob_type, sha1);
1860                 if (strcmp(sha1_to_hex(sha1), patch->new_sha1_prefix))
1861                         return error("binary patch to '%s' creates incorrect result (expecting %s, got %s)", name, patch->new_sha1_prefix, sha1_to_hex(sha1));
1862         }
1863
1864         return 0;
1865 }
1866
1867 static int apply_fragments(struct buffer_desc *desc, struct patch *patch)
1868 {
1869         struct fragment *frag = patch->fragments;
1870         const char *name = patch->old_name ? patch->old_name : patch->new_name;
1871
1872         if (patch->is_binary)
1873                 return apply_binary(desc, patch);
1874
1875         while (frag) {
1876                 if (apply_one_fragment(desc, frag, patch->inaccurate_eof)) {
1877                         error("patch failed: %s:%ld", name, frag->oldpos);
1878                         if (!apply_with_reject)
1879                                 return -1;
1880                         frag->rejected = 1;
1881                 }
1882                 frag = frag->next;
1883         }
1884         return 0;
1885 }
1886
1887 static int apply_data(struct patch *patch, struct stat *st, struct cache_entry *ce)
1888 {
1889         char *buf;
1890         unsigned long size, alloc;
1891         struct buffer_desc desc;
1892
1893         size = 0;
1894         alloc = 0;
1895         buf = NULL;
1896         if (cached) {
1897                 if (ce) {
1898                         char type[20];
1899                         buf = read_sha1_file(ce->sha1, type, &size);
1900                         if (!buf)
1901                                 return error("read of %s failed",
1902                                              patch->old_name);
1903                         alloc = size;
1904                 }
1905         }
1906         else if (patch->old_name) {
1907                 size = st->st_size;
1908                 alloc = size + 8192;
1909                 buf = xmalloc(alloc);
1910                 if (read_old_data(st, patch->old_name, buf, alloc) != size)
1911                         return error("read of %s failed", patch->old_name);
1912         }
1913
1914         desc.size = size;
1915         desc.alloc = alloc;
1916         desc.buffer = buf;
1917
1918         if (apply_fragments(&desc, patch) < 0)
1919                 return -1; /* note with --reject this succeeds. */
1920
1921         /* NUL terminate the result */
1922         if (desc.alloc <= desc.size)
1923                 desc.buffer = xrealloc(desc.buffer, desc.size + 1);
1924         desc.buffer[desc.size] = 0;
1925
1926         patch->result = desc.buffer;
1927         patch->resultsize = desc.size;
1928
1929         if (0 < patch->is_delete && patch->resultsize)
1930                 return error("removal patch leaves file contents");
1931
1932         return 0;
1933 }
1934
1935 static int check_patch(struct patch *patch, struct patch *prev_patch)
1936 {
1937         struct stat st;
1938         const char *old_name = patch->old_name;
1939         const char *new_name = patch->new_name;
1940         const char *name = old_name ? old_name : new_name;
1941         struct cache_entry *ce = NULL;
1942         int ok_if_exists;
1943
1944         patch->rejected = 1; /* we will drop this after we succeed */
1945         if (old_name) {
1946                 int changed = 0;
1947                 int stat_ret = 0;
1948                 unsigned st_mode = 0;
1949
1950                 if (!cached)
1951                         stat_ret = lstat(old_name, &st);
1952                 if (check_index) {
1953                         int pos = cache_name_pos(old_name, strlen(old_name));
1954                         if (pos < 0)
1955                                 return error("%s: does not exist in index",
1956                                              old_name);
1957                         ce = active_cache[pos];
1958                         if (stat_ret < 0) {
1959                                 struct checkout costate;
1960                                 if (errno != ENOENT)
1961                                         return error("%s: %s", old_name,
1962                                                      strerror(errno));
1963                                 /* checkout */
1964                                 costate.base_dir = "";
1965                                 costate.base_dir_len = 0;
1966                                 costate.force = 0;
1967                                 costate.quiet = 0;
1968                                 costate.not_new = 0;
1969                                 costate.refresh_cache = 1;
1970                                 if (checkout_entry(ce,
1971                                                    &costate,
1972                                                    NULL) ||
1973                                     lstat(old_name, &st))
1974                                         return -1;
1975                         }
1976                         if (!cached)
1977                                 changed = ce_match_stat(ce, &st, 1);
1978                         if (changed)
1979                                 return error("%s: does not match index",
1980                                              old_name);
1981                         if (cached)
1982                                 st_mode = ntohl(ce->ce_mode);
1983                 }
1984                 else if (stat_ret < 0)
1985                         return error("%s: %s", old_name, strerror(errno));
1986
1987                 if (!cached)
1988                         st_mode = ntohl(create_ce_mode(st.st_mode));
1989
1990                 if (patch->is_new < 0)
1991                         patch->is_new = 0;
1992                 if (!patch->old_mode)
1993                         patch->old_mode = st_mode;
1994                 if ((st_mode ^ patch->old_mode) & S_IFMT)
1995                         return error("%s: wrong type", old_name);
1996                 if (st_mode != patch->old_mode)
1997                         fprintf(stderr, "warning: %s has type %o, expected %o\n",
1998                                 old_name, st_mode, patch->old_mode);
1999         }
2000
2001         if (new_name && prev_patch && 0 < prev_patch->is_delete &&
2002             !strcmp(prev_patch->old_name, new_name))
2003                 /* A type-change diff is always split into a patch to
2004                  * delete old, immediately followed by a patch to
2005                  * create new (see diff.c::run_diff()); in such a case
2006                  * it is Ok that the entry to be deleted by the
2007                  * previous patch is still in the working tree and in
2008                  * the index.
2009                  */
2010                 ok_if_exists = 1;
2011         else
2012                 ok_if_exists = 0;
2013
2014         if (new_name &&
2015             ((0 < patch->is_new) | (0 < patch->is_rename) | patch->is_copy)) {
2016                 if (check_index &&
2017                     cache_name_pos(new_name, strlen(new_name)) >= 0 &&
2018                     !ok_if_exists)
2019                         return error("%s: already exists in index", new_name);
2020                 if (!cached) {
2021                         struct stat nst;
2022                         if (!lstat(new_name, &nst)) {
2023                                 if (S_ISDIR(nst.st_mode) || ok_if_exists)
2024                                         ; /* ok */
2025                                 else
2026                                         return error("%s: already exists in working directory", new_name);
2027                         }
2028                         else if ((errno != ENOENT) && (errno != ENOTDIR))
2029                                 return error("%s: %s", new_name, strerror(errno));
2030                 }
2031                 if (!patch->new_mode) {
2032                         if (0 < patch->is_new)
2033                                 patch->new_mode = S_IFREG | 0644;
2034                         else
2035                                 patch->new_mode = patch->old_mode;
2036                 }
2037         }
2038
2039         if (new_name && old_name) {
2040                 int same = !strcmp(old_name, new_name);
2041                 if (!patch->new_mode)
2042                         patch->new_mode = patch->old_mode;
2043                 if ((patch->old_mode ^ patch->new_mode) & S_IFMT)
2044                         return error("new mode (%o) of %s does not match old mode (%o)%s%s",
2045                                 patch->new_mode, new_name, patch->old_mode,
2046                                 same ? "" : " of ", same ? "" : old_name);
2047         }
2048
2049         if (apply_data(patch, &st, ce) < 0)
2050                 return error("%s: patch does not apply", name);
2051         patch->rejected = 0;
2052         return 0;
2053 }
2054
2055 static int check_patch_list(struct patch *patch)
2056 {
2057         struct patch *prev_patch = NULL;
2058         int err = 0;
2059
2060         for (prev_patch = NULL; patch ; patch = patch->next) {
2061                 if (apply_verbosely)
2062                         say_patch_name(stderr,
2063                                        "Checking patch ", patch, "...\n");
2064                 err |= check_patch(patch, prev_patch);
2065                 prev_patch = patch;
2066         }
2067         return err;
2068 }
2069
2070 static void show_index_list(struct patch *list)
2071 {
2072         struct patch *patch;
2073
2074         /* Once we start supporting the reverse patch, it may be
2075          * worth showing the new sha1 prefix, but until then...
2076          */
2077         for (patch = list; patch; patch = patch->next) {
2078                 const unsigned char *sha1_ptr;
2079                 unsigned char sha1[20];
2080                 const char *name;
2081
2082                 name = patch->old_name ? patch->old_name : patch->new_name;
2083                 if (0 < patch->is_new)
2084                         sha1_ptr = null_sha1;
2085                 else if (get_sha1(patch->old_sha1_prefix, sha1))
2086                         die("sha1 information is lacking or useless (%s).",
2087                             name);
2088                 else
2089                         sha1_ptr = sha1;
2090
2091                 printf("%06o %s ",patch->old_mode, sha1_to_hex(sha1_ptr));
2092                 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2093                         quote_c_style(name, NULL, stdout, 0);
2094                 else
2095                         fputs(name, stdout);
2096                 putchar(line_termination);
2097         }
2098 }
2099
2100 static void stat_patch_list(struct patch *patch)
2101 {
2102         int files, adds, dels;
2103
2104         for (files = adds = dels = 0 ; patch ; patch = patch->next) {
2105                 files++;
2106                 adds += patch->lines_added;
2107                 dels += patch->lines_deleted;
2108                 show_stats(patch);
2109         }
2110
2111         printf(" %d files changed, %d insertions(+), %d deletions(-)\n", files, adds, dels);
2112 }
2113
2114 static void numstat_patch_list(struct patch *patch)
2115 {
2116         for ( ; patch; patch = patch->next) {
2117                 const char *name;
2118                 name = patch->new_name ? patch->new_name : patch->old_name;
2119                 printf("%d\t%d\t", patch->lines_added, patch->lines_deleted);
2120                 if (line_termination && quote_c_style(name, NULL, NULL, 0))
2121                         quote_c_style(name, NULL, stdout, 0);
2122                 else
2123                         fputs(name, stdout);
2124                 putchar(line_termination);
2125         }
2126 }
2127
2128 static void show_file_mode_name(const char *newdelete, unsigned int mode, const char *name)
2129 {
2130         if (mode)
2131                 printf(" %s mode %06o %s\n", newdelete, mode, name);
2132         else
2133                 printf(" %s %s\n", newdelete, name);
2134 }
2135
2136 static void show_mode_change(struct patch *p, int show_name)
2137 {
2138         if (p->old_mode && p->new_mode && p->old_mode != p->new_mode) {
2139                 if (show_name)
2140                         printf(" mode change %06o => %06o %s\n",
2141                                p->old_mode, p->new_mode, p->new_name);
2142                 else
2143                         printf(" mode change %06o => %06o\n",
2144                                p->old_mode, p->new_mode);
2145         }
2146 }
2147
2148 static void show_rename_copy(struct patch *p)
2149 {
2150         const char *renamecopy = p->is_rename ? "rename" : "copy";
2151         const char *old, *new;
2152
2153         /* Find common prefix */
2154         old = p->old_name;
2155         new = p->new_name;
2156         while (1) {
2157                 const char *slash_old, *slash_new;
2158                 slash_old = strchr(old, '/');
2159                 slash_new = strchr(new, '/');
2160                 if (!slash_old ||
2161                     !slash_new ||
2162                     slash_old - old != slash_new - new ||
2163                     memcmp(old, new, slash_new - new))
2164                         break;
2165                 old = slash_old + 1;
2166                 new = slash_new + 1;
2167         }
2168         /* p->old_name thru old is the common prefix, and old and new
2169          * through the end of names are renames
2170          */
2171         if (old != p->old_name)
2172                 printf(" %s %.*s{%s => %s} (%d%%)\n", renamecopy,
2173                        (int)(old - p->old_name), p->old_name,
2174                        old, new, p->score);
2175         else
2176                 printf(" %s %s => %s (%d%%)\n", renamecopy,
2177                        p->old_name, p->new_name, p->score);
2178         show_mode_change(p, 0);
2179 }
2180
2181 static void summary_patch_list(struct patch *patch)
2182 {
2183         struct patch *p;
2184
2185         for (p = patch; p; p = p->next) {
2186                 if (p->is_new)
2187                         show_file_mode_name("create", p->new_mode, p->new_name);
2188                 else if (p->is_delete)
2189                         show_file_mode_name("delete", p->old_mode, p->old_name);
2190                 else {
2191                         if (p->is_rename || p->is_copy)
2192                                 show_rename_copy(p);
2193                         else {
2194                                 if (p->score) {
2195                                         printf(" rewrite %s (%d%%)\n",
2196                                                p->new_name, p->score);
2197                                         show_mode_change(p, 0);
2198                                 }
2199                                 else
2200                                         show_mode_change(p, 1);
2201                         }
2202                 }
2203         }
2204 }
2205
2206 static void patch_stats(struct patch *patch)
2207 {
2208         int lines = patch->lines_added + patch->lines_deleted;
2209
2210         if (lines > max_change)
2211                 max_change = lines;
2212         if (patch->old_name) {
2213                 int len = quote_c_style(patch->old_name, NULL, NULL, 0);
2214                 if (!len)
2215                         len = strlen(patch->old_name);
2216                 if (len > max_len)
2217                         max_len = len;
2218         }
2219         if (patch->new_name) {
2220                 int len = quote_c_style(patch->new_name, NULL, NULL, 0);
2221                 if (!len)
2222                         len = strlen(patch->new_name);
2223                 if (len > max_len)
2224                         max_len = len;
2225         }
2226 }
2227
2228 static void remove_file(struct patch *patch)
2229 {
2230         if (write_index) {
2231                 if (remove_file_from_cache(patch->old_name) < 0)
2232                         die("unable to remove %s from index", patch->old_name);
2233                 cache_tree_invalidate_path(active_cache_tree, patch->old_name);
2234         }
2235         if (!cached)
2236                 unlink(patch->old_name);
2237 }
2238
2239 static void add_index_file(const char *path, unsigned mode, void *buf, unsigned long size)
2240 {
2241         struct stat st;
2242         struct cache_entry *ce;
2243         int namelen = strlen(path);
2244         unsigned ce_size = cache_entry_size(namelen);
2245
2246         if (!write_index)
2247                 return;
2248
2249         ce = xcalloc(1, ce_size);
2250         memcpy(ce->name, path, namelen);
2251         ce->ce_mode = create_ce_mode(mode);
2252         ce->ce_flags = htons(namelen);
2253         if (!cached) {
2254                 if (lstat(path, &st) < 0)
2255                         die("unable to stat newly created file %s", path);
2256                 fill_stat_cache_info(ce, &st);
2257         }
2258         if (write_sha1_file(buf, size, blob_type, ce->sha1) < 0)
2259                 die("unable to create backing store for newly created file %s", path);
2260         if (add_cache_entry(ce, ADD_CACHE_OK_TO_ADD) < 0)
2261                 die("unable to add cache entry for %s", path);
2262 }
2263
2264 static int try_create_file(const char *path, unsigned int mode, const char *buf, unsigned long size)
2265 {
2266         int fd;
2267
2268         if (S_ISLNK(mode))
2269                 /* Although buf:size is counted string, it also is NUL
2270                  * terminated.
2271                  */
2272                 return symlink(buf, path);
2273         fd = open(path, O_CREAT | O_EXCL | O_WRONLY, (mode & 0100) ? 0777 : 0666);
2274         if (fd < 0)
2275                 return -1;
2276         while (size) {
2277                 int written = xwrite(fd, buf, size);
2278                 if (written < 0)
2279                         die("writing file %s: %s", path, strerror(errno));
2280                 if (!written)
2281                         die("out of space writing file %s", path);
2282                 buf += written;
2283                 size -= written;
2284         }
2285         if (close(fd) < 0)
2286                 die("closing file %s: %s", path, strerror(errno));
2287         return 0;
2288 }
2289
2290 /*
2291  * We optimistically assume that the directories exist,
2292  * which is true 99% of the time anyway. If they don't,
2293  * we create them and try again.
2294  */
2295 static void create_one_file(char *path, unsigned mode, const char *buf, unsigned long size)
2296 {
2297         if (cached)
2298                 return;
2299         if (!try_create_file(path, mode, buf, size))
2300                 return;
2301
2302         if (errno == ENOENT) {
2303                 if (safe_create_leading_directories(path))
2304                         return;
2305                 if (!try_create_file(path, mode, buf, size))
2306                         return;
2307         }
2308
2309         if (errno == EEXIST || errno == EACCES) {
2310                 /* We may be trying to create a file where a directory
2311                  * used to be.
2312                  */
2313                 struct stat st;
2314                 errno = 0;
2315                 if (!lstat(path, &st) && S_ISDIR(st.st_mode) && !rmdir(path))
2316                         errno = EEXIST;
2317         }
2318
2319         if (errno == EEXIST) {
2320                 unsigned int nr = getpid();
2321
2322                 for (;;) {
2323                         const char *newpath;
2324                         newpath = mkpath("%s~%u", path, nr);
2325                         if (!try_create_file(newpath, mode, buf, size)) {
2326                                 if (!rename(newpath, path))
2327                                         return;
2328                                 unlink(newpath);
2329                                 break;
2330                         }
2331                         if (errno != EEXIST)
2332                                 break;
2333                         ++nr;
2334                 }
2335         }
2336         die("unable to write file %s mode %o", path, mode);
2337 }
2338
2339 static void create_file(struct patch *patch)
2340 {
2341         char *path = patch->new_name;
2342         unsigned mode = patch->new_mode;
2343         unsigned long size = patch->resultsize;
2344         char *buf = patch->result;
2345
2346         if (!mode)
2347                 mode = S_IFREG | 0644;
2348         create_one_file(path, mode, buf, size);
2349         add_index_file(path, mode, buf, size);
2350         cache_tree_invalidate_path(active_cache_tree, path);
2351 }
2352
2353 /* phase zero is to remove, phase one is to create */
2354 static void write_out_one_result(struct patch *patch, int phase)
2355 {
2356         if (patch->is_delete > 0) {
2357                 if (phase == 0)
2358                         remove_file(patch);
2359                 return;
2360         }
2361         if (patch->is_new > 0 || patch->is_copy) {
2362                 if (phase == 1)
2363                         create_file(patch);
2364                 return;
2365         }
2366         /*
2367          * Rename or modification boils down to the same
2368          * thing: remove the old, write the new
2369          */
2370         if (phase == 0)
2371                 remove_file(patch);
2372         if (phase == 1)
2373                 create_file(patch);
2374 }
2375
2376 static int write_out_one_reject(struct patch *patch)
2377 {
2378         FILE *rej;
2379         char namebuf[PATH_MAX];
2380         struct fragment *frag;
2381         int cnt = 0;
2382
2383         for (cnt = 0, frag = patch->fragments; frag; frag = frag->next) {
2384                 if (!frag->rejected)
2385                         continue;
2386                 cnt++;
2387         }
2388
2389         if (!cnt) {
2390                 if (apply_verbosely)
2391                         say_patch_name(stderr,
2392                                        "Applied patch ", patch, " cleanly.\n");
2393                 return 0;
2394         }
2395
2396         /* This should not happen, because a removal patch that leaves
2397          * contents are marked "rejected" at the patch level.
2398          */
2399         if (!patch->new_name)
2400                 die("internal error");
2401
2402         /* Say this even without --verbose */
2403         say_patch_name(stderr, "Applying patch ", patch, " with");
2404         fprintf(stderr, " %d rejects...\n", cnt);
2405
2406         cnt = strlen(patch->new_name);
2407         if (ARRAY_SIZE(namebuf) <= cnt + 5) {
2408                 cnt = ARRAY_SIZE(namebuf) - 5;
2409                 fprintf(stderr,
2410                         "warning: truncating .rej filename to %.*s.rej",
2411                         cnt - 1, patch->new_name);
2412         }
2413         memcpy(namebuf, patch->new_name, cnt);
2414         memcpy(namebuf + cnt, ".rej", 5);
2415
2416         rej = fopen(namebuf, "w");
2417         if (!rej)
2418                 return error("cannot open %s: %s", namebuf, strerror(errno));
2419
2420         /* Normal git tools never deal with .rej, so do not pretend
2421          * this is a git patch by saying --git nor give extended
2422          * headers.  While at it, maybe please "kompare" that wants
2423          * the trailing TAB and some garbage at the end of line ;-).
2424          */
2425         fprintf(rej, "diff a/%s b/%s\t(rejected hunks)\n",
2426                 patch->new_name, patch->new_name);
2427         for (cnt = 1, frag = patch->fragments;
2428              frag;
2429              cnt++, frag = frag->next) {
2430                 if (!frag->rejected) {
2431                         fprintf(stderr, "Hunk #%d applied cleanly.\n", cnt);
2432                         continue;
2433                 }
2434                 fprintf(stderr, "Rejected hunk #%d.\n", cnt);
2435                 fprintf(rej, "%.*s", frag->size, frag->patch);
2436                 if (frag->patch[frag->size-1] != '\n')
2437                         fputc('\n', rej);
2438         }
2439         fclose(rej);
2440         return -1;
2441 }
2442
2443 static int write_out_results(struct patch *list, int skipped_patch)
2444 {
2445         int phase;
2446         int errs = 0;
2447         struct patch *l;
2448
2449         if (!list && !skipped_patch)
2450                 return error("No changes");
2451
2452         for (phase = 0; phase < 2; phase++) {
2453                 l = list;
2454                 while (l) {
2455                         if (l->rejected)
2456                                 errs = 1;
2457                         else {
2458                                 write_out_one_result(l, phase);
2459                                 if (phase == 1 && write_out_one_reject(l))
2460                                         errs = 1;
2461                         }
2462                         l = l->next;
2463                 }
2464         }
2465         return errs;
2466 }
2467
2468 static struct lock_file lock_file;
2469
2470 static struct excludes {
2471         struct excludes *next;
2472         const char *path;
2473 } *excludes;
2474
2475 static int use_patch(struct patch *p)
2476 {
2477         const char *pathname = p->new_name ? p->new_name : p->old_name;
2478         struct excludes *x = excludes;
2479         while (x) {
2480                 if (fnmatch(x->path, pathname, 0) == 0)
2481                         return 0;
2482                 x = x->next;
2483         }
2484         if (0 < prefix_length) {
2485                 int pathlen = strlen(pathname);
2486                 if (pathlen <= prefix_length ||
2487                     memcmp(prefix, pathname, prefix_length))
2488                         return 0;
2489         }
2490         return 1;
2491 }
2492
2493 static int apply_patch(int fd, const char *filename, int inaccurate_eof)
2494 {
2495         unsigned long offset, size;
2496         char *buffer = read_patch_file(fd, &size);
2497         struct patch *list = NULL, **listp = &list;
2498         int skipped_patch = 0;
2499
2500         patch_input_file = filename;
2501         if (!buffer)
2502                 return -1;
2503         offset = 0;
2504         while (size > 0) {
2505                 struct patch *patch;
2506                 int nr;
2507
2508                 patch = xcalloc(1, sizeof(*patch));
2509                 patch->inaccurate_eof = inaccurate_eof;
2510                 nr = parse_chunk(buffer + offset, size, patch);
2511                 if (nr < 0)
2512                         break;
2513                 if (apply_in_reverse)
2514                         reverse_patches(patch);
2515                 if (use_patch(patch)) {
2516                         patch_stats(patch);
2517                         *listp = patch;
2518                         listp = &patch->next;
2519                 } else {
2520                         /* perhaps free it a bit better? */
2521                         free(patch);
2522                         skipped_patch++;
2523                 }
2524                 offset += nr;
2525                 size -= nr;
2526         }
2527
2528         if (whitespace_error && (new_whitespace == error_on_whitespace))
2529                 apply = 0;
2530
2531         write_index = check_index && apply;
2532         if (write_index && newfd < 0)
2533                 newfd = hold_lock_file_for_update(&lock_file,
2534                                                   get_index_file(), 1);
2535         if (check_index) {
2536                 if (read_cache() < 0)
2537                         die("unable to read index file");
2538         }
2539
2540         if ((check || apply) &&
2541             check_patch_list(list) < 0 &&
2542             !apply_with_reject)
2543                 exit(1);
2544
2545         if (apply && write_out_results(list, skipped_patch))
2546                 exit(1);
2547
2548         if (show_index_info)
2549                 show_index_list(list);
2550
2551         if (diffstat)
2552                 stat_patch_list(list);
2553
2554         if (numstat)
2555                 numstat_patch_list(list);
2556
2557         if (summary)
2558                 summary_patch_list(list);
2559
2560         free(buffer);
2561         return 0;
2562 }
2563
2564 static int git_apply_config(const char *var, const char *value)
2565 {
2566         if (!strcmp(var, "apply.whitespace")) {
2567                 apply_default_whitespace = xstrdup(value);
2568                 return 0;
2569         }
2570         return git_default_config(var, value);
2571 }
2572
2573
2574 int cmd_apply(int argc, const char **argv, const char *prefix)
2575 {
2576         int i;
2577         int read_stdin = 1;
2578         int inaccurate_eof = 0;
2579         int errs = 0;
2580
2581         const char *whitespace_option = NULL;
2582
2583         for (i = 1; i < argc; i++) {
2584                 const char *arg = argv[i];
2585                 char *end;
2586                 int fd;
2587
2588                 if (!strcmp(arg, "-")) {
2589                         errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2590                         read_stdin = 0;
2591                         continue;
2592                 }
2593                 if (!strncmp(arg, "--exclude=", 10)) {
2594                         struct excludes *x = xmalloc(sizeof(*x));
2595                         x->path = arg + 10;
2596                         x->next = excludes;
2597                         excludes = x;
2598                         continue;
2599                 }
2600                 if (!strncmp(arg, "-p", 2)) {
2601                         p_value = atoi(arg + 2);
2602                         continue;
2603                 }
2604                 if (!strcmp(arg, "--no-add")) {
2605                         no_add = 1;
2606                         continue;
2607                 }
2608                 if (!strcmp(arg, "--stat")) {
2609                         apply = 0;
2610                         diffstat = 1;
2611                         continue;
2612                 }
2613                 if (!strcmp(arg, "--allow-binary-replacement") ||
2614                     !strcmp(arg, "--binary")) {
2615                         continue; /* now no-op */
2616                 }
2617                 if (!strcmp(arg, "--numstat")) {
2618                         apply = 0;
2619                         numstat = 1;
2620                         continue;
2621                 }
2622                 if (!strcmp(arg, "--summary")) {
2623                         apply = 0;
2624                         summary = 1;
2625                         continue;
2626                 }
2627                 if (!strcmp(arg, "--check")) {
2628                         apply = 0;
2629                         check = 1;
2630                         continue;
2631                 }
2632                 if (!strcmp(arg, "--index")) {
2633                         check_index = 1;
2634                         continue;
2635                 }
2636                 if (!strcmp(arg, "--cached")) {
2637                         check_index = 1;
2638                         cached = 1;
2639                         continue;
2640                 }
2641                 if (!strcmp(arg, "--apply")) {
2642                         apply = 1;
2643                         continue;
2644                 }
2645                 if (!strcmp(arg, "--index-info")) {
2646                         apply = 0;
2647                         show_index_info = 1;
2648                         continue;
2649                 }
2650                 if (!strcmp(arg, "-z")) {
2651                         line_termination = 0;
2652                         continue;
2653                 }
2654                 if (!strncmp(arg, "-C", 2)) {
2655                         p_context = strtoul(arg + 2, &end, 0);
2656                         if (*end != '\0')
2657                                 die("unrecognized context count '%s'", arg + 2);
2658                         continue;
2659                 }
2660                 if (!strncmp(arg, "--whitespace=", 13)) {
2661                         whitespace_option = arg + 13;
2662                         parse_whitespace_option(arg + 13);
2663                         continue;
2664                 }
2665                 if (!strcmp(arg, "-R") || !strcmp(arg, "--reverse")) {
2666                         apply_in_reverse = 1;
2667                         continue;
2668                 }
2669                 if (!strcmp(arg, "--unidiff-zero")) {
2670                         unidiff_zero = 1;
2671                         continue;
2672                 }
2673                 if (!strcmp(arg, "--reject")) {
2674                         apply = apply_with_reject = apply_verbosely = 1;
2675                         continue;
2676                 }
2677                 if (!strcmp(arg, "--verbose")) {
2678                         apply_verbosely = 1;
2679                         continue;
2680                 }
2681                 if (!strcmp(arg, "--inaccurate-eof")) {
2682                         inaccurate_eof = 1;
2683                         continue;
2684                 }
2685
2686                 if (check_index && prefix_length < 0) {
2687                         prefix = setup_git_directory();
2688                         prefix_length = prefix ? strlen(prefix) : 0;
2689                         git_config(git_apply_config);
2690                         if (!whitespace_option && apply_default_whitespace)
2691                                 parse_whitespace_option(apply_default_whitespace);
2692                 }
2693                 if (0 < prefix_length)
2694                         arg = prefix_filename(prefix, prefix_length, arg);
2695
2696                 fd = open(arg, O_RDONLY);
2697                 if (fd < 0)
2698                         usage(apply_usage);
2699                 read_stdin = 0;
2700                 set_default_whitespace_mode(whitespace_option);
2701                 errs |= apply_patch(fd, arg, inaccurate_eof);
2702                 close(fd);
2703         }
2704         set_default_whitespace_mode(whitespace_option);
2705         if (read_stdin)
2706                 errs |= apply_patch(0, "<stdin>", inaccurate_eof);
2707         if (whitespace_error) {
2708                 if (squelch_whitespace_errors &&
2709                     squelch_whitespace_errors < whitespace_error) {
2710                         int squelched =
2711                                 whitespace_error - squelch_whitespace_errors;
2712                         fprintf(stderr, "warning: squelched %d "
2713                                 "whitespace error%s\n",
2714                                 squelched,
2715                                 squelched == 1 ? "" : "s");
2716                 }
2717                 if (new_whitespace == error_on_whitespace)
2718                         die("%d line%s add%s trailing whitespaces.",
2719                             whitespace_error,
2720                             whitespace_error == 1 ? "" : "s",
2721                             whitespace_error == 1 ? "s" : "");
2722                 if (applied_after_stripping)
2723                         fprintf(stderr, "warning: %d line%s applied after"
2724                                 " stripping trailing whitespaces.\n",
2725                                 applied_after_stripping,
2726                                 applied_after_stripping == 1 ? "" : "s");
2727                 else if (whitespace_error)
2728                         fprintf(stderr, "warning: %d line%s add%s trailing"
2729                                 " whitespaces.\n",
2730                                 whitespace_error,
2731                                 whitespace_error == 1 ? "" : "s",
2732                                 whitespace_error == 1 ? "s" : "");
2733         }
2734
2735         if (write_index) {
2736                 if (write_cache(newfd, active_cache, active_nr) ||
2737                     close(newfd) || commit_lock_file(&lock_file))
2738                         die("Unable to write new index file");
2739         }
2740
2741         return !!errs;
2742 }